PNG: Add SSE/AVX version of Sub, Up, Average and Paethfilters

src/ImageSharp/Formats/Png/Filters/AverageFilter.cs

@@ -20,20 +20,75 @@ namespace SixLabors.ImageSharp.Formats.Png.Filters
     internal static class AverageFilter
     {
         /// <summary>
-        /// Decodes the scanline
+        /// Decodes a scanline, which was filtered with the average filter.
         /// </summary>
-        /// <param name="scanline">The scanline to decode</param>
+        /// <param name="scanline">The scanline to decode.</param>
         /// <param name="previousScanline">The previous scanline.</param>
         /// <param name="bytesPerPixel">The bytes per pixel.</param>
         [MethodImpl(MethodImplOptions.AggressiveInlining)]
         public static void Decode(Span<byte> scanline, Span<byte> previousScanline, int bytesPerPixel)
         {
             DebugGuard.MustBeSameSized<byte>(scanline, previousScanline, nameof(scanline));
 
+            // The Avg filter predicts each pixel as the (truncated) average of a and b:
+            // Average(x) + floor((Raw(x-bpp)+Prior(x))/2)
+            // With pixels positioned like this:
+            //  prev:  c b
+            //  row:   a d
+#if SUPPORTS_RUNTIME_INTRINSICS
+            if (Sse2.IsSupported && bytesPerPixel is 4)
+            {
+                DecodeSse2(scanline, previousScanline);
+            }
+            else
+#endif
+            {
+                DecodeScalar(scanline, previousScanline, bytesPerPixel);
+            }
+        }
+
+#if SUPPORTS_RUNTIME_INTRINSICS
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private static void DecodeSse2(Span<byte> scanline, Span<byte> previousScanline)
+        {
+            ref byte scanBaseRef = ref MemoryMarshal.GetReference(scanline);
+            ref byte prevBaseRef = ref MemoryMarshal.GetReference(previousScanline);
+
+            Vector128<byte> d = Vector128<byte>.Zero;
+            var ones = Vector128.Create((byte)1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1);
+
+            int rb = scanline.Length;
+            int offset = 1;
+            while (rb >= 4)
+            {
+                ref byte scanRef = ref Unsafe.Add(ref scanBaseRef, offset);
+                Vector128<byte> a = d;
+                Vector128<byte> b = Sse2.ConvertScalarToVector128Int32(Unsafe.As<byte, int>(ref Unsafe.Add(ref prevBaseRef, offset))).AsByte();
+                d = Sse2.ConvertScalarToVector128Int32(Unsafe.As<byte, int>(ref scanRef)).AsByte();
+
+                // PNG requires a truncating average, so we can't just use _mm_avg_epu8,
+                // but we can fix it up by subtracting off 1 if it rounded up.
+                Vector128<byte> avg = Sse2.Average(a, b);
+                Vector128<byte> xor = Sse2.Xor(a, b);
+                Vector128<byte> and = Sse2.And(xor, ones);
+                avg = Sse2.Subtract(avg, and);
+                d = Sse2.Add(d, avg);
+
+                // Store the result.
+                Unsafe.As<byte, int>(ref scanRef) = Sse2.ConvertToInt32(d.AsInt32());
+
+                rb -= 4;
+                offset += 4;
+            }
+        }
+#endif
+
+        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        private static void DecodeScalar(Span<byte> scanline, Span<byte> previousScanline, int bytesPerPixel)
+        {
             ref byte scanBaseRef = ref MemoryMarshal.GetReference(scanline);
             ref byte prevBaseRef = ref MemoryMarshal.GetReference(previousScanline);
 
-            // Average(x) + floor((Raw(x-bpp)+Prior(x))/2)
             int x = 1;
             for (; x <= bytesPerPixel /* Note the <= because x starts at 1 */; ++x)
             {
@@ -52,13 +107,13 @@ public static void Decode(Span<byte> scanline, Span<byte> previousScanline, int
         }
 
         /// <summary>
-        /// Encodes the scanline
+        /// Encodes a scanline with the average filter applied.
         /// </summary>
-        /// <param name="scanline">The scanline to encode</param>
+        /// <param name="scanline">The scanline to encode.</param>
         /// <param name="previousScanline">The previous scanline.</param>
         /// <param name="result">The filtered scanline result.</param>
         /// <param name="bytesPerPixel">The bytes per pixel.</param>
-        /// <param name="sum">The sum of the total variance of the filtered row</param>
+        /// <param name="sum">The sum of the total variance of the filtered row.</param>
         [MethodImpl(MethodImplOptions.AggressiveInlining)]
         public static void Encode(ReadOnlySpan<byte> scanline, ReadOnlySpan<byte> previousScanline, Span<byte> result, int bytesPerPixel, out int sum)
         {

tests/ImageSharp.Benchmarks/Codecs/Png/DecodeFilteredPng.cs

@@ -16,16 +16,18 @@ public class DecodeFilteredPng
         private byte[] filter1;
         private byte[] filter2;
         private byte[] filter3;
-        private byte[] filter4;
+        private byte[] averageFilter3bpp;
+        private byte[] averageFilter4bpp;
 
         [GlobalSetup]
         public void ReadImages()
         {
             this.filter0 = File.ReadAllBytes(TestImageFullPath(TestImages.Png.Filter0));
             this.filter1 = File.ReadAllBytes(TestImageFullPath(TestImages.Png.Filter1));
             this.filter2 = File.ReadAllBytes(TestImageFullPath(TestImages.Png.Filter2));
-            this.filter3 = File.ReadAllBytes(TestImageFullPath(TestImages.Png.Filter3));
-            this.filter4 = File.ReadAllBytes(TestImageFullPath(TestImages.Png.Filter4));
+            this.filter3 = File.ReadAllBytes(TestImageFullPath(TestImages.Png.Filter4));
+            this.averageFilter3bpp = File.ReadAllBytes(TestImageFullPath(TestImages.Png.AverageFilter3BytesPerPixel));
+            this.averageFilter4bpp = File.ReadAllBytes(TestImageFullPath(TestImages.Png.AverageFilter4BytesPerPixel));
         }
 
         [Benchmark(Baseline = true, Description = "None-filtered PNG file")]
@@ -40,13 +42,17 @@ public Size PngFilter1()
         public Size PngFilter2()
             => LoadPng(this.filter2);
 
-        [Benchmark(Description = "Average-filtered PNG file")]
-        public Size PngFilter3()
-            => LoadPng(this.filter3);
+        [Benchmark(Description = "Average-filtered PNG file (3bpp)")]
+        public Size PngAvgFilter1()
+            => LoadPng(this.averageFilter3bpp);
+
+        [Benchmark(Description = "Average-filtered PNG file (4bpp)")]
+        public Size PngAvgFilter2()
+            => LoadPng(this.averageFilter4bpp);
 
         [Benchmark(Description = "Paeth-filtered PNG file")]
         public Size PngFilter4()
-            => LoadPng(this.filter4);
+            => LoadPng(this.filter3);
 
         [MethodImpl(MethodImplOptions.AggressiveInlining)]
         private static Size LoadPng(byte[] bytes)

tests/ImageSharp.Tests/Formats/Png/PngDecoderFilterTests.cs

@@ -0,0 +1,63 @@
+// Copyright (c) Six Labors.
+// Licensed under the Apache License, Version 2.0.
+
+using SixLabors.ImageSharp.Formats.Png.Filters;
+using SixLabors.ImageSharp.Tests.TestUtilities;
+using Xunit;
+
+namespace SixLabors.ImageSharp.Tests.Formats.Png
+{
+    [Trait("Format", "Png")]
+    public class PngDecoderFilterTests
+    {
+        private static void RunAverageFilterTest()
+        {
+            // arrange
+            byte[] scanline =
+            {
+                3, 39, 39, 39, 0, 4, 4, 4, 0, 1, 1, 1, 0, 1, 1, 1, 0, 2, 2, 2, 0, 2, 2, 2, 0, 2, 2, 2, 0, 4, 4, 4,
+                0, 2, 2, 2, 0, 3, 3, 3, 0, 1, 1, 1, 0, 3, 3, 3, 0, 3, 3, 3, 0, 1, 1, 1, 0, 3, 3, 3, 0, 2, 2, 2, 0,
+                1, 1, 1, 0, 3, 3, 3, 0, 1, 1, 1, 0, 3, 3, 3, 0, 1, 1, 1, 0, 3, 3, 3, 0, 3, 3, 3, 0, 254, 254, 254,
+                0, 6, 6, 6, 14, 71, 71, 71, 157, 254, 254, 254, 28, 251, 251, 251, 0, 4, 4, 4, 0, 2, 2, 2, 0, 11,
+                11, 11, 0, 226, 226, 226, 0, 255, 128, 234
+            };
+
+            byte[] previousScanline =
+            {
+                3, 74, 74, 74, 0, 73, 73, 73, 0, 73, 73, 73, 0, 74, 74, 74, 0, 74, 74, 74, 0, 73, 73, 73, 0, 72, 72,
+                72, 0, 72, 72, 72, 0, 73, 73, 73, 0, 74, 74, 74, 0, 73, 73, 73, 0, 72, 72, 72, 0, 72, 72, 72, 0, 74,
+                74, 74, 0, 72, 72, 72, 0, 73, 73, 73, 0, 75, 75, 75, 0, 73, 73, 73, 0, 74, 74, 74, 0, 72, 72, 72, 0,
+                73, 73, 73, 0, 73, 73, 73, 0, 72, 72, 72, 0, 74, 74, 74, 0, 61, 61, 61, 0, 101, 101, 101, 78, 197,
+                197, 197, 251, 152, 152, 152, 255, 155, 155, 155, 255, 162, 162, 162, 255, 175, 175, 175, 255, 160,
+                160, 160, 255, 139, 128, 134
+            };
+
+            byte[] expected =
+            {
+                3, 76, 76, 76, 0, 78, 78, 78, 0, 76, 76, 76, 0, 76, 76, 76, 0, 77, 77, 77, 0, 77, 77, 77, 0, 76, 76,
+                76, 0, 78, 78, 78, 0, 77, 77, 77, 0, 78, 78, 78, 0, 76, 76, 76, 0, 77, 77, 77, 0, 77, 77, 77, 0, 76,
+                76, 76, 0, 77, 77, 77, 0, 77, 77, 77, 0, 77, 77, 77, 0, 78, 78, 78, 0, 77, 77, 77, 0, 77, 77, 77, 0,
+                76, 76, 76, 0, 77, 77, 77, 0, 77, 77, 77, 0, 73, 73, 73, 0, 73, 73, 73, 14, 158, 158, 158, 203, 175,
+                175, 175, 255, 158, 158, 158, 255, 160, 160, 160, 255, 163, 163, 163, 255, 180, 180, 180, 255, 140,
+                140, 140, 255, 138, 6, 115
+            };
+
+            // act
+            AverageFilter.Decode(scanline, previousScanline, 4);
+
+            // assert
+            Assert.Equal(scanline, expected);
+        }
+
+        [Fact]
+        public void AverageFilter_Works() => RunAverageFilterTest();
+
+#if SUPPORTS_RUNTIME_INTRINSICS
+        [Fact]
+        public void AverageFilter_WithHardwareIntrinsics_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunAverageFilterTest, HwIntrinsics.AllowAll);
+
+        [Fact]
+        public void AverageFilter_WithoutHardwareIntrinsics_Works() => FeatureTestRunner.RunWithHwIntrinsicsFeature(RunAverageFilterTest, HwIntrinsics.DisableHWIntrinsic);
+#endif
+    }
+}

tests/ImageSharp.Tests/Formats/Png/PngDecoderTests.cs

@@ -111,6 +111,17 @@ public void Decode<TPixel>(TestImageProvider<TPixel> provider)
             image.CompareToOriginal(provider, ImageComparer.Exact);
         }
 
+        [Theory]
+        [WithFile(TestImages.Png.AverageFilter3BytesPerPixel, PixelTypes.Rgba64)]
+        [WithFile(TestImages.Png.AverageFilter4BytesPerPixel, PixelTypes.Rgba64)]
+        public void Decode_WithAverageFilter<TPixel>(TestImageProvider<TPixel> provider)
+            where TPixel : unmanaged, IPixel<TPixel>
+        {
+            using Image<TPixel> image = provider.GetImage(PngDecoder);
+            image.DebugSave(provider);
+            image.CompareToOriginal(provider, ImageComparer.Exact);
+        }
+
         [Theory]
         [WithFile(TestImages.Png.GrayA8Bit, PixelTypes.Rgba32)]
         [WithFile(TestImages.Png.Gray1BitTrans, PixelTypes.Rgba32)]

tests/ImageSharp.Tests/Formats/Png/PngFilterTests.cs → tests/ImageSharp.Tests/Formats/Png/PngEncoderFilterTests.cs

@@ -13,15 +13,15 @@
 namespace SixLabors.ImageSharp.Tests.Formats.Png
 {
     [Trait("Format", "Png")]
-    public partial class PngFilterTests : MeasureFixture
+    public class PngEncoderFilterTests : MeasureFixture
     {
 #if BENCHMARKING
         public const int Times = 1000000;
 #else
         public const int Times = 1;
 #endif
 
-        public PngFilterTests(ITestOutputHelper output)
+        public PngEncoderFilterTests(ITestOutputHelper output)
             : base(output)
         {
         }

tests/ImageSharp.Tests/TestImages.cs

@@ -68,9 +68,11 @@ public static class Png
             public const string Filter0 = "Png/filter0.png";
             public const string Filter1 = "Png/filter1.png";
             public const string Filter2 = "Png/filter2.png";
-            public const string Filter3 = "Png/filter3.png";
+            public const string AverageFilter3BytesPerPixel = "Png/filter3.png";
             public const string Filter4 = "Png/filter4.png";
 
+            public const string AverageFilter4BytesPerPixel = "Png/AverageFilter4Bpp.png";
+
             // Paletted images also from http://www.schaik.com/pngsuite/pngsuite_fil_png.html
             public const string PalettedTwoColor = "Png/basn3p01.png";
             public const string PalettedFourColor = "Png/basn3p02.png";
@@ -159,7 +161,7 @@ public static class Bad
             {
                 P1, Pd, Blur, Splash, Cross,
                 Powerpoint, SplashInterlaced, Interlaced,
-                Filter0, Filter1, Filter2, Filter3, Filter4,
+                Filter0, Filter1, Filter2, AverageFilter3BytesPerPixel, Filter4,
                 FilterVar, VimImage1, VimImage2, VersioningImage1,
                 VersioningImage2, Ratio4x1, Ratio1x4
             };