diff --git a/src/ImageSharp/Formats/Jpeg/Components/Decoder/YCbCrToRgbTables.cs b/src/ImageSharp/Formats/Jpeg/Components/Decoder/YCbCrToRgbTables.cs new file mode 100644 index 0000000000..27324b5f68 --- /dev/null +++ b/src/ImageSharp/Formats/Jpeg/Components/Decoder/YCbCrToRgbTables.cs @@ -0,0 +1,109 @@ +// +// Copyright (c) James Jackson-South and contributors. +// Licensed under the Apache License, Version 2.0. +// + +namespace ImageSharp.Formats.Jpg +{ + using System.Runtime.CompilerServices; + using ImageSharp.PixelFormats; + + /// + /// Provides 8-bit lookup tables for converting from YCbCr to Rgb colorspace. + /// Methods to build the tables are based on libjpeg implementation. + /// + internal unsafe struct YCbCrToRgbTables + { + /// + /// The red red-chrominance table + /// + public fixed int CrRTable[256]; + + /// + /// The blue blue-chrominance table + /// + public fixed int CbBTable[256]; + + /// + /// The green red-chrominance table + /// + public fixed int CrGTable[256]; + + /// + /// The green blue-chrominance table + /// + public fixed int CbGTable[256]; + + // Speediest right-shift on some machines and gives us enough accuracy at 4 decimal places. + private const int ScaleBits = 16; + + private const int Half = 1 << (ScaleBits - 1); + + /// + /// Initializes the YCbCr tables + /// + /// The intialized + public static YCbCrToRgbTables Create() + { + YCbCrToRgbTables tables = default(YCbCrToRgbTables); + + for (int i = 0, x = -128; i <= 255; i++, x++) + { + // i is the actual input pixel value, in the range 0..255 + // The Cb or Cr value we are thinking of is x = i - 128 + // Cr=>R value is nearest int to 1.402 * x + tables.CrRTable[i] = RightShift((Fix(1.402F) * x) + Half); + + // Cb=>B value is nearest int to 1.772 * x + tables.CbBTable[i] = RightShift((Fix(1.772F) * x) + Half); + + // Cr=>G value is scaled-up -0.714136286 + tables.CrGTable[i] = (-Fix(0.714136286F)) * x; + + // Cb => G value is scaled - up - 0.344136286 * x + // We also add in Half so that need not do it in inner loop + tables.CbGTable[i] = ((-Fix(0.344136286F)) * x) + Half; + } + + return tables; + } + + /// + /// Optimized method to pack bytes to the image from the YCbCr color space. + /// + /// The pixel format. + /// The packed pixel. + /// The reference to the tables instance. + /// The y luminance component. + /// The cb chroma component. + /// The cr chroma component. + [MethodImpl(MethodImplOptions.AggressiveInlining)] + public static void Pack(ref TPixel packed, YCbCrToRgbTables* tables, byte y, byte cb, byte cr) + where TPixel : struct, IPixel + { + // float r = MathF.Round(y + (1.402F * cr), MidpointRounding.AwayFromZero); + byte r = (byte)(y + tables->CrRTable[cr]).Clamp(0, 255); + + // float g = MathF.Round(y - (0.344136F * cb) - (0.714136F * cr), MidpointRounding.AwayFromZero); + // The values for the G calculation are left scaled up, since we must add them together before rounding. + byte g = (byte)(y + RightShift(tables->CbGTable[cb] + tables->CrGTable[cr])).Clamp(0, 255); + + // float b = MathF.Round(y + (1.772F * cb), MidpointRounding.AwayFromZero); + byte b = (byte)(y + tables->CbBTable[cb]).Clamp(0, 255); + + packed.PackFromBytes(r, g, b, byte.MaxValue); + } + + [MethodImpl(MethodImplOptions.AggressiveInlining)] + private static int Fix(float x) + { + return (int)((x * (1L << ScaleBits)) + 0.5F); + } + + [MethodImpl(MethodImplOptions.AggressiveInlining)] + private static int RightShift(int x) + { + return x >> ScaleBits; + } + } +} \ No newline at end of file diff --git a/src/ImageSharp/Formats/Jpeg/Components/Encoder/RgbToYCbCrTables.cs b/src/ImageSharp/Formats/Jpeg/Components/Encoder/RgbToYCbCrTables.cs new file mode 100644 index 0000000000..94173d3e43 --- /dev/null +++ b/src/ImageSharp/Formats/Jpeg/Components/Encoder/RgbToYCbCrTables.cs @@ -0,0 +1,131 @@ +// +// Copyright (c) James Jackson-South and contributors. +// Licensed under the Apache License, Version 2.0. +// + +namespace ImageSharp.Formats.Jpg +{ + using System.Runtime.CompilerServices; + + /// + /// Provides 8-bit lookup tables for converting from Rgb to YCbCr colorspace. + /// Methods to build the tables are based on libjpeg implementation. + /// + internal unsafe struct RgbToYCbCrTables + { + /// + /// The red luminance table + /// + public fixed int YRTable[256]; + + /// + /// The green luminance table + /// + public fixed int YGTable[256]; + + /// + /// The blue luminance table + /// + public fixed int YBTable[256]; + + /// + /// The red blue-chrominance table + /// + public fixed int CbRTable[256]; + + /// + /// The green blue-chrominance table + /// + public fixed int CbGTable[256]; + + /// + /// The blue blue-chrominance table + /// B=>Cb and R=>Cr are the same + /// + public fixed int CbBTable[256]; + + /// + /// The green red-chrominance table + /// + public fixed int CrGTable[256]; + + /// + /// The blue red-chrominance table + /// + public fixed int CrBTable[256]; + + // Speediest right-shift on some machines and gives us enough accuracy at 4 decimal places. + private const int ScaleBits = 16; + + private const int CBCrOffset = 128 << ScaleBits; + + private const int Half = 1 << (ScaleBits - 1); + + /// + /// Initializes the YCbCr tables + /// + /// The intialized + public static RgbToYCbCrTables Create() + { + RgbToYCbCrTables tables = default(RgbToYCbCrTables); + + for (int i = 0; i <= 255; i++) + { + // The values for the calculations are left scaled up since we must add them together before rounding. + tables.YRTable[i] = Fix(0.299F) * i; + tables.YGTable[i] = Fix(0.587F) * i; + tables.YBTable[i] = (Fix(0.114F) * i) + Half; + tables.CbRTable[i] = (-Fix(0.168735892F)) * i; + tables.CbGTable[i] = (-Fix(0.331264108F)) * i; + + // We use a rounding fudge - factor of 0.5 - epsilon for Cb and Cr. + // This ensures that the maximum output will round to 255 + // not 256, and thus that we don't have to range-limit. + // + // B=>Cb and R=>Cr tables are the same + tables.CbBTable[i] = (Fix(0.5F) * i) + CBCrOffset + Half - 1; + + tables.CrGTable[i] = (-Fix(0.418687589F)) * i; + tables.CrBTable[i] = (-Fix(0.081312411F)) * i; + } + + return tables; + } + + /// + /// Optimized method to allocates the correct y, cb, and cr values to the DCT blocks from the given r, g, b values. + /// + /// The The luminance block. + /// The red chroma block. + /// The blue chroma block. + /// The reference to the tables instance. + /// The current index. + /// The red value. + /// The green value. + /// The blue value. + [MethodImpl(MethodImplOptions.AggressiveInlining)] + public static void Allocate(ref float* yBlockRaw, ref float* cbBlockRaw, ref float* crBlockRaw, ref RgbToYCbCrTables* tables, int index, int r, int g, int b) + { + // float y = (0.299F * r) + (0.587F * g) + (0.114F * b); + yBlockRaw[index] = (tables->YRTable[r] + tables->YGTable[g] + tables->YBTable[b]) >> ScaleBits; + + // float cb = 128F + ((-0.168736F * r) - (0.331264F * g) + (0.5F * b)); + cbBlockRaw[index] = (tables->CbRTable[r] + tables->CbGTable[g] + tables->CbBTable[b]) >> ScaleBits; + + // float b = MathF.Round(y + (1.772F * cb), MidpointRounding.AwayFromZero); + crBlockRaw[index] = (tables->CbBTable[r] + tables->CrGTable[g] + tables->CrBTable[b]) >> ScaleBits; + } + + [MethodImpl(MethodImplOptions.AggressiveInlining)] + private static int Fix(float x) + { + return (int)((x * (1L << ScaleBits)) + 0.5F); + } + + [MethodImpl(MethodImplOptions.AggressiveInlining)] + private static int RightShift(int x) + { + return x >> ScaleBits; + } + } +} \ No newline at end of file diff --git a/src/ImageSharp/Formats/Jpeg/JpegDecoderCore.cs b/src/ImageSharp/Formats/Jpeg/JpegDecoderCore.cs index 186c1e5282..9df21a3b72 100644 --- a/src/ImageSharp/Formats/Jpeg/JpegDecoderCore.cs +++ b/src/ImageSharp/Formats/Jpeg/JpegDecoderCore.cs @@ -38,6 +38,11 @@ namespace ImageSharp.Formats public InputProcessor InputProcessor; #pragma warning restore SA401 + /// + /// Lookup tables for converting YCbCr to Rgb + /// + private static YCbCrToRgbTables yCbCrToRgbTables = YCbCrToRgbTables.Create(); + /// /// The decoder options. /// @@ -251,35 +256,6 @@ namespace ImageSharp.Formats } } - /// - /// Optimized method to pack bytes to the image from the YCbCr color space. - /// This is faster than implicit casting as it avoids double packing. - /// - /// The pixel format. - /// The packed pixel. - /// The y luminance component. - /// The cb chroma component. - /// The cr chroma component. - [MethodImpl(MethodImplOptions.AggressiveInlining)] - private static void PackYcbCr(ref TPixel packed, byte y, byte cb, byte cr) - where TPixel : struct, IPixel - { - int ccb = cb - 128; - int ccr = cr - 128; - - // Speed up the algorithm by removing floating point calculation - // Scale by 65536, add .5F and truncate value. We use bit shifting to divide the result - int r0 = 91881 * ccr; // (1.402F * 65536) + .5F - int g0 = 22554 * ccb; // (0.34414F * 65536) + .5F - int g1 = 46802 * ccr; // (0.71414F * 65536) + .5F - int b0 = 116130 * ccb; // (1.772F * 65536) + .5F - - byte r = (byte)(y + (r0 >> 16)).Clamp(0, 255); - byte g = (byte)(y - (g0 >> 16) - (g1 >> 16)).Clamp(0, 255); - byte b = (byte)(y + (b0 >> 16)).Clamp(0, 255); - packed.PackFromBytes(r, g, b, 255); - } - /// /// Read metadata from stream and read the blocks in the scans into . /// @@ -705,26 +681,34 @@ namespace ImageSharp.Formats using (PixelAccessor pixels = image.Lock()) { Parallel.For( - 0, - image.Height, - image.Configuration.ParallelOptions, - y => - { - // TODO: Simplify + optimize + share duplicate code across converter methods - int yo = this.ycbcrImage.GetRowYOffset(y); - int co = this.ycbcrImage.GetRowCOffset(y); - - for (int x = 0; x < image.Width; x++) - { - byte yy = this.ycbcrImage.YChannel.Pixels[yo + x]; - byte cb = this.ycbcrImage.CbChannel.Pixels[co + (x / scale)]; - byte cr = this.ycbcrImage.CrChannel.Pixels[co + (x / scale)]; - - TPixel packed = default(TPixel); - PackYcbCr(ref packed, yy, cb, cr); - pixels[x, y] = packed; - } - }); + 0, + image.Height, + image.Configuration.ParallelOptions, + y => + { + // TODO. This Parallel loop doesn't give us the boost it should. + ref byte ycRef = ref this.ycbcrImage.YChannel.Pixels[0]; + ref byte cbRef = ref this.ycbcrImage.CbChannel.Pixels[0]; + ref byte crRef = ref this.ycbcrImage.CrChannel.Pixels[0]; + fixed (YCbCrToRgbTables* tables = &yCbCrToRgbTables) + { + // TODO: Simplify + optimize + share duplicate code across converter methods + int yo = this.ycbcrImage.GetRowYOffset(y); + int co = this.ycbcrImage.GetRowCOffset(y); + + for (int x = 0; x < image.Width; x++) + { + int cOff = co + (x / scale); + byte yy = Unsafe.Add(ref ycRef, yo + x); + byte cb = Unsafe.Add(ref cbRef, cOff); + byte cr = Unsafe.Add(ref crRef, cOff); + + TPixel packed = default(TPixel); + YCbCrToRgbTables.Pack(ref packed, tables, yy, cb, cr); + pixels[x, y] = packed; + } + } + }); } this.AssignResolution(image); diff --git a/src/ImageSharp/Formats/Jpeg/JpegEncoderCore.cs b/src/ImageSharp/Formats/Jpeg/JpegEncoderCore.cs index eb083c35d9..0ce59c6dec 100644 --- a/src/ImageSharp/Formats/Jpeg/JpegEncoderCore.cs +++ b/src/ImageSharp/Formats/Jpeg/JpegEncoderCore.cs @@ -5,11 +5,9 @@ namespace ImageSharp.Formats { - using System; using System.Buffers; using System.IO; using System.Runtime.CompilerServices; - using ImageSharp.Formats.Jpg; using ImageSharp.Formats.Jpg.Components; using ImageSharp.PixelFormats; @@ -103,6 +101,11 @@ namespace ImageSharp.Formats } }; + /// + /// Lookup tables for converting Rgb to YCbCr + /// + private static RgbToYCbCrTables rgbToYCbCrTables = RgbToYCbCrTables.Create(); + /// /// A scratch buffer to reduce allocations. /// @@ -285,6 +288,7 @@ namespace ImageSharp.Formats /// /// The pixel format. /// The pixel accessor. + /// The reference to the tables instance. /// The x-position within the image. /// The y-position within the image. /// The luminance block. @@ -293,6 +297,7 @@ namespace ImageSharp.Formats /// Temporal provided by the caller private static void ToYCbCr( PixelAccessor pixels, + RgbToYCbCrTables* tables, int x, int y, Block8x8F* yBlock, @@ -321,29 +326,9 @@ namespace ImageSharp.Formats int g = Unsafe.Add(ref data0, dataIdx + 1); int b = Unsafe.Add(ref data0, dataIdx + 2); - // Speed up the algorithm by removing floating point calculation - // Scale by 65536, add .5F and truncate value. We use bit shifting to divide the result - int y0 = 19595 * r; // (0.299F * 65536) + .5F - int y1 = 38470 * g; // (0.587F * 65536) + .5F - int y2 = 7471 * b; // (0.114F * 65536) + .5F - - int cb0 = -11057 * r; // (-0.168736F * 65536) + .5F - int cb1 = 21710 * g; // (0.331264F * 65536) + .5F - int cb2 = 32768 * b; // (0.5F * 65536) + .5F - - int cr0 = 32768 * r; // (0.5F * 65536) + .5F - int cr1 = 27439 * g; // (0.418688F * 65536) + .5F - int cr2 = 5329 * b; // (0.081312F * 65536) + .5F - - float yy = (y0 + y1 + y2) >> 16; - float cb = 128 + ((cb0 - cb1 + cb2) >> 16); - float cr = 128 + ((cr0 - cr1 - cr2) >> 16); - int index = j8 + i; - yBlockRaw[index] = yy; - cbBlockRaw[index] = cb; - crBlockRaw[index] = cr; + RgbToYCbCrTables.Allocate(ref yBlockRaw, ref cbBlockRaw, ref crBlockRaw, ref tables, index, r, g, b); dataIdx += 3; } @@ -464,38 +449,41 @@ namespace ImageSharp.Formats // ReSharper disable once InconsistentNaming int prevDCY = 0, prevDCCb = 0, prevDCCr = 0; - using (PixelArea rgbBytes = new PixelArea(8, 8, ComponentOrder.Xyz)) + fixed (RgbToYCbCrTables* tables = &rgbToYCbCrTables) { - for (int y = 0; y < pixels.Height; y += 8) + using (PixelArea rgbBytes = new PixelArea(8, 8, ComponentOrder.Xyz)) { - for (int x = 0; x < pixels.Width; x += 8) + for (int y = 0; y < pixels.Height; y += 8) { - ToYCbCr(pixels, x, y, &b, &cb, &cr, rgbBytes); - - prevDCY = this.WriteBlock( - QuantIndex.Luminance, - prevDCY, - &b, - &temp1, - &temp2, - &onStackLuminanceQuantTable, - unzig.Data); - prevDCCb = this.WriteBlock( - QuantIndex.Chrominance, - prevDCCb, - &cb, - &temp1, - &temp2, - &onStackChrominanceQuantTable, - unzig.Data); - prevDCCr = this.WriteBlock( - QuantIndex.Chrominance, - prevDCCr, - &cr, - &temp1, - &temp2, - &onStackChrominanceQuantTable, - unzig.Data); + for (int x = 0; x < pixels.Width; x += 8) + { + ToYCbCr(pixels, tables, x, y, &b, &cb, &cr, rgbBytes); + + prevDCY = this.WriteBlock( + QuantIndex.Luminance, + prevDCY, + &b, + &temp1, + &temp2, + &onStackLuminanceQuantTable, + unzig.Data); + prevDCCb = this.WriteBlock( + QuantIndex.Chrominance, + prevDCCb, + &cb, + &temp1, + &temp2, + &onStackChrominanceQuantTable, + unzig.Data); + prevDCCr = this.WriteBlock( + QuantIndex.Chrominance, + prevDCCr, + &cr, + &temp1, + &temp2, + &onStackChrominanceQuantTable, + unzig.Data); + } } } } @@ -837,49 +825,51 @@ namespace ImageSharp.Formats // ReSharper disable once InconsistentNaming int prevDCY = 0, prevDCCb = 0, prevDCCr = 0; - - using (PixelArea rgbBytes = new PixelArea(8, 8, ComponentOrder.Xyz)) + fixed (RgbToYCbCrTables* tables = &rgbToYCbCrTables) { - for (int y = 0; y < pixels.Height; y += 16) + using (PixelArea rgbBytes = new PixelArea(8, 8, ComponentOrder.Xyz)) { - for (int x = 0; x < pixels.Width; x += 16) + for (int y = 0; y < pixels.Height; y += 16) { - for (int i = 0; i < 4; i++) + for (int x = 0; x < pixels.Width; x += 16) { - int xOff = (i & 1) * 8; - int yOff = (i & 2) * 4; - - ToYCbCr(pixels, x + xOff, y + yOff, &b, cbPtr + i, crPtr + i, rgbBytes); + for (int i = 0; i < 4; i++) + { + int xOff = (i & 1) * 8; + int yOff = (i & 2) * 4; + + ToYCbCr(pixels, tables, x + xOff, y + yOff, &b, cbPtr + i, crPtr + i, rgbBytes); + + prevDCY = this.WriteBlock( + QuantIndex.Luminance, + prevDCY, + &b, + &temp1, + &temp2, + &onStackLuminanceQuantTable, + unzig.Data); + } + + Block8x8F.Scale16X16To8X8(&b, cbPtr); + prevDCCb = this.WriteBlock( + QuantIndex.Chrominance, + prevDCCb, + &b, + &temp1, + &temp2, + &onStackChrominanceQuantTable, + unzig.Data); - prevDCY = this.WriteBlock( - QuantIndex.Luminance, - prevDCY, + Block8x8F.Scale16X16To8X8(&b, crPtr); + prevDCCr = this.WriteBlock( + QuantIndex.Chrominance, + prevDCCr, &b, &temp1, &temp2, - &onStackLuminanceQuantTable, + &onStackChrominanceQuantTable, unzig.Data); } - - Block8x8F.Scale16X16To8X8(&b, cbPtr); - prevDCCb = this.WriteBlock( - QuantIndex.Chrominance, - prevDCCb, - &b, - &temp1, - &temp2, - &onStackChrominanceQuantTable, - unzig.Data); - - Block8x8F.Scale16X16To8X8(&b, crPtr); - prevDCCr = this.WriteBlock( - QuantIndex.Chrominance, - prevDCCr, - &b, - &temp1, - &temp2, - &onStackChrominanceQuantTable, - unzig.Data); } } } diff --git a/tests/ImageSharp.Tests/Colors/RgbaVectorTransformTests.cs b/tests/ImageSharp.Tests/Colors/RgbaVectorTransformTests.cs index bcbc27c7c4..24850954a8 100644 --- a/tests/ImageSharp.Tests/Colors/RgbaVectorTransformTests.cs +++ b/tests/ImageSharp.Tests/Colors/RgbaVectorTransformTests.cs @@ -36,13 +36,13 @@ namespace ImageSharp.Tests.Colors [Fact] public void Multiply() { - Assert.Equal(RgbaVector.Multiply(Backdrop, RgbaVector.Black).ToVector4(), Rgba32.Black.ToVector4(), FloatComparer); + Assert.Equal(RgbaVector.Multiply(Backdrop, RgbaVector.Black).ToVector4(), RgbaVector.Black.ToVector4(), FloatComparer); Assert.Equal(RgbaVector.Multiply(Backdrop, RgbaVector.White).ToVector4(), Backdrop.ToVector4(), FloatComparer); RgbaVector multiply = RgbaVector.Multiply(Backdrop, Source); Assert.Equal(multiply.ToVector4(), new RgbaVector(0, 41, 0).ToVector4(), FloatComparer); } - + [Fact] public void Screen() {