JpegColorConverter.FromYCbCrSimdAvx2 works!

src/ImageSharp/Common/Tuples/Vector4Pair.cs

@@ -71,5 +71,10 @@ namespace SixLabors.ImageSharp.Common.Tuples
             v *= new Vector<float>(1 / 255f);
             self = v;
         }
+
+        public override string ToString()
+        {
+            return $"{this.A}, {this.B}";
+        }
     }
 }

src/ImageSharp/Formats/Jpeg/Common/Block8x8.cs

@@ -1,3 +1,6 @@
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
 using System;
 using System.Diagnostics;
 using System.Runtime.CompilerServices;

src/ImageSharp/Formats/Jpeg/Common/Block8x8F.CopyTo.cs

@@ -1,4 +1,7 @@
-using System.Numerics;
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System.Numerics;
 using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Memory;
 

src/ImageSharp/Formats/Jpeg/Common/Decoder/ColorConverters/JpegColorConverter.FromCmyk.cs

@@ -1,4 +1,7 @@
-using System;
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
 using System.Numerics;
 
 namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters

src/ImageSharp/Formats/Jpeg/Common/Decoder/ColorConverters/JpegColorConverter.FromGrayScale.cs

@@ -1,4 +1,7 @@
-using System;
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
 using System.Numerics;
 
 namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters

src/ImageSharp/Formats/Jpeg/Common/Decoder/ColorConverters/JpegColorConverter.FromRgb.cs

@@ -1,4 +1,7 @@
-using System;
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
 using System.Numerics;
 
 namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters

src/ImageSharp/Formats/Jpeg/Common/Decoder/ColorConverters/JpegColorConverter.FromYCbCrBasic.cs

@@ -1,4 +1,7 @@
-using System;
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
 using System.Numerics;
 
 namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters

src/ImageSharp/Formats/Jpeg/Common/Decoder/ColorConverters/JpegColorConverter.FromYCbCrSimd.cs

@@ -1,4 +1,7 @@
-using System;
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
 using System.Numerics;
 using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Common.Tuples;

src/ImageSharp/Formats/Jpeg/Common/Decoder/ColorConverters/JpegColorConverter.FromYCbCr.cs → src/ImageSharp/Formats/Jpeg/Common/Decoder/ColorConverters/JpegColorConverter.FromYCbCrSimdAvx2.cs

@@ -1,8 +1,12 @@
-using System;
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
 using System.Numerics;
 using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Common.Tuples;
 
+// ReSharper disable ImpureMethodCallOnReadonlyValueField
 namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters
 {
     internal abstract partial class JpegColorConverter
@@ -14,7 +18,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters
             {
             }
 
-            public static bool IsAvailable => Vector.IsHardwareAccelerated && Vector<float>.Count == 8;
+            public static bool IsAvailable => Vector.IsHardwareAccelerated && SimdUtils.IsAvx2CompatibleArchitecture;
 
             public override void ConvertToRGBA(ComponentValues values, Span<Vector4> result)
             {
@@ -56,6 +60,16 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters
                 // Walking 8 elements at one step:
                 int n = result.Length / 8;
 
+                var rr = default(Vector4Pair);
+                var gg = default(Vector4Pair);
+                var bb = default(Vector4Pair);
+
+                ref Vector<float> rrRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref rr);
+                ref Vector<float> ggRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref gg);
+                ref Vector<float> bbRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref bb);
+
+                var scale = new Vector<float>(1 / 255f);
+
                 for (int i = 0; i < n; i++)
                 {
                     // y = yVals[i];
@@ -73,15 +87,16 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters
                     Vector<float> g = y - (cb * new Vector<float>(0.344136F)) - (cr * new Vector<float>(0.714136F));
                     Vector<float> b = y + (cb * new Vector<float>(1.772F));
 
-                    // Vector<float> has no .Clamp(), need to switch to Vector4 for the next operation:
-                    // TODO: Is it worth to use Vector<float> at all?
-                    Vector4Pair rr = Unsafe.As<Vector<float>, Vector4Pair>(ref r);
-                    Vector4Pair gg = Unsafe.As<Vector<float>, Vector4Pair>(ref g);
-                    Vector4Pair bb = Unsafe.As<Vector<float>, Vector4Pair>(ref b);
+                    r = r.FastRound();
+                    g = g.FastRound();
+                    b = b.FastRound();
+                    r *= scale;
+                    g *= scale;
+                    b *= scale;
 
-                    rr.RoundAndDownscaleAvx2();
-                    gg.RoundAndDownscaleAvx2();
-                    bb.RoundAndDownscaleAvx2();
+                    rrRefAsVector = r;
+                    ggRefAsVector = g;
+                    bbRefAsVector = b;
 
                     // Collect (r0,r1...r8) (g0,g1...g8) (b0,b1...b8) vector values in the expected (r0,g0,g1,1), (r1,g1,g2,1) ... order:
                     ref Vector4Octet destination = ref Unsafe.Add(ref resultBase, i);

src/ImageSharp/Formats/Jpeg/Common/Decoder/ColorConverters/JpegColorConverter.FromYccK.cs

@@ -1,11 +1,14 @@
-using System;
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
 using System.Numerics;
 
 namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters
 {
     internal abstract partial class JpegColorConverter
     {
-        internal class FromYccK : ColorConverters.JpegColorConverter
+        internal class FromYccK : JpegColorConverter
         {
             public FromYccK()
                 : base(JpegColorSpace.Ycck)

src/ImageSharp/Formats/Jpeg/Common/Decoder/ColorConverters/JpegColorConverter.cs

@@ -1,3 +1,6 @@
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
 using System;
 using System.Collections.Generic;
 using System.Linq;
@@ -17,7 +20,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters
         /// </summary>
         private static readonly JpegColorConverter[] Converters =
             {
-                new FromYCbCrSimd(), new FromYccK(), new FromCmyk(), new FromGrayScale(), new FromRgb()
+                GetYCbCrConverter(), new FromYccK(), new FromCmyk(), new FromGrayScale(), new FromRgb()
             };
 
         /// <summary>
@@ -54,6 +57,12 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters
         /// <param name="result">The destination buffer of <see cref="Vector4"/> values</param>
         public abstract void ConvertToRGBA(ComponentValues values, Span<Vector4> result);
 
+        /// <summary>
+        /// Returns the <see cref="JpegColorConverter"/> for the YCbCr colorspace that matches the current CPU architecture.
+        /// </summary>
+        private static JpegColorConverter GetYCbCrConverter() =>
+            FromYCbCrSimdAvx2.IsAvailable ? (JpegColorConverter)new FromYCbCrSimdAvx2() : new FromYCbCrSimd();
+
         /// <summary>
         /// A stack-only struct to reference the input buffers using <see cref="ReadOnlySpan{T}"/>-s.
         /// </summary>

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

@@ -1,3 +1,6 @@
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
 using System.Numerics;
 using SixLabors.Primitives;
 

tests/ImageSharp.Benchmarks/Color/Bulk/PackFromVector4.cs

@@ -8,7 +8,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Color.Bulk
     using SixLabors.ImageSharp.Memory;
     using SixLabors.ImageSharp.PixelFormats;
 
-    [Config(typeof(Config.Short))]
+    [Config(typeof(Config.ShortClr))]
     public abstract class PackFromVector4<TPixel>
         where TPixel : struct, IPixel<TPixel>
     {

tests/ImageSharp.Benchmarks/Config.cs

@@ -17,9 +17,9 @@ namespace SixLabors.ImageSharp.Benchmarks
             this.Add(new BenchmarkDotNet.Diagnosers.MemoryDiagnoser());
         }
 
-        public class Short : Config
+        public class ShortClr : Config
         {
-            public Short()
+            public ShortClr()
             {
                 this.Add(
                     Job.Clr.WithLaunchCount(1)

tests/ImageSharp.Benchmarks/General/ArrayCopy.cs

@@ -10,7 +10,7 @@ namespace SixLabors.ImageSharp.Benchmarks.General
 
     using BenchmarkDotNet.Attributes;
     
-    [Config(typeof(Config.Short))]
+    [Config(typeof(Config.ShortClr))]
     public class ArrayCopy
     {
         [Params(10, 100, 1000, 10000)]

tests/ImageSharp.Benchmarks/Image/EncodeBmpMultiple.cs

@@ -13,7 +13,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Image
     using SixLabors.ImageSharp.Formats;
     using SixLabors.ImageSharp.Formats.Bmp;
 
-    [Config(typeof(Config.Short))]
+    [Config(typeof(Config.ShortClr))]
     public class EncodeBmpMultiple : MultiImageBenchmarkBase.WithImagesPreloaded
     {
         protected override IEnumerable<string> InputImageSubfoldersOrFiles => new[] { "Bmp/", "Jpg/baseline" };

tests/ImageSharp.Benchmarks/Image/Jpeg/DecodeJpeg.cs

@@ -18,7 +18,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Image.Jpeg
 
     using CoreSize = SixLabors.Primitives.Size;
     
-    [Config(typeof(Config.Short))]
+    [Config(typeof(Config.ShortClr))]
     public class DecodeJpeg : BenchmarkBase
     {
         private byte[] jpegBytes;

tests/ImageSharp.Benchmarks/Image/Jpeg/DecodeJpegMultiple.cs

@@ -11,7 +11,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Image.Jpeg
 
     using CoreImage = SixLabors.ImageSharp.Image;
 
-    [Config(typeof(Config.Short))]
+    [Config(typeof(Config.ShortClr))]
     public class DecodeJpegMultiple : MultiImageBenchmarkBase
     {
         protected override IEnumerable<string> InputImageSubfoldersOrFiles => new[]

tests/ImageSharp.Benchmarks/Image/Jpeg/EncodeJpegMultiple.cs

@@ -12,7 +12,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Image.Jpeg
 
     using SixLabors.ImageSharp.Formats.Jpeg;
 
-    [Config(typeof(Config.Short))] // It's long enough to iterate through multiple files
+    [Config(typeof(Config.ShortClr))] // It's long enough to iterate through multiple files
     public class EncodeJpegMultiple : MultiImageBenchmarkBase.WithImagesPreloaded
     {
         protected override IEnumerable<string> InputImageSubfoldersOrFiles => new[] { "Bmp/", "Jpg/baseline" };

tests/ImageSharp.Benchmarks/Image/Jpeg/YCbCrColorConversion.cs

@@ -1,15 +1,64 @@
 namespace SixLabors.ImageSharp.Benchmarks.Image.Jpeg
 {
     using System;
+    using System.Numerics;
+
+    using BenchmarkDotNet.Attributes;
 
     using SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder;
+    using SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters;
     using SixLabors.ImageSharp.Memory;
-
-    using JpegColorConverter = SixLabors.ImageSharp.Formats.Jpeg.Common.Decoder.ColorConverters.JpegColorConverter;
-
+    
+    [Config(typeof(Config.ShortClr))]
     public class YCbCrColorConversion
     {
-        private static JpegColorConverter.ComponentValues CreateRandomValues(
+        private Buffer2D<float>[] input;
+
+        private Vector4[] output;
+
+        public const int Count = 64;
+
+        [GlobalSetup]
+        public void Setup()
+        {
+            this.input = CreateRandomValues(3, Count);
+            this.output = new Vector4[Count];
+        }
+
+        [GlobalCleanup]
+        public void Cleanup()
+        {
+            foreach (Buffer2D<float> buffer in this.input)
+            {
+                buffer.Dispose();
+            }
+        }
+
+        [Benchmark(Baseline = true)]
+        public void Scalar()
+        {
+            var values = new JpegColorConverter.ComponentValues(this.input, 0);
+
+            JpegColorConverter.FromYCbCrBasic.ConvertCore(values, this.output);
+        }
+
+        [Benchmark]
+        public void SimdVector4()
+        {
+            var values = new JpegColorConverter.ComponentValues(this.input, 0);
+
+            JpegColorConverter.FromYCbCrSimd.ConvertCore(values, this.output);
+        }
+
+        [Benchmark]
+        public void SimdAvx2()
+        {
+            var values = new JpegColorConverter.ComponentValues(this.input, 0);
+
+            JpegColorConverter.FromYCbCrSimdAvx2.ConvertCore(values, this.output);
+        }
+        
+        private static Buffer2D<float>[] CreateRandomValues(
             int componentCount,
             int inputBufferLength,
             float minVal = 0f,
@@ -29,8 +78,9 @@
                 // no need to dispose when buffer is not array owner
                 buffers[i] = new Buffer2D<float>(values, values.Length, 1);
             }
-            return new JpegColorConverter.ComponentValues(buffers, 0);
-        }
 
+            return buffers;
+        }
+        
     }
 }

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

@@ -89,6 +89,8 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
                 return;
             }
 
+            //JpegColorConverter.FromYCbCrSimdAvx2.LogPlz = s => this.Output.WriteLine(s);
+
             ValidateConversion(
                 new JpegColorConverter.FromYCbCrSimdAvx2(),
                 3,