block scaling bottleneck -> eliminated

src/ImageSharp/Formats/Jpeg/Components/Block8x8F.CopyTo.cs

@@ -16,28 +16,35 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components
         /// </summary>
         [MethodImpl(InliningOptions.ShortMethod)]
         public void CopyTo(in BufferArea<float> area, int horizontalScale, int verticalScale)
+        {
+            ref float areaOrigin = ref area.GetReferenceToOrigin();
+            this.CopyTo(ref areaOrigin, area.Stride, horizontalScale, verticalScale);
+        }
+
+        [MethodImpl(InliningOptions.ShortMethod)]
+        public void CopyTo(ref float areaOrigin, int areaStride, int horizontalScale, int verticalScale)
         {
             if (horizontalScale == 1 && verticalScale == 1)
             {
-                this.Copy1x1Scale(area);
+                this.Copy1x1Scale(ref areaOrigin, areaStride);
                 return;
             }
 
             if (horizontalScale == 2 && verticalScale == 2)
             {
-                this.Copy2x2Scale(area);
+                this.Copy2x2Scale(ref areaOrigin, areaStride);
                 return;
             }
 
             // TODO: Optimize: implement all cases with scale-specific, loopless code!
-            this.CopyArbitraryScale(area, horizontalScale, verticalScale);
+            this.CopyArbitraryScale(ref areaOrigin, areaStride, horizontalScale, verticalScale);
         }
 
-        public void Copy1x1Scale(in BufferArea<float> destination)
+        public void Copy1x1Scale(ref float areaOrigin, int areaStride)
         {
             ref byte selfBase = ref Unsafe.As<Block8x8F, byte>(ref this);
-            ref byte destBase = ref Unsafe.As<float, byte>(ref destination.GetReferenceToOrigin());
+            ref byte destBase = ref Unsafe.As<float, byte>(ref areaOrigin);
-            int destStride = destination.Stride * sizeof(float);
+            int destStride = areaStride * sizeof(float);
 
             CopyRowImpl(ref selfBase, ref destBase, destStride, 0);
             CopyRowImpl(ref selfBase, ref destBase, destStride, 1);
@@ -57,10 +64,10 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components
             Unsafe.CopyBlock(ref d, ref s, 8 * sizeof(float));
         }
 
-        private void Copy2x2Scale(in BufferArea<float> area)
+        private void Copy2x2Scale(ref float areaOrigin, int areaStride)
         {
-            ref Vector2 destBase = ref Unsafe.As<float, Vector2>(ref area.GetReferenceToOrigin());
+            ref Vector2 destBase = ref Unsafe.As<float, Vector2>(ref areaOrigin);
-            int destStride = area.Stride / 2;
+            int destStride = areaStride / 2;
 
             this.WidenCopyRowImpl2x2(ref destBase, 0, destStride);
             this.WidenCopyRowImpl2x2(ref destBase, 1, destStride);
@@ -110,10 +117,8 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components
         }
 
         [MethodImpl(InliningOptions.ColdPath)]
-        private void CopyArbitraryScale(BufferArea<float> area, int horizontalScale, int verticalScale)
+        private void CopyArbitraryScale(ref float areaOrigin, int areaStride, int horizontalScale, int verticalScale)
         {
-            ref float destBase = ref area.GetReferenceToOrigin();
-
             for (int y = 0; y < 8; y++)
             {
                 int yy = y * verticalScale;
@@ -127,12 +132,12 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components
 
                     for (int i = 0; i < verticalScale; i++)
                     {
-                        int baseIdx = ((yy + i) * area.Stride) + xx;
+                        int baseIdx = ((yy + i) * areaStride) + xx;
 
                         for (int j = 0; j < horizontalScale; j++)
                         {
                             // area[xx + j, yy + i] = value;
-                            Unsafe.Add(ref destBase, baseIdx + j) = value;
+                            Unsafe.Add(ref areaOrigin, baseIdx + j) = value;
                         }
                     }
                 }

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

@@ -68,11 +68,13 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
         /// - Copy the resulting color values into 'destArea' scaling up the block by amount defined in <see cref="subSamplingDivisors"/>.
         /// </summary>
         /// <param name="sourceBlock">The source block.</param>
-        /// <param name="destArea">The destination buffer area.</param>
+        /// <param name="destAreaOrigin">Reference to the origin of the destination pixel area.</param>
+        /// <param name="destAreaStride">The width of the destination pixel buffer.</param>
         /// <param name="maximumValue">The maximum value derived from the bitdepth.</param>
         public void ProcessBlockColorsInto(
             ref Block8x8 sourceBlock,
-            in BufferArea<float> destArea,
+            ref float destAreaOrigin,
+            int destAreaStride,
             float maximumValue)
         {
             ref Block8x8F b = ref this.SourceBlock;
@@ -88,7 +90,11 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
             // To be "more accurate", we need to emulate this by rounding!
             this.WorkspaceBlock1.NormalizeColorsAndRoundInplace(maximumValue);
 
-            this.WorkspaceBlock1.CopyTo(destArea, this.subSamplingDivisors.Width, this.subSamplingDivisors.Height);
+            this.WorkspaceBlock1.CopyTo(
+                ref destAreaOrigin,
+                destAreaStride,
+                this.subSamplingDivisors.Width,
+                this.subSamplingDivisors.Height);
         }
     }
-}
+}

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

@@ -79,6 +79,8 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
             var blockPp = new JpegBlockPostProcessor(this.ImagePostProcessor.RawJpeg, this.Component);
             float maximumValue = MathF.Pow(2, this.ImagePostProcessor.RawJpeg.Precision) - 1;
 
+            int destAreaStride = this.ColorBuffer.Width;
+
             for (int y = 0; y < this.BlockRowsPerStep; y++)
             {
                 int yBlock = this.currentComponentRowInBlocks + y;
@@ -90,22 +92,16 @@ namespace SixLabors.ImageSharp.Formats.Jpeg.Components.Decoder
 
                 int yBuffer = y * this.blockAreaSize.Height;
 
+                Span<float> colorBufferRow = this.ColorBuffer.GetRowSpan(yBuffer);
                 Span<Block8x8> blockRow = this.Component.SpectralBlocks.GetRowSpan(yBlock);
 
-                ref Block8x8 blockRowBase = ref MemoryMarshal.GetReference(blockRow);
-
                 for (int xBlock = 0; xBlock < this.SizeInBlocks.Width; xBlock++)
                 {
-                    ref Block8x8 block = ref Unsafe.Add(ref blockRowBase, xBlock);
+                    ref Block8x8 block = ref blockRow[xBlock];
                     int xBuffer = xBlock * this.blockAreaSize.Width;
+                    ref float destAreaOrigin = ref colorBufferRow[xBuffer];
 
-                    BufferArea<float> destArea = this.ColorBuffer.GetArea(
+                    blockPp.ProcessBlockColorsInto(ref block, ref destAreaOrigin, destAreaStride, maximumValue);
-                        xBuffer,
-                        yBuffer,
-                        this.blockAreaSize.Width,
-                        this.blockAreaSize.Height);
-
-                    blockPp.ProcessBlockColorsInto(ref block, destArea, maximumValue);
                 }
             }
 

tests/ImageSharp.Tests/Formats/Jpg/Block8x8FTests.CopyToBufferArea.cs

@@ -44,7 +44,7 @@ namespace SixLabors.ImageSharp.Tests.Formats.Jpg
                 using (Buffer2D<float> buffer = Configuration.Default.MemoryAllocator.Allocate2D<float>(20, 20, AllocationOptions.Clean))
                 {
                     BufferArea<float> area = buffer.GetArea(5, 10, 8, 8);
-                    block.Copy1x1Scale(area);
+                    block.Copy1x1Scale(ref area.GetReferenceToOrigin(), area.Stride);
 
                     Assert.Equal(block[0, 0], buffer[5, 10]);
                     Assert.Equal(block[1, 0], buffer[6, 10]);