Imporved internal rgb -> rcbcr conversion api for 420 subsampling

src/ImageSharp/Formats/Jpeg/Components/Encoder/HuffmanScanEncoder.cs

@@ -146,29 +146,27 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Encoder
                         int yOff = (i & 2) * 4;
 
                         currentRows.Update(pixelBuffer, y + yOff);
-                        pixelConverter.Convert420(frame, x + xOff, y + yOff, ref currentRows, ref temporalBlocks[0], i);
+                        pixelConverter.Convert420(frame, x + xOff, y + yOff, ref currentRows, i);
 
                         prevDCY = this.WriteBlock(
                             QuantIndex.Luminance,
                             prevDCY,
-                            ref temporalBlocks[0],
+                            ref pixelConverter.Y,
                             ref luminanceQuantTable,
                             ref unzig);
                     }
 
-                    pixelConverter.ConvertCbCr(ref temporalBlocks[0], ref temporalBlocks[1]);
-
                     prevDCCb = this.WriteBlock(
                         QuantIndex.Chrominance,
                         prevDCCb,
-                        ref temporalBlocks[0],
+                        ref pixelConverter.Cb,
                         ref chrominanceQuantTable,
                         ref unzig);
 
                     prevDCCr = this.WriteBlock(
                         QuantIndex.Chrominance,
                         prevDCCr,
-                        ref temporalBlocks[1],
+                        ref pixelConverter.Cr,
                         ref chrominanceQuantTable,
                         ref unzig);
                 }

src/ImageSharp/Formats/Jpeg/Components/Encoder/RgbToYCbCrConverterVectorized.cs

@@ -126,12 +126,8 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Encoder
         /// </summary>
         /// <remarks>Total size of rgb span must be 200 bytes</remarks>
         /// <param name="rgbSpan">Span of rgb pixels with size of 64</param>
-        /// <param name="yBlock">8x8 destination matrix of Luminance(Y) converted data</param>
-        /// <param name="rAcc"></param>
-        /// <param name="gAcc"></param>
-        /// <param name="bAcc"></param>
-        /// <param name="idx"></param>
-        public static void Convert420(ReadOnlySpan<Rgb24> rgbSpan, ref Block8x8F yBlock, ref Block8x8F rAcc, ref Block8x8F gAcc, ref Block8x8F bAcc, int idx)
+        /// <param name="yBlock">8x8 destination matrix of Luminance(Y) converted data</param>ф
+        public static void Convert420(ReadOnlySpan<Rgb24> rgbSpan, ref Block8x8F yBlock, ref Block8x8F cbBlock, ref Block8x8F crBlock, int idx)
         {
             Debug.Assert(IsSupported, "AVX2 is required to run this converter");
 
@@ -152,9 +148,8 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Encoder
 
             int destOffset = (idx & 2) * 4 + (idx & 1);
 
-            ref Vector128<float> destRedRef   = ref Unsafe.Add(ref Unsafe.As<Block8x8F, Vector128<float>>(ref rAcc), destOffset);
-            ref Vector128<float> destGreenRef = ref Unsafe.Add(ref Unsafe.As<Block8x8F, Vector128<float>>(ref gAcc), destOffset);
-            ref Vector128<float> destBlueRef  = ref Unsafe.Add(ref Unsafe.As<Block8x8F, Vector128<float>>(ref bAcc), destOffset);
+            ref Vector128<float> destCbRef = ref Unsafe.Add(ref Unsafe.As<Block8x8F, Vector128<float>>(ref cbBlock), destOffset);
+            ref Vector128<float> destCrRef = ref Unsafe.Add(ref Unsafe.As<Block8x8F, Vector128<float>>(ref crBlock), destOffset);
 
             var extractToLanesMask = Unsafe.As<byte, Vector256<uint>>(ref MemoryMarshal.GetReference(MoveFirst24BytesToSeparateLanes));
             var extractRgbMask = Unsafe.As<byte, Vector256<byte>>(ref MemoryMarshal.GetReference(ExtractRgb));
@@ -192,20 +187,19 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Encoder
 
                 int localDestOffset = (i & 1) * 4;
 
-                // red
-                Vector256<float> twoLane = Scale_8x4_4x2(rDataLanes);
-                Unsafe.Add(ref destRedRef, localDestOffset) = twoLane.GetLower();
-                Unsafe.Add(ref destRedRef, localDestOffset + 2) = twoLane.GetUpper();
+                r = Scale_8x4_4x2(rDataLanes);
+                g = Scale_8x4_4x2(gDataLanes);
+                b = Scale_8x4_4x2(bDataLanes);
 
-                // green
-                twoLane = Scale_8x4_4x2(gDataLanes);
-                Unsafe.Add(ref destGreenRef, localDestOffset) = twoLane.GetLower();
-                Unsafe.Add(ref destGreenRef, localDestOffset + 2) = twoLane.GetUpper();
+                // 128F + ((-0.168736F * r) - (0.331264F * g) + (0.5F * b))
+                Vector256<float> cb = Avx.Add(f128, SimdUtils.HwIntrinsics.MultiplyAdd(SimdUtils.HwIntrinsics.MultiplyAdd(Avx.Multiply(f05, b), fn0331264, g), fn0168736, r));
+                Unsafe.Add(ref destCbRef, localDestOffset) = cb.GetLower();
+                Unsafe.Add(ref destCbRef, localDestOffset + 2) = cb.GetUpper();
 
-                // blue
-                twoLane = Scale_8x4_4x2(bDataLanes);
-                Unsafe.Add(ref destBlueRef, localDestOffset) = twoLane.GetLower();
-                Unsafe.Add(ref destBlueRef, localDestOffset + 2) = twoLane.GetUpper();
+                // 128F + ((0.5F * r) - (0.418688F * g) - (0.081312F * b))
+                Vector256<float> cr = Avx.Add(f128, SimdUtils.HwIntrinsics.MultiplyAdd(SimdUtils.HwIntrinsics.MultiplyAdd(Avx.Multiply(fn0081312F, b), fn0418688, g), f05, r));
+                Unsafe.Add(ref destCrRef, localDestOffset) = cr.GetLower();
+                Unsafe.Add(ref destCrRef, localDestOffset + 2) = cr.GetUpper();
             }
 #endif
         }

src/ImageSharp/Formats/Jpeg/Components/Encoder/YCbCrForwardConverter{TPixel}.cs

@@ -88,19 +88,15 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Encoder
         /// <summary>
         /// Converts a 8x8 image area inside 'pixels' at position (x,y) placing the result members of the structure (<see cref="Y"/>, <see cref="Cb"/>, <see cref="Cr"/>)
         /// </summary>
-        public void Convert420(ImageFrame<TPixel> frame, int x, int y, ref RowOctet<TPixel> currentRows, ref Block8x8F yBlock, int idx)
+        public void Convert420(ImageFrame<TPixel> frame, int x, int y, ref RowOctet<TPixel> currentRows, int idx)
         {
             this.pixelBlock.LoadAndStretchEdges(frame.PixelBuffer, x, y, ref currentRows);
 
             PixelOperations<TPixel>.Instance.ToRgb24(frame.GetConfiguration(), this.pixelBlock.AsSpanUnsafe(), this.rgbSpan);
 
-            ref Block8x8F rSub = ref this.Y;
-            ref Block8x8F gSub = ref this.Cb;
-            ref Block8x8F bSub = ref this.Cr;
-
             if (RgbToYCbCrConverterVectorized.IsSupported)
             {
-                RgbToYCbCrConverterVectorized.Convert420(this.rgbSpan, ref yBlock, ref rSub, ref gSub, ref bSub, idx);
+                RgbToYCbCrConverterVectorized.Convert420(this.rgbSpan, ref this.Y, ref this.Cb, ref this.Cr, idx);
             }
             else
             {
@@ -108,17 +104,5 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Encoder
                 //this.colorTables.Convert(this.rgbSpan, ref yBlock, ref cbBlock, ref crBlock);
             }
         }
-
-        public void ConvertCbCr(ref Block8x8F cb, ref Block8x8F cr)
-        {
-            if (RgbToYCbCrConverterVectorized.IsSupported)
-            {
-                RgbToYCbCrConverterVectorized.ConvertCbCr(ref this.Y, ref this.Cb, ref this.Cr, ref cb, ref cr);
-            }
-            else
-            {
-                throw new NotSupportedException("This is not yet implemented");
-            }
-        }
     }
 }