Add AV1 luma transform-size search

src/ImageSharp/Formats/Heif/Av1/Entropy/Av1SymbolContextHelper.cs

@@ -288,6 +288,43 @@ internal static class Av1SymbolContextHelper
         return TransformBlockSkipContexts[(topClass * 5) + leftClass];
     }
 
+    /// <summary>
+    /// Derives the variable-transform partition context from the current node and its adjacent transform edges.
+    /// </summary>
+    /// <param name="aboveTransformWidth">The transform width retained immediately above the current node.</param>
+    /// <param name="leftTransformHeight">The transform height retained immediately left of the current node.</param>
+    /// <param name="blockSize">The containing coding-block size.</param>
+    /// <param name="transformSize">The transform size represented by the current partition node.</param>
+    /// <returns>The variable-transform partition context.</returns>
+    public static int GetTransformPartitionContext(
+        byte aboveTransformWidth,
+        byte leftTransformHeight,
+        Av1BlockSize blockSize,
+        Av1TransformSize transformSize)
+    {
+        if (transformSize <= Av1TransformSize.Size4x4)
+        {
+            return 0;
+        }
+
+        int above = aboveTransformWidth < transformSize.GetWidth() ? 1 : 0;
+        int left = leftTransformHeight < transformSize.GetHeight() ? 1 : 0;
+        int maximumDimension = Math.Max(blockSize.GetWidth(), blockSize.GetHeight());
+        Av1TransformSize maximumSquareTransform = maximumDimension switch
+        {
+            >= 64 => Av1TransformSize.Size64x64,
+            >= 32 => Av1TransformSize.Size32x32,
+            >= 16 => Av1TransformSize.Size16x16,
+            _ => Av1TransformSize.Size8x8
+        };
+
+        int category = (transformSize.GetSquareUpSize() != maximumSquareTransform &&
+            maximumSquareTransform > Av1TransformSize.Size8x8 ? 1 : 0) +
+            ((((int)Av1TransformSize.SquareSizes - 1) - (int)maximumSquareTransform) * 2);
+
+        return (category * 3) + above + left;
+    }
+
     /// <summary>
     /// Reconstructs an end-of-block coefficient position from its token and extra offset.
     /// </summary>
@@ -668,6 +705,37 @@ internal static class Av1SymbolContextHelper
         }
     }
 
+    /// <summary>
+    /// Packs the magnitude class and DC sign retained by neighboring transform blocks.
+    /// </summary>
+    /// <param name="coefficients">The raster-ordered quantized coefficients.</param>
+    /// <param name="transformSize">The transform dimensions.</param>
+    /// <param name="transformType">The transform type selecting scan order.</param>
+    /// <param name="endOfBlock">The one-based final nonzero scan position.</param>
+    /// <returns>The packed coefficient context, or zero for an empty transform.</returns>
+    public static byte GetCoefficientContext(
+        ReadOnlySpan<int> coefficients,
+        Av1TransformSize transformSize,
+        Av1TransformType transformType,
+        ushort endOfBlock)
+    {
+        if (endOfBlock == 0)
+        {
+            return 0;
+        }
+
+        ReadOnlySpan<short> scan = Av1ScanOrderConstants.GetScanOrder(transformSize, transformType).Scan;
+        int culLevel = 0;
+        for (int scanIndex = 0; scanIndex < endOfBlock; scanIndex++)
+        {
+            culLevel += Math.Abs(coefficients[scan[scanIndex]]);
+        }
+
+        culLevel = Math.Min(Av1Constants.CoefficientContextMask, culLevel);
+        SetDcSign(ref culLevel, coefficients[0]);
+        return (byte)culLevel;
+    }
+
     /// <summary>
     /// Converts a one-based end-of-block position to its token and group offset.
     /// </summary>

src/ImageSharp/Formats/Heif/Av1/Entropy/Av1SymbolEncoder.cs

@@ -117,6 +117,11 @@ internal class Av1SymbolEncoder : IDisposable
     /// </summary>
     private readonly Av1Distribution[][] transformSize;
 
+    /// <summary>
+    /// The tile-adaptive variable-transform partition distributions.
+    /// </summary>
+    private readonly Av1Distribution[] transformPartition;
+
     /// <summary>
     /// The tile-adaptive spatial segment-identifier distributions.
     /// </summary>
@@ -200,6 +205,7 @@ internal class Av1SymbolEncoder : IDisposable
         this.intraExtendedTransform = Av1DefaultDistributions.IntraExtendedTransform;
         this.interExtendedTransform = Av1DefaultDistributions.InterExtendedTransform;
         this.transformSize = Av1DefaultDistributions.TransformSize;
+        this.transformPartition = Av1DefaultDistributions.TransformPartition;
         this.segmentId = Av1DefaultDistributions.SegmentId;
         this.angleDelta = Av1DefaultDistributions.AngleDelta;
         this.skip = Av1DefaultDistributions.Skip;
@@ -1208,6 +1214,26 @@ internal class Av1SymbolEncoder : IDisposable
         w.WriteSymbol(selectedDepth, this.transformSize[categoryDepth - 1][context]);
     }
 
+    /// <summary>
+    /// Gets the current fixed-point cost of one variable-transform partition decision.
+    /// </summary>
+    /// <param name="split">Indicates whether the current transform node is split.</param>
+    /// <param name="context">The neighboring variable-transform context.</param>
+    /// <returns>The rate cost in 1/512-bit units.</returns>
+    public int GetTransformPartitionCost(bool split, int context)
+        => Av1ProbabilityCost.GetSymbolCost(this.transformPartition[context], split ? 1 : 0);
+
+    /// <summary>
+    /// Writes one variable-transform partition decision.
+    /// </summary>
+    /// <param name="split">Indicates whether the current transform node is split.</param>
+    /// <param name="context">The neighboring variable-transform context.</param>
+    public void WriteTransformPartition(bool split, int context)
+    {
+        ref Av1SymbolWriter w = ref this.writer;
+        w.WriteSymbol(split ? 1 : 0, this.transformPartition[context]);
+    }
+
     private static int GetTransformSizeDepth(
         Av1BlockSize blockSize,
         Av1TransformSize transformSize,

src/ImageSharp/Formats/Heif/Av1/Pipeline/Av1EncoderModeDecisionWorkspace.cs

@@ -4,6 +4,7 @@
 using System.Runtime.InteropServices;
 using SixLabors.ImageSharp.Formats.Heif.Av1.Prediction;
 using SixLabors.ImageSharp.Formats.Heif.Av1.Prediction.ChromaFromLuma;
+using SixLabors.ImageSharp.Formats.Heif.Av1.Tiling;
 
 namespace SixLabors.ImageSharp.Formats.Heif.Av1.Pipeline;
 
@@ -19,6 +20,11 @@ internal readonly ref struct Av1EncoderModeDecisionWorkspace<TSample>
     /// </summary>
     public const int MaximumSampleCount = 8 * 8;
 
+    /// <summary>
+    /// The number of 4x4 transform blocks covering one 8x8 coding block.
+    /// </summary>
+    public const int CandidateTransformBlockCount = 4;
+
     /// <summary>
     /// The required workspace length in signed-integer storage elements.
     /// </summary>
@@ -31,7 +37,11 @@ internal readonly ref struct Av1EncoderModeDecisionWorkspace<TSample>
     private const int CandidateSampleStorageLength = 2 * MaximumSampleCount * sizeof(ushort) / sizeof(int);
     private const int CandidateCoefficientStorageOffset = CandidateSampleStorageOffset + CandidateSampleStorageLength;
     private const int CandidateCoefficientStorageLength = 2 * MaximumSampleCount;
-    private const int TransientStorageOffset = CandidateCoefficientStorageOffset + CandidateCoefficientStorageLength;
+    private const int CandidateTransformBlockStorageOffset = CandidateCoefficientStorageOffset + CandidateCoefficientStorageLength;
+    private const int CandidateTransformBlockStorageLength = CandidateTransformBlockCount;
+    private const int TransformContextStorageOffset = CandidateTransformBlockStorageOffset + CandidateTransformBlockStorageLength;
+    private const int TransformContextStorageLength = 1;
+    private const int TransientStorageOffset = TransformContextStorageOffset + TransformContextStorageLength;
     private const int ChromaFromLumaSampleCount = Av1ChromaFromLumaContext.BufferLine * 8;
     private const int ChromaFromLumaSampleStorageLength = ChromaFromLumaSampleCount * sizeof(short) / sizeof(int);
     private const int ChromaFromLumaBlueRateOffset = ChromaFromLumaSampleStorageLength;
@@ -81,6 +91,19 @@ internal readonly ref struct Av1EncoderModeDecisionWorkspace<TSample>
     public Av1EncoderPaletteWorkspace<TSample> Palette
         => new(this.storage[TransientStorageOffset..]);
 
+    /// <summary>
+    /// Gets the transform state retained while evaluating a uniform 4x4 luma layout.
+    /// </summary>
+    public Span<Av1EncoderTransformBlockState> CandidateTransformBlocks
+        => MemoryMarshal.Cast<int, Av1EncoderTransformBlockState>(
+            this.storage.Slice(CandidateTransformBlockStorageOffset, CandidateTransformBlockStorageLength));
+
+    /// <summary>
+    /// Gets the two above and two left coefficient contexts used by a uniform 4x4 luma layout.
+    /// </summary>
+    public Span<byte> TransformContexts
+        => MemoryMarshal.AsBytes(this.storage.Slice(TransformContextStorageOffset, TransformContextStorageLength));
+
     /// <summary>
     /// Gets one reference edge including its common-corner prefix.
     /// </summary>

src/ImageSharp/Formats/Heif/Av1/Pipeline/Av1FrameEncoder.cs

@@ -119,7 +119,7 @@ internal static class Av1FrameEncoder
             ErrorResilientMode = true,
             RefreshFrameFlags = byte.MaxValue,
             DisableFrameEndUpdateCdf = true,
-            TransformMode = Av1TransformMode.Largest,
+            TransformMode = effort >= 6 ? Av1TransformMode.Select : Av1TransformMode.Largest,
             ModeInfoColumnCount = modeInfoColumnCount,
             ModeInfoRowCount = modeInfoRowCount,
             TilesInfo = tiles,

src/ImageSharp/Formats/Heif/Av1/Pipeline/Av1IntraSuperblockEncoder.ChromaPaletteModeDecision.cs

@@ -247,6 +247,7 @@ internal static partial class Av1IntraSuperblockEncoder
                     bluePrediction,
                     blueResidual,
                     candidateBlueReconstruction,
+                    transformSize.GetWidth(),
                     candidateBlueCoefficients,
                     transformSize,
                     Av1TransformType.DctDct,
@@ -265,6 +266,7 @@ internal static partial class Av1IntraSuperblockEncoder
                     redPrediction,
                     redResidual,
                     candidateRedReconstruction,
+                    transformSize.GetWidth(),
                     candidateRedCoefficients,
                     transformSize,
                     Av1TransformType.DctDct,

src/ImageSharp/Formats/Heif/Av1/Pipeline/Av1IntraSuperblockEncoder.IntraBlockCopyModeDecision.cs

@@ -128,6 +128,23 @@ internal static partial class Av1IntraSuperblockEncoder
                 BlockSize,
                 LumaTransformSize);
 
+            int transformPartitionRate = 0;
+            if (this.picture.Parent.FrameHeader.TransformMode == Av1TransformMode.Select)
+            {
+                Av1NeighborArrayUnit<byte> transformContexts = this.picture.TransformFunctionContexts[tileIndex];
+                int topIndex = transformContexts.GetTopIndex(blockOrigin);
+                int leftIndex = transformContexts.GetLeftIndex(blockOrigin);
+                int transformPartitionContext = Av1SymbolContextHelper.GetTransformPartitionContext(
+                    transformContexts.Top[topIndex],
+                    transformContexts.Left[leftIndex],
+                    BlockSize,
+                    LumaTransformSize);
+
+                transformPartitionRate = writer.GetTransformPartitionCost(
+                    false,
+                    transformPartitionContext);
+            }
+
             ObuColorConfig colorConfig = this.picture.Sequence.SequenceHeader.ColorConfig;
             int subsamplingX = colorConfig.SubSamplingX ? 1 : 0;
             int subsamplingY = colorConfig.SubSamplingY ? 1 : 0;
@@ -250,6 +267,7 @@ internal static partial class Av1IntraSuperblockEncoder
                 int candidateRate = writer.GetUseIntraBlockCopyCost(true) +
                     displacementRate +
                     writer.GetSkipCost(false, skipContext) +
+                    transformPartitionRate +
                     lumaRate +
                     blueRate +
                     redRate;
@@ -385,6 +403,7 @@ internal static partial class Av1IntraSuperblockEncoder
 
             modeInfo.Block.Mode = Av1PredictionMode.DC;
             modeInfo.Block.UvMode = Av1ChromaPredictionMode.DC;
+            modeInfo.Block.TransformSize = LumaTransformSize;
             modeInfo.Block.Skip = selectedSkip;
             modeInfo.Block.UseIntraBlockCopy = true;
             block.FilterIntraMode = Av1FilterIntraMode.AllFilterIntraModes;
@@ -466,6 +485,7 @@ internal static partial class Av1IntraSuperblockEncoder
                     prediction[..sampleCount],
                     residual[..sampleCount],
                     transformReconstruction[..sampleCount],
+                    transformSize.GetWidth(),
                     transformCoefficients[..sampleCount],
                     transformSize,
                     transformType,

src/ImageSharp/Formats/Heif/Av1/Pipeline/Av1IntraSuperblockEncoder.ModeDecision.cs

@@ -189,23 +189,31 @@ internal static partial class Av1IntraSuperblockEncoder
             int lumaTransformIndex = this.codedAreaLuma /
                 Av1EncoderCoefficientBuffer.TransformBlockUnitCoefficientCount;
 
-            ref Av1EncoderTransformBlockState lumaState = ref lumaTransformBlocks[lumaTransformIndex];
+            Span<Av1EncoderTransformBlockState> retainedLumaStates = lumaTransformBlocks[lumaTransformIndex..];
             modeInfo.Block.Mode = this.SelectLumaMode(
                 writer,
                 macroBlock,
                 blockOrigin,
                 tileIndex,
                 lumaCoefficients[this.codedAreaLuma..],
-                ref lumaState,
+                retainedLumaStates,
                 ref paletteInfo,
                 out int lumaAngleDelta,
                 out Av1FilterIntraMode filterIntraMode,
+                out Av1TransformSize lumaTransformSize,
                 out long lumaCost);
 
             block.PredictionUnit.AngleDelta[(int)Av1PlaneType.Y] = (sbyte)lumaAngleDelta;
             block.FilterIntraMode = filterIntraMode;
+            modeInfo.Block.TransformSize = lumaTransformSize;
+
+            int lumaTransformBlockCount = LumaTransformSize.GetSize2d() / lumaTransformSize.GetSize2d();
+            bool lumaTransformEmpty = true;
+            for (int transformIndex = 0; transformIndex < lumaTransformBlockCount; transformIndex++)
+            {
+                lumaTransformEmpty &= retainedLumaStates[transformIndex].EndOfBlock == 0;
+            }
 
-            bool lumaTransformEmpty = lumaState.EndOfBlock == 0;
             if (this.source.IsMonochrome)
             {
                 int emptyTransformRate = lumaTransformEmpty
@@ -215,8 +223,7 @@ internal static partial class Av1IntraSuperblockEncoder
                         Av1ComponentType.Luminance,
                         blockOrigin,
                         BlockSize,
-                        LumaTransformSize,
+                        lumaTransformSize,
-                        lumaState.TransformType,
                         modeInfo.Block.Mode,
                         block.FilterIntraMode)
                     : 0;
@@ -302,8 +309,7 @@ internal static partial class Av1IntraSuperblockEncoder
                     Av1ComponentType.Luminance,
                     blockOrigin,
                     BlockSize,
-                    LumaTransformSize,
+                    lumaTransformSize,
-                    lumaState.TransformType,
                     modeInfo.Block.Mode,
                     block.FilterIntraMode);
 
@@ -314,7 +320,6 @@ internal static partial class Av1IntraSuperblockEncoder
                     chromaOrigin,
                     chromaBlockSize,
                     chromaTransformSize,
-                    blueState.TransformType,
                     modeInfo.Block.Mode,
                     Av1FilterIntraMode.AllFilterIntraModes);
 
@@ -325,7 +330,6 @@ internal static partial class Av1IntraSuperblockEncoder
                     chromaOrigin,
                     chromaBlockSize,
                     chromaTransformSize,
-                    redState.TransformType,
                     modeInfo.Block.Mode,
                     Av1FilterIntraMode.AllFilterIntraModes);
 
@@ -360,27 +364,73 @@ internal static partial class Av1IntraSuperblockEncoder
             Point blockOrigin,
             Av1BlockSize blockSize,
             Av1TransformSize transformSize,
-            Av1TransformType transformType,
             Av1PredictionMode lumaMode,
             Av1FilterIntraMode filterIntraMode)
         {
-            Av1TransformBlockContext blockContext = Av1TileWriter.GetTransformBlockContexts(
+            int blockWidth = blockSize.Get4x4WideCount();
-                componentType,
+            int blockHeight = blockSize.Get4x4HighCount();
-                coefficientNeighbors,
+            int transformWidth = transformSize.Get4x4WideCount();
-                blockOrigin,
+            int transformHeight = transformSize.Get4x4HighCount();
-                blockSize,
+            if (blockWidth == transformWidth && blockHeight == transformHeight)
-                transformSize);
+            {
+                Av1TransformBlockContext blockContext = Av1TileWriter.GetTransformBlockContexts(
+                    componentType,
+                    coefficientNeighbors,
+                    blockOrigin,
+                    blockSize,
+                    transformSize);
+
+                return writer.GetCoefficientCost(
+                    transformSize,
+                    Av1TransformType.DctDct,
+                    lumaMode,
+                    ReadOnlySpan<int>.Empty,
+                    componentType,
+                    blockContext,
+                    0,
+                    this.picture.Parent.FrameHeader.UseReducedTransformSet,
+                    filterIntraMode);
+            }
 
-            return writer.GetCoefficientCost(
+            Span<byte> contexts = this.blockWorkspace
-                transformSize,
+                .GetModeDecisionWorkspace<TSample>()
-                transformType,
+                .TransformContexts;
-                lumaMode,
+
-                ReadOnlySpan<int>.Empty,
+            Span<byte> topContexts = contexts[..blockWidth];
-                componentType,
+            Span<byte> leftContexts = contexts.Slice(blockWidth, blockHeight);
-                blockContext,
+            int topIndex = coefficientNeighbors.GetTopIndex(blockOrigin);
-                0,
+            int leftIndex = coefficientNeighbors.GetLeftIndex(blockOrigin);
-                this.picture.Parent.FrameHeader.UseReducedTransformSet,
+            coefficientNeighbors.Top.Slice(topIndex, blockWidth).CopyTo(topContexts);
-                filterIntraMode);
+            coefficientNeighbors.Left.Slice(leftIndex, blockHeight).CopyTo(leftContexts);
+            int rate = 0;
+            for (int blockRow = 0; blockRow < blockHeight; blockRow += transformHeight)
+            {
+                for (int blockColumn = 0; blockColumn < blockWidth; blockColumn += transformWidth)
+                {
+                    Av1TransformBlockContext blockContext = Av1TileWriter.GetTransformBlockContexts(
+                        componentType,
+                        topContexts.Slice(blockColumn, transformWidth),
+                        leftContexts.Slice(blockRow, transformHeight),
+                        blockSize,
+                        transformSize);
+
+                    rate += writer.GetCoefficientCost(
+                        transformSize,
+                        Av1TransformType.DctDct,
+                        lumaMode,
+                        ReadOnlySpan<int>.Empty,
+                        componentType,
+                        blockContext,
+                        0,
+                        this.picture.Parent.FrameHeader.UseReducedTransformSet,
+                        filterIntraMode);
+
+                    topContexts.Slice(blockColumn, transformWidth).Clear();
+                    leftContexts.Slice(blockRow, transformHeight).Clear();
+                }
+            }
+
+            return rate;
         }
 
         private Av1PredictionMode SelectLumaMode(
@@ -389,10 +439,11 @@ internal static partial class Av1IntraSuperblockEncoder
             Point blockOrigin,
             ushort tileIndex,
             Span<int> retainedCoefficients,
-            ref Av1EncoderTransformBlockState retainedState,
+            Span<Av1EncoderTransformBlockState> retainedStates,
             ref Av1EncoderPaletteInfo paletteInfo,
             out int selectedAngleDelta,
             out Av1FilterIntraMode selectedFilterIntraMode,
+            out Av1TransformSize selectedTransformSize,
             out long selectedCost)
         {
             const Av1BlockSize BlockSize = Av1BlockSize.Block8x8;
@@ -509,6 +560,16 @@ internal static partial class Av1IntraSuperblockEncoder
                 BlockSize,
                 TransformSize);
 
+            int transformSizeContext = Av1TileWriter.GetTransformSizeContext(
+                this.picture.TransformFunctionContexts[tileIndex],
+                macroBlock,
+                blockOrigin,
+                BlockSize);
+
+            int largestTransformRate = this.picture.Parent.FrameHeader.TransformMode == Av1TransformMode.Select
+                ? writer.GetTransformSizeCost(BlockSize, TransformSize, transformSizeContext)
+                : 0;
+
             int paletteDisabledCost = 0;
             if (this.picture.Parent.FrameHeader.AllowScreenContentTools)
             {
@@ -586,6 +647,7 @@ internal static partial class Av1IntraSuperblockEncoder
                     defaultTransformType,
                     blockContext,
                     paletteDisabledCost,
+                    largestTransformRate,
                     candidateReconstruction,
                     candidateCoefficients,
                     ref candidateState);
@@ -600,7 +662,7 @@ internal static partial class Av1IntraSuperblockEncoder
                         retainedCoefficients,
                         TransformSize,
                         candidateState,
-                        ref retainedState);
+                        ref retainedStates[0]);
 
                     bestCost = candidateCost;
                     bestMode = mode;
@@ -635,6 +697,7 @@ internal static partial class Av1IntraSuperblockEncoder
                     transformType,
                     blockContext,
                     paletteDisabledCost,
+                    largestTransformRate,
                     candidateReconstruction,
                     candidateCoefficients,
                     ref candidateState);
@@ -649,7 +712,7 @@ internal static partial class Av1IntraSuperblockEncoder
                         retainedCoefficients,
                         TransformSize,
                         candidateState,
-                        ref retainedState);
+                        ref retainedStates[0]);
 
                     bestTransformCost = candidateCost;
                 }
@@ -699,6 +762,7 @@ internal static partial class Av1IntraSuperblockEncoder
                             transformType,
                             blockContext,
                             paletteDisabledCost,
+                            largestTransformRate,
                             candidateReconstruction,
                             candidateCoefficients,
                             ref candidateState);
@@ -713,7 +777,7 @@ internal static partial class Av1IntraSuperblockEncoder
                                 retainedCoefficients,
                                 TransformSize,
                                 candidateState,
-                                ref retainedState);
+                                ref retainedStates[0]);
 
                             bestTransformCost = candidateCost;
                             bestMode = Av1PredictionMode.DC;
@@ -735,10 +799,11 @@ internal static partial class Av1IntraSuperblockEncoder
                     tileIndex,
                     transformSetType,
                     blockContext,
+                    largestTransformRate,
                     candidateReconstruction,
                     candidateCoefficients,
                     retainedCoefficients,
-                    ref retainedState,
+                    ref retainedStates[0],
                     ref bestTransformCost,
                     ref paletteInfo))
             {
@@ -747,10 +812,413 @@ internal static partial class Av1IntraSuperblockEncoder
                 selectedFilterIntraMode = Av1FilterIntraMode.AllFilterIntraModes;
             }
 
+            selectedTransformSize = TransformSize;
+            if (this.effort >= 6 &&
+                this.picture.Parent.FrameHeader.TransformMode == Av1TransformMode.Select &&
+                paletteInfo.PaletteSizes[0] == 0 &&
+                selectedFilterIntraMode == Av1FilterIntraMode.AllFilterIntraModes)
+            {
+                long splitCost = this.GetSplitLumaCandidateCost(
+                    writer,
+                    macroBlock,
+                    sourcePlane,
+                    reconstructionPlane,
+                    blockOrigin,
+                    tileIndex,
+                    bestMode,
+                    selectedAngleDelta,
+                    paletteDisabledCost,
+                    transformSizeContext,
+                    bestTransformCost,
+                    candidateReconstruction,
+                    candidateCoefficients,
+                    workspace.CandidateTransformBlocks);
+
+                if (splitCost < bestTransformCost)
+                {
+                    CopySplitCandidate(
+                        candidateReconstruction,
+                        candidateCoefficients,
+                        workspace.CandidateTransformBlocks,
+                        reconstructionPlane,
+                        blockOrigin,
+                        retainedCoefficients,
+                        retainedStates);
+
+                    bestTransformCost = splitCost;
+                    selectedTransformSize = Av1TransformSize.Size4x4;
+                }
+            }
+
             selectedCost = bestTransformCost;
             return bestMode;
         }
 
+        private long GetSplitLumaCandidateCost(
+            Av1SymbolEncoder writer,
+            Av1MacroBlockD macroBlock,
+            Buffer2DRegion<TSample> sourcePlane,
+            Buffer2DRegion<TSample> reconstructionPlane,
+            Point blockOrigin,
+            ushort tileIndex,
+            Av1PredictionMode mode,
+            int angleDelta,
+            int paletteDisabledCost,
+            int transformSizeContext,
+            long costLimit,
+            Span<TSample> candidateReconstruction,
+            Span<int> candidateCoefficients,
+            Span<Av1EncoderTransformBlockState> candidateTransformBlocks)
+        {
+            const Av1BlockSize BlockSize = Av1BlockSize.Block8x8;
+            const Av1TransformSize TransformSize = Av1TransformSize.Size4x4;
+            const int BlockWidth = 8;
+            const int TransformWidth = 4;
+            const int TransformSampleCount = TransformWidth * TransformWidth;
+            Av1EncoderModeDecisionWorkspace<TSample> workspace =
+                this.blockWorkspace.GetModeDecisionWorkspace<TSample>();
+
+            Span<TSample> transformSamples = workspace.GetCandidateReconstruction(1);
+            Span<TSample> prediction = transformSamples[..TransformSampleCount];
+            Span<TSample> transformReconstruction = transformSamples.Slice(
+                TransformSampleCount,
+                TransformSampleCount);
+
+            Span<int> transformCoefficients = workspace.GetCandidateCoefficients(1)[..TransformSampleCount];
+            Span<short> residual = workspace.FilterResidual[..TransformSampleCount];
+            Span<byte> contexts = workspace.TransformContexts;
+            Span<byte> topContexts = contexts[..2];
+            Span<byte> leftContexts = contexts[2..4];
+            Av1NeighborArrayUnit<byte> coefficientNeighbors =
+                this.picture.LuminanceDcSignLevelCoefficientNeighbors[tileIndex];
+
+            int topIndex = coefficientNeighbors.GetTopIndex(blockOrigin);
+            int leftIndex = coefficientNeighbors.GetLeftIndex(blockOrigin);
+            coefficientNeighbors.Top.Slice(topIndex, 2).CopyTo(topContexts);
+            coefficientNeighbors.Left.Slice(leftIndex, 2).CopyTo(leftContexts);
+            bool useReducedTransformSet = this.picture.Parent.FrameHeader.UseReducedTransformSet;
+            Av1TransformSetType transformSetType = Av1SymbolContextHelper.GetExtendedTransformSetType(
+                TransformSize,
+                useReducedTransformSet);
+
+            int rate = Av1TileWriter.GetLumaModeCost(writer, macroBlock, BlockSize, mode, angleDelta);
+            rate += writer.GetTransformSizeCost(BlockSize, TransformSize, transformSizeContext);
+            if (mode == Av1PredictionMode.DC)
+            {
+                rate += paletteDisabledCost;
+                if (this.picture.Sequence.SequenceHeader.EnableFilterIntra)
+                {
+                    rate += writer.GetFilterIntraModeCost(
+                        Av1FilterIntraMode.AllFilterIntraModes,
+                        BlockSize);
+                }
+            }
+
+            long distortion = 0;
+            for (int transformRow = 0; transformRow < 2; transformRow++)
+            {
+                for (int transformColumn = 0; transformColumn < 2; transformColumn++)
+                {
+                    int transformIndex = (transformRow * 2) + transformColumn;
+                    int reconstructionOffset =
+                        (transformRow * TransformWidth * BlockWidth) + (transformColumn * TransformWidth);
+
+                    Point transformOrigin = blockOrigin + new Size(
+                        transformColumn * TransformWidth,
+                        transformRow * TransformWidth);
+
+                    Span<TSample> aboveStorage = workspace.GetReferenceSamples(0);
+                    Span<TSample> leftStorage = workspace.GetReferenceSamples(1);
+                    this.PrepareSplitLumaReferenceSamples(
+                        reconstructionPlane,
+                        blockOrigin,
+                        macroBlock,
+                        transformRow,
+                        transformColumn,
+                        candidateReconstruction,
+                        aboveStorage,
+                        leftStorage,
+                        out bool hasLeft,
+                        out bool hasAbove);
+
+                    TOperator.PrepareIntra(
+                        this.blockWorkspace,
+                        sourcePlane,
+                        transformOrigin,
+                        prediction,
+                        aboveStorage.Slice(1, TransformWidth * 2),
+                        leftStorage.Slice(1, TransformWidth * 2),
+                        hasLeft,
+                        hasAbove,
+                        mode,
+                        angleDelta,
+                        residual,
+                        TransformSize,
+                        this.bitDepth);
+
+                    Av1TransformBlockContext blockContext = Av1TileWriter.GetTransformBlockContexts(
+                        Av1ComponentType.Luminance,
+                        topContexts.Slice(transformColumn, 1),
+                        leftContexts.Slice(transformRow, 1),
+                        BlockSize,
+                        TransformSize);
+
+                    long bestTransformCost = long.MaxValue;
+                    Av1TransformType bestTransformType = Av1TransformType.DctDct;
+                    int bestTransformRate = 0;
+                    long bestTransformDistortion = 0;
+                    Av1EncoderTransformBlockState bestTransformState = default;
+                    Span<int> retainedTransformCoefficients = candidateCoefficients.Slice(
+                        transformIndex * TransformSampleCount,
+                        TransformSampleCount);
+
+                    for (Av1TransformType transformType = Av1TransformType.DctDct;
+                        transformType < Av1TransformType.AllTransformTypes;
+                        transformType++)
+                    {
+                        if (!transformType.IsExtendedSetUsed(transformSetType))
+                        {
+                            continue;
+                        }
+
+                        Av1EncoderTransformBlockState candidateState = default;
+                        long candidateDistortion = TOperator.EncodePredictionCandidate(
+                            this.blockWorkspace,
+                            sourcePlane,
+                            transformOrigin,
+                            prediction,
+                            residual,
+                            transformReconstruction,
+                            TransformWidth,
+                            transformCoefficients,
+                            TransformSize,
+                            transformType,
+                            Av1Plane.Y,
+                            this.quantization.QIndex[0],
+                            this.quantization.DeltaQDc[(int)Av1Plane.Y],
+                            this.quantization.DeltaQAc[(int)Av1Plane.Y],
+                            this.bitDepth,
+                            ref candidateState);
+
+                        int candidateRate = writer.GetCoefficientCost(
+                            TransformSize,
+                            transformType,
+                            mode,
+                            transformCoefficients,
+                            Av1ComponentType.Luminance,
+                            blockContext,
+                            candidateState.EndOfBlock,
+                            useReducedTransformSet,
+                            Av1FilterIntraMode.AllFilterIntraModes);
+
+                        long candidateCost = Av1RateDistortion.GetCost(
+                            this.rateMultiplier,
+                            candidateRate,
+                            candidateDistortion);
+
+                        if (candidateCost < bestTransformCost)
+                        {
+                            // Preserve the improving 4x4 trial in the block mosaic. Later transform predictions
+                            // consume that reconstruction, and copying the compact result avoids another transform.
+                            transformCoefficients.CopyTo(retainedTransformCoefficients);
+                            for (int row = 0; row < TransformWidth; row++)
+                            {
+                                transformReconstruction.Slice(row * TransformWidth, TransformWidth)
+                                    .CopyTo(
+                                        candidateReconstruction.Slice(
+                                            reconstructionOffset + (row * BlockWidth),
+                                            TransformWidth));
+                            }
+
+                            bestTransformCost = candidateCost;
+                            bestTransformType = transformType;
+                            bestTransformRate = candidateRate;
+                            bestTransformDistortion = candidateDistortion;
+                            bestTransformState = candidateState;
+                        }
+                    }
+
+                    rate += bestTransformRate;
+                    distortion += bestTransformDistortion;
+                    candidateTransformBlocks[transformIndex] = bestTransformState;
+                    byte coefficientContext = Av1SymbolContextHelper.GetCoefficientContext(
+                        retainedTransformCoefficients,
+                        TransformSize,
+                        bestTransformType,
+                        bestTransformState.EndOfBlock);
+
+                    topContexts[transformColumn] = coefficientContext;
+                    leftContexts[transformRow] = coefficientContext;
+
+                    // Every remaining transform can only add nonnegative rate and distortion.
+                    if (Av1RateDistortion.GetCost(this.rateMultiplier, rate, distortion) >= costLimit)
+                    {
+                        return long.MaxValue;
+                    }
+                }
+            }
+
+            return Av1RateDistortion.GetCost(this.rateMultiplier, rate, distortion);
+        }
+
+        private void PrepareSplitLumaReferenceSamples(
+            Buffer2DRegion<TSample> reconstructionPlane,
+            Point blockOrigin,
+            Av1MacroBlockD macroBlock,
+            int transformRow,
+            int transformColumn,
+            ReadOnlySpan<TSample> candidateReconstruction,
+            Span<TSample> aboveStorage,
+            Span<TSample> leftStorage,
+            out bool hasLeft,
+            out bool hasAbove)
+        {
+            const Av1BlockSize BlockSize = Av1BlockSize.Block8x8;
+            const Av1TransformSize TransformSize = Av1TransformSize.Size4x4;
+            const int BlockWidth = 8;
+            const int TransformWidth = 4;
+            int rowOffset = transformRow * TransformWidth;
+            int columnOffset = transformColumn * TransformWidth;
+            int modeInfoRow = blockOrigin.Y >> Av1Constants.ModeInfoSizeLog2;
+            int modeInfoColumn = blockOrigin.X >> Av1Constants.ModeInfoSizeLog2;
+            hasAbove = transformRow > 0 || macroBlock.IsUpAvailable;
+            hasLeft = transformColumn > 0 || macroBlock.IsLeftAvailable;
+            bool rightAvailable =
+                modeInfoColumn + transformColumn + TransformSize.Get4x4WideCount() <
+                macroBlock.Tile.ModeInfoColumnEnd;
+
+            bool bottomAvailable =
+                modeInfoRow + transformRow + TransformSize.Get4x4HighCount() <
+                macroBlock.Tile.ModeInfoRowEnd;
+
+            bool hasTopRight = Av1IntraReferenceAvailability.HasTopRight(
+                this.picture.Sequence.SequenceHeader.SuperblockSize,
+                BlockSize,
+                modeInfoRow,
+                modeInfoColumn,
+                hasAbove,
+                rightAvailable,
+                Av1PartitionType.None,
+                TransformSize,
+                transformRow,
+                transformColumn,
+                0,
+                0);
+
+            bool hasBottomLeft = Av1IntraReferenceAvailability.HasBottomLeft(
+                this.picture.Sequence.SequenceHeader.SuperblockSize,
+                BlockSize,
+                modeInfoRow,
+                modeInfoColumn,
+                bottomAvailable,
+                hasLeft,
+                Av1PartitionType.None,
+                TransformSize,
+                transformRow,
+                transformColumn,
+                0,
+                0);
+
+            Span<TSample> above = aboveStorage.Slice(1, TransformWidth * 2);
+            Span<TSample> left = leftStorage.Slice(1, TransformWidth * 2);
+            if (hasAbove)
+            {
+                if (transformRow > 0)
+                {
+                    candidateReconstruction
+                        .Slice(((rowOffset - 1) * BlockWidth) + columnOffset, TransformWidth)
+                        .CopyTo(above);
+                }
+                else
+                {
+                    reconstructionPlane.DangerousGetRowSpan(blockOrigin.Y - 1)
+                        .Slice(blockOrigin.X + columnOffset, TransformWidth)
+                        .CopyTo(above);
+                }
+            }
+
+            if (hasLeft)
+            {
+                if (transformColumn > 0)
+                {
+                    for (int row = 0; row < TransformWidth; row++)
+                    {
+                        left[row] = candidateReconstruction[((rowOffset + row) * BlockWidth) + columnOffset - 1];
+                    }
+                }
+                else
+                {
+                    for (int row = 0; row < TransformWidth; row++)
+                    {
+                        left[row] = reconstructionPlane
+                            .DangerousGetRowSpan(blockOrigin.Y + rowOffset + row)[blockOrigin.X - 1];
+                    }
+                }
+            }
+
+            int midpoint = 128 << (this.bitDepth.GetBitCount() - 8);
+            if (!hasAbove)
+            {
+                above[..TransformWidth].Fill(hasLeft ? left[0] : TOperator.CreateSample(midpoint - 1));
+            }
+
+            if (!hasLeft)
+            {
+                left[..TransformWidth].Fill(hasAbove ? above[0] : TOperator.CreateSample(midpoint + 1));
+            }
+
+            if (hasTopRight)
+            {
+                if (transformRow > 0)
+                {
+                    candidateReconstruction
+                        .Slice(
+                            ((rowOffset - 1) * BlockWidth) + columnOffset + TransformWidth,
+                            TransformWidth)
+                        .CopyTo(above[TransformWidth..]);
+                }
+                else
+                {
+                    reconstructionPlane.DangerousGetRowSpan(blockOrigin.Y - 1)
+                        .Slice(blockOrigin.X + columnOffset + TransformWidth, TransformWidth)
+                        .CopyTo(above[TransformWidth..]);
+                }
+            }
+            else
+            {
+                above[TransformWidth..].Fill(above[TransformWidth - 1]);
+            }
+
+            if (hasBottomLeft)
+            {
+                for (int row = TransformWidth; row < TransformWidth * 2; row++)
+                {
+                    left[row] = reconstructionPlane
+                        .DangerousGetRowSpan(blockOrigin.Y + rowOffset + row)[blockOrigin.X - 1];
+                }
+            }
+            else
+            {
+                left[TransformWidth..].Fill(left[TransformWidth - 1]);
+            }
+
+            // Only an interior transform corner belongs to decision scratch. Boundary corners continue
+            // to read the already reconstructed neighboring block so candidate trials remain isolated.
+            TSample corner = hasAbove && hasLeft
+                ? transformRow > 0 && transformColumn > 0
+                    ? candidateReconstruction[((rowOffset - 1) * BlockWidth) + columnOffset - 1]
+                    : reconstructionPlane.DangerousGetRowSpan(blockOrigin.Y + rowOffset - 1)[
+                        blockOrigin.X + columnOffset - 1]
+                : hasAbove
+                    ? above[0]
+                    : hasLeft
+                        ? left[0]
+                        : TOperator.CreateSample(midpoint);
+
+            aboveStorage[0] = corner;
+            leftStorage[0] = corner;
+        }
+
         private long GetLumaCandidateCost(
             Av1SymbolEncoder writer,
             Av1MacroBlockD macroBlock,
@@ -765,6 +1233,7 @@ internal static partial class Av1IntraSuperblockEncoder
             Av1TransformType transformType,
             Av1TransformBlockContext blockContext,
             int paletteDisabledCost,
+            int transformSizeRate,
             Span<TSample> candidateReconstruction,
             Span<int> candidateCoefficients,
             ref Av1EncoderTransformBlockState candidateState)
@@ -793,6 +1262,7 @@ internal static partial class Av1IntraSuperblockEncoder
                 ref candidateState);
 
             int rate = Av1TileWriter.GetLumaModeCost(writer, macroBlock, BlockSize, mode, angleDelta);
+            rate += transformSizeRate;
             if (mode == Av1PredictionMode.DC)
             {
                 rate += paletteDisabledCost;
@@ -828,6 +1298,7 @@ internal static partial class Av1IntraSuperblockEncoder
             Av1TransformType transformType,
             Av1TransformBlockContext blockContext,
             int paletteDisabledCost,
+            int transformSizeRate,
             Span<TSample> candidateReconstruction,
             Span<int> candidateCoefficients,
             ref Av1EncoderTransformBlockState candidateState)
@@ -841,6 +1312,7 @@ internal static partial class Av1IntraSuperblockEncoder
                 prediction,
                 residual,
                 candidateReconstruction,
+                TransformSize.GetWidth(),
                 candidateCoefficients,
                 TransformSize,
                 transformType,
@@ -858,6 +1330,7 @@ internal static partial class Av1IntraSuperblockEncoder
                 Av1PredictionMode.DC,
                 0);
 
+            rate += transformSizeRate;
             rate += paletteDisabledCost;
             rate += writer.GetFilterIntraModeCost(filterIntraMode, BlockSize);
             rate += writer.GetCoefficientCost(
@@ -895,5 +1368,30 @@ internal static partial class Av1IntraSuperblockEncoder
 
             retainedState = candidateState;
         }
+
+        private static void CopySplitCandidate(
+            ReadOnlySpan<TSample> candidateReconstruction,
+            ReadOnlySpan<int> candidateCoefficients,
+            ReadOnlySpan<Av1EncoderTransformBlockState> candidateTransformBlocks,
+            Buffer2DRegion<TSample> reconstructionPlane,
+            Point blockOrigin,
+            Span<int> retainedCoefficients,
+            Span<Av1EncoderTransformBlockState> retainedTransformBlocks)
+        {
+            const int BlockWidth = 8;
+            const int SampleCount = BlockWidth * BlockWidth;
+            candidateCoefficients[..SampleCount].CopyTo(retainedCoefficients);
+            candidateTransformBlocks[..Av1EncoderModeDecisionWorkspace<TSample>.CandidateTransformBlockCount]
+                .CopyTo(retainedTransformBlocks);
+
+            for (int row = 0; row < BlockWidth; row++)
+            {
+                candidateReconstruction.Slice(row * BlockWidth, BlockWidth)
+                    .CopyTo(
+                        reconstructionPlane
+                            .DangerousGetRowSpan(blockOrigin.Y + row)
+                            .Slice(blockOrigin.X, BlockWidth));
+            }
+        }
     }
 }

src/ImageSharp/Formats/Heif/Av1/Pipeline/Av1IntraSuperblockEncoder.Operator.cs

@@ -191,6 +191,37 @@ internal static partial class Av1IntraSuperblockEncoder
             Av1BitDepth bitDepth,
             ref Av1EncoderTransformBlockState state);
 
+        /// <summary>
+        /// Builds one spatial intra prediction and its source residual for reuse across transform candidates.
+        /// </summary>
+        /// <param name="workspace">The reusable block workspace.</param>
+        /// <param name="source">The coded source plane.</param>
+        /// <param name="blockOrigin">The transform-block origin in plane samples.</param>
+        /// <param name="prediction">The contiguous prediction destination.</param>
+        /// <param name="above">The top reference samples, with prefix storage for the shared corner.</param>
+        /// <param name="left">The left reference samples.</param>
+        /// <param name="hasLeft">Whether the left reference is available.</param>
+        /// <param name="hasAbove">Whether the top reference is available.</param>
+        /// <param name="mode">The intra prediction mode.</param>
+        /// <param name="angleDelta">The signed directional-angle adjustment.</param>
+        /// <param name="residual">The contiguous source-minus-prediction destination.</param>
+        /// <param name="transformSize">The prediction dimensions.</param>
+        /// <param name="bitDepth">The coded sample bit depth.</param>
+        public static abstract void PrepareIntra(
+            Av1EncoderBlockWorkspace workspace,
+            Buffer2DRegion<TSample> source,
+            Point blockOrigin,
+            Span<TSample> prediction,
+            ReadOnlySpan<TSample> above,
+            ReadOnlySpan<TSample> left,
+            bool hasLeft,
+            bool hasAbove,
+            Av1PredictionMode mode,
+            int angleDelta,
+            Span<short> residual,
+            Av1TransformSize transformSize,
+            Av1BitDepth bitDepth);
+
         /// <summary>
         /// Builds one filter-intra prediction for reuse across transform candidates.
         /// </summary>
@@ -247,7 +278,8 @@ internal static partial class Av1IntraSuperblockEncoder
         /// <param name="blockOrigin">The transform-block origin in plane samples.</param>
         /// <param name="prediction">The contiguous prediction samples.</param>
         /// <param name="residual">The contiguous source-minus-prediction samples.</param>
-        /// <param name="reconstruction">The contiguous candidate reconstruction.</param>
+        /// <param name="reconstruction">The candidate reconstruction.</param>
+        /// <param name="reconstructionStride">The number of reconstruction samples between rows.</param>
         /// <param name="quantizedCoefficients">The candidate entropy-coding coefficients.</param>
         /// <param name="transformSize">The transform dimensions.</param>
         /// <param name="transformType">The compound transform applied to the residual.</param>
@@ -265,6 +297,7 @@ internal static partial class Av1IntraSuperblockEncoder
             ReadOnlySpan<TSample> prediction,
             ReadOnlySpan<short> residual,
             Span<TSample> reconstruction,
+            int reconstructionStride,
             Span<int> quantizedCoefficients,
             Av1TransformSize transformSize,
             Av1TransformType transformType,
@@ -704,6 +737,36 @@ internal static partial class Av1IntraSuperblockEncoder
                 plane,
                 ref state);
 
+        /// <inheritdoc/>
+        public static void PrepareIntra(
+            Av1EncoderBlockWorkspace workspace,
+            Buffer2DRegion<byte> source,
+            Point blockOrigin,
+            Span<byte> prediction,
+            ReadOnlySpan<byte> above,
+            ReadOnlySpan<byte> left,
+            bool hasLeft,
+            bool hasAbove,
+            Av1PredictionMode mode,
+            int angleDelta,
+            Span<short> residual,
+            Av1TransformSize transformSize,
+            Av1BitDepth bitDepth)
+            => Av1TransformBlockEncoder.PrepareIntraPrediction(
+                workspace,
+                Av1TransformBlockEncoder.GetPlaneSpan(source, blockOrigin),
+                source.Stride,
+                prediction,
+                transformSize.GetWidth(),
+                above,
+                left,
+                hasLeft,
+                hasAbove,
+                mode,
+                angleDelta,
+                residual,
+                transformSize);
+
         /// <inheritdoc/>
         public static void PrepareFilterIntra(
             Av1EncoderBlockWorkspace workspace,
@@ -782,6 +845,7 @@ internal static partial class Av1IntraSuperblockEncoder
             ReadOnlySpan<byte> prediction,
             ReadOnlySpan<short> residual,
             Span<byte> reconstruction,
+            int reconstructionStride,
             Span<int> quantizedCoefficients,
             Av1TransformSize transformSize,
             Av1TransformType transformType,
@@ -798,6 +862,7 @@ internal static partial class Av1IntraSuperblockEncoder
                 prediction,
                 residual,
                 reconstruction,
+                reconstructionStride,
                 quantizedCoefficients,
                 transformSize,
                 transformType,
@@ -1213,6 +1278,37 @@ internal static partial class Av1IntraSuperblockEncoder
                 bitDepth,
                 ref state);
 
+        /// <inheritdoc/>
+        public static void PrepareIntra(
+            Av1EncoderBlockWorkspace workspace,
+            Buffer2DRegion<ushort> source,
+            Point blockOrigin,
+            Span<ushort> prediction,
+            ReadOnlySpan<ushort> above,
+            ReadOnlySpan<ushort> left,
+            bool hasLeft,
+            bool hasAbove,
+            Av1PredictionMode mode,
+            int angleDelta,
+            Span<short> residual,
+            Av1TransformSize transformSize,
+            Av1BitDepth bitDepth)
+            => Av1TransformBlockEncoder.PrepareIntraPrediction(
+                workspace,
+                Av1TransformBlockEncoder.GetPlaneSpan(source, blockOrigin),
+                source.Stride,
+                prediction,
+                transformSize.GetWidth(),
+                above,
+                left,
+                hasLeft,
+                hasAbove,
+                mode,
+                angleDelta,
+                residual,
+                transformSize,
+                bitDepth);
+
         /// <inheritdoc/>
         public static void PrepareFilterIntra(
             Av1EncoderBlockWorkspace workspace,
@@ -1299,6 +1395,7 @@ internal static partial class Av1IntraSuperblockEncoder
             ReadOnlySpan<ushort> prediction,
             ReadOnlySpan<short> residual,
             Span<ushort> reconstruction,
+            int reconstructionStride,
             Span<int> quantizedCoefficients,
             Av1TransformSize transformSize,
             Av1TransformType transformType,
@@ -1315,6 +1412,7 @@ internal static partial class Av1IntraSuperblockEncoder
                 prediction,
                 residual,
                 reconstruction,
+                reconstructionStride,
                 quantizedCoefficients,
                 transformSize,
                 transformType,

src/ImageSharp/Formats/Heif/Av1/Pipeline/Av1IntraSuperblockEncoder.PaletteModeDecision.cs

@@ -28,6 +28,7 @@ internal static partial class Av1IntraSuperblockEncoder
             ushort tileIndex,
             Av1TransformSetType transformSetType,
             Av1TransformBlockContext blockContext,
+            int transformSizeRate,
             Span<TSample> candidateReconstruction,
             Span<int> candidateCoefficients,
             Span<int> retainedCoefficients,
@@ -138,6 +139,7 @@ internal static partial class Av1IntraSuperblockEncoder
                     blockOrigin,
                     transformSetType,
                     blockContext,
+                    transformSizeRate,
                     samples,
                     rows,
                     columns,
@@ -167,6 +169,7 @@ internal static partial class Av1IntraSuperblockEncoder
                     blockOrigin,
                     transformSetType,
                     blockContext,
+                    transformSizeRate,
                     samples,
                     rows,
                     columns,
@@ -205,6 +208,7 @@ internal static partial class Av1IntraSuperblockEncoder
                         blockOrigin,
                         transformSetType,
                         blockContext,
+                        transformSizeRate,
                         samples,
                         rows,
                         columns,
@@ -243,6 +247,7 @@ internal static partial class Av1IntraSuperblockEncoder
             Point blockOrigin,
             Av1TransformSetType transformSetType,
             Av1TransformBlockContext blockContext,
+            int transformSizeRate,
             ReadOnlySpan<short> samples,
             int rows,
             int columns,
@@ -345,6 +350,7 @@ internal static partial class Av1IntraSuperblockEncoder
                 Av1PredictionMode.DC,
                 0);
 
+            rate += transformSizeRate;
             rate += writer.GetPaletteYModeCost(true, blockSizeContext, neighborContext);
             rate += writer.GetPaletteSizeCost(paletteSize, blockSizeContext, Av1PlaneType.Y);
             rate += Av1SymbolEncoder.GetPaletteYColorCost(colorCache, paletteColors, bitDepth);
@@ -372,6 +378,7 @@ internal static partial class Av1IntraSuperblockEncoder
                     prediction,
                     residual,
                     candidateReconstruction,
+                    TransformSize.GetWidth(),
                     candidateCoefficients,
                     TransformSize,
                     transformType,

src/ImageSharp/Formats/Heif/Av1/Pipeline/Av1TransformBlockEncoder.cs

@@ -166,7 +166,8 @@ internal static class Av1TransformBlockEncoder
     /// <param name="blockOrigin">The block origin in plane samples.</param>
     /// <param name="prediction">The contiguous prediction samples.</param>
     /// <param name="residual">The contiguous source-minus-prediction samples.</param>
-    /// <param name="reconstruction">The contiguous candidate reconstruction.</param>
+    /// <param name="reconstruction">The candidate reconstruction.</param>
+    /// <param name="reconstructionStride">The number of reconstruction samples between rows.</param>
     /// <param name="quantizedCoefficients">The candidate entropy-coding coefficients.</param>
     /// <param name="transformSize">The selected transform dimensions.</param>
     /// <param name="transformType">The selected compound transform type.</param>
@@ -183,6 +184,7 @@ internal static class Av1TransformBlockEncoder
         ReadOnlySpan<byte> prediction,
         ReadOnlySpan<short> residual,
         Span<byte> reconstruction,
+        int reconstructionStride,
         Span<int> quantizedCoefficients,
         Av1TransformSize transformSize,
         Av1TransformType transformType,
@@ -198,7 +200,12 @@ internal static class Av1TransformBlockEncoder
         ReadOnlySpan<byte> sourceSamples = GetPlaneSpan(source, blockOrigin);
 
         // Each transform trial mutates reconstruction and residual scratch, so restore both prepared inputs.
-        prediction[..sampleCount].CopyTo(reconstruction);
+        // Row copies preserve a larger candidate surface without materializing a second compact block.
+        for (int row = 0; row < height; row++)
+        {
+            prediction.Slice(row * width, width).CopyTo(reconstruction.Slice(row * reconstructionStride, width));
+        }
+
         residual[..sampleCount].CopyTo(workspace.Residual);
         EncodeLossy(
             workspace,
@@ -216,7 +223,7 @@ internal static class Av1TransformBlockEncoder
             Av1InverseTransformer.Reconstruct8Bit(
                 workspace.DequantizedCoefficients,
                 reconstruction,
-                width,
+                reconstructionStride,
                 transformSize,
                 transformType,
                 (int)plane,
@@ -229,7 +236,7 @@ internal static class Av1TransformBlockEncoder
             sourceSamples,
             source.Stride,
             reconstruction,
-            width,
+            reconstructionStride,
             workspace.Residual,
             width,
             width,
@@ -490,7 +497,8 @@ internal static class Av1TransformBlockEncoder
     /// <param name="blockOrigin">The block origin in plane samples.</param>
     /// <param name="prediction">The contiguous prediction samples.</param>
     /// <param name="residual">The contiguous source-minus-prediction samples.</param>
-    /// <param name="reconstruction">The contiguous candidate reconstruction.</param>
+    /// <param name="reconstruction">The candidate reconstruction.</param>
+    /// <param name="reconstructionStride">The number of reconstruction samples between rows.</param>
     /// <param name="quantizedCoefficients">The candidate entropy-coding coefficients.</param>
     /// <param name="transformSize">The selected transform dimensions.</param>
     /// <param name="transformType">The selected compound transform type.</param>
@@ -508,6 +516,7 @@ internal static class Av1TransformBlockEncoder
         ReadOnlySpan<ushort> prediction,
         ReadOnlySpan<short> residual,
         Span<ushort> reconstruction,
+        int reconstructionStride,
         Span<int> quantizedCoefficients,
         Av1TransformSize transformSize,
         Av1TransformType transformType,
@@ -524,7 +533,12 @@ internal static class Av1TransformBlockEncoder
         ReadOnlySpan<ushort> sourceSamples = GetPlaneSpan(source, blockOrigin);
 
         // Each transform trial mutates reconstruction and residual scratch, so restore both prepared inputs.
-        prediction[..sampleCount].CopyTo(reconstruction);
+        // Row copies preserve a larger candidate surface without materializing a second compact block.
+        for (int row = 0; row < height; row++)
+        {
+            prediction.Slice(row * width, width).CopyTo(reconstruction.Slice(row * reconstructionStride, width));
+        }
+
         residual[..sampleCount].CopyTo(workspace.Residual);
         EncodeLossy(
             workspace,
@@ -542,7 +556,7 @@ internal static class Av1TransformBlockEncoder
             Av1InverseTransformer.ReconstructHighBitDepth(
                 workspace.DequantizedCoefficients,
                 MemoryMarshal.Cast<ushort, short>(reconstruction),
-                width,
+                reconstructionStride,
                 transformSize,
                 transformType,
                 (int)plane,
@@ -556,7 +570,7 @@ internal static class Av1TransformBlockEncoder
             sourceSamples,
             source.Stride,
             reconstruction,
-            width,
+            reconstructionStride,
             workspace.Residual,
             width,
             width,
@@ -678,65 +692,54 @@ internal static class Av1TransformBlockEncoder
     }
 
     /// <summary>
-    /// Encodes and reconstructs one eight-bit lossy intra block.
+    /// Builds an eight-bit intra prediction and its compact source residual.
     /// </summary>
-    /// <param name="workspace">The reusable residual, coefficient, and transform storage.</param>
+    /// <param name="workspace">The reusable prediction scratch.</param>
     /// <param name="source">The source samples.</param>
     /// <param name="sourceStride">The number of source samples between rows.</param>
-    /// <param name="reconstruction">The reconstructed frame samples and prediction destination.</param>
+    /// <param name="prediction">The prediction destination.</param>
-    /// <param name="reconstructionStride">The number of reconstruction samples between rows.</param>
+    /// <param name="predictionStride">The number of prediction samples between rows.</param>
     /// <param name="above">The contiguous top reference samples, with prefix storage for the shared corner.</param>
     /// <param name="left">The contiguous left reference samples.</param>
     /// <param name="hasLeft">Whether the left reference is available.</param>
     /// <param name="hasAbove">Whether the top reference is available.</param>
     /// <param name="mode">The intra prediction mode.</param>
     /// <param name="angleDelta">The signed directional-angle adjustment.</param>
-    /// <param name="quantizedCoefficients">The retained entropy-coding coefficients.</param>
+    /// <param name="residual">The compact source-minus-prediction destination.</param>
-    /// <param name="transformSize">The selected transform dimensions.</param>
+    /// <param name="transformSize">The prediction dimensions.</param>
-    /// <param name="transformType">The selected compound transform type.</param>
+    public static void PrepareIntraPrediction(
-    /// <param name="qIndex">The segment quantizer index.</param>
-    /// <param name="dcDeltaQ">The plane DC quantizer adjustment.</param>
-    /// <param name="acDeltaQ">The plane AC quantizer adjustment.</param>
-    /// <param name="plane">The component plane containing the block.</param>
-    /// <param name="state">The retained transform type and end-of-block syntax.</param>
-    private static void EncodeIntraLossyContiguous(
         Av1EncoderBlockWorkspace workspace,
         ReadOnlySpan<byte> source,
         int sourceStride,
-        Span<byte> reconstruction,
+        Span<byte> prediction,
-        int reconstructionStride,
+        int predictionStride,
         ReadOnlySpan<byte> above,
         ReadOnlySpan<byte> left,
         bool hasLeft,
         bool hasAbove,
         Av1PredictionMode mode,
         int angleDelta,
-        Span<int> quantizedCoefficients,
+        Span<short> residual,
-        Av1TransformSize transformSize,
+        Av1TransformSize transformSize)
-        Av1TransformType transformType,
-        int qIndex,
-        int dcDeltaQ,
-        int acDeltaQ,
-        Av1Plane plane,
-        ref Av1EncoderTransformBlockState state)
     {
         int width = transformSize.GetWidth();
         int height = transformSize.GetHeight();
 
-        // Prediction and subtraction stay in their SIMD-first operators while this method owns the required block-stage ordering.
+        // Prediction remains in the specialized SIMD-first kernels. This boundary only shares the prepared
+        // samples and residual across transform trials that differ in transform size or type.
         if (mode == Av1PredictionMode.DC)
         {
-            Av1DcIntraPredictor.Predict(hasLeft, hasAbove, reconstruction, reconstructionStride, above, left, width, height);
+            Av1DcIntraPredictor.Predict(hasLeft, hasAbove, prediction, predictionStride, above, left, width, height);
         }
         else if (mode.IsDirectional())
         {
-            // The current encoder disables intra-edge filtering in sequence syntax. Reusing transform workspace for
+            // The current encoder disables intra-edge filtering in sequence syntax. Zone-three transposition
-            // zone-three transposition keeps directional prediction allocation-free before the transform overwrites it.
+            // borrows transform scratch because prediction completes before forward transformation starts.
             Span<byte> directionalScratch = MemoryMarshal.AsBytes(workspace.TransformWorkspace)[..(width * height)];
 
             Av1DirectionalIntraPredictor.Predict(
-                reconstruction,
+                prediction,
-                reconstructionStride,
+                predictionStride,
                 transformSize,
                 above,
                 left,
@@ -748,10 +751,163 @@ internal static class Av1TransformBlockEncoder
         else
         {
             Av1NonDirectionalIntraPredictorBase.GetPredictor(mode)
-                .Predict(reconstruction, reconstructionStride, above, left, width, height);
+                .Predict(prediction, predictionStride, above, left, width, height);
         }
 
-        Av1ResidualBuilder.Subtract(source, sourceStride, reconstruction, reconstructionStride, workspace.Residual, width, width, height);
+        Av1ResidualBuilder.Subtract(
+            source,
+            sourceStride,
+            prediction,
+            predictionStride,
+            residual,
+            width,
+            width,
+            height);
+    }
+
+    /// <summary>
+    /// Builds a high-bit-depth intra prediction and its compact source residual.
+    /// </summary>
+    /// <param name="workspace">The reusable prediction scratch.</param>
+    /// <param name="source">The source samples.</param>
+    /// <param name="sourceStride">The number of source samples between rows.</param>
+    /// <param name="prediction">The prediction destination.</param>
+    /// <param name="predictionStride">The number of prediction samples between rows.</param>
+    /// <param name="above">The contiguous top reference samples, with prefix storage for the shared corner.</param>
+    /// <param name="left">The contiguous left reference samples.</param>
+    /// <param name="hasLeft">Whether the left reference is available.</param>
+    /// <param name="hasAbove">Whether the top reference is available.</param>
+    /// <param name="mode">The intra prediction mode.</param>
+    /// <param name="angleDelta">The signed directional-angle adjustment.</param>
+    /// <param name="residual">The compact source-minus-prediction destination.</param>
+    /// <param name="transformSize">The prediction dimensions.</param>
+    /// <param name="bitDepth">The coded sample bit depth.</param>
+    public static void PrepareIntraPrediction(
+        Av1EncoderBlockWorkspace workspace,
+        ReadOnlySpan<ushort> source,
+        int sourceStride,
+        Span<ushort> prediction,
+        int predictionStride,
+        ReadOnlySpan<ushort> above,
+        ReadOnlySpan<ushort> left,
+        bool hasLeft,
+        bool hasAbove,
+        Av1PredictionMode mode,
+        int angleDelta,
+        Span<short> residual,
+        Av1TransformSize transformSize,
+        Av1BitDepth bitDepth)
+    {
+        int width = transformSize.GetWidth();
+        int height = transformSize.GetHeight();
+
+        // Valid high-bit-depth samples remain below the sign bit, so the predictor kernels can share
+        // the unsigned frame storage with their signed transform-domain implementation.
+        Span<short> signedPrediction = MemoryMarshal.Cast<ushort, short>(prediction);
+        ReadOnlySpan<short> signedAbove = MemoryMarshal.Cast<ushort, short>(above);
+        ReadOnlySpan<short> signedLeft = MemoryMarshal.Cast<ushort, short>(left);
+        if (mode == Av1PredictionMode.DC)
+        {
+            Av1DcIntraPredictor.Predict(
+                hasLeft,
+                hasAbove,
+                signedPrediction,
+                predictionStride,
+                signedAbove,
+                signedLeft,
+                width,
+                height,
+                bitDepth.GetBitCount());
+        }
+        else if (mode.IsDirectional())
+        {
+            Span<short> directionalScratch = MemoryMarshal.Cast<int, short>(workspace.TransformWorkspace)[..(width * height)];
+
+            Av1DirectionalIntraPredictor.Predict(
+                signedPrediction,
+                predictionStride,
+                transformSize,
+                signedAbove,
+                signedLeft,
+                false,
+                false,
+                mode.ToAngle() + (angleDelta * Av1Constants.AngleStep),
+                directionalScratch);
+        }
+        else
+        {
+            Av1NonDirectionalIntraPredictorBase.GetPredictor(mode)
+                .Predict(signedPrediction, predictionStride, signedAbove, signedLeft, width, height);
+        }
+
+        Av1ResidualBuilder.Subtract(
+            source,
+            sourceStride,
+            prediction,
+            predictionStride,
+            residual,
+            width,
+            width,
+            height);
+    }
+
+    /// <summary>
+    /// Encodes and reconstructs one eight-bit lossy intra block.
+    /// </summary>
+    /// <param name="workspace">The reusable residual, coefficient, and transform storage.</param>
+    /// <param name="source">The source samples.</param>
+    /// <param name="sourceStride">The number of source samples between rows.</param>
+    /// <param name="reconstruction">The reconstructed frame samples and prediction destination.</param>
+    /// <param name="reconstructionStride">The number of reconstruction samples between rows.</param>
+    /// <param name="above">The contiguous top reference samples, with prefix storage for the shared corner.</param>
+    /// <param name="left">The contiguous left reference samples.</param>
+    /// <param name="hasLeft">Whether the left reference is available.</param>
+    /// <param name="hasAbove">Whether the top reference is available.</param>
+    /// <param name="mode">The intra prediction mode.</param>
+    /// <param name="angleDelta">The signed directional-angle adjustment.</param>
+    /// <param name="quantizedCoefficients">The retained entropy-coding coefficients.</param>
+    /// <param name="transformSize">The selected transform dimensions.</param>
+    /// <param name="transformType">The selected compound transform type.</param>
+    /// <param name="qIndex">The segment quantizer index.</param>
+    /// <param name="dcDeltaQ">The plane DC quantizer adjustment.</param>
+    /// <param name="acDeltaQ">The plane AC quantizer adjustment.</param>
+    /// <param name="plane">The component plane containing the block.</param>
+    /// <param name="state">The retained transform type and end-of-block syntax.</param>
+    private static void EncodeIntraLossyContiguous(
+        Av1EncoderBlockWorkspace workspace,
+        ReadOnlySpan<byte> source,
+        int sourceStride,
+        Span<byte> reconstruction,
+        int reconstructionStride,
+        ReadOnlySpan<byte> above,
+        ReadOnlySpan<byte> left,
+        bool hasLeft,
+        bool hasAbove,
+        Av1PredictionMode mode,
+        int angleDelta,
+        Span<int> quantizedCoefficients,
+        Av1TransformSize transformSize,
+        Av1TransformType transformType,
+        int qIndex,
+        int dcDeltaQ,
+        int acDeltaQ,
+        Av1Plane plane,
+        ref Av1EncoderTransformBlockState state)
+    {
+        PrepareIntraPrediction(
+            workspace,
+            source,
+            sourceStride,
+            reconstruction,
+            reconstructionStride,
+            above,
+            left,
+            hasLeft,
+            hasAbove,
+            mode,
+            angleDelta,
+            workspace.Residual,
+            transformSize);
 
         EncodeLossy(
             workspace,
@@ -825,56 +981,21 @@ internal static class Av1TransformBlockEncoder
         Av1BitDepth bitDepth,
         ref Av1EncoderTransformBlockState state)
     {
-        int width = transformSize.GetWidth();
+        PrepareIntraPrediction(
-        int height = transformSize.GetHeight();
+            workspace,
-
-        // Valid high-bit-depth samples remain below the sign bit, so signed transform lanes can share the unsigned frame storage.
-        Span<short> signedReconstruction = MemoryMarshal.Cast<ushort, short>(reconstruction);
-        ReadOnlySpan<short> signedAbove = MemoryMarshal.Cast<ushort, short>(above);
-        ReadOnlySpan<short> signedLeft = MemoryMarshal.Cast<ushort, short>(left);
-        if (mode == Av1PredictionMode.DC)
-        {
-            Av1DcIntraPredictor.Predict(
-                hasLeft,
-                hasAbove,
-                signedReconstruction,
-                reconstructionStride,
-                signedAbove,
-                signedLeft,
-                width,
-                height,
-                bitDepth.GetBitCount());
-        }
-        else if (mode.IsDirectional())
-        {
-            Span<short> directionalScratch = MemoryMarshal.Cast<int, short>(workspace.TransformWorkspace)[..(width * height)];
-
-            Av1DirectionalIntraPredictor.Predict(
-                signedReconstruction,
-                reconstructionStride,
-                transformSize,
-                signedAbove,
-                signedLeft,
-                false,
-                false,
-                mode.ToAngle() + (angleDelta * Av1Constants.AngleStep),
-                directionalScratch);
-        }
-        else
-        {
-            Av1NonDirectionalIntraPredictorBase.GetPredictor(mode)
-                .Predict(signedReconstruction, reconstructionStride, signedAbove, signedLeft, width, height);
-        }
-
-        Av1ResidualBuilder.Subtract(
             source,
             sourceStride,
             reconstruction,
             reconstructionStride,
+            above,
+            left,
+            hasLeft,
+            hasAbove,
+            mode,
+            angleDelta,
             workspace.Residual,
-            width,
+            transformSize,
-            width,
+            bitDepth);
-            height);
 
         EncodeLossy(
             workspace,
@@ -892,7 +1013,7 @@ internal static class Av1TransformBlockEncoder
             // Reconstructing the quantized result makes later predictions use exactly the samples a decoder will reproduce.
             Av1InverseTransformer.ReconstructHighBitDepth(
                 workspace.DequantizedCoefficients,
-                signedReconstruction,
+                MemoryMarshal.Cast<ushort, short>(reconstruction),
                 reconstructionStride,
                 transformSize,
                 transformType,

src/ImageSharp/Formats/Heif/Av1/Tiling/Av1TileWriter.cs

@@ -910,6 +910,50 @@ internal partial class Av1TileWriter
         UpdateNeighbors(pcs, entropyCodingContext, blockOrigin, ref blk_ptr, tile_idx, blockSize);
     }
 
+    /// <summary>
+    /// Derives the uniform intra transform-size context from the current above and left edges.
+    /// </summary>
+    /// <param name="transformContexts">The retained transform widths and heights.</param>
+    /// <param name="macroBlock">The reusable macroblock edge and neighbor state.</param>
+    /// <param name="blockOrigin">The block origin in samples.</param>
+    /// <param name="blockSize">The block size defining the maximum transform.</param>
+    /// <returns>The uniform transform-size context.</returns>
+    public static int GetTransformSizeContext(
+        Av1NeighborArrayUnit<byte> transformContexts,
+        Av1MacroBlockD macroBlock,
+        Point blockOrigin,
+        Av1BlockSize blockSize)
+    {
+        Av1TransformSize maximumTransformSize = blockSize.GetMaximumTransformSize();
+        int above = transformContexts.Top[transformContexts.GetTopIndex(blockOrigin)] >= maximumTransformSize.GetWidth() ? 1 : 0;
+        int left = transformContexts.Left[transformContexts.GetLeftIndex(blockOrigin)] >= maximumTransformSize.GetHeight() ? 1 : 0;
+
+        // Inter neighbors contribute their coding-block extent, not their residual-transform extent.
+        if (macroBlock.IsUpAvailable)
+        {
+            ref Av1MacroBlockModeInfo aboveModeInfo =
+                ref macroBlock.GetRelativeModeInfo(-macroBlock.ModeInfoStride);
+
+            if (aboveModeInfo.Block.UseIntraBlockCopy)
+            {
+                above = aboveModeInfo.Block.BlockSize.GetWidth() >= maximumTransformSize.GetWidth() ? 1 : 0;
+            }
+        }
+
+        if (macroBlock.IsLeftAvailable)
+        {
+            ref Av1MacroBlockModeInfo leftModeInfo = ref macroBlock.GetRelativeModeInfo(-1);
+            if (leftModeInfo.Block.UseIntraBlockCopy)
+            {
+                left = leftModeInfo.Block.BlockSize.GetHeight() >= maximumTransformSize.GetHeight() ? 1 : 0;
+            }
+        }
+
+        return macroBlock.IsUpAvailable
+            ? macroBlock.IsLeftAvailable ? above + left : above
+            : macroBlock.IsLeftAvailable ? left : 0;
+    }
+
     /// <summary>
     /// Writes or derives the block transform size and publishes its edge contexts.
     /// </summary>
@@ -931,28 +975,42 @@ internal partial class Av1TileWriter
     {
         ObuFrameHeader frameHeader = pcs.Parent.FrameHeader;
         bool isLossless = frameHeader.LosslessArray[macroBlockModeInfo.Block.SegmentId];
-        bool writesTransformSize = !isLossless &&
+        bool isInter = macroBlockModeInfo.Block.UseIntraBlockCopy;
+        bool writesUniformTransformSize = !isLossless &&
             frameHeader.TransformMode == Av1TransformMode.Select &&
+            !isInter &&
             blockSize > Av1BlockSize.Block4x4;
+        bool writesVariableTransformSize = !isLossless &&
+            frameHeader.TransformMode == Av1TransformMode.Select &&
+            isInter &&
+            !macroBlockModeInfo.Block.Skip;
+
         Av1TransformSize transformSize = isLossless
             ? Av1TransformSize.Size4x4
-            : writesTransformSize
+            : writesUniformTransformSize || writesVariableTransformSize
                 ? macroBlockModeInfo.Block.TransformSize
                 : blockSize.GetMaximumTransformSize();
 
         macroBlockModeInfo.Block.TransformSize = transformSize;
         Av1NeighborArrayUnit<byte> transformContexts = pcs.TransformFunctionContexts[tileIndex];
-        if (writesTransformSize)
+        if (writesUniformTransformSize)
         {
-            Av1TransformSize maximumTransformSize = blockSize.GetMaximumTransformSize();
+            int context = GetTransformSizeContext(transformContexts, macroBlock, blockOrigin, blockSize);
-            int above = transformContexts.Top[transformContexts.GetTopIndex(blockOrigin)] >= maximumTransformSize.GetWidth() ? 1 : 0;
-            int left = transformContexts.Left[transformContexts.GetLeftIndex(blockOrigin)] >= maximumTransformSize.GetHeight() ? 1 : 0;
-            int context = macroBlock.IsUpAvailable
-                ? macroBlock.IsLeftAvailable ? above + left : above
-                : macroBlock.IsLeftAvailable ? left : 0;
-
             writer.WriteTransformSize(blockSize, transformSize, context);
         }
+        else if (writesVariableTransformSize)
+        {
+            int topIndex = transformContexts.GetTopIndex(blockOrigin);
+            int leftIndex = transformContexts.GetLeftIndex(blockOrigin);
+            int context = Av1SymbolContextHelper.GetTransformPartitionContext(
+                transformContexts.Top[topIndex],
+                transformContexts.Left[leftIndex],
+                blockSize,
+                blockSize.GetMaximumTransformSize());
+
+            // Intra-block copy currently retains the maximum transform, so its variable-transform tree has one unsplit root.
+            writer.WriteTransformPartition(false, context);
+        }
 
         Size blockDimensions = new(blockSize.GetWidth(), blockSize.GetHeight());
 
@@ -1863,6 +1921,26 @@ internal partial class Av1TileWriter
         int transformBlockHeight = transformSize.Get4x4HighCount();
         ReadOnlySpan<byte> topContexts = dcSignLevelCoefficientNeighborArray.Top.Slice(topIndex, transformBlockWidth);
         ReadOnlySpan<byte> leftContexts = dcSignLevelCoefficientNeighborArray.Left.Slice(leftIndex, transformBlockHeight);
+
+        return GetTransformBlockContexts(plane, topContexts, leftContexts, planeBlockSize, transformSize);
+    }
+
+    /// <summary>
+    /// Derives coefficient skip and DC-sign contexts from explicit transform-edge contexts.
+    /// </summary>
+    /// <param name="plane">The luma or chroma component class.</param>
+    /// <param name="topContexts">The packed contexts immediately above the transform block.</param>
+    /// <param name="leftContexts">The packed contexts immediately left of the transform block.</param>
+    /// <param name="planeBlockSize">The containing block size on the target plane.</param>
+    /// <param name="transformSize">The transform size.</param>
+    /// <returns>The coefficient skip and DC-sign contexts selected by both transform edges.</returns>
+    public static Av1TransformBlockContext GetTransformBlockContexts(
+        Av1ComponentType plane,
+        ReadOnlySpan<byte> topContexts,
+        ReadOnlySpan<byte> leftContexts,
+        Av1BlockSize planeBlockSize,
+        Av1TransformSize transformSize)
+    {
         int dcSign = 0;
         int top = 0;
         int left = 0;

tests/ImageSharp.Tests/Formats/Heif/Av1/Av1CoefficientsEntropyTests.cs

@@ -594,6 +594,53 @@ public class Av1CoefficientsEntropyTests
         }
     }
 
+    [Fact]
+    public void TransformSizeContextUsesIntraBlockCopyNeighborExtents()
+    {
+        Av1PictureControlSet picture = CreateEncoderPicture(16, 16);
+        Point blockOrigin = new(16, 16);
+        Av1MacroBlockD macroBlock = new()
+        {
+            Tile = new Av1TileInfo(0, 0, picture.Parent.FrameHeader),
+            IsUpAvailable = true,
+            IsLeftAvailable = true
+        };
+
+        int modeInfoIndex =
+            ((blockOrigin.Y >> Av1Constants.ModeInfoSizeLog2) * picture.ModeInfoStride) +
+            (blockOrigin.X >> Av1Constants.ModeInfoSizeLog2);
+
+        macroBlock.ModeInfoStride = picture.ModeInfoStride;
+        macroBlock.SetModeInfoGrid(picture.ModeInfoGrid, picture.ModeInfoAllocation, modeInfoIndex);
+
+        ref Av1MacroBlockModeInfo aboveModeInfo = ref macroBlock.GetRelativeModeInfo(-macroBlock.ModeInfoStride);
+        aboveModeInfo.Block.BlockSize = Av1BlockSize.Block16x8;
+        aboveModeInfo.Block.UseIntraBlockCopy = true;
+        ref Av1MacroBlockModeInfo leftModeInfo = ref macroBlock.GetRelativeModeInfo(-1);
+        leftModeInfo.Block.BlockSize = Av1BlockSize.Block8x16;
+        leftModeInfo.Block.UseIntraBlockCopy = true;
+
+        using Av1NeighborArrayUnit<byte> transforms = new(
+            Configuration.Default,
+            leftSize: 64,
+            topSize: 64)
+        {
+            GranularityNormalLog2 = Av1Constants.ModeInfoSizeLog2
+        };
+
+        // Residual contexts report 8x8, but libaom derives 16x16 availability from the IBC coding blocks.
+        transforms.Top[transforms.GetTopIndex(blockOrigin)] = 8;
+        transforms.Left[transforms.GetLeftIndex(blockOrigin)] = 8;
+
+        Assert.Equal(
+            2,
+            Av1TileWriter.GetTransformSizeContext(
+                transforms,
+                macroBlock,
+                blockOrigin,
+                Av1BlockSize.Block16x16));
+    }
+
     [Fact]
     public void SelectedTransformSizeRoundTripsAndPublishesRectangularEdgeContexts()
     {
@@ -611,6 +658,14 @@ public class Av1CoefficientsEntropyTests
             IsLeftAvailable = true
         };
 
+        int modeInfoIndex =
+            ((blockOrigin.Y >> Av1Constants.ModeInfoSizeLog2) * picture.ModeInfoStride) +
+            (blockOrigin.X >> Av1Constants.ModeInfoSizeLog2);
+
+        // Uniform-size context substitutes coding-block extents for inter neighbors, so mirror production mode-info setup.
+        macroBlock.ModeInfoStride = picture.ModeInfoStride;
+        macroBlock.SetModeInfoGrid(picture.ModeInfoGrid, picture.ModeInfoAllocation, modeInfoIndex);
+
         using Av1NeighborArrayUnit<byte> transforms = new(
             Configuration.Default,
             leftSize: 64,

tests/ImageSharp.Tests/Formats/Heif/Av1/Av1EncoderFrameTests.cs

@@ -7,6 +7,7 @@ using SixLabors.ImageSharp.Formats.Heif.Av1.OpenBitstreamUnit;
 using SixLabors.ImageSharp.Formats.Heif.Av1.Pipeline;
 using SixLabors.ImageSharp.Formats.Heif.Av1.Prediction;
 using SixLabors.ImageSharp.Formats.Heif.Av1.Tiling;
+using SixLabors.ImageSharp.Formats.Heif.Av1.Transform;
 using SixLabors.ImageSharp.Formats.Heif.Components;
 using SixLabors.ImageSharp.Formats.Heif.Components.Alpha;
 using SixLabors.ImageSharp.Memory;
@@ -488,13 +489,18 @@ public class Av1EncoderFrameTests
     }
 
     [Theory]
-    [InlineData(0, false, false)]
+    [InlineData(0, false, false, false)]
-    [InlineData(1, false, false)]
+    [InlineData(1, false, false, false)]
-    [InlineData(2, false, false)]
+    [InlineData(2, false, false, false)]
-    [InlineData(3, false, false)]
+    [InlineData(3, false, false, false)]
-    [InlineData(4, true, false)]
+    [InlineData(4, true, false, false)]
-    [InlineData(5, true, true)]
+    [InlineData(5, true, true, false)]
-    public void EncodeEffortControlsSearchFeatures(int effort, bool enableFilterIntra, bool enableScreenContentTools)
+    [InlineData(6, true, true, true)]
+    public void EncodeEffortControlsSearchFeatures(
+        int effort,
+        bool enableFilterIntra,
+        bool enableScreenContentTools,
+        bool selectTransformSize)
     {
         const int width = 16;
         const int height = 16;
@@ -529,6 +535,9 @@ public class Av1EncoderFrameTests
         Assert.Equal(enableFilterIntra, sequenceHeader.EnableFilterIntra);
         Assert.Equal(enableScreenContentTools, frameHeader.AllowScreenContentTools);
         Assert.Equal(enableScreenContentTools, frameHeader.AllowIntraBlockCopy);
+        Assert.Equal(
+            selectTransformSize ? Av1TransformMode.Select : Av1TransformMode.Largest,
+            frameHeader.TransformMode);
         Assert.Equal(new Size(width, height), decoded.Size);
 
         int modeCount = 0;
@@ -573,7 +582,64 @@ public class Av1EncoderFrameTests
     }
 
     [Fact]
-    public void EncodeSelectsIntraBlockCopyForRepeatedScreenContent()
+    public void EncodeEffortSixSelectsFourByFourLumaTransforms()
+    {
+        const int Width = 16;
+        const int Height = 16;
+        using Image<Rgba32> source = new(Width, Height);
+        for (int row = 0; row < Height; row++)
+        {
+            Span<Rgba32> pixels = source.Frames.RootFrame.PixelBuffer.DangerousGetRowSpan(row);
+            for (int column = 0; column < Width; column++)
+            {
+                byte value = (byte)(16 + ((((row >> 2) * 4) + (column >> 2)) * 14));
+                pixels[column] = new Rgba32(value, value, value);
+            }
+        }
+
+        ObuColorConfig colorConfig = CreateColorConfig(Av1BitDepth.EightBit, Av1ColorFormat.Yuv400);
+        using MemoryStream stream = new();
+        _ = Av1FrameEncoder.Encode(
+            Configuration.Default,
+            source.Frames.RootFrame,
+            stream,
+            colorConfig,
+            qIndex: 37,
+            effort: 6);
+
+        byte[] payload = stream.ToArray();
+        using Av1Decoder decoder = new(Configuration.Default);
+        using Image<L8> decoded = decoder.Decode<L8>(payload);
+        Assert.NotNull(decoder.FrameHeader);
+        Assert.Equal(Av1TransformMode.Select, decoder.FrameHeader.TransformMode);
+        Assert.NotNull(decoder.FrameInfo);
+        bool foundSplitTransform = false;
+        foreach (Av1BlockModeInfo modeInfo in decoder.FrameInfo.GetSuperblock(Point.Empty).GetModeInfos())
+        {
+            foundSplitTransform |= modeInfo.GetTransformUnitCount(Av1Plane.Y) == 4;
+        }
+
+        Assert.True(foundSplitTransform);
+        Assert.Equal(new Size(Width, Height), decoded.Size);
+
+        string outputDirectory = Path.Combine(
+            TestEnvironment.ActualOutputDirectoryFullPath,
+            "Formats",
+            "Heif",
+            "Av1");
+
+        Directory.CreateDirectory(outputDirectory);
+        File.WriteAllBytes(
+            Path.Combine(outputDirectory, "encoder-frame-16x16-8b-400-transform-size-select.obu"),
+            payload);
+    }
+
+    [Theory]
+    [InlineData(5, false)]
+    [InlineData(6, true)]
+    public void EncodeSelectsIntraBlockCopyForRepeatedScreenContent(
+        int effort,
+        bool selectTransformSize)
     {
         const int Width = 328;
         const int Height = 16;
@@ -599,7 +665,7 @@ public class Av1EncoderFrameTests
             stream,
             colorConfig,
             qIndex: 37,
-            effort: 5);
+            effort);
 
         byte[] payload = stream.ToArray();
         using Av1Decoder decoder = new(Configuration.Default);
@@ -607,6 +673,9 @@ public class Av1EncoderFrameTests
         Assert.NotNull(decoder.FrameHeader);
         Assert.True(decoder.FrameHeader.AllowScreenContentTools);
         Assert.True(decoder.FrameHeader.AllowIntraBlockCopy);
+        Assert.Equal(
+            selectTransformSize ? Av1TransformMode.Select : Av1TransformMode.Largest,
+            decoder.FrameHeader.TransformMode);
         Assert.NotNull(decoder.FrameInfo);
         Av1SuperblockInfo targetSuperblock = decoder.FrameInfo.GetSuperblock(new Point(5, 0));
         bool usesIntraBlockCopy = false;
@@ -625,7 +694,11 @@ public class Av1EncoderFrameTests
             "Av1");
 
         Directory.CreateDirectory(outputDirectory);
-        File.WriteAllBytes(Path.Combine(outputDirectory, "encoder-frame-328x16-8b-444-intrabc.obu"), payload);
+        string fileName = effort == 5
+            ? "encoder-frame-328x16-8b-444-intrabc.obu"
+            : "encoder-frame-328x16-8b-444-intrabc-effort-6.obu";
+
+        File.WriteAllBytes(Path.Combine(outputDirectory, fileName), payload);
     }
 
     [Fact]

tests/ImageSharp.Tests/Formats/Heif/Av1/Av1EntropyTests.cs

@@ -134,6 +134,15 @@ public class Av1EntropyTests
             Av1ProbabilityCost.GetSymbolCost(transformSize, 1),
             encoder.GetTransformSizeCost(BlockSize, Av1TransformSize.Size4x4, SkipContext));
 
+        Av1Distribution transformPartition = Av1DefaultDistributions.TransformPartition[SkipContext];
+        Assert.Equal(
+            Av1ProbabilityCost.GetSymbolCost(transformPartition, 0),
+            encoder.GetTransformPartitionCost(false, SkipContext));
+
+        Assert.Equal(
+            Av1ProbabilityCost.GetSymbolCost(transformPartition, 1),
+            encoder.GetTransformPartitionCost(true, SkipContext));
+
         Av1Distribution transformSkip = Av1DefaultDistributions
             .GetTransformBlockSkip(BaseQIndex)[(int)TransformSize][SkipContext];
 

tests/ImageSharp.Tests/Formats/Heif/Av1/Av1SymbolContextTests.cs

@@ -76,6 +76,47 @@ public class Av1SymbolContextTests
         Assert.Equal((Av1TransformType)expectedValue, actual);
     }
 
+    [Theory]
+    [InlineData(0, 0, 0)]
+    [InlineData(-1, 4, 12)]
+    [InlineData(1, 4, 20)]
+    public void CoefficientContextMatchesCurrentLibaom(int dcCoefficient, ushort endOfBlock, byte expected)
+    {
+        Span<int> coefficients = stackalloc int[16];
+        coefficients.Fill(1);
+        coefficients[0] = dcCoefficient;
+
+        byte actual = Av1SymbolContextHelper.GetCoefficientContext(
+            coefficients,
+            Av1TransformSize.Size4x4,
+            Av1TransformType.DctDct,
+            endOfBlock);
+
+        Assert.Equal(expected, actual);
+    }
+
+    [Theory]
+    [InlineData(8, 8, (int)Av1BlockSize.Block8x8, (int)Av1TransformSize.Size8x8, 18)]
+    [InlineData(4, 8, (int)Av1BlockSize.Block8x8, (int)Av1TransformSize.Size8x8, 19)]
+    [InlineData(4, 4, (int)Av1BlockSize.Block8x8, (int)Av1TransformSize.Size8x8, 20)]
+    [InlineData(4, 4, (int)Av1BlockSize.Block16x16, (int)Av1TransformSize.Size8x8, 17)]
+    [InlineData(0, 0, (int)Av1BlockSize.Block8x8, (int)Av1TransformSize.Size4x4, 0)]
+    public void TransformPartitionContextMatchesCurrentLibaom(
+        byte aboveWidth,
+        byte leftHeight,
+        int blockSizeValue,
+        int transformSizeValue,
+        int expected)
+    {
+        int actual = Av1SymbolContextHelper.GetTransformPartitionContext(
+            aboveWidth,
+            leftHeight,
+            (Av1BlockSize)blockSizeValue,
+            (Av1TransformSize)transformSizeValue);
+
+        Assert.Equal(expected, actual);
+    }
+
     public static TheoryData<int, int, int> GetLowLevelContextEndOfBlockData()
     {
         TheoryData<int, int, int> result = [];