Reintroduce scanline optimizations

src/ImageSharp/Formats/Png/PngDecoderCore.cs

@@ -567,7 +567,7 @@ internal sealed class PngDecoderCore : IImageDecoderInternals
         if (frameControl is { } control)
         {
             PngFrameMetadata frameMetadata = image.Frames.RootFrame.Metadata.GetPngFrameMetadata();
-            frameMetadata.FromChunk(control);
+            frameMetadata.FromChunk(in control);
         }
 
         this.bytesPerPixel = this.CalculateBytesPerPixel();
@@ -837,7 +837,13 @@ internal sealed class PngDecoderCore : IImageDecoderInternals
                 }
 
                 Span<TPixel> rowSpan = imageBuffer.DangerousGetRowSpan(currentRow);
-                this.ProcessInterlacedDefilteredScanline(frameControl, this.scanline.GetSpan(), rowSpan, pngMetadata, pixelOffset: Adam7.FirstColumn[pass], increment: Adam7.ColumnIncrement[pass]);
+                this.ProcessInterlacedDefilteredScanline(
+                    frameControl,
+                    this.scanline.GetSpan(),
+                    rowSpan,
+                    pngMetadata,
+                    pixelOffset: Adam7.FirstColumn[pass],
+                    increment: Adam7.ColumnIncrement[pass]);
 
                 this.SwapScanlineBuffers();
 
@@ -935,6 +941,7 @@ internal sealed class PngDecoderCore : IImageDecoderInternals
 
                 case PngColorType.RgbWithAlpha:
                     PngScanlineProcessor.ProcessRgbaScanline(
+                        this.configuration,
                         this.header.BitDepth,
                         in frameControl,
                         scanlineSpan,
@@ -961,7 +968,13 @@ internal sealed class PngDecoderCore : IImageDecoderInternals
     /// <param name="pngMetadata">The png metadata.</param>
     /// <param name="pixelOffset">The column start index. Always 0 for none interlaced images.</param>
     /// <param name="increment">The column increment. Always 1 for none interlaced images.</param>
-    private void ProcessInterlacedDefilteredScanline<TPixel>(in FrameControl frameControl, ReadOnlySpan<byte> defilteredScanline, Span<TPixel> rowSpan, PngMetadata pngMetadata, int pixelOffset = 0, int increment = 1)
+    private void ProcessInterlacedDefilteredScanline<TPixel>(
+        in FrameControl frameControl,
+        ReadOnlySpan<byte> defilteredScanline,
+        Span<TPixel> rowSpan,
+        PngMetadata pngMetadata,
+        int pixelOffset = 0,
+        int increment = 1)
         where TPixel : unmanaged, IPixel<TPixel>
     {
         // Trim the first marker byte from the buffer
@@ -1034,6 +1047,7 @@ internal sealed class PngDecoderCore : IImageDecoderInternals
 
                 case PngColorType.RgbWithAlpha:
                     PngScanlineProcessor.ProcessInterlacedRgbaScanline(
+                        this.configuration,
                         this.header.BitDepth,
                         in frameControl,
                         scanlineSpan,

src/ImageSharp/Formats/Png/PngFrameMetadata.cs

@@ -53,7 +53,7 @@ public class PngFrameMetadata : IDeepCloneable
     /// Initializes a new instance of the <see cref="PngFrameMetadata"/> class.
     /// </summary>
     /// <param name="frameControl">The chunk to create an instance from.</param>
-    internal void FromChunk(FrameControl frameControl)
+    internal void FromChunk(in FrameControl frameControl)
     {
         this.DelayNumerator = frameControl.DelayNumerator;
         this.DelayDenominator = frameControl.DelayDenominator;

src/ImageSharp/Formats/Png/PngScanlineProcessor.cs

@@ -216,18 +216,18 @@ internal static class PngScanlineProcessor
         int bytesPerPixel,
         int bytesPerSample,
         Color? transparentColor)
-        where TPixel : unmanaged, IPixel<TPixel> =>
-        ProcessInterlacedRgbScanline(
-            configuration,
-            bitDepth,
-            frameControl,
-            scanlineSpan,
-            rowSpan,
-            0,
-            1,
-            bytesPerPixel,
-            bytesPerSample,
-            transparentColor);
+       where TPixel : unmanaged, IPixel<TPixel> =>
+       ProcessInterlacedRgbScanline(
+           configuration,
+           bitDepth,
+           frameControl,
+           scanlineSpan,
+           rowSpan,
+           0,
+           1,
+           bytesPerPixel,
+           bytesPerSample,
+           transparentColor);
 
     public static void ProcessInterlacedRgbScanline<TPixel>(
         Configuration configuration,
@@ -264,9 +264,16 @@ internal static class PngScanlineProcessor
                     Unsafe.Add(ref rowSpanRef, x) = pixel;
                 }
             }
+            else if (pixelOffset == 0 && increment == 1)
+            {
+                PixelOperations<TPixel>.Instance.FromRgb24Bytes(
+                    configuration,
+                    scanlineSpan[..(int)(frameControl.Width * bytesPerPixel)],
+                    rowSpan.Slice((int)frameControl.XOffset, (int)frameControl.Width),
+                    (int)frameControl.Width);
+            }
             else
             {
-                // TODO: Investigate reintroducing bulk operations optimization here.
                 Rgb24 rgb = default;
                 int o = 0;
                 for (nuint x = offset; x < frameControl.XMax; x += increment, o += bytesPerPixel)
@@ -323,24 +330,27 @@ internal static class PngScanlineProcessor
     }
 
     public static void ProcessRgbaScanline<TPixel>(
+        Configuration configuration,
         int bitDepth,
         in FrameControl frameControl,
         ReadOnlySpan<byte> scanlineSpan,
         Span<TPixel> rowSpan,
         int bytesPerPixel,
         int bytesPerSample)
-        where TPixel : unmanaged, IPixel<TPixel> =>
-        ProcessInterlacedRgbaScanline(
-            bitDepth,
-            frameControl,
-            scanlineSpan,
-            rowSpan,
-            0,
-            1,
-            bytesPerPixel,
-            bytesPerSample);
+       where TPixel : unmanaged, IPixel<TPixel> =>
+       ProcessInterlacedRgbaScanline(
+           configuration,
+           bitDepth,
+           frameControl,
+           scanlineSpan,
+           rowSpan,
+           0,
+           1,
+           bytesPerPixel,
+           bytesPerSample);
 
     public static void ProcessInterlacedRgbaScanline<TPixel>(
+        Configuration configuration,
         int bitDepth,
         in FrameControl frameControl,
         ReadOnlySpan<byte> scanlineSpan,
@@ -370,9 +380,16 @@ internal static class PngScanlineProcessor
                 Unsafe.Add(ref rowSpanRef, x) = pixel;
             }
         }
+        else if (pixelOffset == 0 && increment == 1)
+        {
+            PixelOperations<TPixel>.Instance.FromRgba32Bytes(
+                configuration,
+                scanlineSpan[..(int)(frameControl.Width * bytesPerPixel)],
+                rowSpan.Slice((int)frameControl.XOffset, (int)frameControl.Width),
+                (int)frameControl.Width);
+        }
         else
         {
-            // TODO: Investigate reintroducing bulk operations optimization here.
             ref byte scanlineSpanRef = ref MemoryMarshal.GetReference(scanlineSpan);
             Rgba32 rgba = default;
             int o = 0;