public static TiffBasePlanarColorDecoder <TPixel> CreatePlanar(
TiffColorType colorType,
TiffBitsPerSample bitsPerSample,
ushort[] colorMap,
Rational[] referenceBlackAndWhite,
Rational[] ycbcrCoefficients,
ushort[] ycbcrSubSampling,
ByteOrder byteOrder)
{
switch (colorType)
{
case TiffColorType.Rgb888Planar:
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new RgbPlanarTiffColor <TPixel>(bitsPerSample));
case TiffColorType.YCbCrPlanar:
return(new YCbCrPlanarTiffColor <TPixel>(referenceBlackAndWhite, ycbcrCoefficients, ycbcrSubSampling));
case TiffColorType.Rgb161616Planar:
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new Rgb16PlanarTiffColor <TPixel>(byteOrder == ByteOrder.BigEndian));
case TiffColorType.Rgb242424Planar:
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new Rgb24PlanarTiffColor <TPixel>(byteOrder == ByteOrder.BigEndian));
case TiffColorType.Rgb323232Planar:
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new Rgb32PlanarTiffColor <TPixel>(byteOrder == ByteOrder.BigEndian));
default:
throw TiffThrowHelper.InvalidColorType(colorType.ToString());
}
}
public virtual void Write(TiffBaseCompressor compressor, int rowsPerStrip)
{
DebugGuard.IsTrue(this.BytesPerRow == compressor.BytesPerRow, "bytes per row of the compressor does not match tiff color writer");
int stripsCount = (this.Image.Height + rowsPerStrip - 1) / rowsPerStrip;
uint[] stripOffsets = new uint[stripsCount];
uint[] stripByteCounts = new uint[stripsCount];
int stripIndex = 0;
compressor.Initialize(rowsPerStrip);
for (int y = 0; y < this.Image.Height; y += rowsPerStrip)
{
long offset = compressor.Output.Position;
int height = Math.Min(rowsPerStrip, this.Image.Height - y);
this.EncodeStrip(y, height, compressor);
long endOffset = compressor.Output.Position;
stripOffsets[stripIndex] = (uint)offset;
stripByteCounts[stripIndex] = (uint)(endOffset - offset);
stripIndex++;
}
DebugGuard.IsTrue(stripIndex == stripsCount, "stripIndex and stripsCount should match");
this.AddStripTags(rowsPerStrip, stripOffsets, stripByteCounts);
}
/// <summary>
/// Initializes a new instance of the <see cref="TiffCcittCompressor" /> class.
/// </summary>
/// <param name="output">The output.</param>
/// <param name="allocator">The allocator.</param>
/// <param name="width">The width.</param>
/// <param name="bitsPerPixel">The bits per pixel.</param>
protected TiffCcittCompressor(Stream output, MemoryAllocator allocator, int width, int bitsPerPixel)
: base(output, allocator, width, bitsPerPixel)
{
DebugGuard.IsTrue(bitsPerPixel == 1, nameof(bitsPerPixel), "CCITT compression requires one bit per pixel");
this.bytePosition = 0;
this.bitPosition = 0;
}
/// <inheritdoc/>
public CieXyz Convert(LinearRgb input)
{
DebugGuard.IsTrue(input.WorkingSpace.Equals(this.SourceWorkingSpace), nameof(input.WorkingSpace), "Input and source working spaces must be equal.");
Vector3 vector = Vector3.Transform(input.Vector, this.conversionMatrix);
return(new CieXyz(vector));
}
public static TiffBaseCompressor Create(
TiffCompression method,
Stream output,
MemoryAllocator allocator,
int width,
int bitsPerPixel,
DeflateCompressionLevel compressionLevel,
TiffPredictor predictor)
{
switch (method)
{
// The following compression types are not implemented in the encoder and will default to no compression instead.
case TiffCompression.ItuTRecT43:
case TiffCompression.ItuTRecT82:
case TiffCompression.OldJpeg:
case TiffCompression.OldDeflate:
case TiffCompression.None:
DebugGuard.IsTrue(compressionLevel == DeflateCompressionLevel.DefaultCompression, "No deflate compression level is expected to be set");
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
return(new NoCompressor(output, allocator, width, bitsPerPixel));
case TiffCompression.Jpeg:
DebugGuard.IsTrue(compressionLevel == DeflateCompressionLevel.DefaultCompression, "No deflate compression level is expected to be set");
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
return(new TiffJpegCompressor(output, allocator, width, bitsPerPixel));
case TiffCompression.PackBits:
DebugGuard.IsTrue(compressionLevel == DeflateCompressionLevel.DefaultCompression, "No deflate compression level is expected to be set");
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
return(new PackBitsCompressor(output, allocator, width, bitsPerPixel));
case TiffCompression.Deflate:
return(new DeflateCompressor(output, allocator, width, bitsPerPixel, predictor, compressionLevel));
case TiffCompression.Lzw:
DebugGuard.IsTrue(compressionLevel == DeflateCompressionLevel.DefaultCompression, "No deflate compression level is expected to be set");
return(new LzwCompressor(output, allocator, width, bitsPerPixel, predictor));
case TiffCompression.CcittGroup3Fax:
DebugGuard.IsTrue(compressionLevel == DeflateCompressionLevel.DefaultCompression, "No deflate compression level is expected to be set");
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
return(new T4BitCompressor(output, allocator, width, bitsPerPixel, false));
case TiffCompression.CcittGroup4Fax:
DebugGuard.IsTrue(compressionLevel == DeflateCompressionLevel.DefaultCompression, "No deflate compression level is expected to be set");
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
return(new T6BitCompressor(output, allocator, width, bitsPerPixel));
case TiffCompression.Ccitt1D:
DebugGuard.IsTrue(compressionLevel == DeflateCompressionLevel.DefaultCompression, "No deflate compression level is expected to be set");
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
return(new T4BitCompressor(output, allocator, width, bitsPerPixel, true));
default:
throw TiffThrowHelper.NotSupportedCompressor(method.ToString());
}
}
public void IsTrue_IsFalse_ThrowsException()
{
var exception = Assert.Throws <ArgumentException>(() =>
{
DebugGuard.IsTrue(false, "myParamName", "myTestMessage");
});
Assert.Equal("myParamName", exception.ParamName);
Assert.True(exception.Message.Contains("myTestMessage"));
}
/// <summary>
/// Stores code in table[0], table[step], table[2*step], ..., table[end-step].
/// Assumes that end is an integer multiple of step.
/// </summary>
private static void ReplicateValue(Span <HuffmanCode> table, int step, int end, HuffmanCode code)
{
DebugGuard.IsTrue(end % step == 0, nameof(end), "end must be a multiple of step");
do
{
end -= step;
table[end] = code;
}while (end > 0);
}
#pragma warning restore SA1310, SA1311, IDE1006
/// <summary>
/// Apply floating point FDCT inplace using simd operations.
/// </summary>
/// <param name="block">Input block.</param>
private static void FDCT8x8_Avx(ref Block8x8F block)
{
DebugGuard.IsTrue(Avx.IsSupported, "Avx support is required to execute this operation.");
// First pass - process columns
FDCT8x8_1D_Avx(ref block);
// Second pass - process rows
block.TransposeInplace();
FDCT8x8_1D_Avx(ref block);
public static TiffBaseDecompressor Create(
Configuration configuration,
TiffDecoderCompressionType method,
MemoryAllocator allocator,
TiffPhotometricInterpretation photometricInterpretation,
int width,
int bitsPerPixel,
TiffColorType colorType,
TiffPredictor predictor,
FaxCompressionOptions faxOptions,
byte[] jpegTables,
TiffFillOrder fillOrder,
ByteOrder byteOrder)
{
switch (method)
{
case TiffDecoderCompressionType.None:
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
DebugGuard.IsTrue(faxOptions == FaxCompressionOptions.None, "No fax compression options are expected");
return(new NoneTiffCompression(allocator, width, bitsPerPixel));
case TiffDecoderCompressionType.PackBits:
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
DebugGuard.IsTrue(faxOptions == FaxCompressionOptions.None, "No fax compression options are expected");
return(new PackBitsTiffCompression(allocator, width, bitsPerPixel));
case TiffDecoderCompressionType.Deflate:
DebugGuard.IsTrue(faxOptions == FaxCompressionOptions.None, "No fax compression options are expected");
return(new DeflateTiffCompression(allocator, width, bitsPerPixel, colorType, predictor, byteOrder == ByteOrder.BigEndian));
case TiffDecoderCompressionType.Lzw:
DebugGuard.IsTrue(faxOptions == FaxCompressionOptions.None, "No fax compression options are expected");
return(new LzwTiffCompression(allocator, width, bitsPerPixel, colorType, predictor, byteOrder == ByteOrder.BigEndian));
case TiffDecoderCompressionType.T4:
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
return(new T4TiffCompression(allocator, fillOrder, width, bitsPerPixel, faxOptions, photometricInterpretation));
case TiffDecoderCompressionType.T6:
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
return(new T6TiffCompression(allocator, fillOrder, width, bitsPerPixel, photometricInterpretation));
case TiffDecoderCompressionType.HuffmanRle:
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
return(new ModifiedHuffmanTiffCompression(allocator, fillOrder, width, bitsPerPixel, photometricInterpretation));
case TiffDecoderCompressionType.Jpeg:
DebugGuard.IsTrue(predictor == TiffPredictor.None, "Predictor should only be used with lzw or deflate compression");
return(new JpegTiffCompression(configuration, allocator, width, bitsPerPixel, jpegTables, photometricInterpretation));
default:
throw TiffThrowHelper.NotSupportedDecompressor(nameof(method));
}
}
/// <inheritdoc />
protected override void Dispose(bool disposing)
{
DebugGuard.IsTrue(disposing, nameof(disposing), "Unmanaged buffers should not have finalizer!");
if (Interlocked.Exchange(ref this.disposed, 1) == 1)
{
// Already disposed
return;
}
this.lifetimeGuard.Dispose();
}
private static void ApplyHorizontalPrediction8Bit(Span <byte> rows, int width)
{
DebugGuard.IsTrue(rows.Length % width == 0, "Values must be equals");
int height = rows.Length / width;
for (int y = 0; y < height; y++)
{
Span <byte> rowSpan = rows.Slice(y * width, width);
for (int x = rowSpan.Length - 1; x >= 1; x--)
{
rowSpan[x] -= rowSpan[x - 1];
}
}
}
/// <inheritdoc/>
public override void CompressStrip(Span <byte> rows, int height)
{
DebugGuard.IsTrue(rows.Length % height == 0, "Invalid height");
DebugGuard.IsTrue(this.BytesPerRow == rows.Length / height, "The widths must match");
Span <byte> span = this.pixelData.GetSpan();
for (int i = 0; i < height; i++)
{
Span <byte> row = rows.Slice(i * this.BytesPerRow, this.BytesPerRow);
int size = PackBitsWriter.PackBits(row, span);
this.Output.Write(span.Slice(0, size));
}
}
/// <summary>
/// Apply 1D floating point FDCT inplace using AVX operations on 8x8 matrix.
/// </summary>
/// <remarks>
/// Requires Avx support.
/// </remarks>
/// <param name="block">Input matrix.</param>
public static void FDCT8x8_Avx(ref Block8x8F block)
{
DebugGuard.IsTrue(Avx.IsSupported, "Avx support is required to execute this operation.");
Vector256 <float> tmp0 = Avx.Add(block.V0, block.V7);
Vector256 <float> tmp7 = Avx.Subtract(block.V0, block.V7);
Vector256 <float> tmp1 = Avx.Add(block.V1, block.V6);
Vector256 <float> tmp6 = Avx.Subtract(block.V1, block.V6);
Vector256 <float> tmp2 = Avx.Add(block.V2, block.V5);
Vector256 <float> tmp5 = Avx.Subtract(block.V2, block.V5);
Vector256 <float> tmp3 = Avx.Add(block.V3, block.V4);
Vector256 <float> tmp4 = Avx.Subtract(block.V3, block.V4);
// Even part
Vector256 <float> tmp10 = Avx.Add(tmp0, tmp3);
Vector256 <float> tmp13 = Avx.Subtract(tmp0, tmp3);
Vector256 <float> tmp11 = Avx.Add(tmp1, tmp2);
Vector256 <float> tmp12 = Avx.Subtract(tmp1, tmp2);
block.V0 = Avx.Add(tmp10, tmp11);
block.V4 = Avx.Subtract(tmp10, tmp11);
Vector256 <float> z1 = Avx.Multiply(Avx.Add(tmp12, tmp13), mm256_F_0_7071);
block.V2 = Avx.Add(tmp13, z1);
block.V6 = Avx.Subtract(tmp13, z1);
// Odd part
tmp10 = Avx.Add(tmp4, tmp5);
tmp11 = Avx.Add(tmp5, tmp6);
tmp12 = Avx.Add(tmp6, tmp7);
Vector256 <float> z5 = Avx.Multiply(Avx.Subtract(tmp10, tmp12), mm256_F_0_3826);
Vector256 <float> z2 = SimdUtils.HwIntrinsics.MultiplyAdd(z5, mm256_F_0_5411, tmp10);
Vector256 <float> z4 = SimdUtils.HwIntrinsics.MultiplyAdd(z5, mm256_F_1_3065, tmp12);
Vector256 <float> z3 = Avx.Multiply(tmp11, mm256_F_0_7071);
Vector256 <float> z11 = Avx.Add(tmp7, z3);
Vector256 <float> z13 = Avx.Subtract(tmp7, z3);
block.V5 = Avx.Add(z13, z2);
block.V3 = Avx.Subtract(z13, z2);
block.V1 = Avx.Add(z11, z4);
block.V7 = Avx.Subtract(z11, z4);
}
/// <summary>
/// Writes a image compressed with CCITT T6 to the stream.
/// </summary>
/// <param name="pixelsAsGray">The pixels as 8-bit gray array.</param>
/// <param name="height">The strip height.</param>
public override void CompressStrip(Span <byte> pixelsAsGray, int height)
{
DebugGuard.IsTrue(pixelsAsGray.Length / height == this.Width, "Values must be equals");
DebugGuard.IsTrue(pixelsAsGray.Length % height == 0, "Values must be equals");
this.compressedDataBuffer.Clear();
Span <byte> compressedData = this.compressedDataBuffer.GetSpan();
this.bytePosition = 0;
this.bitPosition = 0;
this.CompressStrip(pixelsAsGray, height, compressedData);
// Write the compressed data to the stream.
int bytesToWrite = this.bitPosition != 0 ? this.bytePosition + 1 : this.bytePosition;
this.Output.Write(compressedData.Slice(0, bytesToWrite));
}
private static void ApplyHorizontalPrediction24Bit(Span <byte> rows, int width)
{
DebugGuard.IsTrue(rows.Length % width == 0, "Values must be equals");
int height = rows.Length / width;
for (int y = 0; y < height; y++)
{
Span <byte> rowSpan = rows.Slice(y * width, width);
Span <Rgb24> rowRgb = MemoryMarshal.Cast <byte, Rgb24>(rowSpan);
for (int x = rowRgb.Length - 1; x >= 1; x--)
{
byte r = (byte)(rowRgb[x].R - rowRgb[x - 1].R);
byte g = (byte)(rowRgb[x].G - rowRgb[x - 1].G);
byte b = (byte)(rowRgb[x].B - rowRgb[x - 1].B);
var rgb = new Rgb24(r, g, b);
rowRgb[x].FromRgb24(rgb);
}
}
}
private IMemoryOwner <ulong> ConvertNumbers(Array array, out Span <ulong> span)
{
if (array is Number[] numbers)
{
IMemoryOwner <ulong> memory = this.memoryAllocator.Allocate <ulong>(numbers.Length);
span = memory.GetSpan();
for (int i = 0; i < numbers.Length; i++)
{
span[i] = (uint)numbers[i];
}
return(memory);
}
else
{
DebugGuard.IsTrue(array is ulong[], $"Expected {nameof(UInt64)} array.");
span = (ulong[])array;
return(null);
}
}
/// <summary>
/// SIMD optimized bulk implementation of <see cref="IPixel.PackFromVector4(Vector4)"/>
/// that works only with `count` divisible by <see cref="Vector{UInt32}.Count"/>.
/// </summary>
/// <param name="sourceColors">The <see cref="Span{T}"/> to the source colors.</param>
/// <param name="destVectors">The <see cref="Span{T}"/> to the dstination vectors.</param>
/// <param name="count">The number of pixels to convert.</param>
/// <remarks>
/// Implementation adapted from:
/// <see>
/// <cref>http://stackoverflow.com/a/5362789</cref>
/// </see>
/// TODO: We can replace this implementation in the future using new Vector API-s:
/// <see>
/// <cref>https://github.com/dotnet/corefx/issues/15957</cref>
/// </see>
/// </remarks>
internal static void ToVector4SimdAligned(ReadOnlySpan <Rgba32> sourceColors, Span <Vector4> destVectors, int count)
{
if (!Vector.IsHardwareAccelerated)
{
throw new InvalidOperationException(
"Rgba32.PixelOperations.ToVector4SimdAligned() should not be called when Vector.IsHardwareAccelerated == false!");
}
DebugGuard.IsTrue(
count % Vector <uint> .Count == 0,
nameof(count),
"Argument 'count' should divisible by Vector<uint>.Count!");
var bVec = new Vector <float>(256.0f / 255.0f);
var magicFloat = new Vector <float>(32768.0f);
var magicInt = new Vector <uint>(1191182336); // reinterpreded value of 32768.0f
var mask = new Vector <uint>(255);
int unpackedRawCount = count * 4;
ref uint sourceBase = ref Unsafe.As <Rgba32, uint>(ref MemoryMarshal.GetReference(sourceColors));
/// <summary>
/// SIMD optimized bulk implementation of <see cref="IPixel.PackFromVector4(Vector4)"/>
/// that works only with `count` divisible by <see cref="Vector{UInt32}.Count"/>.
/// </summary>
/// <param name="sourceColors">The <see cref="BufferSpan{T}"/> to the source colors.</param>
/// <param name="destVectors">The <see cref="BufferSpan{T}"/> to the dstination vectors.</param>
/// <param name="count">The number of pixels to convert.</param>
/// <remarks>
/// Implementation adapted from:
/// <see>
/// <cref>http://stackoverflow.com/a/5362789</cref>
/// </see>
/// TODO: We can replace this implementation in the future using new Vector API-s:
/// <see>
/// <cref>https://github.com/dotnet/corefx/issues/15957</cref>
/// </see>
/// </remarks>
internal static unsafe void ToVector4SimdAligned(BufferSpan <Rgba32> sourceColors, BufferSpan <Vector4> destVectors, int count)
{
if (!Vector.IsHardwareAccelerated)
{
throw new InvalidOperationException(
"Rgba32.BulkOperations.ToVector4SimdAligned() should not be called when Vector.IsHardwareAccelerated == false!");
}
int vecSize = Vector <uint> .Count;
DebugGuard.IsTrue(
count % vecSize == 0,
nameof(count),
"Argument 'count' should divisible by Vector<uint>.Count!");
Vector <float> bVec = new Vector <float>(256.0f / 255.0f);
Vector <float> magicFloat = new Vector <float>(32768.0f);
Vector <uint> magicInt = new Vector <uint>(1191182336); // reinterpreded value of 32768.0f
Vector <uint> mask = new Vector <uint>(255);
int unpackedRawCount = count * 4;
ref uint src = ref Unsafe.As <Rgba32, uint>(ref sourceColors.DangerousGetPinnableReference());
/// <summary>
/// Calculates the size (in bytes) for a pixel buffer using the determined color format.
/// </summary>
/// <param name="width">The width for the desired pixel buffer.</param>
/// <param name="height">The height for the desired pixel buffer.</param>
/// <param name="plane">The index of the plane for planar image configuration (or zero for chunky).</param>
/// <returns>The size (in bytes) of the required pixel buffer.</returns>
private int CalculateStripBufferSize(int width, int height, int plane = -1)
{
DebugGuard.MustBeLessThanOrEqualTo(plane, 3, nameof(plane));
int bitsPerPixel = 0;
if (this.PlanarConfiguration == TiffPlanarConfiguration.Chunky)
{
DebugGuard.IsTrue(plane == -1, "Expected Chunky planar.");
bitsPerPixel = this.BitsPerPixel;
}
else
{
switch (plane)
{
case 0:
bitsPerPixel = this.BitsPerSample.Channel0;
break;
case 1:
bitsPerPixel = this.BitsPerSample.Channel1;
break;
case 2:
bitsPerPixel = this.BitsPerSample.Channel2;
break;
default:
TiffThrowHelper.ThrowNotSupported("More then 3 color channels are not supported");
break;
}
}
int bytesPerRow = ((width * bitsPerPixel) + 7) / 8;
return(bytesPerRow * height);
}
public void IsTrue_IsTrue_ThrowsNoException()
{
DebugGuard.IsTrue(true, "myParamName", "myTestMessage");
}
/// <summary>
/// Writes a image compressed with CCITT T4 to the stream.
/// </summary>
/// <param name="pixelsAsGray">The pixels as 8-bit gray array.</param>
/// <param name="height">The strip height.</param>
public override void CompressStrip(Span <byte> pixelsAsGray, int height)
{
DebugGuard.IsTrue(pixelsAsGray.Length / height == this.Width, "Values must be equals");
DebugGuard.IsTrue(pixelsAsGray.Length % height == 0, "Values must be equals");
this.compressedDataBuffer.Clear();
Span <byte> compressedData = this.compressedDataBuffer.GetSpan();
this.bytePosition = 0;
this.bitPosition = 0;
if (!this.useModifiedHuffman)
{
// An EOL code is expected at the start of the data.
this.WriteCode(12, 1, compressedData);
}
for (int y = 0; y < height; y++)
{
bool isWhiteRun = true;
bool isStartOrRow = true;
int x = 0;
Span <byte> row = pixelsAsGray.Slice(y * this.Width, this.Width);
while (x < this.Width)
{
uint runLength = 0;
for (int i = x; i < this.Width; i++)
{
if (isWhiteRun && row[i] != 255)
{
break;
}
if (isWhiteRun && row[i] == 255)
{
runLength++;
continue;
}
if (!isWhiteRun && row[i] != 0)
{
break;
}
if (!isWhiteRun && row[i] == 0)
{
runLength++;
}
}
if (isStartOrRow && runLength == 0)
{
this.WriteCode(8, WhiteZeroRunTermCode, compressedData);
isWhiteRun = false;
isStartOrRow = false;
continue;
}
uint code;
uint codeLength;
if (runLength <= 63)
{
code = this.GetTermCode(runLength, out codeLength, isWhiteRun);
this.WriteCode(codeLength, code, compressedData);
x += (int)runLength;
}
else
{
runLength = this.GetBestFittingMakeupRunLength(runLength);
code = this.GetMakeupCode(runLength, out codeLength, isWhiteRun);
this.WriteCode(codeLength, code, compressedData);
x += (int)runLength;
// If we are at the end of the line with a makeup code, we need to write a final term code with a length of zero.
if (x == this.Width)
{
if (isWhiteRun)
{
this.WriteCode(8, WhiteZeroRunTermCode, compressedData);
}
else
{
this.WriteCode(10, BlackZeroRunTermCode, compressedData);
}
}
continue;
}
isStartOrRow = false;
isWhiteRun = !isWhiteRun;
}
this.WriteEndOfLine(compressedData);
}
// Write the compressed data to the stream.
int bytesToWrite = this.bitPosition != 0 ? this.bytePosition + 1 : this.bytePosition;
this.Output.Write(compressedData.Slice(0, bytesToWrite));
}
/// <summary>
/// Applies zig zag ordering for given 8x8 matrix using SSE cpu intrinsics.
/// </summary>
/// <param name="block">Input matrix.</param>
public static unsafe void ApplyZigZagOrderingSsse3(ref Block8x8 block)
{
DebugGuard.IsTrue(Ssse3.IsSupported, "Ssse3 support is required to run this operation!");
fixed(byte *maskPtr = SseShuffleMasks)
{
Vector128 <byte> rowA = block.V0.AsByte();
Vector128 <byte> rowB = block.V1.AsByte();
Vector128 <byte> rowC = block.V2.AsByte();
Vector128 <byte> rowD = block.V3.AsByte();
Vector128 <byte> rowE = block.V4.AsByte();
Vector128 <byte> rowF = block.V5.AsByte();
Vector128 <byte> rowG = block.V6.AsByte();
Vector128 <byte> rowH = block.V7.AsByte();
// row0 - A0 A1 B0 C0 B1 A2 A3 B2
Vector128 <short> rowA0 = Ssse3.Shuffle(rowA, Sse2.LoadVector128(maskPtr + (16 * 0))).AsInt16();
Vector128 <short> rowB0 = Ssse3.Shuffle(rowB, Sse2.LoadVector128(maskPtr + (16 * 1))).AsInt16();
Vector128 <short> row0 = Sse2.Or(rowA0, rowB0);
Vector128 <short> rowC0 = Ssse3.Shuffle(rowC, Sse2.LoadVector128(maskPtr + (16 * 2))).AsInt16();
row0 = Sse2.Or(row0, rowC0);
// row1 - C1 D0 E0 D1 C2 B3 A4 A5
Vector128 <short> rowA1 = Ssse3.Shuffle(rowA, Sse2.LoadVector128(maskPtr + (16 * 3))).AsInt16();
Vector128 <short> rowC1 = Ssse3.Shuffle(rowC, Sse2.LoadVector128(maskPtr + (16 * 4))).AsInt16();
Vector128 <short> row1 = Sse2.Or(rowA1, rowC1);
Vector128 <short> rowD1 = Ssse3.Shuffle(rowD, Sse2.LoadVector128(maskPtr + (16 * 5))).AsInt16();
row1 = Sse2.Or(row1, rowD1);
row1 = Sse2.Insert(row1.AsUInt16(), Sse2.Extract(rowB.AsUInt16(), 3), 5).AsInt16();
row1 = Sse2.Insert(row1.AsUInt16(), Sse2.Extract(rowE.AsUInt16(), 0), 2).AsInt16();
// row2
Vector128 <short> rowE2 = Ssse3.Shuffle(rowE, Sse2.LoadVector128(maskPtr + (16 * 6))).AsInt16();
Vector128 <short> rowF2 = Ssse3.Shuffle(rowF, Sse2.LoadVector128(maskPtr + (16 * 7))).AsInt16();
Vector128 <short> row2 = Sse2.Or(rowE2, rowF2);
row2 = Sse2.Insert(row2.AsUInt16(), Sse2.Extract(rowB.AsUInt16(), 4), 0).AsInt16();
row2 = Sse2.Insert(row2.AsUInt16(), Sse2.Extract(rowC.AsUInt16(), 3), 1).AsInt16();
row2 = Sse2.Insert(row2.AsUInt16(), Sse2.Extract(rowD.AsUInt16(), 2), 2).AsInt16();
row2 = Sse2.Insert(row2.AsUInt16(), Sse2.Extract(rowG.AsUInt16(), 0), 5).AsInt16();
// row3
Vector128 <short> rowA3 = Ssse3.Shuffle(rowA, Sse2.LoadVector128(maskPtr + (16 * 8))).AsInt16().AsInt16();
Vector128 <short> rowB3 = Ssse3.Shuffle(rowB, Sse2.LoadVector128(maskPtr + (16 * 9))).AsInt16().AsInt16();
Vector128 <short> row3 = Sse2.Or(rowA3, rowB3);
Vector128 <short> rowC3 = Ssse3.Shuffle(rowC, Sse2.LoadVector128(maskPtr + (16 * 10))).AsInt16();
row3 = Sse2.Or(row3, rowC3);
Vector128 <byte> shuffleRowD3EF = Sse2.LoadVector128(maskPtr + (16 * 11));
Vector128 <short> rowD3 = Ssse3.Shuffle(rowD, shuffleRowD3EF).AsInt16();
row3 = Sse2.Or(row3, rowD3);
// row4
Vector128 <short> rowE4 = Ssse3.Shuffle(rowE, shuffleRowD3EF).AsInt16();
Vector128 <short> rowF4 = Ssse3.Shuffle(rowF, Sse2.LoadVector128(maskPtr + (16 * 12))).AsInt16();
Vector128 <short> row4 = Sse2.Or(rowE4, rowF4);
Vector128 <short> rowG4 = Ssse3.Shuffle(rowG, Sse2.LoadVector128(maskPtr + (16 * 13))).AsInt16();
row4 = Sse2.Or(row4, rowG4);
Vector128 <short> rowH4 = Ssse3.Shuffle(rowH, Sse2.LoadVector128(maskPtr + (16 * 14))).AsInt16();
row4 = Sse2.Or(row4, rowH4);
// row5
Vector128 <short> rowC5 = Ssse3.Shuffle(rowC, Sse2.LoadVector128(maskPtr + (16 * 15))).AsInt16();
Vector128 <short> rowD5 = Ssse3.Shuffle(rowD, Sse2.LoadVector128(maskPtr + (16 * 16))).AsInt16();
Vector128 <short> row5 = Sse2.Or(rowC5, rowD5);
row5 = Sse2.Insert(row5.AsUInt16(), Sse2.Extract(rowB.AsUInt16(), 7), 2).AsInt16();
row5 = Sse2.Insert(row5.AsUInt16(), Sse2.Extract(rowE.AsUInt16(), 5), 5).AsInt16();
row5 = Sse2.Insert(row5.AsUInt16(), Sse2.Extract(rowF.AsUInt16(), 4), 6).AsInt16();
row5 = Sse2.Insert(row5.AsUInt16(), Sse2.Extract(rowG.AsUInt16(), 3), 7).AsInt16();
// row6
Vector128 <short> rowE6 = Ssse3.Shuffle(rowE, Sse2.LoadVector128(maskPtr + (16 * 17))).AsInt16();
Vector128 <short> rowF6 = Ssse3.Shuffle(rowF, Sse2.LoadVector128(maskPtr + (16 * 18))).AsInt16();
Vector128 <short> row6 = Sse2.Or(rowE6, rowF6);
Vector128 <short> rowH6 = Ssse3.Shuffle(rowH, Sse2.LoadVector128(maskPtr + (16 * 19))).AsInt16();
row6 = Sse2.Or(row6, rowH6);
row6 = Sse2.Insert(row6.AsUInt16(), Sse2.Extract(rowD.AsUInt16(), 7), 5).AsInt16();
row6 = Sse2.Insert(row6.AsUInt16(), Sse2.Extract(rowG.AsUInt16(), 4), 2).AsInt16();
// row7
Vector128 <short> rowG7 = Ssse3.Shuffle(rowG, Sse2.LoadVector128(maskPtr + (16 * 20))).AsInt16();
Vector128 <short> rowH7 = Ssse3.Shuffle(rowH, Sse2.LoadVector128(maskPtr + (16 * 21))).AsInt16();
Vector128 <short> row7 = Sse2.Or(rowG7, rowH7);
row7 = Sse2.Insert(row7.AsUInt16(), Sse2.Extract(rowF.AsUInt16(), 7), 4).AsInt16();
block.V0 = row0;
block.V1 = row1;
block.V2 = row2;
block.V3 = row3;
block.V4 = row4;
block.V5 = row5;
block.V6 = row6;
block.V7 = row7;
}
}
private static unsafe void MultiplyIntoInt16_Avx2(ref Block8x8F a, ref Block8x8F b, ref Block8x8 dest)
{
DebugGuard.IsTrue(Avx2.IsSupported, "Avx2 support is required to run this operation!");
ref Vector256 <float> aBase = ref a.V0;
/// <summary>
/// Applies zig zag ordering for given 8x8 matrix using AVX cpu intrinsics.
/// </summary>
/// <param name="block">Input matrix.</param>
public static unsafe void ApplyZigZagOrderingAvx2(ref Block8x8 block)
{
DebugGuard.IsTrue(Avx2.IsSupported, "Avx2 support is required to run this operation!");
fixed(byte *shuffleVectorsPtr = AvxShuffleMasks)
{
Vector256 <byte> rowsAB = block.V01.AsByte();
Vector256 <byte> rowsCD = block.V23.AsByte();
Vector256 <byte> rowsEF = block.V45.AsByte();
Vector256 <byte> rowsGH = block.V67.AsByte();
// rows 0 1
Vector256 <int> rows_AB01_EF01_CD23_shuffleMask = Avx.LoadVector256(shuffleVectorsPtr + (0 * 32)).AsInt32();
Vector256 <byte> row01_AB = Avx2.PermuteVar8x32(rowsAB.AsInt32(), rows_AB01_EF01_CD23_shuffleMask).AsByte();
row01_AB = Avx2.Shuffle(row01_AB, Avx.LoadVector256(shuffleVectorsPtr + (1 * 32))).AsByte();
Vector256 <int> rows_CD01_GH23_shuffleMask = Avx.LoadVector256(shuffleVectorsPtr + (2 * 32)).AsInt32();
Vector256 <byte> row01_CD = Avx2.PermuteVar8x32(rowsCD.AsInt32(), rows_CD01_GH23_shuffleMask).AsByte();
row01_CD = Avx2.Shuffle(row01_CD, Avx.LoadVector256(shuffleVectorsPtr + (3 * 32))).AsByte();
Vector256 <byte> row0123_EF = Avx2.PermuteVar8x32(rowsEF.AsInt32(), rows_AB01_EF01_CD23_shuffleMask).AsByte();
Vector256 <byte> row01_EF = Avx2.Shuffle(row0123_EF, Avx.LoadVector256(shuffleVectorsPtr + (4 * 32))).AsByte();
Vector256 <byte> row01 = Avx2.Or(Avx2.Or(row01_AB, row01_CD), row01_EF);
// rows 2 3
Vector256 <int> rows_AB23_CD45_EF67_shuffleMask = Avx.LoadVector256(shuffleVectorsPtr + (5 * 32)).AsInt32();
Vector256 <byte> row2345_AB = Avx2.PermuteVar8x32(rowsAB.AsInt32(), rows_AB23_CD45_EF67_shuffleMask).AsByte();
Vector256 <byte> row23_AB = Avx2.Shuffle(row2345_AB, Avx.LoadVector256(shuffleVectorsPtr + (6 * 32))).AsByte();
Vector256 <byte> row23_CD = Avx2.PermuteVar8x32(rowsCD.AsInt32(), rows_AB01_EF01_CD23_shuffleMask).AsByte();
row23_CD = Avx2.Shuffle(row23_CD, Avx.LoadVector256(shuffleVectorsPtr + (7 * 32))).AsByte();
Vector256 <byte> row23_EF = Avx2.Shuffle(row0123_EF, Avx.LoadVector256(shuffleVectorsPtr + (8 * 32))).AsByte();
Vector256 <byte> row2345_GH = Avx2.PermuteVar8x32(rowsGH.AsInt32(), rows_CD01_GH23_shuffleMask).AsByte();
Vector256 <byte> row23_GH = Avx2.Shuffle(row2345_GH, Avx.LoadVector256(shuffleVectorsPtr + (9 * 32)).AsByte());
Vector256 <byte> row23 = Avx2.Or(Avx2.Or(row23_AB, row23_CD), Avx2.Or(row23_EF, row23_GH));
// rows 4 5
Vector256 <byte> row45_AB = Avx2.Shuffle(row2345_AB, Avx.LoadVector256(shuffleVectorsPtr + (10 * 32)).AsByte());
Vector256 <byte> row4567_CD = Avx2.PermuteVar8x32(rowsCD.AsInt32(), rows_AB23_CD45_EF67_shuffleMask).AsByte();
Vector256 <byte> row45_CD = Avx2.Shuffle(row4567_CD, Avx.LoadVector256(shuffleVectorsPtr + (11 * 32)).AsByte());
Vector256 <int> rows_EF45_GH67_shuffleMask = Avx.LoadVector256(shuffleVectorsPtr + (12 * 32)).AsInt32();
Vector256 <byte> row45_EF = Avx2.PermuteVar8x32(rowsEF.AsInt32(), rows_EF45_GH67_shuffleMask).AsByte();
row45_EF = Avx2.Shuffle(row45_EF, Avx.LoadVector256(shuffleVectorsPtr + (13 * 32)).AsByte());
Vector256 <byte> row45_GH = Avx2.Shuffle(row2345_GH, Avx.LoadVector256(shuffleVectorsPtr + (14 * 32)).AsByte());
Vector256 <byte> row45 = Avx2.Or(Avx2.Or(row45_AB, row45_CD), Avx2.Or(row45_EF, row45_GH));
// rows 6 7
Vector256 <byte> row67_CD = Avx2.Shuffle(row4567_CD, Avx.LoadVector256(shuffleVectorsPtr + (15 * 32)).AsByte());
Vector256 <byte> row67_EF = Avx2.PermuteVar8x32(rowsEF.AsInt32(), rows_AB23_CD45_EF67_shuffleMask).AsByte();
row67_EF = Avx2.Shuffle(row67_EF, Avx.LoadVector256(shuffleVectorsPtr + (16 * 32)).AsByte());
Vector256 <byte> row67_GH = Avx2.PermuteVar8x32(rowsGH.AsInt32(), rows_EF45_GH67_shuffleMask).AsByte();
row67_GH = Avx2.Shuffle(row67_GH, Avx.LoadVector256(shuffleVectorsPtr + (17 * 32)).AsByte());
Vector256 <byte> row67 = Avx2.Or(Avx2.Or(row67_CD, row67_EF), row67_GH);
block.V01 = row01.AsInt16();
block.V23 = row23.AsInt16();
block.V45 = row45.AsInt16();
block.V67 = row67.AsInt16();
}
}
/// <summary>
/// SIMD convert using buffers of sizes divisable by 8.
/// </summary>
internal static void ConvertCore(ComponentValues values, Span <Vector4> result)
{
DebugGuard.IsTrue(result.Length % 8 == 0, nameof(result), "result.Length should be divisable by 8!");
ref Vector4Pair yBase =
public static TiffBaseColorDecoder <TPixel> Create(
Configuration configuration,
MemoryAllocator memoryAllocator,
TiffColorType colorType,
TiffBitsPerSample bitsPerSample,
ushort[] colorMap,
Rational[] referenceBlackAndWhite,
Rational[] ycbcrCoefficients,
ushort[] ycbcrSubSampling,
ByteOrder byteOrder)
{
switch (colorType)
{
case TiffColorType.WhiteIsZero:
DebugGuard.IsTrue(bitsPerSample.Channels == 1, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new WhiteIsZeroTiffColor <TPixel>(bitsPerSample));
case TiffColorType.WhiteIsZero1:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 1, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new WhiteIsZero1TiffColor <TPixel>());
case TiffColorType.WhiteIsZero4:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 4, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new WhiteIsZero4TiffColor <TPixel>());
case TiffColorType.WhiteIsZero8:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 8, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new WhiteIsZero8TiffColor <TPixel>());
case TiffColorType.WhiteIsZero16:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 16, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new WhiteIsZero16TiffColor <TPixel>(byteOrder == ByteOrder.BigEndian));
case TiffColorType.WhiteIsZero24:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 24, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new WhiteIsZero24TiffColor <TPixel>(byteOrder == ByteOrder.BigEndian));
case TiffColorType.WhiteIsZero32:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 32, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new WhiteIsZero32TiffColor <TPixel>(byteOrder == ByteOrder.BigEndian));
case TiffColorType.WhiteIsZero32Float:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 32, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new WhiteIsZero32FloatTiffColor <TPixel>(byteOrder == ByteOrder.BigEndian));
case TiffColorType.BlackIsZero:
DebugGuard.IsTrue(bitsPerSample.Channels == 1, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new BlackIsZeroTiffColor <TPixel>(bitsPerSample));
case TiffColorType.BlackIsZero1:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 1, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new BlackIsZero1TiffColor <TPixel>());
case TiffColorType.BlackIsZero4:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 4, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new BlackIsZero4TiffColor <TPixel>());
case TiffColorType.BlackIsZero8:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 8, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new BlackIsZero8TiffColor <TPixel>(configuration));
case TiffColorType.BlackIsZero16:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 16, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new BlackIsZero16TiffColor <TPixel>(configuration, byteOrder == ByteOrder.BigEndian));
case TiffColorType.BlackIsZero24:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 24, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new BlackIsZero24TiffColor <TPixel>(byteOrder == ByteOrder.BigEndian));
case TiffColorType.BlackIsZero32:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 32, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new BlackIsZero32TiffColor <TPixel>(byteOrder == ByteOrder.BigEndian));
case TiffColorType.BlackIsZero32Float:
DebugGuard.IsTrue(bitsPerSample.Channels == 1 && bitsPerSample.Channel0 == 32, "bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new BlackIsZero32FloatTiffColor <TPixel>(byteOrder == ByteOrder.BigEndian));
case TiffColorType.Rgb:
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new RgbTiffColor <TPixel>(bitsPerSample));
case TiffColorType.Rgb222:
DebugGuard.IsTrue(
bitsPerSample.Channels == 3 &&
bitsPerSample.Channel2 == 2 &&
bitsPerSample.Channel1 == 2 &&
bitsPerSample.Channel0 == 2,
"bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new RgbTiffColor <TPixel>(bitsPerSample));
case TiffColorType.Rgb444:
DebugGuard.IsTrue(
bitsPerSample.Channels == 3 &&
bitsPerSample.Channel2 == 4 &&
bitsPerSample.Channel1 == 4 &&
bitsPerSample.Channel0 == 4,
"bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new Rgb444TiffColor <TPixel>());
case TiffColorType.Rgb888:
DebugGuard.IsTrue(
bitsPerSample.Channels == 3 &&
bitsPerSample.Channel2 == 8 &&
bitsPerSample.Channel1 == 8 &&
bitsPerSample.Channel0 == 8,
"bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new Rgb888TiffColor <TPixel>(configuration));
case TiffColorType.Rgb101010:
DebugGuard.IsTrue(
bitsPerSample.Channels == 3 &&
bitsPerSample.Channel2 == 10 &&
bitsPerSample.Channel1 == 10 &&
bitsPerSample.Channel0 == 10,
"bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new RgbTiffColor <TPixel>(bitsPerSample));
case TiffColorType.Rgb121212:
DebugGuard.IsTrue(
bitsPerSample.Channels == 3 &&
bitsPerSample.Channel2 == 12 &&
bitsPerSample.Channel1 == 12 &&
bitsPerSample.Channel0 == 12,
"bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new RgbTiffColor <TPixel>(bitsPerSample));
case TiffColorType.Rgb141414:
DebugGuard.IsTrue(
bitsPerSample.Channels == 3 &&
bitsPerSample.Channel2 == 14 &&
bitsPerSample.Channel1 == 14 &&
bitsPerSample.Channel0 == 14,
"bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new RgbTiffColor <TPixel>(bitsPerSample));
case TiffColorType.Rgb161616:
DebugGuard.IsTrue(
bitsPerSample.Channels == 3 &&
bitsPerSample.Channel2 == 16 &&
bitsPerSample.Channel1 == 16 &&
bitsPerSample.Channel0 == 16,
"bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new Rgb161616TiffColor <TPixel>(configuration, isBigEndian: byteOrder == ByteOrder.BigEndian));
case TiffColorType.Rgb242424:
DebugGuard.IsTrue(
bitsPerSample.Channels == 3 &&
bitsPerSample.Channel2 == 24 &&
bitsPerSample.Channel1 == 24 &&
bitsPerSample.Channel0 == 24,
"bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new Rgb242424TiffColor <TPixel>(isBigEndian: byteOrder == ByteOrder.BigEndian));
case TiffColorType.Rgb323232:
DebugGuard.IsTrue(
bitsPerSample.Channels == 3 &&
bitsPerSample.Channel2 == 32 &&
bitsPerSample.Channel1 == 32 &&
bitsPerSample.Channel0 == 32,
"bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new Rgb323232TiffColor <TPixel>(isBigEndian: byteOrder == ByteOrder.BigEndian));
case TiffColorType.RgbFloat323232:
DebugGuard.IsTrue(
bitsPerSample.Channels == 3 &&
bitsPerSample.Channel2 == 32 &&
bitsPerSample.Channel1 == 32 &&
bitsPerSample.Channel0 == 32,
"bitsPerSample");
DebugGuard.IsTrue(colorMap == null, "colorMap");
return(new RgbFloat323232TiffColor <TPixel>(isBigEndian: byteOrder == ByteOrder.BigEndian));
case TiffColorType.PaletteColor:
DebugGuard.NotNull(colorMap, "colorMap");
return(new PaletteTiffColor <TPixel>(bitsPerSample, colorMap));
case TiffColorType.YCbCr:
return(new YCbCrTiffColor <TPixel>(memoryAllocator, referenceBlackAndWhite, ycbcrCoefficients, ycbcrSubSampling));
default:
throw TiffThrowHelper.InvalidColorType(colorType.ToString());
}
}
/// <summary>
/// Applies zig zag ordering for given 8x8 matrix using SSE cpu intrinsics.
/// </summary>
/// <param name="block">Input matrix.</param>
public static unsafe void ApplyTransposingZigZagOrderingSsse3(ref Block8x8 block)
{
DebugGuard.IsTrue(Ssse3.IsSupported, "Ssse3 support is required to run this operation!");
fixed(byte *shuffleVectorsPtr = &MemoryMarshal.GetReference(SseShuffleMasks))
{
Vector128 <byte> rowA = block.V0.AsByte();
Vector128 <byte> rowB = block.V1.AsByte();
Vector128 <byte> rowC = block.V2.AsByte();
Vector128 <byte> rowD = block.V3.AsByte();
Vector128 <byte> rowE = block.V4.AsByte();
Vector128 <byte> rowF = block.V5.AsByte();
Vector128 <byte> rowG = block.V6.AsByte();
Vector128 <byte> rowH = block.V7.AsByte();
// row0 - A0 B0 A1 A2 B1 C0 D0 C1
Vector128 <short> row0_A = Ssse3.Shuffle(rowA, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 0))).AsInt16();
Vector128 <short> row0_B = Ssse3.Shuffle(rowB, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 1))).AsInt16();
Vector128 <short> row0_C = Ssse3.Shuffle(rowC, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 2))).AsInt16();
Vector128 <short> row0 = Sse2.Or(Sse2.Or(row0_A, row0_B), row0_C);
row0 = Sse2.Insert(row0.AsUInt16(), Sse2.Extract(rowD.AsUInt16(), 0), 6).AsInt16();
// row1 - B2 A3 A4 B3 C2 D1 E0 F0
Vector128 <short> row1_A = Ssse3.Shuffle(rowA, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 3))).AsInt16();
Vector128 <short> row1_B = Ssse3.Shuffle(rowB, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 4))).AsInt16();
Vector128 <short> row1 = Sse2.Or(row1_A, row1_B);
row1 = Sse2.Insert(row1.AsUInt16(), Sse2.Extract(rowC.AsUInt16(), 2), 4).AsInt16();
row1 = Sse2.Insert(row1.AsUInt16(), Sse2.Extract(rowD.AsUInt16(), 1), 5).AsInt16();
row1 = Sse2.Insert(row1.AsUInt16(), Sse2.Extract(rowE.AsUInt16(), 0), 6).AsInt16();
row1 = Sse2.Insert(row1.AsUInt16(), Sse2.Extract(rowF.AsUInt16(), 0), 7).AsInt16();
// row2 - E1 D2 C3 B4 A5 A6 B5 C4
Vector128 <short> row2_A = Ssse3.Shuffle(rowA, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 5))).AsInt16();
Vector128 <short> row2_B = Ssse3.Shuffle(rowB, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 6))).AsInt16();
Vector128 <short> row2_C = Ssse3.Shuffle(rowC, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 7))).AsInt16();
Vector128 <short> row2 = Sse2.Or(Sse2.Or(row2_A, row2_B), row2_C);
row2 = Sse2.Insert(row2.AsUInt16(), Sse2.Extract(rowD.AsUInt16(), 2), 1).AsInt16();
row2 = Sse2.Insert(row2.AsUInt16(), Sse2.Extract(rowE.AsUInt16(), 1), 0).AsInt16();
// row3 - D3 E2 F1 G0 H0 G1 F2 E3
Vector128 <short> row3_E = Ssse3.Shuffle(rowE, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 8))).AsInt16();
Vector128 <short> row3_F = Ssse3.Shuffle(rowF, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 9))).AsInt16();
Vector128 <short> row3_G = Ssse3.Shuffle(rowG, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 10))).AsInt16();
Vector128 <short> row3 = Sse2.Or(Sse2.Or(row3_E, row3_F), row3_G);
row3 = Sse2.Insert(row3.AsUInt16(), Sse2.Extract(rowD.AsUInt16(), 3), 0).AsInt16();
row3 = Sse2.Insert(row3.AsUInt16(), Sse2.Extract(rowH.AsUInt16(), 0), 4).AsInt16();
// row4 - D4 C5 B6 A7 B7 C6 D5 E4
Vector128 <short> row4_B = Ssse3.Shuffle(rowB, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 11))).AsInt16();
Vector128 <short> row4_C = Ssse3.Shuffle(rowC, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 12))).AsInt16();
Vector128 <short> row4_D = Ssse3.Shuffle(rowD, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 13))).AsInt16();
Vector128 <short> row4 = Sse2.Or(Sse2.Or(row4_B, row4_C), row4_D);
row4 = Sse2.Insert(row4.AsUInt16(), Sse2.Extract(rowA.AsUInt16(), 7), 3).AsInt16();
row4 = Sse2.Insert(row4.AsUInt16(), Sse2.Extract(rowE.AsUInt16(), 4), 7).AsInt16();
// row5 - F3 G2 H1 H2 G3 F4 E5 D6
Vector128 <short> row5_F = Ssse3.Shuffle(rowF, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 14))).AsInt16();
Vector128 <short> row5_G = Ssse3.Shuffle(rowG, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 15))).AsInt16();
Vector128 <short> row5_H = Ssse3.Shuffle(rowH, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 16))).AsInt16();
Vector128 <short> row5 = Sse2.Or(Sse2.Or(row5_F, row5_G), row5_H);
row5 = Sse2.Insert(row5.AsUInt16(), Sse2.Extract(rowD.AsUInt16(), 6), 7).AsInt16();
row5 = Sse2.Insert(row5.AsUInt16(), Sse2.Extract(rowE.AsUInt16(), 5), 6).AsInt16();
// row6 - C7 D7 E6 F5 G4 H3 H4 G5
Vector128 <short> row6_G = Ssse3.Shuffle(rowG, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 17))).AsInt16();
Vector128 <short> row6_H = Ssse3.Shuffle(rowH, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 18))).AsInt16();
Vector128 <short> row6 = Sse2.Or(row6_G, row6_H);
row6 = Sse2.Insert(row6.AsUInt16(), Sse2.Extract(rowC.AsUInt16(), 7), 0).AsInt16();
row6 = Sse2.Insert(row6.AsUInt16(), Sse2.Extract(rowD.AsUInt16(), 7), 1).AsInt16();
row6 = Sse2.Insert(row6.AsUInt16(), Sse2.Extract(rowE.AsUInt16(), 6), 2).AsInt16();
row6 = Sse2.Insert(row6.AsUInt16(), Sse2.Extract(rowF.AsUInt16(), 5), 3).AsInt16();
// row7 - F6 E7 F7 G6 H5 H6 G7 H7
Vector128 <short> row7_F = Ssse3.Shuffle(rowF, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 19))).AsInt16();
Vector128 <short> row7_G = Ssse3.Shuffle(rowG, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 20))).AsInt16();
Vector128 <short> row7_H = Ssse3.Shuffle(rowH, Sse2.LoadVector128(shuffleVectorsPtr + (16 * 21))).AsInt16();
Vector128 <short> row7 = Sse2.Or(Sse2.Or(row7_F, row7_G), row7_H);
row7 = Sse2.Insert(row7.AsUInt16(), Sse2.Extract(rowE.AsUInt16(), 7), 1).AsInt16();
block.V0 = row0;
block.V1 = row1;
block.V2 = row2;
block.V3 = row3;
block.V4 = row4;
block.V5 = row5;
block.V6 = row6;
block.V7 = row7;
}
}
/// <summary>
/// Applies zig zag ordering for given 8x8 matrix using AVX cpu intrinsics.
/// </summary>
/// <param name="block">Input matrix.</param>
public static unsafe void ApplyTransposingZigZagOrderingAvx2(ref Block8x8 block)
{
DebugGuard.IsTrue(Avx2.IsSupported, "Avx2 support is required to run this operation!");
fixed(byte *shuffleVectorsPtr = &MemoryMarshal.GetReference(AvxShuffleMasks))
{
Vector256 <byte> rowAB = block.V01.AsByte();
Vector256 <byte> rowCD = block.V23.AsByte();
Vector256 <byte> rowEF = block.V45.AsByte();
Vector256 <byte> rowGH = block.V67.AsByte();
/* row01 - A0 B0 A1 A2 B1 C0 D0 C1 | B2 A3 A4 B3 C2 D1 E0 F0 */
Vector256 <int> crln_01_AB_CD = Avx.LoadVector256(shuffleVectorsPtr + (0 * 32)).AsInt32();
Vector256 <byte> row01_AB = Avx2.PermuteVar8x32(rowAB.AsInt32(), crln_01_AB_CD).AsByte();
row01_AB = Avx2.Shuffle(row01_AB, Avx.LoadVector256(shuffleVectorsPtr + (1 * 32))).AsByte();
Vector256 <byte> row01_CD = Avx2.PermuteVar8x32(rowCD.AsInt32(), crln_01_AB_CD).AsByte();
row01_CD = Avx2.Shuffle(row01_CD, Avx.LoadVector256(shuffleVectorsPtr + (2 * 32))).AsByte();
Vector256 <int> crln_01_23_EF_23_CD = Avx.LoadVector256(shuffleVectorsPtr + (3 * 32)).AsInt32();
Vector256 <byte> row01_23_EF = Avx2.PermuteVar8x32(rowEF.AsInt32(), crln_01_23_EF_23_CD).AsByte();
Vector256 <byte> row01_EF = Avx2.Shuffle(row01_23_EF, Avx.LoadVector256(shuffleVectorsPtr + (4 * 32))).AsByte();
Vector256 <byte> row01 = Avx2.Or(row01_AB, Avx2.Or(row01_CD, row01_EF));
/* row23 - E1 D2 C3 B4 A5 A6 B5 C4 | D3 E2 F1 G0 H0 G1 F2 E3 */
Vector256 <int> crln_23_AB_23_45_GH = Avx.LoadVector256(shuffleVectorsPtr + (5 * 32)).AsInt32();
Vector256 <byte> row23_45_AB = Avx2.PermuteVar8x32(rowAB.AsInt32(), crln_23_AB_23_45_GH).AsByte();
Vector256 <byte> row23_AB = Avx2.Shuffle(row23_45_AB, Avx.LoadVector256(shuffleVectorsPtr + (6 * 32))).AsByte();
Vector256 <byte> row23_CD = Avx2.PermuteVar8x32(rowCD.AsInt32(), crln_01_23_EF_23_CD).AsByte();
row23_CD = Avx2.Shuffle(row23_CD, Avx.LoadVector256(shuffleVectorsPtr + (7 * 32))).AsByte();
Vector256 <byte> row23_EF = Avx2.Shuffle(row01_23_EF, Avx.LoadVector256(shuffleVectorsPtr + (8 * 32))).AsByte();
Vector256 <byte> row23_45_GH = Avx2.PermuteVar8x32(rowGH.AsInt32(), crln_23_AB_23_45_GH).AsByte();
Vector256 <byte> row23_GH = Avx2.Shuffle(row23_45_GH, Avx.LoadVector256(shuffleVectorsPtr + (9 * 32))).AsByte();
Vector256 <byte> row23 = Avx2.Or(Avx2.Or(row23_AB, row23_CD), Avx2.Or(row23_EF, row23_GH));
/* row45 - D4 C5 B6 A7 B7 C6 D5 E4 | F3 G2 H1 H2 G3 F4 E5 D6 */
Vector256 <byte> row45_AB = Avx2.Shuffle(row23_45_AB, Avx.LoadVector256(shuffleVectorsPtr + (10 * 32))).AsByte();
Vector256 <int> crln_45_67_CD_45_EF = Avx.LoadVector256(shuffleVectorsPtr + (11 * 32)).AsInt32();
Vector256 <byte> row45_67_CD = Avx2.PermuteVar8x32(rowCD.AsInt32(), crln_45_67_CD_45_EF).AsByte();
Vector256 <byte> row45_CD = Avx2.Shuffle(row45_67_CD, Avx.LoadVector256(shuffleVectorsPtr + (12 * 32))).AsByte();
Vector256 <byte> row45_EF = Avx2.PermuteVar8x32(rowEF.AsInt32(), crln_45_67_CD_45_EF).AsByte();
row45_EF = Avx2.Shuffle(row45_EF, Avx.LoadVector256(shuffleVectorsPtr + (13 * 32))).AsByte();
Vector256 <byte> row45_GH = Avx2.Shuffle(row23_45_GH, Avx.LoadVector256(shuffleVectorsPtr + (14 * 32))).AsByte();
Vector256 <byte> row45 = Avx2.Or(Avx2.Or(row45_AB, row45_CD), Avx2.Or(row45_EF, row45_GH));
/* row67 - C7 D7 E6 F5 G4 H3 H4 G5 | F6 E7 F7 G6 H5 H6 G7 H7 */
Vector256 <byte> row67_CD = Avx2.Shuffle(row45_67_CD, Avx.LoadVector256(shuffleVectorsPtr + (15 * 32))).AsByte();
Vector256 <int> crln_67_EF_67_GH = Avx.LoadVector256(shuffleVectorsPtr + (16 * 32)).AsInt32();
Vector256 <byte> row67_EF = Avx2.PermuteVar8x32(rowEF.AsInt32(), crln_67_EF_67_GH).AsByte();
row67_EF = Avx2.Shuffle(row67_EF, Avx.LoadVector256(shuffleVectorsPtr + (17 * 32))).AsByte();
Vector256 <byte> row67_GH = Avx2.PermuteVar8x32(rowGH.AsInt32(), crln_67_EF_67_GH).AsByte();
row67_GH = Avx2.Shuffle(row67_GH, Avx.LoadVector256(shuffleVectorsPtr + (18 * 32))).AsByte();
Vector256 <byte> row67 = Avx2.Or(row67_CD, Avx2.Or(row67_EF, row67_GH));
block.V01 = row01.AsInt16();
block.V23 = row23.AsInt16();
block.V45 = row45.AsInt16();
block.V67 = row67.AsInt16();
}
}
