Super-optimized GenericBlock8x8<T> to replace PixelArea<T> in JpegEncoder

src/ImageSharp/Formats/Jpeg/Common/GenericBlock8x8.Generated.cs

@@ -0,0 +1,26 @@
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System.Numerics;
+using System.Runtime.CompilerServices;
+
+// <auto-generated />
+namespace SixLabors.ImageSharp.Formats.Jpeg.Common
+{
+	internal unsafe partial struct GenericBlock8x8<T>
+    {
+        #pragma warning disable 169
+
+        // It's not allowed use fix-sized buffers with generics, need to place all the fields manually:
+        private T _y0_x0, _y0_x1, _y0_x2, _y0_x3, _y0_x4, _y0_x5, _y0_x6, _y0_x7;
+        private T _y1_x0, _y1_x1, _y1_x2, _y1_x3, _y1_x4, _y1_x5, _y1_x6, _y1_x7;
+        private T _y2_x0, _y2_x1, _y2_x2, _y2_x3, _y2_x4, _y2_x5, _y2_x6, _y2_x7;
+        private T _y3_x0, _y3_x1, _y3_x2, _y3_x3, _y3_x4, _y3_x5, _y3_x6, _y3_x7;
+        private T _y4_x0, _y4_x1, _y4_x2, _y4_x3, _y4_x4, _y4_x5, _y4_x6, _y4_x7;
+        private T _y5_x0, _y5_x1, _y5_x2, _y5_x3, _y5_x4, _y5_x5, _y5_x6, _y5_x7;
+        private T _y6_x0, _y6_x1, _y6_x2, _y6_x3, _y6_x4, _y6_x5, _y6_x6, _y6_x7;
+        private T _y7_x0, _y7_x1, _y7_x2, _y7_x3, _y7_x4, _y7_x5, _y7_x6, _y7_x7;
+
+        #pragma warning restore 169
+	}
+}

src/ImageSharp/Formats/Jpeg/Common/GenericBlock8x8.Generated.tt

@@ -0,0 +1,43 @@
+<#
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+#>
+<#@ template debug="false" hostspecific="false" language="C#" #>
+<#@ assembly name="System.Core" #>
+<#@ import namespace="System.Linq" #>
+<#@ import namespace="System.Text" #>
+<#@ import namespace="System.Collections.Generic" #>
+<#@ output extension=".cs" #>
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System.Numerics;
+using System.Runtime.CompilerServices;
+
+// <auto-generated />
+namespace SixLabors.ImageSharp.Formats.Jpeg.Common
+{
+	internal unsafe partial struct GenericBlock8x8<T>
+    {
+        #pragma warning disable 169
+
+        // It's not allowed use fix-sized buffers with generics, need to place all the fields manually:
+    <#
+			PushIndent("        ");
+            Write("    ");
+            for (int y = 0; y < 8; y++)
+            {
+                Write("private T ");
+                for (int x = 0; x < 8; x++)
+                {
+					Write($"_y{y}_x{x}");
+                    if (x < 7) Write(", ");
+                }
+                WriteLine(";");
+            }
+			PopIndent();
+	#>
+
+        #pragma warning restore 169
+	}
+}

src/ImageSharp/Formats/Jpeg/Common/GenericBlock8x8.cs

@@ -0,0 +1,128 @@
+using System;
+using System.Runtime.CompilerServices;
+using SixLabors.ImageSharp.Advanced;
+using SixLabors.ImageSharp.Memory;
+using SixLabors.ImageSharp.PixelFormats;
+using SixLabors.Primitives;
+
+namespace SixLabors.ImageSharp.Formats.Jpeg.Common
+{
+    /// <summary>
+    /// A generic 8x8 block implementation, useful for manipulating custom 8x8 pixel data.
+    /// </summary>
+    // ReSharper disable once InconsistentNaming
+    internal unsafe partial struct GenericBlock8x8<T>
+        where T : struct
+    {
+        public const int Size = 64;
+
+        public const int SizeInBytes = Size * 3;
+
+        public void LoadAndStretchEdges<TPixel>(IPixelSource<TPixel> source, int sourceX, int sourceY)
+            where TPixel : struct, IPixel<TPixel>
+        {
+            var buffer = source.PixelBuffer as Buffer2D<T>;
+            if (buffer == null)
+            {
+                throw new InvalidOperationException("LoadAndStretchEdges<TPixels>() is only valid for TPixel == T !");
+            }
+
+            this.LoadAndStretchEdges(buffer, sourceX, sourceY);
+        }
+
+        /// <summary>
+        /// Load a 8x8 region of an image into the block.
+        /// The "outlying" area of the block will be stretched out with pixels on the right and bottom edge of the image.
+        /// </summary>
+        public void LoadAndStretchEdges(Buffer2D<T> source, int sourceX, int sourceY)
+        {
+            int width = Math.Min(8, source.Width - sourceX);
+            int height = Math.Min(8, source.Height - sourceY);
+
+            if (width <= 0 || height <= 0)
+            {
+                return;
+            }
+
+            uint byteWidth = (uint)width * (uint)Unsafe.SizeOf<T>();
+            int remainderXCount = 8 - width;
+
+            ref byte blockStart = ref Unsafe.As<GenericBlock8x8<T>, byte>(ref this);
+            ref byte imageStart = ref Unsafe.As<T, byte>(
+                                      ref Unsafe.Add(ref source.GetRowSpan(sourceY).DangerousGetPinnableReference(), sourceX)
+                                  );
+
+            int blockRowSizeInBytes = 8 * Unsafe.SizeOf<T>();
+            int imageRowSizeInBytes = source.Width * Unsafe.SizeOf<T>();
+
+            for (int y = 0; y < height; y++)
+            {
+                ref byte s = ref Unsafe.Add(ref imageStart, y * imageRowSizeInBytes);
+                ref byte d = ref Unsafe.Add(ref blockStart, y * blockRowSizeInBytes);
+
+                Unsafe.CopyBlock(ref d, ref s, byteWidth);
+
+                ref T last = ref Unsafe.Add(ref Unsafe.As<byte, T>(ref d), width - 1);
+
+                for (int x = 1; x <= remainderXCount; x++)
+                {
+                    Unsafe.Add(ref last, x) = last;
+                }
+            }
+
+            int remainderYCount = 8 - height;
+
+            if (remainderYCount == 0)
+            {
+                return;
+            }
+
+            ref byte lastRowStart = ref Unsafe.Add(ref blockStart, (height - 1) * blockRowSizeInBytes);
+
+            for (int y = 1; y <= remainderYCount; y++)
+            {
+                ref byte remStart = ref Unsafe.Add(ref lastRowStart, blockRowSizeInBytes * y);
+                Unsafe.CopyBlock(ref remStart, ref lastRowStart, (uint)blockRowSizeInBytes);
+            }
+        }
+
+        /// <summary>
+        /// ONLY FOR GenericBlock instances living on the stack!
+        /// </summary>
+        public Span<T> AsSpanUnsafe() => new Span<T>(Unsafe.AsPointer(ref this), Size);
+
+        /// <summary>
+        /// FOR TESTING ONLY!
+        /// Gets or sets a <see cref="Rgb24"/> value at the given index
+        /// </summary>
+        /// <param name="idx">The index</param>
+        /// <returns>The value</returns>
+        public T this[int idx]
+        {
+            get
+            {
+                ref T selfRef = ref Unsafe.As<GenericBlock8x8<T>, T>(ref this);
+                return Unsafe.Add(ref selfRef, idx);
+            }
+
+            set
+            {
+                ref T selfRef = ref Unsafe.As<GenericBlock8x8<T>, T>(ref this);
+                Unsafe.Add(ref selfRef, idx) = value;
+            }
+        }
+
+        /// <summary>
+        /// FOR TESTING ONLY!
+        /// Gets or sets a value in a row+coulumn of the 8x8 block
+        /// </summary>
+        /// <param name="x">The x position index in the row</param>
+        /// <param name="y">The column index</param>
+        /// <returns>The value</returns>
+        public T this[int x, int y]
+        {
+            get => this[(y * 8) + x];
+            set => this[(y * 8) + x] = value;
+        }
+    }
+}

src/ImageSharp/Formats/Jpeg/GolangPort/Components/Encoder/RgbToYCbCrTables.cs

@@ -8,6 +8,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.GolangPort.Components.Encoder
     /// <summary>
     /// Provides 8-bit lookup tables for converting from Rgb to YCbCr colorspace.
     /// Methods to build the tables are based on libjpeg implementation.
+    /// TODO: Replace this logic with SIMD conversion (similar to the one in the decoder)!
     /// </summary>
     internal unsafe struct RgbToYCbCrTables
     {
@@ -91,6 +92,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.GolangPort.Components.Encoder
         }
 
         /// <summary>
+        /// TODO: Replace this logic with SIMD conversion (similar to the one in the decoder)!
         /// Optimized method to allocates the correct y, cb, and cr values to the DCT blocks from the given r, g, b values.
         /// </summary>
         /// <param name="yBlockRaw">The The luminance block.</param>
@@ -102,7 +104,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.GolangPort.Components.Encoder
         /// <param name="g">The green value.</param>
         /// <param name="b">The blue value.</param>
         [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)
+        public static void Rgb2YCbCr(float* yBlockRaw, float* cbBlockRaw, 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;

src/ImageSharp/Formats/Jpeg/GolangPort/JpegEncoderCore.cs

@@ -331,7 +331,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.GolangPort
 
                     int index = j8 + i;
 
-                    RgbToYCbCrTables.Allocate(ref yBlockRaw, ref cbBlockRaw, ref crBlockRaw, ref tables, index, r, g, b);
+                    RgbToYCbCrTables.Rgb2YCbCr(yBlockRaw, cbBlockRaw, crBlockRaw, ref tables, index, r, g, b);
 
                     dataIdx += 3;
                 }

src/ImageSharp/Formats/Jpeg/GolangPort/Utils/OrigJpegUtils.cs

@@ -8,7 +8,7 @@ using SixLabors.ImageSharp.PixelFormats;
 namespace SixLabors.ImageSharp.Formats.Jpeg.GolangPort.Utils
 {
     /// <summary>
-    ///     Jpeg specific utilities and extension methods
+    /// Jpeg specific utilities and extension methods
     /// </summary>
     internal static class OrigJpegUtils
     {

src/ImageSharp/ImageSharp.csproj

@@ -62,6 +62,10 @@
       <Generator>TextTemplatingFileGenerator</Generator>
       <LastGenOutput>Block8x8F.Generated.cs</LastGenOutput>
     </None>
+    <None Update="Formats\Jpeg\Common\GenericBlock8x8.Generated.tt">
+      <Generator>TextTemplatingFileGenerator</Generator>
+      <LastGenOutput>GenericBlock8x8.Generated.cs</LastGenOutput>
+    </None>
     <None Update="Formats\Jpeg\Components\Block8x8F.Generated.tt">
       <Generator>TextTemplatingFileGenerator</Generator>
       <LastGenOutput>Block8x8F.Generated.cs</LastGenOutput>
@@ -92,6 +96,11 @@
       <AutoGen>True</AutoGen>
       <DependentUpon>Block8x8F.Generated.tt</DependentUpon>
     </Compile>
+    <Compile Update="Formats\Jpeg\Common\GenericBlock8x8.Generated.cs">
+      <DesignTime>True</DesignTime>
+      <AutoGen>True</AutoGen>
+      <DependentUpon>GenericBlock8x8.Generated.tt</DependentUpon>
+    </Compile>
     <Compile Update="Formats\Jpeg\Components\Block8x8F.Generated.cs">
       <DesignTime>True</DesignTime>
       <AutoGen>True</AutoGen>

tests/ImageSharp.Tests/Formats/Jpg/GenericBlock8x8Tests.cs

@@ -0,0 +1,126 @@
+// ReSharper disable InconsistentNaming
+namespace SixLabors.ImageSharp.Tests.Formats.Jpg
+{
+    using System;
+    using System.Numerics;
+
+    using SixLabors.ImageSharp.Formats.Jpeg.Common;
+    using SixLabors.ImageSharp.Helpers;
+    using SixLabors.ImageSharp.PixelFormats;
+    using SixLabors.Primitives;
+
+    using Xunit;
+
+    public class GenericBlock8x8Tests
+    {
+        public static Image<TPixel> CreateTestImage<TPixel>()
+            where TPixel : struct, IPixel<TPixel>
+        {
+            var image = new Image<TPixel>(10, 10);
+            using (PixelAccessor<TPixel> pixels = image.Lock())
+            {
+                for (int i = 0; i < 10; i++)
+                {
+                    for (int j = 0; j < 10; j++)
+                    {
+                        var rgba = new Rgba32((byte)(i+1), (byte)(j+1), (byte)200, (byte)255);
+                        var color = default(TPixel);
+                        color.PackFromRgba32(rgba);
+
+                        pixels[i, j] = color;
+                    }
+                }
+            }
+
+            return image;
+        }
+
+        [Theory]
+        [WithMemberFactory(nameof(CreateTestImage), PixelTypes.Rgb24 | PixelTypes.Rgba32 /* | PixelTypes.Rgba32 | PixelTypes.Argb32*/)]
+        public void LoadAndStretchCorners_FromOrigo<TPixel>(TestImageProvider<TPixel> provider)
+            where TPixel : struct, IPixel<TPixel>
+        {
+            using (Image<TPixel> s = provider.GetImage())
+            {
+                var d = default(GenericBlock8x8<TPixel>);
+                d.LoadAndStretchEdges(s.Frames.RootFrame, 0, 0);
+
+                TPixel a = s.Frames.RootFrame[0, 0];
+                TPixel b = d[0, 0];
+
+                Assert.Equal(s[0, 0], d[0, 0]);
+                Assert.Equal(s[1, 0], d[1, 0]);
+                Assert.Equal(s[7, 0], d[7, 0]);
+                Assert.Equal(s[0, 1], d[0, 1]);
+                Assert.Equal(s[1, 1], d[1, 1]);
+                Assert.Equal(s[7, 0], d[7, 0]);
+                Assert.Equal(s[0, 7], d[0, 7]);
+                Assert.Equal(s[7, 7], d[7, 7]);
+            }
+        }
+
+        [Theory]
+        [WithMemberFactory(nameof(CreateTestImage), PixelTypes.Rgb24 | PixelTypes.Rgba32)]
+        public unsafe void LoadAndStretchCorners_WithOffset<TPixel>(TestImageProvider<TPixel> provider)
+            where TPixel : struct, IPixel<TPixel>
+        {
+            using (Image<TPixel> s = provider.GetImage())
+            {
+                var d = default(GenericBlock8x8<TPixel>);
+                d.LoadAndStretchEdges(s.Frames.RootFrame, 6, 7);
+
+                Assert.Equal(s[6, 7], d[0, 0]);
+                Assert.Equal(s[6, 8], d[0, 1]);
+                Assert.Equal(s[7, 8], d[1, 1]);
+
+                Assert.Equal(s[6, 9], d[0, 2]);
+                Assert.Equal(s[6, 9], d[0, 3]);
+                Assert.Equal(s[6, 9], d[0, 7]);
+
+                Assert.Equal(s[7, 9], d[1, 2]);
+                Assert.Equal(s[7, 9], d[1, 3]);
+                Assert.Equal(s[7, 9], d[1, 7]);
+
+                Assert.Equal(s[9, 9], d[3, 2]);
+                Assert.Equal(s[9, 9], d[3, 3]);
+                Assert.Equal(s[9, 9], d[3, 7]);
+
+                Assert.Equal(s[9, 7], d[3, 0]);
+                Assert.Equal(s[9, 7], d[4, 0]);
+                Assert.Equal(s[9, 7], d[7, 0]);
+
+                Assert.Equal(s[9, 9], d[3, 2]);
+                Assert.Equal(s[9, 9], d[4, 2]);
+                Assert.Equal(s[9, 9], d[7, 2]);
+
+                Assert.Equal(s[9, 9], d[4, 3]);
+                Assert.Equal(s[9, 9], d[7, 7]);
+            }
+        }
+
+        [Fact]
+        public void Indexer()
+        {
+            var block = default(GenericBlock8x8<Rgb24>);
+            Span<Rgb24> span = block.AsSpanUnsafe();
+            Assert.Equal(64, span.Length);
+
+            for (int i = 0; i < 64; i++)
+            {
+                span[i] = new Rgb24((byte)i, (byte)(2 * i), (byte)(3 * i));
+            }
+
+            Rgb24 expected00 = new Rgb24(0, 0, 0);
+            Rgb24 expected07 = new Rgb24(7, 14, 21);
+            Rgb24 expected11 = new Rgb24(9, 18, 27);
+            Rgb24 expected77 = new Rgb24(63, 126, 189);
+            Rgb24 expected67 = new Rgb24(62, 124, 186);
+
+            Assert.Equal(expected00, block[0, 0]);
+            Assert.Equal(expected07, block[7, 0]);
+            Assert.Equal(expected11, block[1, 1]);
+            Assert.Equal(expected67, block[6, 7]);
+            Assert.Equal(expected77, block[7, 7]);
+        }
+    }
+}

tests/ImageSharp.Tests/Formats/Jpg/JpegUtilsTests.cs

@@ -1,18 +1,14 @@
 // Copyright (c) Six Labors and contributors.
 // Licensed under the Apache License, Version 2.0.
 
-
+using System.Numerics;
+using SixLabors.ImageSharp.Formats.Jpeg.GolangPort.Utils;
+using SixLabors.ImageSharp.PixelFormats;
+using Xunit;
 
 // ReSharper disable InconsistentNaming
 namespace SixLabors.ImageSharp.Tests.Formats.Jpg
 {
-    using System.Numerics;
-
-    using SixLabors.ImageSharp.Formats.Jpeg.GolangPort.Utils;
-    using SixLabors.ImageSharp.PixelFormats;
-
-    using Xunit;
-
     public class JpegUtilsTests
     {
         public static Image<TPixel> CreateTestImage<TPixel>()