Improved performance of Sum on Images

HPPH.Benchmark/BenchmarkHelper.cs

@@ -21,4 +21,18 @@ internal static class BenchmarkHelper
 
         return colors;
     }
+
+    public static List<IImage<T>> GetSampleDataImages<T>()
+        where T : struct, IColor
+    {
+        if (!Directory.Exists(SAMPLE_DATA_DIR)) throw new Exception("sample data not found!");
+
+        List<IImage<T>> colors = [];
+
+        IEnumerable<string> files = Directory.EnumerateFiles(SAMPLE_DATA_DIR, "*.png", SearchOption.AllDirectories);
+        foreach (string file in files)
+            colors.Add(ImageHelper.LoadImage(file).ConvertTo<T>());
+
+        return colors;
+    }
 }

HPPH.Benchmark/SumBenchmarks.cs

@@ -13,6 +13,8 @@ public class SumBenchmarks
 
     private readonly List<ColorRGB[]> _colors3bpp;
     private readonly List<ColorRGBA[]> _colors4bpp;
+    private readonly List<IImage<ColorRGB>> _images3bpp;
+    private readonly List<IImage<ColorRGBA>> _images4bpp;
 
     #endregion
 
@@ -22,6 +24,9 @@ public class SumBenchmarks
     {
         _colors3bpp = BenchmarkHelper.GetSampleData<ColorRGB>();
         _colors4bpp = BenchmarkHelper.GetSampleData<ColorRGBA>();
+
+        _images3bpp = BenchmarkHelper.GetSampleDataImages<ColorRGB>();
+        _images4bpp = BenchmarkHelper.GetSampleDataImages<ColorRGBA>();
     }
 
     #endregion
@@ -48,6 +53,26 @@ public class SumBenchmarks
         return sums;
     }
 
+    [Benchmark]
+    public ISum[] PixelHelper_3BPP_Image()
+    {
+        ISum[] sums = new ISum[_colors3bpp.Count];
+        for (int i = 0; i < _images3bpp.Count; i++)
+            sums[i] = _images3bpp[i].Sum();
+
+        return sums;
+    }
+
+    [Benchmark]
+    public ISum[] PixelHelper_4BPP_Image()
+    {
+        ISum[] sums = new ISum[_images4bpp.Count];
+        for (int i = 0; i < _images4bpp.Count; i++)
+            sums[i] = _images4bpp[i].Sum();
+
+        return sums;
+    }
+
     [Benchmark]
     public ISum[] Reference_3BPP()
     {

HPPH.Generators/Features/Sum.cs

@@ -106,6 +106,7 @@ internal class Sum : IGeneratorFeature
         return $$"""
                  #nullable enable
                  
+                 using System.Runtime.CompilerServices;
                  using System.Runtime.InteropServices;
                  
                  namespace HPPH;
@@ -114,7 +115,8 @@ internal class Sum : IGeneratorFeature
                  {
                      #region Methods
                  
-                     unsafe ISum IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum<Color{{colorFormat.Format}}, Sum{{colorFormat.Format}}>(MemoryMarshal.Cast<byte, Color{{colorFormat.Format}}>(data));
+                     unsafe Generic4LongData IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum(MemoryMarshal.Cast<byte, Generic{{colorFormat.Bpp}}ByteData>(data));
+                     unsafe ISum IColorFormat.ToSum(Generic4LongData data) => Unsafe.BitCast<Generic4LongData, Sum{{colorFormat.Format}}>(data);
                  
                      #endregion
                  }
@@ -130,7 +132,8 @@ internal class Sum : IGeneratorFeature
                
                public partial interface IColorFormat
                {
-                   internal ISum Sum(ReadOnlySpan<byte> data);
+                   internal Generic4LongData Sum(ReadOnlySpan<byte> data);
+                   internal ISum ToSum(Generic4LongData data);
                }
                """;
     }

HPPH/Generated/HPPH.Generators/HPPH.Generators.ColorSourceGenerator/ColorFormatABGR.Sum.g.cs

@@ -1,5 +1,6 @@
 #nullable enable
 
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 
 namespace HPPH;
@@ -8,7 +9,8 @@ public sealed partial class ColorFormatABGR
 {
     #region Methods
 
-    unsafe ISum IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum<ColorABGR, SumABGR>(MemoryMarshal.Cast<byte, ColorABGR>(data));
+    unsafe Generic4LongData IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum(MemoryMarshal.Cast<byte, Generic4ByteData>(data));
+    unsafe ISum IColorFormat.ToSum(Generic4LongData data) => Unsafe.BitCast<Generic4LongData, SumABGR>(data);
 
     #endregion
 }

HPPH/Generated/HPPH.Generators/HPPH.Generators.ColorSourceGenerator/ColorFormatARGB.Sum.g.cs

@@ -1,5 +1,6 @@
 #nullable enable
 
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 
 namespace HPPH;
@@ -8,7 +9,8 @@ public sealed partial class ColorFormatARGB
 {
     #region Methods
 
-    unsafe ISum IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum<ColorARGB, SumARGB>(MemoryMarshal.Cast<byte, ColorARGB>(data));
+    unsafe Generic4LongData IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum(MemoryMarshal.Cast<byte, Generic4ByteData>(data));
+    unsafe ISum IColorFormat.ToSum(Generic4LongData data) => Unsafe.BitCast<Generic4LongData, SumARGB>(data);
 
     #endregion
 }

HPPH/Generated/HPPH.Generators/HPPH.Generators.ColorSourceGenerator/ColorFormatBGR.Sum.g.cs

@@ -1,5 +1,6 @@
 #nullable enable
 
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 
 namespace HPPH;
@@ -8,7 +9,8 @@ public sealed partial class ColorFormatBGR
 {
     #region Methods
 
-    unsafe ISum IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum<ColorBGR, SumBGR>(MemoryMarshal.Cast<byte, ColorBGR>(data));
+    unsafe Generic4LongData IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum(MemoryMarshal.Cast<byte, Generic3ByteData>(data));
+    unsafe ISum IColorFormat.ToSum(Generic4LongData data) => Unsafe.BitCast<Generic4LongData, SumBGR>(data);
 
     #endregion
 }

HPPH/Generated/HPPH.Generators/HPPH.Generators.ColorSourceGenerator/ColorFormatBGRA.Sum.g.cs

@@ -1,5 +1,6 @@
 #nullable enable
 
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 
 namespace HPPH;
@@ -8,7 +9,8 @@ public sealed partial class ColorFormatBGRA
 {
     #region Methods
 
-    unsafe ISum IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum<ColorBGRA, SumBGRA>(MemoryMarshal.Cast<byte, ColorBGRA>(data));
+    unsafe Generic4LongData IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum(MemoryMarshal.Cast<byte, Generic4ByteData>(data));
+    unsafe ISum IColorFormat.ToSum(Generic4LongData data) => Unsafe.BitCast<Generic4LongData, SumBGRA>(data);
 
     #endregion
 }

HPPH/Generated/HPPH.Generators/HPPH.Generators.ColorSourceGenerator/ColorFormatRGB.Sum.g.cs

@@ -1,5 +1,6 @@
 #nullable enable
 
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 
 namespace HPPH;
@@ -8,7 +9,8 @@ public sealed partial class ColorFormatRGB
 {
     #region Methods
 
-    unsafe ISum IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum<ColorRGB, SumRGB>(MemoryMarshal.Cast<byte, ColorRGB>(data));
+    unsafe Generic4LongData IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum(MemoryMarshal.Cast<byte, Generic3ByteData>(data));
+    unsafe ISum IColorFormat.ToSum(Generic4LongData data) => Unsafe.BitCast<Generic4LongData, SumRGB>(data);
 
     #endregion
 }

HPPH/Generated/HPPH.Generators/HPPH.Generators.ColorSourceGenerator/ColorFormatRGBA.Sum.g.cs

@@ -1,5 +1,6 @@
 #nullable enable
 
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 
 namespace HPPH;
@@ -8,7 +9,8 @@ public sealed partial class ColorFormatRGBA
 {
     #region Methods
 
-    unsafe ISum IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum<ColorRGBA, SumRGBA>(MemoryMarshal.Cast<byte, ColorRGBA>(data));
+    unsafe Generic4LongData IColorFormat.Sum(ReadOnlySpan<byte> data) => PixelHelper.Sum(MemoryMarshal.Cast<byte, Generic4ByteData>(data));
+    unsafe ISum IColorFormat.ToSum(Generic4LongData data) => Unsafe.BitCast<Generic4LongData, SumRGBA>(data);
 
     #endregion
 }

HPPH/Generated/HPPH.Generators/HPPH.Generators.ColorSourceGenerator/IColorFormat.Sum.g.cs

@@ -4,5 +4,6 @@ namespace HPPH;
 
 public partial interface IColorFormat
 {
-    internal ISum Sum(ReadOnlySpan<byte> data);
+    internal Generic4LongData Sum(ReadOnlySpan<byte> data);
+    internal ISum ToSum(Generic4LongData data);
 }

HPPH/Images/ImageRow.cs

@@ -156,6 +156,9 @@ internal sealed class IColorImageRow<T> : IImageRow
 
     #region Methods
 
+    /// <inheritdoc />
+    public ReadOnlySpan<byte> AsByteSpan() => _buffer.AsSpan().Slice(_start, SizeInBytes);
+
     /// <inheritdoc />
     public void CopyTo(Span<IColor> destination)
     {

HPPH/Images/Interfaces/IImageRow.cs

@@ -22,6 +22,8 @@ public interface IImageRow : IEnumerable<IColor>
     /// <returns>The <see cref="IColor"/> at the specified location.</returns>
     IColor this[int x] { get; }
 
+    ReadOnlySpan<byte> AsByteSpan();
+
     void CopyTo(Span<IColor> destination);
 
     /// <summary>

HPPH/PixelHelper.Quantize.cs

@@ -245,8 +245,6 @@ public static partial class PixelHelper
         ColorCube<T>[] cubes = new ColorCube<T>[1 << splits];
         cubes[0] = new ColorCube<T>(0, colors.Length, SortTarget.None);
 
-        ParallelOptions parallelOptions = new() { MaxDegreeOfParallelism = Environment.ProcessorCount };
-
         int colorsLength = colors.Length;
         fixed (T* colorsPtr = colors)
         {
@@ -257,7 +255,7 @@ public static partial class PixelHelper
             {
                 int currentCubeCount = 1 << i;
 
-                Parallel.For(0, currentCubeCount, parallelOptions, CreateCubes);
+                Parallel.For(0, currentCubeCount, PARALLEL_OPTIONS, CreateCubes);
 
                 void CreateCubes(int index)
                 {

HPPH/PixelHelper.Sum.cs

@@ -1,7 +1,6 @@
 using System.Runtime.InteropServices;
 using System.Runtime.Intrinsics.X86;
 using System.Runtime.Intrinsics;
-using System.Buffers;
 using System.Runtime.CompilerServices;
 
 namespace HPPH;
@@ -14,29 +13,20 @@ public static unsafe partial class PixelHelper
     {
         ArgumentNullException.ThrowIfNull(image);
 
-        int dataLength = image.SizeInBytes;
+        IColorFormat colorFormat = image.ColorFormat;
 
-        if (dataLength <= 1024)
+        if (image.Height == 0) return colorFormat.ToSum(new Generic4LongData(0, 0, 0, 0));
-        {
+        if (image.Height == 1) return colorFormat.ToSum(colorFormat.Sum(image.Rows[0].AsByteSpan()));
-            Span<byte> buffer = stackalloc byte[dataLength];
 
-            image.CopyTo(buffer);
+        Vector256<long> result = Vector256<long>.Zero;
-            return image.ColorFormat.Sum(buffer);
+        for (int y = 0; y < image.Height; y++)
-        }
-        else
         {
-            byte[] array = ArrayPool<byte>.Shared.Rent(dataLength);
+            Generic4LongData rowSum = colorFormat.Sum(image.Rows[y].AsByteSpan());
-            Span<byte> buffer = array.AsSpan()[..dataLength];
+            Vector256<long> rowSumVector = Vector256.LoadUnsafe(ref Unsafe.As<Generic4LongData, long>(ref rowSum));
-            try
+            result = Vector256.Add(result, rowSumVector);
-            {
-                image.CopyTo(buffer);
-                return image.ColorFormat.Sum(buffer);
-            }
-            finally
-            {
-                ArrayPool<byte>.Shared.Return(array);
-            }
         }
+
+        return colorFormat.ToSum(Unsafe.BitCast<Vector256<long>, Generic4LongData>(result));
     }
 
     public static ISum Sum<T>(this IImage<T> image)
@@ -50,56 +40,37 @@ public static unsafe partial class PixelHelper
     public static ISum Sum<T>(this RefImage<T> image)
         where T : struct, IColor
     {
-        int dataLength = image.Width * image.Height;
+        IColorFormat colorFormat = T.ColorFormat;
-        int sizeInBytes = dataLength * T.ColorFormat.BytesPerPixel;
 
-        if (sizeInBytes <= 1024)
+        if (image.Height == 0) return colorFormat.ToSum(new Generic4LongData(0, 0, 0, 0));
-        {
+        if (image.Height == 1) return colorFormat.ToSum(colorFormat.Sum(image.Rows[0].AsByteSpan()));
-            Span<T> buffer = MemoryMarshal.Cast<byte, T>(stackalloc byte[sizeInBytes]);
 
-            image.CopyTo(buffer);
+        Vector256<long> result = Vector256<long>.Zero;
-            return Sum(buffer);
+        for (int y = 0; y < image.Height; y++)
-        }
-        else
         {
-            T[] array = ArrayPool<T>.Shared.Rent(dataLength);
+            Generic4LongData rowSum = colorFormat.Sum(image.Rows[y].AsByteSpan());
-            Span<T> buffer = array.AsSpan()[..(dataLength)];
+            Vector256<long> rowSumVector = Vector256.LoadUnsafe(ref Unsafe.As<Generic4LongData, long>(ref rowSum));
-            try
+            result = Vector256.Add(result, rowSumVector);
-            {
-                image.CopyTo(buffer);
-                return Sum(buffer);
-            }
-            finally
-            {
-                ArrayPool<T>.Shared.Return(array);
-            }
         }
+
+        return colorFormat.ToSum(Unsafe.BitCast<Vector256<long>, Generic4LongData>(result));
     }
 
     public static ISum Sum<T>(this ReadOnlySpan<T> colors)
         where T : struct, IColor
-        => T.ColorFormat.Sum(MemoryMarshal.AsBytes(colors));
+    {
+        IColorFormat colorFormat = T.ColorFormat;
+        return colorFormat.ToSum(colorFormat.Sum(MemoryMarshal.AsBytes(colors)));
+    }
 
     public static ISum Sum<T>(this Span<T> colors)
         where T : struct, IColor
-        => T.ColorFormat.Sum(MemoryMarshal.AsBytes(colors));
-
-    internal static ISum Sum<T, TSum>(ReadOnlySpan<T> colors)
-        where T : struct, IColor
-        where TSum : struct, ISum
     {
-        if (colors == null) throw new ArgumentNullException(nameof(colors));
+        IColorFormat colorFormat = T.ColorFormat;
-
+        return colorFormat.ToSum(colorFormat.Sum(MemoryMarshal.AsBytes(colors)));
-        return T.ColorFormat.BytesPerPixel switch
-        {
-            // DarthAffe 05.07.2024: Important: The sum of 3-byte colors result in 4 byte data!
-            3 => Unsafe.BitCast<Generic4LongData, TSum>(Sum(MemoryMarshal.Cast<T, Generic3ByteData>(colors))),
-            4 => Unsafe.BitCast<Generic4LongData, TSum>(Sum(MemoryMarshal.Cast<T, Generic4ByteData>(colors))),
-            _ => throw new NotSupportedException("Data is not of a supported valid color-type.")
-        };
     }
-
+    
-    private static Generic4LongData Sum(ReadOnlySpan<Generic3ByteData> data)
+    internal static Generic4LongData Sum(ReadOnlySpan<Generic3ByteData> data)
     {
         long b1Sum = 0, b2Sum = 0, b3Sum = 0;
 
@@ -215,7 +186,7 @@ public static unsafe partial class PixelHelper
         return new Generic4LongData(b1Sum, b2Sum, b3Sum, data.Length * 255);
     }
 
-    private static Generic4LongData Sum(ReadOnlySpan<Generic4ByteData> data)
+    internal static Generic4LongData Sum(ReadOnlySpan<Generic4ByteData> data)
     {
         long b1Sum, b2Sum, b3Sum, b4Sum;
         int i = 0;

HPPH/PixelHelper.cs

@@ -3,4 +3,7 @@
 namespace HPPH;
 
 [SkipLocalsInit]
-public static partial class PixelHelper;
+public static partial class PixelHelper
+{
+    private static readonly ParallelOptions PARALLEL_OPTIONS = new() { MaxDegreeOfParallelism = Environment.ProcessorCount };
+}