inplace ycck conversion

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

@@ -103,8 +103,72 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
 #endif
             }
 
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
+            {
+#if SUPPORTS_RUNTIME_INTRINSICS
+                ref Vector256<float> c0Base =
+                    ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component0));
+                ref Vector256<float> c1Base =
+                    ref Unsafe.As<float, Vector256<float>>(ref MemoryMarshal.GetReference(values.Component1));
+                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));
+
+                // Used for the color conversion
+                var chromaOffset = Vector256.Create(-this.HalfValue);
+                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);
+
+                // Walking 8 elements at one step:
+                int n = values.Component0.Length / 8;
+                for (int i = 0; i < n; i++)
+                {
+                    // y = yVals[i];
+                    // cb = cbVals[i] - 128F;
+                    // cr = crVals[i] - 128F;
+                    // k = kVals[i] / 256F;
+                    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);
+                    // b = y + (1.772F * cb);
+                    // Adding & multiplying 8 elements at one time:
+                    Vector256<float> r = HwIntrinsics.MultiplyAdd(y, cr, rCrMult);
+                    Vector256<float> g =
+                        HwIntrinsics.MultiplyAdd(HwIntrinsics.MultiplyAdd(y, cb, gCbMult), cr, gCrMult);
+                    Vector256<float> b = HwIntrinsics.MultiplyAdd(y, cb, bCbMult);
+
+                    r = Avx.Subtract(max, Avx.RoundToNearestInteger(r));
+                    g = Avx.Subtract(max, Avx.RoundToNearestInteger(g));
+                    b = Avx.Subtract(max, Avx.RoundToNearestInteger(b));
+
+                    r = Avx.Multiply(r, scaledK);
+                    g = Avx.Multiply(g, scaledK);
+                    b = Avx.Multiply(b, scaledK);
+
+                    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

@@ -20,6 +20,9 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                 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)
             {
                 // TODO: We can optimize a lot here with Vector<float> and SRCS.Unsafe()!
@@ -50,6 +53,30 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                     result[i] = v;
                 }
             }
+
+            internal static void ConvertCoreInplace(in ComponentValues values, float maxValue, float halfValue)
+            {
+                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 scale = 1 / (maxValue * maxValue);
+
+                for (int i = 0; i < values.Component0.Length; i++)
+                {
+                    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

@@ -84,8 +84,63 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder.ColorConverters
                 }
             }
 
+            protected override void ConvertCoreVectorizedInplace(in ComponentValues values)
+            {
+                ref Vector<float> c0Base =
+                    ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component0));
+                ref Vector<float> c1Base =
+                    ref Unsafe.As<float, Vector<float>>(ref MemoryMarshal.GetReference(values.Component1));
+                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));
+
+                var chromaOffset = new Vector<float>(-this.HalfValue);
+
+                // Walking 8 elements at one step:
+                int n = values.Component0.Length / 8;
+
+                var scale = new Vector<float>(1 / (this.MaximumValue * this.MaximumValue));
+                var max = new Vector<float>(this.MaximumValue);
+
+                for (int i = 0; i < n; i++)
+                {
+                    // y = yVals[i];
+                    // cb = cbVals[i] - 128F;
+                    // cr = crVals[i] - 128F;
+                    // k = kVals[i] / 256F;
+                    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);
+                    // b = y + (1.772F * cb);
+                    // Adding & multiplying 8 elements at one time:
+                    Vector<float> r = y + (cr * new Vector<float>(1.402F));
+                    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()) * scaledK;
+                    g = (max - g.FastRound()) * scaledK;
+                    b = (max - b.FastRound()) * scaledK;
+
+                    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);
         }
     }
 }