Merge branch 'master' into js/quantization-tweaks

src/ImageSharp/Advanced/IRowIntervalOperation.cs

@@ -0,0 +1,21 @@
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
+using System.Runtime.CompilerServices;
+using SixLabors.ImageSharp.Memory;
+
+namespace SixLabors.ImageSharp.Advanced
+{
+    /// <summary>
+    /// Defines the contract for an action that operates on a row interval.
+    /// </summary>
+    public interface IRowIntervalOperation
+    {
+        /// <summary>
+        /// Invokes the method passing the row interval.
+        /// </summary>
+        /// <param name="rows">The row interval.</param>
+        void Invoke(in RowInterval rows);
+    }
+}

src/ImageSharp/Advanced/IRowIntervalOperation{TBuffer}.cs

@@ -0,0 +1,25 @@
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
+using System.Buffers;
+using System.Runtime.CompilerServices;
+using SixLabors.ImageSharp.Memory;
+
+namespace SixLabors.ImageSharp.Advanced
+{
+    /// <summary>
+    /// Defines the contract for an action that operates on a row interval with a temporary buffer.
+    /// </summary>
+    /// <typeparam name="TBuffer">The type of buffer elements.</typeparam>
+    public interface IRowIntervalOperation<TBuffer>
+        where TBuffer : unmanaged
+    {
+        /// <summary>
+        /// Invokes the method passing the row interval and a buffer.
+        /// </summary>
+        /// <param name="rows">The row interval.</param>
+        /// <param name="span">The contiguous region of memory.</param>
+        void Invoke(in RowInterval rows, Span<TBuffer> span);
+    }
+}

src/ImageSharp/Advanced/ParallelUtils/ParallelExecutionSettings.cs → src/ImageSharp/Advanced/ParallelExecutionSettings.cs

@@ -1,4 +1,4 @@
-// Copyright (c) Six Labors and contributors.
+// Copyright (c) Six Labors and contributors.
 // Licensed under the Apache License, Version 2.0.
 
 using System;
@@ -6,10 +6,10 @@ using System.Threading.Tasks;
 
 using SixLabors.ImageSharp.Memory;
 
-namespace SixLabors.ImageSharp.Advanced.ParallelUtils
+namespace SixLabors.ImageSharp.Advanced
 {
     /// <summary>
-    /// Defines execution settings for methods in <see cref="ParallelHelper"/>.
+    /// Defines execution settings for methods in <see cref="ParallelRowIterator"/>.
     /// </summary>
     public readonly struct ParallelExecutionSettings
     {
@@ -89,7 +89,7 @@ namespace SixLabors.ImageSharp.Advanced.ParallelUtils
         }
 
         /// <summary>
-        /// Get the default <see cref="SixLabors.ImageSharp.Advanced.ParallelUtils.ParallelExecutionSettings"/> for a <see cref="SixLabors.ImageSharp.Configuration"/>
+        /// Get the default <see cref="SixLabors.ImageSharp.Advanced.ParallelExecutionSettings"/> for a <see cref="SixLabors.ImageSharp.Configuration"/>
         /// </summary>
         /// <param name="configuration">The <see cref="Configuration"/>.</param>
         /// <returns>The <see cref="ParallelExecutionSettings"/>.</returns>

src/ImageSharp/Advanced/ParallelRowIterator.Wrappers.cs

@@ -0,0 +1,110 @@
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
+using System.Buffers;
+using System.Runtime.CompilerServices;
+using System.Threading.Tasks;
+using SixLabors.ImageSharp.Memory;
+
+namespace SixLabors.ImageSharp.Advanced
+{
+    /// <content>
+    /// Utility methods for batched processing of pixel row intervals.
+    /// Parallel execution is optimized for image processing based on values defined
+    /// <see cref="ParallelExecutionSettings"/> or <see cref="Configuration"/>.
+    /// Using this class is preferred over direct usage of <see cref="Parallel"/> utility methods.
+    /// </content>
+    public static partial class ParallelRowIterator
+    {
+        private readonly struct IterationParameters
+        {
+            public readonly int MinY;
+            public readonly int MaxY;
+            public readonly int StepY;
+            public readonly int Width;
+
+            public IterationParameters(int minY, int maxY, int stepY)
+                : this(minY, maxY, stepY, 0)
+            {
+            }
+
+            public IterationParameters(int minY, int maxY, int stepY, int width)
+            {
+                this.MinY = minY;
+                this.MaxY = maxY;
+                this.StepY = stepY;
+                this.Width = width;
+            }
+        }
+
+        private readonly struct RowIntervalOperationWrapper<T>
+            where T : struct, IRowIntervalOperation
+        {
+            private readonly IterationParameters info;
+            private readonly T operation;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperationWrapper(in IterationParameters info, in T operation)
+            {
+                this.info = info;
+                this.operation = operation;
+            }
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(int i)
+            {
+                int yMin = this.info.MinY + (i * this.info.StepY);
+
+                if (yMin >= this.info.MaxY)
+                {
+                    return;
+                }
+
+                int yMax = Math.Min(yMin + this.info.StepY, this.info.MaxY);
+                var rows = new RowInterval(yMin, yMax);
+
+                // Skip the safety copy when invoking a potentially impure method on a readonly field
+                Unsafe.AsRef(in this.operation).Invoke(in rows);
+            }
+        }
+
+        private readonly struct RowIntervalOperationWrapper<T, TBuffer>
+            where T : struct, IRowIntervalOperation<TBuffer>
+            where TBuffer : unmanaged
+        {
+            private readonly IterationParameters info;
+            private readonly MemoryAllocator allocator;
+            private readonly T operation;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperationWrapper(
+                in IterationParameters info,
+                MemoryAllocator allocator,
+                in T operation)
+            {
+                this.info = info;
+                this.allocator = allocator;
+                this.operation = operation;
+            }
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(int i)
+            {
+                int yMin = this.info.MinY + (i * this.info.StepY);
+
+                if (yMin >= this.info.MaxY)
+                {
+                    return;
+                }
+
+                int yMax = Math.Min(yMin + this.info.StepY, this.info.MaxY);
+                var rows = new RowInterval(yMin, yMax);
+
+                using IMemoryOwner<TBuffer> buffer = this.allocator.Allocate<TBuffer>(this.info.Width);
+
+                Unsafe.AsRef(in this.operation).Invoke(in rows, buffer.Memory.Span);
+            }
+        }
+    }
+}

src/ImageSharp/Advanced/ParallelRowIterator.cs

@@ -0,0 +1,162 @@
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
+using System.Buffers;
+using System.Runtime.CompilerServices;
+using System.Threading.Tasks;
+using SixLabors.ImageSharp.Memory;
+
+namespace SixLabors.ImageSharp.Advanced
+{
+    /// <summary>
+    /// Utility methods for batched processing of pixel row intervals.
+    /// Parallel execution is optimized for image processing based on values defined
+    /// <see cref="ParallelExecutionSettings"/> or <see cref="Configuration"/>.
+    /// Using this class is preferred over direct usage of <see cref="Parallel"/> utility methods.
+    /// </summary>
+    public static partial class ParallelRowIterator
+    {
+        /// <summary>
+        /// Iterate through the rows of a rectangle in optimized batches defined by <see cref="RowInterval"/>-s.
+        /// </summary>
+        /// <typeparam name="T">The type of row operation to perform.</typeparam>
+        /// <param name="configuration">The <see cref="Configuration"/> to get the parallel settings from.</param>
+        /// <param name="rectangle">The <see cref="Rectangle"/>.</param>
+        /// <param name="operation">The operation defining the iteration logic on a single <see cref="RowInterval"/>.</param>
+        [MethodImpl(InliningOptions.ShortMethod)]
+        public static void IterateRows<T>(Configuration configuration, Rectangle rectangle, in T operation)
+            where T : struct, IRowIntervalOperation
+        {
+            var parallelSettings = ParallelExecutionSettings.FromConfiguration(configuration);
+            IterateRows(rectangle, in parallelSettings, in operation);
+        }
+
+        /// <summary>
+        /// Iterate through the rows of a rectangle in optimized batches defined by <see cref="RowInterval"/>-s.
+        /// </summary>
+        /// <typeparam name="T">The type of row operation to perform.</typeparam>
+        /// <param name="rectangle">The <see cref="Rectangle"/>.</param>
+        /// <param name="parallelSettings">The <see cref="ParallelExecutionSettings"/>.</param>
+        /// <param name="operation">The operation defining the iteration logic on a single <see cref="RowInterval"/>.</param>
+        public static void IterateRows<T>(
+            Rectangle rectangle,
+            in ParallelExecutionSettings parallelSettings,
+            in T operation)
+            where T : struct, IRowIntervalOperation
+        {
+            ValidateRectangle(rectangle);
+
+            int top = rectangle.Top;
+            int bottom = rectangle.Bottom;
+            int width = rectangle.Width;
+            int height = rectangle.Height;
+
+            int maxSteps = DivideCeil(width * height, parallelSettings.MinimumPixelsProcessedPerTask);
+            int numOfSteps = Math.Min(parallelSettings.MaxDegreeOfParallelism, maxSteps);
+
+            // Avoid TPL overhead in this trivial case:
+            if (numOfSteps == 1)
+            {
+                var rows = new RowInterval(top, bottom);
+                Unsafe.AsRef(in operation).Invoke(in rows);
+                return;
+            }
+
+            int verticalStep = DivideCeil(rectangle.Height, numOfSteps);
+            var parallelOptions = new ParallelOptions { MaxDegreeOfParallelism = numOfSteps };
+            var info = new IterationParameters(top, bottom, verticalStep);
+            var wrappingOperation = new RowIntervalOperationWrapper<T>(in info, in operation);
+
+            Parallel.For(
+                0,
+                numOfSteps,
+                parallelOptions,
+                wrappingOperation.Invoke);
+        }
+
+        /// <summary>
+        /// Iterate through the rows of a rectangle in optimized batches defined by <see cref="RowInterval"/>-s
+        /// instantiating a temporary buffer for each <paramref name="operation"/> invocation.
+        /// </summary>
+        /// <typeparam name="T">The type of row operation to perform.</typeparam>
+        /// <typeparam name="TBuffer">The type of buffer elements.</typeparam>
+        /// <param name="configuration">The <see cref="Configuration"/> to get the parallel settings from.</param>
+        /// <param name="rectangle">The <see cref="Rectangle"/>.</param>
+        /// <param name="operation">The operation defining the iteration logic on a single <see cref="RowInterval"/>.</param>
+        public static void IterateRows<T, TBuffer>(Configuration configuration, Rectangle rectangle, in T operation)
+            where T : struct, IRowIntervalOperation<TBuffer>
+            where TBuffer : unmanaged
+        {
+            var parallelSettings = ParallelExecutionSettings.FromConfiguration(configuration);
+            IterateRows<T, TBuffer>(rectangle, in parallelSettings, in operation);
+        }
+
+        /// <summary>
+        /// Iterate through the rows of a rectangle in optimized batches defined by <see cref="RowInterval"/>-s
+        /// instantiating a temporary buffer for each <paramref name="operation"/> invocation.
+        /// </summary>
+        /// <typeparam name="T">The type of row operation to perform.</typeparam>
+        /// <typeparam name="TBuffer">The type of buffer elements.</typeparam>
+        /// <param name="rectangle">The <see cref="Rectangle"/>.</param>
+        /// <param name="parallelSettings">The <see cref="ParallelExecutionSettings"/>.</param>
+        /// <param name="operation">The operation defining the iteration logic on a single <see cref="RowInterval"/>.</param>
+        public static void IterateRows<T, TBuffer>(
+            Rectangle rectangle,
+            in ParallelExecutionSettings parallelSettings,
+            in T operation)
+            where T : struct, IRowIntervalOperation<TBuffer>
+            where TBuffer : unmanaged
+        {
+            ValidateRectangle(rectangle);
+
+            int top = rectangle.Top;
+            int bottom = rectangle.Bottom;
+            int width = rectangle.Width;
+            int height = rectangle.Height;
+
+            int maxSteps = DivideCeil(width * height, parallelSettings.MinimumPixelsProcessedPerTask);
+            int numOfSteps = Math.Min(parallelSettings.MaxDegreeOfParallelism, maxSteps);
+            MemoryAllocator allocator = parallelSettings.MemoryAllocator;
+
+            // Avoid TPL overhead in this trivial case:
+            if (numOfSteps == 1)
+            {
+                var rows = new RowInterval(top, bottom);
+                using (IMemoryOwner<TBuffer> buffer = allocator.Allocate<TBuffer>(width))
+                {
+                    Unsafe.AsRef(operation).Invoke(in rows, buffer.Memory.Span);
+                }
+
+                return;
+            }
+
+            int verticalStep = DivideCeil(height, numOfSteps);
+            var parallelOptions = new ParallelOptions { MaxDegreeOfParallelism = numOfSteps };
+            var info = new IterationParameters(top, bottom, verticalStep, width);
+            var wrappingOperation = new RowIntervalOperationWrapper<T, TBuffer>(in info, allocator, in operation);
+
+            Parallel.For(
+                0,
+                numOfSteps,
+                parallelOptions,
+                wrappingOperation.Invoke);
+        }
+
+        [MethodImpl(InliningOptions.ShortMethod)]
+        private static int DivideCeil(int dividend, int divisor) => 1 + ((dividend - 1) / divisor);
+
+        private static void ValidateRectangle(Rectangle rectangle)
+        {
+            Guard.MustBeGreaterThan(
+                rectangle.Width,
+                0,
+                $"{nameof(rectangle)}.{nameof(rectangle.Width)}");
+
+            Guard.MustBeGreaterThan(
+                rectangle.Height,
+                0,
+                $"{nameof(rectangle)}.{nameof(rectangle.Height)}");
+        }
+    }
+}

src/ImageSharp/Advanced/ParallelUtils/ParallelHelper.cs

@@ -1,178 +0,0 @@
-// Copyright (c) Six Labors and contributors.
-// Licensed under the Apache License, Version 2.0.
-
-using System;
-using System.Buffers;
-using System.Runtime.CompilerServices;
-using System.Threading.Tasks;
-
-using SixLabors.ImageSharp.Memory;
-
-namespace SixLabors.ImageSharp.Advanced.ParallelUtils
-{
-    /// <summary>
-    /// Utility methods for batched processing of pixel row intervals.
-    /// Parallel execution is optimized for image processing based on values defined
-    /// <see cref="ParallelExecutionSettings"/> or <see cref="Configuration"/>.
-    /// Using this class is preferred over direct usage of <see cref="Parallel"/> utility methods.
-    /// </summary>
-    public static class ParallelHelper
-    {
-        /// <summary>
-        /// Iterate through the rows of a rectangle in optimized batches defined by <see cref="RowInterval"/>-s.
-        /// </summary>
-        /// <param name="rectangle">The <see cref="Rectangle"/>.</param>
-        /// <param name="configuration">The <see cref="Configuration"/> to get the parallel settings from.</param>
-        /// <param name="body">The method body defining the iteration logic on a single <see cref="RowInterval"/>.</param>
-        public static void IterateRows(Rectangle rectangle, Configuration configuration, Action<RowInterval> body)
-        {
-            var parallelSettings = ParallelExecutionSettings.FromConfiguration(configuration);
-
-            IterateRows(rectangle, parallelSettings, body);
-        }
-
-        /// <summary>
-        /// Iterate through the rows of a rectangle in optimized batches defined by <see cref="RowInterval"/>-s.
-        /// </summary>
-        /// <param name="rectangle">The <see cref="Rectangle"/>.</param>
-        /// <param name="parallelSettings">The <see cref="ParallelExecutionSettings"/>.</param>
-        /// <param name="body">The method body defining the iteration logic on a single <see cref="RowInterval"/>.</param>
-        public static void IterateRows(
-            Rectangle rectangle,
-            in ParallelExecutionSettings parallelSettings,
-            Action<RowInterval> body)
-        {
-            ValidateRectangle(rectangle);
-
-            int maxSteps = DivideCeil(
-                rectangle.Width * rectangle.Height,
-                parallelSettings.MinimumPixelsProcessedPerTask);
-
-            int numOfSteps = Math.Min(parallelSettings.MaxDegreeOfParallelism, maxSteps);
-
-            // Avoid TPL overhead in this trivial case:
-            if (numOfSteps == 1)
-            {
-                var rows = new RowInterval(rectangle.Top, rectangle.Bottom);
-                body(rows);
-                return;
-            }
-
-            int verticalStep = DivideCeil(rectangle.Height, numOfSteps);
-
-            var parallelOptions = new ParallelOptions { MaxDegreeOfParallelism = numOfSteps };
-
-            int top = rectangle.Top;
-            int bottom = rectangle.Bottom;
-
-            Parallel.For(
-                0,
-                numOfSteps,
-                parallelOptions,
-                i =>
-                    {
-                        int yMin = top + (i * verticalStep);
-
-                        if (yMin >= bottom)
-                        {
-                            return;
-                        }
-
-                        int yMax = Math.Min(yMin + verticalStep, bottom);
-
-                        var rows = new RowInterval(yMin, yMax);
-                        body(rows);
-                    });
-        }
-
-        /// <summary>
-        /// Iterate through the rows of a rectangle in optimized batches defined by <see cref="RowInterval"/>-s
-        /// instantiating a temporary buffer for each <paramref name="body"/> invocation.
-        /// </summary>
-        internal static void IterateRowsWithTempBuffer<T>(
-            Rectangle rectangle,
-            in ParallelExecutionSettings parallelSettings,
-            Action<RowInterval, Memory<T>> body)
-            where T : unmanaged
-        {
-            ValidateRectangle(rectangle);
-
-            int maxSteps = DivideCeil(rectangle.Width * rectangle.Height, parallelSettings.MinimumPixelsProcessedPerTask);
-
-            int numOfSteps = Math.Min(parallelSettings.MaxDegreeOfParallelism, maxSteps);
-
-            MemoryAllocator memoryAllocator = parallelSettings.MemoryAllocator;
-
-            // Avoid TPL overhead in this trivial case:
-            if (numOfSteps == 1)
-            {
-                var rows = new RowInterval(rectangle.Top, rectangle.Bottom);
-                using (IMemoryOwner<T> buffer = memoryAllocator.Allocate<T>(rectangle.Width))
-                {
-                    body(rows, buffer.Memory);
-                }
-
-                return;
-            }
-
-            int verticalStep = DivideCeil(rectangle.Height, numOfSteps);
-
-            var parallelOptions = new ParallelOptions { MaxDegreeOfParallelism = numOfSteps };
-
-            int top = rectangle.Top;
-            int bottom = rectangle.Bottom;
-
-            Parallel.For(
-                0,
-                numOfSteps,
-                parallelOptions,
-                i =>
-                    {
-                        int yMin = top + (i * verticalStep);
-
-                        if (yMin >= bottom)
-                        {
-                            return;
-                        }
-
-                        int yMax = Math.Min(yMin + verticalStep, rectangle.Bottom);
-
-                        var rows = new RowInterval(yMin, yMax);
-
-                        using (IMemoryOwner<T> buffer = memoryAllocator.Allocate<T>(rectangle.Width))
-                        {
-                            body(rows, buffer.Memory);
-                        }
-                    });
-        }
-
-        /// <summary>
-        /// Iterate through the rows of a rectangle in optimized batches defined by <see cref="RowInterval"/>-s
-        /// instantiating a temporary buffer for each <paramref name="body"/> invocation.
-        /// </summary>
-        internal static void IterateRowsWithTempBuffer<T>(
-            Rectangle rectangle,
-            Configuration configuration,
-            Action<RowInterval, Memory<T>> body)
-            where T : unmanaged
-        {
-            IterateRowsWithTempBuffer(rectangle, ParallelExecutionSettings.FromConfiguration(configuration), body);
-        }
-
-        [MethodImpl(MethodImplOptions.AggressiveInlining)]
-        private static int DivideCeil(int dividend, int divisor) => 1 + ((dividend - 1) / divisor);
-
-        private static void ValidateRectangle(Rectangle rectangle)
-        {
-            Guard.MustBeGreaterThan(
-                rectangle.Width,
-                0,
-                $"{nameof(rectangle)}.{nameof(rectangle.Width)}");
-
-            Guard.MustBeGreaterThan(
-                rectangle.Height,
-                0,
-                $"{nameof(rectangle)}.{nameof(rectangle.Height)}");
-        }
-    }
-}

src/ImageSharp/Common/Extensions/EnumerableExtensions.cs

@@ -1,10 +1,10 @@
-// Copyright (c) Six Labors and contributors.
+// Copyright (c) Six Labors and contributors.
 // Licensed under the Apache License, Version 2.0.
 
 using System;
 using System.Collections.Generic;
 
-namespace SixLabors.ImageSharp.Common
+namespace SixLabors.ImageSharp
 {
     /// <summary>
     /// Encapsulates a series of time saving extension methods to the <see cref="T:System.Collections.IEnumerable"/> interface.
@@ -34,15 +34,11 @@ namespace SixLabors.ImageSharp.Common
         /// <summary>
         /// Generates a sequence of integral numbers within a specified range.
         /// </summary>
-        /// <param name="fromInclusive">
-        /// The start index, inclusive.
-        /// </param>
+        /// <param name="fromInclusive">The start index, inclusive.</param>
         /// <param name="toDelegate">
         /// A method that has one parameter and returns a <see cref="bool"/> calculating the end index.
         /// </param>
-        /// <param name="step">
-        /// The incremental step.
-        /// </param>
+        /// <param name="step">The incremental step.</param>
         /// <returns>
         /// The <see cref="IEnumerable{Int32}"/> that contains a range of sequential integral numbers.
         /// </returns>
@@ -56,4 +52,4 @@ namespace SixLabors.ImageSharp.Common
             }
         }
     }
-}
+}

src/ImageSharp/ImageFrame{TPixel}.cs

@@ -4,9 +4,7 @@
 using System;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
-
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.Metadata;
 using SixLabors.ImageSharp.PixelFormats;
@@ -261,19 +259,12 @@ namespace SixLabors.ImageSharp
             }
 
             var target = new ImageFrame<TPixel2>(configuration, this.Width, this.Height, this.Metadata.DeepClone());
+            var operation = new RowIntervalOperation<TPixel2>(this, target, configuration);
 
-            ParallelHelper.IterateRows(
-                this.Bounds(),
+            ParallelRowIterator.IterateRows(
                 configuration,
-                rows =>
-                    {
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            Span<TPixel> sourceRow = this.GetPixelRowSpan(y);
-                            Span<TPixel2> targetRow = target.GetPixelRowSpan(y);
-                            PixelOperations<TPixel>.Instance.To(configuration, sourceRow, targetRow);
-                        }
-                    });
+                this.Bounds(),
+                in operation);
 
             return target;
         }
@@ -295,5 +286,39 @@ namespace SixLabors.ImageSharp
                 span.Fill(value);
             }
         }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the clone logic for <see cref="ImageFrame{TPixel}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation<TPixel2> : IRowIntervalOperation
+            where TPixel2 : struct, IPixel<TPixel2>
+        {
+            private readonly ImageFrame<TPixel> source;
+            private readonly ImageFrame<TPixel2> target;
+            private readonly Configuration configuration;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                ImageFrame<TPixel> source,
+                ImageFrame<TPixel2> target,
+                Configuration configuration)
+            {
+                this.source = source;
+                this.target = target;
+                this.configuration = configuration;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> sourceRow = this.source.GetPixelRowSpan(y);
+                    Span<TPixel2> targetRow = this.target.GetPixelRowSpan(y);
+                    PixelOperations<TPixel>.Instance.To(this.configuration, sourceRow, targetRow);
+                }
+            }
+        }
     }
 }

src/ImageSharp/ImageSharp.csproj

@@ -24,9 +24,9 @@
   </ItemGroup>
 
   <ItemGroup Condition=" $(TargetFramework.StartsWith('netstandard')) OR '$(TargetFramework)' == 'net472'">
-    <PackageReference Include="System.Numerics.Vectors"/>
-    <PackageReference Include="System.Buffers"/>
-    <PackageReference Include="System.Memory"/>
+    <PackageReference Include="System.Numerics.Vectors" />
+    <PackageReference Include="System.Buffers" />
+    <PackageReference Include="System.Memory" />
   </ItemGroup>
 
   <ItemGroup Condition=" '$(TargetFramework)' == 'netstandard1.3'">

src/ImageSharp/PixelFormats/Utils/Vector4Converters.RgbaCompatible.cs

@@ -1,4 +1,4 @@
-// Copyright (c) Six Labors and contributors.
+// Copyright (c) Six Labors and contributors.
 // Licensed under the Apache License, Version 2.0.
 
 using System;
@@ -126,4 +126,4 @@ namespace SixLabors.ImageSharp.PixelFormats.Utils
             }
         }
     }
-}
+}

src/ImageSharp/Primitives/Rectangle.cs

@@ -460,4 +460,4 @@ namespace SixLabors.ImageSharp
             this.Width.Equals(other.Width) &&
             this.Height.Equals(other.Height);
     }
-}
+}

src/ImageSharp/Processing/Processors/Binarization/BinaryThresholdProcessor{TPixel}.cs

@@ -2,9 +2,9 @@
 // Licensed under the Apache License, Version 2.0.
 
 using System;
-
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Binarization
@@ -42,36 +42,66 @@ namespace SixLabors.ImageSharp.Processing.Processors.Binarization
             Configuration configuration = this.Configuration;
 
             var interest = Rectangle.Intersect(sourceRectangle, source.Bounds());
-            int startY = interest.Y;
-            int endY = interest.Bottom;
-            int startX = interest.X;
-            int endX = interest.Right;
-
             bool isAlphaOnly = typeof(TPixel) == typeof(A8);
 
-            var workingRect = Rectangle.FromLTRB(startX, startY, endX, endY);
-
-            ParallelHelper.IterateRows(
-                workingRect,
+            var operation = new RowIntervalOperation(interest, source, upper, lower, threshold, isAlphaOnly);
+            ParallelRowIterator.IterateRows(
                 configuration,
-                rows =>
-                    {
-                        Rgba32 rgba = default;
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            Span<TPixel> row = source.GetPixelRowSpan(y);
+                interest,
+                in operation);
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the clone logic for <see cref="BinaryThresholdProcessor{TPixel}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly ImageFrame<TPixel> source;
+            private readonly TPixel upper;
+            private readonly TPixel lower;
+            private readonly byte threshold;
+            private readonly int minX;
+            private readonly int maxX;
+            private readonly bool isAlphaOnly;
 
-                            for (int x = startX; x < endX; x++)
-                            {
-                                ref TPixel color = ref row[x];
-                                color.ToRgba32(ref rgba);
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Rectangle bounds,
+                ImageFrame<TPixel> source,
+                TPixel upper,
+                TPixel lower,
+                byte threshold,
+                bool isAlphaOnly)
+            {
+                this.source = source;
+                this.upper = upper;
+                this.lower = lower;
+                this.threshold = threshold;
+                this.minX = bounds.X;
+                this.maxX = bounds.Right;
+                this.isAlphaOnly = isAlphaOnly;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                Rgba32 rgba = default;
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> row = this.source.GetPixelRowSpan(y);
+
+                    for (int x = this.minX; x < this.maxX; x++)
+                    {
+                        ref TPixel color = ref row[x];
+                        color.ToRgba32(ref rgba);
 
-                                // Convert to grayscale using ITU-R Recommendation BT.709 if required
-                                byte luminance = isAlphaOnly ? rgba.A : ImageMaths.Get8BitBT709Luminance(rgba.R, rgba.G, rgba.B);
-                                color = luminance >= threshold ? upper : lower;
-                            }
-                        }
-                    });
+                        // Convert to grayscale using ITU-R Recommendation BT.709 if required
+                        byte luminance = this.isAlphaOnly ? rgba.A : ImageMaths.Get8BitBT709Luminance(rgba.R, rgba.G, rgba.B);
+                        color = luminance >= this.threshold ? this.upper : this.lower;
+                    }
+                }
+            }
         }
     }
 }

src/ImageSharp/Processing/Processors/Convolution/BokehBlurProcessor{TPixel}.cs

@@ -7,8 +7,7 @@ using System.Collections.Generic;
 using System.Numerics;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
-
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Advanced;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 using SixLabors.ImageSharp.Processing.Processors.Convolution.Parameters;
@@ -269,17 +268,26 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
             // Preliminary gamma highlight pass
-            this.ApplyGammaExposure(source.PixelBuffer, this.SourceRectangle, this.Configuration);
+            var gammaOperation = new ApplyGammaExposureRowIntervalOperation(this.SourceRectangle, source.PixelBuffer, this.Configuration, this.gamma);
+            ParallelRowIterator.IterateRows<ApplyGammaExposureRowIntervalOperation, Vector4>(
+                this.Configuration,
+                this.SourceRectangle,
+                in gammaOperation);
 
             // Create a 0-filled buffer to use to store the result of the component convolutions
-            using (Buffer2D<Vector4> processingBuffer = this.Configuration.MemoryAllocator.Allocate2D<Vector4>(source.Size(), AllocationOptions.Clean))
-            {
-                // Perform the 1D convolutions on all the kernel components and accumulate the results
-                this.OnFrameApplyCore(source, this.SourceRectangle, this.Configuration, processingBuffer);
+            using Buffer2D<Vector4> processingBuffer = this.Configuration.MemoryAllocator.Allocate2D<Vector4>(source.Size(), AllocationOptions.Clean);
 
-                // Apply the inverse gamma exposure pass, and write the final pixel data
-                this.ApplyInverseGammaExposure(source.PixelBuffer, processingBuffer, this.SourceRectangle, this.Configuration);
-            }
+            // Perform the 1D convolutions on all the kernel components and accumulate the results
+            this.OnFrameApplyCore(source, this.SourceRectangle, this.Configuration, processingBuffer);
+
+            float inverseGamma = 1 / this.gamma;
+
+            // Apply the inverse gamma exposure pass, and write the final pixel data
+            var operation = new ApplyInverseGammaExposureRowIntervalOperation(this.SourceRectangle, source.PixelBuffer, processingBuffer, this.Configuration, inverseGamma);
+            ParallelRowIterator.IterateRows(
+                this.Configuration,
+                this.SourceRectangle,
+                in operation);
         }
 
         /// <summary>
@@ -295,216 +303,223 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
             Configuration configuration,
             Buffer2D<Vector4> processingBuffer)
         {
-            using (Buffer2D<ComplexVector4> firstPassBuffer = this.Configuration.MemoryAllocator.Allocate2D<ComplexVector4>(source.Size()))
+            // Allocate the buffer with the intermediate convolution results
+            using Buffer2D<ComplexVector4> firstPassBuffer = this.Configuration.MemoryAllocator.Allocate2D<ComplexVector4>(source.Size());
+
+            // Perform two 1D convolutions for each component in the current instance
+            ref Complex64[] baseRef = ref MemoryMarshal.GetReference(this.kernels.AsSpan());
+            ref Vector4 paramsRef = ref MemoryMarshal.GetReference(this.kernelParameters.AsSpan());
+            for (int i = 0; i < this.kernels.Length; i++)
             {
-                // Perform two 1D convolutions for each component in the current instance
-                ref Complex64[] baseRef = ref MemoryMarshal.GetReference(this.kernels.AsSpan());
-                ref Vector4 paramsRef = ref MemoryMarshal.GetReference(this.kernelParameters.AsSpan());
-                for (int i = 0; i < this.kernels.Length; i++)
-                {
-                    // Compute the resulting complex buffer for the current component
-                    var interest = Rectangle.Intersect(sourceRectangle, source.Bounds());
-                    Complex64[] kernel = Unsafe.Add(ref baseRef, i);
-                    Vector4 parameters = Unsafe.Add(ref paramsRef, i);
-
-                    // Compute the two 1D convolutions and accumulate the partial results on the target buffer
-                    this.ApplyConvolution(firstPassBuffer, source.PixelBuffer, interest, kernel, configuration);
-                    this.ApplyConvolution(processingBuffer, firstPassBuffer, interest, kernel, configuration, parameters.Z, parameters.W);
-                }
+                // Compute the resulting complex buffer for the current component
+                Complex64[] kernel = Unsafe.Add(ref baseRef, i);
+                Vector4 parameters = Unsafe.Add(ref paramsRef, i);
+
+                // Compute the vertical 1D convolution
+                var verticalOperation = new ApplyVerticalConvolutionRowIntervalOperation(sourceRectangle, firstPassBuffer, source.PixelBuffer, kernel);
+                ParallelRowIterator.IterateRows(
+                    configuration,
+                    sourceRectangle,
+                    in verticalOperation);
+
+                // Compute the horizontal 1D convolutions and accumulate the partial results on the target buffer
+                var horizontalOperation = new ApplyHorizontalConvolutionRowIntervalOperation(sourceRectangle, processingBuffer, firstPassBuffer, kernel, parameters.Z, parameters.W);
+                ParallelRowIterator.IterateRows(
+                    configuration,
+                    sourceRectangle,
+                    in horizontalOperation);
             }
         }
 
         /// <summary>
-        /// Applies the process to the specified portion of the specified <see cref="ImageFrame{TPixel}"/> at the specified location
-        /// and with the specified size.
+        /// A <see langword="struct"/> implementing the vertical convolution logic for <see cref="BokehBlurProcessor{T}"/>.
         /// </summary>
-        /// <param name="targetValues">The target <see cref="ComplexVector4"/> values to use to store the results.</param>
-        /// <param name="sourcePixels">The source pixels. Cannot be null.</param>
-        /// <param name="sourceRectangle">
-        /// The <see cref="Rectangle"/> structure that specifies the portion of the image object to draw.
-        /// </param>
-        /// <param name="kernel">The 1D kernel.</param>
-        /// <param name="configuration">The <see cref="Configuration"/></param>
-        private void ApplyConvolution(
-            Buffer2D<ComplexVector4> targetValues,
-            Buffer2D<TPixel> sourcePixels,
-            Rectangle sourceRectangle,
-            Complex64[] kernel,
-            Configuration configuration)
+        private readonly struct ApplyVerticalConvolutionRowIntervalOperation : IRowIntervalOperation
         {
-            int startY = sourceRectangle.Y;
-            int endY = sourceRectangle.Bottom;
-            int startX = sourceRectangle.X;
-            int endX = sourceRectangle.Right;
-            int maxY = endY - 1;
-            int maxX = endX - 1;
-
-            var workingRectangle = Rectangle.FromLTRB(startX, startY, endX, endY);
-            int width = workingRectangle.Width;
-
-            ParallelHelper.IterateRows(
-                workingRectangle,
-                configuration,
-                rows =>
+            private readonly Rectangle bounds;
+            private readonly Buffer2D<ComplexVector4> targetValues;
+            private readonly Buffer2D<TPixel> sourcePixels;
+            private readonly Complex64[] kernel;
+            private readonly int maxY;
+            private readonly int maxX;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public ApplyVerticalConvolutionRowIntervalOperation(
+                Rectangle bounds,
+                Buffer2D<ComplexVector4> targetValues,
+                Buffer2D<TPixel> sourcePixels,
+                Complex64[] kernel)
+            {
+                this.bounds = bounds;
+                this.maxY = this.bounds.Bottom - 1;
+                this.maxX = this.bounds.Right - 1;
+                this.targetValues = targetValues;
+                this.sourcePixels = sourcePixels;
+                this.kernel = kernel;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
                 {
-                    for (int y = rows.Min; y < rows.Max; y++)
-                    {
-                        Span<ComplexVector4> targetRowSpan = targetValues.GetRowSpan(y).Slice(startX);
+                    Span<ComplexVector4> targetRowSpan = this.targetValues.GetRowSpan(y).Slice(this.bounds.X);
 
-                        for (int x = 0; x < width; x++)
-                        {
-                            Buffer2DUtils.Convolve4(kernel, sourcePixels, targetRowSpan, y, x, startY, maxY, startX, maxX);
-                        }
+                    for (int x = 0; x < this.bounds.Width; x++)
+                    {
+                        Buffer2DUtils.Convolve4(this.kernel, this.sourcePixels, targetRowSpan, y, x, this.bounds.Y, this.maxY, this.bounds.X, this.maxX);
                     }
-                });
+                }
+            }
         }
 
         /// <summary>
-        /// Applies the process to the specified portion of the specified <see cref="Buffer2D{T}"/> buffer at the specified location
-        /// and with the specified size.
+        /// A <see langword="struct"/> implementing the horizontal convolution logic for <see cref="BokehBlurProcessor{T}"/>.
         /// </summary>
-        /// <param name="targetValues">The target <see cref="Vector4"/> values to use to store the results.</param>
-        /// <param name="sourceValues">The source complex values. Cannot be null.</param>
-        /// <param name="sourceRectangle">The <see cref="Rectangle"/> structure that specifies the portion of the image object to draw.</param>
-        /// <param name="kernel">The 1D kernel.</param>
-        /// <param name="configuration">The <see cref="Configuration"/></param>
-        /// <param name="z">The weight factor for the real component of the complex pixel values.</param>
-        /// <param name="w">The weight factor for the imaginary component of the complex pixel values.</param>
-        private void ApplyConvolution(
-            Buffer2D<Vector4> targetValues,
-            Buffer2D<ComplexVector4> sourceValues,
-            Rectangle sourceRectangle,
-            Complex64[] kernel,
-            Configuration configuration,
-            float z,
-            float w)
+        private readonly struct ApplyHorizontalConvolutionRowIntervalOperation : IRowIntervalOperation
         {
-            int startY = sourceRectangle.Y;
-            int endY = sourceRectangle.Bottom;
-            int startX = sourceRectangle.X;
-            int endX = sourceRectangle.Right;
-            int maxY = endY - 1;
-            int maxX = endX - 1;
-
-            var workingRectangle = Rectangle.FromLTRB(startX, startY, endX, endY);
-            int width = workingRectangle.Width;
-
-            ParallelHelper.IterateRows(
-                workingRectangle,
-                configuration,
-                rows =>
+            private readonly Rectangle bounds;
+            private readonly Buffer2D<Vector4> targetValues;
+            private readonly Buffer2D<ComplexVector4> sourceValues;
+            private readonly Complex64[] kernel;
+            private readonly float z;
+            private readonly float w;
+            private readonly int maxY;
+            private readonly int maxX;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public ApplyHorizontalConvolutionRowIntervalOperation(
+                Rectangle bounds,
+                Buffer2D<Vector4> targetValues,
+                Buffer2D<ComplexVector4> sourceValues,
+                Complex64[] kernel,
+                float z,
+                float w)
+            {
+                this.bounds = bounds;
+                this.maxY = this.bounds.Bottom - 1;
+                this.maxX = this.bounds.Right - 1;
+                this.targetValues = targetValues;
+                this.sourceValues = sourceValues;
+                this.kernel = kernel;
+                this.z = z;
+                this.w = w;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
                 {
-                    for (int y = rows.Min; y < rows.Max; y++)
-                    {
-                        Span<Vector4> targetRowSpan = targetValues.GetRowSpan(y).Slice(startX);
+                    Span<Vector4> targetRowSpan = this.targetValues.GetRowSpan(y).Slice(this.bounds.X);
 
-                        for (int x = 0; x < width; x++)
-                        {
-                            Buffer2DUtils.Convolve4AndAccumulatePartials(kernel, sourceValues, targetRowSpan, y, x, startY, maxY, startX, maxX, z, w);
-                        }
+                    for (int x = 0; x < this.bounds.Width; x++)
+                    {
+                        Buffer2DUtils.Convolve4AndAccumulatePartials(this.kernel, this.sourceValues, targetRowSpan, y, x, this.bounds.Y, this.maxY, this.bounds.X, this.maxX, this.z, this.w);
                     }
-                });
+                }
+            }
         }
 
         /// <summary>
-        /// Applies the gamma correction/highlight to the input pixel buffer.
+        /// A <see langword="struct"/> implementing the gamma exposure logic for <see cref="BokehBlurProcessor{T}"/>.
         /// </summary>
-        /// <param name="targetPixels">The target pixel buffer to adjust.</param>
-        /// <param name="sourceRectangle">
-        /// The <see cref="Rectangle"/> structure that specifies the portion of the image object to draw.
-        /// </param>
-        /// <param name="configuration">The <see cref="Configuration"/></param>
-        private void ApplyGammaExposure(
-            Buffer2D<TPixel> targetPixels,
-            Rectangle sourceRectangle,
-            Configuration configuration)
+        private readonly struct ApplyGammaExposureRowIntervalOperation : IRowIntervalOperation<Vector4>
         {
-            int startY = sourceRectangle.Y;
-            int endY = sourceRectangle.Bottom;
-            int startX = sourceRectangle.X;
-            int endX = sourceRectangle.Right;
-
-            var workingRectangle = Rectangle.FromLTRB(startX, startY, endX, endY);
-            int width = workingRectangle.Width;
-            float exp = this.gamma;
-
-            ParallelHelper.IterateRowsWithTempBuffer<Vector4>(
-                workingRectangle,
-                configuration,
-                (rows, vectorBuffer) =>
+            private readonly Rectangle bounds;
+            private readonly Buffer2D<TPixel> targetPixels;
+            private readonly Configuration configuration;
+            private readonly float gamma;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public ApplyGammaExposureRowIntervalOperation(
+                Rectangle bounds,
+                Buffer2D<TPixel> targetPixels,
+                Configuration configuration,
+                float gamma)
+            {
+                this.bounds = bounds;
+                this.targetPixels = targetPixels;
+                this.configuration = configuration;
+                this.gamma = gamma;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows, Span<Vector4> span)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> targetRowSpan = this.targetPixels.GetRowSpan(y).Slice(this.bounds.X);
+                    PixelOperations<TPixel>.Instance.ToVector4(this.configuration, targetRowSpan.Slice(0, span.Length), span, PixelConversionModifiers.Premultiply);
+                    ref Vector4 baseRef = ref MemoryMarshal.GetReference(span);
+
+                    for (int x = 0; x < this.bounds.Width; x++)
                     {
-                        Span<Vector4> vectorSpan = vectorBuffer.Span;
-                        int length = vectorSpan.Length;
-
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            Span<TPixel> targetRowSpan = targetPixels.GetRowSpan(y).Slice(startX);
-                            PixelOperations<TPixel>.Instance.ToVector4(configuration, targetRowSpan.Slice(0, length), vectorSpan, PixelConversionModifiers.Premultiply);
-                            ref Vector4 baseRef = ref MemoryMarshal.GetReference(vectorSpan);
-
-                            for (int x = 0; x < width; x++)
-                            {
-                                ref Vector4 v = ref Unsafe.Add(ref baseRef, x);
-                                v.X = MathF.Pow(v.X, exp);
-                                v.Y = MathF.Pow(v.Y, exp);
-                                v.Z = MathF.Pow(v.Z, exp);
-                            }
-
-                            PixelOperations<TPixel>.Instance.FromVector4Destructive(configuration, vectorSpan.Slice(0, length), targetRowSpan);
-                        }
-                    });
+                        ref Vector4 v = ref Unsafe.Add(ref baseRef, x);
+                        v.X = MathF.Pow(v.X, this.gamma);
+                        v.Y = MathF.Pow(v.Y, this.gamma);
+                        v.Z = MathF.Pow(v.Z, this.gamma);
+                    }
+
+                    PixelOperations<TPixel>.Instance.FromVector4Destructive(this.configuration, span, targetRowSpan);
+                }
+            }
         }
 
         /// <summary>
-        /// Applies the inverse gamma correction/highlight pass, and converts the input <see cref="Vector4"/> buffer into pixel values.
+        /// A <see langword="struct"/> implementing the inverse gamma exposure logic for <see cref="BokehBlurProcessor{T}"/>.
         /// </summary>
-        /// <param name="targetPixels">The target pixels to apply the process to.</param>
-        /// <param name="sourceValues">The source <see cref="Vector4"/> values. Cannot be null.</param>
-        /// <param name="sourceRectangle">
-        /// The <see cref="Rectangle"/> structure that specifies the portion of the image object to draw.
-        /// </param>
-        /// <param name="configuration">The <see cref="Configuration"/></param>
-        private void ApplyInverseGammaExposure(
-            Buffer2D<TPixel> targetPixels,
-            Buffer2D<Vector4> sourceValues,
-            Rectangle sourceRectangle,
-            Configuration configuration)
+        private readonly struct ApplyInverseGammaExposureRowIntervalOperation : IRowIntervalOperation
         {
-            int startY = sourceRectangle.Y;
-            int endY = sourceRectangle.Bottom;
-            int startX = sourceRectangle.X;
-            int endX = sourceRectangle.Right;
-
-            var workingRectangle = Rectangle.FromLTRB(startX, startY, endX, endY);
-            int width = workingRectangle.Width;
-            float expGamma = 1 / this.gamma;
-
-            ParallelHelper.IterateRows(
-                workingRectangle,
-                configuration,
-                rows =>
+            private readonly Rectangle bounds;
+            private readonly Buffer2D<TPixel> targetPixels;
+            private readonly Buffer2D<Vector4> sourceValues;
+            private readonly Configuration configuration;
+            private readonly float inverseGamma;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public ApplyInverseGammaExposureRowIntervalOperation(
+                Rectangle bounds,
+                Buffer2D<TPixel> targetPixels,
+                Buffer2D<Vector4> sourceValues,
+                Configuration configuration,
+                float inverseGamma)
+            {
+                this.bounds = bounds;
+                this.targetPixels = targetPixels;
+                this.sourceValues = sourceValues;
+                this.configuration = configuration;
+                this.inverseGamma = inverseGamma;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                Vector4 low = Vector4.Zero;
+                var high = new Vector4(float.PositiveInfinity, float.PositiveInfinity, float.PositiveInfinity, float.PositiveInfinity);
+
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> targetPixelSpan = this.targetPixels.GetRowSpan(y).Slice(this.bounds.X);
+                    Span<Vector4> sourceRowSpan = this.sourceValues.GetRowSpan(y).Slice(this.bounds.X);
+                    ref Vector4 sourceRef = ref MemoryMarshal.GetReference(sourceRowSpan);
+
+                    for (int x = 0; x < this.bounds.Width; x++)
                     {
-                        Vector4 low = Vector4.Zero;
-                        var high = new Vector4(float.PositiveInfinity, float.PositiveInfinity, float.PositiveInfinity, float.PositiveInfinity);
-
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            Span<TPixel> targetPixelSpan = targetPixels.GetRowSpan(y).Slice(startX);
-                            Span<Vector4> sourceRowSpan = sourceValues.GetRowSpan(y).Slice(startX);
-                            ref Vector4 sourceRef = ref MemoryMarshal.GetReference(sourceRowSpan);
-
-                            for (int x = 0; x < width; x++)
-                            {
-                                ref Vector4 v = ref Unsafe.Add(ref sourceRef, x);
-                                var clamp = Vector4.Clamp(v, low, high);
-                                v.X = MathF.Pow(clamp.X, expGamma);
-                                v.Y = MathF.Pow(clamp.Y, expGamma);
-                                v.Z = MathF.Pow(clamp.Z, expGamma);
-                            }
-
-                            PixelOperations<TPixel>.Instance.FromVector4Destructive(configuration, sourceRowSpan.Slice(0, width), targetPixelSpan, PixelConversionModifiers.Premultiply);
-                        }
-                    });
+                        ref Vector4 v = ref Unsafe.Add(ref sourceRef, x);
+                        var clamp = Vector4.Clamp(v, low, high);
+                        v.X = MathF.Pow(clamp.X, this.inverseGamma);
+                        v.Y = MathF.Pow(clamp.Y, this.inverseGamma);
+                        v.Z = MathF.Pow(clamp.Z, this.inverseGamma);
+                    }
+
+                    PixelOperations<TPixel>.Instance.FromVector4Destructive(this.configuration, sourceRowSpan.Slice(0, this.bounds.Width), targetPixelSpan, PixelConversionModifiers.Premultiply);
+                }
+            }
         }
     }
 }

src/ImageSharp/Processing/Processors/Convolution/BoxBlurProcessor{TPixel}.cs

@@ -40,10 +40,9 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
-            using (var processor = new Convolution2PassProcessor<TPixel>(this.Configuration, this.KernelX, this.KernelY, false, this.Source, this.SourceRectangle))
-            {
-                processor.Apply(source);
-            }
+            using var processor = new Convolution2PassProcessor<TPixel>(this.Configuration, this.KernelX, this.KernelY, false, this.Source, this.SourceRectangle);
+
+            processor.Apply(source);
         }
 
         /// <summary>

src/ImageSharp/Processing/Processors/Convolution/Convolution2DProcessor{TPixel}.cs

@@ -3,9 +3,9 @@
 
 using System;
 using System.Numerics;
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
-
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Advanced;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -60,79 +60,105 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
-            DenseMatrix<float> matrixY = this.KernelY;
-            DenseMatrix<float> matrixX = this.KernelX;
-            bool preserveAlpha = this.PreserveAlpha;
+            using Buffer2D<TPixel> targetPixels = this.Configuration.MemoryAllocator.Allocate2D<TPixel>(source.Width, source.Height);
+
+            source.CopyTo(targetPixels);
 
             var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
-            int startY = interest.Y;
-            int endY = interest.Bottom;
-            int startX = interest.X;
-            int endX = interest.Right;
-            int maxY = endY - 1;
-            int maxX = endX - 1;
-
-            using (Buffer2D<TPixel> targetPixels = this.Configuration.MemoryAllocator.Allocate2D<TPixel>(source.Width, source.Height))
+            var operation = new RowIntervalOperation(interest, targetPixels, source.PixelBuffer, this.KernelY, this.KernelX, this.Configuration, this.PreserveAlpha);
+
+            ParallelRowIterator.IterateRows<RowIntervalOperation, Vector4>(
+                this.Configuration,
+                interest,
+                in operation);
+
+            Buffer2D<TPixel>.SwapOrCopyContent(source.PixelBuffer, targetPixels);
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the convolution logic for <see cref="Convolution2DProcessor{T}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation : IRowIntervalOperation<Vector4>
+        {
+            private readonly Rectangle bounds;
+            private readonly int maxY;
+            private readonly int maxX;
+            private readonly Buffer2D<TPixel> targetPixels;
+            private readonly Buffer2D<TPixel> sourcePixels;
+            private readonly DenseMatrix<float> kernelY;
+            private readonly DenseMatrix<float> kernelX;
+            private readonly Configuration configuration;
+            private readonly bool preserveAlpha;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Rectangle bounds,
+                Buffer2D<TPixel> targetPixels,
+                Buffer2D<TPixel> sourcePixels,
+                DenseMatrix<float> kernelY,
+                DenseMatrix<float> kernelX,
+                Configuration configuration,
+                bool preserveAlpha)
             {
-                source.CopyTo(targetPixels);
+                this.bounds = bounds;
+                this.maxY = this.bounds.Bottom - 1;
+                this.maxX = this.bounds.Right - 1;
+                this.targetPixels = targetPixels;
+                this.sourcePixels = sourcePixels;
+                this.kernelY = kernelY;
+                this.kernelX = kernelX;
+                this.configuration = configuration;
+                this.preserveAlpha = preserveAlpha;
+            }
 
-                var workingRectangle = Rectangle.FromLTRB(startX, startY, endX, endY);
-                int width = workingRectangle.Width;
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows, Span<Vector4> span)
+            {
+                ref Vector4 spanRef = ref MemoryMarshal.GetReference(span);
 
-                ParallelHelper.IterateRowsWithTempBuffer<Vector4>(
-                    workingRectangle,
-                    this.Configuration,
-                    (rows, vectorBuffer) =>
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> targetRowSpan = this.targetPixels.GetRowSpan(y).Slice(this.bounds.X);
+                    PixelOperations<TPixel>.Instance.ToVector4(this.configuration, targetRowSpan.Slice(0, span.Length), span);
+
+                    if (this.preserveAlpha)
+                    {
+                        for (int x = 0; x < this.bounds.Width; x++)
+                        {
+                            DenseMatrixUtils.Convolve2D3(
+                                in this.kernelY,
+                                in this.kernelX,
+                                this.sourcePixels,
+                                ref spanRef,
+                                y,
+                                x,
+                                this.bounds.Y,
+                                this.maxY,
+                                this.bounds.X,
+                                this.maxX);
+                        }
+                    }
+                    else
+                    {
+                        for (int x = 0; x < this.bounds.Width; x++)
                         {
-                            Span<Vector4> vectorSpan = vectorBuffer.Span;
-                            int length = vectorSpan.Length;
-                            ref Vector4 vectorSpanRef = ref MemoryMarshal.GetReference(vectorSpan);
-
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                Span<TPixel> targetRowSpan = targetPixels.GetRowSpan(y).Slice(startX);
-                                PixelOperations<TPixel>.Instance.ToVector4(this.Configuration, targetRowSpan.Slice(0, length), vectorSpan);
-
-                                if (preserveAlpha)
-                                {
-                                    for (int x = 0; x < width; x++)
-                                    {
-                                        DenseMatrixUtils.Convolve2D3(
-                                            in matrixY,
-                                            in matrixX,
-                                            source.PixelBuffer,
-                                            ref vectorSpanRef,
-                                            y,
-                                            x,
-                                            startY,
-                                            maxY,
-                                            startX,
-                                            maxX);
-                                    }
-                                }
-                                else
-                                {
-                                    for (int x = 0; x < width; x++)
-                                    {
-                                        DenseMatrixUtils.Convolve2D4(
-                                            in matrixY,
-                                            in matrixX,
-                                            source.PixelBuffer,
-                                            ref vectorSpanRef,
-                                            y,
-                                            x,
-                                            startY,
-                                            maxY,
-                                            startX,
-                                            maxX);
-                                    }
-                                }
-
-                                PixelOperations<TPixel>.Instance.FromVector4Destructive(this.Configuration, vectorSpan, targetRowSpan);
-                            }
-                        });
-
-                Buffer2D<TPixel>.SwapOrCopyContent(source.PixelBuffer, targetPixels);
+                            DenseMatrixUtils.Convolve2D4(
+                                in this.kernelY,
+                                in this.kernelX,
+                                this.sourcePixels,
+                                ref spanRef,
+                                y,
+                                x,
+                                this.bounds.Y,
+                                this.maxY,
+                                this.bounds.X,
+                                this.maxX);
+                        }
+                    }
+
+                    PixelOperations<TPixel>.Instance.FromVector4Destructive(this.configuration, span, targetRowSpan);
+                }
             }
         }
     }

src/ImageSharp/Processing/Processors/Convolution/Convolution2PassProcessor{TPixel}.cs

@@ -3,9 +3,9 @@
 
 using System;
 using System.Numerics;
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
-
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Advanced;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -59,95 +59,104 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
-            using (Buffer2D<TPixel> firstPassPixels = this.Configuration.MemoryAllocator.Allocate2D<TPixel>(source.Size()))
-            {
-                var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
-                this.ApplyConvolution(firstPassPixels, source.PixelBuffer, interest, this.KernelX, this.Configuration);
-                this.ApplyConvolution(source.PixelBuffer, firstPassPixels, interest, this.KernelY, this.Configuration);
-            }
+            using Buffer2D<TPixel> firstPassPixels = this.Configuration.MemoryAllocator.Allocate2D<TPixel>(source.Size());
+
+            var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
+
+            // Horizontal convolution
+            var horizontalOperation = new RowIntervalOperation(interest, firstPassPixels, source.PixelBuffer, this.KernelX, this.Configuration, this.PreserveAlpha);
+            ParallelRowIterator.IterateRows<RowIntervalOperation, Vector4>(
+                this.Configuration,
+                interest,
+                in horizontalOperation);
+
+            // Vertical convolution
+            var verticalOperation = new RowIntervalOperation(interest, source.PixelBuffer, firstPassPixels, this.KernelY, this.Configuration, this.PreserveAlpha);
+            ParallelRowIterator.IterateRows<RowIntervalOperation, Vector4>(
+                this.Configuration,
+                interest,
+                in verticalOperation);
         }
 
         /// <summary>
-        /// Applies the process to the specified portion of the specified <see cref="ImageFrame{TPixel}"/> at the specified location
-        /// and with the specified size.
+        /// A <see langword="struct"/> implementing the convolution logic for <see cref="Convolution2PassProcessor{T}"/>.
         /// </summary>
-        /// <param name="targetPixels">The target pixels to apply the process to.</param>
-        /// <param name="sourcePixels">The source pixels. Cannot be null.</param>
-        /// <param name="sourceRectangle">
-        /// The <see cref="Rectangle"/> structure that specifies the portion of the image object to draw.
-        /// </param>
-        /// <param name="kernel">The kernel operator.</param>
-        /// <param name="configuration">The <see cref="Configuration"/></param>
-        private void ApplyConvolution(
-            Buffer2D<TPixel> targetPixels,
-            Buffer2D<TPixel> sourcePixels,
-            Rectangle sourceRectangle,
-            in DenseMatrix<float> kernel,
-            Configuration configuration)
+        private readonly struct RowIntervalOperation : IRowIntervalOperation<Vector4>
         {
-            DenseMatrix<float> matrix = kernel;
-            bool preserveAlpha = this.PreserveAlpha;
-
-            int startY = sourceRectangle.Y;
-            int endY = sourceRectangle.Bottom;
-            int startX = sourceRectangle.X;
-            int endX = sourceRectangle.Right;
-            int maxY = endY - 1;
-            int maxX = endX - 1;
-
-            var workingRectangle = Rectangle.FromLTRB(startX, startY, endX, endY);
-            int width = workingRectangle.Width;
-
-            ParallelHelper.IterateRowsWithTempBuffer<Vector4>(
-                workingRectangle,
-                configuration,
-                (rows, vectorBuffer) =>
-                    {
-                        Span<Vector4> vectorSpan = vectorBuffer.Span;
-                        int length = vectorSpan.Length;
-                        ref Vector4 vectorSpanRef = ref MemoryMarshal.GetReference(vectorSpan);
+            private readonly Rectangle bounds;
+            private readonly Buffer2D<TPixel> targetPixels;
+            private readonly Buffer2D<TPixel> sourcePixels;
+            private readonly DenseMatrix<float> kernel;
+            private readonly Configuration configuration;
+            private readonly bool preserveAlpha;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Rectangle bounds,
+                Buffer2D<TPixel> targetPixels,
+                Buffer2D<TPixel> sourcePixels,
+                DenseMatrix<float> kernel,
+                Configuration configuration,
+                bool preserveAlpha)
+            {
+                this.bounds = bounds;
+                this.targetPixels = targetPixels;
+                this.sourcePixels = sourcePixels;
+                this.kernel = kernel;
+                this.configuration = configuration;
+                this.preserveAlpha = preserveAlpha;
+            }
 
-                        for (int y = rows.Min; y < rows.Max; y++)
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows, Span<Vector4> span)
+            {
+                ref Vector4 spanRef = ref MemoryMarshal.GetReference(span);
+
+                int maxY = this.bounds.Bottom - 1;
+                int maxX = this.bounds.Right - 1;
+
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> targetRowSpan = this.targetPixels.GetRowSpan(y).Slice(this.bounds.X);
+                    PixelOperations<TPixel>.Instance.ToVector4(this.configuration, targetRowSpan.Slice(0, span.Length), span);
+
+                    if (this.preserveAlpha)
+                    {
+                        for (int x = 0; x < this.bounds.Width; x++)
                         {
-                            Span<TPixel> targetRowSpan = targetPixels.GetRowSpan(y).Slice(startX);
-                            PixelOperations<TPixel>.Instance.ToVector4(configuration, targetRowSpan.Slice(0, length), vectorSpan);
-
-                            if (preserveAlpha)
-                            {
-                                for (int x = 0; x < width; x++)
-                                {
-                                    DenseMatrixUtils.Convolve3(
-                                        in matrix,
-                                        sourcePixels,
-                                        ref vectorSpanRef,
-                                        y,
-                                        x,
-                                        startY,
-                                        maxY,
-                                        startX,
-                                        maxX);
-                                }
-                            }
-                            else
-                            {
-                                for (int x = 0; x < width; x++)
-                                {
-                                    DenseMatrixUtils.Convolve4(
-                                        in matrix,
-                                        sourcePixels,
-                                        ref vectorSpanRef,
-                                        y,
-                                        x,
-                                        startY,
-                                        maxY,
-                                        startX,
-                                        maxX);
-                                }
-                            }
-
-                            PixelOperations<TPixel>.Instance.FromVector4Destructive(configuration, vectorSpan, targetRowSpan);
+                            DenseMatrixUtils.Convolve3(
+                                in this.kernel,
+                                this.sourcePixels,
+                                ref spanRef,
+                                y,
+                                x,
+                                this.bounds.Y,
+                                maxY,
+                                this.bounds.X,
+                                maxX);
                         }
-                    });
+                    }
+                    else
+                    {
+                        for (int x = 0; x < this.bounds.Width; x++)
+                        {
+                            DenseMatrixUtils.Convolve4(
+                                in this.kernel,
+                                this.sourcePixels,
+                                ref spanRef,
+                                y,
+                                x,
+                                this.bounds.Y,
+                                maxY,
+                                this.bounds.X,
+                                maxX);
+                        }
+                    }
+
+                    PixelOperations<TPixel>.Instance.FromVector4Destructive(this.configuration, span, targetRowSpan);
+                }
+            }
         }
     }
 }

src/ImageSharp/Processing/Processors/Convolution/ConvolutionProcessor{TPixel}.cs

@@ -3,9 +3,9 @@
 
 using System;
 using System.Numerics;
+using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
-
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Advanced;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -51,76 +51,99 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
-            DenseMatrix<float> matrix = this.KernelXY;
-            bool preserveAlpha = this.PreserveAlpha;
+            using Buffer2D<TPixel> targetPixels = this.Configuration.MemoryAllocator.Allocate2D<TPixel>(source.Size());
+
+            source.CopyTo(targetPixels);
 
             var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
-            int startY = interest.Y;
-            int endY = interest.Bottom;
-            int startX = interest.X;
-            int endX = interest.Right;
-            int maxY = endY - 1;
-            int maxX = endX - 1;
+            var operation = new RowIntervalOperation(interest, targetPixels, source.PixelBuffer, this.KernelXY, this.Configuration, this.PreserveAlpha);
+            ParallelRowIterator.IterateRows<RowIntervalOperation, Vector4>(
+                this.Configuration,
+                interest,
+                in operation);
 
-            using (Buffer2D<TPixel> targetPixels = this.Configuration.MemoryAllocator.Allocate2D<TPixel>(source.Size()))
-            {
-                source.CopyTo(targetPixels);
+            Buffer2D<TPixel>.SwapOrCopyContent(source.PixelBuffer, targetPixels);
+        }
 
-                var workingRectangle = Rectangle.FromLTRB(startX, startY, endX, endY);
-                int width = workingRectangle.Width;
+        /// <summary>
+        /// A <see langword="struct"/> implementing the convolution logic for <see cref="ConvolutionProcessor{T}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation : IRowIntervalOperation<Vector4>
+        {
+            private readonly Rectangle bounds;
+            private readonly int maxY;
+            private readonly int maxX;
+            private readonly Buffer2D<TPixel> targetPixels;
+            private readonly Buffer2D<TPixel> sourcePixels;
+            private readonly DenseMatrix<float> kernel;
+            private readonly Configuration configuration;
+            private readonly bool preserveAlpha;
 
-                ParallelHelper.IterateRowsWithTempBuffer<Vector4>(
-                    workingRectangle,
-                    this.Configuration,
-                    (rows, vectorBuffer) =>
-                        {
-                            Span<Vector4> vectorSpan = vectorBuffer.Span;
-                            int length = vectorSpan.Length;
-                            ref Vector4 vectorSpanRef = ref MemoryMarshal.GetReference(vectorSpan);
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Rectangle bounds,
+                Buffer2D<TPixel> targetPixels,
+                Buffer2D<TPixel> sourcePixels,
+                DenseMatrix<float> kernel,
+                Configuration configuration,
+                bool preserveAlpha)
+            {
+                this.bounds = bounds;
+                this.maxY = this.bounds.Bottom - 1;
+                this.maxX = this.bounds.Right - 1;
+                this.targetPixels = targetPixels;
+                this.sourcePixels = sourcePixels;
+                this.kernel = kernel;
+                this.configuration = configuration;
+                this.preserveAlpha = preserveAlpha;
+            }
 
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                Span<TPixel> targetRowSpan = targetPixels.GetRowSpan(y).Slice(startX);
-                                PixelOperations<TPixel>.Instance.ToVector4(this.Configuration, targetRowSpan.Slice(0, length), vectorSpan);
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows, Span<Vector4> span)
+            {
+                ref Vector4 spanRef = ref MemoryMarshal.GetReference(span);
 
-                                if (preserveAlpha)
-                                {
-                                    for (int x = 0; x < width; x++)
-                                    {
-                                        DenseMatrixUtils.Convolve3(
-                                            in matrix,
-                                            source.PixelBuffer,
-                                            ref vectorSpanRef,
-                                            y,
-                                            x,
-                                            startY,
-                                            maxY,
-                                            startX,
-                                            maxX);
-                                    }
-                                }
-                                else
-                                {
-                                    for (int x = 0; x < width; x++)
-                                    {
-                                        DenseMatrixUtils.Convolve4(
-                                            in matrix,
-                                            source.PixelBuffer,
-                                            ref vectorSpanRef,
-                                            y,
-                                            x,
-                                            startY,
-                                            maxY,
-                                            startX,
-                                            maxX);
-                                    }
-                                }
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> targetRowSpan = this.targetPixels.GetRowSpan(y).Slice(this.bounds.X);
+                    PixelOperations<TPixel>.Instance.ToVector4(this.configuration, targetRowSpan.Slice(0, span.Length), span);
 
-                                PixelOperations<TPixel>.Instance.FromVector4Destructive(this.Configuration, vectorSpan, targetRowSpan);
-                            }
-                        });
+                    if (this.preserveAlpha)
+                    {
+                        for (int x = 0; x < this.bounds.Width; x++)
+                        {
+                            DenseMatrixUtils.Convolve3(
+                                in this.kernel,
+                                this.sourcePixels,
+                                ref spanRef,
+                                y,
+                                x,
+                                this.bounds.Y,
+                                this.maxY,
+                                this.bounds.X,
+                                this.maxX);
+                        }
+                    }
+                    else
+                    {
+                        for (int x = 0; x < this.bounds.Width; x++)
+                        {
+                            DenseMatrixUtils.Convolve4(
+                                in this.kernel,
+                                this.sourcePixels,
+                                ref spanRef,
+                                y,
+                                x,
+                                this.bounds.Y,
+                                this.maxY,
+                                this.bounds.X,
+                                this.maxX);
+                        }
+                    }
 
-                Buffer2D<TPixel>.SwapOrCopyContent(source.PixelBuffer, targetPixels);
+                    PixelOperations<TPixel>.Instance.FromVector4Destructive(this.configuration, span, targetRowSpan);
+                }
             }
         }
     }

src/ImageSharp/Processing/Processors/Convolution/EdgeDetector2DProcessor{TPixel}.cs

@@ -63,10 +63,9 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// <inheritdoc />
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
-            using (var processor = new Convolution2DProcessor<TPixel>(this.Configuration, this.KernelX, this.KernelY, true, this.Source, this.SourceRectangle))
-            {
-                processor.Apply(source);
-            }
+            using var processor = new Convolution2DProcessor<TPixel>(this.Configuration, this.KernelX, this.KernelY, true, this.Source, this.SourceRectangle);
+
+            processor.Apply(source);
         }
     }
 }

src/ImageSharp/Processing/Processors/Convolution/EdgeDetectorCompassProcessor{TPixel}.cs

@@ -1,12 +1,10 @@
 // Copyright (c) Six Labors and contributors.
 // Licensed under the Apache License, Version 2.0.
 
-using System;
 using System.Numerics;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
-
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Advanced;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 using SixLabors.ImageSharp.Processing.Processors.Filters;
@@ -55,88 +53,79 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         {
             DenseMatrix<float>[] kernels = this.Kernels.Flatten();
 
-            int startY = this.SourceRectangle.Y;
-            int endY = this.SourceRectangle.Bottom;
-            int startX = this.SourceRectangle.X;
-            int endX = this.SourceRectangle.Right;
+            var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
 
-            // Align start/end positions.
-            int minX = Math.Max(0, startX);
-            int maxX = Math.Min(source.Width, endX);
-            int minY = Math.Max(0, startY);
-            int maxY = Math.Min(source.Height, endY);
+            // We need a clean copy for each pass to start from
+            using ImageFrame<TPixel> cleanCopy = source.Clone();
 
-            // we need a clean copy for each pass to start from
-            using (ImageFrame<TPixel> cleanCopy = source.Clone())
+            using (var processor = new ConvolutionProcessor<TPixel>(this.Configuration, kernels[0], true, this.Source, interest))
             {
-                using (var processor = new ConvolutionProcessor<TPixel>(this.Configuration, kernels[0], true, this.Source, this.SourceRectangle))
-                {
-                    processor.Apply(source);
-                }
+                processor.Apply(source);
+            }
 
-                if (kernels.Length == 1)
-                {
-                    return;
-                }
+            if (kernels.Length == 1)
+            {
+                return;
+            }
 
-                int shiftY = startY;
-                int shiftX = startX;
+            // Additional runs
+            for (int i = 1; i < kernels.Length; i++)
+            {
+                using ImageFrame<TPixel> pass = cleanCopy.Clone();
 
-                // Reset offset if necessary.
-                if (minX > 0)
+                using (var processor = new ConvolutionProcessor<TPixel>(this.Configuration, kernels[i], true, this.Source, interest))
                 {
-                    shiftX = 0;
+                    processor.Apply(pass);
                 }
 
-                if (minY > 0)
-                {
-                    shiftY = 0;
-                }
+                var operation = new RowIntervalOperation(source.PixelBuffer, pass.PixelBuffer, interest);
+                ParallelRowIterator.IterateRows(
+                    this.Configuration,
+                    interest,
+                    in operation);
+            }
+        }
 
-                var workingRect = Rectangle.FromLTRB(minX, minY, maxX, maxY);
+        /// <summary>
+        /// A <see langword="struct"/> implementing the convolution logic for <see cref="EdgeDetectorCompassProcessor{T}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Buffer2D<TPixel> targetPixels;
+            private readonly Buffer2D<TPixel> passPixels;
+            private readonly int minX;
+            private readonly int maxX;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Buffer2D<TPixel> targetPixels,
+                Buffer2D<TPixel> passPixels,
+                Rectangle bounds)
+            {
+                this.targetPixels = targetPixels;
+                this.passPixels = passPixels;
+                this.minX = bounds.X;
+                this.maxX = bounds.Right;
+            }
 
-                // Additional runs.
-                // ReSharper disable once ForCanBeConvertedToForeach
-                for (int i = 1; i < kernels.Length; i++)
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
                 {
-                    using (ImageFrame<TPixel> pass = cleanCopy.Clone())
+                    ref TPixel passPixelsBase = ref MemoryMarshal.GetReference(this.passPixels.GetRowSpan(y));
+                    ref TPixel targetPixelsBase = ref MemoryMarshal.GetReference(this.targetPixels.GetRowSpan(y));
+
+                    for (int x = this.minX; x < this.maxX; x++)
                     {
-                        using (var processor = new ConvolutionProcessor<TPixel>(this.Configuration, kernels[i], true, this.Source, this.SourceRectangle))
-                        {
-                            processor.Apply(pass);
-                        }
-
-                        Buffer2D<TPixel> passPixels = pass.PixelBuffer;
-                        Buffer2D<TPixel> targetPixels = source.PixelBuffer;
-
-                        ParallelHelper.IterateRows(
-                            workingRect,
-                            this.Configuration,
-                            rows =>
-                                {
-                                    for (int y = rows.Min; y < rows.Max; y++)
-                                    {
-                                        int offsetY = y - shiftY;
-
-                                        ref TPixel passPixelsBase = ref MemoryMarshal.GetReference(passPixels.GetRowSpan(offsetY));
-                                        ref TPixel targetPixelsBase = ref MemoryMarshal.GetReference(targetPixels.GetRowSpan(offsetY));
-
-                                        for (int x = minX; x < maxX; x++)
-                                        {
-                                            int offsetX = x - shiftX;
-
-                                            // Grab the max components of the two pixels
-                                            ref TPixel currentPassPixel = ref Unsafe.Add(ref passPixelsBase, offsetX);
-                                            ref TPixel currentTargetPixel = ref Unsafe.Add(ref targetPixelsBase, offsetX);
-
-                                            var pixelValue = Vector4.Max(
-                                                currentPassPixel.ToVector4(),
-                                                currentTargetPixel.ToVector4());
-
-                                            currentTargetPixel.FromVector4(pixelValue);
-                                        }
-                                    }
-                                });
+                        // Grab the max components of the two pixels
+                        ref TPixel currentPassPixel = ref Unsafe.Add(ref passPixelsBase, x);
+                        ref TPixel currentTargetPixel = ref Unsafe.Add(ref targetPixelsBase, x);
+
+                        var pixelValue = Vector4.Max(currentPassPixel.ToVector4(), currentTargetPixel.ToVector4());
+
+                        currentTargetPixel.FromVector4(pixelValue);
                     }
                 }
             }

src/ImageSharp/Processing/Processors/Convolution/GaussianBlurProcessor{TPixel}.cs

@@ -44,10 +44,9 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
-            using (var processor = new Convolution2PassProcessor<TPixel>(this.Configuration, this.KernelX, this.KernelY, false, this.Source, this.SourceRectangle))
-            {
-                processor.Apply(source);
-            }
+            using var processor = new Convolution2PassProcessor<TPixel>(this.Configuration, this.KernelX, this.KernelY, false, this.Source, this.SourceRectangle);
+
+            processor.Apply(source);
         }
     }
 }

src/ImageSharp/Processing/Processors/Convolution/GaussianSharpenProcessor{TPixel}.cs

@@ -44,10 +44,9 @@ namespace SixLabors.ImageSharp.Processing.Processors.Convolution
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
-            using (var processor = new Convolution2PassProcessor<TPixel>(this.Configuration, this.KernelX, this.KernelY, false, this.Source, this.SourceRectangle))
-            {
-                processor.Apply(source);
-            }
+            using var processor = new Convolution2PassProcessor<TPixel>(this.Configuration, this.KernelX, this.KernelY, false, this.Source, this.SourceRectangle);
+
+            processor.Apply(source);
         }
     }
 }

src/ImageSharp/Processing/Processors/Drawing/DrawImageProcessor{TPixelBg,TPixelFg}.cs

@@ -2,9 +2,9 @@
 // Licensed under the Apache License, Version 2.0.
 
 using System;
-
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Drawing
@@ -99,18 +99,62 @@ namespace SixLabors.ImageSharp.Processing.Processors.Drawing
                     "Cannot draw image because the source image does not overlap the target image.");
             }
 
-            ParallelHelper.IterateRows(
-                workingRect,
+            var operation = new RowIntervalOperation(source, targetImage, blender, configuration, minX, width, locationY, targetX, this.Opacity);
+            ParallelRowIterator.IterateRows(
                 configuration,
-                rows =>
+                workingRect,
+                in operation);
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the draw logic for <see cref="DrawImageProcessor{TPixelBg,TPixelFg}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly ImageFrame<TPixelBg> sourceFrame;
+            private readonly Image<TPixelFg> targetImage;
+            private readonly PixelBlender<TPixelBg> blender;
+            private readonly Configuration configuration;
+            private readonly int minX;
+            private readonly int width;
+            private readonly int locationY;
+            private readonly int targetX;
+            private readonly float opacity;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                ImageFrame<TPixelBg> sourceFrame,
+                Image<TPixelFg> targetImage,
+                PixelBlender<TPixelBg> blender,
+                Configuration configuration,
+                int minX,
+                int width,
+                int locationY,
+                int targetX,
+                float opacity)
+            {
+                this.sourceFrame = sourceFrame;
+                this.targetImage = targetImage;
+                this.blender = blender;
+                this.configuration = configuration;
+                this.minX = minX;
+                this.width = width;
+                this.locationY = locationY;
+                this.targetX = targetX;
+                this.opacity = opacity;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
                 {
-                    for (int y = rows.Min; y < rows.Max; y++)
-                    {
-                        Span<TPixelBg> background = source.GetPixelRowSpan(y).Slice(minX, width);
-                        Span<TPixelFg> foreground = targetImage.GetPixelRowSpan(y - locationY).Slice(targetX, width);
-                        blender.Blend<TPixelFg>(configuration, background, background, foreground, this.Opacity);
-                    }
-                });
+                    Span<TPixelBg> background = this.sourceFrame.GetPixelRowSpan(y).Slice(this.minX, this.width);
+                    Span<TPixelFg> foreground = this.targetImage.GetPixelRowSpan(y - this.locationY).Slice(this.targetX, this.width);
+                    this.blender.Blend<TPixelFg>(this.configuration, background, background, foreground, this.opacity);
+                }
+            }
         }
     }
 }

src/ImageSharp/Processing/Processors/Effects/IPixelRowDelegate.cs

@@ -0,0 +1,21 @@
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
+using System.Numerics;
+
+namespace SixLabors.ImageSharp.Processing.Processors.Effects
+{
+    /// <summary>
+    /// An <see langword="interface"/> used by the row delegates for a given <see cref="PixelRowDelegateProcessor{TPixel,TDelegate}"/> instance
+    /// </summary>
+    public interface IPixelRowDelegate
+    {
+        /// <summary>
+        /// Applies the current pixel row delegate to a target row of preprocessed pixels.
+        /// </summary>
+        /// <param name="span">The target row of <see cref="Vector4"/> pixels to process.</param>
+        /// <param name="offset">The initial horizontal and vertical offset for the input pixels to process.</param>
+        void Invoke(Span<Vector4> span, Point offset);
+    }
+}

src/ImageSharp/Processing/Processors/Effects/OilPaintingProcessor{TPixel}.cs

@@ -5,9 +5,7 @@ using System;
 using System.Buffers;
 using System.Numerics;
 using System.Runtime.CompilerServices;
-
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -45,122 +43,145 @@ namespace SixLabors.ImageSharp.Processing.Processors.Effects
                 throw new ArgumentOutOfRangeException(nameof(brushSize));
             }
 
-            int startY = this.SourceRectangle.Y;
-            int endY = this.SourceRectangle.Bottom;
-            int startX = this.SourceRectangle.X;
-            int endX = this.SourceRectangle.Right;
-            int maxY = endY - 1;
-            int maxX = endX - 1;
-
-            int radius = brushSize >> 1;
-            int levels = this.definition.Levels;
-            int rowWidth = source.Width;
-            int rectangleWidth = this.SourceRectangle.Width;
-
-            Configuration configuration = this.Configuration;
-
             using Buffer2D<TPixel> targetPixels = this.Configuration.MemoryAllocator.Allocate2D<TPixel>(source.Size());
+
             source.CopyTo(targetPixels);
 
-            var workingRect = Rectangle.FromLTRB(startX, startY, endX, endY);
-            ParallelHelper.IterateRows(
-                workingRect,
+            var operation = new RowIntervalOperation(this.SourceRectangle, targetPixels, source, this.Configuration, brushSize >> 1, this.definition.Levels);
+            ParallelRowIterator.IterateRows(
                 this.Configuration,
-                (rows) =>
-                {
-                    /* Allocate the two temporary Vector4 buffers, one for the source row and one for the target row.
-                     * The ParallelHelper.IterateRowsWithTempBuffers overload is not used in this case because
-                     * the two allocated buffers have a length equal to the width of the source image,
-                     * and not just equal to the width of the target rectangle to process.
-                     * Furthermore, there are two buffers being allocated in this case, so using that overload would
-                     * have still required the explicit allocation of the secondary buffer.
-                     * Similarly, one temporary float buffer is also allocated from the pool, and that is used
-                     * to create the target bins for all the color channels being processed.
-                     * This buffer is only rented once outside of the main processing loop, and its contents
-                     * are cleared for each loop iteration, to avoid the repeated allocation for each processed pixel. */
-                    using (IMemoryOwner<Vector4> sourceRowBuffer = configuration.MemoryAllocator.Allocate<Vector4>(rowWidth))
-                    using (IMemoryOwner<Vector4> targetRowBuffer = configuration.MemoryAllocator.Allocate<Vector4>(rowWidth))
-                    using (IMemoryOwner<float> bins = configuration.MemoryAllocator.Allocate<float>(levels * 4))
-                    {
-                        Span<Vector4> sourceRowVector4Span = sourceRowBuffer.Memory.Span;
-                        Span<Vector4> sourceRowAreaVector4Span = sourceRowVector4Span.Slice(startX, rectangleWidth);
+                this.SourceRectangle,
+                in operation);
 
-                        Span<Vector4> targetRowVector4Span = targetRowBuffer.Memory.Span;
-                        Span<Vector4> targetRowAreaVector4Span = targetRowVector4Span.Slice(startX, rectangleWidth);
-
-                        ref float binsRef = ref bins.GetReference();
-                        ref int intensityBinRef = ref Unsafe.As<float, int>(ref binsRef);
-                        ref float redBinRef = ref Unsafe.Add(ref binsRef, levels);
-                        ref float blueBinRef = ref Unsafe.Add(ref redBinRef, levels);
-                        ref float greenBinRef = ref Unsafe.Add(ref blueBinRef, levels);
-
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            Span<TPixel> sourceRowPixelSpan = source.GetPixelRowSpan(y);
-                            Span<TPixel> sourceRowAreaPixelSpan = sourceRowPixelSpan.Slice(startX, rectangleWidth);
+            Buffer2D<TPixel>.SwapOrCopyContent(source.PixelBuffer, targetPixels);
+        }
 
-                            PixelOperations<TPixel>.Instance.ToVector4(configuration, sourceRowAreaPixelSpan, sourceRowAreaVector4Span);
+        /// <summary>
+        /// A <see langword="struct"/> implementing the convolution logic for <see cref="OilPaintingProcessor{T}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Rectangle bounds;
+            private readonly Buffer2D<TPixel> targetPixels;
+            private readonly ImageFrame<TPixel> source;
+            private readonly Configuration configuration;
+            private readonly int radius;
+            private readonly int levels;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Rectangle bounds,
+                Buffer2D<TPixel> targetPixels,
+                ImageFrame<TPixel> source,
+                Configuration configuration,
+                int radius,
+                int levels)
+            {
+                this.bounds = bounds;
+                this.targetPixels = targetPixels;
+                this.source = source;
+                this.configuration = configuration;
+                this.radius = radius;
+                this.levels = levels;
+            }
 
-                            for (int x = startX; x < endX; x++)
-                            {
-                                int maxIntensity = 0;
-                                int maxIndex = 0;
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                int maxY = this.bounds.Bottom - 1;
+                int maxX = this.bounds.Right - 1;
+
+                /* Allocate the two temporary Vector4 buffers, one for the source row and one for the target row.
+                 * The ParallelHelper.IterateRowsWithTempBuffers overload is not used in this case because
+                 * the two allocated buffers have a length equal to the width of the source image,
+                 * and not just equal to the width of the target rectangle to process.
+                 * Furthermore, there are two buffers being allocated in this case, so using that overload would
+                 * have still required the explicit allocation of the secondary buffer.
+                 * Similarly, one temporary float buffer is also allocated from the pool, and that is used
+                 * to create the target bins for all the color channels being processed.
+                 * This buffer is only rented once outside of the main processing loop, and its contents
+                 * are cleared for each loop iteration, to avoid the repeated allocation for each processed pixel. */
+                using IMemoryOwner<Vector4> sourceRowBuffer = this.configuration.MemoryAllocator.Allocate<Vector4>(this.source.Width);
+                using IMemoryOwner<Vector4> targetRowBuffer = this.configuration.MemoryAllocator.Allocate<Vector4>(this.source.Width);
+                using IMemoryOwner<float> bins = this.configuration.MemoryAllocator.Allocate<float>(this.levels * 4);
+
+                Span<Vector4> sourceRowVector4Span = sourceRowBuffer.Memory.Span;
+                Span<Vector4> sourceRowAreaVector4Span = sourceRowVector4Span.Slice(this.bounds.X, this.bounds.Width);
+
+                Span<Vector4> targetRowVector4Span = targetRowBuffer.Memory.Span;
+                Span<Vector4> targetRowAreaVector4Span = targetRowVector4Span.Slice(this.bounds.X, this.bounds.Width);
+
+                ref float binsRef = ref bins.GetReference();
+                ref int intensityBinRef = ref Unsafe.As<float, int>(ref binsRef);
+                ref float redBinRef = ref Unsafe.Add(ref binsRef, this.levels);
+                ref float blueBinRef = ref Unsafe.Add(ref redBinRef, this.levels);
+                ref float greenBinRef = ref Unsafe.Add(ref blueBinRef, this.levels);
+
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> sourceRowPixelSpan = this.source.GetPixelRowSpan(y);
+                    Span<TPixel> sourceRowAreaPixelSpan = sourceRowPixelSpan.Slice(this.bounds.X, this.bounds.Width);
 
-                                // Clear the current shared buffer before processing each target pixel
-                                bins.Memory.Span.Clear();
+                    PixelOperations<TPixel>.Instance.ToVector4(this.configuration, sourceRowAreaPixelSpan, sourceRowAreaVector4Span);
 
-                                for (int fy = 0; fy <= radius; fy++)
-                                {
-                                    int fyr = fy - radius;
-                                    int offsetY = y + fyr;
+                    for (int x = this.bounds.X; x < this.bounds.Right; x++)
+                    {
+                        int maxIntensity = 0;
+                        int maxIndex = 0;
 
-                                    offsetY = offsetY.Clamp(0, maxY);
+                        // Clear the current shared buffer before processing each target pixel
+                        bins.Memory.Span.Clear();
 
-                                    Span<TPixel> sourceOffsetRow = source.GetPixelRowSpan(offsetY);
+                        for (int fy = 0; fy <= this.radius; fy++)
+                        {
+                            int fyr = fy - this.radius;
+                            int offsetY = y + fyr;
 
-                                    for (int fx = 0; fx <= radius; fx++)
-                                    {
-                                        int fxr = fx - radius;
-                                        int offsetX = x + fxr;
-                                        offsetX = offsetX.Clamp(0, maxX);
+                            offsetY = offsetY.Clamp(0, maxY);
 
-                                        var vector = sourceOffsetRow[offsetX].ToVector4();
+                            Span<TPixel> sourceOffsetRow = this.source.GetPixelRowSpan(offsetY);
 
-                                        float sourceRed = vector.X;
-                                        float sourceBlue = vector.Z;
-                                        float sourceGreen = vector.Y;
+                            for (int fx = 0; fx <= this.radius; fx++)
+                            {
+                                int fxr = fx - this.radius;
+                                int offsetX = x + fxr;
+                                offsetX = offsetX.Clamp(0, maxX);
 
-                                        int currentIntensity = (int)MathF.Round((sourceBlue + sourceGreen + sourceRed) / 3F * (levels - 1));
+                                var vector = sourceOffsetRow[offsetX].ToVector4();
 
-                                        Unsafe.Add(ref intensityBinRef, currentIntensity)++;
-                                        Unsafe.Add(ref redBinRef, currentIntensity) += sourceRed;
-                                        Unsafe.Add(ref blueBinRef, currentIntensity) += sourceBlue;
-                                        Unsafe.Add(ref greenBinRef, currentIntensity) += sourceGreen;
+                                float sourceRed = vector.X;
+                                float sourceBlue = vector.Z;
+                                float sourceGreen = vector.Y;
 
-                                        if (Unsafe.Add(ref intensityBinRef, currentIntensity) > maxIntensity)
-                                        {
-                                            maxIntensity = Unsafe.Add(ref intensityBinRef, currentIntensity);
-                                            maxIndex = currentIntensity;
-                                        }
-                                    }
+                                int currentIntensity = (int)MathF.Round((sourceBlue + sourceGreen + sourceRed) / 3F * (this.levels - 1));
 
-                                    float red = MathF.Abs(Unsafe.Add(ref redBinRef, maxIndex) / maxIntensity);
-                                    float blue = MathF.Abs(Unsafe.Add(ref blueBinRef, maxIndex) / maxIntensity);
-                                    float green = MathF.Abs(Unsafe.Add(ref greenBinRef, maxIndex) / maxIntensity);
-                                    float alpha = sourceRowVector4Span[x].W;
+                                Unsafe.Add(ref intensityBinRef, currentIntensity)++;
+                                Unsafe.Add(ref redBinRef, currentIntensity) += sourceRed;
+                                Unsafe.Add(ref blueBinRef, currentIntensity) += sourceBlue;
+                                Unsafe.Add(ref greenBinRef, currentIntensity) += sourceGreen;
 
-                                    targetRowVector4Span[x] = new Vector4(red, green, blue, alpha);
+                                if (Unsafe.Add(ref intensityBinRef, currentIntensity) > maxIntensity)
+                                {
+                                    maxIntensity = Unsafe.Add(ref intensityBinRef, currentIntensity);
+                                    maxIndex = currentIntensity;
                                 }
                             }
 
-                            Span<TPixel> targetRowAreaPixelSpan = targetPixels.GetRowSpan(y).Slice(startX, rectangleWidth);
+                            float red = MathF.Abs(Unsafe.Add(ref redBinRef, maxIndex) / maxIntensity);
+                            float blue = MathF.Abs(Unsafe.Add(ref blueBinRef, maxIndex) / maxIntensity);
+                            float green = MathF.Abs(Unsafe.Add(ref greenBinRef, maxIndex) / maxIntensity);
+                            float alpha = sourceRowVector4Span[x].W;
 
-                            PixelOperations<TPixel>.Instance.FromVector4Destructive(configuration, targetRowAreaVector4Span, targetRowAreaPixelSpan);
+                            targetRowVector4Span[x] = new Vector4(red, green, blue, alpha);
                         }
                     }
-                });
 
-            Buffer2D<TPixel>.SwapOrCopyContent(source.PixelBuffer, targetPixels);
+                    Span<TPixel> targetRowAreaPixelSpan = this.targetPixels.GetRowSpan(y).Slice(this.bounds.X, this.bounds.Width);
+
+                    PixelOperations<TPixel>.Instance.FromVector4Destructive(this.configuration, targetRowAreaVector4Span, targetRowAreaPixelSpan);
+                }
+            }
         }
     }
 }

src/ImageSharp/Processing/Processors/Effects/PixelRowDelegateProcessor.cs

@@ -1,6 +1,9 @@
 // Copyright (c) Six Labors and contributors.
 // Licensed under the Apache License, Version 2.0.
 
+using System;
+using System.Numerics;
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Effects
@@ -34,6 +37,31 @@ namespace SixLabors.ImageSharp.Processing.Processors.Effects
         /// <inheritdoc />
         public IImageProcessor<TPixel> CreatePixelSpecificProcessor<TPixel>(Configuration configuration, Image<TPixel> source, Rectangle sourceRectangle)
             where TPixel : struct, IPixel<TPixel>
-            => new PixelRowDelegateProcessor<TPixel>(configuration, this, source, sourceRectangle);
+        {
+            return new PixelRowDelegateProcessor<TPixel, PixelRowDelegate>(
+                new PixelRowDelegate(this.PixelRowOperation),
+                configuration,
+                this.Modifiers,
+                source,
+                sourceRectangle);
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the row processing logic for <see cref="PixelRowDelegateProcessor"/>.
+        /// </summary>
+        public readonly struct PixelRowDelegate : IPixelRowDelegate
+        {
+            private readonly PixelRowOperation pixelRowOperation;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public PixelRowDelegate(PixelRowOperation pixelRowOperation)
+            {
+                this.pixelRowOperation = pixelRowOperation;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(Span<Vector4> span, Point offset) => this.pixelRowOperation(span);
+        }
     }
 }

src/ImageSharp/Processing/Processors/Effects/PixelRowDelegateProcessorBase{TPixel}.cs

@@ -1,70 +0,0 @@
-// Copyright (c) Six Labors and contributors.
-// Licensed under the Apache License, Version 2.0.
-
-using System;
-using System.Numerics;
-using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
-using SixLabors.ImageSharp.PixelFormats;
-
-namespace SixLabors.ImageSharp.Processing.Processors.Effects
-{
-    /// <summary>
-    /// The base class for all processors that accept a user defined row processing delegate.
-    /// </summary>
-    /// <typeparam name="TPixel">The pixel format.</typeparam>
-    internal abstract class PixelRowDelegateProcessorBase<TPixel> : ImageProcessor<TPixel>
-        where TPixel : struct, IPixel<TPixel>
-    {
-        /// <summary>
-        /// The <see cref="PixelConversionModifiers"/> to apply during the pixel conversions.
-        /// </summary>
-        private readonly PixelConversionModifiers modifiers;
-
-        /// <summary>
-        /// Initializes a new instance of the <see cref="PixelRowDelegateProcessorBase{TPixel}"/> class.
-        /// </summary>
-        /// <param name="configuration">The configuration which allows altering default behaviour or extending the library.</param>
-        /// <param name="modifiers">The <see cref="PixelConversionModifiers"/> to apply during the pixel conversions.</param>
-        /// <param name="source">The source <see cref="Image{TPixel}"/> for the current processor instance.</param>
-        /// <param name="sourceRectangle">The source area to process for the current processor instance.</param>
-        protected PixelRowDelegateProcessorBase(Configuration configuration, PixelConversionModifiers modifiers, Image<TPixel> source, Rectangle sourceRectangle)
-            : base(configuration, source, sourceRectangle)
-            => this.modifiers = modifiers;
-
-        /// <inheritdoc/>
-        protected override void OnFrameApply(ImageFrame<TPixel> source)
-        {
-            var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
-            int startX = interest.X;
-            Configuration configuration = this.Configuration;
-            PixelConversionModifiers modifiers = this.modifiers;
-
-            ParallelHelper.IterateRowsWithTempBuffer<Vector4>(
-                interest,
-                this.Configuration,
-                (rows, vectorBuffer) =>
-                {
-                    for (int y = rows.Min; y < rows.Max; y++)
-                    {
-                        Span<Vector4> vectorSpan = vectorBuffer.Span;
-                        int length = vectorSpan.Length;
-                        Span<TPixel> rowSpan = source.GetPixelRowSpan(y).Slice(startX, length);
-                        PixelOperations<TPixel>.Instance.ToVector4(configuration, rowSpan, vectorSpan, modifiers);
-
-                        // Run the user defined pixel shader to the current row of pixels
-                        this.ApplyPixelRowDelegate(vectorSpan, new Point(startX, y));
-
-                        PixelOperations<TPixel>.Instance.FromVector4Destructive(configuration, vectorSpan, rowSpan, modifiers);
-                    }
-                });
-        }
-
-        /// <summary>
-        /// Applies the current pixel row delegate to a target row of preprocessed pixels.
-        /// </summary>
-        /// <param name="span">The target row of <see cref="Vector4"/> pixels to process.</param>
-        /// <param name="offset">The initial horizontal and vertical offset for the input pixels to process.</param>
-        protected abstract void ApplyPixelRowDelegate(Span<Vector4> span, Point offset);
-    }
-}

src/ImageSharp/Processing/Processors/Effects/PixelRowDelegateProcessor{TPixel,TDelegate}.cs

@@ -0,0 +1,104 @@
+// Copyright (c) Six Labors and contributors.
+// Licensed under the Apache License, Version 2.0.
+
+using System;
+using System.Numerics;
+using System.Runtime.CompilerServices;
+using SixLabors.ImageSharp.Advanced;
+using SixLabors.ImageSharp.Memory;
+using SixLabors.ImageSharp.PixelFormats;
+
+namespace SixLabors.ImageSharp.Processing.Processors.Effects
+{
+    /// <summary>
+    /// The base class for all processors that accept a user defined row processing delegate.
+    /// </summary>
+    /// <typeparam name="TPixel">The pixel format.</typeparam>
+    /// <typeparam name="TDelegate">The row processor type.</typeparam>
+    internal sealed class PixelRowDelegateProcessor<TPixel, TDelegate> : ImageProcessor<TPixel>
+        where TPixel : struct, IPixel<TPixel>
+        where TDelegate : struct, IPixelRowDelegate
+    {
+        private readonly TDelegate rowDelegate;
+
+        /// <summary>
+        /// The <see cref="PixelConversionModifiers"/> to apply during the pixel conversions.
+        /// </summary>
+        private readonly PixelConversionModifiers modifiers;
+
+        /// <summary>
+        /// Initializes a new instance of the <see cref="PixelRowDelegateProcessor{TPixel,TDelegate}"/> class.
+        /// </summary>
+        /// <param name="rowDelegate">The row processor to use to process each pixel row</param>
+        /// <param name="configuration">The configuration which allows altering default behaviour or extending the library.</param>
+        /// <param name="modifiers">The <see cref="PixelConversionModifiers"/> to apply during the pixel conversions.</param>
+        /// <param name="source">The source <see cref="Image{TPixel}"/> for the current processor instance.</param>
+        /// <param name="sourceRectangle">The source area to process for the current processor instance.</param>
+        public PixelRowDelegateProcessor(
+            in TDelegate rowDelegate,
+            Configuration configuration,
+            PixelConversionModifiers modifiers,
+            Image<TPixel> source,
+            Rectangle sourceRectangle)
+            : base(configuration, source, sourceRectangle)
+        {
+            this.rowDelegate = rowDelegate;
+            this.modifiers = modifiers;
+        }
+
+        /// <inheritdoc/>
+        protected override void OnFrameApply(ImageFrame<TPixel> source)
+        {
+            var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
+            var operation = new RowIntervalOperation(interest.X, source, this.Configuration, this.modifiers, this.rowDelegate);
+
+            ParallelRowIterator.IterateRows<RowIntervalOperation, Vector4>(
+                this.Configuration,
+                interest,
+                in operation);
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the convolution logic for <see cref="PixelRowDelegateProcessor{TPixel,TDelegate}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation : IRowIntervalOperation<Vector4>
+        {
+            private readonly int startX;
+            private readonly ImageFrame<TPixel> source;
+            private readonly Configuration configuration;
+            private readonly PixelConversionModifiers modifiers;
+            private readonly TDelegate rowProcessor;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                int startX,
+                ImageFrame<TPixel> source,
+                Configuration configuration,
+                PixelConversionModifiers modifiers,
+                in TDelegate rowProcessor)
+            {
+                this.startX = startX;
+                this.source = source;
+                this.configuration = configuration;
+                this.modifiers = modifiers;
+                this.rowProcessor = rowProcessor;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows, Span<Vector4> span)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> rowSpan = this.source.GetPixelRowSpan(y).Slice(this.startX, span.Length);
+                    PixelOperations<TPixel>.Instance.ToVector4(this.configuration, rowSpan, span, this.modifiers);
+
+                    // Run the user defined pixel shader to the current row of pixels
+                    Unsafe.AsRef(this.rowProcessor).Invoke(span, new Point(this.startX, y));
+
+                    PixelOperations<TPixel>.Instance.FromVector4Destructive(this.configuration, span, rowSpan, this.modifiers);
+                }
+            }
+        }
+    }
+}

src/ImageSharp/Processing/Processors/Effects/PixelRowDelegateProcessor{TPixel}.cs

@@ -1,39 +0,0 @@
-// Copyright (c) Six Labors and contributors.
-// Licensed under the Apache License, Version 2.0.
-
-using System;
-using System.Numerics;
-
-using SixLabors.ImageSharp.PixelFormats;
-
-namespace SixLabors.ImageSharp.Processing.Processors.Effects
-{
-    /// <summary>
-    /// Applies a user defined row processing delegate to the image.
-    /// </summary>
-    /// <typeparam name="TPixel">The pixel format.</typeparam>
-    internal sealed class PixelRowDelegateProcessor<TPixel> : PixelRowDelegateProcessorBase<TPixel>
-        where TPixel : struct, IPixel<TPixel>
-    {
-        /// <summary>
-        /// The user defined pixel row processing delegate.
-        /// </summary>
-        private readonly PixelRowOperation pixelRowOperation;
-
-        /// <summary>
-        /// Initializes a new instance of the <see cref="PixelRowDelegateProcessor{TPixel}"/> class.
-        /// </summary>
-        /// <param name="configuration">The configuration which allows altering default behaviour or extending the library.</param>
-        /// <param name="definition">The <see cref="PixelRowDelegateProcessor"/> defining the processor parameters.</param>
-        /// <param name="source">The source <see cref="Image{TPixel}"/> for the current processor instance.</param>
-        /// <param name="sourceRectangle">The source area to process for the current processor instance.</param>
-        public PixelRowDelegateProcessor(Configuration configuration, PixelRowDelegateProcessor definition, Image<TPixel> source, Rectangle sourceRectangle)
-            : base(configuration, definition.Modifiers, source, sourceRectangle)
-        {
-            this.pixelRowOperation = definition.PixelRowOperation;
-        }
-
-        /// <inheritdoc/>
-        protected override void ApplyPixelRowDelegate(Span<Vector4> span, Point offset) => this.pixelRowOperation(span);
-    }
-}

src/ImageSharp/Processing/Processors/Effects/PixelateProcessor{TPixel}.cs

@@ -3,10 +3,9 @@
 
 using System;
 using System.Collections.Generic;
+using System.Runtime.CompilerServices;
 using System.Threading.Tasks;
-
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Common;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Effects
@@ -29,86 +28,76 @@ namespace SixLabors.ImageSharp.Processing.Processors.Effects
         /// <param name="sourceRectangle">The source area to process for the current processor instance.</param>
         public PixelateProcessor(Configuration configuration, PixelateProcessor definition, Image<TPixel> source, Rectangle sourceRectangle)
             : base(configuration, source, sourceRectangle)
-        {
-            this.definition = definition;
-        }
+            => this.definition = definition;
 
         private int Size => this.definition.Size;
 
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
-            if (this.Size <= 0 || this.Size > source.Height || this.Size > source.Width)
-            {
-                throw new ArgumentOutOfRangeException(nameof(this.Size));
-            }
-
-            int startY = this.SourceRectangle.Y;
-            int endY = this.SourceRectangle.Bottom;
-            int startX = this.SourceRectangle.X;
-            int endX = this.SourceRectangle.Right;
+            var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
             int size = this.Size;
-            int offset = this.Size / 2;
 
-            // Align start/end positions.
-            int minX = Math.Max(0, startX);
-            int maxX = Math.Min(source.Width, endX);
-            int minY = Math.Max(0, startY);
-            int maxY = Math.Min(source.Height, endY);
+            Guard.MustBeBetweenOrEqualTo(size, 0, interest.Width, nameof(size));
+            Guard.MustBeBetweenOrEqualTo(size, 0, interest.Height, nameof(size));
 
-            // Reset offset if necessary.
-            if (minX > 0)
+            // Get the range on the y-plane to choose from.
+            // TODO: It would be nice to be able to pool this somehow but neither Memory<T> nor Span<T>
+            // implement IEnumerable<T>.
+            IEnumerable<int> range = EnumerableExtensions.SteppedRange(interest.Y, i => i < interest.Bottom, size);
+            Parallel.ForEach(
+            range,
+            this.Configuration.GetParallelOptions(),
+            new RowOperation(interest, size, source).Invoke);
+        }
+
+        private readonly struct RowOperation
+        {
+            private readonly int minX;
+            private readonly int maxX;
+            private readonly int maxXIndex;
+            private readonly int maxY;
+            private readonly int maxYIndex;
+            private readonly int size;
+            private readonly int radius;
+            private readonly ImageFrame<TPixel> source;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowOperation(
+                Rectangle bounds,
+                int size,
+                ImageFrame<TPixel> source)
             {
-                startX = 0;
+                this.minX = bounds.X;
+                this.maxX = bounds.Right;
+                this.maxXIndex = bounds.Right - 1;
+                this.maxY = bounds.Bottom;
+                this.maxYIndex = bounds.Bottom - 1;
+                this.size = size;
+                this.radius = size >> 1;
+                this.source = source;
             }
 
-            if (minY > 0)
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(int y)
             {
-                startY = 0;
-            }
+                Span<TPixel> rowSpan = this.source.GetPixelRowSpan(Math.Min(y + this.radius, this.maxYIndex));
 
-            // Get the range on the y-plane to choose from.
-            IEnumerable<int> range = EnumerableExtensions.SteppedRange(minY, i => i < maxY, size);
+                for (int x = this.minX; x < this.maxX; x += this.size)
+                {
+                    // Get the pixel color in the centre of the soon to be pixelated area.
+                    TPixel pixel = rowSpan[Math.Min(x + this.radius, this.maxXIndex)];
 
-            Parallel.ForEach(
-                range,
-                this.Configuration.GetParallelOptions(),
-                y =>
+                    // For each pixel in the pixelate size, set it to the centre color.
+                    for (int oY = y; oY < y + this.size && oY < this.maxY; oY++)
                     {
-                        int offsetY = y - startY;
-                        int offsetPy = offset;
-
-                        // Make sure that the offset is within the boundary of the image.
-                        while (offsetY + offsetPy >= maxY)
+                        for (int oX = x; oX < x + this.size && oX < this.maxX; oX++)
                         {
-                            offsetPy--;
+                            this.source[oX, oY] = pixel;
                         }
-
-                        Span<TPixel> row = source.GetPixelRowSpan(offsetY + offsetPy);
-
-                        for (int x = minX; x < maxX; x += size)
-                        {
-                            int offsetX = x - startX;
-                            int offsetPx = offset;
-
-                            while (x + offsetPx >= maxX)
-                            {
-                                offsetPx--;
-                            }
-
-                            // Get the pixel color in the centre of the soon to be pixelated area.
-                            TPixel pixel = row[offsetX + offsetPx];
-
-                            // For each pixel in the pixelate size, set it to the centre color.
-                            for (int l = offsetY; l < offsetY + size && l < maxY; l++)
-                            {
-                                for (int k = offsetX; k < offsetX + size && k < maxX; k++)
-                                {
-                                    source[k, l] = pixel;
-                                }
-                            }
-                        }
-                    });
+                    }
+                }
+            }
         }
     }
 }

src/ImageSharp/Processing/Processors/Effects/PositionAwarePixelRowDelegateProcessor.cs

@@ -1,6 +1,9 @@
 // Copyright (c) Six Labors and contributors.
 // Licensed under the Apache License, Version 2.0.
 
+using System;
+using System.Numerics;
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Effects
@@ -34,6 +37,31 @@ namespace SixLabors.ImageSharp.Processing.Processors.Effects
         /// <inheritdoc />
         public IImageProcessor<TPixel> CreatePixelSpecificProcessor<TPixel>(Configuration configuration, Image<TPixel> source, Rectangle sourceRectangle)
             where TPixel : struct, IPixel<TPixel>
-            => new PositionAwarePixelRowDelegateProcessor<TPixel>(configuration, this, source, sourceRectangle);
+        {
+            return new PixelRowDelegateProcessor<TPixel, PixelRowDelegate>(
+                new PixelRowDelegate(this.PixelRowOperation),
+                configuration,
+                this.Modifiers,
+                source,
+                sourceRectangle);
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the row processing logic for <see cref="PositionAwarePixelRowDelegateProcessor"/>.
+        /// </summary>
+        public readonly struct PixelRowDelegate : IPixelRowDelegate
+        {
+            private readonly PixelRowOperation<Point> pixelRowOperation;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public PixelRowDelegate(PixelRowOperation<Point> pixelRowOperation)
+            {
+                this.pixelRowOperation = pixelRowOperation;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(Span<Vector4> span, Point offset) => this.pixelRowOperation(span, offset);
+        }
     }
 }

src/ImageSharp/Processing/Processors/Effects/PositionAwarePixelRowDelegateProcessor{TPixel}.cs

@@ -1,36 +0,0 @@
-// Copyright (c) Six Labors and contributors.
-// Licensed under the Apache License, Version 2.0.
-
-using System;
-using System.Numerics;
-
-using SixLabors.ImageSharp.PixelFormats;
-
-namespace SixLabors.ImageSharp.Processing.Processors.Effects
-{
-    /// <summary>
-    /// Applies a user defined, position aware, row processing delegate to the image.
-    /// </summary>
-    /// <typeparam name="TPixel">The pixel format.</typeparam>
-    internal sealed class PositionAwarePixelRowDelegateProcessor<TPixel> : PixelRowDelegateProcessorBase<TPixel>
-        where TPixel : struct, IPixel<TPixel>
-    {
-        private readonly PixelRowOperation<Point> pixelRowOperation;
-
-        /// <summary>
-        /// Initializes a new instance of the <see cref="PositionAwarePixelRowDelegateProcessor{TPixel}"/> class.
-        /// </summary>
-        /// <param name="configuration">The configuration which allows altering default behaviour or extending the library.</param>
-        /// <param name="definition">The <see cref="PositionAwarePixelRowDelegateProcessor"/> defining the processor parameters.</param>
-        /// <param name="source">The source <see cref="Image{TPixel}"/> for the current processor instance.</param>
-        /// <param name="sourceRectangle">The source area to process for the current processor instance.</param>
-        public PositionAwarePixelRowDelegateProcessor(Configuration configuration, PositionAwarePixelRowDelegateProcessor definition, Image<TPixel> source, Rectangle sourceRectangle)
-            : base(configuration, definition.Modifiers, source, sourceRectangle)
-        {
-            this.pixelRowOperation = definition.PixelRowOperation;
-        }
-
-        /// <inheritdoc/>
-        protected override void ApplyPixelRowDelegate(Span<Vector4> span, Point offset) => this.pixelRowOperation(span, offset);
-    }
-}

src/ImageSharp/Processing/Processors/Filters/FilterProcessor{TPixel}.cs

@@ -3,8 +3,9 @@
 
 using System;
 using System.Numerics;
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Filters
@@ -35,27 +36,51 @@ namespace SixLabors.ImageSharp.Processing.Processors.Filters
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
             var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
-            int startX = interest.X;
+            var operation = new RowIntervalOperation(interest.X, source, this.definition.Matrix, this.Configuration);
 
-            ColorMatrix matrix = this.definition.Matrix;
-
-            ParallelHelper.IterateRowsWithTempBuffer<Vector4>(
-                interest,
+            ParallelRowIterator.IterateRows<RowIntervalOperation, Vector4>(
                 this.Configuration,
-                (rows, vectorBuffer) =>
-                    {
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            Span<Vector4> vectorSpan = vectorBuffer.Span;
-                            int length = vectorSpan.Length;
-                            Span<TPixel> rowSpan = source.GetPixelRowSpan(y).Slice(startX, length);
-                            PixelOperations<TPixel>.Instance.ToVector4(this.Configuration, rowSpan, vectorSpan);
-
-                            Vector4Utils.Transform(vectorSpan, ref matrix);
-
-                            PixelOperations<TPixel>.Instance.FromVector4Destructive(this.Configuration, vectorSpan, rowSpan);
-                        }
-                    });
+                interest,
+                in operation);
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the convolution logic for <see cref="FilterProcessor{TPixel}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation : IRowIntervalOperation<Vector4>
+        {
+            private readonly int startX;
+            private readonly ImageFrame<TPixel> source;
+            private readonly ColorMatrix matrix;
+            private readonly Configuration configuration;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                int startX,
+                ImageFrame<TPixel> source,
+                ColorMatrix matrix,
+                Configuration configuration)
+            {
+                this.startX = startX;
+                this.source = source;
+                this.matrix = matrix;
+                this.configuration = configuration;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows, Span<Vector4> span)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> rowSpan = this.source.GetPixelRowSpan(y).Slice(this.startX, span.Length);
+                    PixelOperations<TPixel>.Instance.ToVector4(this.configuration, rowSpan, span);
+
+                    Vector4Utils.Transform(span, ref Unsafe.AsRef(this.matrix));
+
+                    PixelOperations<TPixel>.Instance.FromVector4Destructive(this.configuration, span, rowSpan);
+                }
+            }
         }
     }
 }

src/ImageSharp/Processing/Processors/Normalization/AdaptiveHistogramEqualizationProcessor{TPixel}.cs

@@ -7,8 +7,7 @@ using System.Collections.Generic;
 using System.Numerics;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
-using System.Threading.Tasks;
-
+using SixLabors.ImageSharp.Advanced;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -81,56 +80,11 @@ namespace SixLabors.ImageSharp.Processing.Processors.Normalization
                     yStart += tileHeight;
                 }
 
-                Parallel.For(
-                    0,
-                    tileYStartPositions.Count,
-                    new ParallelOptions { MaxDegreeOfParallelism = this.Configuration.MaxDegreeOfParallelism },
-                    index =>
-                        {
-                            int y = tileYStartPositions[index].y;
-                            int cdfYY = tileYStartPositions[index].cdfY;
-
-                            // It's unfortunate that we have to do this per iteration.
-                            ref TPixel sourceBase = ref source.GetPixelReference(0, 0);
-
-                            int cdfX = 0;
-                            int x = halfTileWidth;
-                            for (int tile = 0; tile < tileCount - 1; tile++)
-                            {
-                                int tileY = 0;
-                                int yEnd = Math.Min(y + tileHeight, sourceHeight);
-                                int xEnd = Math.Min(x + tileWidth, sourceWidth);
-                                for (int dy = y; dy < yEnd; dy++)
-                                {
-                                    int dyOffSet = dy * sourceWidth;
-                                    int tileX = 0;
-                                    for (int dx = x; dx < xEnd; dx++)
-                                    {
-                                        ref TPixel pixel = ref Unsafe.Add(ref sourceBase, dyOffSet + dx);
-                                        float luminanceEqualized = InterpolateBetweenFourTiles(
-                                            pixel,
-                                            cdfData,
-                                            tileCount,
-                                            tileCount,
-                                            tileX,
-                                            tileY,
-                                            cdfX,
-                                            cdfYY,
-                                            tileWidth,
-                                            tileHeight,
-                                            luminanceLevels);
-
-                                        pixel.FromVector4(new Vector4(luminanceEqualized, luminanceEqualized, luminanceEqualized, pixel.ToVector4().W));
-                                        tileX++;
-                                    }
-
-                                    tileY++;
-                                }
-
-                                cdfX++;
-                                x += tileWidth;
-                            }
-                        });
+                var operation = new RowIntervalOperation(cdfData, tileYStartPositions, tileWidth, tileHeight, tileCount, halfTileWidth, luminanceLevels, source);
+                ParallelRowIterator.IterateRows(
+                    this.Configuration,
+                    new Rectangle(0, 0, sourceWidth, tileYStartPositions.Count),
+                    in operation);
 
                 ref TPixel pixelsBase = ref source.GetPixelReference(0, 0);
 
@@ -416,6 +370,95 @@ namespace SixLabors.ImageSharp.Processing.Processors.Normalization
         private static float LinearInterpolation(float left, float right, float t)
             => left + ((right - left) * t);
 
+        private readonly struct RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly CdfTileData cdfData;
+            private readonly List<(int y, int cdfY)> tileYStartPositions;
+            private readonly int tileWidth;
+            private readonly int tileHeight;
+            private readonly int tileCount;
+            private readonly int halfTileWidth;
+            private readonly int luminanceLevels;
+            private readonly ImageFrame<TPixel> source;
+            private readonly int sourceWidth;
+            private readonly int sourceHeight;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                CdfTileData cdfData,
+                List<(int y, int cdfY)> tileYStartPositions,
+                int tileWidth,
+                int tileHeight,
+                int tileCount,
+                int halfTileWidth,
+                int luminanceLevels,
+                ImageFrame<TPixel> source)
+            {
+                this.cdfData = cdfData;
+                this.tileYStartPositions = tileYStartPositions;
+                this.tileWidth = tileWidth;
+                this.tileHeight = tileHeight;
+                this.tileCount = tileCount;
+                this.halfTileWidth = halfTileWidth;
+                this.luminanceLevels = luminanceLevels;
+                this.source = source;
+                this.sourceWidth = source.Width;
+                this.sourceHeight = source.Height;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                ref TPixel sourceBase = ref this.source.GetPixelReference(0, 0);
+
+                for (int index = rows.Min; index < rows.Max; index++)
+                {
+                    (int y, int cdfY) tileYStartPosition = this.tileYStartPositions[index];
+                    int y = tileYStartPosition.y;
+                    int cdfYY = tileYStartPosition.cdfY;
+
+                    int cdfX = 0;
+                    int x = this.halfTileWidth;
+                    for (int tile = 0; tile < this.tileCount - 1; tile++)
+                    {
+                        int tileY = 0;
+                        int yEnd = Math.Min(y + this.tileHeight, this.sourceHeight);
+                        int xEnd = Math.Min(x + this.tileWidth, this.sourceWidth);
+                        for (int dy = y; dy < yEnd; dy++)
+                        {
+                            int dyOffSet = dy * this.sourceWidth;
+                            int tileX = 0;
+                            for (int dx = x; dx < xEnd; dx++)
+                            {
+                                ref TPixel pixel = ref Unsafe.Add(ref sourceBase, dyOffSet + dx);
+                                float luminanceEqualized = InterpolateBetweenFourTiles(
+                                    pixel,
+                                    this.cdfData,
+                                    this.tileCount,
+                                    this.tileCount,
+                                    tileX,
+                                    tileY,
+                                    cdfX,
+                                    cdfYY,
+                                    this.tileWidth,
+                                    this.tileHeight,
+                                    this.luminanceLevels);
+
+                                pixel.FromVector4(new Vector4(luminanceEqualized, luminanceEqualized, luminanceEqualized, pixel.ToVector4().W));
+                                tileX++;
+                            }
+
+                            tileY++;
+                        }
+
+                        cdfX++;
+                        x += this.tileWidth;
+                    }
+                }
+            }
+        }
+
         /// <summary>
         /// Contains the results of the cumulative distribution function for all tiles.
         /// </summary>
@@ -431,7 +474,6 @@ namespace SixLabors.ImageSharp.Processing.Processors.Normalization
             private readonly Buffer2D<int> cdfLutBuffer2D;
             private readonly int pixelsInTile;
             private readonly int sourceWidth;
-            private readonly int sourceHeight;
             private readonly int tileWidth;
             private readonly int tileHeight;
             private readonly int luminanceLevels;
@@ -453,7 +495,6 @@ namespace SixLabors.ImageSharp.Processing.Processors.Normalization
                 this.cdfMinBuffer2D = this.memoryAllocator.Allocate2D<int>(tileCountX, tileCountY);
                 this.cdfLutBuffer2D = this.memoryAllocator.Allocate2D<int>(tileCountX * luminanceLevels, tileCountY);
                 this.sourceWidth = sourceWidth;
-                this.sourceHeight = sourceHeight;
                 this.tileWidth = tileWidth;
                 this.tileHeight = tileHeight;
                 this.pixelsInTile = tileWidth * tileHeight;
@@ -470,57 +511,21 @@ namespace SixLabors.ImageSharp.Processing.Processors.Normalization
 
             public void CalculateLookupTables(ImageFrame<TPixel> source, HistogramEqualizationProcessor<TPixel> processor)
             {
-                int sourceWidth = this.sourceWidth;
-                int sourceHeight = this.sourceHeight;
-                int tileWidth = this.tileWidth;
-                int tileHeight = this.tileHeight;
-                int luminanceLevels = this.luminanceLevels;
-
-                Parallel.For(
-                    0,
-                    this.tileYStartPositions.Count,
-                    new ParallelOptions { MaxDegreeOfParallelism = this.configuration.MaxDegreeOfParallelism },
-                    index =>
-                        {
-                            int cdfX = 0;
-                            int cdfY = this.tileYStartPositions[index].cdfY;
-                            int y = this.tileYStartPositions[index].y;
-                            int endY = Math.Min(y + tileHeight, sourceHeight);
-                            ref TPixel sourceBase = ref source.GetPixelReference(0, 0);
-                            ref int cdfMinBase = ref MemoryMarshal.GetReference(this.cdfMinBuffer2D.GetRowSpan(cdfY));
-
-                            using (IMemoryOwner<int> histogramBuffer = this.memoryAllocator.Allocate<int>(luminanceLevels))
-                            {
-                                Span<int> histogram = histogramBuffer.GetSpan();
-                                ref int histogramBase = ref MemoryMarshal.GetReference(histogram);
-
-                                for (int x = 0; x < sourceWidth; x += tileWidth)
-                                {
-                                    histogram.Clear();
-                                    ref int cdfBase = ref MemoryMarshal.GetReference(this.GetCdfLutSpan(cdfX, index));
-
-                                    int xlimit = Math.Min(x + tileWidth, sourceWidth);
-                                    for (int dy = y; dy < endY; dy++)
-                                    {
-                                        int dyOffset = dy * sourceWidth;
-                                        for (int dx = x; dx < xlimit; dx++)
-                                        {
-                                            int luminance = GetLuminance(Unsafe.Add(ref sourceBase, dyOffset + dx), luminanceLevels);
-                                            histogram[luminance]++;
-                                        }
-                                    }
-
-                                    if (processor.ClipHistogramEnabled)
-                                    {
-                                        processor.ClipHistogram(histogram, processor.ClipLimit);
-                                    }
-
-                                    Unsafe.Add(ref cdfMinBase, cdfX) = processor.CalculateCdf(ref cdfBase, ref histogramBase, histogram.Length - 1);
-
-                                    cdfX++;
-                                }
-                            }
-                        });
+                var operation = new RowIntervalOperation(
+                    processor,
+                    this.memoryAllocator,
+                    this.cdfMinBuffer2D,
+                    this.cdfLutBuffer2D,
+                    this.tileYStartPositions,
+                    this.tileWidth,
+                    this.tileHeight,
+                    this.luminanceLevels,
+                    source);
+
+                ParallelRowIterator.IterateRows(
+                    this.configuration,
+                    new Rectangle(0, 0, this.sourceWidth, this.tileYStartPositions.Count),
+                    in operation);
             }
 
             [MethodImpl(InliningOptions.ShortMethod)]
@@ -548,6 +553,93 @@ namespace SixLabors.ImageSharp.Processing.Processors.Normalization
                 this.cdfMinBuffer2D.Dispose();
                 this.cdfLutBuffer2D.Dispose();
             }
+
+            private readonly struct RowIntervalOperation : IRowIntervalOperation
+            {
+                private readonly HistogramEqualizationProcessor<TPixel> processor;
+                private readonly MemoryAllocator allocator;
+                private readonly Buffer2D<int> cdfMinBuffer2D;
+                private readonly Buffer2D<int> cdfLutBuffer2D;
+                private readonly List<(int y, int cdfY)> tileYStartPositions;
+                private readonly int tileWidth;
+                private readonly int tileHeight;
+                private readonly int luminanceLevels;
+                private readonly ImageFrame<TPixel> source;
+                private readonly int sourceWidth;
+                private readonly int sourceHeight;
+
+                [MethodImpl(InliningOptions.ShortMethod)]
+                public RowIntervalOperation(
+                    HistogramEqualizationProcessor<TPixel> processor,
+                    MemoryAllocator allocator,
+                    Buffer2D<int> cdfMinBuffer2D,
+                    Buffer2D<int> cdfLutBuffer2D,
+                    List<(int y, int cdfY)> tileYStartPositions,
+                    int tileWidth,
+                    int tileHeight,
+                    int luminanceLevels,
+                    ImageFrame<TPixel> source)
+                {
+                    this.processor = processor;
+                    this.allocator = allocator;
+                    this.cdfMinBuffer2D = cdfMinBuffer2D;
+                    this.cdfLutBuffer2D = cdfLutBuffer2D;
+                    this.tileYStartPositions = tileYStartPositions;
+                    this.tileWidth = tileWidth;
+                    this.tileHeight = tileHeight;
+                    this.luminanceLevels = luminanceLevels;
+                    this.source = source;
+                    this.sourceWidth = source.Width;
+                    this.sourceHeight = source.Height;
+                }
+
+                /// <inheritdoc/>
+                [MethodImpl(InliningOptions.ShortMethod)]
+                public void Invoke(in RowInterval rows)
+                {
+                    ref TPixel sourceBase = ref this.source.GetPixelReference(0, 0);
+
+                    for (int index = rows.Min; index < rows.Max; index++)
+                    {
+                        int cdfX = 0;
+                        int cdfY = this.tileYStartPositions[index].cdfY;
+                        int y = this.tileYStartPositions[index].y;
+                        int endY = Math.Min(y + this.tileHeight, this.sourceHeight);
+                        ref int cdfMinBase = ref MemoryMarshal.GetReference(this.cdfMinBuffer2D.GetRowSpan(cdfY));
+
+                        using IMemoryOwner<int> histogramBuffer = this.allocator.Allocate<int>(this.luminanceLevels);
+                        Span<int> histogram = histogramBuffer.GetSpan();
+                        ref int histogramBase = ref MemoryMarshal.GetReference(histogram);
+
+                        for (int x = 0; x < this.sourceWidth; x += this.tileWidth)
+                        {
+                            histogram.Clear();
+                            Span<int> cdfLutSpan = this.cdfLutBuffer2D.GetRowSpan(index).Slice(cdfX * this.luminanceLevels, this.luminanceLevels);
+                            ref int cdfBase = ref MemoryMarshal.GetReference(cdfLutSpan);
+
+                            int xlimit = Math.Min(x + this.tileWidth, this.sourceWidth);
+                            for (int dy = y; dy < endY; dy++)
+                            {
+                                int dyOffset = dy * this.sourceWidth;
+                                for (int dx = x; dx < xlimit; dx++)
+                                {
+                                    int luminance = GetLuminance(Unsafe.Add(ref sourceBase, dyOffset + dx), this.luminanceLevels);
+                                    histogram[luminance]++;
+                                }
+                            }
+
+                            if (this.processor.ClipHistogramEnabled)
+                            {
+                                this.processor.ClipHistogram(histogram, this.processor.ClipLimit);
+                            }
+
+                            Unsafe.Add(ref cdfMinBase, cdfX) = this.processor.CalculateCdf(ref cdfBase, ref histogramBase, histogram.Length - 1);
+
+                            cdfX++;
+                        }
+                    }
+                }
+            }
         }
     }
 }

src/ImageSharp/Processing/Processors/Normalization/AdaptiveHistogramEqualizationSlidingWindowProcessor{TPixel}.cs

@@ -66,191 +66,101 @@ namespace SixLabors.ImageSharp.Processing.Processors.Normalization
             int halfTileWidth = halfTileHeight;
             var slidingWindowInfos = new SlidingWindowInfos(tileWidth, tileHeight, halfTileWidth, halfTileHeight, pixelInTile);
 
-            using (Buffer2D<TPixel> targetPixels = this.Configuration.MemoryAllocator.Allocate2D<TPixel>(source.Width, source.Height))
-            {
-                // Process the inner tiles, which do not require to check the borders.
-                Parallel.For(
-                    halfTileWidth,
-                    source.Width - halfTileWidth,
-                    parallelOptions,
-                    this.ProcessSlidingWindow(
-                        source,
-                        memoryAllocator,
-                        targetPixels,
-                        slidingWindowInfos,
-                        yStart: halfTileHeight,
-                        yEnd: source.Height - halfTileHeight,
-                        useFastPath: true,
-                        this.Configuration));
-
-                // Process the left border of the image.
-                Parallel.For(
-                    0,
-                    halfTileWidth,
-                    parallelOptions,
-                    this.ProcessSlidingWindow(
-                        source,
-                        memoryAllocator,
-                        targetPixels,
-                        slidingWindowInfos,
-                        yStart: 0,
-                        yEnd: source.Height,
-                        useFastPath: false,
-                        this.Configuration));
-
-                // Process the right border of the image.
-                Parallel.For(
-                    source.Width - halfTileWidth,
-                    source.Width,
-                    parallelOptions,
-                    this.ProcessSlidingWindow(
-                        source,
-                        memoryAllocator,
-                        targetPixels,
-                        slidingWindowInfos,
-                        yStart: 0,
-                        yEnd: source.Height,
-                        useFastPath: false,
-                        this.Configuration));
-
-                // Process the top border of the image.
-                Parallel.For(
-                    halfTileWidth,
-                    source.Width - halfTileWidth,
-                    parallelOptions,
-                    this.ProcessSlidingWindow(
-                        source,
-                        memoryAllocator,
-                        targetPixels,
-                        slidingWindowInfos,
-                        yStart: 0,
-                        yEnd: halfTileHeight,
-                        useFastPath: false,
-                        this.Configuration));
-
-                // Process the bottom border of the image.
-                Parallel.For(
-                    halfTileWidth,
-                    source.Width - halfTileWidth,
-                    parallelOptions,
-                    this.ProcessSlidingWindow(
-                        source,
-                        memoryAllocator,
-                        targetPixels,
-                        slidingWindowInfos,
-                        yStart: source.Height - halfTileHeight,
-                        yEnd: source.Height,
-                        useFastPath: false,
-                        this.Configuration));
-
-                Buffer2D<TPixel>.SwapOrCopyContent(source.PixelBuffer, targetPixels);
-            }
-        }
-
-        /// <summary>
-        /// Applies the sliding window equalization to one column of the image. The window is moved from top to bottom.
-        /// Moving the window one pixel down requires to remove one row from the top of the window from the histogram and
-        /// adding a new row at the bottom.
-        /// </summary>
-        /// <param name="source">The source image.</param>
-        /// <param name="memoryAllocator">The memory allocator.</param>
-        /// <param name="targetPixels">The target pixels.</param>
-        /// <param name="swInfos"><see cref="SlidingWindowInfos"/> about the sliding window dimensions.</param>
-        /// <param name="yStart">The y start position.</param>
-        /// <param name="yEnd">The y end position.</param>
-        /// <param name="useFastPath">if set to true the borders of the image will not be checked.</param>
-        /// <param name="configuration">The configuration.</param>
-        /// <returns>Action Delegate.</returns>
-        private Action<int> ProcessSlidingWindow(
-            ImageFrame<TPixel> source,
-            MemoryAllocator memoryAllocator,
-            Buffer2D<TPixel> targetPixels,
-            SlidingWindowInfos swInfos,
-            int yStart,
-            int yEnd,
-            bool useFastPath,
-            Configuration configuration)
-        {
-            return x =>
-                {
-                    using (IMemoryOwner<int> histogramBuffer = memoryAllocator.Allocate<int>(this.LuminanceLevels, AllocationOptions.Clean))
-                    using (IMemoryOwner<int> histogramBufferCopy = memoryAllocator.Allocate<int>(this.LuminanceLevels, AllocationOptions.Clean))
-                    using (IMemoryOwner<int> cdfBuffer = memoryAllocator.Allocate<int>(this.LuminanceLevels, AllocationOptions.Clean))
-                    using (IMemoryOwner<Vector4> pixelRowBuffer = memoryAllocator.Allocate<Vector4>(swInfos.TileWidth, AllocationOptions.Clean))
-                    {
-                        Span<int> histogram = histogramBuffer.GetSpan();
-                        ref int histogramBase = ref MemoryMarshal.GetReference(histogram);
-
-                        Span<int> histogramCopy = histogramBufferCopy.GetSpan();
-                        ref int histogramCopyBase = ref MemoryMarshal.GetReference(histogramCopy);
-
-                        ref int cdfBase = ref MemoryMarshal.GetReference(cdfBuffer.GetSpan());
-
-                        Span<Vector4> pixelRow = pixelRowBuffer.GetSpan();
-                        ref Vector4 pixelRowBase = ref MemoryMarshal.GetReference(pixelRow);
-
-                        // Build the initial histogram of grayscale values.
-                        for (int dy = yStart - swInfos.HalfTileHeight; dy < yStart + swInfos.HalfTileHeight; dy++)
-                        {
-                            if (useFastPath)
-                            {
-                                this.CopyPixelRowFast(source, pixelRow, x - swInfos.HalfTileWidth, dy, swInfos.TileWidth, configuration);
-                            }
-                            else
-                            {
-                                this.CopyPixelRow(source, pixelRow, x - swInfos.HalfTileWidth, dy, swInfos.TileWidth, configuration);
-                            }
-
-                            this.AddPixelsToHistogram(ref pixelRowBase, ref histogramBase, this.LuminanceLevels, pixelRow.Length);
-                        }
-
-                        for (int y = yStart; y < yEnd; y++)
-                        {
-                            if (this.ClipHistogramEnabled)
-                            {
-                                // Clipping the histogram, but doing it on a copy to keep the original un-clipped values for the next iteration.
-                                histogram.CopyTo(histogramCopy);
-                                this.ClipHistogram(histogramCopy, this.ClipLimit);
-                            }
-
-                            // Calculate the cumulative distribution function, which will map each input pixel in the current tile to a new value.
-                            int cdfMin = this.ClipHistogramEnabled
-                                             ? this.CalculateCdf(ref cdfBase, ref histogramCopyBase, histogram.Length - 1)
-                                             : this.CalculateCdf(ref cdfBase, ref histogramBase, histogram.Length - 1);
-
-                            float numberOfPixelsMinusCdfMin = swInfos.PixelInTile - cdfMin;
-
-                            // Map the current pixel to the new equalized value.
-                            int luminance = GetLuminance(source[x, y], this.LuminanceLevels);
-                            float luminanceEqualized = Unsafe.Add(ref cdfBase, luminance) / numberOfPixelsMinusCdfMin;
-                            targetPixels[x, y].FromVector4(new Vector4(luminanceEqualized, luminanceEqualized, luminanceEqualized, source[x, y].ToVector4().W));
-
-                            // Remove top most row from the histogram, mirroring rows which exceeds the borders.
-                            if (useFastPath)
-                            {
-                                this.CopyPixelRowFast(source, pixelRow, x - swInfos.HalfTileWidth, y - swInfos.HalfTileWidth, swInfos.TileWidth, configuration);
-                            }
-                            else
-                            {
-                                this.CopyPixelRow(source, pixelRow, x - swInfos.HalfTileWidth, y - swInfos.HalfTileWidth, swInfos.TileWidth, configuration);
-                            }
-
-                            this.RemovePixelsFromHistogram(ref pixelRowBase, ref histogramBase, this.LuminanceLevels, pixelRow.Length);
-
-                            // Add new bottom row to the histogram, mirroring rows which exceeds the borders.
-                            if (useFastPath)
-                            {
-                                this.CopyPixelRowFast(source, pixelRow, x - swInfos.HalfTileWidth, y + swInfos.HalfTileWidth, swInfos.TileWidth, configuration);
-                            }
-                            else
-                            {
-                                this.CopyPixelRow(source, pixelRow, x - swInfos.HalfTileWidth, y + swInfos.HalfTileWidth, swInfos.TileWidth, configuration);
-                            }
-
-                            this.AddPixelsToHistogram(ref pixelRowBase, ref histogramBase, this.LuminanceLevels, pixelRow.Length);
-                        }
-                    }
-                };
+            // TODO: If the process was able to be switched to operate in parallel rows instead of columns
+            // then we could take advantage of batching and allocate per-row buffers only once per batch.
+            using Buffer2D<TPixel> targetPixels = this.Configuration.MemoryAllocator.Allocate2D<TPixel>(source.Width, source.Height);
+
+            // Process the inner tiles, which do not require to check the borders.
+            var innerOperation = new SlidingWindowOperation(
+                    this.Configuration,
+                    this,
+                    source,
+                    memoryAllocator,
+                    targetPixels,
+                    slidingWindowInfos,
+                    yStart: halfTileHeight,
+                    yEnd: source.Height - halfTileHeight,
+                    useFastPath: true);
+
+            Parallel.For(
+                halfTileWidth,
+                source.Width - halfTileWidth,
+                parallelOptions,
+                innerOperation.Invoke);
+
+            // Process the left border of the image.
+            var leftBorderOperation = new SlidingWindowOperation(
+                    this.Configuration,
+                    this,
+                    source,
+                    memoryAllocator,
+                    targetPixels,
+                    slidingWindowInfos,
+                    yStart: 0,
+                    yEnd: source.Height,
+                    useFastPath: false);
+
+            Parallel.For(
+                0,
+                halfTileWidth,
+                parallelOptions,
+                leftBorderOperation.Invoke);
+
+            // Process the right border of the image.
+            var rightBorderOperation = new SlidingWindowOperation(
+                    this.Configuration,
+                    this,
+                    source,
+                    memoryAllocator,
+                    targetPixels,
+                    slidingWindowInfos,
+                    yStart: 0,
+                    yEnd: source.Height,
+                    useFastPath: false);
+
+            Parallel.For(
+                source.Width - halfTileWidth,
+                source.Width,
+                parallelOptions,
+                rightBorderOperation.Invoke);
+
+            // Process the top border of the image.
+            var topBorderOperation = new SlidingWindowOperation(
+                    this.Configuration,
+                    this,
+                    source,
+                    memoryAllocator,
+                    targetPixels,
+                    slidingWindowInfos,
+                    yStart: 0,
+                    yEnd: halfTileHeight,
+                    useFastPath: false);
+
+            Parallel.For(
+                halfTileWidth,
+                source.Width - halfTileWidth,
+                parallelOptions,
+                topBorderOperation.Invoke);
+
+            // Process the bottom border of the image.
+            var bottomBorderOperation = new SlidingWindowOperation(
+                    this.Configuration,
+                    this,
+                    source,
+                    memoryAllocator,
+                    targetPixels,
+                    slidingWindowInfos,
+                    yStart: source.Height - halfTileHeight,
+                    yEnd: source.Height,
+                    useFastPath: false);
+
+            Parallel.For(
+                halfTileWidth,
+                source.Width - halfTileWidth,
+                parallelOptions,
+                bottomBorderOperation.Invoke);
+
+            Buffer2D<TPixel>.SwapOrCopyContent(source.PixelBuffer, targetPixels);
         }
 
         /// <summary>
@@ -371,6 +281,141 @@ namespace SixLabors.ImageSharp.Processing.Processors.Normalization
             }
         }
 
+        /// <summary>
+        /// Applies the sliding window equalization to one column of the image. The window is moved from top to bottom.
+        /// Moving the window one pixel down requires to remove one row from the top of the window from the histogram and
+        /// adding a new row at the bottom.
+        /// </summary>
+        private readonly struct SlidingWindowOperation
+        {
+            private readonly Configuration configuration;
+            private readonly AdaptiveHistogramEqualizationSlidingWindowProcessor<TPixel> processor;
+            private readonly ImageFrame<TPixel> source;
+            private readonly MemoryAllocator memoryAllocator;
+            private readonly Buffer2D<TPixel> targetPixels;
+            private readonly SlidingWindowInfos swInfos;
+            private readonly int yStart;
+            private readonly int yEnd;
+            private readonly bool useFastPath;
+
+            /// <summary>
+            /// Initializes a new instance of the <see cref="SlidingWindowOperation"/> struct.
+            /// </summary>
+            /// <param name="configuration">The configuration.</param>
+            /// <param name="processor">The histogram processor.</param>
+            /// <param name="source">The source image.</param>
+            /// <param name="memoryAllocator">The memory allocator.</param>
+            /// <param name="targetPixels">The target pixels.</param>
+            /// <param name="swInfos"><see cref="SlidingWindowInfos"/> about the sliding window dimensions.</param>
+            /// <param name="yStart">The y start position.</param>
+            /// <param name="yEnd">The y end position.</param>
+            /// <param name="useFastPath">if set to true the borders of the image will not be checked.</param>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public SlidingWindowOperation(
+                Configuration configuration,
+                AdaptiveHistogramEqualizationSlidingWindowProcessor<TPixel> processor,
+                ImageFrame<TPixel> source,
+                MemoryAllocator memoryAllocator,
+                Buffer2D<TPixel> targetPixels,
+                SlidingWindowInfos swInfos,
+                int yStart,
+                int yEnd,
+                bool useFastPath)
+            {
+                this.configuration = configuration;
+                this.processor = processor;
+                this.source = source;
+                this.memoryAllocator = memoryAllocator;
+                this.targetPixels = targetPixels;
+                this.swInfos = swInfos;
+                this.yStart = yStart;
+                this.yEnd = yEnd;
+                this.useFastPath = useFastPath;
+            }
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(int x)
+            {
+                using (IMemoryOwner<int> histogramBuffer = this.memoryAllocator.Allocate<int>(this.processor.LuminanceLevels, AllocationOptions.Clean))
+                using (IMemoryOwner<int> histogramBufferCopy = this.memoryAllocator.Allocate<int>(this.processor.LuminanceLevels, AllocationOptions.Clean))
+                using (IMemoryOwner<int> cdfBuffer = this.memoryAllocator.Allocate<int>(this.processor.LuminanceLevels, AllocationOptions.Clean))
+                using (IMemoryOwner<Vector4> pixelRowBuffer = this.memoryAllocator.Allocate<Vector4>(this.swInfos.TileWidth, AllocationOptions.Clean))
+                {
+                    Span<int> histogram = histogramBuffer.GetSpan();
+                    ref int histogramBase = ref MemoryMarshal.GetReference(histogram);
+
+                    Span<int> histogramCopy = histogramBufferCopy.GetSpan();
+                    ref int histogramCopyBase = ref MemoryMarshal.GetReference(histogramCopy);
+
+                    ref int cdfBase = ref MemoryMarshal.GetReference(cdfBuffer.GetSpan());
+
+                    Span<Vector4> pixelRow = pixelRowBuffer.GetSpan();
+                    ref Vector4 pixelRowBase = ref MemoryMarshal.GetReference(pixelRow);
+
+                    // Build the initial histogram of grayscale values.
+                    for (int dy = this.yStart - this.swInfos.HalfTileHeight; dy < this.yStart + this.swInfos.HalfTileHeight; dy++)
+                    {
+                        if (this.useFastPath)
+                        {
+                            this.processor.CopyPixelRowFast(this.source, pixelRow, x - this.swInfos.HalfTileWidth, dy, this.swInfos.TileWidth, this.configuration);
+                        }
+                        else
+                        {
+                            this.processor.CopyPixelRow(this.source, pixelRow, x - this.swInfos.HalfTileWidth, dy, this.swInfos.TileWidth, this.configuration);
+                        }
+
+                        this.processor.AddPixelsToHistogram(ref pixelRowBase, ref histogramBase, this.processor.LuminanceLevels, pixelRow.Length);
+                    }
+
+                    for (int y = this.yStart; y < this.yEnd; y++)
+                    {
+                        if (this.processor.ClipHistogramEnabled)
+                        {
+                            // Clipping the histogram, but doing it on a copy to keep the original un-clipped values for the next iteration.
+                            histogram.CopyTo(histogramCopy);
+                            this.processor.ClipHistogram(histogramCopy, this.processor.ClipLimit);
+                        }
+
+                        // Calculate the cumulative distribution function, which will map each input pixel in the current tile to a new value.
+                        int cdfMin = this.processor.ClipHistogramEnabled
+                                         ? this.processor.CalculateCdf(ref cdfBase, ref histogramCopyBase, histogram.Length - 1)
+                                         : this.processor.CalculateCdf(ref cdfBase, ref histogramBase, histogram.Length - 1);
+
+                        float numberOfPixelsMinusCdfMin = this.swInfos.PixelInTile - cdfMin;
+
+                        // Map the current pixel to the new equalized value.
+                        int luminance = GetLuminance(this.source[x, y], this.processor.LuminanceLevels);
+                        float luminanceEqualized = Unsafe.Add(ref cdfBase, luminance) / numberOfPixelsMinusCdfMin;
+                        this.targetPixels[x, y].FromVector4(new Vector4(luminanceEqualized, luminanceEqualized, luminanceEqualized, this.source[x, y].ToVector4().W));
+
+                        // Remove top most row from the histogram, mirroring rows which exceeds the borders.
+                        if (this.useFastPath)
+                        {
+                            this.processor.CopyPixelRowFast(this.source, pixelRow, x - this.swInfos.HalfTileWidth, y - this.swInfos.HalfTileWidth, this.swInfos.TileWidth, this.configuration);
+                        }
+                        else
+                        {
+                            this.processor.CopyPixelRow(this.source, pixelRow, x - this.swInfos.HalfTileWidth, y - this.swInfos.HalfTileWidth, this.swInfos.TileWidth, this.configuration);
+                        }
+
+                        this.processor.RemovePixelsFromHistogram(ref pixelRowBase, ref histogramBase, this.processor.LuminanceLevels, pixelRow.Length);
+
+                        // Add new bottom row to the histogram, mirroring rows which exceeds the borders.
+                        if (this.useFastPath)
+                        {
+                            this.processor.CopyPixelRowFast(this.source, pixelRow, x - this.swInfos.HalfTileWidth, y + this.swInfos.HalfTileWidth, this.swInfos.TileWidth, this.configuration);
+                        }
+                        else
+                        {
+                            this.processor.CopyPixelRow(this.source, pixelRow, x - this.swInfos.HalfTileWidth, y + this.swInfos.HalfTileWidth, this.swInfos.TileWidth, this.configuration);
+                        }
+
+                        this.processor.AddPixelsToHistogram(ref pixelRowBase, ref histogramBase, this.processor.LuminanceLevels, pixelRow.Length);
+                    }
+                }
+            }
+        }
+
         private class SlidingWindowInfos
         {
             public SlidingWindowInfos(int tileWidth, int tileHeight, int halfTileWidth, int halfTileHeight, int pixelInTile)
@@ -382,15 +427,15 @@ namespace SixLabors.ImageSharp.Processing.Processors.Normalization
                 this.PixelInTile = pixelInTile;
             }
 
-            public int TileWidth { get; private set; }
+            public int TileWidth { get; }
 
-            public int TileHeight { get; private set; }
+            public int TileHeight { get; }
 
-            public int PixelInTile { get; private set; }
+            public int PixelInTile { get; }
 
-            public int HalfTileWidth { get; private set; }
+            public int HalfTileWidth { get; }
 
-            public int HalfTileHeight { get; private set; }
+            public int HalfTileHeight { get; }
         }
     }
 }

src/ImageSharp/Processing/Processors/Normalization/GlobalHistogramEqualizationProcessor{TPixel}.cs

@@ -6,9 +6,7 @@ using System.Buffers;
 using System.Numerics;
 using System.Runtime.CompilerServices;
 using System.Runtime.InteropServices;
-
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -49,64 +47,125 @@ namespace SixLabors.ImageSharp.Processing.Processors.Normalization
         {
             MemoryAllocator memoryAllocator = this.Configuration.MemoryAllocator;
             int numberOfPixels = source.Width * source.Height;
-            var workingRect = new Rectangle(0, 0, source.Width, source.Height);
+            var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
+
+            using IMemoryOwner<int> histogramBuffer = memoryAllocator.Allocate<int>(this.LuminanceLevels, AllocationOptions.Clean);
+
+            // Build the histogram of the grayscale levels
+            var grayscaleOperation = new GrayscaleLevelsRowIntervalOperation(interest, histogramBuffer, source, this.LuminanceLevels);
+            ParallelRowIterator.IterateRows(
+                this.Configuration,
+                interest,
+                in grayscaleOperation);
+
+            Span<int> histogram = histogramBuffer.GetSpan();
+            if (this.ClipHistogramEnabled)
+            {
+                this.ClipHistogram(histogram, this.ClipLimit);
+            }
+
+            using IMemoryOwner<int> cdfBuffer = memoryAllocator.Allocate<int>(this.LuminanceLevels, AllocationOptions.Clean);
+
+            // Calculate the cumulative distribution function, which will map each input pixel to a new value.
+            int cdfMin = this.CalculateCdf(
+                ref MemoryMarshal.GetReference(cdfBuffer.GetSpan()),
+                ref MemoryMarshal.GetReference(histogram),
+                histogram.Length - 1);
+
+            float numberOfPixelsMinusCdfMin = numberOfPixels - cdfMin;
+
+            // Apply the cdf to each pixel of the image
+            var cdfOperation = new CdfApplicationRowIntervalOperation(interest, cdfBuffer, source, this.LuminanceLevels, numberOfPixelsMinusCdfMin);
+            ParallelRowIterator.IterateRows(
+                this.Configuration,
+                interest,
+                in cdfOperation);
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the grayscale levels logic for <see cref="GlobalHistogramEqualizationProcessor{TPixel}"/>.
+        /// </summary>
+        private readonly struct GrayscaleLevelsRowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Rectangle bounds;
+            private readonly IMemoryOwner<int> histogramBuffer;
+            private readonly ImageFrame<TPixel> source;
+            private readonly int luminanceLevels;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public GrayscaleLevelsRowIntervalOperation(
+                Rectangle bounds,
+                IMemoryOwner<int> histogramBuffer,
+                ImageFrame<TPixel> source,
+                int luminanceLevels)
+            {
+                this.bounds = bounds;
+                this.histogramBuffer = histogramBuffer;
+                this.source = source;
+                this.luminanceLevels = luminanceLevels;
+            }
 
-            using (IMemoryOwner<int> histogramBuffer = memoryAllocator.Allocate<int>(this.LuminanceLevels, AllocationOptions.Clean))
-            using (IMemoryOwner<int> cdfBuffer = memoryAllocator.Allocate<int>(this.LuminanceLevels, AllocationOptions.Clean))
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
             {
-                // Build the histogram of the grayscale levels.
-                ParallelHelper.IterateRows(
-                    workingRect,
-                    this.Configuration,
-                    rows =>
-                        {
-                            ref int histogramBase = ref MemoryMarshal.GetReference(histogramBuffer.GetSpan());
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                ref TPixel pixelBase = ref MemoryMarshal.GetReference(source.GetPixelRowSpan(y));
-
-                                for (int x = 0; x < workingRect.Width; x++)
-                                {
-                                    int luminance = GetLuminance(Unsafe.Add(ref pixelBase, x), this.LuminanceLevels);
-                                    Unsafe.Add(ref histogramBase, luminance)++;
-                                }
-                            }
-                        });
-
-                Span<int> histogram = histogramBuffer.GetSpan();
-                if (this.ClipHistogramEnabled)
+                ref int histogramBase = ref MemoryMarshal.GetReference(this.histogramBuffer.GetSpan());
+                for (int y = rows.Min; y < rows.Max; y++)
                 {
-                    this.ClipHistogram(histogram, this.ClipLimit);
+                    ref TPixel pixelBase = ref MemoryMarshal.GetReference(this.source.GetPixelRowSpan(y));
+
+                    for (int x = 0; x < this.bounds.Width; x++)
+                    {
+                        int luminance = GetLuminance(Unsafe.Add(ref pixelBase, x), this.luminanceLevels);
+                        Unsafe.Add(ref histogramBase, luminance)++;
+                    }
                 }
+            }
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the cdf application levels logic for <see cref="GlobalHistogramEqualizationProcessor{TPixel}"/>.
+        /// </summary>
+        private readonly struct CdfApplicationRowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Rectangle bounds;
+            private readonly IMemoryOwner<int> cdfBuffer;
+            private readonly ImageFrame<TPixel> source;
+            private readonly int luminanceLevels;
+            private readonly float numberOfPixelsMinusCdfMin;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public CdfApplicationRowIntervalOperation(
+                Rectangle bounds,
+                IMemoryOwner<int> cdfBuffer,
+                ImageFrame<TPixel> source,
+                int luminanceLevels,
+                float numberOfPixelsMinusCdfMin)
+            {
+                this.bounds = bounds;
+                this.cdfBuffer = cdfBuffer;
+                this.source = source;
+                this.luminanceLevels = luminanceLevels;
+                this.numberOfPixelsMinusCdfMin = numberOfPixelsMinusCdfMin;
+            }
 
-                // Calculate the cumulative distribution function, which will map each input pixel to a new value.
-                int cdfMin = this.CalculateCdf(
-                    ref MemoryMarshal.GetReference(cdfBuffer.GetSpan()),
-                    ref MemoryMarshal.GetReference(histogram),
-                    histogram.Length - 1);
-
-                float numberOfPixelsMinusCdfMin = numberOfPixels - cdfMin;
-
-                // Apply the cdf to each pixel of the image
-                ParallelHelper.IterateRows(
-                    workingRect,
-                    this.Configuration,
-                    rows =>
-                        {
-                            ref int cdfBase = ref MemoryMarshal.GetReference(cdfBuffer.GetSpan());
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                ref TPixel pixelBase = ref MemoryMarshal.GetReference(source.GetPixelRowSpan(y));
-
-                                for (int x = 0; x < workingRect.Width; x++)
-                                {
-                                    ref TPixel pixel = ref Unsafe.Add(ref pixelBase, x);
-                                    int luminance = GetLuminance(pixel, this.LuminanceLevels);
-                                    float luminanceEqualized = Unsafe.Add(ref cdfBase, luminance) / numberOfPixelsMinusCdfMin;
-                                    pixel.FromVector4(new Vector4(luminanceEqualized, luminanceEqualized, luminanceEqualized, pixel.ToVector4().W));
-                                }
-                            }
-                        });
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                ref int cdfBase = ref MemoryMarshal.GetReference(this.cdfBuffer.GetSpan());
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    ref TPixel pixelBase = ref MemoryMarshal.GetReference(this.source.GetPixelRowSpan(y));
+
+                    for (int x = 0; x < this.bounds.Width; x++)
+                    {
+                        ref TPixel pixel = ref Unsafe.Add(ref pixelBase, x);
+                        int luminance = GetLuminance(pixel, this.luminanceLevels);
+                        float luminanceEqualized = Unsafe.Add(ref cdfBase, luminance) / this.numberOfPixelsMinusCdfMin;
+                        pixel.FromVector4(new Vector4(luminanceEqualized, luminanceEqualized, luminanceEqualized, pixel.ToVector4().W));
+                    }
+                }
             }
         }
     }

src/ImageSharp/Processing/Processors/Overlays/BackgroundColorProcessor{TPixel}.cs

@@ -3,9 +3,8 @@
 
 using System;
 using System.Buffers;
-
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -29,9 +28,7 @@ namespace SixLabors.ImageSharp.Processing.Processors.Overlays
         /// <param name="sourceRectangle">The source area to process for the current processor instance.</param>
         public BackgroundColorProcessor(Configuration configuration, BackgroundColorProcessor definition, Image<TPixel> source, Rectangle sourceRectangle)
             : base(configuration, source, sourceRectangle)
-        {
-            this.definition = definition;
-        }
+            => this.definition = definition;
 
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source)
@@ -39,65 +36,70 @@ namespace SixLabors.ImageSharp.Processing.Processors.Overlays
             TPixel color = this.definition.Color.ToPixel<TPixel>();
             GraphicsOptions graphicsOptions = this.definition.GraphicsOptions;
 
-            int startY = this.SourceRectangle.Y;
-            int endY = this.SourceRectangle.Bottom;
-            int startX = this.SourceRectangle.X;
-            int endX = this.SourceRectangle.Right;
-
-            // Align start/end positions.
-            int minX = Math.Max(0, startX);
-            int maxX = Math.Min(source.Width, endX);
-            int minY = Math.Max(0, startY);
-            int maxY = Math.Min(source.Height, endY);
-
-            // Reset offset if necessary.
-            if (minX > 0)
-            {
-                startX = 0;
-            }
-
-            if (minY > 0)
-            {
-                startY = 0;
-            }
-
-            int width = maxX - minX;
+            var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
 
-            var workingRect = Rectangle.FromLTRB(minX, minY, maxX, maxY);
             Configuration configuration = this.Configuration;
             MemoryAllocator memoryAllocator = configuration.MemoryAllocator;
 
-            using (IMemoryOwner<TPixel> colors = memoryAllocator.Allocate<TPixel>(width))
-            using (IMemoryOwner<float> amount = memoryAllocator.Allocate<float>(width))
-            {
-                // Be careful! Do not capture colorSpan & amountSpan in the lambda below!
-                Span<TPixel> colorSpan = colors.GetSpan();
-                Span<float> amountSpan = amount.GetSpan();
+            using IMemoryOwner<TPixel> colors = memoryAllocator.Allocate<TPixel>(interest.Width);
+            using IMemoryOwner<float> amount = memoryAllocator.Allocate<float>(interest.Width);
 
-                colorSpan.Fill(color);
-                amountSpan.Fill(graphicsOptions.BlendPercentage);
+            colors.GetSpan().Fill(color);
+            amount.GetSpan().Fill(graphicsOptions.BlendPercentage);
 
-                PixelBlender<TPixel> blender = PixelOperations<TPixel>.Instance.GetPixelBlender(graphicsOptions);
+            PixelBlender<TPixel> blender = PixelOperations<TPixel>.Instance.GetPixelBlender(graphicsOptions);
 
-                ParallelHelper.IterateRows(
-                    workingRect,
-                    configuration,
-                    rows =>
-                        {
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                Span<TPixel> destination =
-                                    source.GetPixelRowSpan(y - startY).Slice(minX - startX, width);
+            var operation = new RowIntervalOperation(configuration, interest, blender, amount, colors, source);
+            ParallelRowIterator.IterateRows(
+                configuration,
+                interest,
+                in operation);
+        }
 
-                                // This switched color & destination in the 2nd and 3rd places because we are applying the target color under the current one
-                                blender.Blend(
-                                    configuration,
-                                    destination,
-                                    colors.GetSpan(),
-                                    destination,
-                                    amount.GetSpan());
-                            }
-                        });
+        private readonly struct RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Configuration configuration;
+            private readonly Rectangle bounds;
+            private readonly PixelBlender<TPixel> blender;
+            private readonly IMemoryOwner<float> amount;
+            private readonly IMemoryOwner<TPixel> colors;
+            private readonly ImageFrame<TPixel> source;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Configuration configuration,
+                Rectangle bounds,
+                PixelBlender<TPixel> blender,
+                IMemoryOwner<float> amount,
+                IMemoryOwner<TPixel> colors,
+                ImageFrame<TPixel> source)
+            {
+                this.configuration = configuration;
+                this.bounds = bounds;
+                this.blender = blender;
+                this.amount = amount;
+                this.colors = colors;
+                this.source = source;
+            }
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> destination =
+                        this.source.GetPixelRowSpan(y)
+                                   .Slice(this.bounds.X, this.bounds.Width);
+
+                    // Switch color & destination in the 2nd and 3rd places because we are
+                    // applying the target color under the current one.
+                    this.blender.Blend(
+                        this.configuration,
+                        destination,
+                        this.colors.GetSpan(),
+                        destination,
+                        this.amount.GetSpan());
+                }
             }
         }
     }

src/ImageSharp/Processing/Processors/Overlays/GlowProcessor{TPixel}.cs

@@ -4,9 +4,8 @@
 using System;
 using System.Buffers;
 using System.Numerics;
-
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -39,78 +38,84 @@ namespace SixLabors.ImageSharp.Processing.Processors.Overlays
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
-            // TODO: can we simplify the rectangle calculation?
-            int startY = this.SourceRectangle.Y;
-            int endY = this.SourceRectangle.Bottom;
-            int startX = this.SourceRectangle.X;
-            int endX = this.SourceRectangle.Right;
             TPixel glowColor = this.definition.GlowColor.ToPixel<TPixel>();
-            Vector2 center = Rectangle.Center(this.SourceRectangle);
+            float blendPercent = this.definition.GraphicsOptions.BlendPercentage;
 
-            float finalRadius = this.definition.Radius.Calculate(source.Size());
+            var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
 
+            Vector2 center = Rectangle.Center(interest);
+            float finalRadius = this.definition.Radius.Calculate(interest.Size);
             float maxDistance = finalRadius > 0
-                                    ? MathF.Min(finalRadius, this.SourceRectangle.Width * .5F)
-                                    : this.SourceRectangle.Width * .5F;
+                ? MathF.Min(finalRadius, interest.Width * .5F)
+                : interest.Width * .5F;
 
-            // Align start/end positions.
-            int minX = Math.Max(0, startX);
-            int maxX = Math.Min(source.Width, endX);
-            int minY = Math.Max(0, startY);
-            int maxY = Math.Min(source.Height, endY);
+            Configuration configuration = this.Configuration;
+            MemoryAllocator allocator = configuration.MemoryAllocator;
 
-            // Reset offset if necessary.
-            if (minX > 0)
-            {
-                startX = 0;
-            }
+            using IMemoryOwner<TPixel> rowColors = allocator.Allocate<TPixel>(interest.Width);
+            rowColors.GetSpan().Fill(glowColor);
 
-            if (minY > 0)
+            var operation = new RowIntervalOperation(configuration, interest, rowColors, this.blender, center, maxDistance, blendPercent, source);
+            ParallelRowIterator.IterateRows<RowIntervalOperation, float>(
+                configuration,
+                interest,
+                in operation);
+        }
+
+        private readonly struct RowIntervalOperation : IRowIntervalOperation<float>
+        {
+            private readonly Configuration configuration;
+            private readonly Rectangle bounds;
+            private readonly PixelBlender<TPixel> blender;
+            private readonly Vector2 center;
+            private readonly float maxDistance;
+            private readonly float blendPercent;
+            private readonly IMemoryOwner<TPixel> colors;
+            private readonly ImageFrame<TPixel> source;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Configuration configuration,
+                Rectangle bounds,
+                IMemoryOwner<TPixel> colors,
+                PixelBlender<TPixel> blender,
+                Vector2 center,
+                float maxDistance,
+                float blendPercent,
+                ImageFrame<TPixel> source)
             {
-                startY = 0;
+                this.configuration = configuration;
+                this.bounds = bounds;
+                this.colors = colors;
+                this.blender = blender;
+                this.center = center;
+                this.maxDistance = maxDistance;
+                this.blendPercent = blendPercent;
+                this.source = source;
             }
 
-            int width = maxX - minX;
-            int offsetX = minX - startX;
-
-            var workingRect = Rectangle.FromLTRB(minX, minY, maxX, maxY);
-
-            float blendPercentage = this.definition.GraphicsOptions.BlendPercentage;
-            Configuration configuration = this.Configuration;
-            MemoryAllocator memoryAllocator = configuration.MemoryAllocator;
-
-            using (IMemoryOwner<TPixel> rowColors = memoryAllocator.Allocate<TPixel>(width))
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows, Span<float> span)
             {
-                rowColors.GetSpan().Fill(glowColor);
-
-                ParallelHelper.IterateRowsWithTempBuffer<float>(
-                    workingRect,
-                    configuration,
-                    (rows, amounts) =>
-                        {
-                            Span<float> amountsSpan = amounts.Span;
-
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                int offsetY = y - startY;
-
-                                for (int i = 0; i < width; i++)
-                                {
-                                    float distance = Vector2.Distance(center, new Vector2(i + offsetX, offsetY));
-                                    amountsSpan[i] =
-                                        (blendPercentage * (1 - (.95F * (distance / maxDistance)))).Clamp(0, 1);
-                                }
-
-                                Span<TPixel> destination = source.GetPixelRowSpan(offsetY).Slice(offsetX, width);
-
-                                this.blender.Blend(
-                                    configuration,
-                                    destination,
-                                    destination,
-                                    rowColors.GetSpan(),
-                                    amountsSpan);
-                            }
-                        });
+                Span<TPixel> colorSpan = this.colors.GetSpan();
+
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    for (int i = 0; i < this.bounds.Width; i++)
+                    {
+                        float distance = Vector2.Distance(this.center, new Vector2(i + this.bounds.X, y));
+                        span[i] = (this.blendPercent * (1 - (.95F * (distance / this.maxDistance)))).Clamp(0, 1);
+                    }
+
+                    Span<TPixel> destination = this.source.GetPixelRowSpan(y).Slice(this.bounds.X, this.bounds.Width);
+
+                    this.blender.Blend(
+                        this.configuration,
+                        destination,
+                        destination,
+                        colorSpan,
+                        span);
+                }
             }
         }
     }

src/ImageSharp/Processing/Processors/Overlays/VignetteProcessor{TPixel}.cs

@@ -4,9 +4,8 @@
 using System;
 using System.Buffers;
 using System.Numerics;
-
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -20,7 +19,6 @@ namespace SixLabors.ImageSharp.Processing.Processors.Overlays
         where TPixel : struct, IPixel<TPixel>
     {
         private readonly PixelBlender<TPixel> blender;
-
         private readonly VignetteProcessor definition;
 
         /// <summary>
@@ -40,80 +38,92 @@ namespace SixLabors.ImageSharp.Processing.Processors.Overlays
         /// <inheritdoc/>
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
-            int startY = this.SourceRectangle.Y;
-            int endY = this.SourceRectangle.Bottom;
-            int startX = this.SourceRectangle.X;
-            int endX = this.SourceRectangle.Right;
             TPixel vignetteColor = this.definition.VignetteColor.ToPixel<TPixel>();
-            Vector2 centre = Rectangle.Center(this.SourceRectangle);
+            float blendPercent = this.definition.GraphicsOptions.BlendPercentage;
+
+            var interest = Rectangle.Intersect(this.SourceRectangle, source.Bounds());
+
+            Vector2 center = Rectangle.Center(interest);
+            float finalRadiusX = this.definition.RadiusX.Calculate(interest.Size);
+            float finalRadiusY = this.definition.RadiusY.Calculate(interest.Size);
 
-            Size sourceSize = source.Size();
-            float finalRadiusX = this.definition.RadiusX.Calculate(sourceSize);
-            float finalRadiusY = this.definition.RadiusY.Calculate(sourceSize);
             float rX = finalRadiusX > 0
-                           ? MathF.Min(finalRadiusX, this.SourceRectangle.Width * .5F)
-                           : this.SourceRectangle.Width * .5F;
+                ? MathF.Min(finalRadiusX, interest.Width * .5F)
+                : interest.Width * .5F;
+
             float rY = finalRadiusY > 0
-                           ? MathF.Min(finalRadiusY, this.SourceRectangle.Height * .5F)
-                           : this.SourceRectangle.Height * .5F;
+                ? MathF.Min(finalRadiusY, interest.Height * .5F)
+                : interest.Height * .5F;
+
             float maxDistance = MathF.Sqrt((rX * rX) + (rY * rY));
 
-            // Align start/end positions.
-            int minX = Math.Max(0, startX);
-            int maxX = Math.Min(source.Width, endX);
-            int minY = Math.Max(0, startY);
-            int maxY = Math.Min(source.Height, endY);
+            Configuration configuration = this.Configuration;
+            MemoryAllocator allocator = configuration.MemoryAllocator;
 
-            // Reset offset if necessary.
-            if (minX > 0)
-            {
-                startX = 0;
-            }
+            using IMemoryOwner<TPixel> rowColors = allocator.Allocate<TPixel>(interest.Width);
+            rowColors.GetSpan().Fill(vignetteColor);
+
+            var operation = new RowIntervalOperation(configuration, interest, rowColors, this.blender, center, maxDistance, blendPercent, source);
+            ParallelRowIterator.IterateRows<RowIntervalOperation, float>(
+                configuration,
+                interest,
+                in operation);
+        }
 
-            if (minY > 0)
+        private readonly struct RowIntervalOperation : IRowIntervalOperation<float>
+        {
+            private readonly Configuration configuration;
+            private readonly Rectangle bounds;
+            private readonly PixelBlender<TPixel> blender;
+            private readonly Vector2 center;
+            private readonly float maxDistance;
+            private readonly float blendPercent;
+            private readonly IMemoryOwner<TPixel> colors;
+            private readonly ImageFrame<TPixel> source;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Configuration configuration,
+                Rectangle bounds,
+                IMemoryOwner<TPixel> colors,
+                PixelBlender<TPixel> blender,
+                Vector2 center,
+                float maxDistance,
+                float blendPercent,
+                ImageFrame<TPixel> source)
             {
-                startY = 0;
+                this.configuration = configuration;
+                this.bounds = bounds;
+                this.colors = colors;
+                this.blender = blender;
+                this.center = center;
+                this.maxDistance = maxDistance;
+                this.blendPercent = blendPercent;
+                this.source = source;
             }
 
-            int width = maxX - minX;
-            int offsetX = minX - startX;
-
-            var workingRect = Rectangle.FromLTRB(minX, minY, maxX, maxY);
-            float blendPercentage = this.definition.GraphicsOptions.BlendPercentage;
-            Configuration configuration = this.Configuration;
-            MemoryAllocator memoryAllocator = configuration.MemoryAllocator;
-
-            using (IMemoryOwner<TPixel> rowColors = memoryAllocator.Allocate<TPixel>(width))
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows, Span<float> span)
             {
-                rowColors.GetSpan().Fill(vignetteColor);
-
-                ParallelHelper.IterateRowsWithTempBuffer<float>(
-                    workingRect,
-                    configuration,
-                    (rows, amounts) =>
-                        {
-                            Span<float> amountsSpan = amounts.Span;
-
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                int offsetY = y - startY;
-
-                                for (int i = 0; i < width; i++)
-                                {
-                                    float distance = Vector2.Distance(centre, new Vector2(i + offsetX, offsetY));
-                                    amountsSpan[i] = (blendPercentage * (.9F * (distance / maxDistance))).Clamp(0, 1);
-                                }
-
-                                Span<TPixel> destination = source.GetPixelRowSpan(offsetY).Slice(offsetX, width);
-
-                                this.blender.Blend(
-                                    configuration,
-                                    destination,
-                                    destination,
-                                    rowColors.GetSpan(),
-                                    amountsSpan);
-                            }
-                        });
+                Span<TPixel> colorSpan = this.colors.GetSpan();
+
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    for (int i = 0; i < this.bounds.Width; i++)
+                    {
+                        float distance = Vector2.Distance(this.center, new Vector2(i + this.bounds.X, y));
+                        span[i] = (this.blendPercent * (.9F * (distance / this.maxDistance))).Clamp(0, 1);
+                    }
+
+                    Span<TPixel> destination = this.source.GetPixelRowSpan(y).Slice(this.bounds.X, this.bounds.Width);
+
+                    this.blender.Blend(
+                        this.configuration,
+                        destination,
+                        destination,
+                        colorSpan,
+                        span);
+                }
             }
         }
     }

src/ImageSharp/Processing/Processors/Quantization/QuantizeProcessor{TPixel}.cs

@@ -2,8 +2,9 @@
 // Licensed under the Apache License, Version 2.0.
 
 using System;
-
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
+using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Quantization
@@ -35,27 +36,50 @@ namespace SixLabors.ImageSharp.Processing.Processors.Quantization
         protected override void OnFrameApply(ImageFrame<TPixel> source)
         {
             Configuration configuration = this.Configuration;
-            using (IFrameQuantizer<TPixel> frameQuantizer = this.quantizer.CreateFrameQuantizer<TPixel>(configuration))
-            using (IQuantizedFrame<TPixel> quantized = frameQuantizer.QuantizeFrame(source))
-            {
-                int paletteCount = quantized.Palette.Length - 1;
+            using IFrameQuantizer<TPixel> frameQuantizer = this.quantizer.CreateFrameQuantizer<TPixel>(configuration);
+            using IQuantizedFrame<TPixel> quantized = frameQuantizer.QuantizeFrame(source);
 
-                // Not parallel to remove "quantized" closure allocation.
-                // We can operate directly on the source here as we've already read it to get the
-                // quantized result
-                for (int y = 0; y < source.Height; y++)
-                {
-                    Span<TPixel> row = source.GetPixelRowSpan(y);
-                    ReadOnlySpan<byte> quantizedPixelSpan = quantized.GetPixelSpan();
+            var operation = new RowIntervalOperation(this.SourceRectangle, source, quantized);
+            ParallelRowIterator.IterateRows(
+                configuration,
+                this.SourceRectangle,
+                in operation);
+        }
 
-                    ReadOnlySpan<TPixel> paletteSpan = quantized.Palette.Span;
+        private readonly struct RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Rectangle bounds;
+            private readonly ImageFrame<TPixel> source;
+            private readonly IQuantizedFrame<TPixel> quantized;
+            private readonly int maxPaletteIndex;
 
-                    int yy = y * source.Width;
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Rectangle bounds,
+                ImageFrame<TPixel> source,
+                IQuantizedFrame<TPixel> quantized)
+            {
+                this.bounds = bounds;
+                this.source = source;
+                this.quantized = quantized;
+                this.maxPaletteIndex = quantized.Palette.Length - 1;
+            }
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                ReadOnlySpan<byte> quantizedPixelSpan = this.quantized.GetPixelSpan();
+                ReadOnlySpan<TPixel> paletteSpan = this.quantized.Palette.Span;
+
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> row = this.source.GetPixelRowSpan(y);
+                    int yy = y * this.bounds.Width;
 
-                    for (int x = 0; x < source.Width; x++)
+                    for (int x = this.bounds.X; x < this.bounds.Right; x++)
                     {
                         int i = x + yy;
-                        row[x] = paletteSpan[Math.Min(paletteCount, quantizedPixelSpan[i])];
+                        row[x] = paletteSpan[Math.Min(this.maxPaletteIndex, quantizedPixelSpan[i])];
                     }
                 }
             }

src/ImageSharp/Processing/Processors/Transforms/AffineTransformProcessor{TPixel}.cs

@@ -3,8 +3,9 @@
 
 using System;
 using System.Numerics;
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Transforms
@@ -16,8 +17,8 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
     internal class AffineTransformProcessor<TPixel> : TransformProcessor<TPixel>
         where TPixel : struct, IPixel<TPixel>
     {
-        private Size targetSize;
-        private Matrix3x2 transformMatrix;
+        private readonly Size targetSize;
+        private readonly Matrix3x2 transformMatrix;
         private readonly IResampler resampler;
 
         /// <summary>
@@ -49,7 +50,6 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
             }
 
             int width = this.targetSize.Width;
-            Rectangle sourceBounds = this.SourceRectangle;
             var targetBounds = new Rectangle(Point.Empty, this.targetSize);
             Configuration configuration = this.Configuration;
 
@@ -58,70 +58,130 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
 
             if (this.resampler is NearestNeighborResampler)
             {
-                ParallelHelper.IterateRows(
-                    targetBounds,
+                var nnOperation = new NearestNeighborRowIntervalOperation(this.SourceRectangle, ref matrix, width, source, destination);
+                ParallelRowIterator.IterateRows(
                     configuration,
-                    rows =>
-                        {
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                Span<TPixel> destRow = destination.GetPixelRowSpan(y);
-
-                                for (int x = 0; x < width; x++)
-                                {
-                                    var point = Point.Transform(new Point(x, y), matrix);
-                                    if (sourceBounds.Contains(point.X, point.Y))
-                                    {
-                                        destRow[x] = source[point.X, point.Y];
-                                    }
-                                }
-                            }
-                        });
+                    targetBounds,
+                    in nnOperation);
 
                 return;
             }
 
-            var kernel = new TransformKernelMap(configuration, source.Size(), destination.Size(), this.resampler);
+            using var kernelMap = new TransformKernelMap(configuration, source.Size(), destination.Size(), this.resampler);
 
-            try
+            var operation = new RowIntervalOperation(configuration, kernelMap, ref matrix, width, source, destination);
+            ParallelRowIterator.IterateRows<RowIntervalOperation, Vector4>(
+                configuration,
+                targetBounds,
+                in operation);
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the nearest neighbor resampler logic for <see cref="AffineTransformProcessor{T}"/>.
+        /// </summary>
+        private readonly struct NearestNeighborRowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Rectangle bounds;
+            private readonly Matrix3x2 matrix;
+            private readonly int maxX;
+            private readonly ImageFrame<TPixel> source;
+            private readonly ImageFrame<TPixel> destination;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public NearestNeighborRowIntervalOperation(
+                Rectangle bounds,
+                ref Matrix3x2 matrix,
+                int maxX,
+                ImageFrame<TPixel> source,
+                ImageFrame<TPixel> destination)
             {
-                ParallelHelper.IterateRowsWithTempBuffer<Vector4>(
-                    targetBounds,
-                    configuration,
-                    (rows, vectorBuffer) =>
+                this.bounds = bounds;
+                this.matrix = matrix;
+                this.maxX = maxX;
+                this.source = source;
+                this.destination = destination;
+            }
+
+            /// <inheritdoc/>
+            /// <param name="rows"></param>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> destRow = this.destination.GetPixelRowSpan(y);
+
+                    for (int x = 0; x < this.maxX; x++)
+                    {
+                        var point = Point.Transform(new Point(x, y), this.matrix);
+                        if (this.bounds.Contains(point.X, point.Y))
                         {
-                            Span<Vector4> vectorSpan = vectorBuffer.Span;
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                Span<TPixel> targetRowSpan = destination.GetPixelRowSpan(y);
-                                PixelOperations<TPixel>.Instance.ToVector4(configuration, targetRowSpan, vectorSpan);
-                                ref float ySpanRef = ref kernel.GetYStartReference(y);
-                                ref float xSpanRef = ref kernel.GetXStartReference(y);
-
-                                for (int x = 0; x < width; x++)
-                                {
-                                    // Use the single precision position to calculate correct bounding pixels
-                                    // otherwise we get rogue pixels outside of the bounds.
-                                    var point = Vector2.Transform(new Vector2(x, y), matrix);
-                                    kernel.Convolve(
-                                        point,
-                                        x,
-                                        ref ySpanRef,
-                                        ref xSpanRef,
-                                        source.PixelBuffer,
-                                        vectorSpan);
-                                }
-
-                                PixelOperations<TPixel>.Instance.FromVector4Destructive(
-                                    configuration,
-                                    vectorSpan,
-                                    targetRowSpan);
-                            }
-                        });
+                            destRow[x] = this.source[point.X, point.Y];
+                        }
+                    }
+                }
             }
-            finally
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the transformation logic for <see cref="AffineTransformProcessor{T}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation : IRowIntervalOperation<Vector4>
+        {
+            private readonly Configuration configuration;
+            private readonly TransformKernelMap kernelMap;
+            private readonly Matrix3x2 matrix;
+            private readonly int maxX;
+            private readonly ImageFrame<TPixel> source;
+            private readonly ImageFrame<TPixel> destination;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Configuration configuration,
+                TransformKernelMap kernelMap,
+                ref Matrix3x2 matrix,
+                int maxX,
+                ImageFrame<TPixel> source,
+                ImageFrame<TPixel> destination)
+            {
+                this.configuration = configuration;
+                this.kernelMap = kernelMap;
+                this.matrix = matrix;
+                this.maxX = maxX;
+                this.source = source;
+                this.destination = destination;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows, Span<Vector4> span)
             {
-                kernel.Dispose();
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> targetRowSpan = this.destination.GetPixelRowSpan(y);
+                    PixelOperations<TPixel>.Instance.ToVector4(this.configuration, targetRowSpan, span);
+                    ref float ySpanRef = ref this.kernelMap.GetYStartReference(y);
+                    ref float xSpanRef = ref this.kernelMap.GetXStartReference(y);
+
+                    for (int x = 0; x < this.maxX; x++)
+                    {
+                        // Use the single precision position to calculate correct bounding pixels
+                        // otherwise we get rogue pixels outside of the bounds.
+                        var point = Vector2.Transform(new Vector2(x, y), this.matrix);
+                        this.kernelMap.Convolve(
+                            point,
+                            x,
+                            ref ySpanRef,
+                            ref xSpanRef,
+                            this.source.PixelBuffer,
+                            span);
+                    }
+
+                    PixelOperations<TPixel>.Instance.FromVector4Destructive(
+                        this.configuration,
+                        span,
+                        targetRowSpan);
+                }
             }
         }
     }

src/ImageSharp/Processing/Processors/Transforms/CropProcessor{TPixel}.cs

@@ -2,8 +2,9 @@
 // Licensed under the Apache License, Version 2.0.
 
 using System;
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Transforms
@@ -15,7 +16,7 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
     internal class CropProcessor<TPixel> : TransformProcessor<TPixel>
         where TPixel : struct, IPixel<TPixel>
     {
-        private Rectangle cropRectangle;
+        private readonly Rectangle cropRectangle;
 
         /// <summary>
         /// Initializes a new instance of the <see cref="CropProcessor{TPixel}"/> class.
@@ -47,21 +48,51 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
             Rectangle bounds = this.cropRectangle;
 
             // Copying is cheap, we should process more pixels per task:
-            ParallelExecutionSettings parallelSettings = ParallelExecutionSettings.FromConfiguration(this.Configuration)
-                .MultiplyMinimumPixelsPerTask(4);
+            ParallelExecutionSettings parallelSettings =
+                ParallelExecutionSettings.FromConfiguration(this.Configuration).MultiplyMinimumPixelsPerTask(4);
 
-            ParallelHelper.IterateRows(
+            var operation = new RowIntervalOperation(bounds, source, destination);
+
+            ParallelRowIterator.IterateRows(
                 bounds,
-                parallelSettings,
-                rows =>
-                    {
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            Span<TPixel> sourceRow = source.GetPixelRowSpan(y).Slice(bounds.Left);
-                            Span<TPixel> targetRow = destination.GetPixelRowSpan(y - bounds.Top);
-                            sourceRow.Slice(0, bounds.Width).CopyTo(targetRow);
-                        }
-                    });
+                in parallelSettings,
+                in operation);
+        }
+
+        /// <summary>
+        /// A <see langword="struct"/> implementing the processor logic for <see cref="CropProcessor{T}"/>.
+        /// </summary>
+        private readonly struct RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Rectangle bounds;
+            private readonly ImageFrame<TPixel> source;
+            private readonly ImageFrame<TPixel> destination;
+
+            /// <summary>
+            /// Initializes a new instance of the <see cref="RowIntervalOperation"/> struct.
+            /// </summary>
+            /// <param name="bounds">The target processing bounds for the current instance.</param>
+            /// <param name="source">The source <see cref="Image{TPixel}"/> for the current instance.</param>
+            /// <param name="destination">The destination <see cref="Image{TPixel}"/> for the current instance.</param>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(Rectangle bounds, ImageFrame<TPixel> source, ImageFrame<TPixel> destination)
+            {
+                this.bounds = bounds;
+                this.source = source;
+                this.destination = destination;
+            }
+
+            /// <inheritdoc/>
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> sourceRow = this.source.GetPixelRowSpan(y).Slice(this.bounds.Left);
+                    Span<TPixel> targetRow = this.destination.GetPixelRowSpan(y - this.bounds.Top);
+                    sourceRow.Slice(0, this.bounds.Width).CopyTo(targetRow);
+                }
+            }
         }
     }
 }

src/ImageSharp/Processing/Processors/Transforms/FlipProcessor{TPixel}.cs

@@ -3,9 +3,9 @@
 
 using System;
 using System.Buffers;
-
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Transforms
@@ -55,20 +55,17 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
         private void FlipX(ImageFrame<TPixel> source, Configuration configuration)
         {
             int height = source.Height;
+            using IMemoryOwner<TPixel> tempBuffer = configuration.MemoryAllocator.Allocate<TPixel>(source.Width);
+            Span<TPixel> temp = tempBuffer.Memory.Span;
 
-            using (IMemoryOwner<TPixel> tempBuffer = configuration.MemoryAllocator.Allocate<TPixel>(source.Width))
+            for (int yTop = 0; yTop < height / 2; yTop++)
             {
-                Span<TPixel> temp = tempBuffer.Memory.Span;
-
-                for (int yTop = 0; yTop < height / 2; yTop++)
-                {
-                    int yBottom = height - yTop - 1;
-                    Span<TPixel> topRow = source.GetPixelRowSpan(yBottom);
-                    Span<TPixel> bottomRow = source.GetPixelRowSpan(yTop);
-                    topRow.CopyTo(temp);
-                    bottomRow.CopyTo(topRow);
-                    temp.CopyTo(bottomRow);
-                }
+                int yBottom = height - yTop - 1;
+                Span<TPixel> topRow = source.GetPixelRowSpan(yBottom);
+                Span<TPixel> bottomRow = source.GetPixelRowSpan(yTop);
+                topRow.CopyTo(temp);
+                bottomRow.CopyTo(topRow);
+                temp.CopyTo(bottomRow);
             }
         }
 
@@ -79,16 +76,28 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
         /// <param name="configuration">The configuration.</param>
         private void FlipY(ImageFrame<TPixel> source, Configuration configuration)
         {
-            ParallelHelper.IterateRows(
-                source.Bounds(),
+            var operation = new RowIntervalOperation(source);
+            ParallelRowIterator.IterateRows(
                 configuration,
-                rows =>
-                    {
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            source.GetPixelRowSpan(y).Reverse();
-                        }
-                    });
+                source.Bounds(),
+                in operation);
+        }
+
+        private readonly struct RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly ImageFrame<TPixel> source;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(ImageFrame<TPixel> source) => this.source = source;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    this.source.GetPixelRowSpan(y).Reverse();
+                }
+            }
         }
     }
 }

src/ImageSharp/Processing/Processors/Transforms/ProjectiveTransformProcessor{TPixel}.cs

@@ -3,9 +3,9 @@
 
 using System;
 using System.Numerics;
-
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
 namespace SixLabors.ImageSharp.Processing.Processors.Transforms
@@ -17,9 +17,9 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
     internal class ProjectiveTransformProcessor<TPixel> : TransformProcessor<TPixel>
         where TPixel : struct, IPixel<TPixel>
     {
-        private Size targetSize;
+        private readonly Size targetSize;
         private readonly IResampler resampler;
-        private Matrix4x4 transformMatrix;
+        private readonly Matrix4x4 transformMatrix;
 
         /// <summary>
         /// Initializes a new instance of the <see cref="ProjectiveTransformProcessor{TPixel}"/> class.
@@ -50,7 +50,6 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
             }
 
             int width = this.targetSize.Width;
-            Rectangle sourceBounds = this.SourceRectangle;
             var targetBounds = new Rectangle(Point.Empty, this.targetSize);
             Configuration configuration = this.Configuration;
 
@@ -59,73 +58,126 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
 
             if (this.resampler is NearestNeighborResampler)
             {
-                ParallelHelper.IterateRows(
-                    targetBounds,
+                Rectangle sourceBounds = this.SourceRectangle;
+
+                var nnOperation = new NearestNeighborRowIntervalOperation(sourceBounds, ref matrix, width, source, destination);
+                ParallelRowIterator.IterateRows(
                     configuration,
-                    rows =>
-                        {
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                Span<TPixel> destRow = destination.GetPixelRowSpan(y);
-
-                                for (int x = 0; x < width; x++)
-                                {
-                                    Vector2 point = TransformUtils.ProjectiveTransform2D(x, y, matrix);
-                                    int px = (int)MathF.Round(point.X);
-                                    int py = (int)MathF.Round(point.Y);
-
-                                    if (sourceBounds.Contains(px, py))
-                                    {
-                                        destRow[x] = source[px, py];
-                                    }
-                                }
-                            }
-                        });
+                    targetBounds,
+                    in nnOperation);
 
                 return;
             }
 
-            var kernel = new TransformKernelMap(configuration, source.Size(), destination.Size(), this.resampler);
+            using var kernelMap = new TransformKernelMap(configuration, source.Size(), destination.Size(), this.resampler);
 
-            try
+            var operation = new RowIntervalOperation(configuration, kernelMap, ref matrix, width, source, destination);
+            ParallelRowIterator.IterateRows<RowIntervalOperation, Vector4>(
+                configuration,
+                targetBounds,
+                in operation);
+        }
+
+        private readonly struct NearestNeighborRowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Rectangle bounds;
+            private readonly Matrix4x4 matrix;
+            private readonly int maxX;
+            private readonly ImageFrame<TPixel> source;
+            private readonly ImageFrame<TPixel> destination;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public NearestNeighborRowIntervalOperation(
+                Rectangle bounds,
+                ref Matrix4x4 matrix,
+                int maxX,
+                ImageFrame<TPixel> source,
+                ImageFrame<TPixel> destination)
             {
-                ParallelHelper.IterateRowsWithTempBuffer<Vector4>(
-                    targetBounds,
-                    configuration,
-                    (rows, vectorBuffer) =>
+                this.bounds = bounds;
+                this.matrix = matrix;
+                this.maxX = maxX;
+                this.source = source;
+                this.destination = destination;
+            }
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> destRow = this.destination.GetPixelRowSpan(y);
+
+                    for (int x = 0; x < this.maxX; x++)
+                    {
+                        Vector2 point = TransformUtils.ProjectiveTransform2D(x, y, this.matrix);
+                        int px = (int)MathF.Round(point.X);
+                        int py = (int)MathF.Round(point.Y);
+
+                        if (this.bounds.Contains(px, py))
                         {
-                            Span<Vector4> vectorSpan = vectorBuffer.Span;
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                Span<TPixel> targetRowSpan = destination.GetPixelRowSpan(y);
-                                PixelOperations<TPixel>.Instance.ToVector4(configuration, targetRowSpan, vectorSpan);
-                                ref float ySpanRef = ref kernel.GetYStartReference(y);
-                                ref float xSpanRef = ref kernel.GetXStartReference(y);
-
-                                for (int x = 0; x < width; x++)
-                                {
-                                    // Use the single precision position to calculate correct bounding pixels
-                                    // otherwise we get rogue pixels outside of the bounds.
-                                    Vector2 point = TransformUtils.ProjectiveTransform2D(x, y, matrix);
-                                    kernel.Convolve(
-                                        point,
-                                        x,
-                                        ref ySpanRef,
-                                        ref xSpanRef,
-                                        source.PixelBuffer,
-                                        vectorSpan);
-                                }
-
-                                PixelOperations<TPixel>.Instance.FromVector4Destructive(
-                                    configuration,
-                                    vectorSpan,
-                                    targetRowSpan);
-                            }
-                        });
+                            destRow[x] = this.source[px, py];
+                        }
+                    }
+                }
+            }
+        }
+
+        private readonly struct RowIntervalOperation : IRowIntervalOperation<Vector4>
+        {
+            private readonly Configuration configuration;
+            private readonly TransformKernelMap kernelMap;
+            private readonly Matrix4x4 matrix;
+            private readonly int maxX;
+            private readonly ImageFrame<TPixel> source;
+            private readonly ImageFrame<TPixel> destination;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Configuration configuration,
+                TransformKernelMap kernelMap,
+                ref Matrix4x4 matrix,
+                int maxX,
+                ImageFrame<TPixel> source,
+                ImageFrame<TPixel> destination)
+            {
+                this.configuration = configuration;
+                this.kernelMap = kernelMap;
+                this.matrix = matrix;
+                this.maxX = maxX;
+                this.source = source;
+                this.destination = destination;
             }
-            finally
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows, Span<Vector4> span)
             {
-                kernel.Dispose();
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> targetRowSpan = this.destination.GetPixelRowSpan(y);
+                    PixelOperations<TPixel>.Instance.ToVector4(this.configuration, targetRowSpan, span);
+                    ref float ySpanRef = ref this.kernelMap.GetYStartReference(y);
+                    ref float xSpanRef = ref this.kernelMap.GetXStartReference(y);
+
+                    for (int x = 0; x < this.maxX; x++)
+                    {
+                        // Use the single precision position to calculate correct bounding pixels
+                        // otherwise we get rogue pixels outside of the bounds.
+                        Vector2 point = TransformUtils.ProjectiveTransform2D(x, y, this.matrix);
+                        this.kernelMap.Convolve(
+                            point,
+                            x,
+                            ref ySpanRef,
+                            ref xSpanRef,
+                            this.source.PixelBuffer,
+                            span);
+                    }
+
+                    PixelOperations<TPixel>.Instance.FromVector4Destructive(
+                        this.configuration,
+                        span,
+                        targetRowSpan);
+                }
             }
         }
     }

src/ImageSharp/Processing/Processors/Transforms/Resize/ResizeKernelMap.cs

@@ -1,4 +1,4 @@
-// Copyright (c) Six Labors and contributors.
+// Copyright (c) Six Labors and contributors.
 // Licensed under the Apache License, Version 2.0.
 
 using System;
@@ -248,4 +248,4 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
             return new ResizeKernel(left, rowPtr, length);
         }
     }
-}
+}

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

@@ -2,9 +2,8 @@
 // Licensed under the Apache License, Version 2.0.
 
 using System;
-
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -24,7 +23,7 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
         private readonly int targetWidth;
         private readonly int targetHeight;
         private readonly IResampler resampler;
-        private Rectangle targetRectangle;
+        private readonly Rectangle targetRectangle;
         private readonly bool compand;
 
         // The following fields are not immutable but are optionally created on demand.
@@ -77,7 +76,9 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
             Configuration configuration = this.Configuration;
 
             // Handle resize dimensions identical to the original
-            if (source.Width == destination.Width && source.Height == destination.Height && sourceRectangle == this.targetRectangle)
+            if (source.Width == destination.Width
+                && source.Height == destination.Height
+                && sourceRectangle == this.targetRectangle)
             {
                 // The cloned will be blank here copy all the pixel data over
                 source.GetPixelSpan().CopyTo(destination.GetPixelSpan());
@@ -86,14 +87,7 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
 
             int width = this.targetWidth;
             int height = this.targetHeight;
-            int sourceX = sourceRectangle.X;
-            int sourceY = sourceRectangle.Y;
-            int startY = this.targetRectangle.Y;
-            int startX = this.targetRectangle.X;
-
-            var targetWorkingRect = Rectangle.Intersect(
-                this.targetRectangle,
-                new Rectangle(0, 0, width, height));
+            var interest = Rectangle.Intersect(this.targetRectangle, new Rectangle(0, 0, width, height));
 
             if (this.resampler is NearestNeighborResampler)
             {
@@ -101,24 +95,11 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
                 float widthFactor = sourceRectangle.Width / (float)this.targetRectangle.Width;
                 float heightFactor = sourceRectangle.Height / (float)this.targetRectangle.Height;
 
-                ParallelHelper.IterateRows(
-                    targetWorkingRect,
+                var operation = new RowIntervalOperation(sourceRectangle, this.targetRectangle, widthFactor, heightFactor, source, destination);
+                ParallelRowIterator.IterateRows(
                     configuration,
-                    rows =>
-                    {
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            // Y coordinates of source points
-                            Span<TPixel> sourceRow = source.GetPixelRowSpan((int)(((y - startY) * heightFactor) + sourceY));
-                            Span<TPixel> targetRow = destination.GetPixelRowSpan(y);
-
-                            for (int x = targetWorkingRect.Left; x < targetWorkingRect.Right; x++)
-                            {
-                                // X coordinates of source points
-                                targetRow[x] = sourceRow[(int)(((x - startX) * widthFactor) + sourceX)];
-                            }
-                        }
-                    });
+                    interest,
+                    in operation);
 
                 return;
             }
@@ -137,12 +118,12 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
                 this.horizontalKernelMap,
                 this.verticalKernelMap,
                 width,
-                targetWorkingRect,
+                interest,
                 this.targetRectangle.Location))
             {
                 worker.Initialize();
 
-                var workingInterval = new RowInterval(targetWorkingRect.Top, targetWorkingRect.Bottom);
+                var workingInterval = new RowInterval(interest.Top, interest.Bottom);
                 worker.FillDestinationPixels(workingInterval, destination.PixelBuffer);
             }
         }
@@ -166,5 +147,56 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
             this.isDisposed = true;
             base.Dispose(disposing);
         }
+
+        private readonly struct RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Rectangle sourceBounds;
+            private readonly Rectangle destinationBounds;
+            private readonly float widthFactor;
+            private readonly float heightFactor;
+            private readonly ImageFrame<TPixel> source;
+            private readonly ImageFrame<TPixel> destination;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public RowIntervalOperation(
+                Rectangle sourceBounds,
+                Rectangle destinationBounds,
+                float widthFactor,
+                float heightFactor,
+                ImageFrame<TPixel> source,
+                ImageFrame<TPixel> destination)
+            {
+                this.sourceBounds = sourceBounds;
+                this.destinationBounds = destinationBounds;
+                this.widthFactor = widthFactor;
+                this.heightFactor = heightFactor;
+                this.source = source;
+                this.destination = destination;
+            }
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                int sourceX = this.sourceBounds.X;
+                int sourceY = this.sourceBounds.Y;
+                int destX = this.destinationBounds.X;
+                int destY = this.destinationBounds.Y;
+                int destLeft = this.destinationBounds.Left;
+                int destRight = this.destinationBounds.Right;
+
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    // Y coordinates of source points
+                    Span<TPixel> sourceRow = this.source.GetPixelRowSpan((int)(((y - destY) * this.heightFactor) + sourceY));
+                    Span<TPixel> targetRow = this.destination.GetPixelRowSpan(y);
+
+                    for (int x = destLeft; x < destRight; x++)
+                    {
+                        // X coordinates of source points
+                        targetRow[x] = sourceRow[(int)(((x - destX) * this.widthFactor) + sourceX)];
+                    }
+                }
+            }
+        }
     }
 }

src/ImageSharp/Processing/Processors/Transforms/Resize/ResizeWorker.cs

@@ -19,7 +19,7 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
     /// When sliding the window, the contents of the bottom window band are copied to the new top band.
     /// For more details, and visual explanation, see "ResizeWorker.pptx".
     /// </summary>
-    internal class ResizeWorker<TPixel> : IDisposable
+    internal sealed class ResizeWorker<TPixel> : IDisposable
         where TPixel : struct, IPixel<TPixel>
     {
         private readonly Buffer2D<Vector4> transposedFirstPassBuffer;

src/ImageSharp/Processing/Processors/Transforms/RotateProcessor{TPixel}.cs

@@ -2,9 +2,9 @@
 // Licensed under the Apache License, Version 2.0.
 
 using System;
-
+using System.Runtime.CompilerServices;
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.Metadata.Profiles.Exif;
 using SixLabors.ImageSharp.PixelFormats;
 
@@ -17,6 +17,8 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
     internal class RotateProcessor<TPixel> : AffineTransformProcessor<TPixel>
         where TPixel : struct, IPixel<TPixel>
     {
+        private readonly float degrees;
+
         /// <summary>
         /// Initializes a new instance of the <see cref="RotateProcessor{TPixel}"/> class.
         /// </summary>
@@ -26,11 +28,18 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
         /// <param name="sourceRectangle">The source area to process for the current processor instance.</param>
         public RotateProcessor(Configuration configuration, RotateProcessor definition, Image<TPixel> source, Rectangle sourceRectangle)
             : base(configuration, definition, source, sourceRectangle)
+            => this.degrees = definition.Degrees;
+
+        /// <inheritdoc/>
+        protected override void OnFrameApply(ImageFrame<TPixel> source, ImageFrame<TPixel> destination)
         {
-            this.Degrees = definition.Degrees;
-        }
+            if (this.OptimizedApply(source, destination, this.Configuration))
+            {
+                return;
+            }
 
-        private float Degrees { get; }
+            base.OnFrameApply(source, destination);
+        }
 
         /// <inheritdoc/>
         protected override void AfterImageApply(Image<TPixel> destination)
@@ -41,7 +50,7 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
                 return;
             }
 
-            if (MathF.Abs(WrapDegrees(this.Degrees)) < Constants.Epsilon)
+            if (MathF.Abs(WrapDegrees(this.degrees)) < Constants.Epsilon)
             {
                 // No need to do anything so return.
                 return;
@@ -52,17 +61,6 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
             base.AfterImageApply(destination);
         }
 
-        /// <inheritdoc/>
-        protected override void OnFrameApply(ImageFrame<TPixel> source, ImageFrame<TPixel> destination)
-        {
-            if (this.OptimizedApply(source, destination, this.Configuration))
-            {
-                return;
-            }
-
-            base.OnFrameApply(source, destination);
-        }
-
         /// <summary>
         /// Wraps a given angle in degrees so that it falls withing the 0-360 degree range
         /// </summary>
@@ -95,7 +93,7 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
             Configuration configuration)
         {
             // Wrap the degrees to keep within 0-360 so we can apply optimizations when possible.
-            float degrees = WrapDegrees(this.Degrees);
+            float degrees = WrapDegrees(this.degrees);
 
             if (MathF.Abs(degrees) < Constants.Epsilon)
             {
@@ -133,25 +131,11 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
         /// <param name="configuration">The configuration.</param>
         private void Rotate180(ImageFrame<TPixel> source, ImageFrame<TPixel> destination, Configuration configuration)
         {
-            int width = source.Width;
-            int height = source.Height;
-
-            ParallelHelper.IterateRows(
-                source.Bounds(),
+            var operation = new Rotate180RowIntervalOperation(source.Width, source.Height, source, destination);
+            ParallelRowIterator.IterateRows(
                 configuration,
-                rows =>
-                    {
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            Span<TPixel> sourceRow = source.GetPixelRowSpan(y);
-                            Span<TPixel> targetRow = destination.GetPixelRowSpan(height - y - 1);
-
-                            for (int x = 0; x < width; x++)
-                            {
-                                targetRow[width - x - 1] = sourceRow[x];
-                            }
-                        }
-                    });
+                source.Bounds(),
+                in operation);
         }
 
         /// <summary>
@@ -162,31 +146,11 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
         /// <param name="configuration">The configuration.</param>
         private void Rotate270(ImageFrame<TPixel> source, ImageFrame<TPixel> destination, Configuration configuration)
         {
-            int width = source.Width;
-            int height = source.Height;
-            Rectangle destinationBounds = destination.Bounds();
-
-            ParallelHelper.IterateRows(
-                source.Bounds(),
+            var operation = new Rotate270RowIntervalOperation(destination.Bounds(), source.Width, source.Height, source, destination);
+            ParallelRowIterator.IterateRows(
                 configuration,
-                rows =>
-                    {
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            Span<TPixel> sourceRow = source.GetPixelRowSpan(y);
-                            for (int x = 0; x < width; x++)
-                            {
-                                int newX = height - y - 1;
-                                newX = height - newX - 1;
-                                int newY = width - x - 1;
-
-                                if (destinationBounds.Contains(newX, newY))
-                                {
-                                    destination[newX, newY] = sourceRow[x];
-                                }
-                            }
-                        }
-                    });
+                source.Bounds(),
+                in operation);
         }
 
         /// <summary>
@@ -197,28 +161,132 @@ namespace SixLabors.ImageSharp.Processing.Processors.Transforms
         /// <param name="configuration">The configuration.</param>
         private void Rotate90(ImageFrame<TPixel> source, ImageFrame<TPixel> destination, Configuration configuration)
         {
-            int width = source.Width;
-            int height = source.Height;
-            Rectangle destinationBounds = destination.Bounds();
-
-            ParallelHelper.IterateRows(
-                source.Bounds(),
+            var operation = new Rotate90RowIntervalOperation(destination.Bounds(), source.Width, source.Height, source, destination);
+            ParallelRowIterator.IterateRows(
                 configuration,
-                rows =>
+                source.Bounds(),
+                in operation);
+        }
+
+        private readonly struct Rotate180RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly int width;
+            private readonly int height;
+            private readonly ImageFrame<TPixel> source;
+            private readonly ImageFrame<TPixel> destination;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public Rotate180RowIntervalOperation(
+                int width,
+                int height,
+                ImageFrame<TPixel> source,
+                ImageFrame<TPixel> destination)
+            {
+                this.width = width;
+                this.height = height;
+                this.source = source;
+                this.destination = destination;
+            }
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> sourceRow = this.source.GetPixelRowSpan(y);
+                    Span<TPixel> targetRow = this.destination.GetPixelRowSpan(this.height - y - 1);
+
+                    for (int x = 0; x < this.width; x++)
+                    {
+                        targetRow[this.width - x - 1] = sourceRow[x];
+                    }
+                }
+            }
+        }
+
+        private readonly struct Rotate270RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Rectangle bounds;
+            private readonly int width;
+            private readonly int height;
+            private readonly ImageFrame<TPixel> source;
+            private readonly ImageFrame<TPixel> destination;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public Rotate270RowIntervalOperation(
+                Rectangle bounds,
+                int width,
+                int height,
+                ImageFrame<TPixel> source,
+                ImageFrame<TPixel> destination)
+            {
+                this.bounds = bounds;
+                this.width = width;
+                this.height = height;
+                this.source = source;
+                this.destination = destination;
+            }
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> sourceRow = this.source.GetPixelRowSpan(y);
+                    for (int x = 0; x < this.width; x++)
+                    {
+                        int newX = this.height - y - 1;
+                        newX = this.height - newX - 1;
+                        int newY = this.width - x - 1;
+
+                        if (this.bounds.Contains(newX, newY))
+                        {
+                            this.destination[newX, newY] = sourceRow[x];
+                        }
+                    }
+                }
+            }
+        }
+
+        private readonly struct Rotate90RowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Rectangle bounds;
+            private readonly int width;
+            private readonly int height;
+            private readonly ImageFrame<TPixel> source;
+            private readonly ImageFrame<TPixel> destination;
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public Rotate90RowIntervalOperation(
+                Rectangle bounds,
+                int width,
+                int height,
+                ImageFrame<TPixel> source,
+                ImageFrame<TPixel> destination)
+            {
+                this.bounds = bounds;
+                this.width = width;
+                this.height = height;
+                this.source = source;
+                this.destination = destination;
+            }
+
+            [MethodImpl(InliningOptions.ShortMethod)]
+            public void Invoke(in RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    Span<TPixel> sourceRow = this.source.GetPixelRowSpan(y);
+                    int newX = this.height - y - 1;
+                    for (int x = 0; x < this.width; x++)
                     {
-                        for (int y = rows.Min; y < rows.Max; y++)
+                        if (this.bounds.Contains(newX, x))
                         {
-                            Span<TPixel> sourceRow = source.GetPixelRowSpan(y);
-                            int newX = height - y - 1;
-                            for (int x = 0; x < width; x++)
-                            {
-                                if (destinationBounds.Contains(newX, x))
-                                {
-                                    destination[newX, x] = sourceRow[x];
-                                }
-                            }
+                            this.destination[newX, x] = sourceRow[x];
                         }
-                    });
+                    }
+                }
+            }
         }
     }
 }

tests/Directory.Build.targets

@@ -26,6 +26,7 @@
   <ItemGroup>
     <!--Test Dependencies-->
     <PackageReference Update="BenchmarkDotNet" Version="0.12.0" />
+    <PackageReference Update="BenchmarkDotNet.Diagnostics.Windows" Version="0.12.0" Condition="'$(OS)' == 'Windows_NT'" />    
     <PackageReference Update="Colourful" Version="2.0.3" />
     <PackageReference Update="coverlet.collector" Version="1.2.0" PrivateAssets="All"/>
     <PackageReference Update="Magick.NET-Q16-AnyCPU" Version="7.14.4" />

tests/ImageSharp.Benchmarks/Config.cs

@@ -1,6 +1,10 @@
 // Copyright (c) Six Labors and contributors.
 // Licensed under the Apache License, Version 2.0.
 
+#if Windows_NT
+using System.Security.Principal;
+using BenchmarkDotNet.Diagnostics.Windows;
+#endif
 using BenchmarkDotNet.Configs;
 using BenchmarkDotNet.Diagnosers;
 using BenchmarkDotNet.Environments;
@@ -13,6 +17,14 @@ namespace SixLabors.ImageSharp.Benchmarks
         public Config()
         {
             this.Add(MemoryDiagnoser.Default);
+
+#if Windows_NT
+            if (this.IsElevated)
+            {
+                this.Add(new NativeMemoryProfiler());
+            }
+#endif
+
         }
 
         public class ShortClr : Config
@@ -25,5 +37,15 @@ namespace SixLabors.ImageSharp.Benchmarks
                     Job.Default.With(CoreRuntime.Core21).WithLaunchCount(1).WithWarmupCount(3).WithIterationCount(3));
             }
         }
+
+#if Windows_NT
+        private bool IsElevated
+        {
+            get
+            {
+                return new WindowsPrincipal(WindowsIdentity.GetCurrent()).IsInRole(WindowsBuiltInRole.Administrator);
+            }
+        }
+#endif
     }
 }

tests/ImageSharp.Benchmarks/ImageSharp.Benchmarks.csproj

@@ -19,6 +19,7 @@
   <ItemGroup>
     <PackageReference Include="Magick.NET-Q16-AnyCPU" />
     <PackageReference Include="BenchmarkDotNet" />
+    <PackageReference Include="BenchmarkDotNet.Diagnostics.Windows" Condition="'$(OS)' == 'Windows_NT'" />
     <PackageReference Include="Colourful" />
     <PackageReference Include="Pfim" />
     <PackageReference Include="System.Drawing.Common" />

tests/ImageSharp.Benchmarks/Samplers/Crop.cs

@@ -13,6 +13,7 @@ using SDSize = System.Drawing.Size;
 
 namespace SixLabors.ImageSharp.Benchmarks
 {
+    [Config(typeof(Config.ShortClr))]
     public class Crop : BenchmarkBase
     {
         [Benchmark(Baseline = true, Description = "System.Drawing Crop")]

tests/ImageSharp.Tests/Helpers/ParallelExecutionSettingsTests.cs

@@ -2,7 +2,7 @@
 // Licensed under the Apache License, Version 2.0.
 
 using System;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Advanced;
 using Xunit;
 
 namespace SixLabors.ImageSharp.Tests.Helpers

tests/ImageSharp.Tests/Helpers/ParallelHelperTests.cs → tests/ImageSharp.Tests/Helpers/ParallelRowIteratorTests.cs

@@ -2,25 +2,25 @@
 // Licensed under the Apache License, Version 2.0.
 
 using System;
-using System.Collections.Concurrent;
 using System.Linq;
 using System.Numerics;
 using System.Threading;
 
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
+using SixLabors.ImageSharp.Advanced;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
-
 using Xunit;
 using Xunit.Abstractions;
 
 namespace SixLabors.ImageSharp.Tests.Helpers
 {
-    public class ParallelHelperTests
+    public class ParallelRowIteratorTests
     {
+        public delegate void RowIntervalAction<T>(RowInterval rows, Span<T> span);
+
         private readonly ITestOutputHelper output;
 
-        public ParallelHelperTests(ITestOutputHelper output)
+        public ParallelRowIteratorTests(ITestOutputHelper output)
         {
             this.output = output;
         }
@@ -30,20 +30,20 @@ namespace SixLabors.ImageSharp.Tests.Helpers
         /// </summary>
         public static TheoryData<int, int, int, int, int, int> IterateRows_OverMinimumPixelsLimit_Data =
             new TheoryData<int, int, int, int, int, int>
-                {
-                    { 1, 0, 100, -1, 100, 1 },
-                    { 2, 0, 9, 5, 4, 2 },
-                    { 4, 0, 19, 5, 4, 4 },
-                    { 2, 10, 19, 5, 4, 2 },
-                    { 4, 0, 200, 50, 50, 4 },
-                    { 4, 123, 323, 50, 50, 4 },
-                    { 4, 0, 1201, 301, 298, 4 },
-                    { 8, 10, 236, 29, 23, 8 },
-                    { 16, 0, 209, 14, 13, 15 },
-                    { 24, 0, 209, 9, 2, 24 },
-                    { 32, 0, 209, 7, 6, 30 },
-                    { 64, 0, 209, 4, 1, 53 },
-                };
+            {
+                { 1, 0, 100, -1, 100, 1 },
+                { 2, 0, 9, 5, 4, 2 },
+                { 4, 0, 19, 5, 4, 4 },
+                { 2, 10, 19, 5, 4, 2 },
+                { 4, 0, 200, 50, 50, 4 },
+                { 4, 123, 323, 50, 50, 4 },
+                { 4, 0, 1201, 301, 298, 4 },
+                { 8, 10, 236, 29, 23, 8 },
+                { 16, 0, 209, 14, 13, 15 },
+                { 24, 0, 209, 9, 2, 24 },
+                { 32, 0, 209, 7, 6, 30 },
+                { 64, 0, 209, 4, 1, 53 },
+            };
 
         [Theory]
         [MemberData(nameof(IterateRows_OverMinimumPixelsLimit_Data))]
@@ -64,20 +64,24 @@ namespace SixLabors.ImageSharp.Tests.Helpers
 
             int actualNumberOfSteps = 0;
 
-            ParallelHelper.IterateRows(
-                rectangle,
-                parallelSettings,
-                rows =>
-                    {
-                        Assert.True(rows.Min >= minY);
-                        Assert.True(rows.Max <= maxY);
+            void RowAction(RowInterval rows)
+            {
+                Assert.True(rows.Min >= minY);
+                Assert.True(rows.Max <= maxY);
+
+                int step = rows.Max - rows.Min;
+                int expected = rows.Max < maxY ? expectedStepLength : expectedLastStepLength;
+
+                Interlocked.Increment(ref actualNumberOfSteps);
+                Assert.Equal(expected, step);
+            }
 
-                        int step = rows.Max - rows.Min;
-                        int expected = rows.Max < maxY ? expectedStepLength : expectedLastStepLength;
+            var operation = new TestRowIntervalOperation(RowAction);
 
-                        Interlocked.Increment(ref actualNumberOfSteps);
-                        Assert.Equal(expected, step);
-                    });
+            ParallelRowIterator.IterateRows(
+                rectangle,
+                in parallelSettings,
+                in operation);
 
             Assert.Equal(expectedNumberOfSteps, actualNumberOfSteps);
         }
@@ -102,16 +106,20 @@ namespace SixLabors.ImageSharp.Tests.Helpers
             int[] expectedData = Enumerable.Repeat(0, minY).Concat(Enumerable.Range(minY, maxY - minY)).ToArray();
             var actualData = new int[maxY];
 
-            ParallelHelper.IterateRows(
+            void RowAction(RowInterval rows)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    actualData[y] = y;
+                }
+            }
+
+            var operation = new TestRowIntervalOperation(RowAction);
+
+            ParallelRowIterator.IterateRows(
                 rectangle,
-                parallelSettings,
-                rows =>
-                    {
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            actualData[y] = y;
-                        }
-                    });
+                in parallelSettings,
+                in operation);
 
             Assert.Equal(expectedData, actualData);
         }
@@ -133,30 +141,28 @@ namespace SixLabors.ImageSharp.Tests.Helpers
 
             var rectangle = new Rectangle(0, minY, 10, maxY - minY);
 
-            var bufferHashes = new ConcurrentBag<int>();
-
             int actualNumberOfSteps = 0;
-            ParallelHelper.IterateRowsWithTempBuffer(
-                rectangle,
-                parallelSettings,
-                (RowInterval rows, Memory<Vector4> buffer) =>
-                    {
-                        Assert.True(rows.Min >= minY);
-                        Assert.True(rows.Max <= maxY);
 
-                        bufferHashes.Add(buffer.GetHashCode());
+            void RowAction(RowInterval rows, Span<Vector4> buffer)
+            {
+                Assert.True(rows.Min >= minY);
+                Assert.True(rows.Max <= maxY);
 
-                        int step = rows.Max - rows.Min;
-                        int expected = rows.Max < maxY ? expectedStepLength : expectedLastStepLength;
+                int step = rows.Max - rows.Min;
+                int expected = rows.Max < maxY ? expectedStepLength : expectedLastStepLength;
 
-                        Interlocked.Increment(ref actualNumberOfSteps);
-                        Assert.Equal(expected, step);
-                    });
+                Interlocked.Increment(ref actualNumberOfSteps);
+                Assert.Equal(expected, step);
+            }
 
-            Assert.Equal(expectedNumberOfSteps, actualNumberOfSteps);
+            var operation = new TestRowIntervalOperation<Vector4>(RowAction);
+
+            ParallelRowIterator.IterateRows<TestRowIntervalOperation<Vector4>, Vector4>(
+                rectangle,
+                in parallelSettings,
+                in operation);
 
-            int numberOfDifferentBuffers = bufferHashes.Distinct().Count();
-            Assert.Equal(actualNumberOfSteps, numberOfDifferentBuffers);
+            Assert.Equal(expectedNumberOfSteps, actualNumberOfSteps);
         }
 
         [Theory]
@@ -179,31 +185,35 @@ namespace SixLabors.ImageSharp.Tests.Helpers
             int[] expectedData = Enumerable.Repeat(0, minY).Concat(Enumerable.Range(minY, maxY - minY)).ToArray();
             var actualData = new int[maxY];
 
-            ParallelHelper.IterateRowsWithTempBuffer(
+            void RowAction(RowInterval rows, Span<Vector4> buffer)
+            {
+                for (int y = rows.Min; y < rows.Max; y++)
+                {
+                    actualData[y] = y;
+                }
+            }
+
+            var operation = new TestRowIntervalOperation<Vector4>(RowAction);
+
+            ParallelRowIterator.IterateRows<TestRowIntervalOperation<Vector4>, Vector4>(
                 rectangle,
-                parallelSettings,
-                (RowInterval rows, Memory<Vector4> buffer) =>
-                    {
-                        for (int y = rows.Min; y < rows.Max; y++)
-                        {
-                            actualData[y] = y;
-                        }
-                    });
+                in parallelSettings,
+                in operation);
 
             Assert.Equal(expectedData, actualData);
         }
 
         public static TheoryData<int, int, int, int, int, int, int> IterateRows_WithEffectiveMinimumPixelsLimit_Data =
             new TheoryData<int, int, int, int, int, int, int>
-                {
-                    { 2, 200, 50, 2, 1, -1, 2 },
-                    { 2, 200, 200, 1, 1, -1, 1 },
-                    { 4, 200, 100, 4, 2, 2, 2 },
-                    { 4, 300, 100, 8, 3, 3, 2 },
-                    { 2, 5000, 1, 4500, 1, -1, 4500 },
-                    { 2, 5000, 1, 5000, 1, -1, 5000 },
-                    { 2, 5000, 1, 5001, 2, 2501, 2500 },
-                };
+            {
+                { 2, 200, 50, 2, 1, -1, 2 },
+                { 2, 200, 200, 1, 1, -1, 1 },
+                { 4, 200, 100, 4, 2, 2, 2 },
+                { 4, 300, 100, 8, 3, 3, 2 },
+                { 2, 5000, 1, 4500, 1, -1, 4500 },
+                { 2, 5000, 1, 5000, 1, -1, 5000 },
+                { 2, 5000, 1, 5001, 2, 2501, 2500 },
+            };
 
         [Theory]
         [MemberData(nameof(IterateRows_WithEffectiveMinimumPixelsLimit_Data))]
@@ -225,20 +235,24 @@ namespace SixLabors.ImageSharp.Tests.Helpers
 
             int actualNumberOfSteps = 0;
 
-            ParallelHelper.IterateRows(
-                rectangle,
-                parallelSettings,
-                rows =>
-                    {
-                        Assert.True(rows.Min >= 0);
-                        Assert.True(rows.Max <= height);
+            void RowAction(RowInterval rows)
+            {
+                Assert.True(rows.Min >= 0);
+                Assert.True(rows.Max <= height);
 
-                        int step = rows.Max - rows.Min;
-                        int expected = rows.Max < height ? expectedStepLength : expectedLastStepLength;
+                int step = rows.Max - rows.Min;
+                int expected = rows.Max < height ? expectedStepLength : expectedLastStepLength;
 
-                        Interlocked.Increment(ref actualNumberOfSteps);
-                        Assert.Equal(expected, step);
-                    });
+                Interlocked.Increment(ref actualNumberOfSteps);
+                Assert.Equal(expected, step);
+            }
+
+            var operation = new TestRowIntervalOperation(RowAction);
+
+            ParallelRowIterator.IterateRows(
+                rectangle,
+                in parallelSettings,
+                in operation);
 
             Assert.Equal(expectedNumberOfSteps, actualNumberOfSteps);
         }
@@ -262,33 +276,38 @@ namespace SixLabors.ImageSharp.Tests.Helpers
             var rectangle = new Rectangle(0, 0, width, height);
 
             int actualNumberOfSteps = 0;
-            ParallelHelper.IterateRowsWithTempBuffer(
-                rectangle,
-                parallelSettings,
-                (RowInterval rows, Memory<Vector4> buffer) =>
-                    {
-                        Assert.True(rows.Min >= 0);
-                        Assert.True(rows.Max <= height);
 
-                        int step = rows.Max - rows.Min;
-                        int expected = rows.Max < height ? expectedStepLength : expectedLastStepLength;
+            void RowAction(RowInterval rows, Span<Vector4> buffer)
+            {
+                Assert.True(rows.Min >= 0);
+                Assert.True(rows.Max <= height);
+
+                int step = rows.Max - rows.Min;
+                int expected = rows.Max < height ? expectedStepLength : expectedLastStepLength;
+
+                Interlocked.Increment(ref actualNumberOfSteps);
+                Assert.Equal(expected, step);
+            }
 
-                        Interlocked.Increment(ref actualNumberOfSteps);
-                        Assert.Equal(expected, step);
-                    });
+            var operation = new TestRowIntervalOperation<Vector4>(RowAction);
+
+            ParallelRowIterator.IterateRows<TestRowIntervalOperation<Vector4>, Vector4>(
+                rectangle,
+                in parallelSettings,
+                in operation);
 
             Assert.Equal(expectedNumberOfSteps, actualNumberOfSteps);
         }
 
         public static readonly TheoryData<int, int, int, int, int, int, int> IterateRectangularBuffer_Data =
             new TheoryData<int, int, int, int, int, int, int>
-                {
-                    { 8, 582, 453, 10, 10, 291, 226 }, // boundary data from DetectEdgesTest.DetectEdges_InBox
-                    { 2, 582, 453, 10, 10, 291, 226 },
-                    { 16, 582, 453, 10, 10, 291, 226 },
-                    { 16, 582, 453, 10, 10, 1, 226 },
-                    { 16, 1, 453, 0, 10, 1, 226 },
-                };
+            {
+                { 8, 582, 453, 10, 10, 291, 226 }, // boundary data from DetectEdgesTest.DetectEdges_InBox
+                { 2, 582, 453, 10, 10, 291, 226 },
+                { 16, 582, 453, 10, 10, 291, 226 },
+                { 16, 582, 453, 10, 10, 1, 226 },
+                { 16, 1, 453, 0, 10, 1, 226 },
+            };
 
         [Theory]
         [MemberData(nameof(IterateRectangularBuffer_Data))]
@@ -325,17 +344,21 @@ namespace SixLabors.ImageSharp.Tests.Helpers
                 // Fill actual data using IterateRows:
                 var settings = new ParallelExecutionSettings(maxDegreeOfParallelism, memoryAllocator);
 
-                ParallelHelper.IterateRows(
+                void RowAction(RowInterval rows)
+                {
+                    this.output.WriteLine(rows.ToString());
+                    for (int y = rows.Min; y < rows.Max; y++)
+                    {
+                        FillRow(y, actual);
+                    }
+                }
+
+                var operation = new TestRowIntervalOperation(RowAction);
+
+                ParallelRowIterator.IterateRows(
                     rect,
                     settings,
-                    rows =>
-                        {
-                            this.output.WriteLine(rows.ToString());
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                FillRow(y, actual);
-                            }
-                        });
+                    in operation);
 
                 // Assert:
                 TestImageExtensions.CompareBuffers(expected.GetSpan(), actual.GetSpan());
@@ -353,8 +376,14 @@ namespace SixLabors.ImageSharp.Tests.Helpers
 
             var rect = new Rectangle(0, 0, width, height);
 
+            void RowAction(RowInterval rows)
+            {
+            }
+
+            var operation = new TestRowIntervalOperation(RowAction);
+
             ArgumentOutOfRangeException ex = Assert.Throws<ArgumentOutOfRangeException>(
-                () => ParallelHelper.IterateRows(rect, parallelSettings, rows => { }));
+                () => ParallelRowIterator.IterateRows(rect, in parallelSettings, in operation));
 
             Assert.Contains(width <= 0 ? "Width" : "Height", ex.Message);
         }
@@ -370,10 +399,38 @@ namespace SixLabors.ImageSharp.Tests.Helpers
 
             var rect = new Rectangle(0, 0, width, height);
 
+            void RowAction(RowInterval rows, Span<Rgba32> memory)
+            {
+            }
+
+            var operation = new TestRowIntervalOperation<Rgba32>(RowAction);
+
             ArgumentOutOfRangeException ex = Assert.Throws<ArgumentOutOfRangeException>(
-                () => ParallelHelper.IterateRowsWithTempBuffer<Rgba32>(rect, parallelSettings, (rows, memory) => { }));
+                () => ParallelRowIterator.IterateRows<TestRowIntervalOperation<Rgba32>, Rgba32>(rect, in parallelSettings, in operation));
 
             Assert.Contains(width <= 0 ? "Width" : "Height", ex.Message);
         }
+
+        private readonly struct TestRowIntervalOperation : IRowIntervalOperation
+        {
+            private readonly Action<RowInterval> action;
+
+            public TestRowIntervalOperation(Action<RowInterval> action)
+                => this.action = action;
+
+            public void Invoke(in RowInterval rows) => this.action(rows);
+        }
+
+        private readonly struct TestRowIntervalOperation<TBuffer> : IRowIntervalOperation<TBuffer>
+            where TBuffer : unmanaged
+        {
+            private readonly RowIntervalAction<TBuffer> action;
+
+            public TestRowIntervalOperation(RowIntervalAction<TBuffer> action)
+                => this.action = action;
+
+            public void Invoke(in RowInterval rows, Span<TBuffer> span)
+                => this.action(rows, span);
+        }
     }
 }

tests/ImageSharp.Tests/TestUtilities/TestImageExtensions.cs

@@ -7,7 +7,6 @@ using System.IO;
 using System.Numerics;
 
 using SixLabors.ImageSharp.Advanced;
-using SixLabors.ImageSharp.Advanced.ParallelUtils;
 using SixLabors.ImageSharp.Formats;
 using SixLabors.ImageSharp.Memory;
 using SixLabors.ImageSharp.PixelFormats;
@@ -703,25 +702,43 @@ namespace SixLabors.ImageSharp.Tests
             {
                 Rectangle sourceRectangle = this.SourceRectangle;
                 Configuration configuration = this.Configuration;
-                ParallelHelper.IterateRowsWithTempBuffer<Vector4>(
-                    sourceRectangle,
+
+                var operation = new RowOperation(configuration, sourceRectangle, source);
+
+                ParallelRowIterator.IterateRows<RowOperation, Vector4>(
                     configuration,
-                    (rows, temp) =>
+                    sourceRectangle,
+                    in operation);
+            }
+
+            private readonly struct RowOperation : IRowIntervalOperation<Vector4>
+            {
+                private readonly Configuration configuration;
+                private readonly Rectangle bounds;
+                private readonly ImageFrame<TPixel> source;
+
+                public RowOperation(Configuration configuration, Rectangle bounds, ImageFrame<TPixel> source)
+                {
+                    this.configuration = configuration;
+                    this.bounds = bounds;
+                    this.source = source;
+                }
+
+                public void Invoke(in RowInterval rows, Span<Vector4> span)
+                {
+                    for (int y = rows.Min; y < rows.Max; y++)
+                    {
+                        Span<TPixel> rowSpan = this.source.GetPixelRowSpan(y).Slice(this.bounds.Left, this.bounds.Width);
+                        PixelOperations<TPixel>.Instance.ToVector4(this.configuration, rowSpan, span, PixelConversionModifiers.Scale);
+                        for (int i = 0; i < span.Length; i++)
                         {
-                            Span<Vector4> tempSpan = temp.Span;
-                            for (int y = rows.Min; y < rows.Max; y++)
-                            {
-                                Span<TPixel> rowSpan = source.GetPixelRowSpan(y).Slice(sourceRectangle.Left, sourceRectangle.Width);
-                                PixelOperations<TPixel>.Instance.ToVector4(configuration, rowSpan, tempSpan, PixelConversionModifiers.Scale);
-                                for (int i = 0; i < tempSpan.Length; i++)
-                                {
-                                    ref Vector4 v = ref tempSpan[i];
-                                    v.W = 1F;
-                                }
-
-                                PixelOperations<TPixel>.Instance.FromVector4Destructive(configuration, tempSpan, rowSpan, PixelConversionModifiers.Scale);
-                            }
-                        });
+                            ref Vector4 v = ref span[i];
+                            v.W = 1F;
+                        }
+
+                        PixelOperations<TPixel>.Instance.FromVector4Destructive(this.configuration, span, rowSpan, PixelConversionModifiers.Scale);
+                    }
+                }
             }
         }
     }