Enable Sse2, simplify

Author: JimBobSquarePants

src/ImageSharp/Common/Helpers/SimdUtils.HwIntrinsics.cs

@@ -1,6 +1,7 @@
 // Copyright (c) Six Labors.
 // Licensed under the Six Labors Split License.
 
+using System.Numerics;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
 using System.Runtime.Intrinsics;
@@ -656,6 +657,36 @@ public static Vector128<byte> BlendVariable(Vector128<byte> left, Vector128<byte
             return AdvSimd.BitwiseSelect(signedMask, right.AsInt16(), left.AsInt16()).AsByte();
         }
 
+        /// <summary>
+        /// Blend packed 32-bit unsigned integers from <paramref name="left"/> and <paramref name="right"/> using <paramref name="mask"/>.
+        /// The high bit of each corresponding <paramref name="mask"/> byte determines the selection.
+        /// If the high bit is set the element of <paramref name="left"/> is selected.
+        /// The element of <paramref name="right"/> is selected otherwise.
+        /// </summary>
+        /// <param name="left">The left vector.</param>
+        /// <param name="right">The right vector.</param>
+        /// <param name="mask">The mask vector.</param>
+        /// <returns>The <see cref="Vector256{T}"/>.</returns>
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        public static Vector128<uint> BlendVariable(Vector128<uint> left, Vector128<uint> right, Vector128<uint> mask)
+            => BlendVariable(left.AsByte(), right.AsByte(), mask.AsByte()).AsUInt32();
+
+        /// <summary>
+        /// Count the number of leading zero bits in a mask.
+        /// Similar in behavior to the x86 instruction LZCNT.
+        /// </summary>
+        /// <param name="value">The value.</param>
+        public static ushort LeadingZeroCount(ushort value)
+            => (ushort)(BitOperations.LeadingZeroCount(value) - 16);
+
+        /// <summary>
+        /// Count the number of trailing zero bits in an integer value.
+        /// Similar in behavior to the x86 instruction TZCNT.
+        /// </summary>
+        /// <param name="value">The value.</param>
+        public static ushort TrailingZeroCount(ushort value)
+            => (ushort)(BitOperations.TrailingZeroCount(value << 16) - 16);
+
         /// <summary>
         /// <see cref="ByteToNormalizedFloat"/> as many elements as possible, slicing them down (keeping the remainder).
         /// </summary>

src/ImageSharp/Formats/AnimationUtilities.cs

@@ -128,6 +128,52 @@ public static (bool Difference, Rectangle Bounds) DeDuplicatePixels<TPixel>(
                 }
             }
 
+            if (Sse2.IsSupported && remaining >= 4)
+            {
+                Vector128<uint> r128 = previousFrame != null ? Vector128.Create(bg.PackedValue) : Vector128<uint>.Zero;
+                Vector128<uint> vmb128 = Vector128<uint>.Zero;
+                if (blend)
+                {
+                    vmb128 = Sse2.CompareEqual(vmb128, vmb128);
+                }
+
+                while (remaining >= 4)
+                {
+                    Vector128<uint> p = Unsafe.Add(ref Unsafe.As<Vector256<byte>, Vector128<uint>>(ref previousBase), x);
+                    Vector128<uint> c = Unsafe.Add(ref Unsafe.As<Vector256<byte>, Vector128<uint>>(ref currentBase), x);
+
+                    Vector128<uint> eq = Sse2.CompareEqual(p, c);
+                    Vector128<uint> r = SimdUtils.HwIntrinsics.BlendVariable(c, r128, Sse2.And(eq, vmb128));
+
+                    if (nextFrame != null)
+                    {
+                        Vector128<int> n = Sse2.ShiftRightLogical(Unsafe.Add(ref Unsafe.As<Vector256<byte>, Vector128<uint>>(ref nextBase), x), 24).AsInt32();
+                        eq = Sse2.AndNot(Sse2.CompareGreaterThan(Sse2.ShiftRightLogical(c, 24).AsInt32(), n).AsUInt32(), eq);
+                    }
+
+                    Unsafe.Add(ref Unsafe.As<Vector256<byte>, Vector128<byte>>(ref resultBase), x) = r.AsByte();
+
+                    ushort msk = (ushort)(uint)Sse2.MoveMask(eq.AsByte());
+                    msk = (ushort)~msk;
+                    if (msk != 0)
+                    {
+                        // If is diff is found, the left side is marked by the min of previously found left side and the start position.
+                        // The right is the max of the previously found right side and the end position.
+                        int start = i + (SimdUtils.HwIntrinsics.TrailingZeroCount(msk) / sizeof(uint));
+                        int end = i + (4 - (SimdUtils.HwIntrinsics.LeadingZeroCount(msk) / sizeof(uint)));
+                        left = Math.Min(left, start);
+                        right = Math.Max(right, end);
+                        hasRowDiff = true;
+                        hasDiff = true;
+                    }
+
+                    x++;
+                    i += 4;
+                    remaining -= 4;
+                }
+            }
+
+            // TODO: AdvSimd ??
             for (i = remaining; i > 0; i--)
             {
                 x = (uint)(length - i);

src/ImageSharp/Formats/Gif/GifEncoderCore.cs

@@ -168,7 +168,7 @@ public void Encode<TPixel>(Image<TPixel> image, Stream stream, CancellationToken
 
         stream.WriteByte(GifConstants.EndIntroducer);
 
-        quantized.Dispose();
+        quantized?.Dispose();
     }
 
     private static GifMetadata GetGifMetadata<TPixel>(Image<TPixel> image)
@@ -251,6 +251,7 @@ private void EncodeAdditionalFrames<TPixel>(
         {
             // Gather the metadata for this frame.
             ImageFrame<TPixel> currentFrame = image.Frames[i];
+            ImageFrame<TPixel>? nextFrame = i < image.Frames.Count - 1 ? image.Frames[i + 1] : null;
             GifFrameMetadata gifMetadata = GetGifFrameMetadata(currentFrame, globalTransparencyIndex);
             bool useLocal = this.colorTableMode == GifColorTableMode.Local || (gifMetadata.ColorTableMode == GifColorTableMode.Local);
 
@@ -264,8 +265,6 @@ private void EncodeAdditionalFrames<TPixel>(
                 hasPaletteQuantizer = true;
             }
 
-            ImageFrame<TPixel>? nextFrame = i < image.Frames.Count - 1 ? image.Frames[i + 1] : null;
-
             this.EncodeAdditionalFrame(
                 stream,
                 previousFrame,

src/ImageSharp/Formats/Png/PngMetadata.cs

@@ -90,29 +90,26 @@ internal static PngMetadata FromAnimatedMetadata(AnimatedImageMetadata metadata)
     {
         // Should the conversion be from a format that uses a 24bit palette entries (gif)
         // we need to clone and adjust the color table to allow for transparency.
-        ReadOnlyMemory<Color>? colorTable = metadata.ColorTable;
-        if (metadata.ColorTable.HasValue)
+        Color[]? colorTable = metadata.ColorTable.HasValue ? metadata.ColorTable.Value.ToArray() : null;
+        if (colorTable != null)
         {
-            Color[] clone = metadata.ColorTable.Value.ToArray();
-            for (int i = 0; i < clone.Length; i++)
+            for (int i = 0; i < colorTable.Length; i++)
             {
-                ref Color c = ref clone[i];
+                ref Color c = ref colorTable[i];
                 if (c == metadata.BackgroundColor)
                 {
                     // Png treats background as fully empty
                     c = Color.Transparent;
                     break;
                 }
             }
-
-            colorTable = clone;
         }
 
         return new()
         {
-            ColorType = colorTable.HasValue ? PngColorType.Palette : null,
-            BitDepth = colorTable.HasValue
-                        ? (PngBitDepth)Numerics.Clamp(ColorNumerics.GetBitsNeededForColorDepth(colorTable.Value.Length), 1, 8)
+            ColorType = colorTable != null ? PngColorType.Palette : null,
+            BitDepth = colorTable != null
+                        ? (PngBitDepth)Numerics.Clamp(ColorNumerics.GetBitsNeededForColorDepth(colorTable.Length), 1, 8)
                         : null,
             ColorTable = colorTable,
             RepeatCount = metadata.RepeatCount,

tests/ImageSharp.Tests/TestUtilities/TestEnvironment.Formats.cs

@@ -1,7 +1,6 @@
 // Copyright (c) Six Labors.
 // Licensed under the Six Labors Split License.
 
-using System.Diagnostics.CodeAnalysis;
 using SixLabors.ImageSharp.Formats;
 using SixLabors.ImageSharp.Formats.Bmp;
 using SixLabors.ImageSharp.Formats.Gif;