Merge branch 'master' into js/v2-builds

src/ImageSharp/Common/Helpers/InliningOptions.cs

@@ -12,6 +12,10 @@ namespace SixLabors.ImageSharp
     /// </summary>
     internal static class InliningOptions
     {
+        /// <summary>
+        /// <see cref="MethodImplOptions.AggressiveInlining"/> regardless of the build conditions.
+        /// </summary>
+        public const MethodImplOptions AlwaysInline = MethodImplOptions.AggressiveInlining;
 #if PROFILING
         public const MethodImplOptions HotPath = MethodImplOptions.NoInlining;
         public const MethodImplOptions ShortMethod = MethodImplOptions.NoInlining;

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

@@ -537,7 +537,7 @@ namespace SixLabors.ImageSharp
             /// <param name="vm0">The first vector to multiply.</param>
             /// <param name="vm1">The second vector to multiply.</param>
             /// <returns>The <see cref="Vector256{T}"/>.</returns>
-            [MethodImpl(InliningOptions.ShortMethod)]
+            [MethodImpl(InliningOptions.AlwaysInline)]
             public static Vector256<float> MultiplyAdd(
                 in Vector256<float> va,
                 in Vector256<float> vm0,

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromCmykAvx2.cs

@@ -22,60 +22,39 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            protected override void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result)
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
             {
 #if SUPPORTS_RUNTIME_INTRINSICS
-                ref Vector256<float> cBase =
+                ref Vector256<float> c0Base =
                                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component0));
-                ref Vector256<float> mBase =
+                ref Vector256<float> c1Base =
                                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component1));
-                ref Vector256<float> yBase =
+                ref Vector256<float> c2Base =
                                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component2));
-                ref Vector256<float> kBase =
+                ref Vector256<float> c3Base =
                                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component3));
 
-                ref Vector256<float> resultBase =
-                    ref Unsafe.As<Vector4, Vector256<float>>(ref MemoryMarshal.GetReference(result));
-
                 // Used for the color conversion
                 var scale = Vector256.Create(1 / this.MaximumValue);
-                var one = Vector256.Create(1F);
-
-                // Used for packing
-                ref byte control = ref MemoryMarshal.GetReference(HwIntrinsics.PermuteMaskEvenOdd8x32);
-                Vector256<int> vcontrol = Unsafe.As<byte, Vector256<int>>(ref control);
 
-                int n = result.Length / 8;
-                for (int i = 0; i < n; i++)
+                nint n = values.Component0.Length / 8;
+                for (nint i = 0; i < n; i++)
                 {
-                    Vector256<float> k = Avx2.PermuteVar8x32(Unsafe.Add(ref kBase, i), vcontrol);
-                    Vector256<float> c = Avx2.PermuteVar8x32(Unsafe.Add(ref cBase, i), vcontrol);
-                    Vector256<float> m = Avx2.PermuteVar8x32(Unsafe.Add(ref mBase, i), vcontrol);
-                    Vector256<float> y = Avx2.PermuteVar8x32(Unsafe.Add(ref yBase, i), vcontrol);
+                    ref Vector256<float> c = ref Unsafe.Add(ref c0Base, i);
+                    ref Vector256<float> m = ref Unsafe.Add(ref c1Base, i);
+                    ref Vector256<float> y = ref Unsafe.Add(ref c2Base, i);
+                    Vector256<float> k = Unsafe.Add(ref c3Base, i);
 
                     k = Avx.Multiply(k, scale);
-
                     c = Avx.Multiply(Avx.Multiply(c, k), scale);
                     m = Avx.Multiply(Avx.Multiply(m, k), scale);
                     y = Avx.Multiply(Avx.Multiply(y, k), scale);
-
-                    Vector256<float> cmLo = Avx.UnpackLow(c, m);
-                    Vector256<float> yoLo = Avx.UnpackLow(y, one);
-                    Vector256<float> cmHi = Avx.UnpackHigh(c, m);
-                    Vector256<float> yoHi = Avx.UnpackHigh(y, one);
-
-                    ref Vector256<float> destination = ref Unsafe.Add(ref resultBase, i * 4);
-
-                    destination = Avx.Shuffle(cmLo, yoLo, 0b01_00_01_00);
-                    Unsafe.Add(ref destination, 1) = Avx.Shuffle(cmLo, yoLo, 0b11_10_11_10);
-                    Unsafe.Add(ref destination, 2) = Avx.Shuffle(cmHi, yoHi, 0b01_00_01_00);
-                    Unsafe.Add(ref destination, 3) = Avx.Shuffle(cmHi, yoHi, 0b11_10_11_10);
                 }
 #endif
             }
 
-            protected override void ConvertCore(in ComponentValues values, Span<Vector4> result) =>
-                FromCmykBasic.ConvertCore(values, result, this.MaximumValue);
+            protected override void ConvertCoreInplace(in ComponentValues values) =>
+                FromCmykBasic.ConvertCoreInplace(values, this.MaximumValue);
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromCmykBasic.cs

@@ -15,38 +15,27 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            public override void ConvertToRgba(in ComponentValues values, Span<Vector4> result)
-            {
-                ConvertCore(values, result, this.MaximumValue);
-            }
+            public override void ConvertToRgbInplace(in ComponentValues values) =>
+                ConvertCoreInplace(values, this.MaximumValue);
 
-            internal static void ConvertCore(in ComponentValues values, Span<Vector4> result, float maxValue)
+            internal static void ConvertCoreInplace(in ComponentValues values, float maxValue)
             {
-                ReadOnlySpan<float> cVals = values.Component0;
-                ReadOnlySpan<float> mVals = values.Component1;
-                ReadOnlySpan<float> yVals = values.Component2;
-                ReadOnlySpan<float> kVals = values.Component3;
-
-                var v = new Vector4(0, 0, 0, 1F);
-
-                var maximum = 1 / maxValue;
-                var scale = new Vector4(maximum, maximum, maximum, 1F);
+                Span<float> c0 = values.Component0;
+                Span<float> c1 = values.Component1;
+                Span<float> c2 = values.Component2;
+                Span<float> c3 = values.Component3;
 
-                for (int i = 0; i < result.Length; i++)
+                float scale = 1 / maxValue;
+                for (int i = 0; i < c0.Length; i++)
                 {
-                    float c = cVals[i];
-                    float m = mVals[i];
-                    float y = yVals[i];
-                    float k = kVals[i] / maxValue;
-
-                    v.X = c * k;
-                    v.Y = m * k;
-                    v.Z = y * k;
-                    v.W = 1F;
-
-                    v *= scale;
-
-                    result[i] = v;
+                    float c = c0[i];
+                    float m = c1[i];
+                    float y = c2[i];
+                    float k = c3[i] / maxValue;
+
+                    c0[i] = c * k * scale;
+                    c1[i] = m * k * scale;
+                    c2[i] = y * k * scale;
                 }
             }
         }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromCmykVector8.cs

@@ -18,7 +18,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            protected override void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result)
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
             {
                 ref Vector<float> cBase =
                                     ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component0));
@@ -29,43 +29,25 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                 ref Vector<float> kBase =
                                     ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component3));
 
-                ref Vector4Octet resultBase =
-                    ref Unsafe.As<Vector4, Vector4Octet>(ref MemoryMarshal.GetReference(result));
-
-                Vector4Pair cc = default;
-                Vector4Pair mm = default;
-                Vector4Pair yy = default;
-                ref Vector<float> ccRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref cc);
-                ref Vector<float> mmRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref mm);
-                ref Vector<float> yyRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref yy);
-
                 var scale = new Vector<float>(1 / this.MaximumValue);
 
                 // Walking 8 elements at one step:
-                int n = result.Length / 8;
-                for (int i = 0; i < n; i++)
+                nint n = values.Component0.Length / 8;
+                for (nint i = 0; i < n; i++)
                 {
-                    Vector<float> c = Unsafe.Add(ref cBase, i);
-                    Vector<float> m = Unsafe.Add(ref mBase, i);
-                    Vector<float> y = Unsafe.Add(ref yBase, i);
+                    ref Vector<float> c = ref Unsafe.Add(ref cBase, i);
+                    ref Vector<float> m = ref Unsafe.Add(ref mBase, i);
+                    ref Vector<float> y = ref Unsafe.Add(ref yBase, i);
                     Vector<float> k = Unsafe.Add(ref kBase, i) * scale;
 
                     c = (c * k) * scale;
                     m = (m * k) * scale;
                     y = (y * k) * scale;
-
-                    ccRefAsVector = c;
-                    mmRefAsVector = m;
-                    yyRefAsVector = y;
-
-                    // Collect (c0,c1...c8) (m0,m1...m8) (y0,y1...y8) vector values in the expected (r0,g0,g1,1), (r1,g1,g2,1) ... order:
-                    ref Vector4Octet destination = ref Unsafe.Add(ref resultBase, i);
-                    destination.Pack(ref cc, ref mm, ref yy);
                 }
             }
 
-            protected override void ConvertCore(in ComponentValues values, Span<Vector4> result) =>
-                FromCmykBasic.ConvertCore(values, result, this.MaximumValue);
+            protected override void ConvertCoreInplace(in ComponentValues values) =>
+                FromCmykBasic.ConvertCoreInplace(values, this.MaximumValue);
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromGrayScaleAvx2.cs

@@ -22,42 +22,26 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            protected override void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result)
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
             {
 #if SUPPORTS_RUNTIME_INTRINSICS
-                ref Vector256<float> gBase =
+                ref Vector256<float> c0Base =
                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component0));
 
-                ref Vector256<float> resultBase =
-                    ref Unsafe.As<Vector4, Vector256<float>>(ref MemoryMarshal.GetReference(result));
-
                 // Used for the color conversion
                 var scale = Vector256.Create(1 / this.MaximumValue);
-                var one = Vector256.Create(1F);
-
-                // Used for packing
-                ref byte control = ref MemoryMarshal.GetReference(HwIntrinsics.PermuteMaskEvenOdd8x32);
-                Vector256<int> vcontrol = Unsafe.As<byte, Vector256<int>>(ref control);
 
-                int n = result.Length / 8;
-                for (int i = 0; i < n; i++)
+                nint n = values.Component0.Length / 8;
+                for (nint i = 0; i < n; i++)
                 {
-                    Vector256<float> g = Avx.Multiply(Unsafe.Add(ref gBase, i), scale);
-
-                    g = Avx2.PermuteVar8x32(g, vcontrol);
-
-                    ref Vector256<float> destination = ref Unsafe.Add(ref resultBase, i * 4);
-
-                    destination = Avx.Blend(Avx.Permute(g, 0b00_00_00_00), one, 0b1000_1000);
-                    Unsafe.Add(ref destination, 1) = Avx.Blend(Avx.Shuffle(g, g, 0b01_01_01_01), one, 0b1000_1000);
-                    Unsafe.Add(ref destination, 2) = Avx.Blend(Avx.Shuffle(g, g, 0b10_10_10_10), one, 0b1000_1000);
-                    Unsafe.Add(ref destination, 3) = Avx.Blend(Avx.Shuffle(g, g, 0b11_11_11_11), one, 0b1000_1000);
+                    ref Vector256<float> c0 = ref Unsafe.Add(ref c0Base, i);
+                    c0 = Avx.Multiply(c0, scale);
                 }
 #endif
             }
 
-            protected override void ConvertCore(in ComponentValues values, Span<Vector4> result) =>
-                FromGrayscaleBasic.ConvertCore(values, result, this.MaximumValue);
+            protected override void ConvertCoreInplace(in ComponentValues values) =>
+                FromGrayscaleBasic.ScaleValues(values.Component0, this.MaximumValue);
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromGrayScaleBasic.cs

@@ -17,25 +17,35 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            public override void ConvertToRgba(in ComponentValues values, Span<Vector4> result)
-            {
-                ConvertCore(values, result, this.MaximumValue);
-            }
+            public override void ConvertToRgbInplace(in ComponentValues values) =>
+                ScaleValues(values.Component0, this.MaximumValue);
 
-            internal static void ConvertCore(in ComponentValues values, Span<Vector4> result, float maxValue)
+            internal static void ScaleValues(Span<float> values, float maxValue)
             {
-                var maximum = 1 / maxValue;
-                var scale = new Vector4(maximum, maximum, maximum, 1F);
+                Span<Vector4> vecValues = MemoryMarshal.Cast<float, Vector4>(values);
 
-                ref float sBase = ref MemoryMarshal.GetReference(values.Component0);
-                ref Vector4 dBase = ref MemoryMarshal.GetReference(result);
+                var scaleVector = new Vector4(1 / maxValue);
 
-                for (int i = 0; i < result.Length; i++)
+                for (int i = 0; i < vecValues.Length; i++)
                 {
-                    var v = new Vector4(Unsafe.Add(ref sBase, i));
-                    v.W = 1f;
-                    v *= scale;
-                    Unsafe.Add(ref dBase, i) = v;
+                    vecValues[i] *= scaleVector;
+                }
+
+                values = values.Slice(vecValues.Length * 4);
+                if (!values.IsEmpty)
+                {
+                    float scaleValue = 1f / maxValue;
+                    values[0] *= scaleValue;
+
+                    if ((uint)values.Length > 1)
+                    {
+                        values[1] *= scaleValue;
+
+                        if ((uint)values.Length > 2)
+                        {
+                            values[2] *= scaleValue;
+                        }
+                    }
                 }
             }
         }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromRgbAvx2.cs

@@ -22,7 +22,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            protected override void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result)
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
             {
 #if SUPPORTS_RUNTIME_INTRINSICS
                 ref Vector256<float> rBase =
@@ -32,41 +32,23 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                 ref Vector256<float> bBase =
                                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component2));
 
-                ref Vector256<float> resultBase =
-                    ref Unsafe.As<Vector4, Vector256<float>>(ref MemoryMarshal.GetReference(result));
-
                 // Used for the color conversion
                 var scale = Vector256.Create(1 / this.MaximumValue);
-                var one = Vector256.Create(1F);
-
-                // Used for packing
-                ref byte control = ref MemoryMarshal.GetReference(HwIntrinsics.PermuteMaskEvenOdd8x32);
-                Vector256<int> vcontrol = Unsafe.As<byte, Vector256<int>>(ref control);
-
-                int n = result.Length / 8;
-                for (int i = 0; i < n; i++)
+                nint n = values.Component0.Length / 8;
+                for (nint i = 0; i < n; i++)
                 {
-                    Vector256<float> r = Avx.Multiply(Avx2.PermuteVar8x32(Unsafe.Add(ref rBase, i), vcontrol), scale);
-                    Vector256<float> g = Avx.Multiply(Avx2.PermuteVar8x32(Unsafe.Add(ref gBase, i), vcontrol), scale);
-                    Vector256<float> b = Avx.Multiply(Avx2.PermuteVar8x32(Unsafe.Add(ref bBase, i), vcontrol), scale);
-
-                    Vector256<float> rgLo = Avx.UnpackLow(r, g);
-                    Vector256<float> boLo = Avx.UnpackLow(b, one);
-                    Vector256<float> rgHi = Avx.UnpackHigh(r, g);
-                    Vector256<float> boHi = Avx.UnpackHigh(b, one);
-
-                    ref Vector256<float> destination = ref Unsafe.Add(ref resultBase, i * 4);
-
-                    destination = Avx.Shuffle(rgLo, boLo, 0b01_00_01_00);
-                    Unsafe.Add(ref destination, 1) = Avx.Shuffle(rgLo, boLo, 0b11_10_11_10);
-                    Unsafe.Add(ref destination, 2) = Avx.Shuffle(rgHi, boHi, 0b01_00_01_00);
-                    Unsafe.Add(ref destination, 3) = Avx.Shuffle(rgHi, boHi, 0b11_10_11_10);
+                    ref Vector256<float> r = ref Unsafe.Add(ref rBase, i);
+                    ref Vector256<float> g = ref Unsafe.Add(ref gBase, i);
+                    ref Vector256<float> b = ref Unsafe.Add(ref bBase, i);
+                    r = Avx.Multiply(r, scale);
+                    g = Avx.Multiply(g, scale);
+                    b = Avx.Multiply(b, scale);
                 }
 #endif
             }
 
-            protected override void ConvertCore(in ComponentValues values, Span<Vector4> result) =>
-                FromRgbBasic.ConvertCore(values, result, this.MaximumValue);
+            protected override void ConvertCoreInplace(in ComponentValues values) =>
+                FromRgbBasic.ConvertCoreInplace(values, this.MaximumValue);
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromRgbBasic.cs

@@ -3,6 +3,7 @@
 
 using System;
 using System.Numerics;
+using System.Runtime.InteropServices;
 
 namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
 {
@@ -15,36 +16,16 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            public override void ConvertToRgba(in ComponentValues values, Span<Vector4> result)
+            public override void ConvertToRgbInplace(in ComponentValues values)
             {
-                ConvertCore(values, result, this.MaximumValue);
+                ConvertCoreInplace(values, this.MaximumValue);
             }
 
-            internal static void ConvertCore(in ComponentValues values, Span<Vector4> result, float maxValue)
+            internal static void ConvertCoreInplace(ComponentValues values, float maxValue)
             {
-                ReadOnlySpan<float> rVals = values.Component0;
-                ReadOnlySpan<float> gVals = values.Component1;
-                ReadOnlySpan<float> bVals = values.Component2;
-
-                var v = new Vector4(0, 0, 0, 1);
-
-                var maximum = 1 / maxValue;
-                var scale = new Vector4(maximum, maximum, maximum, 1F);
-
-                for (int i = 0; i < result.Length; i++)
-                {
-                    float r = rVals[i];
-                    float g = gVals[i];
-                    float b = bVals[i];
-
-                    v.X = r;
-                    v.Y = g;
-                    v.Z = b;
-
-                    v *= scale;
-
-                    result[i] = v;
-                }
+                FromGrayscaleBasic.ScaleValues(values.Component0, maxValue);
+                FromGrayscaleBasic.ScaleValues(values.Component1, maxValue);
+                FromGrayscaleBasic.ScaleValues(values.Component2, maxValue);
             }
         }
     }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromRgbVector8.cs

@@ -18,50 +18,32 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            protected override void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result)
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
             {
                 ref Vector<float> rBase =
-                                    ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component0));
+                    ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component0));
                 ref Vector<float> gBase =
-                                    ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component1));
+                    ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component1));
                 ref Vector<float> bBase =
-                                    ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component2));
-
-                ref Vector4Octet resultBase =
-                    ref Unsafe.As<Vector4, Vector4Octet>(ref MemoryMarshal.GetReference(result));
-
-                Vector4Pair rr = default;
-                Vector4Pair gg = default;
-                Vector4Pair bb = default;
-                ref Vector<float> rrRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref rr);
-                ref Vector<float> ggRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref gg);
-                ref Vector<float> bbRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref bb);
+                    ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component2));
 
                 var scale = new Vector<float>(1 / this.MaximumValue);
 
                 // Walking 8 elements at one step:
-                int n = result.Length / 8;
-                for (int i = 0; i < n; i++)
+                nint n = values.Component0.Length / 8;
+                for (nint i = 0; i < n; i++)
                 {
-                    Vector<float> r = Unsafe.Add(ref rBase, i);
-                    Vector<float> g = Unsafe.Add(ref gBase, i);
-                    Vector<float> b = Unsafe.Add(ref bBase, i);
+                    ref Vector<float> r = ref Unsafe.Add(ref rBase, i);
+                    ref Vector<float> g = ref Unsafe.Add(ref gBase, i);
+                    ref Vector<float> b = ref Unsafe.Add(ref bBase, i);
                     r *= scale;
                     g *= scale;
                     b *= scale;
-
-                    rrRefAsVector = r;
-                    ggRefAsVector = g;
-                    bbRefAsVector = b;
-
-                    // Collect (r0,r1...r8) (g0,g1...g8) (b0,b1...b8) vector values in the expected (r0,g0,g1,1), (r1,g1,g2,1) ... order:
-                    ref Vector4Octet destination = ref Unsafe.Add(ref resultBase, i);
-                    destination.Pack(ref rr, ref gg, ref bb);
                 }
             }
 
-            protected override void ConvertCore(in ComponentValues values, Span<Vector4> result) =>
-                FromRgbBasic.ConvertCore(values, result, this.MaximumValue);
+            protected override void ConvertCoreInplace(in ComponentValues values) =>
+                FromRgbBasic.ConvertCoreInplace(values, this.MaximumValue);
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromYCbCrAvx2.cs

@@ -23,19 +23,16 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            protected override void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result)
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
             {
-                #if SUPPORTS_RUNTIME_INTRINSICS
-                ref Vector256<float> yBase =
+#if SUPPORTS_RUNTIME_INTRINSICS
+                ref Vector256<float> c0Base =
                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component0));
-                ref Vector256<float> cbBase =
+                ref Vector256<float> c1Base =
                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component1));
-                ref Vector256<float> crBase =
+                ref Vector256<float> c2Base =
                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component2));
 
-                ref Vector256<float> resultBase =
-                    ref Unsafe.As<Vector4, Vector256<float>>(ref MemoryMarshal.GetReference(result));
-
                 // Used for the color conversion
                 var chromaOffset = Vector256.Create(-this.HalfValue);
                 var scale = Vector256.Create(1 / this.MaximumValue);
@@ -50,19 +47,19 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                 Vector256<int> vcontrol = Unsafe.As<byte, Vector256<int>>(ref control);
 
                 // Walking 8 elements at one step:
-                int n = result.Length / 8;
-                for (int i = 0; i < n; i++)
+                nint n = values.Component0.Length / 8;
+                for (nint i = 0; i < n; i++)
                 {
                     // y = yVals[i];
                     // cb = cbVals[i] - 128F;
                     // cr = crVals[i] - 128F;
-                    Vector256<float> y = Unsafe.Add(ref yBase, i);
-                    Vector256<float> cb = Avx.Add(Unsafe.Add(ref cbBase, i), chromaOffset);
-                    Vector256<float> cr = Avx.Add(Unsafe.Add(ref crBase, i), chromaOffset);
+                    ref Vector256<float> c0 = ref Unsafe.Add(ref c0Base, i);
+                    ref Vector256<float> c1 = ref Unsafe.Add(ref c1Base, i);
+                    ref Vector256<float> c2 = ref Unsafe.Add(ref c2Base, i);
 
-                    y = Avx2.PermuteVar8x32(y, vcontrol);
-                    cb = Avx2.PermuteVar8x32(cb, vcontrol);
-                    cr = Avx2.PermuteVar8x32(cr, vcontrol);
+                    Vector256<float> y = c0;
+                    Vector256<float> cb = Avx.Add(c1, chromaOffset);
+                    Vector256<float> cr = Avx.Add(c2, chromaOffset);
 
                     // r = y + (1.402F * cr);
                     // g = y - (0.344136F * cb) - (0.714136F * cr);
@@ -72,30 +69,19 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                     Vector256<float> g = HwIntrinsics.MultiplyAdd(HwIntrinsics.MultiplyAdd(y, cb, gCbMult), cr, gCrMult);
                     Vector256<float> b = HwIntrinsics.MultiplyAdd(y, cb, bCbMult);
 
-                    // TODO: We should be saving to RGBA not Vector4
                     r = Avx.Multiply(Avx.RoundToNearestInteger(r), scale);
                     g = Avx.Multiply(Avx.RoundToNearestInteger(g), scale);
                     b = Avx.Multiply(Avx.RoundToNearestInteger(b), scale);
 
-                    Vector256<float> vte = Avx.UnpackLow(r, b);
-                    Vector256<float> vto = Avx.UnpackLow(g, va);
-
-                    ref Vector256<float> destination = ref Unsafe.Add(ref resultBase, i * 4);
-
-                    destination = Avx.UnpackLow(vte, vto);
-                    Unsafe.Add(ref destination, 1) = Avx.UnpackHigh(vte, vto);
-
-                    vte = Avx.UnpackHigh(r, b);
-                    vto = Avx.UnpackHigh(g, va);
-
-                    Unsafe.Add(ref destination, 2) = Avx.UnpackLow(vte, vto);
-                    Unsafe.Add(ref destination, 3) = Avx.UnpackHigh(vte, vto);
+                    c0 = r;
+                    c1 = g;
+                    c2 = b;
                 }
 #endif
             }
 
-            protected override void ConvertCore(in ComponentValues values, Span<Vector4> result) =>
-                FromYCbCrBasic.ConvertCore(values, result, this.MaximumValue, this.HalfValue);
+            protected override void ConvertCoreInplace(in ComponentValues values) =>
+                FromYCbCrBasic.ConvertCoreInplace(values, this.MaximumValue, this.HalfValue);
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromYCbCrBasic.cs

@@ -15,35 +15,26 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            public override void ConvertToRgba(in ComponentValues values, Span<Vector4> result)
-            {
-                ConvertCore(values, result, this.MaximumValue, this.HalfValue);
-            }
+            public override void ConvertToRgbInplace(in ComponentValues values)
+                => ConvertCoreInplace(values, this.MaximumValue, this.HalfValue);
 
-            internal static void ConvertCore(in ComponentValues values, Span<Vector4> result, float maxValue, float halfValue)
+            internal static void ConvertCoreInplace(in ComponentValues values, float maxValue, float halfValue)
             {
-                // TODO: We can optimize a lot here with Vector<float> and SRCS.Unsafe()!
-                ReadOnlySpan<float> yVals = values.Component0;
-                ReadOnlySpan<float> cbVals = values.Component1;
-                ReadOnlySpan<float> crVals = values.Component2;
-
-                var v = new Vector4(0, 0, 0, 1);
+                Span<float> c0 = values.Component0;
+                Span<float> c1 = values.Component1;
+                Span<float> c2 = values.Component2;
 
-                var scale = new Vector4(1 / maxValue, 1 / maxValue, 1 / maxValue, 1F);
+                var scale = 1 / maxValue;
 
-                for (int i = 0; i < result.Length; i++)
+                for (int i = 0; i < c0.Length; i++)
                 {
-                    float y = yVals[i];
-                    float cb = cbVals[i] - halfValue;
-                    float cr = crVals[i] - halfValue;
-
-                    v.X = MathF.Round(y + (1.402F * cr), MidpointRounding.AwayFromZero);
-                    v.Y = MathF.Round(y - (0.344136F * cb) - (0.714136F * cr), MidpointRounding.AwayFromZero);
-                    v.Z = MathF.Round(y + (1.772F * cb), MidpointRounding.AwayFromZero);
-
-                    v *= scale;
+                    float y = c0[i];
+                    float cb = c1[i] - halfValue;
+                    float cr = c2[i] - halfValue;
 
-                    result[i] = v;
+                    c0[i] = MathF.Round(y + (1.402F * cr), MidpointRounding.AwayFromZero) * scale;
+                    c1[i] = MathF.Round(y - (0.344136F * cb) - (0.714136F * cr), MidpointRounding.AwayFromZero) * scale;
+                    c2[i] = MathF.Round(y + (1.772F * cb), MidpointRounding.AwayFromZero) * scale;
                 }
             }
         }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromYCbCrVector4.cs

@@ -20,58 +20,54 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
 
             protected override bool IsAvailable => SimdUtils.HasVector4;
 
-            protected override void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result)
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
             {
-                // TODO: Find a way to properly run & test this path on AVX2 PC-s! (Have I already mentioned that Vector<T> is terrible?)
-                DebugGuard.IsTrue(result.Length % 8 == 0, nameof(result), "result.Length should be divisible by 8!");
+                DebugGuard.IsTrue(values.Component0.Length % 8 == 0, nameof(values), "Length should be divisible by 8!");
 
-                ref Vector4Pair yBase =
+                ref Vector4Pair c0Base =
                     ref Unsafe.As<float, Vector4Pair>(ref MemoryMarshal.GetReference(values.Component0));
-                ref Vector4Pair cbBase =
+                ref Vector4Pair c1Base =
                     ref Unsafe.As<float, Vector4Pair>(ref MemoryMarshal.GetReference(values.Component1));
-                ref Vector4Pair crBase =
+                ref Vector4Pair c2Base =
                     ref Unsafe.As<float, Vector4Pair>(ref MemoryMarshal.GetReference(values.Component2));
 
-                ref Vector4Octet resultBase =
-                    ref Unsafe.As<Vector4, Vector4Octet>(ref MemoryMarshal.GetReference(result));
-
                 var chromaOffset = new Vector4(-this.HalfValue);
                 var maxValue = this.MaximumValue;
 
                 // Walking 8 elements at one step:
-                int n = result.Length / 8;
+                nint n = values.Component0.Length / 8;
 
-                for (int i = 0; i < n; i++)
+                for (nint i = 0; i < n; i++)
                 {
                     // y = yVals[i];
-                    Vector4Pair y = Unsafe.Add(ref yBase, i);
+                    ref Vector4Pair c0 = ref Unsafe.Add(ref c0Base, i);
 
                     // cb = cbVals[i] - halfValue);
-                    Vector4Pair cb = Unsafe.Add(ref cbBase, i);
-                    cb.AddInplace(chromaOffset);
+                    ref Vector4Pair c1 = ref Unsafe.Add(ref c1Base, i);
+                    c1.AddInplace(chromaOffset);
 
                     // cr = crVals[i] - halfValue;
-                    Vector4Pair cr = Unsafe.Add(ref crBase, i);
-                    cr.AddInplace(chromaOffset);
+                    ref Vector4Pair c2 = ref Unsafe.Add(ref c2Base, i);
+                    c2.AddInplace(chromaOffset);
 
                     // r = y + (1.402F * cr);
-                    Vector4Pair r = y;
-                    Vector4Pair tmp = cr;
+                    Vector4Pair r = c0;
+                    Vector4Pair tmp = c2;
                     tmp.MultiplyInplace(1.402F);
                     r.AddInplace(ref tmp);
 
                     // g = y - (0.344136F * cb) - (0.714136F * cr);
-                    Vector4Pair g = y;
-                    tmp = cb;
+                    Vector4Pair g = c0;
+                    tmp = c1;
                     tmp.MultiplyInplace(-0.344136F);
                     g.AddInplace(ref tmp);
-                    tmp = cr;
+                    tmp = c2;
                     tmp.MultiplyInplace(-0.714136F);
                     g.AddInplace(ref tmp);
 
                     // b = y + (1.772F * cb);
-                    Vector4Pair b = y;
-                    tmp = cb;
+                    Vector4Pair b = c0;
+                    tmp = c1;
                     tmp.MultiplyInplace(1.772F);
                     b.AddInplace(ref tmp);
 
@@ -79,14 +75,14 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                     g.RoundAndDownscalePreVector8(maxValue);
                     b.RoundAndDownscalePreVector8(maxValue);
 
-                    // Collect (r0,r1...r8) (g0,g1...g8) (b0,b1...b8) vector values in the expected (r0,g0,g1,1), (r1,g1,g2,1) ... order:
-                    ref Vector4Octet destination = ref Unsafe.Add(ref resultBase, i);
-                    destination.Pack(ref r, ref g, ref b);
+                    c0 = r;
+                    c1 = g;
+                    c2 = b;
                 }
             }
 
-            protected override void ConvertCore(in ComponentValues values, Span<Vector4> result) =>
-                FromYCbCrBasic.ConvertCore(values, result, this.MaximumValue, this.HalfValue);
+            protected override void ConvertCoreInplace(in ComponentValues values)
+                => FromYCbCrBasic.ConvertCoreInplace(values, this.MaximumValue, this.HalfValue);
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromYCbCrVector8.cs

@@ -19,41 +19,32 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            protected override void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result)
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
             {
-                ref Vector<float> yBase =
+                ref Vector<float> c0Base =
                     ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component0));
-                ref Vector<float> cbBase =
+                ref Vector<float> c1Base =
                     ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component1));
-                ref Vector<float> crBase =
+                ref Vector<float> c2Base =
                     ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component2));
 
-                ref Vector4Octet resultBase =
-                    ref Unsafe.As<Vector4, Vector4Octet>(ref MemoryMarshal.GetReference(result));
-
                 var chromaOffset = new Vector<float>(-this.HalfValue);
 
                 // Walking 8 elements at one step:
-                int n = result.Length / 8;
-
-                Vector4Pair rr = default;
-                Vector4Pair gg = default;
-                Vector4Pair bb = default;
-
-                ref Vector<float> rrRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref rr);
-                ref Vector<float> ggRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref gg);
-                ref Vector<float> bbRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref bb);
-
+                nint n = values.Component0.Length / 8;
                 var scale = new Vector<float>(1 / this.MaximumValue);
 
-                for (int i = 0; i < n; i++)
+                for (nint i = 0; i < n; i++)
                 {
                     // y = yVals[i];
                     // cb = cbVals[i] - 128F;
                     // cr = crVals[i] - 128F;
-                    Vector<float> y = Unsafe.Add(ref yBase, i);
-                    Vector<float> cb = Unsafe.Add(ref cbBase, i) + chromaOffset;
-                    Vector<float> cr = Unsafe.Add(ref crBase, i) + chromaOffset;
+                    ref Vector<float> c0 = ref Unsafe.Add(ref c0Base, i);
+                    ref Vector<float> c1 = ref Unsafe.Add(ref c1Base, i);
+                    ref Vector<float> c2 = ref Unsafe.Add(ref c2Base, i);
+                    Vector<float> y = Unsafe.Add(ref c0Base, i);
+                    Vector<float> cb = Unsafe.Add(ref c1Base, i) + chromaOffset;
+                    Vector<float> cr = Unsafe.Add(ref c2Base, i) + chromaOffset;
 
                     // r = y + (1.402F * cr);
                     // g = y - (0.344136F * cb) - (0.714136F * cr);
@@ -70,18 +61,14 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                     g *= scale;
                     b *= scale;
 
-                    rrRefAsVector = r;
-                    ggRefAsVector = g;
-                    bbRefAsVector = b;
-
-                    // Collect (r0,r1...r8) (g0,g1...g8) (b0,b1...b8) vector values in the expected (r0,g0,g1,1), (r1,g1,g2,1) ... order:
-                    ref Vector4Octet destination = ref Unsafe.Add(ref resultBase, i);
-                    destination.Pack(ref rr, ref gg, ref bb);
+                    c0 = r;
+                    c1 = g;
+                    c2 = b;
                 }
             }
 
-            protected override void ConvertCore(in ComponentValues values, Span<Vector4> result) =>
-                FromYCbCrBasic.ConvertCore(values, result, this.MaximumValue, this.HalfValue);
+            protected override void ConvertCoreInplace(in ComponentValues values) =>
+                FromYCbCrBasic.ConvertCoreInplace(values, this.MaximumValue, this.HalfValue);
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromYccKAvx2.cs

@@ -22,52 +22,42 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            protected override void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result)
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
             {
 #if SUPPORTS_RUNTIME_INTRINSICS
-                ref Vector256<float> yBase =
+                ref Vector256<float> c0Base =
                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component0));
-                ref Vector256<float> cbBase =
+                ref Vector256<float> c1Base =
                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component1));
-                ref Vector256<float> crBase =
+                ref Vector256<float> c2Base =
                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component2));
                 ref Vector256<float> kBase =
                     ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component3));
 
-                ref Vector256<float> resultBase =
-                    ref Unsafe.As<Vector4, Vector256<float>>(ref MemoryMarshal.GetReference(result));
-
                 // Used for the color conversion
                 var chromaOffset = Vector256.Create(-this.HalfValue);
-                var scale = Vector256.Create(1 / this.MaximumValue);
+                var scale = Vector256.Create(1 / (this.MaximumValue * this.MaximumValue));
                 var max = Vector256.Create(this.MaximumValue);
                 var rCrMult = Vector256.Create(1.402F);
                 var gCbMult = Vector256.Create(-0.344136F);
                 var gCrMult = Vector256.Create(-0.714136F);
                 var bCbMult = Vector256.Create(1.772F);
 
-                // Used for packing.
-                var va = Vector256.Create(1F);
-                ref byte control = ref MemoryMarshal.GetReference(HwIntrinsics.PermuteMaskEvenOdd8x32);
-                Vector256<int> vcontrol = Unsafe.As<byte, Vector256<int>>(ref control);
-
                 // Walking 8 elements at one step:
-                int n = result.Length / 8;
-                for (int i = 0; i < n; i++)
+                nint n = values.Component0.Length / 8;
+                for (nint i = 0; i < n; i++)
                 {
                     // y = yVals[i];
                     // cb = cbVals[i] - 128F;
                     // cr = crVals[i] - 128F;
                     // k = kVals[i] / 256F;
-                    Vector256<float> y = Unsafe.Add(ref yBase, i);
-                    Vector256<float> cb = Avx.Add(Unsafe.Add(ref cbBase, i), chromaOffset);
-                    Vector256<float> cr = Avx.Add(Unsafe.Add(ref crBase, i), chromaOffset);
-                    Vector256<float> k = Avx.Divide(Unsafe.Add(ref kBase, i), max);
-
-                    y = Avx2.PermuteVar8x32(y, vcontrol);
-                    cb = Avx2.PermuteVar8x32(cb, vcontrol);
-                    cr = Avx2.PermuteVar8x32(cr, vcontrol);
-                    k = Avx2.PermuteVar8x32(k, vcontrol);
+                    ref Vector256<float> c0 = ref Unsafe.Add(ref c0Base, i);
+                    ref Vector256<float> c1 = ref Unsafe.Add(ref c1Base, i);
+                    ref Vector256<float> c2 = ref Unsafe.Add(ref c2Base, i);
+                    Vector256<float> y = c0;
+                    Vector256<float> cb = Avx.Add(c1, chromaOffset);
+                    Vector256<float> cr = Avx.Add(c2, chromaOffset);
+                    Vector256<float> scaledK = Avx.Multiply(Unsafe.Add(ref kBase, i), scale);
 
                     // r = y + (1.402F * cr);
                     // g = y - (0.344136F * cb) - (0.714136F * cr);
@@ -82,29 +72,19 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                     g = Avx.Subtract(max, Avx.RoundToNearestInteger(g));
                     b = Avx.Subtract(max, Avx.RoundToNearestInteger(b));
 
-                    r = Avx.Multiply(Avx.Multiply(r, k), scale);
-                    g = Avx.Multiply(Avx.Multiply(g, k), scale);
-                    b = Avx.Multiply(Avx.Multiply(b, k), scale);
-
-                    Vector256<float> vte = Avx.UnpackLow(r, b);
-                    Vector256<float> vto = Avx.UnpackLow(g, va);
-
-                    ref Vector256<float> destination = ref Unsafe.Add(ref resultBase, i * 4);
-
-                    destination = Avx.UnpackLow(vte, vto);
-                    Unsafe.Add(ref destination, 1) = Avx.UnpackHigh(vte, vto);
-
-                    vte = Avx.UnpackHigh(r, b);
-                    vto = Avx.UnpackHigh(g, va);
+                    r = Avx.Multiply(r, scaledK);
+                    g = Avx.Multiply(g, scaledK);
+                    b = Avx.Multiply(b, scaledK);
 
-                    Unsafe.Add(ref destination, 2) = Avx.UnpackLow(vte, vto);
-                    Unsafe.Add(ref destination, 3) = Avx.UnpackHigh(vte, vto);
+                    c0 = r;
+                    c1 = g;
+                    c2 = b;
                 }
 #endif
             }
 
-            protected override void ConvertCore(in ComponentValues values, Span<Vector4> result) =>
-                FromYccKBasic.ConvertCore(values, result, this.MaximumValue, this.HalfValue);
+            protected override void ConvertCoreInplace(in ComponentValues values) =>
+                FromYccKBasic.ConvertCoreInplace(values, this.MaximumValue, this.HalfValue);
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromYccKBasic.cs

@@ -15,39 +15,30 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            public override void ConvertToRgba(in ComponentValues values, Span<Vector4> result)
-            {
-                ConvertCore(values, result, this.MaximumValue, this.HalfValue);
-            }
+            public override void ConvertToRgbInplace(in ComponentValues values) =>
+                ConvertCoreInplace(values, this.MaximumValue, this.HalfValue);
 
-            internal static void ConvertCore(in ComponentValues values, Span<Vector4> result, float maxValue, float halfValue)
+            internal static void ConvertCoreInplace(in ComponentValues values, float maxValue, float halfValue)
             {
-                // TODO: We can optimize a lot here with Vector<float> and SRCS.Unsafe()!
-                ReadOnlySpan<float> yVals = values.Component0;
-                ReadOnlySpan<float> cbVals = values.Component1;
-                ReadOnlySpan<float> crVals = values.Component2;
-                ReadOnlySpan<float> kVals = values.Component3;
+                Span<float> c0 = values.Component0;
+                Span<float> c1 = values.Component1;
+                Span<float> c2 = values.Component2;
+                Span<float> c3 = values.Component3;
 
                 var v = new Vector4(0, 0, 0, 1F);
 
-                var maximum = 1 / maxValue;
-                var scale = new Vector4(maximum, maximum, maximum, 1F);
+                var scale = 1 / (maxValue * maxValue);
 
-                for (int i = 0; i < result.Length; i++)
+                for (int i = 0; i < values.Component0.Length; i++)
                 {
-                    float y = yVals[i];
-                    float cb = cbVals[i] - halfValue;
-                    float cr = crVals[i] - halfValue;
-                    float k = kVals[i] / maxValue;
-
-                    v.X = (maxValue - MathF.Round(y + (1.402F * cr), MidpointRounding.AwayFromZero)) * k;
-                    v.Y = (maxValue - MathF.Round(y - (0.344136F * cb) - (0.714136F * cr), MidpointRounding.AwayFromZero)) * k;
-                    v.Z = (maxValue - MathF.Round(y + (1.772F * cb), MidpointRounding.AwayFromZero)) * k;
-                    v.W = 1F;
-
-                    v *= scale;
-
-                    result[i] = v;
+                    float y = c0[i];
+                    float cb = c1[i] - halfValue;
+                    float cr = c2[i] - halfValue;
+                    float scaledK = c3[i] * scale;
+
+                    c0[i] = (maxValue - MathF.Round(y + (1.402F * cr), MidpointRounding.AwayFromZero)) * scaledK;
+                    c1[i] = (maxValue - MathF.Round(y - (0.344136F * cb) - (0.714136F * cr), MidpointRounding.AwayFromZero)) * scaledK;
+                    c2[i] = (maxValue - MathF.Round(y + (1.772F * cb), MidpointRounding.AwayFromZero)) * scaledK;
                 }
             }
         }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.FromYccKVector8.cs

@@ -18,46 +18,39 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             {
             }
 
-            protected override void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result)
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
             {
-                ref Vector<float> yBase =
+                ref Vector<float> c0Base =
                     ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component0));
-                ref Vector<float> cbBase =
+                ref Vector<float> c1Base =
                     ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component1));
-                ref Vector<float> crBase =
+                ref Vector<float> c2Base =
                     ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component2));
                 ref Vector<float> kBase =
                     ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component3));
 
-                ref Vector4Octet resultBase =
-                    ref Unsafe.As<Vector4, Vector4Octet>(ref MemoryMarshal.GetReference(result));
-
                 var chromaOffset = new Vector<float>(-this.HalfValue);
 
                 // Walking 8 elements at one step:
-                int n = result.Length / 8;
-
-                Vector4Pair rr = default;
-                Vector4Pair gg = default;
-                Vector4Pair bb = default;
-
-                ref Vector<float> rrRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref rr);
-                ref Vector<float> ggRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref gg);
-                ref Vector<float> bbRefAsVector = ref Unsafe.As<Vector4Pair, Vector<float>>(ref bb);
+                nint n = values.Component0.Length / 8;
 
-                var scale = new Vector<float>(1 / this.MaximumValue);
                 var max = new Vector<float>(this.MaximumValue);
+                var scale = new Vector<float>(1f) / (max * max);
 
-                for (int i = 0; i < n; i++)
+                for (nint i = 0; i < n; i++)
                 {
                     // y = yVals[i];
                     // cb = cbVals[i] - 128F;
                     // cr = crVals[i] - 128F;
                     // k = kVals[i] / 256F;
-                    Vector<float> y = Unsafe.Add(ref yBase, i);
-                    Vector<float> cb = Unsafe.Add(ref cbBase, i) + chromaOffset;
-                    Vector<float> cr = Unsafe.Add(ref crBase, i) + chromaOffset;
-                    Vector<float> k = Unsafe.Add(ref kBase, i) / max;
+                    ref Vector<float> c0 = ref Unsafe.Add(ref c0Base, i);
+                    ref Vector<float> c1 = ref Unsafe.Add(ref c1Base, i);
+                    ref Vector<float> c2 = ref Unsafe.Add(ref c2Base, i);
+
+                    Vector<float> y = c0;
+                    Vector<float> cb = c1 + chromaOffset;
+                    Vector<float> cr = c2 + chromaOffset;
+                    Vector<float> scaledK = Unsafe.Add(ref kBase, i) * scale;
 
                     // r = y + (1.402F * cr);
                     // g = y - (0.344136F * cb) - (0.714136F * cr);
@@ -67,25 +60,18 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                     Vector<float> g = y - (cb * new Vector<float>(0.344136F)) - (cr * new Vector<float>(0.714136F));
                     Vector<float> b = y + (cb * new Vector<float>(1.772F));
 
-                    r = (max - r.FastRound()) * k;
-                    g = (max - g.FastRound()) * k;
-                    b = (max - b.FastRound()) * k;
-                    r *= scale;
-                    g *= scale;
-                    b *= scale;
-
-                    rrRefAsVector = r;
-                    ggRefAsVector = g;
-                    bbRefAsVector = b;
+                    r = (max - r.FastRound()) * scaledK;
+                    g = (max - g.FastRound()) * scaledK;
+                    b = (max - b.FastRound()) * scaledK;
 
-                    // Collect (r0,r1...r8) (g0,g1...g8) (b0,b1...b8) vector values in the expected (r0,g0,g1,1), (r1,g1,g2,1) ... order:
-                    ref Vector4Octet destination = ref Unsafe.Add(ref resultBase, i);
-                    destination.Pack(ref rr, ref gg, ref bb);
+                    c0 = r;
+                    c1 = g;
+                    c2 = b;
                 }
             }
 
-            protected override void ConvertCore(in ComponentValues values, Span<Vector4> result) =>
-                FromYccKBasic.ConvertCore(values, result, this.MaximumValue, this.HalfValue);
+            protected override void ConvertCoreInplace(in ComponentValues values) =>
+                FromYccKBasic.ConvertCoreInplace(values, this.MaximumValue, this.HalfValue);
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.VectorizedJpegColorConverter.cs

@@ -18,10 +18,11 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                 this.vectorSize = vectorSize;
             }
 
-            public sealed override void ConvertToRgba(in ComponentValues values, Span<Vector4> result)
+            public override void ConvertToRgbInplace(in ComponentValues values)
             {
-                int remainder = result.Length % this.vectorSize;
-                int simdCount = result.Length - remainder;
+                int length = values.Component0.Length;
+                int remainder = values.Component0.Length % this.vectorSize;
+                int simdCount = length - remainder;
                 if (simdCount > 0)
                 {
                     // This implementation is actually AVX specific.
@@ -32,15 +33,15 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                             "This converter can be used only on architecture having 256 byte floating point SIMD registers!");
                     }
 
-                    this.ConvertCoreVectorized(values.Slice(0, simdCount), result.Slice(0, simdCount));
+                    this.ConvertCoreVectorizedInplace(values.Slice(0, simdCount));
                 }
 
-                this.ConvertCore(values.Slice(simdCount, remainder), result.Slice(simdCount, remainder));
+                this.ConvertCoreInplace(values.Slice(simdCount, remainder));
             }
 
-            protected abstract void ConvertCoreVectorized(in ComponentValues values, Span<Vector4> result);
+            protected virtual void ConvertCoreVectorizedInplace(in ComponentValues values) => throw new NotImplementedException();
 
-            protected abstract void ConvertCore(in ComponentValues values, Span<Vector4> result);
+            protected virtual void ConvertCoreInplace(in ComponentValues values) => throw new NotImplementedException();
         }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/ColorConverters/JpegColorConverter.cs

@@ -76,11 +76,10 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
         }
 
         /// <summary>
-        /// He implementation of the conversion.
+        /// Converts planar jpeg component values in <paramref name="values"/> to RGB color space inplace.
         /// </summary>
-        /// <param name="values">The input as a stack-only <see cref="ComponentValues"/> struct</param>
-        /// <param name="result">The destination buffer of <see cref="Vector4"/> values</param>
-        public abstract void ConvertToRgba(in ComponentValues values, Span<Vector4> result);
+        /// <param name="values">The input/ouptut as a stack-only <see cref="ComponentValues"/> struct</param>
+        public abstract void ConvertToRgbInplace(in ComponentValues values);
 
         /// <summary>
         /// Returns the <see cref="JpegColorConverter"/>s for all supported colorspaces and precisions.
@@ -181,22 +180,22 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
             /// <summary>
             /// The component 0 (eg. Y)
             /// </summary>
-            public readonly ReadOnlySpan<float> Component0;
+            public readonly Span<float> Component0;
 
             /// <summary>
-            /// The component 1 (eg. Cb)
+            /// The component 1 (eg. Cb). In case of grayscale, it points to <see cref="Component0"/>.
             /// </summary>
-            public readonly ReadOnlySpan<float> Component1;
+            public readonly Span<float> Component1;
 
             /// <summary>
-            /// The component 2 (eg. Cr)
+            /// The component 2 (eg. Cr). In case of grayscale, it points to <see cref="Component0"/>.
             /// </summary>
-            public readonly ReadOnlySpan<float> Component2;
+            public readonly Span<float> Component2;
 
             /// <summary>
             /// The component 4
             /// </summary>
-            public readonly ReadOnlySpan<float> Component3;
+            public readonly Span<float> Component3;
 
             /// <summary>
             /// Initializes a new instance of the <see cref="ComponentValues"/> struct.
@@ -208,30 +207,19 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                 this.ComponentCount = componentBuffers.Count;
 
                 this.Component0 = componentBuffers[0].GetRowSpan(row);
-                this.Component1 = Span<float>.Empty;
-                this.Component2 = Span<float>.Empty;
-                this.Component3 = Span<float>.Empty;
-
-                if (this.ComponentCount > 1)
-                {
-                    this.Component1 = componentBuffers[1].GetRowSpan(row);
-                    if (this.ComponentCount > 2)
-                    {
-                        this.Component2 = componentBuffers[2].GetRowSpan(row);
-                        if (this.ComponentCount > 3)
-                        {
-                            this.Component3 = componentBuffers[3].GetRowSpan(row);
-                        }
-                    }
-                }
+
+                // In case of grayscale, Component1 and Component2 point to Component0 memory area
+                this.Component1 = this.ComponentCount > 1 ? componentBuffers[1].GetRowSpan(row) : this.Component0;
+                this.Component2 = this.ComponentCount > 2 ? componentBuffers[2].GetRowSpan(row) : this.Component0;
+                this.Component3 = this.ComponentCount > 3 ? componentBuffers[3].GetRowSpan(row) : Span<float>.Empty;
             }
 
-            private ComponentValues(
+            internal ComponentValues(
                 int componentCount,
-                ReadOnlySpan<float> c0,
-                ReadOnlySpan<float> c1,
-                ReadOnlySpan<float> c2,
-                ReadOnlySpan<float> c3)
+                Span<float> c0,
+                Span<float> c1,
+                Span<float> c2,
+                Span<float> c3)
             {
                 this.ComponentCount = componentCount;
                 this.Component0 = c0;
@@ -242,111 +230,13 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
 
             public ComponentValues Slice(int start, int length)
             {
-                ReadOnlySpan<float> c0 = this.Component0.Slice(start, length);
-                ReadOnlySpan<float> c1 = this.ComponentCount > 1 ? this.Component1.Slice(start, length) : ReadOnlySpan<float>.Empty;
-                ReadOnlySpan<float> c2 = this.ComponentCount > 2 ? this.Component2.Slice(start, length) : ReadOnlySpan<float>.Empty;
-                ReadOnlySpan<float> c3 = this.ComponentCount > 3 ? this.Component3.Slice(start, length) : ReadOnlySpan<float>.Empty;
+                Span<float> c0 = this.Component0.Slice(start, length);
+                Span<float> c1 = this.Component1.Length > 0 ? this.Component1.Slice(start, length) : Span<float>.Empty;
+                Span<float> c2 = this.Component2.Length > 0 ? this.Component2.Slice(start, length) : Span<float>.Empty;
+                Span<float> c3 = this.Component3.Length > 0 ? this.Component3.Slice(start, length) : Span<float>.Empty;
 
                 return new ComponentValues(this.ComponentCount, c0, c1, c2, c3);
             }
         }
-
-        internal struct Vector4Octet
-        {
-#pragma warning disable SA1132 // Do not combine fields
-            public Vector4 V0, V1, V2, V3, V4, V5, V6, V7;
-
-            /// <summary>
-            /// Pack (r0,r1...r7) (g0,g1...g7) (b0,b1...b7) vector values as (r0,g0,b0,1), (r1,g1,b1,1) ...
-            /// </summary>
-            public void Pack(ref Vector4Pair r, ref Vector4Pair g, ref Vector4Pair b)
-            {
-                this.V0.X = r.A.X;
-                this.V0.Y = g.A.X;
-                this.V0.Z = b.A.X;
-                this.V0.W = 1f;
-
-                this.V1.X = r.A.Y;
-                this.V1.Y = g.A.Y;
-                this.V1.Z = b.A.Y;
-                this.V1.W = 1f;
-
-                this.V2.X = r.A.Z;
-                this.V2.Y = g.A.Z;
-                this.V2.Z = b.A.Z;
-                this.V2.W = 1f;
-
-                this.V3.X = r.A.W;
-                this.V3.Y = g.A.W;
-                this.V3.Z = b.A.W;
-                this.V3.W = 1f;
-
-                this.V4.X = r.B.X;
-                this.V4.Y = g.B.X;
-                this.V4.Z = b.B.X;
-                this.V4.W = 1f;
-
-                this.V5.X = r.B.Y;
-                this.V5.Y = g.B.Y;
-                this.V5.Z = b.B.Y;
-                this.V5.W = 1f;
-
-                this.V6.X = r.B.Z;
-                this.V6.Y = g.B.Z;
-                this.V6.Z = b.B.Z;
-                this.V6.W = 1f;
-
-                this.V7.X = r.B.W;
-                this.V7.Y = g.B.W;
-                this.V7.Z = b.B.W;
-                this.V7.W = 1f;
-            }
-
-            /// <summary>
-            /// Pack (g0,g1...g7) vector values as (g0,g0,g0,1), (g1,g1,g1,1) ...
-            /// </summary>
-            public void Pack(ref Vector4Pair g)
-            {
-                this.V0.X = g.A.X;
-                this.V0.Y = g.A.X;
-                this.V0.Z = g.A.X;
-                this.V0.W = 1f;
-
-                this.V1.X = g.A.Y;
-                this.V1.Y = g.A.Y;
-                this.V1.Z = g.A.Y;
-                this.V1.W = 1f;
-
-                this.V2.X = g.A.Z;
-                this.V2.Y = g.A.Z;
-                this.V2.Z = g.A.Z;
-                this.V2.W = 1f;
-
-                this.V3.X = g.A.W;
-                this.V3.Y = g.A.W;
-                this.V3.Z = g.A.W;
-                this.V3.W = 1f;
-
-                this.V4.X = g.B.X;
-                this.V4.Y = g.B.X;
-                this.V4.Z = g.B.X;
-                this.V4.W = 1f;
-
-                this.V5.X = g.B.Y;
-                this.V5.Y = g.B.Y;
-                this.V5.Z = g.B.Y;
-                this.V5.W = 1f;
-
-                this.V6.X = g.B.Z;
-                this.V6.Y = g.B.Z;
-                this.V6.Z = g.B.Z;
-                this.V6.W = 1f;
-
-                this.V7.X = g.B.W;
-                this.V7.Y = g.B.W;
-                this.V7.Z = g.B.W;
-                this.V7.W = 1f;
-            }
-        }
     }
 }

src/ImageSharp/Formats/Jpeg/Components/Decoder/HuffmanScanBuffer.cs

@@ -80,7 +80,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
         [MethodImpl(InliningOptions.ShortMethod)]
         public bool HasBadMarker() => this.Marker != JpegConstants.Markers.XFF && !this.HasRestartMarker();
 
-        [MethodImpl(InliningOptions.ShortMethod)]
+        [MethodImpl(InliningOptions.AlwaysInline)]
         public void FillBuffer()
         {
             // Attempt to load at least the minimum number of required bits into the buffer.
@@ -130,7 +130,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
         [MethodImpl(InliningOptions.ShortMethod)]
         public int PeekBits(int nbits) => (int)ExtractBits(this.data, this.remainingBits - nbits, nbits);
 
-        [MethodImpl(InliningOptions.ShortMethod)]
+        [MethodImpl(InliningOptions.AlwaysInline)]
         private static ulong ExtractBits(ulong value, int offset, int size) => (value >> offset) & (ulong)((1 << size) - 1);
 
         [MethodImpl(InliningOptions.ShortMethod)]
@@ -207,7 +207,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
             }
         }
 
-        [MethodImpl(InliningOptions.ShortMethod)]
+        [MethodImpl(InliningOptions.AlwaysInline)]
         private int ReadStream()
         {
             int value = this.badData ? 0 : this.stream.ReadByte();

src/ImageSharp/Formats/Jpeg/Components/Decoder/SpectralConverter{TPixel}.cs

@@ -22,7 +22,10 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
 
         private JpegColorConverter colorConverter;
 
-        private IMemoryOwner<Vector4> rgbaBuffer;
+        // private IMemoryOwner<Vector4> rgbaBuffer;
+        private IMemoryOwner<byte> rgbBuffer;
+
+        private IMemoryOwner<TPixel> paddedProxyPixelRow;
 
         private Buffer2D<TPixel> pixelBuffer;
 
@@ -40,23 +43,20 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
 
         private bool Converted => this.pixelRowCounter >= this.pixelBuffer.Height;
 
-        public Buffer2D<TPixel> PixelBuffer
+        public Buffer2D<TPixel> GetPixelBuffer()
         {
-            get
+            if (!this.Converted)
             {
-                if (!this.Converted)
-                {
-                    int steps = (int)Math.Ceiling(this.pixelBuffer.Height / (float)this.pixelRowsPerStep);
+                int steps = (int)Math.Ceiling(this.pixelBuffer.Height / (float)this.pixelRowsPerStep);
 
-                    for (int step = 0; step < steps; step++)
-                    {
-                        this.cancellationToken.ThrowIfCancellationRequested();
-                        this.ConvertNextStride(step);
-                    }
+                for (int step = 0; step < steps; step++)
+                {
+                    this.cancellationToken.ThrowIfCancellationRequested();
+                    this.ConvertNextStride(step);
                 }
-
-                return this.pixelBuffer;
             }
+
+            return this.pixelBuffer;
         }
 
         /// <inheritdoc/>
@@ -72,7 +72,8 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
             this.pixelRowsPerStep = this.blockRowsPerStep * blockPixelHeight;
 
             // pixel buffer for resulting image
-            this.pixelBuffer = allocator.Allocate2D<TPixel>(frame.PixelWidth, frame.PixelHeight, AllocationOptions.Clean);
+            this.pixelBuffer = allocator.Allocate2D<TPixel>(frame.PixelWidth, frame.PixelHeight);
+            this.paddedProxyPixelRow = allocator.Allocate<TPixel>(frame.PixelWidth + 3);
 
             // component processors from spectral to Rgba32
             var postProcessorBufferSize = new Size(c0.SizeInBlocks.Width * 8, this.pixelRowsPerStep);
@@ -83,7 +84,8 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
             }
 
             // single 'stride' rgba32 buffer for conversion between spectral and TPixel
-            this.rgbaBuffer = allocator.Allocate<Vector4>(frame.PixelWidth);
+            // this.rgbaBuffer = allocator.Allocate<Vector4>(frame.PixelWidth);
+            this.rgbBuffer = allocator.Allocate<byte>(frame.PixelWidth * 3);
 
             // color converter from Rgba32 to TPixel
             this.colorConverter = this.GetColorConverter(frame, jpegData);
@@ -115,7 +117,8 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
                 }
             }
 
-            this.rgbaBuffer?.Dispose();
+            this.rgbBuffer?.Dispose();
+            this.paddedProxyPixelRow?.Dispose();
         }
 
         private void ConvertNextStride(int spectralStep)
@@ -129,17 +132,38 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
                 buffers[i] = this.componentProcessors[i].ColorBuffer;
             }
 
+            int width = this.pixelBuffer.Width;
+
             for (int yy = this.pixelRowCounter; yy < maxY; yy++)
             {
                 int y = yy - this.pixelRowCounter;
 
                 var values = new JpegColorConverter.ComponentValues(buffers, y);
-                this.colorConverter.ConvertToRgba(values, this.rgbaBuffer.GetSpan());
 
-                Span<TPixel> destRow = this.pixelBuffer.GetRowSpan(yy);
+                this.colorConverter.ConvertToRgbInplace(values);
+                values = values.Slice(0, width); // slice away Jpeg padding
 
-                // TODO: Investigate if slicing is actually necessary
-                PixelOperations<TPixel>.Instance.FromVector4Destructive(this.configuration, this.rgbaBuffer.GetSpan().Slice(0, destRow.Length), destRow);
+                Span<byte> r = this.rgbBuffer.Slice(0, width);
+                Span<byte> g = this.rgbBuffer.Slice(width, width);
+                Span<byte> b = this.rgbBuffer.Slice(width * 2, width);
+
+                SimdUtils.NormalizedFloatToByteSaturate(values.Component0, r);
+                SimdUtils.NormalizedFloatToByteSaturate(values.Component1, g);
+                SimdUtils.NormalizedFloatToByteSaturate(values.Component2, b);
+
+                // PackFromRgbPlanes expects the destination to be padded, so try to get padded span containing extra elements from the next row.
+                // If we can't get such a padded row because we are on a MemoryGroup boundary or at the last row,
+                // pack pixels to a temporary, padded proxy buffer, then copy the relevant values to the destination row.
+                if (this.pixelBuffer.TryGetPaddedRowSpan(yy, 3, out Span<TPixel> destRow))
+                {
+                    PixelOperations<TPixel>.Instance.PackFromRgbPlanes(this.configuration, r, g, b, destRow);
+                }
+                else
+                {
+                    Span<TPixel> proxyRow = this.paddedProxyPixelRow.GetSpan();
+                    PixelOperations<TPixel>.Instance.PackFromRgbPlanes(this.configuration, r, g, b, proxyRow);
+                    proxyRow.Slice(0, width).CopyTo(this.pixelBuffer.GetRowSpan(yy));
+                }
             }
 
             this.pixelRowCounter += this.pixelRowsPerStep;

src/ImageSharp/Formats/Jpeg/JpegDecoder.cs

@@ -28,7 +28,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg
 
         /// <inheritdoc />
         public Image Decode(Configuration configuration, Stream stream)
-            => this.Decode<Rgba32>(configuration, stream);
+            => this.Decode<Rgb24>(configuration, stream);
 
         /// <inheritdoc/>
         public Task<Image<TPixel>> DecodeAsync<TPixel>(Configuration configuration, Stream stream, CancellationToken cancellationToken)

src/ImageSharp/Formats/Jpeg/JpegDecoderCore.cs

@@ -185,7 +185,7 @@ namespace SixLabors.ImageSharp.Formats.Jpeg
             this.InitIptcProfile();
             this.InitDerivedMetadataProperties();
 
-            return new Image<TPixel>(this.Configuration, spectralConverter.PixelBuffer, this.Metadata);
+            return new Image<TPixel>(this.Configuration, spectralConverter.GetPixelBuffer(), this.Metadata);
         }
 
         /// <inheritdoc/>

src/ImageSharp/Formats/Tiff/Compression/Decompressors/JpegTiffCompression.cs

@@ -65,7 +65,7 @@ namespace SixLabors.ImageSharp.Formats.Tiff.Compression.Decompressors
                 scanDecoder.ResetInterval = 0;
                 jpegDecoder.ParseStream(stream, scanDecoder, CancellationToken.None);
 
-                CopyImageBytesToBuffer(buffer, spectralConverter.PixelBuffer);
+                CopyImageBytesToBuffer(buffer, spectralConverter.GetPixelBuffer());
             }
             else
             {

src/ImageSharp/Memory/Buffer2D{T}.cs

@@ -116,6 +116,36 @@ namespace SixLabors.ImageSharp.Memory
                 : this.GetRowMemorySlow(y).Span;
         }
 
+        internal bool TryGetPaddedRowSpan(int y, int padding, out Span<T> paddedSpan)
+        {
+            DebugGuard.MustBeGreaterThanOrEqualTo(y, 0, nameof(y));
+            DebugGuard.MustBeLessThan(y, this.Height, nameof(y));
+
+            int stride = this.Width + padding;
+            if (this.cachedMemory.Length > 0)
+            {
+                paddedSpan = this.cachedMemory.Span.Slice(y * this.Width);
+                if (paddedSpan.Length < stride)
+                {
+                    return false;
+                }
+
+                paddedSpan = paddedSpan.Slice(0, stride);
+                return true;
+            }
+
+            Memory<T> memory = this.FastMemoryGroup.GetRemainingSliceOfBuffer(y * (long)this.Width);
+
+            if (memory.Length < stride)
+            {
+                paddedSpan = default;
+                return false;
+            }
+
+            paddedSpan = memory.Span.Slice(0, stride);
+            return true;
+        }
+
         [MethodImpl(InliningOptions.ShortMethod)]
         internal ref T GetElementUnsafe(int x, int y)
         {

src/ImageSharp/Memory/DiscontiguousBuffers/MemoryGroupExtensions.cs

@@ -39,12 +39,8 @@ namespace SixLabors.ImageSharp.Memory
 
             int bufferIdx = (int)(start / group.BufferLength);
 
-            if (bufferIdx < 0)
-            {
-                throw new ArgumentOutOfRangeException(nameof(start));
-            }
-
-            if (bufferIdx >= group.Count)
+            // if (bufferIdx < 0 || bufferIdx >= group.Count)
+            if ((uint)bufferIdx >= group.Count)
             {
                 throw new ArgumentOutOfRangeException(nameof(start));
             }
@@ -61,6 +57,31 @@ namespace SixLabors.ImageSharp.Memory
             return memory.Slice(bufferStart, length);
         }
 
+        /// <summary>
+        /// Returns the slice of the buffer starting at global index <paramref name="start"/> that goes until the end of the buffer.
+        /// </summary>
+        internal static Memory<T> GetRemainingSliceOfBuffer<T>(this IMemoryGroup<T> group, long start)
+            where T : struct
+        {
+            Guard.NotNull(group, nameof(group));
+            Guard.IsTrue(group.IsValid, nameof(group), "Group must be valid!");
+            Guard.MustBeLessThan(start, group.TotalLength, nameof(start));
+
+            int bufferIdx = (int)(start / group.BufferLength);
+
+            // if (bufferIdx < 0 || bufferIdx >= group.Count)
+            if ((uint)bufferIdx >= group.Count)
+            {
+                throw new ArgumentOutOfRangeException(nameof(start));
+            }
+
+            int bufferStart = (int)(start % group.BufferLength);
+
+            Memory<T> memory = group[bufferIdx];
+
+            return memory.Slice(bufferStart);
+        }
+
         internal static void CopyTo<T>(this IMemoryGroup<T> source, Span<T> target)
             where T : struct
         {

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

@@ -32,9 +32,7 @@ namespace SixLabors.ImageSharp.PixelFormats
             {
                 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!");
+                GuardPackFromRgbPlanes(greenChannel, blueChannel, destination, count);
 
                 SimdUtils.PackFromRgbPlanes(configuration, redChannel, greenChannel, blueChannel, destination);
             }

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

@@ -67,9 +67,7 @@ namespace SixLabors.ImageSharp.PixelFormats
             {
                 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, nameof(destination), "'destination' span should not be shorter than the source channels!");
+                GuardPackFromRgbPlanes(greenChannel, blueChannel, destination, count);
 
                 SimdUtils.PackFromRgbPlanes(configuration, redChannel, greenChannel, blueChannel, destination);
             }

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

@@ -181,11 +181,7 @@ namespace SixLabors.ImageSharp.PixelFormats
             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));
+            GuardPackFromRgbPlanes(greenChannel, blueChannel, destination, count);
 
             Rgb24 rgb24 = default;
             ref byte r = ref MemoryMarshal.GetReference(redChannel);
@@ -201,5 +197,13 @@ namespace SixLabors.ImageSharp.PixelFormats
                 Unsafe.Add(ref d, i).FromRgb24(rgb24);
             }
         }
+
+        [MethodImpl(InliningOptions.ShortMethod)]
+        internal static void GuardPackFromRgbPlanes(ReadOnlySpan<byte> greenChannel, ReadOnlySpan<byte> blueChannel, Span<TPixel> destination, int count)
+        {
+            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!");
+        }
     }
 }

src/ImageSharp/PixelFormats/Utils/Vector4Converters.cs

@@ -13,7 +13,7 @@ namespace SixLabors.ImageSharp.PixelFormats.Utils
         /// <summary>
         /// Apply modifiers used requested by ToVector4() conversion.
         /// </summary>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        [MethodImpl(InliningOptions.ShortMethod)]
         internal static void ApplyForwardConversionModifiers(Span<Vector4> vectors, PixelConversionModifiers modifiers)
         {
             if (modifiers.IsDefined(PixelConversionModifiers.SRgbCompand))
@@ -30,7 +30,7 @@ namespace SixLabors.ImageSharp.PixelFormats.Utils
         /// <summary>
         /// Apply modifiers used requested by FromVector4() conversion.
         /// </summary>
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
+        [MethodImpl(InliningOptions.ShortMethod)]
         internal static void ApplyBackwardConversionModifiers(Span<Vector4> vectors, PixelConversionModifiers modifiers)
         {
             if (modifiers.IsDefined(PixelConversionModifiers.Premultiply))

src/ImageSharp/Processing/Processors/Transforms/Resize/ResizeProcessor{TPixel}.cs

@@ -4,6 +4,7 @@
 using System;
 using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
+using SixLabors.ImageSharp.Formats;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -187,6 +188,11 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
             bool compand,
             bool premultiplyAlpha)
         {
+            PixelAlphaRepresentation? alphaRepresentation = PixelOperations<TPixel>.Instance.GetPixelTypeInfo()?.AlphaRepresentation;
+
+            // Premultiply only if alpha representation is unknown or Unassociated:
+            bool needsPremultiplication = alphaRepresentation == null || alphaRepresentation.Value == PixelAlphaRepresentation.Unassociated;
+            premultiplyAlpha &= needsPremultiplication;
             PixelConversionModifiers conversionModifiers = GetModifiers(compand, premultiplyAlpha);
 
             Buffer2DRegion<TPixel> sourceRegion = source.PixelBuffer.GetRegion(sourceRectangle);

tests/ImageSharp.Benchmarks/Codecs/Jpeg/CmykColorConversion.cs

@@ -19,7 +19,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromCmykBasic(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromCmykBasic(8).ConvertToRgbInplace(values);
         }
 
         [Benchmark]
@@ -27,7 +27,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromCmykVector8(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromCmykVector8(8).ConvertToRgbInplace(values);
         }
 
         [Benchmark]
@@ -35,7 +35,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromCmykAvx2(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromCmykAvx2(8).ConvertToRgbInplace(values);
         }
     }
 }

tests/ImageSharp.Benchmarks/Codecs/Jpeg/ColorConversionBenchmark.cs

@@ -19,13 +19,10 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
 
         protected Buffer2D<float>[] Input { get; private set; }
 
-        protected Vector4[] Output { get; private set; }
-
         [GlobalSetup]
         public void Setup()
         {
             this.Input = CreateRandomValues(this.componentCount, Count);
-            this.Output = new Vector4[Count];
         }
 
         [GlobalCleanup]

tests/ImageSharp.Benchmarks/Codecs/Jpeg/DecodeJpeg_ImageSpecific.cs

@@ -44,8 +44,8 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
             }
         }
 
-        [Benchmark(Baseline = true, Description = "Decode Jpeg - System.Drawing")]
-        public SDSize JpegSystemDrawing()
+        [Benchmark(Baseline = true)]
+        public SDSize SystemDrawing()
         {
             using (var memoryStream = new MemoryStream(this.jpegBytes))
             {
@@ -56,12 +56,12 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
             }
         }
 
-        [Benchmark(Description = "Decode Jpeg - ImageSharp")]
-        public Size JpegImageSharp()
+        [Benchmark]
+        public Size ImageSharp()
         {
             using (var memoryStream = new MemoryStream(this.jpegBytes))
             {
-                using (var image = Image.Load<Rgba32>(memoryStream, new JpegDecoder { IgnoreMetadata = true }))
+                using (var image = Image.Load(memoryStream, new JpegDecoder { IgnoreMetadata = true }))
                 {
                     return new Size(image.Width, image.Height);
                 }

tests/ImageSharp.Benchmarks/Codecs/Jpeg/GrayscaleColorConversion.cs

@@ -19,7 +19,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromGrayscaleBasic(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromGrayscaleBasic(8).ConvertToRgbInplace(values);
         }
 
         [Benchmark]
@@ -27,7 +27,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromGrayscaleAvx2(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromGrayscaleAvx2(8).ConvertToRgbInplace(values);
         }
     }
 }

tests/ImageSharp.Benchmarks/Codecs/Jpeg/RgbColorConversion.cs

@@ -19,7 +19,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromRgbBasic(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromRgbBasic(8).ConvertToRgbInplace(values);
         }
 
         [Benchmark]
@@ -27,7 +27,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromRgbVector8(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromRgbVector8(8).ConvertToRgbInplace(values);
         }
 
         [Benchmark]
@@ -35,7 +35,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromRgbAvx2(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromRgbAvx2(8).ConvertToRgbInplace(values);
         }
     }
 }

tests/ImageSharp.Benchmarks/Codecs/Jpeg/YCbCrColorConversion.cs

@@ -19,7 +19,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromYCbCrBasic(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromYCbCrBasic(8).ConvertToRgbInplace(values);
         }
 
         [Benchmark(Baseline = true)]
@@ -27,7 +27,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromYCbCrVector4(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromYCbCrVector4(8).ConvertToRgbInplace(values);
         }
 
         [Benchmark]
@@ -35,7 +35,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromYCbCrVector8(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromYCbCrVector8(8).ConvertToRgbInplace(values);
         }
 
         [Benchmark]
@@ -43,7 +43,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromYCbCrAvx2(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromYCbCrAvx2(8).ConvertToRgbInplace(values);
         }
     }
 }

tests/ImageSharp.Benchmarks/Codecs/Jpeg/YccKColorConverter.cs

@@ -19,7 +19,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromYccKBasic(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromYccKBasic(8).ConvertToRgbInplace(values);
         }
 
         [Benchmark]
@@ -27,7 +27,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromYccKVector8(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromYccKVector8(8).ConvertToRgbInplace(values);
         }
 
         [Benchmark]
@@ -35,7 +35,7 @@ namespace SixLabors.ImageSharp.Benchmarks.Codecs.Jpeg
         {
             var values = new JpegColorConverter.ComponentValues(this.Input, 0);
 
-            new JpegColorConverter.FromYccKAvx2(8).ConvertToRgba(values, this.Output);
+            new JpegColorConverter.FromYccKAvx2(8).ConvertToRgbInplace(values);
         }
     }
 }

tests/ImageSharp.Benchmarks/Config.cs

@@ -33,7 +33,6 @@ namespace SixLabors.ImageSharp.Benchmarks
         {
             public MultiFramework() => this.AddJob(
                     Job.Default.WithRuntime(ClrRuntime.Net472),
-                    Job.Default.WithRuntime(CoreRuntime.Core21),
                     Job.Default.WithRuntime(CoreRuntime.Core31));
         }
 
@@ -41,7 +40,6 @@ namespace SixLabors.ImageSharp.Benchmarks
         {
             public ShortMultiFramework() => this.AddJob(
                     Job.Default.WithRuntime(ClrRuntime.Net472).WithLaunchCount(1).WithWarmupCount(3).WithIterationCount(3),
-                    Job.Default.WithRuntime(CoreRuntime.Core21).WithLaunchCount(1).WithWarmupCount(3).WithIterationCount(3),
                     Job.Default.WithRuntime(CoreRuntime.Core31).WithLaunchCount(1).WithWarmupCount(3).WithIterationCount(3));
         }
 

tests/ImageSharp.Benchmarks/LoadResizeSave/LoadResizeSaveStressRunner.cs

@@ -31,7 +31,6 @@ namespace SixLabors.ImageSharp.Benchmarks.LoadResizeSave
 
     public class LoadResizeSaveStressRunner
     {
-        private const int ThumbnailSize = 150;
         private const int Quality = 75;
         private const string ImageSharp = nameof(ImageSharp);
         private const string SystemDrawing = nameof(SystemDrawing);
@@ -58,6 +57,8 @@ namespace SixLabors.ImageSharp.Benchmarks.LoadResizeSave
 
         public JpegKind Filter { get; set; }
 
+        public int ThumbnailSize { get; set; } = 150;
+
         private static readonly string[] ProgressiveFiles =
         {
             "ancyloscelis-apiformis-m-paraguay-face_2014-08-08-095255-zs-pmax_15046500892_o.jpg",

tests/ImageSharp.Benchmarks/Processing/Resize.cs

@@ -215,4 +215,35 @@ namespace SixLabors.ImageSharp.Benchmarks
         //                             SystemDrawing | Core |    Core |       3032 |      400 | 118.3 ms |  6.899 ms | 0.3781 ms |  1.00 |    0.00 |           - |           - |           - |                96 B |
         //  'ImageSharp, MaxDegreeOfParallelism = 1' | Core |    Core |       3032 |      400 | 122.4 ms | 15.069 ms | 0.8260 ms |  1.03 |    0.01 |           - |           - |           - |             15712 B |
     }
+
+    public class Resize_Bicubic_Compare_Rgba32_Rgb24
+    {
+        private Resize_Bicubic_Rgb24 rgb24;
+        private Resize_Bicubic_Rgba32 rgba32;
+
+        [GlobalSetup]
+        public void Setup()
+        {
+            this.rgb24 = new Resize_Bicubic_Rgb24();
+            this.rgb24.Setup();
+            this.rgba32 = new Resize_Bicubic_Rgba32();
+            this.rgba32.Setup();
+        }
+
+        [GlobalCleanup]
+        public void Cleanup()
+        {
+            this.rgb24.Cleanup();
+            this.rgba32.Cleanup();
+        }
+
+        [Benchmark]
+        public void SystemDrawing() => this.rgba32.SystemDrawing();
+
+        [Benchmark(Baseline = true)]
+        public void Rgba32() => this.rgba32.ImageSharp_P1();
+
+        [Benchmark]
+        public void Rgb24() => this.rgb24.ImageSharp_P1();
+    }
 }

tests/ImageSharp.Tests.ProfilingSandbox/LoadResizeSaveParallelMemoryStress.cs

@@ -18,7 +18,7 @@ namespace SixLabors.ImageSharp.Tests.ProfilingSandbox
             this.benchmarks = new LoadResizeSaveStressRunner()
             {
                 // MaxDegreeOfParallelism = 10,
-                // Filter = JpegKind.Baseline
+                Filter = JpegKind.Baseline,
             };
             this.benchmarks.Init();
         }
@@ -58,6 +58,7 @@ namespace SixLabors.ImageSharp.Tests.ProfilingSandbox
                         lrs.SystemDrawingBenchmarkParallel();
                         break;
                     case ConsoleKey.D2:
+                        Console.WriteLine($"Images: {lrs.benchmarks.Images.Length}");
                         lrs.ImageSharpBenchmarkParallel();
                         break;
                     case ConsoleKey.D3:

tests/ImageSharp.Tests/Formats/Jpg/JpegColorConverterTests.cs

@@ -355,13 +355,13 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
             JpegColorConverter converter = simd ? (JpegColorConverter)new JpegColorConverter.FromYCbCrVector4(8) : new JpegColorConverter.FromYCbCrBasic(8);
 
             // Warm up:
-            converter.ConvertToRgba(values, result);
+            converter.ConvertToRgbInplace(values);
 
             using (new MeasureGuard(this.Output, $"{converter.GetType().Name} x {times}"))
             {
                 for (int i = 0; i < times; i++)
                 {
-                    converter.ConvertToRgba(values, result);
+                    converter.ConvertToRgbInplace(values);
                 }
             }
         }
@@ -374,6 +374,7 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
             float maxVal = 255f)
         {
             var rnd = new Random(seed);
+
             var buffers = new Buffer2D<float>[componentCount];
             for (int i = 0; i < componentCount; i++)
             {
@@ -415,39 +416,48 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
             int resultBufferLength,
             int seed)
         {
-            JpegColorConverter.ComponentValues values = CreateRandomValues(componentCount, inputBufferLength, seed);
-            var result = new Vector4[resultBufferLength];
+            JpegColorConverter.ComponentValues original = CreateRandomValues(componentCount, inputBufferLength, seed);
+            JpegColorConverter.ComponentValues values = Copy(original);
 
-            converter.ConvertToRgba(values, result);
+            converter.ConvertToRgbInplace(values);
 
             for (int i = 0; i < resultBufferLength; i++)
             {
-                Validate(converter.ColorSpace, values, result, i);
+                Validate(converter.ColorSpace, original, values, i);
+            }
+
+            static JpegColorConverter.ComponentValues Copy(JpegColorConverter.ComponentValues values)
+            {
+                Span<float> c0 = values.Component0.ToArray();
+                Span<float> c1 = values.ComponentCount > 1 ? values.Component1.ToArray().AsSpan() : c0;
+                Span<float> c2 = values.ComponentCount > 2 ? values.Component2.ToArray().AsSpan() : c0;
+                Span<float> c3 = values.ComponentCount > 3 ? values.Component3.ToArray().AsSpan() : Span<float>.Empty;
+                return new JpegColorConverter.ComponentValues(values.ComponentCount, c0, c1, c2, c3);
             }
         }
 
         private static void Validate(
             JpegColorSpace colorSpace,
-            in JpegColorConverter.ComponentValues values,
-            Vector4[] result,
+            in JpegColorConverter.ComponentValues original,
+            in JpegColorConverter.ComponentValues result,
             int i)
         {
             switch (colorSpace)
             {
                 case JpegColorSpace.Grayscale:
-                    ValidateGrayScale(values, result, i);
+                    ValidateGrayScale(original, result, i);
                     break;
                 case JpegColorSpace.Ycck:
-                    ValidateCyyK(values, result, i);
+                    ValidateCyyK(original, result, i);
                     break;
                 case JpegColorSpace.Cmyk:
-                    ValidateCmyk(values, result, i);
+                    ValidateCmyk(original, result, i);
                     break;
                 case JpegColorSpace.RGB:
-                    ValidateRgb(values, result, i);
+                    ValidateRgb(original, result, i);
                     break;
                 case JpegColorSpace.YCbCr:
-                    ValidateYCbCr(values, result, i);
+                    ValidateYCbCr(original, result, i);
                     break;
                 default:
                     Assert.True(false, $"Colorspace {colorSpace} not supported!");
@@ -455,22 +465,20 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
             }
         }
 
-        private static void ValidateYCbCr(in JpegColorConverter.ComponentValues values, Vector4[] result, int i)
+        private static void ValidateYCbCr(in JpegColorConverter.ComponentValues values, in JpegColorConverter.ComponentValues result, int i)
         {
             float y = values.Component0[i];
             float cb = values.Component1[i];
             float cr = values.Component2[i];
             var ycbcr = new YCbCr(y, cb, cr);
 
-            Vector4 rgba = result[i];
-            var actual = new Rgb(rgba.X, rgba.Y, rgba.Z);
+            var actual = new Rgb(result.Component0[i], result.Component1[i], result.Component2[i]);
             var expected = ColorSpaceConverter.ToRgb(ycbcr);
 
             Assert.Equal(expected, actual, ColorSpaceComparer);
-            Assert.Equal(1, rgba.W);
         }
 
-        private static void ValidateCyyK(in JpegColorConverter.ComponentValues values, Vector4[] result, int i)
+        private static void ValidateCyyK(in JpegColorConverter.ComponentValues values, in JpegColorConverter.ComponentValues result, int i)
         {
             var v = new Vector4(0, 0, 0, 1F);
             var scale = new Vector4(1 / 255F, 1 / 255F, 1 / 255F, 1F);
@@ -489,39 +497,34 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
 
             v *= scale;
 
-            Vector4 rgba = result[i];
-            var actual = new Rgb(rgba.X, rgba.Y, rgba.Z);
+            var actual = new Rgb(result.Component0[i], result.Component1[i], result.Component2[i]);
             var expected = new Rgb(v.X, v.Y, v.Z);
 
             Assert.Equal(expected, actual, ColorSpaceComparer);
-            Assert.Equal(1, rgba.W);
         }
 
-        private static void ValidateRgb(in JpegColorConverter.ComponentValues values, Vector4[] result, int i)
+        private static void ValidateRgb(in JpegColorConverter.ComponentValues values, in JpegColorConverter.ComponentValues result, int i)
         {
             float r = values.Component0[i];
             float g = values.Component1[i];
             float b = values.Component2[i];
-            Vector4 rgba = result[i];
-            var actual = new Rgb(rgba.X, rgba.Y, rgba.Z);
+
+            var actual = new Rgb(result.Component0[i], result.Component1[i], result.Component2[i]);
             var expected = new Rgb(r / 255F, g / 255F, b / 255F);
 
             Assert.Equal(expected, actual, ColorSpaceComparer);
-            Assert.Equal(1, rgba.W);
         }
 
-        private static void ValidateGrayScale(in JpegColorConverter.ComponentValues values, Vector4[] result, int i)
+        private static void ValidateGrayScale(in JpegColorConverter.ComponentValues values, in JpegColorConverter.ComponentValues result, int i)
         {
             float y = values.Component0[i];
-            Vector4 rgba = result[i];
-            var actual = new Rgb(rgba.X, rgba.Y, rgba.Z);
+            var actual = new Rgb(result.Component0[i], result.Component0[i], result.Component0[i]);
             var expected = new Rgb(y / 255F, y / 255F, y / 255F);
 
             Assert.Equal(expected, actual, ColorSpaceComparer);
-            Assert.Equal(1, rgba.W);
         }
 
-        private static void ValidateCmyk(in JpegColorConverter.ComponentValues values, Vector4[] result, int i)
+        private static void ValidateCmyk(in JpegColorConverter.ComponentValues values, in JpegColorConverter.ComponentValues result, int i)
         {
             var v = new Vector4(0, 0, 0, 1F);
             var scale = new Vector4(1 / 255F, 1 / 255F, 1 / 255F, 1F);
@@ -538,12 +541,10 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
 
             v *= scale;
 
-            Vector4 rgba = result[i];
-            var actual = new Rgb(rgba.X, rgba.Y, rgba.Z);
+            var actual = new Rgb(result.Component0[i], result.Component1[i], result.Component2[i]);
             var expected = new Rgb(v.X, v.Y, v.Z);
 
             Assert.Equal(expected, actual, ColorSpaceComparer);
-            Assert.Equal(1, rgba.W);
         }
     }
 }

tests/ImageSharp.Tests/Formats/Jpg/JpegDecoderTests.Baseline.cs

@@ -13,9 +13,9 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
     public partial class JpegDecoderTests
     {
         [Theory]
-        [WithFileCollection(nameof(BaselineTestJpegs), PixelTypes.Rgba32, false)]
-        [WithFile(TestImages.Jpeg.Baseline.Calliphora, PixelTypes.Rgba32, true)]
-        [WithFile(TestImages.Jpeg.Baseline.Turtle420, PixelTypes.Rgba32, true)]
+        [WithFileCollection(nameof(BaselineTestJpegs), PixelTypes.Rgb24, false)]
+        [WithFile(TestImages.Jpeg.Baseline.Calliphora, PixelTypes.Rgb24, true)]
+        [WithFile(TestImages.Jpeg.Baseline.Turtle420, PixelTypes.Rgb24, true)]
         public void DecodeBaselineJpeg<TPixel>(TestImageProvider<TPixel> provider, bool enforceDiscontiguousBuffers)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -26,7 +26,7 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
 
                 if (!string.IsNullOrEmpty(nonContiguousBuffersStr))
                 {
-                    provider.LimitAllocatorBufferCapacity().InPixels(1000 * 8);
+                    provider.LimitAllocatorBufferCapacity().InPixels(16_000);
                 }
 
                 using Image<TPixel> image = provider.GetImage(JpegDecoder);

tests/ImageSharp.Tests/Formats/Jpg/JpegDecoderTests.Metadata.cs

@@ -82,7 +82,7 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
             using (var stream = new MemoryStream(testFile.Bytes, false))
             {
                 var decoder = new JpegDecoder();
-                using (Image<Rgba32> image = decoder.Decode<Rgba32>(Configuration.Default, stream))
+                using (Image image = decoder.Decode(Configuration.Default, stream))
                 {
                     ImageMetadata meta = image.Metadata;
                     Assert.Equal(xResolution, meta.HorizontalResolution);
@@ -129,7 +129,7 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
             var testFile = TestFile.Create(imagePath);
             using (var stream = new MemoryStream(testFile.Bytes, false))
             {
-                using (Image<Rgba32> image = JpegDecoder.Decode<Rgba32>(Configuration.Default, stream))
+                using (Image image = JpegDecoder.Decode(Configuration.Default, stream))
                 {
                     JpegMetadata meta = image.Metadata.GetJpegMetadata();
                     Assert.Equal(quality, meta.Quality);
@@ -158,11 +158,11 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
         }
 
         [Theory]
-        [WithFile(TestImages.Jpeg.Baseline.Floorplan, PixelTypes.Rgba32, JpegColorType.Luminance)]
-        [WithFile(TestImages.Jpeg.Baseline.Jpeg420Small, PixelTypes.Rgba32, JpegColorType.YCbCrRatio420)]
-        [WithFile(TestImages.Jpeg.Baseline.Jpeg444, PixelTypes.Rgba32, JpegColorType.YCbCrRatio444)]
-        [WithFile(TestImages.Jpeg.Baseline.JpegRgb, PixelTypes.Rgba32, JpegColorType.Rgb)]
-        [WithFile(TestImages.Jpeg.Baseline.Cmyk, PixelTypes.Rgba32, JpegColorType.Cmyk)]
+        [WithFile(TestImages.Jpeg.Baseline.Floorplan, PixelTypes.Rgb24, JpegColorType.Luminance)]
+        [WithFile(TestImages.Jpeg.Baseline.Jpeg420Small, PixelTypes.Rgb24, JpegColorType.YCbCrRatio420)]
+        [WithFile(TestImages.Jpeg.Baseline.Jpeg444, PixelTypes.Rgb24, JpegColorType.YCbCrRatio444)]
+        [WithFile(TestImages.Jpeg.Baseline.JpegRgb, PixelTypes.Rgb24, JpegColorType.Rgb)]
+        [WithFile(TestImages.Jpeg.Baseline.Cmyk, PixelTypes.Rgb24, JpegColorType.Cmyk)]
         public void Decode_DetectsCorrectColorType<TPixel>(TestImageProvider<TPixel> provider, JpegColorType expectedColorType)
             where TPixel : unmanaged, IPixel<TPixel>
         {

tests/ImageSharp.Tests/Formats/Jpg/JpegDecoderTests.Progressive.cs

@@ -15,7 +15,7 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
         public const string DecodeProgressiveJpegOutputName = "DecodeProgressiveJpeg";
 
         [Theory]
-        [WithFileCollection(nameof(ProgressiveTestJpegs), PixelTypes.Rgba32)]
+        [WithFileCollection(nameof(ProgressiveTestJpegs), PixelTypes.Rgb24)]
         public void DecodeProgressiveJpeg<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -30,17 +30,17 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
         }
 
         [Theory]
-        [WithFile(TestImages.Jpeg.Progressive.Progress, PixelTypes.Rgba32)]
-        public void DecodeProgressiveJpeg_WithLimitedAllocatorBufferCapacity(TestImageProvider<Rgba32> provider)
+        [WithFile(TestImages.Jpeg.Progressive.Progress, PixelTypes.Rgb24)]
+        public void DecodeProgressiveJpeg_WithLimitedAllocatorBufferCapacity(TestImageProvider<Rgb24> provider)
         {
             static void RunTest(string providerDump, string nonContiguousBuffersStr)
             {
-                TestImageProvider<Rgba32> provider =
-                    BasicSerializer.Deserialize<TestImageProvider<Rgba32>>(providerDump);
+                TestImageProvider<Rgb24> provider =
+                    BasicSerializer.Deserialize<TestImageProvider<Rgb24>>(providerDump);
 
                 provider.LimitAllocatorBufferCapacity().InBytesSqrt(200);
 
-                using Image<Rgba32> image = provider.GetImage(JpegDecoder);
+                using Image<Rgb24> image = provider.GetImage(JpegDecoder);
                 image.DebugSave(provider, nonContiguousBuffersStr);
 
                 provider.Utility.TestName = DecodeProgressiveJpegOutputName;

tests/ImageSharp.Tests/Formats/Jpg/JpegDecoderTests.cs

@@ -25,7 +25,7 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
     [Trait("Format", "Jpg")]
     public partial class JpegDecoderTests
     {
-        public const PixelTypes CommonNonDefaultPixelTypes = PixelTypes.Rgba32 | PixelTypes.Argb32 | PixelTypes.RgbaVector;
+        public const PixelTypes CommonNonDefaultPixelTypes = PixelTypes.Rgba32 | PixelTypes.Argb32 | PixelTypes.Bgr24 | PixelTypes.RgbaVector;
 
         private const float BaselineTolerance = 0.001F / 100;
 
@@ -85,6 +85,14 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
 
         public const string DecodeBaselineJpegOutputName = "DecodeBaselineJpeg";
 
+        [Fact]
+        public void Decode_NonGeneric_CreatesRgb24Image()
+        {
+            string file = Path.Combine(TestEnvironment.InputImagesDirectoryFullPath, TestImages.Jpeg.Baseline.Jpeg420Small);
+            using var image = Image.Load(file);
+            Assert.IsType<Image<Rgb24>>(image);
+        }
+
         [Theory]
         [WithFile(TestImages.Jpeg.Baseline.Calliphora, CommonNonDefaultPixelTypes)]
         public void JpegDecoder_IsNotBoundToSinglePixelType<TPixel>(TestImageProvider<TPixel> provider)

tests/ImageSharp.Tests/Formats/Jpg/SpectralToPixelConversionTests.cs

@@ -53,7 +53,7 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
             provider.Utility.TestName = JpegDecoderTests.DecodeBaselineJpegOutputName;
 
             // Comparison
-            using (Image<TPixel> image = new Image<TPixel>(Configuration.Default, converter.PixelBuffer, new ImageMetadata()))
+            using (Image<TPixel> image = new Image<TPixel>(Configuration.Default, converter.GetPixelBuffer(), new ImageMetadata()))
             using (Image<TPixel> referenceImage = provider.GetReferenceOutputImage<TPixel>(appendPixelTypeToFileName: false))
             {
                 ImageSimilarityReport report = ImageComparer.Exact.CompareImagesOrFrames(referenceImage, image);

tests/ImageSharp.Tests/Memory/Buffer2DTests.cs

@@ -117,6 +117,32 @@ namespace SixLabors.ImageSharp.Tests.Memory
             }
         }
 
+        [Theory]
+        [InlineData(10, 0, 0, 0)]
+        [InlineData(10, 0, 2, 0)]
+        [InlineData(10, 1, 2, 0)]
+        [InlineData(10, 1, 3, 0)]
+        [InlineData(10, 1, 5, -1)]
+        [InlineData(10, 2, 2, -1)]
+        [InlineData(10, 3, 2, 1)]
+        [InlineData(10, 4, 2, -1)]
+        [InlineData(30, 3, 2, 0)]
+        [InlineData(30, 4, 1, -1)]
+        public void TryGetPaddedRowSpanY(int bufferCapacity, int y, int padding, int expectedBufferIndex)
+        {
+            this.MemoryAllocator.BufferCapacityInBytes = bufferCapacity;
+            using Buffer2D<byte> buffer = this.MemoryAllocator.Allocate2D<byte>(3, 5);
+
+            bool expectSuccess = expectedBufferIndex >= 0;
+            bool success = buffer.TryGetPaddedRowSpan(y, padding, out Span<byte> paddedSpan);
+            Xunit.Assert.Equal(expectSuccess, success);
+            if (success)
+            {
+                int expectedSubBufferOffset = (3 * y) - (expectedBufferIndex * buffer.FastMemoryGroup.BufferLength);
+                Assert.SpanPointsTo(paddedSpan, buffer.FastMemoryGroup[expectedBufferIndex], expectedSubBufferOffset);
+            }
+        }
+
         public static TheoryData<int, int, int, int> GetRowSpanY_OutOfRange_Data = new TheoryData<int, int, int, int>()
         {
             { Big, 10, 8, -1 },

tests/ImageSharp.Tests/Processing/Processors/Transforms/ResizeTests.cs

@@ -19,9 +19,7 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
     public class ResizeTests
     {
         private const PixelTypes CommonNonDefaultPixelTypes =
-            PixelTypes.Rgba32 | PixelTypes.Bgra32 | PixelTypes.RgbaVector;
-
-        private const PixelTypes DefaultPixelType = PixelTypes.Rgba32;
+            PixelTypes.Rgba32 | PixelTypes.Bgra32 | PixelTypes.Bgr24 | PixelTypes.RgbaVector;
 
         public static readonly string[] AllResamplerNames = TestUtils.GetAllResamplerNames();
 
@@ -188,7 +186,7 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
         }
 
         [Theory]
-        [WithTestPatternImages(100, 100, DefaultPixelType)]
+        [WithTestPatternImages(100, 100, PixelTypes.Rgba32)]
         public void Resize_Compand<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -202,8 +200,8 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
         }
 
         [Theory]
-        [WithFile(TestImages.Png.Kaboom, DefaultPixelType, false)]
-        [WithFile(TestImages.Png.Kaboom, DefaultPixelType, true)]
+        [WithFile(TestImages.Png.Kaboom, PixelTypes.Rgba32, false)]
+        [WithFile(TestImages.Png.Kaboom, PixelTypes.Rgba32, true)]
         public void Resize_DoesNotBleedAlphaPixels<TPixel>(TestImageProvider<TPixel> provider, bool compand)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -217,8 +215,8 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
         }
 
         [Theory]
-        [WithFile(TestImages.Png.Kaboom, DefaultPixelType, false)]
-        [WithFile(TestImages.Png.Kaboom, DefaultPixelType, true)]
+        [WithFile(TestImages.Png.Kaboom, PixelTypes.Rgba32, false)]
+        [WithFile(TestImages.Png.Kaboom, PixelTypes.Rgba32, true)]
         public void Resize_PremultiplyAlpha<TPixel>(TestImageProvider<TPixel> provider, bool premultiplyAlpha)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -243,7 +241,7 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
         }
 
         [Theory]
-        [WithFile(TestImages.Gif.Giphy, DefaultPixelType)]
+        [WithFile(TestImages.Gif.Giphy, PixelTypes.Rgba32)]
         public void Resize_IsAppliedToAllFrames<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -265,7 +263,7 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32)]
         public void Resize_ThrowsForWrappedMemoryImage<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -283,10 +281,10 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType, 1)]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType, 4)]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType, 8)]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType, -1)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24, 1)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24, 4)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24, 8)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24, -1)]
         public void Resize_WorksWithAllParallelismLevels<TPixel>(
             TestImageProvider<TPixel> provider,
             int maxDegreeOfParallelism)
@@ -305,27 +303,27 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), nameof(AllResamplerNames), DefaultPixelType, 0.5f, null, null)]
+        [WithFileCollection(nameof(CommonTestImages), nameof(AllResamplerNames), PixelTypes.Rgba32 | PixelTypes.Rgb24, 0.5f, null, null)]
         [WithFileCollection(
             nameof(CommonTestImages),
             nameof(SmokeTestResamplerNames),
-            DefaultPixelType,
+            PixelTypes.Rgba32 | PixelTypes.Rgb24,
             0.3f,
             null,
             null)]
         [WithFileCollection(
             nameof(CommonTestImages),
             nameof(SmokeTestResamplerNames),
-            DefaultPixelType,
+            PixelTypes.Rgba32 | PixelTypes.Rgb24,
             1.8f,
             null,
             null)]
-        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 100, 100, DefaultPixelType, 0.5f, null, null)]
-        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 100, 100, DefaultPixelType, 1f, null, null)]
-        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 50, 50, DefaultPixelType, 8f, null, null)]
-        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 201, 199, DefaultPixelType, null, 100, 99)]
-        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 301, 1180, DefaultPixelType, null, 300, 480)]
-        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 49, 80, DefaultPixelType, null, 301, 100)]
+        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 100, 100, PixelTypes.Rgba32, 0.5f, null, null)]
+        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 100, 100, PixelTypes.Rgba32, 1f, null, null)]
+        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 50, 50, PixelTypes.Rgba32, 8f, null, null)]
+        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 201, 199, PixelTypes.Rgba32, null, 100, 99)]
+        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 301, 1180, PixelTypes.Rgba32, null, 300, 480)]
+        [WithTestPatternImages(nameof(SmokeTestResamplerNames), 49, 80, PixelTypes.Rgba32, null, 301, 100)]
         public void Resize_WorksWithAllResamplers<TPixel>(
             TestImageProvider<TPixel> provider,
             string samplerName,
@@ -382,7 +380,7 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeFromSourceRectangle<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -405,12 +403,12 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
                         false));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeHeightAndKeepAspect<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -419,12 +417,12 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
                 image.Mutate(x => x.Resize(0, image.Height / 3, false));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithTestPatternImages(10, 100, DefaultPixelType)]
+        [WithTestPatternImages(10, 100, PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeHeightCannotKeepAspectKeepsOnePixel<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -437,7 +435,7 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeWidthAndKeepAspect<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -446,12 +444,12 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
                 image.Mutate(x => x.Resize(image.Width / 3, 0, false));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithTestPatternImages(100, 10, DefaultPixelType)]
+        [WithTestPatternImages(100, 10, PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeWidthCannotKeepAspectKeepsOnePixel<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -464,7 +462,7 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeWithBoxPadMode<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -479,12 +477,12 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
                 image.Mutate(x => x.Resize(options));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeWithCropHeightMode<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -495,12 +493,12 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
                 image.Mutate(x => x.Resize(options));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeWithCropWidthMode<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -511,12 +509,12 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
                 image.Mutate(x => x.Resize(options));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithFile(TestImages.Jpeg.Issues.IncorrectResize1006, DefaultPixelType)]
+        [WithFile(TestImages.Jpeg.Issues.IncorrectResize1006, PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void CanResizeLargeImageWithCropMode<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -531,12 +529,12 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
                 image.Mutate(x => x.Resize(options));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeWithMaxMode<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -547,12 +545,12 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
                 image.Mutate(x => x.Resize(options));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeWithMinMode<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -560,21 +558,19 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
             {
                 var options = new ResizeOptions
                 {
-                    Size = new Size(
-                                          (int)Math.Round(image.Width * .75F),
-                                          (int)Math.Round(image.Height * .95F)),
+                    Size = new Size((int)Math.Round(image.Width * .75F), (int)Math.Round(image.Height * .95F)),
                     Mode = ResizeMode.Min
                 };
 
                 image.Mutate(x => x.Resize(options));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeWithPadMode<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -589,12 +585,12 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
                 image.Mutate(x => x.Resize(options));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithFileCollection(nameof(CommonTestImages), DefaultPixelType)]
+        [WithFileCollection(nameof(CommonTestImages), PixelTypes.Rgba32 | PixelTypes.Rgb24)]
         public void ResizeWithStretchMode<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {
@@ -609,14 +605,14 @@ namespace SixLabors.ImageSharp.Tests.Processing.Processors.Transforms
                 image.Mutate(x => x.Resize(options));
 
                 image.DebugSave(provider);
-                image.CompareToReferenceOutput(ValidatorComparer, provider);
+                image.CompareToReferenceOutput(ValidatorComparer, provider, appendPixelTypeToFileName: false);
             }
         }
 
         [Theory]
-        [WithFile(TestImages.Jpeg.Issues.ExifDecodeOutOfRange694, DefaultPixelType)]
-        [WithFile(TestImages.Jpeg.Issues.ExifGetString750Transform, DefaultPixelType)]
-        [WithFile(TestImages.Jpeg.Issues.ExifResize1049, DefaultPixelType)]
+        [WithFile(TestImages.Jpeg.Issues.ExifDecodeOutOfRange694, PixelTypes.Rgb24)]
+        [WithFile(TestImages.Jpeg.Issues.ExifGetString750Transform, PixelTypes.Rgb24)]
+        [WithFile(TestImages.Jpeg.Issues.ExifResize1049, PixelTypes.Rgb24)]
         public void CanResizeExifIssueImages<TPixel>(TestImageProvider<TPixel> provider)
             where TPixel : unmanaged, IPixel<TPixel>
         {

tests/Images/External/ReferenceOutput/ResizeTests/Resize_IsNotBoundToSinglePixelType_Bgr24_TestPattern50x50.png

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4a9940410cca3fe98a6d7aaf0e2184779f908c569a5a34f9965fb3a4f9e6fa8f
+size 1066