add benchmark results and fix PixelOperations

src/ImageSharp/PixelFormats/PixelImplementations/PixelOperations/Rgb24.PixelOperations.cs

@@ -30,6 +30,7 @@ namespace SixLabors.ImageSharp.PixelFormats
                 ReadOnlySpan<byte> blueChannel,
                 Span<Rgb24> destination)
             {
+                Guard.NotNull(configuration, nameof(configuration));
                 int count = redChannel.Length;
                 Guard.IsTrue(greenChannel.Length == count, nameof(greenChannel), "Channels must be of same size!");
                 Guard.IsTrue(blueChannel.Length == count, nameof(blueChannel), "Channels must be of same size!");

src/ImageSharp/PixelFormats/PixelImplementations/PixelOperations/Rgba32.PixelOperations.cs

@@ -65,6 +65,7 @@ namespace SixLabors.ImageSharp.PixelFormats
                 ReadOnlySpan<byte> blueChannel,
                 Span<Rgba32> destination)
             {
+                Guard.NotNull(configuration, nameof(configuration));
                 int count = redChannel.Length;
                 Guard.IsTrue(greenChannel.Length == count, nameof(greenChannel), "Channels must be of same size!");
                 Guard.IsTrue(blueChannel.Length == count, nameof(blueChannel), "Channels must be of same size!");

src/ImageSharp/PixelFormats/PixelOperations{TPixel}.cs

@@ -4,6 +4,8 @@
 using System;
 using System.Buffers;
 using System.Numerics;
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
 using SixLabors.ImageSharp.Formats;
 using SixLabors.ImageSharp.Memory;
 
@@ -177,13 +179,26 @@ namespace SixLabors.ImageSharp.PixelFormats
             Span<TPixel> destination)
         {
             Guard.NotNull(configuration, nameof(configuration));
+
+            int count = redChannel.Length;
+            Guard.IsTrue(greenChannel.Length == count, nameof(greenChannel), "Channels must be of same size!");
+            Guard.IsTrue(blueChannel.Length == count, nameof(blueChannel), "Channels must be of same size!");
+            Guard.IsTrue(destination.Length > count + 2, nameof(destination), "'destination' must contain a padding of 3 elements!");
+
             Guard.DestinationShouldNotBeTooShort(redChannel, destination, nameof(destination));
 
-            for (int i = 0; i < destination.Length; i++)
-            {
-                var rgb24 = new Rgb24(redChannel[i], greenChannel[i], blueChannel[i]);
+            Rgb24 rgb24 = default;
+            ref byte r = ref MemoryMarshal.GetReference(redChannel);
+            ref byte g = ref MemoryMarshal.GetReference(greenChannel);
+            ref byte b = ref MemoryMarshal.GetReference(blueChannel);
+            ref TPixel d = ref MemoryMarshal.GetReference(destination);
 
-                destination[i].FromRgb24(rgb24);
+            for (int i = 0; i < count; i++)
+            {
+                rgb24.R = Unsafe.Add(ref r, i);
+                rgb24.G = Unsafe.Add(ref g, i);
+                rgb24.B = Unsafe.Add(ref b, i);
+                Unsafe.Add(ref d, i).FromRgb24(rgb24);
             }
         }
     }

tests/ImageSharp.Benchmarks/General/PixelConversion/PixelConversion_PackFromRgbPlanes.cs

@@ -268,5 +268,19 @@ namespace SixLabors.ImageSharp.Benchmarks.General.PixelConversion
             public byte V0, V1, V2, V3;
         }
 #pragma warning restore
+
+        // Results @ Anton's PC, 2020 Dec 05
+        // .NET Core 3.1.1
+        // Intel Core i7-7700HQ CPU 2.80GHz (Kaby Lake), 1 CPU, 8 logical and 4 physical cores
+        //
+        // |                           Method | Count |       Mean |    Error |   StdDev | Ratio | RatioSD |
+        // |--------------------------------- |------ |-----------:|---------:|---------:|------:|--------:|
+        // |     Rgb24_Scalar_PerElement_Span |  1024 | 1,634.6 ns | 26.56 ns | 24.84 ns |  3.12 |    0.05 |
+        // |   Rgb24_Scalar_PerElement_Unsafe |  1024 | 1,284.7 ns |  4.70 ns |  4.16 ns |  2.46 |    0.01 |
+        // | Rgb24_Scalar_PerElement_Batched8 |  1024 | 1,182.3 ns |  5.12 ns |  4.27 ns |  2.26 |    0.01 |
+        // | Rgb24_Scalar_PerElement_Batched4 |  1024 | 1,146.2 ns | 16.38 ns | 14.52 ns |  2.19 |    0.02 |
+        // |                Rgba32_Avx2_Float |  1024 |   522.7 ns |  1.78 ns |  1.39 ns |  1.00 |    0.00 |
+        // |                Rgba24_Avx2_Bytes |  1024 |   243.3 ns |  1.56 ns |  1.30 ns |  0.47 |    0.00 |
+        // |                Rgba32_Avx2_Bytes |  1024 |   146.0 ns |  2.48 ns |  2.32 ns |  0.28 |    0.01 |
     }
 }

tests/ImageSharp.Tests/Common/SimdUtilsTests.cs

@@ -443,10 +443,10 @@ namespace SixLabors.ImageSharp.Tests.Common
                 expected[i].FromRgb24(new Rgb24(r[i], g[i], b[i]));
             }
 
-            TPixel[] actual = new TPixel[count];
+            TPixel[] actual = new TPixel[count + 3]; // padding for Rgb24 AVX2
             packMethod(r, g, b, actual);
 
-            Assert.Equal(expected, actual);
+            Assert.True(expected.AsSpan().SequenceEqual(actual.AsSpan().Slice(0, count)));
         }
 
         private static void TestImpl_BulkConvertNormalizedFloatToByteClampOverflows(