Advance HEIF AV1 encoder foundation

HEIF_IMPLEMENTATION_PLAN.md

@@ -754,6 +754,21 @@ Final decoder stream, presentation, and public-registration evidence on 2026-09-
   `git diff --check` passes, and `.gitattributes` is unchanged. Every VSTest invocation returned
   normally with no surviving test host and no Windows application-error dialog.
 
+SIMD traversal consistency evidence on 2026-09-02:
+
+- [x] The shared `Numerics` vector-count helpers now cover same-lane spans and all fixed hardware widths.
+  AV1 decoder and current encoder hot paths use those helpers for complete-vector traversal instead of
+  repeating local modulo or last-vector calculations. Reverse-source indexing and algorithm-specific
+  partial-output groups remain explicit because they are not vector-count calculations.
+- [x] Forward quantization, palette prediction, and scaled inter prediction construct width-specific SIMD
+  constants only when at least one vector batch will execute. Narrower dispatch tiers consume only the
+  remainder left by wider tiers before the scalar tail.
+- [x] The net11.0 Release production assembly builds with zero warnings and zero errors. The test project
+  builds with zero errors while retaining the existing repository warning set. Roslynk reports zero
+  compiler errors, `git diff --check` passes, and `.gitattributes` is unchanged.
+  Foreground VSTest passes 63 of 63 focused quantizer, forward-transform, CDEF, restoration, palette,
+  intra, inter, film-grain, and super-resolution cases.
+
 Decoder exit gate:
 
 - [x] Every supported native format and AV1 tool has exact current-main libaom production-path evidence.
@@ -768,28 +783,49 @@ Writer primitives are not an encoder. The public encoder remains incomplete unti
 
 ### 5. Define and enforce the encoder contract
 
+- [x] Use official libaom `main` at `a40ed1ea9e4ecc3df58a5bccb76623f2c94ae727` as the encoder syntax, probability-model, transform, quantization, filtering, and bitstream reference.
+- [x] Use the existing PNG, TIFF, and JPEG encoders as the ImageSharp architecture reference: generic `Image<TPixel>` input, encoder options taking precedence over converted format metadata and codec defaults, allocator-owned temporary storage, and deterministic disposal.
+- [x] Treat source pixel type, source alpha representation, and decoded source bit depth as conversion inputs, never as output-eligibility checks. Do not pre-scan pixels before encoding.
+- [x] Resolve output configuration once from explicit encoder options, converted `HeifMetadata`, and AV1 defaults in that order. Sanitize only combinations that cannot describe a legal requested output, and never write resolved values back to source metadata.
 - [ ] Finalize observable options for quality, effort, lossless mode, bit depth, chroma subsampling, alpha quality, metadata, and bounded sequences.
 - [ ] Preserve high-bit-depth source precision through 16-bit RGB and native 10/12-bit component planes.
-- [ ] Reject unsupported combinations at the public boundary before writing output.
+- [ ] Reject only genuinely unsupported output combinations at the public boundary before writing output.
 - [ ] Register only capabilities that the completed encoder proves.
 
+Encoder data-flow contract:
+
+1. Resolve immutable frame and sequence output settings before allocating codec state.
+2. Convert each generic `ImageFrame<TPixel>` once through `PixelOperations<TPixel>` and the SIMD-first HEIF planar converter into native 8, 10, or 12-bit planes. Alpha is encoded as an auxiliary image when requested by the resolved output contract; it is not discarded through a source scan.
+3. Reuse allocator-owned plane, row, block, transform, quantization, entropy, and reconstruction workspaces for the complete frame. No active path may allocate per row, block, transform, scanline, or SIMD tail.
+4. Analyze and encode tiles directly from those planes, retaining reconstructed reference frames only for the bounded sequence lifetime.
+5. Stream OBUs and container extents through allocator-backed chunked storage. Every ownership transfer is explicit, every owner is disposed exactly once, and no `ToArray` or file-sized copy crosses a layer boundary.
+6. Iterate image frames using ImageSharp frame metadata and format-connecting metadata. Root-frame-only behavior is permitted only for an explicitly static output contract.
+
+Encoder verification contract:
+
+- Exercise source pixel formats independently from requested AV1 bit depth, chroma subsampling, alpha, and lossless/lossy mode.
+- Run every SIMD operator through FeatureTestRunner at Vector512, Vector256, Vector128, and scalar tiers against an independent scalar oracle shaped from the same libaom revision.
+- Cover discontiguous allocator buffers, constrained memory groups, cancellation, non-seekable output, multiple extents, auxiliary alpha, and bounded sequences.
+- Validate produced AV1 payloads with current-main libaom and compare native planes before using ImageSharp self-decode as supplemental container coverage.
+
 ### 6. Build the complete AV1 frame encoder
 
 - [~] SIMD-first RGB-to-native-plane conversion exists locally.
 - [~] Forward transform families and transform workspace exist locally.
 - [~] Symbol writer, coefficient writer, and tile writer fragments exist locally.
-- [ ] Connect a frame-owned encoder lifecycle using ImageSharp allocators and pools.
+- [~] A non-owning encoder-frame view now separates visible conversion regions from coded regions and performs complete left, top, right, bottom, and corner extension across each bordered plane. Current libaom uses 8-sample-aligned coded dimensions, a 32-sample-aligned luma stride with chroma stride derived from it, and a 64-pixel luma border for non-resized all-intra encoding. Allocator-backed luma and 4:2:0 chroma extension passed direct net11 VSTest; the containing encode operation still needs to connect matching plane rents with ordinary `using` lifetimes.
-- [ ] Write compliant temporal delimiter, sequence header, frame header, tile group, metadata, and padding OBUs as required.
+- [~] Temporal delimiter, sequence header, frame header, and combined-frame tile-group writing exist locally. The remaining required metadata, padding, and encoder-wide syntax paths are not complete.
 - [ ] Implement superblock and partition analysis for every permitted block size and partition.
 - [ ] Implement intra mode search, chroma mode search, palette, filter intra, chroma-from-luma, and intra-block copy decisions.
 - [ ] Implement inter mode search for bounded sequences, including reference selection and the decoder-supported inter tools.
-- [ ] Implement transform-size/type search, forward transform, quantization, coefficient optimization, and lossless behavior.
+- [~] Current-libaom `av1_quantize_fp_no_qmatrix` arithmetic is implemented as a closed generic forward-quantizer family with Vector512, Vector256, Vector128, and scalar paths, raster-order output, coded 64-point coefficient limits, and scan-order EOB selection. Transform search, coefficient optimization, and lossless behavior remain.
 - [ ] Implement real rate-distortion selection and make quality and effort change work, size, and output quality.
 - [ ] Implement tile-local entropy coding and CDF update behavior.
 - [ ] Implement legal deblocking, CDEF, restoration, super-resolution, and film-grain signaling decisions.
-- [ ] Remove per-transform and per-block managed allocations from active encoder paths.
+- [~] The coefficient symbol encoder now reuses tile-lifetime level and context workspaces instead of allocating per transform. Every remaining encoder fragment must be audited before it becomes active.
-- [ ] Use descending SIMD dispatch: Vector512, Vector256, Vector128, then scalar.
+- [~] The planar conversion, forward transform, and forward quantizer use descending SIMD dispatch: Vector512, Vector256, Vector128, then scalar. Apply the same rule to every later hot-path family.
-- [ ] Verify every SIMD operator with FeatureTestRunner and an independent scalar oracle shaped from the same current-main libaom behavior.
+- [~] Forward-quantizer FeatureTestRunner and zero-allocation tests compare every hardware tier with an independent scan-order scalar oracle shaped from current-main libaom. Both passed direct net11 VSTest in Release.
+- [~] The combined-frame writer now completes the byte-counted uncompressed frame header before starting the optional multi-tile tile-group flag, matching current libaom's separate frame-header and tile-group writers. A non-uniform two-tile round trip verifies the explicit boundaries, both tile payloads, and complete stream consumption through direct net11 VSTest in Release.
 
 ### 7. Write complete AVIF output
 

src/ImageSharp/Common/Helpers/Numerics.cs

@@ -1024,6 +1024,46 @@ internal static class Numerics
         where TVector : struct
         => (uint)span.Length / (uint)Vector512<TVector>.Count;
 
+    /// <summary>
+    /// Gets the count of vectors that safely fit into a span whose element type matches the vector lane type.
+    /// </summary>
+    /// <typeparam name="TVector">The type of the span elements and vector lanes.</typeparam>
+    /// <param name="span">The given span.</param>
+    /// <returns>Count of vectors that safely fit into the span.</returns>
+    public static nuint Vector128Count<TVector>(this ReadOnlySpan<TVector> span)
+        where TVector : struct
+        => (uint)span.Length / (uint)Vector128<TVector>.Count;
+
+    /// <summary>
+    /// Gets the count of vectors that safely fit into a span whose element type matches the vector lane type.
+    /// </summary>
+    /// <typeparam name="TVector">The type of the span elements and vector lanes.</typeparam>
+    /// <param name="span">The given span.</param>
+    /// <returns>Count of vectors that safely fit into the span.</returns>
+    public static nuint Vector256Count<TVector>(this ReadOnlySpan<TVector> span)
+        where TVector : struct
+        => (uint)span.Length / (uint)Vector256<TVector>.Count;
+
+    /// <summary>
+    /// Gets the count of vectors that safely fit into a span whose element type matches the vector lane type.
+    /// </summary>
+    /// <typeparam name="TVector">The type of the span elements and vector lanes.</typeparam>
+    /// <param name="span">The given span.</param>
+    /// <returns>Count of vectors that safely fit into the span.</returns>
+    public static nuint Vector512Count<TVector>(this ReadOnlySpan<TVector> span)
+        where TVector : struct
+        => (uint)span.Length / (uint)Vector512<TVector>.Count;
+
+    /// <summary>
+    /// Gets the count of vectors that safely fit into the given length.
+    /// </summary>
+    /// <typeparam name="TVector">The type of the vector.</typeparam>
+    /// <param name="length">The given length.</param>
+    /// <returns>Count of vectors that safely fit into the length.</returns>
+    public static nuint Vector128Count<TVector>(int length)
+        where TVector : struct
+        => (uint)length / (uint)Vector128<TVector>.Count;
+
     /// <summary>
     /// Gets the count of vectors that safely fit into length.
     /// </summary>

src/ImageSharp/Formats/Heif/Av1/Av1BitStreamWriter.cs

@@ -187,10 +187,11 @@ internal ref struct Av1BitStreamWriter
         }
         else
         {
-            uint extraBit = ((value + m) >> 1) - value;
+            // libaom partitions the upper values into a shorter prefix followed by the low bit of the offset from m.
-            uint k = (value + m - extraBit) >> 1;
+            uint offset = value - m;
+            uint k = m + (offset >> 1);
             this.WriteLiteral(k, w - 1);
-            this.WriteLiteral(extraBit, 1);
+            this.WriteLiteral(offset & 1, 1);
         }
     }
 
@@ -236,7 +237,7 @@ internal ref struct Av1BitStreamWriter
         DebugGuard.IsTrue(Av1Math.Modulus8(this.BitPosition) == 0, "Writing of Tile Data only allowed on byte alignment");
 
         int wordPosition = this.BitPosition >> 3;
-        if (this.span.Length <= wordPosition + tileData.Length)
+        if (this.span.Length < wordPosition + tileData.Length)
         {
             this.memory.GetSpan(wordPosition + tileData.Length);
             this.span = this.memory.GetEntireSpan();

src/ImageSharp/Formats/Heif/Av1/Color/Av1PresentationSampleBuffer.cs

@@ -475,7 +475,9 @@ internal sealed class Av1PresentationSampleBuffer<TSample, TBuffer> : IDisposabl
             ref ushort bottomSourceBase = ref MemoryMarshal.GetReference(bottomSource);
             ref ushort topDestinationBase = ref MemoryMarshal.GetReference(topDestination);
             ref ushort bottomDestinationBase = ref MemoryMarshal.GetReference(bottomDestination);
-            for (; x + Vector128<ushort>.Count <= lastSource; x += Vector128<ushort>.Count)
+            nuint vectorCount = topSource[..lastSource].Vector128Count<ushort>();
+
+            for (; vectorCount > 0; vectorCount--, x += Vector128<ushort>.Count)
             {
                 Vector128<ushort> top0 = Vector128.LoadUnsafe(ref topSourceBase, (nuint)x);
                 Vector128<ushort> top1 = Vector128.LoadUnsafe(ref topSourceBase, (nuint)(x + 1));

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

@@ -16,6 +16,11 @@ namespace SixLabors.ImageSharp.Formats.Heif.Av1.Entropy;
 /// </summary>
 internal class Av1SymbolEncoder : IDisposable
 {
+    /// <summary>
+    /// The largest coefficient-context plane required after AV1 removes the uncoded half of 64-point transforms.
+    /// </summary>
+    private const int MaximumCoefficientContextCount = (Av1Constants.MaxTransformSize / 2) * (Av1Constants.MaxTransformSize / 2);
+
     /// <summary>
     /// The tile-adaptive intra-block-copy distribution.
     /// </summary>
@@ -132,9 +137,14 @@ internal class Av1SymbolEncoder : IDisposable
     private bool isDisposed;
 
     /// <summary>
-    /// The configuration providing output and coefficient-context memory.
+    /// The reusable padded coefficient levels used to derive entropy contexts.
+    /// </summary>
+    private readonly Av1LevelBuffer levels;
+
+    /// <summary>
+    /// The reusable raster-order coefficient contexts for one transform.
     /// </summary>
-    private readonly Configuration configuration;
+    private readonly IMemoryOwner<sbyte> coefficientContexts;
 
     /// <summary>
     /// The range writer producing the current tile payload.
@@ -178,7 +188,8 @@ internal class Av1SymbolEncoder : IDisposable
         this.coefficientsBaseEndOfBlock = Av1DefaultDistributions.GetBaseEndOfBlock(qIndex);
         this.dcSign = Av1DefaultDistributions.GetDcSign(qIndex);
         this.endOfBlockExtra = Av1DefaultDistributions.GetEndOfBlockExtra(qIndex);
-        this.configuration = configuration;
+        this.levels = new(configuration);
+        this.coefficientContexts = configuration.MemoryAllocator.Allocate<sbyte>(MaximumCoefficientContextCount);
         this.writer = new(configuration, initialSize, updateCdf);
         this.baseQIndex = qIndex;
     }
@@ -275,9 +286,11 @@ internal class Av1SymbolEncoder : IDisposable
 
         ref Av1SymbolWriter w = ref this.writer;
 
-        // AV1 omits high-frequency coefficients beyond 32 samples on every 64-point transform dimension.
+        // AV1 omits high-frequency coefficients beyond 32 samples on every 64-point transform dimension. The tile
-        using Av1LevelBuffer levels = new(this.configuration, new Size(width, height));
+        // owns maximum-sized workspaces so repeated transform coding changes only their active views.
-        Span<sbyte> coefficientContexts = new sbyte[width * height];
+        this.levels.Reset(new Size(width, height));
+        Span<sbyte> coefficientContexts = this.coefficientContexts.Memory.Span[..(width * height)];
+        coefficientContexts.Clear();
 
         Guard.MustBeLessThan((int)transformSizeContext, (int)Av1TransformSize.AllSizes, nameof(transformSizeContext));
 
@@ -288,7 +301,7 @@ internal class Av1SymbolEncoder : IDisposable
             return 0;
         }
 
-        levels.Initialize(coefficientBuffer);
+        this.levels.Initialize(coefficientBuffer);
         if (componentType == Av1ComponentType.Luminance)
         {
             this.WriteTransformType(transformType, transformSize, useReducedTransformSet, this.baseQIndex, filterIntraMode, intraDirection);
@@ -296,14 +309,14 @@ internal class Av1SymbolEncoder : IDisposable
 
         this.WriteEndOfBlockPosition(endOfBlock, componentType, transformClass, transformSize, transformSizeContext);
 
-        Av1SymbolContextHelper.GetNzMapContexts(levels, scan, endOfBlock, transformSize, transformClass, coefficientContexts);
+        Av1SymbolContextHelper.GetNzMapContexts(this.levels, scan, endOfBlock, transformSize, transformClass, coefficientContexts);
         int limitedTransformSizeContext = Math.Min((int)transformSizeContext, (int)Av1TransformSize.Size32x32);
         for (c = endOfBlock - 1; c >= 0; --c)
         {
             short pos = scan[c];
             int v = coefficientBuffer[pos];
             short coeffContext = coefficientContexts[pos];
-            Point position = levels.GetPosition(pos);
+            Point position = this.levels.GetPosition(pos);
             int level = Math.Abs(v);
 
             if (c == endOfBlock - 1)
@@ -319,7 +332,7 @@ internal class Av1SymbolEncoder : IDisposable
             {
                 // Base-range symbols extend levels above the two base levels in fixed-size chunks.
                 int baseRange = level - 1 - Av1Constants.BaseLevelsCount;
-                int baseRangeContext = Av1SymbolContextHelper.GetBaseRangeContext(levels, position, transformClass);
+                int baseRangeContext = Av1SymbolContextHelper.GetBaseRangeContext(this.levels, position, transformClass);
                 for (int idx = 0; idx < Av1Constants.CoefficientBaseRange; idx += Av1Constants.BaseRangeSizeMinus1)
                 {
                     int k = Math.Min(baseRange - idx, Av1Constants.BaseRangeSizeMinus1);
@@ -429,6 +442,8 @@ internal class Av1SymbolEncoder : IDisposable
     {
         if (!this.isDisposed)
         {
+            this.coefficientContexts.Dispose();
+            this.levels.Dispose();
             this.writer.Dispose();
             this.isDisposed = true;
         }

src/ImageSharp/Formats/Heif/Av1/IAv1TileWriter.cs

@@ -9,11 +9,11 @@ namespace SixLabors.ImageSharp.Formats.Heif.Av1;
 internal interface IAv1TileWriter
 {
     /// <summary>
-    /// Write the information for a single tile.
+    /// Gets the encoded bytes for a single tile.
     /// </summary>
-    /// <param name="tileNum">The index of the tile that is to be read.</param>
+    /// <param name="tileNum">The index of the encoded tile.</param>
     /// <returns>
     /// The bytes of encoded data in the bitstream dedicated to this tile.
     /// </returns>
-    Span<byte> WriteTile(int tileNum);
+    ReadOnlySpan<byte> GetTileData(int tileNum);
 }

src/ImageSharp/Formats/Heif/Av1/OpenBitstreamUnit/ObuWriter.cs

@@ -29,7 +29,7 @@ internal class ObuWriter
         // The allocation expands when necessary; this initial size avoids repeated growth for
         // the small headers and tiles produced by the current still-image encoder.
         int initialBufferSize = 2000;
-        AutoExpandingMemory<byte> buffer = new(configuration, initialBufferSize);
+        using AutoExpandingMemory<byte> buffer = new(configuration, initialBufferSize);
         Av1BitStreamWriter writer = new(buffer);
         WriteObuHeaderAndSize(stream, ObuType.TemporalDelimiter, []);
 
@@ -173,9 +173,6 @@ internal class ObuWriter
         {
             // AV1 fixes this RGB identity-matrix combination to full-range 4:4:4 and omits
             // the range and subsampling fields used by YUV configurations.
-            colorConfig.ColorRange = true;
-            colorConfig.SubSamplingX = false;
-            colorConfig.SubSamplingY = false;
         }
         else
         {
@@ -335,13 +332,18 @@ internal class ObuWriter
         else
         {
             int startSuperBlock = 0;
-            int i = 0;
+            for (int i = 0; i < tileInfo.TileColumnCount; i++)
-            for (; startSuperBlock < superblockColumnCount; i++)
             {
-                uint widthInSuperBlocks = (uint)((tileInfo.TileColumnStartModeInfo[i] >> superblockShift) - startSuperBlock);
+                int endSuperBlock = i == tileInfo.TileColumnCount - 1
+                    ? superblockColumnCount
+                    : tileInfo.TileColumnStartModeInfo[i + 1] >> superblockShift;
+
+                // The stored terminal boundary is clipped to the visible mode-info width. libaom retains the exact
+                // superblock endpoint, so the final tile uses the derived frame-wide superblock count instead.
+                uint widthInSuperBlocks = (uint)(endSuperBlock - startSuperBlock);
                 uint maxWidth = (uint)Math.Min(superblockColumnCount - startSuperBlock, tileInfo.MaxTileWidthSuperblock);
                 writer.WriteNonSymmetric(widthInSuperBlocks - 1, maxWidth);
-                startSuperBlock += (int)widthInSuperBlocks;
+                startSuperBlock = endSuperBlock;
             }
 
             if (startSuperBlock != superblockColumnCount)
@@ -350,12 +352,17 @@ internal class ObuWriter
             }
 
             startSuperBlock = 0;
-            for (i = 0; startSuperBlock < superblockRowCount; i++)
+            for (int i = 0; i < tileInfo.TileRowCount; i++)
             {
-                uint heightInSuperBlocks = (uint)((tileInfo.TileRowStartModeInfo[i] >> superblockShift) - startSuperBlock);
+                int endSuperBlock = i == tileInfo.TileRowCount - 1
+                    ? superblockRowCount
+                    : tileInfo.TileRowStartModeInfo[i + 1] >> superblockShift;
+
+                // As with columns, the final visible mode-info boundary may end inside its containing superblock.
+                uint heightInSuperBlocks = (uint)(endSuperBlock - startSuperBlock);
                 uint maxHeight = (uint)Math.Min(superblockRowCount - startSuperBlock, tileInfo.MaxTileHeightSuperblock);
                 writer.WriteNonSymmetric(heightInSuperBlocks - 1, maxHeight);
-                startSuperBlock += (int)heightInSuperBlocks;
+                startSuperBlock = endSuperBlock;
             }
 
             if (startSuperBlock != superblockRowCount)
@@ -521,6 +528,11 @@ internal class ObuWriter
     private static void WriteTileGroup(ref Av1BitStreamWriter writer, ObuTileGroupHeader tileInfo, IAv1TileWriter tileWriter)
     {
         int tileCount = tileInfo.TileColumnCount * tileInfo.TileRowCount;
+
+        // libaom starts the tile-group header at the next byte after the uncompressed frame header. This
+        // boundary is required before the optional flag because the flag belongs to tile_group_obu syntax.
+        AlignToByteBoundary(ref writer);
+
         if (tileCount > 1)
         {
             // A combined OBU_FRAME has implicit complete-frame tile bounds. The reference decoder still
@@ -544,7 +556,7 @@ internal class ObuWriter
         int tileCount = tileInfo.TileColumnCount * tileInfo.TileRowCount;
         for (int tileNum = 0; tileNum < tileCount; tileNum++)
         {
-            Span<byte> tileData = tileWriter.WriteTile(tileNum);
+            ReadOnlySpan<byte> tileData = tileWriter.GetTileData(tileNum);
             if (tileNum != tileCount - 1 && tileCount > 1)
             {
                 writer.WriteLittleEndian((uint)tileData.Length - 1U, tileInfo.TileSizeBytes);

src/ImageSharp/Formats/Heif/Av1/Pipeline/Av1EncoderFrame.cs

@@ -0,0 +1,334 @@
+// Copyright (c) Six Labors.
+// Licensed under the Six Labors Split License.
+
+using SixLabors.ImageSharp.Formats.Heif.Components;
+using SixLabors.ImageSharp.Memory;
+
+namespace SixLabors.ImageSharp.Formats.Heif.Av1.Pipeline;
+
+/// <summary>
+/// Provides non-owning visible and coded views over operation-scoped AV1 component planes.
+/// </summary>
+/// <typeparam name="TSample">The native unsigned sample storage type.</typeparam>
+internal readonly struct Av1EncoderFrame<TSample>
+    where TSample : unmanaged
+{
+    /// <summary>
+    /// The base-two alignment exponent applied to coded frame dimensions.
+    /// </summary>
+    private const int CodedDimensionAlignmentLog2 = 3;
+
+    /// <summary>
+    /// The base-two alignment exponent applied to the physical luma row stride.
+    /// </summary>
+    private const int LumaStrideAlignmentLog2 = 5;
+
+    /// <summary>
+    /// The physical luma border required by non-resized all-intra encoding.
+    /// </summary>
+    public const int LumaBorder = 64;
+
+    /// <summary>
+    /// Initializes a new instance of the <see cref="Av1EncoderFrame{TSample}"/> struct for a monochrome frame.
+    /// </summary>
+    /// <param name="luma">The coded luma region inside the bordered plane owned by the encode operation.</param>
+    /// <param name="width">The visible luma width.</param>
+    /// <param name="height">The visible luma height.</param>
+    /// <param name="bitDepth">The native component precision.</param>
+    public Av1EncoderFrame(Buffer2DRegion<TSample> luma, int width, int height, int bitDepth)
+        : this(luma, default, default, width, height, bitDepth, Av1ColorFormat.Yuv400, 0, 0)
+    {
+    }
+
+    /// <summary>
+    /// Initializes a new instance of the <see cref="Av1EncoderFrame{TSample}"/> struct for a color frame.
+    /// </summary>
+    /// <param name="luma">The coded luma region inside the bordered plane owned by the encode operation.</param>
+    /// <param name="chromaBlue">The coded blue-difference region inside its bordered plane.</param>
+    /// <param name="chromaRed">The coded red-difference region inside its bordered plane.</param>
+    /// <param name="width">The visible luma width.</param>
+    /// <param name="height">The visible luma height.</param>
+    /// <param name="bitDepth">The native component precision.</param>
+    /// <param name="colorFormat">The native luma and chroma sampling layout.</param>
+    /// <param name="chromaPositionX">The horizontal chroma position in half-luma-sample units.</param>
+    /// <param name="chromaPositionY">The vertical chroma position in half-luma-sample units.</param>
+    public Av1EncoderFrame(
+        Buffer2DRegion<TSample> luma,
+        Buffer2DRegion<TSample> chromaBlue,
+        Buffer2DRegion<TSample> chromaRed,
+        int width,
+        int height,
+        int bitDepth,
+        Av1ColorFormat colorFormat,
+        int chromaPositionX,
+        int chromaPositionY)
+    {
+        this.Width = width;
+        this.Height = height;
+        this.LumaBitDepth = bitDepth;
+        this.ChromaBitDepth = bitDepth;
+        this.IsMonochrome = colorFormat == Av1ColorFormat.Yuv400;
+        this.ChromaSubsamplingX = colorFormat is Av1ColorFormat.Yuv420 or Av1ColorFormat.Yuv422 ? 1 : 0;
+        this.ChromaSubsamplingY = colorFormat == Av1ColorFormat.Yuv420 ? 1 : 0;
+        this.ChromaPositionX = chromaPositionX;
+        this.ChromaPositionY = chromaPositionY;
+        this.CodedWidth = luma.Width;
+        this.CodedHeight = luma.Height;
+        int visibleChromaWidth = (width + this.ChromaSubsamplingX) >> this.ChromaSubsamplingX;
+        int visibleChromaHeight = (height + this.ChromaSubsamplingY) >> this.ChromaSubsamplingY;
+        Buffer2DRegion<TSample> visibleChromaBlue = this.IsMonochrome
+            ? default
+            : chromaBlue.GetSubRegion(0, 0, visibleChromaWidth, visibleChromaHeight);
+
+        Buffer2DRegion<TSample> visibleChromaRed = this.IsMonochrome
+            ? default
+            : chromaRed.GetSubRegion(0, 0, visibleChromaWidth, visibleChromaHeight);
+
+        this.View = new PlanarView(
+            luma.GetSubRegion(0, 0, width, height),
+            visibleChromaBlue,
+            visibleChromaRed,
+            width,
+            height,
+            bitDepth,
+            colorFormat,
+            chromaPositionX,
+            chromaPositionY);
+
+        this.CodedView = new PlanarView(
+            luma,
+            chromaBlue,
+            chromaRed,
+            this.CodedWidth,
+            this.CodedHeight,
+            bitDepth,
+            colorFormat,
+            chromaPositionX,
+            chromaPositionY);
+    }
+
+    /// <summary>
+    /// Gets the writable component-plane view used by closed generic conversion and coding operations.
+    /// </summary>
+    public PlanarView View { get; }
+
+    /// <summary>
+    /// Gets the writable coded component planes used by block coding and reconstruction.
+    /// </summary>
+    public PlanarView CodedView { get; }
+
+    /// <summary>
+    /// Gets the visible luma width.
+    /// </summary>
+    public int Width { get; }
+
+    /// <summary>
+    /// Gets the visible luma height.
+    /// </summary>
+    public int Height { get; }
+
+    /// <summary>
+    /// Gets the luma width rounded up to the fixed coding-block boundary.
+    /// </summary>
+    public int CodedWidth { get; }
+
+    /// <summary>
+    /// Gets the luma height rounded up to the fixed coding-block boundary.
+    /// </summary>
+    public int CodedHeight { get; }
+
+    /// <summary>
+    /// Gets the native luma sample precision.
+    /// </summary>
+    public int LumaBitDepth { get; }
+
+    /// <summary>
+    /// Gets the native chroma sample precision.
+    /// </summary>
+    public int ChromaBitDepth { get; }
+
+    /// <summary>
+    /// Gets a value indicating whether the frame contains only luma samples.
+    /// </summary>
+    public bool IsMonochrome { get; }
+
+    /// <summary>
+    /// Gets the horizontal chroma subsampling shift.
+    /// </summary>
+    public int ChromaSubsamplingX { get; }
+
+    /// <summary>
+    /// Gets the vertical chroma subsampling shift.
+    /// </summary>
+    public int ChromaSubsamplingY { get; }
+
+    /// <summary>
+    /// Gets the horizontal chroma position in half-luma-sample units.
+    /// </summary>
+    public int ChromaPositionX { get; }
+
+    /// <summary>
+    /// Gets the vertical chroma position in half-luma-sample units.
+    /// </summary>
+    public int ChromaPositionY { get; }
+
+    /// <summary>
+    /// Calculates the physical dimensions required for an all-intra component plane.
+    /// </summary>
+    /// <param name="width">The visible luma width.</param>
+    /// <param name="height">The visible luma height.</param>
+    /// <param name="subsamplingX">The plane's horizontal subsampling shift.</param>
+    /// <param name="subsamplingY">The plane's vertical subsampling shift.</param>
+    /// <returns>The physical plane dimensions, including its complete border and row padding.</returns>
+    public static Size GetPlaneBufferSize(int width, int height, int subsamplingX, int subsamplingY)
+    {
+        int codedWidth = Av1Math.AlignPowerOf2(width, CodedDimensionAlignmentLog2);
+        int codedHeight = Av1Math.AlignPowerOf2(height, CodedDimensionAlignmentLog2);
+
+        // libaom aligns the complete luma row before deriving a subsampled plane's stride.
+        // Aligning chroma independently would produce a different physical layout for narrow or odd-sized frames.
+        int lumaStride = Av1Math.AlignPowerOf2(codedWidth + (2 * LumaBorder), LumaStrideAlignmentLog2);
+        int planeStride = lumaStride >> subsamplingX;
+        int planeBorderHeight = LumaBorder >> subsamplingY;
+        return new Size(planeStride, (codedHeight >> subsamplingY) + (2 * planeBorderHeight));
+    }
+
+    /// <summary>
+    /// Extends the visible edge samples through the coded padding.
+    /// </summary>
+    public void ExtendBorders()
+        => this.CodedView.ExtendBorders(this.Width, this.Height);
+
+    /// <summary>
+    /// Replicates the visible edge samples through a plane's complete physical border.
+    /// </summary>
+    private static void ExtendPlane(Buffer2DRegion<TSample> plane, int visibleWidth, int visibleHeight)
+    {
+        Buffer2D<TSample> buffer = plane.Buffer;
+        Rectangle bounds = plane.Bounds;
+        for (int y = 0; y < visibleHeight; y++)
+        {
+            Span<TSample> row = buffer.DangerousGetRowSpan(bounds.Y + y);
+
+            // libaom fills both physical borders and the right-hand coded alignment from the nearest visible sample.
+            row[..bounds.X].Fill(row[bounds.X]);
+            row[(bounds.X + visibleWidth)..].Fill(row[bounds.X + visibleWidth - 1]);
+        }
+
+        // Horizontal extension runs first so copying the first and last visible rows also initializes both corners.
+        ReadOnlySpan<TSample> firstVisibleRow = buffer.DangerousGetRowSpan(bounds.Y);
+        for (int y = 0; y < bounds.Y; y++)
+        {
+            firstVisibleRow.CopyTo(buffer.DangerousGetRowSpan(y));
+        }
+
+        ReadOnlySpan<TSample> finalVisibleRow = buffer.DangerousGetRowSpan(bounds.Y + visibleHeight - 1);
+        for (int y = bounds.Y + visibleHeight; y < buffer.Height; y++)
+        {
+            finalVisibleRow.CopyTo(buffer.DangerousGetRowSpan(y));
+        }
+    }
+
+    /// <summary>
+    /// Provides a non-owning component-plane view for generic hot-path operations.
+    /// </summary>
+    internal readonly struct PlanarView : IHeifPlanarSampleBuffer<TSample>
+    {
+        /// <summary>
+        /// The writable luma plane.
+        /// </summary>
+        private readonly Buffer2DRegion<TSample> luma;
+
+        /// <summary>
+        /// The writable blue-difference plane, or the default region for monochrome frames.
+        /// </summary>
+        private readonly Buffer2DRegion<TSample> chromaBlue;
+
+        /// <summary>
+        /// The writable red-difference plane, or the default region for monochrome frames.
+        /// </summary>
+        private readonly Buffer2DRegion<TSample> chromaRed;
+
+        /// <summary>
+        /// Initializes a new instance of the <see cref="PlanarView"/> struct.
+        /// </summary>
+        public PlanarView(
+            Buffer2DRegion<TSample> luma,
+            Buffer2DRegion<TSample> chromaBlue,
+            Buffer2DRegion<TSample> chromaRed,
+            int width,
+            int height,
+            int bitDepth,
+            Av1ColorFormat colorFormat,
+            int chromaPositionX,
+            int chromaPositionY)
+        {
+            this.luma = luma;
+            this.chromaBlue = chromaBlue;
+            this.chromaRed = chromaRed;
+            this.Width = width;
+            this.Height = height;
+            this.LumaBitDepth = bitDepth;
+            this.ChromaBitDepth = bitDepth;
+            this.IsMonochrome = colorFormat == Av1ColorFormat.Yuv400;
+            this.ChromaSubsamplingX = colorFormat is Av1ColorFormat.Yuv420 or Av1ColorFormat.Yuv422 ? 1 : 0;
+            this.ChromaSubsamplingY = colorFormat == Av1ColorFormat.Yuv420 ? 1 : 0;
+            this.ChromaPositionX = chromaPositionX;
+            this.ChromaPositionY = chromaPositionY;
+        }
+
+        /// <inheritdoc/>
+        public int Width { get; }
+
+        /// <inheritdoc/>
+        public int Height { get; }
+
+        /// <inheritdoc/>
+        public int LumaBitDepth { get; }
+
+        /// <inheritdoc/>
+        public int ChromaBitDepth { get; }
+
+        /// <inheritdoc/>
+        public bool IsMonochrome { get; }
+
+        /// <inheritdoc/>
+        public int ChromaSubsamplingX { get; }
+
+        /// <inheritdoc/>
+        public int ChromaSubsamplingY { get; }
+
+        /// <inheritdoc/>
+        public int ChromaPositionX { get; }
+
+        /// <inheritdoc/>
+        public int ChromaPositionY { get; }
+
+        /// <summary>
+        /// Replicates the visible component edges through the coded padding.
+        /// </summary>
+        /// <param name="visibleWidth">The visible luma width.</param>
+        /// <param name="visibleHeight">The visible luma height.</param>
+        public void ExtendBorders(int visibleWidth, int visibleHeight)
+        {
+            ExtendPlane(this.luma, visibleWidth, visibleHeight);
+
+            if (!this.IsMonochrome)
+            {
+                int visibleChromaWidth = (visibleWidth + this.ChromaSubsamplingX) >> this.ChromaSubsamplingX;
+                int visibleChromaHeight = (visibleHeight + this.ChromaSubsamplingY) >> this.ChromaSubsamplingY;
+                ExtendPlane(this.chromaBlue, visibleChromaWidth, visibleChromaHeight);
+                ExtendPlane(this.chromaRed, visibleChromaWidth, visibleChromaHeight);
+            }
+        }
+
+        /// <inheritdoc/>
+        public Span<TSample> GetLumaRowSpan(int row) => this.luma.DangerousGetRowSpan(row);
+
+        /// <inheritdoc/>
+        public Span<TSample> GetChromaBlueRowSpan(int row) => this.chromaBlue.DangerousGetRowSpan(row);
+
+        /// <inheritdoc/>
+        public Span<TSample> GetChromaRedRowSpan(int row) => this.chromaRed.DangerousGetRowSpan(row);
+    }
+}

src/ImageSharp/Formats/Heif/Av1/Pipeline/Cdef/Av1CdefFilter.cs

@@ -60,7 +60,8 @@ internal static partial class Av1CdefFilter
                 int firstDestinationRow = destinationOffset + (row * destinationStride);
                 int secondDestinationRow = firstDestinationRow + destinationStride;
                 int column = 0;
-                for (; column <= width - Vector128<byte>.Count; column += Vector128<byte>.Count)
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<byte> first = Vector128.LoadUnsafe(ref sourceBase, (nuint)(firstSourceRow + column));
                     Vector128<byte> second = Vector128.LoadUnsafe(ref sourceBase, (nuint)(secondSourceRow + column));

src/ImageSharp/Formats/Heif/Av1/Pipeline/FilmGrain/Av1FilmGrainNoise.cs

@@ -264,11 +264,11 @@ internal static class Av1FilmGrainNoise
             int sampleRowOffset = row * sampleStride;
             int grainRowOffset = row * grainStride;
             int column = 0;
-            int vectorEnd = width - Vector256<int>.Count;
+            int vectorEnd = (int)(Numerics.Vector256Count<int>(width) * (nuint)Vector256<int>.Count);
 
             // Samples, grain, and scale indices share the same lane coordinate. No permutation is required between
             // the lookup, fixed-point multiply, clipping, and native-sample store.
-            for (; column <= vectorEnd; column += Vector256<int>.Count)
+            for (; column < vectorEnd; column += Vector256<int>.Count)
             {
                 ref TSample destination = ref Unsafe.Add(ref sampleBase, sampleRowOffset + column);
                 Vector256<int> source = Av1FilmGrainSampleOperations<TSample>.Load8(ref destination);
@@ -319,11 +319,11 @@ internal static class Av1FilmGrainNoise
             int sampleRowOffset = row * sampleStride;
             int grainRowOffset = row * grainStride;
             int column = 0;
-            int vectorEnd = width - Vector128<int>.Count;
+            int vectorEnd = (int)(Numerics.Vector128Count<int>(width) * (nuint)Vector128<int>.Count);
 
             // Four scalar table reads assemble the scale vector; the rest of the normative grain equation remains
             // lane-wise, including interpolation for 10- and 12-bit coordinates.
-            for (; column <= vectorEnd; column += Vector128<int>.Count)
+            for (; column < vectorEnd; column += Vector128<int>.Count)
             {
                 ref TSample destination = ref Unsafe.Add(ref sampleBase, sampleRowOffset + column);
                 Vector128<int> source = Av1FilmGrainSampleOperations<TSample>.Load4(ref destination);
@@ -570,11 +570,11 @@ internal static class Av1FilmGrainNoise
             int chromaRowOffset = row * chromaStride;
             int grainRowOffset = row * grainStride;
             int column = 0;
-            int vectorEnd = width - Vector256<int>.Count;
+            int vectorEnd = (int)(Numerics.Vector256Count<int>(width) * (nuint)Vector256<int>.Count);
 
             // Each lane represents one chroma coordinate and its corresponding reconstructed-luma coordinate. The Q6
             // luma/chroma blend is clamped to a legal sample code before it becomes a scaling-table index.
-            for (; column <= vectorEnd; column += Vector256<int>.Count)
+            for (; column < vectorEnd; column += Vector256<int>.Count)
             {
                 ref TSample lumaSource = ref Unsafe.Add(ref lumaRow, column << subsamplingX);
                 Vector256<int> averageLuma = Av1FilmGrainSampleOperations<TSample>.LoadChromaLuma8(ref lumaSource, subsamplingX);
@@ -701,11 +701,11 @@ internal static class Av1FilmGrainNoise
             int chromaRowOffset = row * chromaStride;
             int grainRowOffset = row * grainStride;
             int column = 0;
-            int vectorEnd = width - Vector128<int>.Count;
+            int vectorEnd = (int)(Numerics.Vector128Count<int>(width) * (nuint)Vector128<int>.Count);
 
             // The four-lane path preserves the same coordinate alignment and Q6 scaling-index arithmetic. Only the
             // table read changes from a hardware gather to four scalar reads assembled into a vector.
-            for (; column <= vectorEnd; column += Vector128<int>.Count)
+            for (; column < vectorEnd; column += Vector128<int>.Count)
             {
                 ref TSample lumaSource = ref Unsafe.Add(ref lumaRow, column << subsamplingX);
                 Vector128<int> averageLuma = Av1FilmGrainSampleOperations<TSample>.LoadChromaLuma4(ref lumaSource, subsamplingX);

src/ImageSharp/Formats/Heif/Av1/Pipeline/FilmGrain/Av1FilmGrainOverlap.cs

@@ -210,8 +210,8 @@ internal static class Av1FilmGrainOverlap
         Vector512<int> rounding = Vector512.Create(16);
         Vector512<int> minima = Vector512.Create(minimum);
         Vector512<int> maxima = Vector512.Create(maximum);
-        int vectorEnd = width - Vector512<int>.Count;
+        nuint vectorCount = Numerics.Vector512Count<int>(width - column);
-        for (; column <= vectorEnd; column += Vector512<int>.Count)
+        for (; vectorCount > 0; vectorCount--, column += Vector512<int>.Count)
         {
             Vector512<int> leftValues = Vector512.LoadUnsafe(ref leftBase, (nuint)column);
             Vector512<int> rightValues = Vector512.LoadUnsafe(ref rightBase, (nuint)column);
@@ -257,8 +257,8 @@ internal static class Av1FilmGrainOverlap
         Vector256<int> rounding = Vector256.Create(16);
         Vector256<int> minima = Vector256.Create(minimum);
         Vector256<int> maxima = Vector256.Create(maximum);
-        int vectorEnd = width - Vector256<int>.Count;
+        nuint vectorCount = Numerics.Vector256Count<int>(width - column);
-        for (; column <= vectorEnd; column += Vector256<int>.Count)
+        for (; vectorCount > 0; vectorCount--, column += Vector256<int>.Count)
         {
             Vector256<int> leftValues = Vector256.LoadUnsafe(ref leftBase, (nuint)column);
             Vector256<int> rightValues = Vector256.LoadUnsafe(ref rightBase, (nuint)column);
@@ -304,8 +304,8 @@ internal static class Av1FilmGrainOverlap
         Vector128<int> rounding = Vector128.Create(16);
         Vector128<int> minima = Vector128.Create(minimum);
         Vector128<int> maxima = Vector128.Create(maximum);
-        int vectorEnd = width - Vector128<int>.Count;
+        nuint vectorCount = Numerics.Vector128Count<int>(width - column);
-        for (; column <= vectorEnd; column += Vector128<int>.Count)
+        for (; vectorCount > 0; vectorCount--, column += Vector128<int>.Count)
         {
             Vector128<int> leftValues = Vector128.LoadUnsafe(ref leftBase, (nuint)column);
             Vector128<int> rightValues = Vector128.LoadUnsafe(ref rightBase, (nuint)column);

src/ImageSharp/Formats/Heif/Av1/Pipeline/LoopRestoration/Av1SelfGuidedFilter.Operations.cs

@@ -252,8 +252,8 @@ internal static partial class Av1SelfGuidedFilter
             Vector256<int> squareCarry = Vector256<int>.Zero;
             Vector256<int> sumCarry = Vector256<int>.Zero;
             int column = 0;
-            int vectorEnd = width - Vector256<int>.Count;
+            nuint vectorCount = Numerics.Vector256Count<int>(width - column);
-            for (; column <= vectorEnd; column += Vector256<int>.Count)
+            for (; vectorCount > 0; vectorCount--, column += Vector256<int>.Count)
             {
                 // Eight packed 16-bit samples become eight 32-bit lanes. The prefix scans mirror
                 // the reference decoder's scan_32, and the replicated carry joins consecutive vector batches.
@@ -323,8 +323,8 @@ internal static partial class Av1SelfGuidedFilter
             Vector128<int> squareCarry = Vector128<int>.Zero;
             Vector128<int> sumCarry = Vector128<int>.Zero;
             int column = 0;
-            int vectorEnd = width - Vector128<int>.Count;
+            nuint vectorCount = Numerics.Vector128Count<int>(width - column);
-            for (; column <= vectorEnd; column += Vector128<int>.Count)
+            for (; vectorCount > 0; vectorCount--, column += Vector128<int>.Count)
             {
                 // Loading through Vector64 avoids reading beyond the four samples owned by this
                 // batch. Widening is normalized by the runtime for both x86 and Arm64 targets.
@@ -827,11 +827,11 @@ internal static partial class Av1SelfGuidedFilter
             int roundingBits = SelfGuidedBits + ((row & 1) == 0 ? 5 : 4) - RestorationBits;
             Vector256<int> rounding = Vector256.Create(1 << (roundingBits - 1));
             int column = 0;
-            int vectorEnd = width - Vector256<int>.Count;
+            int vectorEnd = (int)(Numerics.Vector256Count<int>(width) * (nuint)Vector256<int>.Count);
 
             // Filtered signals use Q4 precision. Projection applies the signaled Q7 weights to their difference from
             // the unfiltered Q4 sample, then performs the combined Q11 rounding shift once before clipping.
-            for (; column <= vectorEnd; column += Vector256<int>.Count)
+            for (; column < vectorEnd; column += Vector256<int>.Count)
             {
                 Vector256<int> factors = CrossSum(blendFactors, coefficientRowOffset + column, bufferStride, row, vector);
                 Vector256<int> means = CrossSum(localMeans, coefficientRowOffset + column, bufferStride, row, vector);
@@ -886,11 +886,11 @@ internal static partial class Av1SelfGuidedFilter
             int roundingBits = SelfGuidedBits + ((row & 1) == 0 ? 5 : 4) - RestorationBits;
             Vector128<int> rounding = Vector128.Create(1 << (roundingBits - 1));
             int column = 0;
-            int vectorEnd = width - Vector128<int>.Count;
+            int vectorEnd = (int)(Numerics.Vector128Count<int>(width) * (nuint)Vector128<int>.Count);
 
             // The 128-bit path uses the same Q4/Q7 projection equation. The four-sample load and store are deliberately
             // 64 bits wide so a tightly strided destination row never requires writable padding.
-            for (; column <= vectorEnd; column += Vector128<int>.Count)
+            for (; column < vectorEnd; column += Vector128<int>.Count)
             {
                 Vector128<int> factors = CrossSum(blendFactors, coefficientRowOffset + column, bufferStride, row, vector);
                 Vector128<int> means = CrossSum(localMeans, coefficientRowOffset + column, bufferStride, row, vector);
@@ -946,8 +946,8 @@ internal static partial class Av1SelfGuidedFilter
             int filteredRowOffset = row * width;
             int coefficientRowOffset = bufferOrigin + (row * bufferStride);
             int column = 0;
-            int vectorEnd = width - Vector256<int>.Count;
+            nuint vectorCount = Numerics.Vector256Count<int>(width - column);
-            for (; column <= vectorEnd; column += Vector256<int>.Count)
+            for (; vectorCount > 0; vectorCount--, column += Vector256<int>.Count)
             {
                 Vector256<int> factors = CrossSum(blendFactors, coefficientRowOffset + column, bufferStride, vector);
                 Vector256<int> means = CrossSum(localMeans, coefficientRowOffset + column, bufferStride, vector);
@@ -1002,8 +1002,8 @@ internal static partial class Av1SelfGuidedFilter
             int filteredRowOffset = row * width;
             int coefficientRowOffset = bufferOrigin + (row * bufferStride);
             int column = 0;
-            int vectorEnd = width - Vector128<int>.Count;
+            nuint vectorCount = Numerics.Vector128Count<int>(width - column);
-            for (; column <= vectorEnd; column += Vector128<int>.Count)
+            for (; vectorCount > 0; vectorCount--, column += Vector128<int>.Count)
             {
                 Vector128<int> factors = CrossSum(blendFactors, coefficientRowOffset + column, bufferStride, vector);
                 Vector128<int> means = CrossSum(localMeans, coefficientRowOffset + column, bufferStride, vector);
@@ -1255,8 +1255,8 @@ internal static partial class Av1SelfGuidedFilter
             int destinationRowOffset = row * destinationStride;
             int filteredRowOffset = row * width;
             int column = 0;
-            int vectorEnd = width - Vector256<int>.Count;
+            nuint vectorCount = Numerics.Vector256Count<int>(width - column);
-            for (; column <= vectorEnd; column += Vector256<int>.Count)
+            for (; vectorCount > 0; vectorCount--, column += Vector256<int>.Count)
             {
                 Vector128<ushort> packed = Vector128.LoadUnsafe(ref sourceBase, (nuint)(sourceRowOffset + column));
                 Vector256<int> samples = Vector256.WidenLower(Vector256.Create(packed, Vector128<ushort>.Zero)).AsInt32();
@@ -1348,8 +1348,8 @@ internal static partial class Av1SelfGuidedFilter
             int destinationRowOffset = row * destinationStride;
             int filteredRowOffset = row * width;
             int column = 0;
-            int vectorEnd = width - Vector128<int>.Count;
+            nuint vectorCount = Numerics.Vector128Count<int>(width - column);
-            for (; column <= vectorEnd; column += Vector128<int>.Count)
+            for (; vectorCount > 0; vectorCount--, column += Vector128<int>.Count)
             {
                 ref ushort sourceReference = ref Unsafe.Add(ref sourceBase, sourceRowOffset + column);
                 Vector64<ushort> packed = Unsafe.As<ushort, Vector64<ushort>>(ref sourceReference);

src/ImageSharp/Formats/Heif/Av1/Pipeline/Quantizers/Av1ForwardQuantizer.Operator.cs

@@ -0,0 +1,205 @@
+// Copyright (c) Six Labors.
+// Licensed under the Six Labors Split License.
+
+using System.Runtime.Intrinsics;
+
+namespace SixLabors.ImageSharp.Formats.Heif.Av1.Pipeline.Quantizers;
+
+/// <content>
+/// Defines the arithmetic operators used by <see cref="Av1ForwardQuantizer"/>.
+/// </content>
+internal static partial class Av1ForwardQuantizer
+{
+    /// <summary>
+    /// Defines one AV1 forward-quantization arithmetic contract across hardware widths.
+    /// </summary>
+    internal interface IForwardQuantizationOperator
+    {
+        /// <summary>
+        /// Quantizes four raster-order transform coefficients.
+        /// </summary>
+        /// <param name="coefficients">The signed transform coefficients.</param>
+        /// <param name="rounding">The positive rounding constant after transform-size scaling.</param>
+        /// <param name="quantizer">The Q16 reciprocal quantizer.</param>
+        /// <param name="dequantizer">The Q3 reconstruction quantizer.</param>
+        /// <param name="logScale">The transform-size quantization scale.</param>
+        /// <param name="dequantizedCoefficients">The signed reconstruction coefficients.</param>
+        /// <returns>The signed entropy-coding coefficients.</returns>
+        public static abstract Vector128<int> Quantize(
+            Vector128<int> coefficients,
+            Vector128<int> rounding,
+            Vector128<int> quantizer,
+            Vector128<int> dequantizer,
+            int logScale,
+            out Vector128<int> dequantizedCoefficients);
+
+        /// <summary>
+        /// Quantizes eight raster-order transform coefficients.
+        /// </summary>
+        /// <param name="coefficients">The signed transform coefficients.</param>
+        /// <param name="rounding">The positive rounding constant after transform-size scaling.</param>
+        /// <param name="quantizer">The Q16 reciprocal quantizer.</param>
+        /// <param name="dequantizer">The Q3 reconstruction quantizer.</param>
+        /// <param name="logScale">The transform-size quantization scale.</param>
+        /// <param name="dequantizedCoefficients">The signed reconstruction coefficients.</param>
+        /// <returns>The signed entropy-coding coefficients.</returns>
+        public static abstract Vector256<int> Quantize(
+            Vector256<int> coefficients,
+            Vector256<int> rounding,
+            Vector256<int> quantizer,
+            Vector256<int> dequantizer,
+            int logScale,
+            out Vector256<int> dequantizedCoefficients);
+
+        /// <summary>
+        /// Quantizes sixteen raster-order transform coefficients.
+        /// </summary>
+        /// <param name="coefficients">The signed transform coefficients.</param>
+        /// <param name="rounding">The positive rounding constant after transform-size scaling.</param>
+        /// <param name="quantizer">The Q16 reciprocal quantizer.</param>
+        /// <param name="dequantizer">The Q3 reconstruction quantizer.</param>
+        /// <param name="logScale">The transform-size quantization scale.</param>
+        /// <param name="dequantizedCoefficients">The signed reconstruction coefficients.</param>
+        /// <returns>The signed entropy-coding coefficients.</returns>
+        public static abstract Vector512<int> Quantize(
+            Vector512<int> coefficients,
+            Vector512<int> rounding,
+            Vector512<int> quantizer,
+            Vector512<int> dequantizer,
+            int logScale,
+            out Vector512<int> dequantizedCoefficients);
+
+        /// <summary>
+        /// Quantizes one transform coefficient.
+        /// </summary>
+        /// <param name="coefficient">The signed transform coefficient.</param>
+        /// <param name="rounding">The positive rounding constant after transform-size scaling.</param>
+        /// <param name="quantizer">The Q16 reciprocal quantizer.</param>
+        /// <param name="dequantizer">The Q3 reconstruction quantizer.</param>
+        /// <param name="logScale">The transform-size quantization scale.</param>
+        /// <param name="dequantizedCoefficient">The signed reconstruction coefficient.</param>
+        /// <returns>The signed entropy-coding coefficient.</returns>
+        public static abstract int Quantize(
+            int coefficient,
+            int rounding,
+            int quantizer,
+            int dequantizer,
+            int logScale,
+            out int dequantizedCoefficient);
+    }
+
+    /// <summary>
+    /// Implements libaom's fast no-matrix quantizer for lossy transform blocks.
+    /// </summary>
+    /// <remarks>
+    /// Every SIMD overload preserves the scalar operation order: magnitude threshold, saturating round, Q16 reciprocal
+    /// multiply, transform-size shift, dequantization, and sign restoration. Each lane owns one raster-order coefficient.
+    /// </remarks>
+    internal readonly struct FastQuantizationOperator : IForwardQuantizationOperator
+    {
+        /// <inheritdoc/>
+        public static Vector128<int> Quantize(
+            Vector128<int> coefficients,
+            Vector128<int> rounding,
+            Vector128<int> quantizer,
+            Vector128<int> dequantizer,
+            int logScale,
+            out Vector128<int> dequantizedCoefficients)
+        {
+            Vector128<int> coefficientSign = Vector128.ShiftRightArithmetic(coefficients, 31);
+            Vector128<int> magnitude = Vector128.Abs(coefficients);
+
+            // libaom retains equality at the dequantizer threshold. Reversing the comparison and complementing its mask
+            // expresses scaledMagnitude >= dequantizer with the vector operations available for every supported ISA.
+            Vector128<int> thresholdMask = ~Vector128.GreaterThan(dequantizer, magnitude << (1 + logScale));
+
+            // Clamp the rounded magnitude to 32,767 so the following Q16 product remains within a signed 32-bit lane.
+            Vector128<int> rounded = Vector128.Min(magnitude + rounding, Vector128.Create((int)short.MaxValue));
+            Vector128<int> quantizedMagnitude = ((rounded * quantizer) >> (16 - logScale)) & thresholdMask;
+
+            // XOR followed by subtraction restores the original sign without a lane-wise branch.
+            Vector128<int> quantized = (quantizedMagnitude ^ coefficientSign) - coefficientSign;
+            Vector128<int> dequantizedMagnitude = (quantizedMagnitude * dequantizer) >> logScale;
+            dequantizedCoefficients = (dequantizedMagnitude ^ coefficientSign) - coefficientSign;
+            return quantized;
+        }
+
+        /// <inheritdoc/>
+        public static Vector256<int> Quantize(
+            Vector256<int> coefficients,
+            Vector256<int> rounding,
+            Vector256<int> quantizer,
+            Vector256<int> dequantizer,
+            int logScale,
+            out Vector256<int> dequantizedCoefficients)
+        {
+            Vector256<int> coefficientSign = Vector256.ShiftRightArithmetic(coefficients, 31);
+            Vector256<int> magnitude = Vector256.Abs(coefficients);
+
+            // Preserve threshold equality by complementing dequantizer > scaledMagnitude.
+            Vector256<int> thresholdMask = ~Vector256.GreaterThan(dequantizer, magnitude << (1 + logScale));
+
+            // Clamp the rounded magnitude to 32,767 so the following Q16 product remains within a signed 32-bit lane.
+            Vector256<int> rounded = Vector256.Min(magnitude + rounding, Vector256.Create((int)short.MaxValue));
+            Vector256<int> quantizedMagnitude = ((rounded * quantizer) >> (16 - logScale)) & thresholdMask;
+
+            // Apply the input sign to both coded and reconstructed magnitudes without branching.
+            Vector256<int> quantized = (quantizedMagnitude ^ coefficientSign) - coefficientSign;
+            Vector256<int> dequantizedMagnitude = (quantizedMagnitude * dequantizer) >> logScale;
+            dequantizedCoefficients = (dequantizedMagnitude ^ coefficientSign) - coefficientSign;
+            return quantized;
+        }
+
+        /// <inheritdoc/>
+        public static Vector512<int> Quantize(
+            Vector512<int> coefficients,
+            Vector512<int> rounding,
+            Vector512<int> quantizer,
+            Vector512<int> dequantizer,
+            int logScale,
+            out Vector512<int> dequantizedCoefficients)
+        {
+            Vector512<int> coefficientSign = Vector512.ShiftRightArithmetic(coefficients, 31);
+            Vector512<int> magnitude = Vector512.Abs(coefficients);
+
+            // Preserve threshold equality by complementing dequantizer > scaledMagnitude.
+            Vector512<int> thresholdMask = ~Vector512.GreaterThan(dequantizer, magnitude << (1 + logScale));
+
+            // Clamp the rounded magnitude to 32,767 so the following Q16 product remains within a signed 32-bit lane.
+            Vector512<int> rounded = Vector512.Min(magnitude + rounding, Vector512.Create((int)short.MaxValue));
+            Vector512<int> quantizedMagnitude = ((rounded * quantizer) >> (16 - logScale)) & thresholdMask;
+
+            // Apply the input sign to both coded and reconstructed magnitudes without branching.
+            Vector512<int> quantized = (quantizedMagnitude ^ coefficientSign) - coefficientSign;
+            Vector512<int> dequantizedMagnitude = (quantizedMagnitude * dequantizer) >> logScale;
+            dequantizedCoefficients = (dequantizedMagnitude ^ coefficientSign) - coefficientSign;
+            return quantized;
+        }
+
+        /// <inheritdoc/>
+        public static int Quantize(
+            int coefficient,
+            int rounding,
+            int quantizer,
+            int dequantizer,
+            int logScale,
+            out int dequantizedCoefficient)
+        {
+            int coefficientSign = coefficient >> 31;
+            int magnitude = (coefficient ^ coefficientSign) - coefficientSign;
+            int quantizedMagnitude = 0;
+
+            // The scalar tail keeps the same threshold, clamp, and fixed-point operation order as every vector lane.
+            if (((long)magnitude << (1 + logScale)) >= dequantizer)
+            {
+                int rounded = Math.Min(magnitude + rounding, short.MaxValue);
+                quantizedMagnitude = (rounded * quantizer) >> (16 - logScale);
+            }
+
+            int quantized = (quantizedMagnitude ^ coefficientSign) - coefficientSign;
+            int dequantizedMagnitude = (quantizedMagnitude * dequantizer) >> logScale;
+            dequantizedCoefficient = (dequantizedMagnitude ^ coefficientSign) - coefficientSign;
+            return quantized;
+        }
+    }
+}

src/ImageSharp/Formats/Heif/Av1/Pipeline/Quantizers/Av1ForwardQuantizer.cs

@@ -0,0 +1,209 @@
+// Copyright (c) Six Labors.
+// Licensed under the Six Labors Split License.
+
+using System.Runtime.CompilerServices;
+using System.Runtime.InteropServices;
+using System.Runtime.Intrinsics;
+using SixLabors.ImageSharp.Formats.Heif.Av1.Transform;
+
+namespace SixLabors.ImageSharp.Formats.Heif.Av1.Pipeline.Quantizers;
+
+/// <summary>
+/// Applies AV1 forward quantization to raster-order transform coefficients.
+/// </summary>
+internal static partial class Av1ForwardQuantizer
+{
+    /// <summary>
+    /// Quantizes one lossy transform block with libaom's fast no-matrix arithmetic.
+    /// </summary>
+    /// <param name="coefficients">The raster-order forward-transform coefficients.</param>
+    /// <param name="quantizedCoefficients">The raster-order entropy-coding coefficients.</param>
+    /// <param name="dequantizedCoefficients">The raster-order reconstruction coefficients.</param>
+    /// <param name="transformSize">The transform-block dimensions.</param>
+    /// <param name="transformType">The 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="bitDepth">The coded sample bit depth.</param>
+    /// <returns>The one-based end position in coefficient scan order.</returns>
+    public static ushort QuantizeLossy(
+        ReadOnlySpan<int> coefficients,
+        Span<int> quantizedCoefficients,
+        Span<int> dequantizedCoefficients,
+        Av1TransformSize transformSize,
+        Av1TransformType transformType,
+        int qIndex,
+        int dcDeltaQ,
+        int acDeltaQ,
+        Av1BitDepth bitDepth)
+        => QuantizeLossy<FastQuantizationOperator>(
+            coefficients,
+            quantizedCoefficients,
+            dequantizedCoefficients,
+            transformSize,
+            transformType,
+            qIndex,
+            dcDeltaQ,
+            acDeltaQ,
+            bitDepth);
+
+    /// <summary>
+    /// Applies a closed generic quantization operator across the widest available hardware widths.
+    /// </summary>
+    internal static ushort QuantizeLossy<TOperator>(
+        ReadOnlySpan<int> coefficients,
+        Span<int> quantizedCoefficients,
+        Span<int> dequantizedCoefficients,
+        Av1TransformSize transformSize,
+        Av1TransformType transformType,
+        int qIndex,
+        int dcDeltaQ,
+        int acDeltaQ,
+        Av1BitDepth bitDepth)
+        where TOperator : struct, IForwardQuantizationOperator
+    {
+        int coefficientCount = transformSize.GetAdjusted().GetSize2d();
+        int logScale = transformSize.GetScale();
+        int dcDequantizer = Av1QuantizationLookup.GetDcQuant(qIndex, dcDeltaQ, bitDepth);
+        int acDequantizer = Av1QuantizationLookup.GetAcQuant(qIndex, acDeltaQ, bitDepth);
+        int dcQuantizer = (1 << 16) / dcDequantizer;
+        int acQuantizer = (1 << 16) / acDequantizer;
+        int dcRounding = RoundPowerOfTwo((64 * dcDequantizer) >> 7, logScale);
+        int acRounding = RoundPowerOfTwo((64 * acDequantizer) >> 7, logScale);
+
+        ref int sourceBase = ref MemoryMarshal.GetReference(coefficients);
+        ref int quantizedBase = ref MemoryMarshal.GetReference(quantizedCoefficients);
+        ref int dequantizedBase = ref MemoryMarshal.GetReference(dequantizedCoefficients);
+
+        // Raster coefficient zero is the only DC coefficient, so it is encoded once with the plane's DC constants
+        // before the AC-only SIMD traversal begins.
+        Unsafe.Add(ref quantizedBase, 0) = TOperator.Quantize(
+            Unsafe.Add(ref sourceBase, 0),
+            dcRounding,
+            dcQuantizer,
+            dcDequantizer,
+            logScale,
+            out Unsafe.Add(ref dequantizedBase, 0));
+
+        int i = 1;
+
+        // Raster traversal keeps loads and stores contiguous. Each narrower tier resumes at the shared offset left by
+        // the previous tier, retaining vector execution for the widest possible remainder without overlapping lanes.
+        if (Vector512.IsHardwareAccelerated)
+        {
+            nuint vectorCount = coefficients[i..coefficientCount].Vector512Count<int>();
+
+            if (vectorCount > 0)
+            {
+                // Width-specific constants are created only when at least one complete vector remains.
+                Vector512<int> rounding = Vector512.Create(acRounding);
+                Vector512<int> quantizer = Vector512.Create(acQuantizer);
+                Vector512<int> dequantizer = Vector512.Create(acDequantizer);
+
+                for (; vectorCount > 0; vectorCount--, i += Vector512<int>.Count)
+                {
+                    Vector512<int> source = Unsafe.As<int, Vector512<int>>(ref Unsafe.Add(ref sourceBase, i));
+                    Vector512<int> quantized = TOperator.Quantize(
+                        source,
+                        rounding,
+                        quantizer,
+                        dequantizer,
+                        logScale,
+                        out Vector512<int> dequantized);
+
+                    Unsafe.As<int, Vector512<int>>(ref Unsafe.Add(ref quantizedBase, i)) = quantized;
+                    Unsafe.As<int, Vector512<int>>(ref Unsafe.Add(ref dequantizedBase, i)) = dequantized;
+                }
+            }
+        }
+
+        if (Vector256.IsHardwareAccelerated)
+        {
+            nuint vectorCount = coefficients[i..coefficientCount].Vector256Count<int>();
+
+            if (vectorCount > 0)
+            {
+                // The shared offset exposes only the remainder left by wider lanes, so no coefficient is revisited.
+                Vector256<int> rounding = Vector256.Create(acRounding);
+                Vector256<int> quantizer = Vector256.Create(acQuantizer);
+                Vector256<int> dequantizer = Vector256.Create(acDequantizer);
+
+                for (; vectorCount > 0; vectorCount--, i += Vector256<int>.Count)
+                {
+                    Vector256<int> source = Unsafe.As<int, Vector256<int>>(ref Unsafe.Add(ref sourceBase, i));
+                    Vector256<int> quantized = TOperator.Quantize(
+                        source,
+                        rounding,
+                        quantizer,
+                        dequantizer,
+                        logScale,
+                        out Vector256<int> dequantized);
+
+                    Unsafe.As<int, Vector256<int>>(ref Unsafe.Add(ref quantizedBase, i)) = quantized;
+                    Unsafe.As<int, Vector256<int>>(ref Unsafe.Add(ref dequantizedBase, i)) = dequantized;
+                }
+            }
+        }
+
+        if (Vector128.IsHardwareAccelerated)
+        {
+            nuint vectorCount = coefficients[i..coefficientCount].Vector128Count<int>();
+
+            if (vectorCount > 0)
+            {
+                // The final SIMD tier consumes complete four-lane groups and leaves fewer than four coefficients.
+                Vector128<int> rounding = Vector128.Create(acRounding);
+                Vector128<int> quantizer = Vector128.Create(acQuantizer);
+                Vector128<int> dequantizer = Vector128.Create(acDequantizer);
+
+                for (; vectorCount > 0; vectorCount--, i += Vector128<int>.Count)
+                {
+                    Vector128<int> source = Unsafe.As<int, Vector128<int>>(ref Unsafe.Add(ref sourceBase, i));
+                    Vector128<int> quantized = TOperator.Quantize(
+                        source,
+                        rounding,
+                        quantizer,
+                        dequantizer,
+                        logScale,
+                        out Vector128<int> dequantized);
+
+                    Unsafe.As<int, Vector128<int>>(ref Unsafe.Add(ref quantizedBase, i)) = quantized;
+                    Unsafe.As<int, Vector128<int>>(ref Unsafe.Add(ref dequantizedBase, i)) = dequantized;
+                }
+            }
+        }
+
+        // On SIMD-capable systems this loop receives only the final zero-to-three AC coefficients.
+        for (; i < coefficientCount; i++)
+        {
+            Unsafe.Add(ref quantizedBase, i) = TOperator.Quantize(
+                Unsafe.Add(ref sourceBase, i),
+                acRounding,
+                acQuantizer,
+                acDequantizer,
+                logScale,
+                out Unsafe.Add(ref dequantizedBase, i));
+        }
+
+        ReadOnlySpan<short> scan = Av1ScanOrderConstants.GetScanOrder(transformSize, transformType).Scan;
+
+        // Quantized coefficients remain in raster order for reconstruction and entropy coding. A reverse scan finds
+        // the final nonzero position without another buffer, and normally exits on its first iteration at high quality.
+        for (int scanIndex = coefficientCount - 1; scanIndex >= 0; scanIndex--)
+        {
+            if (Unsafe.Add(ref quantizedBase, scan[scanIndex]) != 0)
+            {
+                return (ushort)(scanIndex + 1);
+            }
+        }
+
+        return 0;
+    }
+
+    /// <summary>
+    /// Applies libaom's positive round-power-of-two operation to one quantizer constant.
+    /// </summary>
+    [MethodImpl(MethodImplOptions.AggressiveInlining)]
+    private static int RoundPowerOfTwo(int value, int shift)
+        => shift == 0 ? value : (value + (1 << (shift - 1))) >> shift;
+}

src/ImageSharp/Formats/Heif/Av1/Prediction/Av1DcIntraPredictor.Operator.cs

@@ -316,8 +316,8 @@ internal static class Av1DcIntraPredictor
             // length without a separate dispatch tree and leaves only an incomplete final vector to scalar code.
             if (Vector512.IsHardwareAccelerated)
             {
-                int oneVectorFromEnd = samples.Length - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(samples.Length - index);
-                for (; index <= oneVectorFromEnd; index += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, index += Vector512<byte>.Count)
                 {
                     sum += TOperator.Sum(Vector512.LoadUnsafe(ref samplesBase, (nuint)index));
                 }
@@ -325,8 +325,8 @@ internal static class Av1DcIntraPredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int oneVectorFromEnd = samples.Length - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(samples.Length - index);
-                for (; index <= oneVectorFromEnd; index += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, index += Vector256<byte>.Count)
                 {
                     sum += TOperator.Sum(Vector256.LoadUnsafe(ref samplesBase, (nuint)index));
                 }
@@ -334,8 +334,8 @@ internal static class Av1DcIntraPredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int oneVectorFromEnd = samples.Length - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(samples.Length - index);
-                for (; index <= oneVectorFromEnd; index += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, index += Vector128<byte>.Count)
                 {
                     sum += TOperator.Sum(Vector128.LoadUnsafe(ref samplesBase, (nuint)index));
                 }
@@ -362,8 +362,8 @@ internal static class Av1DcIntraPredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int oneVectorFromEnd = samples.Length - Vector512<short>.Count;
+                nuint vectorCount = Numerics.Vector512Count<short>(samples.Length - index);
-                for (; index <= oneVectorFromEnd; index += Vector512<short>.Count)
+                for (; vectorCount > 0; vectorCount--, index += Vector512<short>.Count)
                 {
                     sum += TOperator.Sum(Vector512.LoadUnsafe(ref samplesBase, (nuint)index));
                 }
@@ -371,8 +371,8 @@ internal static class Av1DcIntraPredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int oneVectorFromEnd = samples.Length - Vector256<short>.Count;
+                nuint vectorCount = Numerics.Vector256Count<short>(samples.Length - index);
-                for (; index <= oneVectorFromEnd; index += Vector256<short>.Count)
+                for (; vectorCount > 0; vectorCount--, index += Vector256<short>.Count)
                 {
                     sum += TOperator.Sum(Vector256.LoadUnsafe(ref samplesBase, (nuint)index));
                 }
@@ -380,8 +380,8 @@ internal static class Av1DcIntraPredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int oneVectorFromEnd = samples.Length - Vector128<short>.Count;
+                nuint vectorCount = Numerics.Vector128Count<short>(samples.Length - index);
-                for (; index <= oneVectorFromEnd; index += Vector128<short>.Count)
+                for (; vectorCount > 0; vectorCount--, index += Vector128<short>.Count)
                 {
                     sum += TOperator.Sum(Vector128.LoadUnsafe(ref samplesBase, (nuint)index));
                 }

src/ImageSharp/Formats/Heif/Av1/Prediction/Av1DirectionalIntraPredictor.Operations.cs

@@ -204,8 +204,8 @@ internal static partial class Av1DirectionalIntraPredictor
                 // left by the wider path, so the row is written once without requiring padded destination storage.
                 if (Vector512.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = validCount - Vector512<byte>.Count;
+                    nuint vectorCount = Numerics.Vector512Count<byte>(validCount - index);
-                    for (; index <= oneVectorFromEnd; index += Vector512<byte>.Count)
+                    for (; vectorCount > 0; vectorCount--, index += Vector512<byte>.Count)
                     {
                         Vector512<byte> left = Vector512.LoadUnsafe(ref referenceBase, (nuint)(basis + index));
                         Vector512<byte> right = Vector512.LoadUnsafe(ref referenceBase, (nuint)(basis + index + 1));
@@ -215,8 +215,8 @@ internal static partial class Av1DirectionalIntraPredictor
 
                 if (Vector256.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = validCount - Vector256<byte>.Count;
+                    nuint vectorCount = Numerics.Vector256Count<byte>(validCount - index);
-                    for (; index <= oneVectorFromEnd; index += Vector256<byte>.Count)
+                    for (; vectorCount > 0; vectorCount--, index += Vector256<byte>.Count)
                     {
                         Vector256<byte> left = Vector256.LoadUnsafe(ref referenceBase, (nuint)(basis + index));
                         Vector256<byte> right = Vector256.LoadUnsafe(ref referenceBase, (nuint)(basis + index + 1));
@@ -226,8 +226,8 @@ internal static partial class Av1DirectionalIntraPredictor
 
                 if (Vector128.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = validCount - Vector128<byte>.Count;
+                    nuint vectorCount = Numerics.Vector128Count<byte>(validCount - index);
-                    for (; index <= oneVectorFromEnd; index += Vector128<byte>.Count)
+                    for (; vectorCount > 0; vectorCount--, index += Vector128<byte>.Count)
                     {
                         Vector128<byte> left = Vector128.LoadUnsafe(ref referenceBase, (nuint)(basis + index));
                         Vector128<byte> right = Vector128.LoadUnsafe(ref referenceBase, (nuint)(basis + index + 1));
@@ -312,8 +312,8 @@ internal static partial class Av1DirectionalIntraPredictor
                 // weighted sum. The largest supported 12-bit sample therefore cannot overflow an intermediate lane.
                 if (Vector512.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = validCount - Vector512<short>.Count;
+                    nuint vectorCount = Numerics.Vector512Count<short>(validCount - index);
-                    for (; index <= oneVectorFromEnd; index += Vector512<short>.Count)
+                    for (; vectorCount > 0; vectorCount--, index += Vector512<short>.Count)
                     {
                         Vector512<short> left = Vector512.LoadUnsafe(ref referenceBase, (nuint)(basis + index));
                         Vector512<short> right = Vector512.LoadUnsafe(ref referenceBase, (nuint)(basis + index + 1));
@@ -323,8 +323,8 @@ internal static partial class Av1DirectionalIntraPredictor
 
                 if (Vector256.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = validCount - Vector256<short>.Count;
+                    nuint vectorCount = Numerics.Vector256Count<short>(validCount - index);
-                    for (; index <= oneVectorFromEnd; index += Vector256<short>.Count)
+                    for (; vectorCount > 0; vectorCount--, index += Vector256<short>.Count)
                     {
                         Vector256<short> left = Vector256.LoadUnsafe(ref referenceBase, (nuint)(basis + index));
                         Vector256<short> right = Vector256.LoadUnsafe(ref referenceBase, (nuint)(basis + index + 1));
@@ -334,8 +334,8 @@ internal static partial class Av1DirectionalIntraPredictor
 
                 if (Vector128.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = validCount - Vector128<short>.Count;
+                    nuint vectorCount = Numerics.Vector128Count<short>(validCount - index);
-                    for (; index <= oneVectorFromEnd; index += Vector128<short>.Count)
+                    for (; vectorCount > 0; vectorCount--, index += Vector128<short>.Count)
                     {
                         Vector128<short> left = Vector128.LoadUnsafe(ref referenceBase, (nuint)(basis + index));
                         Vector128<short> right = Vector128.LoadUnsafe(ref referenceBase, (nuint)(basis + index + 1));

src/ImageSharp/Formats/Heif/Av1/Prediction/Av1NonDirectionalIntraPredictor.Operator.cs

@@ -226,7 +226,7 @@ internal abstract partial class Av1NonDirectionalIntraPredictorBase
             // narrower paths consume any complete vectors left before the scalar tail handles the final columns.
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorizedColumns = width - (width % Vector512<byte>.Count);
+                int vectorizedColumns = (int)(Numerics.Vector512Count<byte>(width) * (nuint)Vector512<byte>.Count);
                 if (vectorizedColumns > 0)
                 {
                     Vector512<byte> topLeftVector = usesTopLeft ? Vector512.Create(topLeft) : default;
@@ -255,7 +255,7 @@ internal abstract partial class Av1NonDirectionalIntraPredictorBase
             if (Vector256.IsHardwareAccelerated)
             {
                 int remainingColumns = width - processedColumns;
-                int vectorizedColumns = remainingColumns - (remainingColumns % Vector256<byte>.Count);
+                int vectorizedColumns = (int)(Numerics.Vector256Count<byte>(remainingColumns) * (nuint)Vector256<byte>.Count);
                 if (vectorizedColumns > 0)
                 {
                     Vector256<byte> topLeftVector = usesTopLeft ? Vector256.Create(topLeft) : default;
@@ -285,7 +285,7 @@ internal abstract partial class Av1NonDirectionalIntraPredictorBase
             if (Vector128.IsHardwareAccelerated)
             {
                 int remainingColumns = width - processedColumns;
-                int vectorizedColumns = remainingColumns - (remainingColumns % Vector128<byte>.Count);
+                int vectorizedColumns = (int)(Numerics.Vector128Count<byte>(remainingColumns) * (nuint)Vector128<byte>.Count);
                 if (vectorizedColumns > 0)
                 {
                     Vector128<byte> topLeftVector = usesTopLeft ? Vector128.Create(topLeft) : default;
@@ -355,7 +355,7 @@ internal abstract partial class Av1NonDirectionalIntraPredictorBase
             // column, so the same width-progressive traversal is valid without inter-lane packing or saturation.
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorizedColumns = width - (width % Vector512<short>.Count);
+                int vectorizedColumns = (int)(Numerics.Vector512Count<short>(width) * (nuint)Vector512<short>.Count);
                 if (vectorizedColumns > 0)
                 {
                     Vector512<short> topLeftVector = usesTopLeft ? Vector512.Create(topLeft) : default;
@@ -384,7 +384,7 @@ internal abstract partial class Av1NonDirectionalIntraPredictorBase
             if (Vector256.IsHardwareAccelerated)
             {
                 int remainingColumns = width - processedColumns;
-                int vectorizedColumns = remainingColumns - (remainingColumns % Vector256<short>.Count);
+                int vectorizedColumns = (int)(Numerics.Vector256Count<short>(remainingColumns) * (nuint)Vector256<short>.Count);
                 if (vectorizedColumns > 0)
                 {
                     Vector256<short> topLeftVector = usesTopLeft ? Vector256.Create(topLeft) : default;
@@ -414,7 +414,7 @@ internal abstract partial class Av1NonDirectionalIntraPredictorBase
             if (Vector128.IsHardwareAccelerated)
             {
                 int remainingColumns = width - processedColumns;
-                int vectorizedColumns = remainingColumns - (remainingColumns % Vector128<short>.Count);
+                int vectorizedColumns = (int)(Numerics.Vector128Count<short>(remainingColumns) * (nuint)Vector128<short>.Count);
                 if (vectorizedColumns > 0)
                 {
                     Vector128<short> topLeftVector = usesTopLeft ? Vector128.Create(topLeft) : default;

src/ImageSharp/Formats/Heif/Av1/Prediction/Av1PalettePredictor.Operator.cs

@@ -212,33 +212,43 @@ internal static class Av1PalettePredictor
 
                 if (Vector512.IsHardwareAccelerated)
                 {
-                    Vector256<byte> palette256 = Vector256.Create(palette128, palette128);
+                    nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                    Vector512<byte> palette512 = Vector512.Create(palette256, palette256);
-                    int oneVectorFromEnd = width - Vector512<byte>.Count;
 
-                    for (; column <= oneVectorFromEnd; column += Vector512<byte>.Count)
+                    if (vectorCount > 0)
                     {
-                        Vector512<byte> indices = Vector512.LoadUnsafe(ref mapRow, (nuint)column);
+                        // Replicate the lookup table only when the row has a complete 64-lane batch.
-                        TOperator.Predict(palette512, indices).StoreUnsafe(ref destinationRow, (nuint)column);
+                        Vector256<byte> palette256 = Vector256.Create(palette128, palette128);
+                        Vector512<byte> palette512 = Vector512.Create(palette256, palette256);
+
+                        for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
+                        {
+                            Vector512<byte> indices = Vector512.LoadUnsafe(ref mapRow, (nuint)column);
+                            TOperator.Predict(palette512, indices).StoreUnsafe(ref destinationRow, (nuint)column);
+                        }
                     }
                 }
 
                 if (Vector256.IsHardwareAccelerated)
                 {
-                    Vector256<byte> palette256 = Vector256.Create(palette128, palette128);
+                    nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                    int oneVectorFromEnd = width - Vector256<byte>.Count;
 
-                    for (; column <= oneVectorFromEnd; column += Vector256<byte>.Count)
+                    if (vectorCount > 0)
                     {
-                        Vector256<byte> indices = Vector256.LoadUnsafe(ref mapRow, (nuint)column);
+                        // The narrower table is likewise materialized only for a complete 32-lane remainder.
-                        TOperator.Predict(palette256, indices).StoreUnsafe(ref destinationRow, (nuint)column);
+                        Vector256<byte> palette256 = Vector256.Create(palette128, palette128);
+
+                        for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
+                        {
+                            Vector256<byte> indices = Vector256.LoadUnsafe(ref mapRow, (nuint)column);
+                            TOperator.Predict(palette256, indices).StoreUnsafe(ref destinationRow, (nuint)column);
+                        }
                     }
                 }
 
                 if (Vector128.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = width - Vector128<byte>.Count;
+                    nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                    for (; column <= oneVectorFromEnd; column += Vector128<byte>.Count)
+                    for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                     {
                         Vector128<byte> indices = Vector128.LoadUnsafe(ref mapRow, (nuint)column);
                         TOperator.Predict(palette128, indices).StoreUnsafe(ref destinationRow, (nuint)column);
@@ -296,35 +306,45 @@ internal static class Av1PalettePredictor
 
                 if (Vector512.IsHardwareAccelerated)
                 {
-                    Vector256<byte> palette256 = Vector256.Create(palette128, palette128);
+                    nuint vectorCount = Numerics.Vector512Count<short>(width - column);
-                    Vector512<byte> palette512 = Vector512.Create(palette256, palette256);
-                    int oneVectorFromEnd = width - Vector512<short>.Count;
 
-                    for (; column <= oneVectorFromEnd; column += Vector512<short>.Count)
+                    if (vectorCount > 0)
                     {
-                        (Vector256<ushort> lower, Vector256<ushort> upper) = Vector256.Widen(Vector256.LoadUnsafe(ref mapRow, (nuint)column));
+                        // High-bit-depth output has half as many lanes, so gate table replication with that lane count.
-                        Vector512<ushort> indices = Vector512.Create(lower, upper);
+                        Vector256<byte> palette256 = Vector256.Create(palette128, palette128);
-                        TOperator.Predict(palette512, indices).StoreUnsafe(ref destinationRow, (nuint)column);
+                        Vector512<byte> palette512 = Vector512.Create(palette256, palette256);
+
+                        for (; vectorCount > 0; vectorCount--, column += Vector512<short>.Count)
+                        {
+                            (Vector256<ushort> lower, Vector256<ushort> upper) = Vector256.Widen(Vector256.LoadUnsafe(ref mapRow, (nuint)column));
+                            Vector512<ushort> indices = Vector512.Create(lower, upper);
+                            TOperator.Predict(palette512, indices).StoreUnsafe(ref destinationRow, (nuint)column);
+                        }
                     }
                 }
 
                 if (Vector256.IsHardwareAccelerated)
                 {
-                    Vector256<byte> palette256 = Vector256.Create(palette128, palette128);
+                    nuint vectorCount = Numerics.Vector256Count<short>(width - column);
-                    int oneVectorFromEnd = width - Vector256<short>.Count;
 
-                    for (; column <= oneVectorFromEnd; column += Vector256<short>.Count)
+                    if (vectorCount > 0)
                     {
-                        (Vector128<ushort> lower, Vector128<ushort> upper) = Vector128.Widen(Vector128.LoadUnsafe(ref mapRow, (nuint)column));
+                        // Avoid creating the 256-bit table when the remainder belongs entirely to narrower paths.
-                        Vector256<ushort> indices = Vector256.Create(lower, upper);
+                        Vector256<byte> palette256 = Vector256.Create(palette128, palette128);
-                        TOperator.Predict(palette256, indices).StoreUnsafe(ref destinationRow, (nuint)column);
+
+                        for (; vectorCount > 0; vectorCount--, column += Vector256<short>.Count)
+                        {
+                            (Vector128<ushort> lower, Vector128<ushort> upper) = Vector128.Widen(Vector128.LoadUnsafe(ref mapRow, (nuint)column));
+                            Vector256<ushort> indices = Vector256.Create(lower, upper);
+                            TOperator.Predict(palette256, indices).StoreUnsafe(ref destinationRow, (nuint)column);
+                        }
                     }
                 }
 
                 if (Vector128.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = width - Vector128<short>.Count;
+                    nuint vectorCount = Numerics.Vector128Count<short>(width - column);
-                    for (; column <= oneVectorFromEnd; column += Vector128<short>.Count)
+                    for (; vectorCount > 0; vectorCount--, column += Vector128<short>.Count)
                     {
                         ulong packedIndices = Unsafe.ReadUnaligned<ulong>(ref Unsafe.Add(ref mapRow, column));
                         Vector128<ushort> indices = Vector128.WidenLower(Vector128.CreateScalarUnsafe(packedIndices).AsByte());

src/ImageSharp/Formats/Heif/Av1/Prediction/Av1PredictionDecoder.cs

@@ -1689,11 +1689,11 @@ internal sealed class Av1PredictionDecoder
         // Unsigned 16-bit lanes therefore preserve every normative strength without widening to 32-bit vectors.
         if (Vector128.IsHardwareAccelerated)
         {
-            int eightSamplesFromEnd = outputCount - Vector128<short>.Count;
+            int vectorEnd = (int)(Numerics.Vector128Count<short>(outputCount) * (nuint)Vector128<short>.Count);
             switch (strength)
             {
                 case 1:
-                    for (; processed <= eightSamplesFromEnd; processed += Vector128<short>.Count)
+                    for (; processed < vectorEnd; processed += Vector128<short>.Count)
                     {
                         Vector128<ushort> source0 = Vector128.LoadUnsafe(ref edge, (nuint)(processed + 1)).AsUInt16();
                         Vector128<ushort> source1 = Vector128.LoadUnsafe(ref edge, (nuint)(processed + 2)).AsUInt16();
@@ -1703,7 +1703,7 @@ internal sealed class Av1PredictionDecoder
 
                     break;
                 case 2:
-                    for (; processed <= eightSamplesFromEnd; processed += Vector128<short>.Count)
+                    for (; processed < vectorEnd; processed += Vector128<short>.Count)
                     {
                         Vector128<ushort> source0 = Vector128.LoadUnsafe(ref edge, (nuint)(processed + 1)).AsUInt16();
                         Vector128<ushort> source1 = Vector128.LoadUnsafe(ref edge, (nuint)(processed + 2)).AsUInt16();
@@ -1713,7 +1713,7 @@ internal sealed class Av1PredictionDecoder
 
                     break;
                 default:
-                    for (; processed <= eightSamplesFromEnd; processed += Vector128<short>.Count)
+                    for (; processed < vectorEnd; processed += Vector128<short>.Count)
                     {
                         Vector128<ushort> source0 = Vector128.LoadUnsafe(ref edge, (nuint)processed).AsUInt16();
                         Vector128<ushort> source1 = Vector128.LoadUnsafe(ref edge, (nuint)(processed + 1)).AsUInt16();

src/ImageSharp/Formats/Heif/Av1/Prediction/ChromaFromLuma/Av1ChromaFromLumaContext.Operations.cs

@@ -42,7 +42,8 @@ internal partial class Av1ChromaFromLumaContext
 
                 if (Vector256.IsHardwareAccelerated)
                 {
-                    for (; column <= width - Vector256<byte>.Count; column += Vector256<byte>.Count)
+                    nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
+                    for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                     {
                         (Vector256<ushort> lower, Vector256<ushort> upper) = Vector256.Widen(Vector256.LoadUnsafe(ref inputRow, (nuint)column));
 
@@ -53,7 +54,8 @@ internal partial class Av1ChromaFromLumaContext
 
                 if (Vector128.IsHardwareAccelerated)
                 {
-                    for (; column <= width - Vector128<byte>.Count; column += Vector128<byte>.Count)
+                    nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
+                    for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                     {
                         (Vector128<ushort> lower, Vector128<ushort> upper) = Vector128.Widen(Vector128.LoadUnsafe(ref inputRow, (nuint)column));
 
@@ -103,7 +105,8 @@ internal partial class Av1ChromaFromLumaContext
 
             if (Avx2.IsSupported)
             {
-                for (; column <= width - Vector256<byte>.Count; column += Vector256<byte>.Count)
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<short> sum = Avx2.MultiplyAddAdjacent(Vector256.LoadUnsafe(ref inputRow, (nuint)column), ones256);
                     if (this.subY)
@@ -117,7 +120,8 @@ internal partial class Av1ChromaFromLumaContext
 
             if (Vector128.IsHardwareAccelerated)
             {
-                for (; column <= width - Vector128<byte>.Count; column += Vector128<byte>.Count)
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<short> sum = PairSum(Vector128.LoadUnsafe(ref inputRow, (nuint)column), ones128);
                     if (this.subY)
@@ -193,7 +197,8 @@ internal partial class Av1ChromaFromLumaContext
 
                 if (Vector256.IsHardwareAccelerated)
                 {
-                    for (; column <= width - Vector256<short>.Count; column += Vector256<short>.Count)
+                    nuint vectorCount = Numerics.Vector256Count<short>(width - column);
+                    for (; vectorCount > 0; vectorCount--, column += Vector256<short>.Count)
                     {
                         (Vector256.LoadUnsafe(ref inputRow, (nuint)column) << 3).StoreUnsafe(ref outputRow, (nuint)column);
                     }
@@ -201,7 +206,8 @@ internal partial class Av1ChromaFromLumaContext
 
                 if (Vector128.IsHardwareAccelerated)
                 {
-                    for (; column <= width - Vector128<short>.Count; column += Vector128<short>.Count)
+                    nuint vectorCount = Numerics.Vector128Count<short>(width - column);
+                    for (; vectorCount > 0; vectorCount--, column += Vector128<short>.Count)
                     {
                         (Vector128.LoadUnsafe(ref inputRow, (nuint)column) << 3).StoreUnsafe(ref outputRow, (nuint)column);
                     }
@@ -237,7 +243,8 @@ internal partial class Av1ChromaFromLumaContext
 
             if (Vector256.IsHardwareAccelerated)
             {
-                for (; column <= width - Vector256<short>.Count; column += Vector256<short>.Count)
+                nuint vectorCount = Numerics.Vector256Count<short>(width - column);
+                for (; vectorCount > 0; vectorCount--, column += Vector256<short>.Count)
                 {
                     Vector256<int> sum = Vector256_.MultiplyAddAdjacent(Vector256.LoadUnsafe(ref inputRow, (nuint)column), ones256);
                     if (this.subY)
@@ -251,7 +258,8 @@ internal partial class Av1ChromaFromLumaContext
 
             if (Vector128.IsHardwareAccelerated)
             {
-                for (; column <= width - Vector128<short>.Count; column += Vector128<short>.Count)
+                nuint vectorCount = Numerics.Vector128Count<short>(width - column);
+                for (; vectorCount > 0; vectorCount--, column += Vector128<short>.Count)
                 {
                     Vector128<int> sum = Vector128_.MultiplyAddAdjacent(Vector128.LoadUnsafe(ref inputRow, (nuint)column), ones128);
                     if (this.subY)
@@ -326,7 +334,8 @@ internal partial class Av1ChromaFromLumaContext
             {
                 int rowOffset = row * BufferLine;
                 int column = 0;
-                for (; column <= width - Vector128<short>.Count; column += Vector128<short>.Count)
+                nuint vectorCount = Numerics.Vector128Count<short>(width - column);
+                for (; vectorCount > 0; vectorCount--, column += Vector128<short>.Count)
                 {
                     (Vector128<int> lower, Vector128<int> upper) = Vector128.Widen(Vector128.LoadUnsafe(ref bufferBase, (nuint)(rowOffset + column)));
 
@@ -377,7 +386,8 @@ internal partial class Av1ChromaFromLumaContext
             {
                 int rowOffset = row * BufferLine;
                 int column = 0;
-                for (; column <= width - Vector128<short>.Count; column += Vector128<short>.Count)
+                nuint vectorCount = Numerics.Vector128Count<short>(width - column);
+                for (; vectorCount > 0; vectorCount--, column += Vector128<short>.Count)
                 {
                     (Vector128.LoadUnsafe(ref bufferBase, (nuint)(rowOffset + column)) - average).StoreUnsafe(ref bufferBase, (nuint)(rowOffset + column));
                 }

src/ImageSharp/Formats/Heif/Av1/Prediction/ChromaFromLuma/Av1ChromaFromLumaPredictor.Operator.cs

@@ -216,8 +216,8 @@ internal static partial class Av1ChromaFromLumaPredictor
 
                 if (Vector512.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = width - Vector512<short>.Count;
+                    nuint vectorCount = Numerics.Vector512Count<short>(width - column);
-                    for (; column <= oneVectorFromEnd; column += Vector512<short>.Count)
+                    for (; vectorCount > 0; vectorCount--, column += Vector512<short>.Count)
                     {
                         Vector512<short> prediction = TOperator.Predict(Vector512.LoadUnsafe(ref lumaRow, (nuint)column), dc, alphaQ3, byte.MaxValue);
                         Vector256<byte> packed = Vector512.Narrow(prediction.AsUInt16(), Vector512<ushort>.Zero).GetLower();
@@ -227,8 +227,8 @@ internal static partial class Av1ChromaFromLumaPredictor
 
                 if (Vector256.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = width - Vector256<short>.Count;
+                    nuint vectorCount = Numerics.Vector256Count<short>(width - column);
-                    for (; column <= oneVectorFromEnd; column += Vector256<short>.Count)
+                    for (; vectorCount > 0; vectorCount--, column += Vector256<short>.Count)
                     {
                         Vector256<short> prediction = TOperator.Predict(Vector256.LoadUnsafe(ref lumaRow, (nuint)column), dc, alphaQ3, byte.MaxValue);
                         Vector128<byte> packed = Vector256.Narrow(prediction.AsUInt16(), Vector256<ushort>.Zero).GetLower();
@@ -238,8 +238,8 @@ internal static partial class Av1ChromaFromLumaPredictor
 
                 if (Vector128.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = width - Vector128<short>.Count;
+                    nuint vectorCount = Numerics.Vector128Count<short>(width - column);
-                    for (; column <= oneVectorFromEnd; column += Vector128<short>.Count)
+                    for (; vectorCount > 0; vectorCount--, column += Vector128<short>.Count)
                     {
                         Vector128<short> prediction = TOperator.Predict(Vector128.LoadUnsafe(ref lumaRow, (nuint)column), dc, alphaQ3, byte.MaxValue);
                         Vector64<byte> packed = Vector128.Narrow(prediction.AsUInt16(), Vector128<ushort>.Zero).GetLower();
@@ -282,8 +282,8 @@ internal static partial class Av1ChromaFromLumaPredictor
 
                 if (Vector512.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = width - Vector512<short>.Count;
+                    nuint vectorCount = Numerics.Vector512Count<short>(width - column);
-                    for (; column <= oneVectorFromEnd; column += Vector512<short>.Count)
+                    for (; vectorCount > 0; vectorCount--, column += Vector512<short>.Count)
                     {
                         TOperator.Predict(Vector512.LoadUnsafe(ref lumaRow, (nuint)column), dc, alphaQ3, maximum).StoreUnsafe(ref destinationRow, (nuint)column);
                     }
@@ -291,8 +291,8 @@ internal static partial class Av1ChromaFromLumaPredictor
 
                 if (Vector256.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = width - Vector256<short>.Count;
+                    nuint vectorCount = Numerics.Vector256Count<short>(width - column);
-                    for (; column <= oneVectorFromEnd; column += Vector256<short>.Count)
+                    for (; vectorCount > 0; vectorCount--, column += Vector256<short>.Count)
                     {
                         TOperator.Predict(Vector256.LoadUnsafe(ref lumaRow, (nuint)column), dc, alphaQ3, maximum).StoreUnsafe(ref destinationRow, (nuint)column);
                     }
@@ -300,8 +300,8 @@ internal static partial class Av1ChromaFromLumaPredictor
 
                 if (Vector128.IsHardwareAccelerated)
                 {
-                    int oneVectorFromEnd = width - Vector128<short>.Count;
+                    nuint vectorCount = Numerics.Vector128Count<short>(width - column);
-                    for (; column <= oneVectorFromEnd; column += Vector128<short>.Count)
+                    for (; vectorCount > 0; vectorCount--, column += Vector128<short>.Count)
                     {
                         TOperator.Predict(Vector128.LoadUnsafe(ref lumaRow, (nuint)column), dc, alphaQ3, maximum).StoreUnsafe(ref destinationRow, (nuint)column);
                     }

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1CompoundAveragePredictor.cs

@@ -60,8 +60,8 @@ internal static partial class Av1CompoundAveragePredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<byte> firstVector = Vector512.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector512<byte> secondVector = Vector512.LoadUnsafe(ref secondReference, (nuint)column);
@@ -71,8 +71,8 @@ internal static partial class Av1CompoundAveragePredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<byte> firstVector = Vector256.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector256<byte> secondVector = Vector256.LoadUnsafe(ref secondReference, (nuint)column);
@@ -82,8 +82,8 @@ internal static partial class Av1CompoundAveragePredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<byte> firstVector = Vector128.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector128<byte> secondVector = Vector128.LoadUnsafe(ref secondReference, (nuint)column);
@@ -145,8 +145,8 @@ internal static partial class Av1CompoundAveragePredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<ushort>.Count;
+                nuint vectorCount = Numerics.Vector512Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector512<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<ushort>.Count)
                 {
                     Vector512<ushort> firstVector = Vector512.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector512<ushort> secondVector = Vector512.LoadUnsafe(ref secondReference, (nuint)column);
@@ -156,8 +156,8 @@ internal static partial class Av1CompoundAveragePredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<ushort>.Count;
+                nuint vectorCount = Numerics.Vector256Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector256<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<ushort>.Count)
                 {
                     Vector256<ushort> firstVector = Vector256.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector256<ushort> secondVector = Vector256.LoadUnsafe(ref secondReference, (nuint)column);
@@ -167,8 +167,8 @@ internal static partial class Av1CompoundAveragePredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<ushort>.Count;
+                nuint vectorCount = Numerics.Vector128Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector128<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<ushort>.Count)
                 {
                     Vector128<ushort> firstVector = Vector128.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector128<ushort> secondVector = Vector128.LoadUnsafe(ref secondReference, (nuint)column);

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1CompoundDistanceWeightedPredictor.cs

@@ -62,8 +62,8 @@ internal static partial class Av1CompoundDistanceWeightedPredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<byte> firstVector = Vector512.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector512<byte> secondVector = Vector512.LoadUnsafe(ref secondReference, (nuint)column);
@@ -73,8 +73,8 @@ internal static partial class Av1CompoundDistanceWeightedPredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<byte> firstVector = Vector256.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector256<byte> secondVector = Vector256.LoadUnsafe(ref secondReference, (nuint)column);
@@ -84,8 +84,8 @@ internal static partial class Av1CompoundDistanceWeightedPredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<byte> firstVector = Vector128.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector128<byte> secondVector = Vector128.LoadUnsafe(ref secondReference, (nuint)column);
@@ -147,8 +147,8 @@ internal static partial class Av1CompoundDistanceWeightedPredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<ushort>.Count;
+                nuint vectorCount = Numerics.Vector512Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector512<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<ushort>.Count)
                 {
                     Vector512<ushort> firstVector = Vector512.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector512<ushort> secondVector = Vector512.LoadUnsafe(ref secondReference, (nuint)column);
@@ -158,8 +158,8 @@ internal static partial class Av1CompoundDistanceWeightedPredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<ushort>.Count;
+                nuint vectorCount = Numerics.Vector256Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector256<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<ushort>.Count)
                 {
                     Vector256<ushort> firstVector = Vector256.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector256<ushort> secondVector = Vector256.LoadUnsafe(ref secondReference, (nuint)column);
@@ -169,8 +169,8 @@ internal static partial class Av1CompoundDistanceWeightedPredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<ushort>.Count;
+                nuint vectorCount = Numerics.Vector128Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector128<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<ushort>.Count)
                 {
                     Vector128<ushort> firstVector = Vector128.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector128<ushort> secondVector = Vector128.LoadUnsafe(ref secondReference, (nuint)column);

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1CompoundInterPredictor.cs

@@ -418,8 +418,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<byte> samples = Vector512.LoadUnsafe(ref sourceRow, (nuint)column);
                     TOperator.Copy(samples, roundBits, roundOffset, out Vector512<ushort> lower, out Vector512<ushort> upper);
@@ -430,8 +430,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<byte> samples = Vector256.LoadUnsafe(ref sourceRow, (nuint)column);
                     TOperator.Copy(samples, roundBits, roundOffset, out Vector256<ushort> lower, out Vector256<ushort> upper);
@@ -442,8 +442,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<byte> samples = Vector128.LoadUnsafe(ref sourceRow, (nuint)column);
                     TOperator.Copy(samples, roundBits, roundOffset, out Vector128<ushort> lower, out Vector128<ushort> upper);
@@ -487,8 +487,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<ushort>.Count;
+                nuint vectorCount = Numerics.Vector512Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector512<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<ushort>.Count)
                 {
                     Vector512<ushort> samples = Vector512.LoadUnsafe(ref sourceRow, (nuint)column);
                     TOperator.CopyHighBitDepth(samples, roundBits, roundOffset)
@@ -498,8 +498,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<ushort>.Count;
+                nuint vectorCount = Numerics.Vector256Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector256<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<ushort>.Count)
                 {
                     Vector256<ushort> samples = Vector256.LoadUnsafe(ref sourceRow, (nuint)column);
                     TOperator.CopyHighBitDepth(samples, roundBits, roundOffset)
@@ -509,8 +509,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<ushort>.Count;
+                nuint vectorCount = Numerics.Vector128Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector128<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<ushort>.Count)
                 {
                     Vector128<ushort> samples = Vector128.LoadUnsafe(ref sourceRow, (nuint)column);
                     TOperator.CopyHighBitDepth(samples, roundBits, roundOffset)
@@ -562,8 +562,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -587,8 +587,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -612,8 +612,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -683,8 +683,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<ushort>.Count;
+                nuint vectorCount = Numerics.Vector512Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector512<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<ushort>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -703,8 +703,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<ushort>.Count;
+                nuint vectorCount = Numerics.Vector256Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector256<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<ushort>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -723,8 +723,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<ushort>.Count;
+                nuint vectorCount = Numerics.Vector128Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector128<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<ushort>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -797,8 +797,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -820,8 +820,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -843,8 +843,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -884,8 +884,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<short>.Count;
+                nuint vectorCount = Numerics.Vector512Count<short>(width - column);
-                for (; column <= vectorEnd; column += Vector512<short>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<short>.Count)
                 {
                     Convolve(
                         ref scratchRow,
@@ -903,8 +903,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<short>.Count;
+                nuint vectorCount = Numerics.Vector256Count<short>(width - column);
-                for (; column <= vectorEnd; column += Vector256<short>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<short>.Count)
                 {
                     Convolve(
                         ref scratchRow,
@@ -922,8 +922,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<short>.Count;
+                nuint vectorCount = Numerics.Vector128Count<short>(width - column);
-                for (; column <= vectorEnd; column += Vector128<short>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<short>.Count)
                 {
                     Convolve(
                         ref scratchRow,
@@ -1000,8 +1000,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<ushort>.Count;
+                nuint vectorCount = Numerics.Vector512Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector512<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<ushort>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -1020,8 +1020,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<ushort>.Count;
+                nuint vectorCount = Numerics.Vector256Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector256<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<ushort>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -1040,8 +1040,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<ushort>.Count;
+                nuint vectorCount = Numerics.Vector128Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector128<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<ushort>.Count)
                 {
                     Convolve(
                         ref sourceRow,
@@ -1079,8 +1079,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<short>.Count;
+                nuint vectorCount = Numerics.Vector512Count<short>(width - column);
-                for (; column <= vectorEnd; column += Vector512<short>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<short>.Count)
                 {
                     Convolve(
                         ref scratchRow,
@@ -1099,8 +1099,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<short>.Count;
+                nuint vectorCount = Numerics.Vector256Count<short>(width - column);
-                for (; column <= vectorEnd; column += Vector256<short>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<short>.Count)
                 {
                     Convolve(
                         ref scratchRow,
@@ -1119,8 +1119,8 @@ internal static partial class Av1CompoundInterPredictor
 
             if (useSimd && Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<short>.Count;
+                nuint vectorCount = Numerics.Vector128Count<short>(width - column);
-                for (; column <= vectorEnd; column += Vector128<short>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<short>.Count)
                 {
                     Convolve(
                         ref scratchRow,

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1CompoundIntermediateAveragePredictor.cs

@@ -68,8 +68,8 @@ internal static partial class Av1CompoundIntermediateAveragePredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<ushort> first0 = Vector512.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector512<ushort> first1 = Vector512.LoadUnsafe(ref firstReference, (nuint)(column + Vector512<ushort>.Count));
@@ -82,8 +82,8 @@ internal static partial class Av1CompoundIntermediateAveragePredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<ushort> first0 = Vector256.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector256<ushort> first1 = Vector256.LoadUnsafe(ref firstReference, (nuint)(column + Vector256<ushort>.Count));
@@ -96,8 +96,8 @@ internal static partial class Av1CompoundIntermediateAveragePredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<ushort> first0 = Vector128.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector128<ushort> first1 = Vector128.LoadUnsafe(
@@ -177,8 +177,8 @@ internal static partial class Av1CompoundIntermediateAveragePredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<ushort>.Count;
+                nuint vectorCount = Numerics.Vector512Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector512<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<ushort>.Count)
                 {
                     Vector512<ushort> firstVector = Vector512.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector512<ushort> secondVector = Vector512.LoadUnsafe(ref secondReference, (nuint)column);
@@ -189,8 +189,8 @@ internal static partial class Av1CompoundIntermediateAveragePredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<ushort>.Count;
+                nuint vectorCount = Numerics.Vector256Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector256<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<ushort>.Count)
                 {
                     Vector256<ushort> firstVector = Vector256.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector256<ushort> secondVector = Vector256.LoadUnsafe(ref secondReference, (nuint)column);
@@ -201,8 +201,8 @@ internal static partial class Av1CompoundIntermediateAveragePredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<ushort>.Count;
+                nuint vectorCount = Numerics.Vector128Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector128<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<ushort>.Count)
                 {
                     Vector128<ushort> firstVector = Vector128.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector128<ushort> secondVector = Vector128.LoadUnsafe(ref secondReference, (nuint)column);

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1CompoundIntermediateDifferenceWeightedMaskBuilder.cs

@@ -74,8 +74,8 @@ internal static partial class Av1CompoundIntermediateDifferenceWeightedMaskBuild
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<ushort> first0 = Vector512.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector512<ushort> first1 = Vector512.LoadUnsafe(ref firstReference, (nuint)(column + Vector512<ushort>.Count));
@@ -88,8 +88,8 @@ internal static partial class Av1CompoundIntermediateDifferenceWeightedMaskBuild
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<ushort> first0 = Vector256.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector256<ushort> first1 = Vector256.LoadUnsafe(ref firstReference, (nuint)(column + Vector256<ushort>.Count));
@@ -102,8 +102,8 @@ internal static partial class Av1CompoundIntermediateDifferenceWeightedMaskBuild
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<ushort> first0 = Vector128.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector128<ushort> first1 = Vector128.LoadUnsafe(

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1CompoundIntermediateDistanceWeightedPredictor.cs

@@ -74,8 +74,8 @@ internal static partial class Av1CompoundIntermediateDistanceWeightedPredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<ushort> first0 = Vector512.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector512<ushort> first1 = Vector512.LoadUnsafe(ref firstReference, (nuint)(column + Vector512<ushort>.Count));
@@ -95,8 +95,8 @@ internal static partial class Av1CompoundIntermediateDistanceWeightedPredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<ushort> first0 = Vector256.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector256<ushort> first1 = Vector256.LoadUnsafe(ref firstReference, (nuint)(column + Vector256<ushort>.Count));
@@ -116,8 +116,8 @@ internal static partial class Av1CompoundIntermediateDistanceWeightedPredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<ushort> first0 = Vector128.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector128<ushort> first1 = Vector128.LoadUnsafe(
@@ -215,8 +215,8 @@ internal static partial class Av1CompoundIntermediateDistanceWeightedPredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<ushort>.Count;
+                nuint vectorCount = Numerics.Vector512Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector512<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<ushort>.Count)
                 {
                     Vector512<ushort> firstVector = Vector512.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector512<ushort> secondVector = Vector512.LoadUnsafe(ref secondReference, (nuint)column);
@@ -233,8 +233,8 @@ internal static partial class Av1CompoundIntermediateDistanceWeightedPredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<ushort>.Count;
+                nuint vectorCount = Numerics.Vector256Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector256<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<ushort>.Count)
                 {
                     Vector256<ushort> firstVector = Vector256.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector256<ushort> secondVector = Vector256.LoadUnsafe(ref secondReference, (nuint)column);
@@ -251,8 +251,8 @@ internal static partial class Av1CompoundIntermediateDistanceWeightedPredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<ushort>.Count;
+                nuint vectorCount = Numerics.Vector128Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector128<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<ushort>.Count)
                 {
                     Vector128<ushort> firstVector = Vector128.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector128<ushort> secondVector = Vector128.LoadUnsafe(ref secondReference, (nuint)column);

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1CompoundIntermediateMaskBlendPredictor.cs

@@ -82,8 +82,8 @@ internal static partial class Av1CompoundIntermediateMaskBlendPredictor
             {
                 ref byte maskReference = ref MemoryMarshal.GetReference(mask);
                 int maskRowOffset = row * maskStride;
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<ushort> first0 = Vector512.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector512<ushort> first1 = Vector512.LoadUnsafe(ref firstReference, (nuint)(column + Vector512<ushort>.Count));
@@ -99,8 +99,8 @@ internal static partial class Av1CompoundIntermediateMaskBlendPredictor
             {
                 ref byte maskReference = ref MemoryMarshal.GetReference(mask);
                 int maskRowOffset = row * maskStride;
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<ushort> first0 = Vector256.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector256<ushort> first1 = Vector256.LoadUnsafe(ref firstReference, (nuint)(column + Vector256<ushort>.Count));
@@ -116,8 +116,8 @@ internal static partial class Av1CompoundIntermediateMaskBlendPredictor
             {
                 ref byte maskReference = ref MemoryMarshal.GetReference(mask);
                 int maskRowOffset = row * maskStride;
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<ushort> first0 = Vector128.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector128<ushort> first1 = Vector128.LoadUnsafe(
@@ -221,8 +221,8 @@ internal static partial class Av1CompoundIntermediateMaskBlendPredictor
             {
                 ref byte maskReference = ref MemoryMarshal.GetReference(mask);
                 int maskRowOffset = row * maskStride;
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<ushort> first0 = Vector512.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector512<ushort> first1 = Vector512.LoadUnsafe(
@@ -261,8 +261,8 @@ internal static partial class Av1CompoundIntermediateMaskBlendPredictor
             {
                 ref byte maskReference = ref MemoryMarshal.GetReference(mask);
                 int maskRowOffset = row * maskStride;
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<ushort> first0 = Vector256.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector256<ushort> first1 = Vector256.LoadUnsafe(
@@ -301,8 +301,8 @@ internal static partial class Av1CompoundIntermediateMaskBlendPredictor
             {
                 ref byte maskReference = ref MemoryMarshal.GetReference(mask);
                 int maskRowOffset = row * maskStride;
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<ushort> first0 = Vector128.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector128<ushort> first1 = Vector128.LoadUnsafe(

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1CompoundMaskBlendPredictor.cs

@@ -64,8 +64,8 @@ internal static partial class Av1CompoundMaskBlendPredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<byte> firstVector = Vector512.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector512<byte> secondVector = Vector512.LoadUnsafe(ref secondReference, (nuint)column);
@@ -76,8 +76,8 @@ internal static partial class Av1CompoundMaskBlendPredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<byte> firstVector = Vector256.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector256<byte> secondVector = Vector256.LoadUnsafe(ref secondReference, (nuint)column);
@@ -88,8 +88,8 @@ internal static partial class Av1CompoundMaskBlendPredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<byte> firstVector = Vector128.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector128<byte> secondVector = Vector128.LoadUnsafe(ref secondReference, (nuint)column);
@@ -154,8 +154,8 @@ internal static partial class Av1CompoundMaskBlendPredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<ushort>.Count;
+                nuint vectorCount = Numerics.Vector512Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector512<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<ushort>.Count)
                 {
                     Vector512<ushort> firstVector = Vector512.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector512<ushort> secondVector = Vector512.LoadUnsafe(ref secondReference, (nuint)column);
@@ -166,8 +166,8 @@ internal static partial class Av1CompoundMaskBlendPredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<ushort>.Count;
+                nuint vectorCount = Numerics.Vector256Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector256<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<ushort>.Count)
                 {
                     Vector256<ushort> firstVector = Vector256.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector256<ushort> secondVector = Vector256.LoadUnsafe(ref secondReference, (nuint)column);
@@ -178,8 +178,8 @@ internal static partial class Av1CompoundMaskBlendPredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<ushort>.Count;
+                nuint vectorCount = Numerics.Vector128Count<ushort>(width - column);
-                for (; column <= vectorEnd; column += Vector128<ushort>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<ushort>.Count)
                 {
                     Vector128<ushort> firstVector = Vector128.LoadUnsafe(ref destinationReference, (nuint)column);
                     Vector128<ushort> secondVector = Vector128.LoadUnsafe(ref secondReference, (nuint)column);

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1DifferenceWeightedMaskBuilder.cs

@@ -68,8 +68,8 @@ internal static partial class Av1DifferenceWeightedMaskBuilder
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<byte> firstVector = Vector512.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector512<byte> secondVector = Vector512.LoadUnsafe(ref secondReference, (nuint)column);
@@ -79,8 +79,8 @@ internal static partial class Av1DifferenceWeightedMaskBuilder
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<byte> firstVector = Vector256.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector256<byte> secondVector = Vector256.LoadUnsafe(ref secondReference, (nuint)column);
@@ -90,8 +90,8 @@ internal static partial class Av1DifferenceWeightedMaskBuilder
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<byte> firstVector = Vector128.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector128<byte> secondVector = Vector128.LoadUnsafe(ref secondReference, (nuint)column);
@@ -165,8 +165,8 @@ internal static partial class Av1DifferenceWeightedMaskBuilder
             // temporary buffers before the following vector blend consumes the complete plane block.
             if (Vector512.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector512<byte>.Count;
+                nuint vectorCount = Numerics.Vector512Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<byte>.Count)
                 {
                     Vector512<ushort> first0 = Vector512.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector512<ushort> first1 = Vector512.LoadUnsafe(ref firstReference, (nuint)(column + Vector512<ushort>.Count));
@@ -179,8 +179,8 @@ internal static partial class Av1DifferenceWeightedMaskBuilder
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector256<byte>.Count;
+                nuint vectorCount = Numerics.Vector256Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<byte>.Count)
                 {
                     Vector256<ushort> first0 = Vector256.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector256<ushort> first1 = Vector256.LoadUnsafe(ref firstReference, (nuint)(column + Vector256<ushort>.Count));
@@ -193,8 +193,8 @@ internal static partial class Av1DifferenceWeightedMaskBuilder
 
             if (Vector128.IsHardwareAccelerated)
             {
-                int vectorEnd = width - Vector128<byte>.Count;
+                nuint vectorCount = Numerics.Vector128Count<byte>(width - column);
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector128<byte>.Count)
                 {
                     Vector128<ushort> first0 = Vector128.LoadUnsafe(ref firstReference, (nuint)column);
                     Vector128<ushort> first1 = Vector128.LoadUnsafe(ref firstReference, (nuint)(column + Vector128<ushort>.Count));

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1ScaledInterPredictor.cs

@@ -411,8 +411,8 @@ internal static partial class Av1ScaledInterPredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                int oneVectorFromEnd = width - Vector512<int>.Count;
+                nuint vectorCount = Numerics.Vector512Count<int>(width - column);
-                for (; column <= oneVectorFromEnd; column += Vector512<int>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector512<int>.Count)
                 {
                     Vector512<int> result = FilterScaledHorizontalVector512<TSource, THorizontal>(
                         ref sourceRow,
@@ -431,8 +431,8 @@ internal static partial class Av1ScaledInterPredictor
 
             if (Vector256.IsHardwareAccelerated)
             {
-                int oneVectorFromEnd = width - Vector256<int>.Count;
+                nuint vectorCount = Numerics.Vector256Count<int>(width - column);
-                for (; column <= oneVectorFromEnd; column += Vector256<int>.Count)
+                for (; vectorCount > 0; vectorCount--, column += Vector256<int>.Count)
                 {
                     Vector256<int> result = FilterScaledHorizontalVector256<TSource, THorizontal>(
                         ref sourceRow,
@@ -451,7 +451,8 @@ internal static partial class Av1ScaledInterPredictor
 
             if (Vector128.IsHardwareAccelerated)
             {
-                for (; column <= width - Vector128<int>.Count; column += Vector128<int>.Count)
+                nuint vectorCount = Numerics.Vector128Count<int>(width - column);
+                for (; vectorCount > 0; vectorCount--, column += Vector128<int>.Count)
                 {
                     Vector128<int> result = FilterScaledHorizontalVector128<TSource, THorizontal>(
                         ref sourceRow,
@@ -505,70 +506,89 @@ internal static partial class Av1ScaledInterPredictor
 
             if (Vector512.IsHardwareAccelerated)
             {
-                Vector512<int> initial = Vector512.Create(verticalBias);
+                nuint vectorCount = Numerics.Vector512Count<int>(width - column) / 2;
-                Vector512<int> offset = Vector512.Create(roundOffset);
+
-                int oneVectorFromEnd = width - (Vector512<int>.Count * 2);
+                if (vectorCount > 0)
-                for (; column <= oneVectorFromEnd; column += Vector512<int>.Count * 2)
                 {
-                    NativeOperator.Convolve(
+                    // The scaled vertical kernel emits two register widths per iteration, so constants are needed only
-                        ref scratchRow,
+                    // when the remaining row contains at least two complete vectors.
-                        scratchStride,
+                    Vector512<int> initial = Vector512.Create(verticalBias);
-                        (nuint)column,
+                    Vector512<int> offset = Vector512.Create(roundOffset);
-                        ref coefficientBase,
+
-                        FilterCoefficientCount,
+                    for (; vectorCount > 0; vectorCount--, column += Vector512<int>.Count * 2)
-                        initial,
+                    {
-                        out Vector512<int> result0,
+                        NativeOperator.Convolve(
-                        out Vector512<int> result1);
+                            ref scratchRow,
-
+                            scratchStride,
-                    result0 = RoundPowerOfTwo(result0, round1) - offset;
+                            (nuint)column,
-                    result1 = RoundPowerOfTwo(result1, round1) - offset;
+                            ref coefficientBase,
-                    TOperator.Store(ref destinationRow, column, result0, result1, bitDepth);
+                            FilterCoefficientCount,
+                            initial,
+                            out Vector512<int> result0,
+                            out Vector512<int> result1);
+
+                        result0 = RoundPowerOfTwo(result0, round1) - offset;
+                        result1 = RoundPowerOfTwo(result1, round1) - offset;
+                        TOperator.Store(ref destinationRow, column, result0, result1, bitDepth);
+                    }
                 }
             }
 
             if (Vector256.IsHardwareAccelerated)
             {
-                Vector256<int> initial = Vector256.Create(verticalBias);
+                nuint vectorCount = Numerics.Vector256Count<int>(width - column) / 2;
-                Vector256<int> offset = Vector256.Create(roundOffset);
+
-                int oneVectorFromEnd = width - (Vector256<int>.Count * 2);
+                if (vectorCount > 0)
-                for (; column <= oneVectorFromEnd; column += Vector256<int>.Count * 2)
                 {
-                    NativeOperator.Convolve(
+                    // A single-vector remainder belongs to the next narrower tier and does not materialize YMM constants.
-                        ref scratchRow,
+                    Vector256<int> initial = Vector256.Create(verticalBias);
-                        scratchStride,
+                    Vector256<int> offset = Vector256.Create(roundOffset);
-                        (nuint)column,
+
-                        ref coefficientBase,
+                    for (; vectorCount > 0; vectorCount--, column += Vector256<int>.Count * 2)
-                        FilterCoefficientCount,
+                    {
-                        initial,
+                        NativeOperator.Convolve(
-                        out Vector256<int> result0,
+                            ref scratchRow,
-                        out Vector256<int> result1);
+                            scratchStride,
-
+                            (nuint)column,
-                    result0 = RoundPowerOfTwo(result0, round1) - offset;
+                            ref coefficientBase,
-                    result1 = RoundPowerOfTwo(result1, round1) - offset;
+                            FilterCoefficientCount,
-                    TOperator.Store(ref destinationRow, column, result0, result1, bitDepth);
+                            initial,
+                            out Vector256<int> result0,
+                            out Vector256<int> result1);
+
+                        result0 = RoundPowerOfTwo(result0, round1) - offset;
+                        result1 = RoundPowerOfTwo(result1, round1) - offset;
+                        TOperator.Store(ref destinationRow, column, result0, result1, bitDepth);
+                    }
                 }
             }
 
             if (Vector128.IsHardwareAccelerated)
             {
-                Vector128<int> initial = Vector128.Create(verticalBias);
+                nuint vectorCount = Numerics.Vector128Count<int>(width - column) / 2;
-                Vector128<int> offset = Vector128.Create(roundOffset);
+
-                int oneVectorFromEnd = width - (Vector128<int>.Count * 2);
+                if (vectorCount > 0)
-                for (; column <= oneVectorFromEnd; column += Vector128<int>.Count * 2)
                 {
-                    NativeOperator.Convolve(
+                    // XMM constants are likewise skipped when fewer than eight output samples remain.
-                        ref scratchRow,
+                    Vector128<int> initial = Vector128.Create(verticalBias);
-                        scratchStride,
+                    Vector128<int> offset = Vector128.Create(roundOffset);
-                        (nuint)column,
+
-                        ref coefficientBase,
+                    for (; vectorCount > 0; vectorCount--, column += Vector128<int>.Count * 2)
-                        FilterCoefficientCount,
+                    {
-                        initial,
+                        NativeOperator.Convolve(
-                        out Vector128<int> result0,
+                            ref scratchRow,
-                        out Vector128<int> result1);
+                            scratchStride,
-
+                            (nuint)column,
-                    result0 = RoundPowerOfTwo(result0, round1) - offset;
+                            ref coefficientBase,
-                    result1 = RoundPowerOfTwo(result1, round1) - offset;
+                            FilterCoefficientCount,
-                    TOperator.Store(ref destinationRow, column, result0, result1, bitDepth);
+                            initial,
+                            out Vector128<int> result0,
+                            out Vector128<int> result1);
+
+                        result0 = RoundPowerOfTwo(result0, round1) - offset;
+                        result1 = RoundPowerOfTwo(result1, round1) - offset;
+                        TOperator.Store(ref destinationRow, column, result0, result1, bitDepth);
+                    }
                 }
             }
 

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1TranslationalInterPredictor.Dispatch.cs

@@ -371,14 +371,14 @@ internal static partial class Av1TranslationalInterPredictor
 
         if (Vector512.IsHardwareAccelerated && Vector<int>.Count == Vector512<int>.Count && width >= Vector512<byte>.Count)
         {
-            int vectorEnd = width - Vector512<byte>.Count;
+            int vectorEnd = (int)(Numerics.Vector512Count<byte>(width) * (nuint)Vector512<byte>.Count);
             for (int row = 0; row < height; row++)
             {
                 ref byte sourceRow = ref Unsafe.Add(ref sourceBase, row * sourceStride);
                 ref byte destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
                 int column = 0;
 
-                for (; column <= vectorEnd; column += Vector512<byte>.Count)
+                for (; column < vectorEnd; column += Vector512<byte>.Count)
                 {
                     Vector512.LoadUnsafe(ref sourceRow, (nuint)column).StoreUnsafe(ref destinationRow, (nuint)column);
                 }
@@ -394,14 +394,14 @@ internal static partial class Av1TranslationalInterPredictor
 
         if (Vector256.IsHardwareAccelerated && width >= Vector256<byte>.Count)
         {
-            int vectorEnd = width - Vector256<byte>.Count;
+            int vectorEnd = (int)(Numerics.Vector256Count<byte>(width) * (nuint)Vector256<byte>.Count);
             for (int row = 0; row < height; row++)
             {
                 ref byte sourceRow = ref Unsafe.Add(ref sourceBase, row * sourceStride);
                 ref byte destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
                 int column = 0;
 
-                for (; column <= vectorEnd; column += Vector256<byte>.Count)
+                for (; column < vectorEnd; column += Vector256<byte>.Count)
                 {
                     Vector256.LoadUnsafe(ref sourceRow, (nuint)column).StoreUnsafe(ref destinationRow, (nuint)column);
                 }
@@ -428,9 +428,9 @@ internal static partial class Av1TranslationalInterPredictor
                     continue;
                 }
 
-                int vectorEnd = width - Vector128<byte>.Count;
+                int vectorEnd = (int)(Numerics.Vector128Count<byte>(width) * (nuint)Vector128<byte>.Count);
                 int column = 0;
-                for (; column <= vectorEnd; column += Vector128<byte>.Count)
+                for (; column < vectorEnd; column += Vector128<byte>.Count)
                 {
                     Vector128.LoadUnsafe(ref sourceRow, (nuint)column).StoreUnsafe(ref destinationRow, (nuint)column);
                 }
@@ -464,14 +464,14 @@ internal static partial class Av1TranslationalInterPredictor
 
         if (Vector512.IsHardwareAccelerated && Vector<int>.Count == Vector512<int>.Count && width >= Vector512<ushort>.Count)
         {
-            int vectorEnd = width - Vector512<ushort>.Count;
+            int vectorEnd = (int)(Numerics.Vector512Count<ushort>(width) * (nuint)Vector512<ushort>.Count);
             for (int row = 0; row < height; row++)
             {
                 ref ushort sourceRow = ref Unsafe.Add(ref sourceBase, row * sourceStride);
                 ref ushort destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
                 int column = 0;
 
-                for (; column <= vectorEnd; column += Vector512<ushort>.Count)
+                for (; column < vectorEnd; column += Vector512<ushort>.Count)
                 {
                     Vector512.LoadUnsafe(ref sourceRow, (nuint)column).StoreUnsafe(ref destinationRow, (nuint)column);
                 }
@@ -487,14 +487,14 @@ internal static partial class Av1TranslationalInterPredictor
 
         if (Vector256.IsHardwareAccelerated && width >= Vector256<ushort>.Count)
         {
-            int vectorEnd = width - Vector256<ushort>.Count;
+            int vectorEnd = (int)(Numerics.Vector256Count<ushort>(width) * (nuint)Vector256<ushort>.Count);
             for (int row = 0; row < height; row++)
             {
                 ref ushort sourceRow = ref Unsafe.Add(ref sourceBase, row * sourceStride);
                 ref ushort destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
                 int column = 0;
 
-                for (; column <= vectorEnd; column += Vector256<ushort>.Count)
+                for (; column < vectorEnd; column += Vector256<ushort>.Count)
                 {
                     Vector256.LoadUnsafe(ref sourceRow, (nuint)column).StoreUnsafe(ref destinationRow, (nuint)column);
                 }
@@ -523,9 +523,9 @@ internal static partial class Av1TranslationalInterPredictor
                     continue;
                 }
 
-                int vectorEnd = width - Vector128<ushort>.Count;
+                int vectorEnd = (int)(Numerics.Vector128Count<ushort>(width) * (nuint)Vector128<ushort>.Count);
                 int column = 0;
-                for (; column <= vectorEnd; column += Vector128<ushort>.Count)
+                for (; column < vectorEnd; column += Vector128<ushort>.Count)
                 {
                     Vector128.LoadUnsafe(ref sourceRow, (nuint)column).StoreUnsafe(ref destinationRow, (nuint)column);
                 }

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1TranslationalInterPredictor.OneDimension.cs

@@ -62,8 +62,8 @@ internal static partial class Av1TranslationalInterPredictor
                 continue;
             }
 
-            int vectorEnd = width - Vector128<byte>.Count;
+            nuint vectorCount = Numerics.Vector128Count<byte>(width - processedColumns);
-            for (; processedColumns <= vectorEnd; processedColumns += Vector128<byte>.Count)
+            for (; vectorCount > 0; vectorCount--, processedColumns += Vector128<byte>.Count)
             {
                 Convolve(
                     ref sourceRow,
@@ -107,7 +107,7 @@ internal static partial class Av1TranslationalInterPredictor
         ref byte sourceBase = ref Unsafe.Add(ref MemoryMarshal.GetReference(source), sourceOrigin);
         ref byte destinationBase = ref MemoryMarshal.GetReference(destination);
         ref short coefficientBase = ref MemoryMarshal.GetReference(coefficients);
-        int vectorEnd = width - Vector256<byte>.Count;
+        int vectorEnd = (int)(Numerics.Vector256Count<byte>(width) * (nuint)Vector256<byte>.Count);
 
         for (int row = 0; row < height; row++)
         {
@@ -115,7 +115,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref byte destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector256<byte>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector256<byte>.Count)
             {
                 Convolve(
                     ref sourceRow,
@@ -159,7 +159,7 @@ internal static partial class Av1TranslationalInterPredictor
         ref byte sourceBase = ref Unsafe.Add(ref MemoryMarshal.GetReference(source), sourceOrigin);
         ref byte destinationBase = ref MemoryMarshal.GetReference(destination);
         ref short coefficientBase = ref MemoryMarshal.GetReference(coefficients);
-        int vectorEnd = width - Vector512<byte>.Count;
+        int vectorEnd = (int)(Numerics.Vector512Count<byte>(width) * (nuint)Vector512<byte>.Count);
 
         for (int row = 0; row < height; row++)
         {
@@ -167,7 +167,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref byte destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector512<byte>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector512<byte>.Count)
             {
                 Convolve(
                     ref sourceRow,
@@ -232,8 +232,8 @@ internal static partial class Av1TranslationalInterPredictor
                 continue;
             }
 
-            int vectorEnd = width - Vector128<ushort>.Count;
+            nuint vectorCount = Numerics.Vector128Count<ushort>(width - processedColumns);
-            for (; processedColumns <= vectorEnd; processedColumns += Vector128<ushort>.Count)
+            for (; vectorCount > 0; vectorCount--, processedColumns += Vector128<ushort>.Count)
             {
                 Convolve(ref sourceRow, tapStride, (nuint)processedColumns, ref coefficientBase, tapCount, initial, out Vector128<int> result0, out Vector128<int> result1);
                 Round(ref result0, ref result1, firstRound, secondRound);
@@ -268,7 +268,7 @@ internal static partial class Av1TranslationalInterPredictor
         ref ushort destinationBase = ref MemoryMarshal.GetReference(destination);
         ref short coefficientBase = ref MemoryMarshal.GetReference(coefficients);
         int maximum = (1 << bitDepth) - 1;
-        int vectorEnd = width - Vector256<ushort>.Count;
+        int vectorEnd = (int)(Numerics.Vector256Count<ushort>(width) * (nuint)Vector256<ushort>.Count);
 
         for (int row = 0; row < height; row++)
         {
@@ -277,7 +277,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref ushort destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector256<ushort>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector256<ushort>.Count)
             {
                 Convolve(ref sourceRow, tapStride, (nuint)processedColumns, ref coefficientBase, tapCount, initial, out Vector256<int> result0, out Vector256<int> result1);
                 Round(ref result0, ref result1, firstRound, secondRound);
@@ -312,7 +312,7 @@ internal static partial class Av1TranslationalInterPredictor
         ref ushort destinationBase = ref MemoryMarshal.GetReference(destination);
         ref short coefficientBase = ref MemoryMarshal.GetReference(coefficients);
         int maximum = (1 << bitDepth) - 1;
-        int vectorEnd = width - Vector512<ushort>.Count;
+        int vectorEnd = (int)(Numerics.Vector512Count<ushort>(width) * (nuint)Vector512<ushort>.Count);
 
         for (int row = 0; row < height; row++)
         {
@@ -321,7 +321,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref ushort destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector512<ushort>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector512<ushort>.Count)
             {
                 Convolve(ref sourceRow, tapStride, (nuint)processedColumns, ref coefficientBase, tapCount, initial, out Vector512<int> result0, out Vector512<int> result1);
                 Round(ref result0, ref result1, firstRound, secondRound);

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1TranslationalInterPredictor.TwoDimensions.Byte.cs

@@ -68,8 +68,8 @@ internal static partial class Av1TranslationalInterPredictor
                 continue;
             }
 
-            int vectorEnd = width - Vector128<byte>.Count;
+            nuint vectorCount = Numerics.Vector128Count<byte>(width - processedColumns);
-            for (; processedColumns <= vectorEnd; processedColumns += Vector128<byte>.Count)
+            for (; vectorCount > 0; vectorCount--, processedColumns += Vector128<byte>.Count)
             {
                 Convolve(
                     ref sourceRow,
@@ -135,8 +135,8 @@ internal static partial class Av1TranslationalInterPredictor
                 continue;
             }
 
-            int vectorEnd = width - Vector128<byte>.Count;
+            nuint vectorCount = Numerics.Vector128Count<byte>(width - processedColumns);
-            for (; processedColumns <= vectorEnd; processedColumns += Vector128<byte>.Count)
+            for (; vectorCount > 0; vectorCount--, processedColumns += Vector128<byte>.Count)
             {
                 Convolve(
                     ref scratchRow,
@@ -199,7 +199,7 @@ internal static partial class Av1TranslationalInterPredictor
         int scratchStride = Math.Max(width, MinimumScratchStride);
         int intermediateHeight = height + verticalTapCount - 1;
         Vector256<int> horizontalInitial = initial + Vector256.Create(1 << (bitDepth + FilterBits - 1));
-        int vectorEnd = width - Vector256<byte>.Count;
+        int vectorEnd = (int)(Numerics.Vector256Count<byte>(width) * (nuint)Vector256<byte>.Count);
 
         for (int row = 0; row < intermediateHeight; row++)
         {
@@ -207,7 +207,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref short scratchRow = ref Unsafe.Add(ref scratchBase, row * scratchStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector256<byte>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector256<byte>.Count)
             {
                 Convolve(
                     ref sourceRow,
@@ -243,7 +243,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref byte destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector256<byte>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector256<byte>.Count)
             {
                 Convolve(
                     ref scratchRow,
@@ -306,7 +306,7 @@ internal static partial class Av1TranslationalInterPredictor
         int scratchStride = Math.Max(width, MinimumScratchStride);
         int intermediateHeight = height + verticalTapCount - 1;
         Vector512<int> horizontalInitial = initial + Vector512.Create(1 << (bitDepth + FilterBits - 1));
-        int vectorEnd = width - Vector512<byte>.Count;
+        int vectorEnd = (int)(Numerics.Vector512Count<byte>(width) * (nuint)Vector512<byte>.Count);
 
         for (int row = 0; row < intermediateHeight; row++)
         {
@@ -314,7 +314,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref short scratchRow = ref Unsafe.Add(ref scratchBase, row * scratchStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector512<byte>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector512<byte>.Count)
             {
                 Convolve(
                     ref sourceRow,
@@ -350,7 +350,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref byte destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector512<byte>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector512<byte>.Count)
             {
                 Convolve(
                     ref scratchRow,

src/ImageSharp/Formats/Heif/Av1/Prediction/Inter/Av1TranslationalInterPredictor.TwoDimensions.UInt16.cs

@@ -74,8 +74,8 @@ internal static partial class Av1TranslationalInterPredictor
                 continue;
             }
 
-            int vectorEnd = width - Vector128<ushort>.Count;
+            nuint vectorCount = Numerics.Vector128Count<ushort>(width - processedColumns);
-            for (; processedColumns <= vectorEnd; processedColumns += Vector128<ushort>.Count)
+            for (; vectorCount > 0; vectorCount--, processedColumns += Vector128<ushort>.Count)
             {
                 Convolve(
                     ref sourceRow,
@@ -138,8 +138,8 @@ internal static partial class Av1TranslationalInterPredictor
                 continue;
             }
 
-            int vectorEnd = width - Vector128<ushort>.Count;
+            nuint vectorCount = Numerics.Vector128Count<ushort>(width - processedColumns);
-            for (; processedColumns <= vectorEnd; processedColumns += Vector128<ushort>.Count)
+            for (; vectorCount > 0; vectorCount--, processedColumns += Vector128<ushort>.Count)
             {
                 Convolve(
                     ref scratchRow,
@@ -199,7 +199,7 @@ internal static partial class Av1TranslationalInterPredictor
         int scratchStride = Math.Max(width, MinimumScratchStride);
         int intermediateHeight = height + verticalTapCount - 1;
         Vector256<int> horizontalInitial = initial + Vector256.Create(1 << (bitDepth + FilterBits - 1));
-        int vectorEnd = width - Vector256<ushort>.Count;
+        int vectorEnd = (int)(Numerics.Vector256Count<ushort>(width) * (nuint)Vector256<ushort>.Count);
 
         for (int row = 0; row < intermediateHeight; row++)
         {
@@ -211,7 +211,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref short scratchRow = ref Unsafe.Add(ref scratchBase, row * scratchStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector256<ushort>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector256<ushort>.Count)
             {
                 Convolve(
                     ref sourceRow,
@@ -253,7 +253,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref ushort destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector256<ushort>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector256<ushort>.Count)
             {
                 Convolve(
                     ref scratchRow,
@@ -313,7 +313,7 @@ internal static partial class Av1TranslationalInterPredictor
         int scratchStride = Math.Max(width, MinimumScratchStride);
         int intermediateHeight = height + verticalTapCount - 1;
         Vector512<int> horizontalInitial = initial + Vector512.Create(1 << (bitDepth + FilterBits - 1));
-        int vectorEnd = width - Vector512<ushort>.Count;
+        int vectorEnd = (int)(Numerics.Vector512Count<ushort>(width) * (nuint)Vector512<ushort>.Count);
 
         for (int row = 0; row < intermediateHeight; row++)
         {
@@ -325,7 +325,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref short scratchRow = ref Unsafe.Add(ref scratchBase, row * scratchStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector512<ushort>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector512<ushort>.Count)
             {
                 Convolve(
                     ref sourceRow,
@@ -367,7 +367,7 @@ internal static partial class Av1TranslationalInterPredictor
             ref ushort destinationRow = ref Unsafe.Add(ref destinationBase, row * destinationStride);
             int processedColumns = 0;
 
-            for (; processedColumns <= vectorEnd; processedColumns += Vector512<ushort>.Count)
+            for (; processedColumns < vectorEnd; processedColumns += Vector512<ushort>.Count)
             {
                 Convolve(
                     ref scratchRow,

src/ImageSharp/Formats/Heif/Av1/Prediction/IntraBlockCopy/Av1IntraBlockCopyBilinearPredictor.cs

@@ -110,7 +110,7 @@ internal static partial class Av1IntraBlockCopyBilinearPredictor
         // cumulative narrower tiers preserve the same contract for future legal widths without over-reading a tail.
         if (Vector512.IsHardwareAccelerated)
         {
-            int vectorizedColumns = width - (width % Vector512<byte>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector512Count<byte>(width) * (nuint)Vector512<byte>.Count);
             if (vectorizedColumns > 0)
             {
                 for (int row = 0; row < height; row++)
@@ -136,7 +136,7 @@ internal static partial class Av1IntraBlockCopyBilinearPredictor
         if (Vector256.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector256<byte>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector256Count<byte>(remainingColumns) * (nuint)Vector256<byte>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)
@@ -161,7 +161,7 @@ internal static partial class Av1IntraBlockCopyBilinearPredictor
         if (Vector128.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector128<byte>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector128Count<byte>(remainingColumns) * (nuint)Vector128<byte>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)
@@ -244,7 +244,7 @@ internal static partial class Av1IntraBlockCopyBilinearPredictor
         // continuation as the byte path.
         if (Vector512.IsHardwareAccelerated)
         {
-            int vectorizedColumns = width - (width % Vector512<short>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector512Count<short>(width) * (nuint)Vector512<short>.Count);
             if (vectorizedColumns > 0)
             {
                 for (int row = 0; row < height; row++)
@@ -270,7 +270,7 @@ internal static partial class Av1IntraBlockCopyBilinearPredictor
         if (Vector256.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector256<short>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector256Count<short>(remainingColumns) * (nuint)Vector256<short>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)
@@ -295,7 +295,7 @@ internal static partial class Av1IntraBlockCopyBilinearPredictor
         if (Vector128.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector128<short>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector128Count<short>(remainingColumns) * (nuint)Vector128<short>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)

src/ImageSharp/Formats/Heif/Av1/Prediction/IntraBlockCopy/Av1IntraBlockCopyHorizontalPredictor.cs

@@ -108,7 +108,7 @@ internal static partial class Av1IntraBlockCopyHorizontalPredictor
         // cumulative narrower tiers preserve the same contract for future legal widths without over-reading a tail.
         if (Vector512.IsHardwareAccelerated)
         {
-            int vectorizedColumns = width - (width % Vector512<byte>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector512Count<byte>(width) * (nuint)Vector512<byte>.Count);
             if (vectorizedColumns > 0)
             {
                 for (int row = 0; row < height; row++)
@@ -132,7 +132,7 @@ internal static partial class Av1IntraBlockCopyHorizontalPredictor
         if (Vector256.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector256<byte>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector256Count<byte>(remainingColumns) * (nuint)Vector256<byte>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)
@@ -155,7 +155,7 @@ internal static partial class Av1IntraBlockCopyHorizontalPredictor
         if (Vector128.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector128<byte>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector128Count<byte>(remainingColumns) * (nuint)Vector128<byte>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)
@@ -232,7 +232,7 @@ internal static partial class Av1IntraBlockCopyHorizontalPredictor
         // continuation as the byte path.
         if (Vector512.IsHardwareAccelerated)
         {
-            int vectorizedColumns = width - (width % Vector512<short>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector512Count<short>(width) * (nuint)Vector512<short>.Count);
             if (vectorizedColumns > 0)
             {
                 for (int row = 0; row < height; row++)
@@ -256,7 +256,7 @@ internal static partial class Av1IntraBlockCopyHorizontalPredictor
         if (Vector256.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector256<short>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector256Count<short>(remainingColumns) * (nuint)Vector256<short>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)
@@ -279,7 +279,7 @@ internal static partial class Av1IntraBlockCopyHorizontalPredictor
         if (Vector128.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector128<short>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector128Count<short>(remainingColumns) * (nuint)Vector128<short>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)

src/ImageSharp/Formats/Heif/Av1/Prediction/IntraBlockCopy/Av1IntraBlockCopyVerticalPredictor.cs

@@ -108,7 +108,7 @@ internal static partial class Av1IntraBlockCopyVerticalPredictor
         // cumulative narrower tiers preserve the same contract for future legal widths without over-reading a tail.
         if (Vector512.IsHardwareAccelerated)
         {
-            int vectorizedColumns = width - (width % Vector512<byte>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector512Count<byte>(width) * (nuint)Vector512<byte>.Count);
             if (vectorizedColumns > 0)
             {
                 for (int row = 0; row < height; row++)
@@ -132,7 +132,7 @@ internal static partial class Av1IntraBlockCopyVerticalPredictor
         if (Vector256.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector256<byte>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector256Count<byte>(remainingColumns) * (nuint)Vector256<byte>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)
@@ -155,7 +155,7 @@ internal static partial class Av1IntraBlockCopyVerticalPredictor
         if (Vector128.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector128<byte>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector128Count<byte>(remainingColumns) * (nuint)Vector128<byte>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)
@@ -232,7 +232,7 @@ internal static partial class Av1IntraBlockCopyVerticalPredictor
         // continuation as the byte path.
         if (Vector512.IsHardwareAccelerated)
         {
-            int vectorizedColumns = width - (width % Vector512<short>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector512Count<short>(width) * (nuint)Vector512<short>.Count);
             if (vectorizedColumns > 0)
             {
                 for (int row = 0; row < height; row++)
@@ -256,7 +256,7 @@ internal static partial class Av1IntraBlockCopyVerticalPredictor
         if (Vector256.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector256<short>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector256Count<short>(remainingColumns) * (nuint)Vector256<short>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)
@@ -279,7 +279,7 @@ internal static partial class Av1IntraBlockCopyVerticalPredictor
         if (Vector128.IsHardwareAccelerated)
         {
             int remainingColumns = width - processedColumns;
-            int vectorizedColumns = remainingColumns - (remainingColumns % Vector128<short>.Count);
+            int vectorizedColumns = (int)(Numerics.Vector128Count<short>(remainingColumns) * (nuint)Vector128<short>.Count);
             int endColumn = processedColumns + vectorizedColumns;
 
             for (int row = 0; row < height; row++)

src/ImageSharp/Formats/Heif/Av1/Transform/Av1ForwardTransformer.cs

@@ -1046,7 +1046,8 @@ internal static partial class Av1ForwardTransformer
 
             if (Avx512BW.IsSupported)
             {
-                for (; column <= width - Vector512<short>.Count; column += Vector512<short>.Count)
+                nuint vector512Count = Numerics.Vector512Count<short>(width - column);
+                for (; vector512Count > 0; vector512Count--, column += Vector512<short>.Count)
                 {
                     (Vector512<int> lower, Vector512<int> upper) = Vector512.Widen(Vector512.LoadUnsafe(ref sourceRow, (nuint)column));
 
@@ -1057,7 +1058,8 @@ internal static partial class Av1ForwardTransformer
 
             if (Avx2.IsSupported)
             {
-                for (; column <= width - Vector256<short>.Count; column += Vector256<short>.Count)
+                nuint vector256Count = Numerics.Vector256Count<short>(width - column);
+                for (; vector256Count > 0; vector256Count--, column += Vector256<short>.Count)
                 {
                     (Vector256<int> lower, Vector256<int> upper) = Vector256.Widen(Vector256.LoadUnsafe(ref sourceRow, (nuint)column));
 
@@ -1066,7 +1068,8 @@ internal static partial class Av1ForwardTransformer
                 }
             }
 
-            for (; column <= width - Vector128<short>.Count; column += Vector128<short>.Count)
+            nuint vector128Count = Numerics.Vector128Count<short>(width - column);
+            for (; vector128Count > 0; vector128Count--, column += Vector128<short>.Count)
             {
                 (Vector128<int> lower, Vector128<int> upper) = Vector128.Widen(Vector128.LoadUnsafe(ref sourceRow, (nuint)column));
 

src/ImageSharp/Formats/Heif/HeifEncoder.cs

@@ -31,8 +31,8 @@ public sealed class HeifEncoder : AnimatedImageEncoder
 
     /// <summary>
     /// Gets the lossy compression quality, or <see langword="null"/> to use the compression method's default quality.
-    /// Valid values range from 0 for the lowest quality to 100 for the highest quality. Legacy JPEG image items
+    /// Valid values range from 0 for the lowest quality to 100 for the highest quality. A value of 100 does not
-    /// support values from 1 through 100. A value of 100 does not enable <see cref="Lossless"/> encoding.
+    /// enable <see cref="Lossless"/> encoding.
     /// </summary>
     /// <exception cref="ArgumentException">The quality is outside the range 0 to 100.</exception>
     public int? Quality

src/ImageSharp/Formats/Heif/HeifEncoderCore.cs

@@ -64,9 +64,6 @@ internal sealed class HeifEncoderCore
         this.WriteMetadataBox(items, links, stream);
         this.WriteMediaDataBox(compressedPixels, stream);
         stream.Flush();
-
-        HeifMetadata meta = image.Metadata.GetHeifMetadata();
-        meta.CompressionMethod = this.encoder.CompressionMethod;
     }
 
     /// <summary>
@@ -448,13 +445,6 @@ internal sealed class HeifEncoderCore
             throw new NotSupportedException("Legacy JPEG image items support only 8-bit component encoding.");
         }
 
-        if (this.encoder.Quality == 0)
-        {
-            // Zero is meaningful to the AV1 quality scale, but ImageSharp's JPEG encoder deliberately
-            // exposes the JPEG quality scale as 1 through 100. Reject the codec-specific mismatch at this boundary.
-            throw new NotSupportedException("Legacy JPEG image items support quality values in the range [1..100].");
-        }
-
         JpegColorType colorType = this.encoder.ChromaSubsampling switch
         {
             null or HeifChromaSubsampling.Yuv420 => JpegColorType.YCbCrRatio420,
@@ -467,7 +457,9 @@ internal sealed class HeifEncoderCore
         ChunkedMemoryStream stream = new(this.configuration.MemoryAllocator);
         JpegEncoder encoder = new()
         {
-            Quality = this.encoder.Quality,
+            // The HEIF quality scale includes zero while the JPEG payload encoder starts at one.
+            // Map the lowest HEIF setting to the lowest representable JPEG setting.
+            Quality = this.encoder.Quality == 0 ? 1 : this.encoder.Quality,
             ColorType = colorType
         };
 

tests/ImageSharp.Tests/Formats/Heif/Av1/Av1BitStreamTests.cs

@@ -175,6 +175,7 @@ public class Av1BitStreamTests
 
     [InlineData(4, 0, 1, 2, 3)]
     [InlineData(5, 0, 1, 2, 3)]
+    [InlineData(5, 1, 2, 3, 4)]
     [InlineData(8, 0, 1, 2, 3)]
     [InlineData(8, 4, 5, 6, 7)]
     [InlineData(16, 15, 0, 5, 8)]

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

@@ -0,0 +1,98 @@
+// Copyright (c) Six Labors.
+// Licensed under the Six Labors Split License.
+
+using SixLabors.ImageSharp.Formats.Heif.Av1;
+using SixLabors.ImageSharp.Formats.Heif.Av1.Pipeline;
+using SixLabors.ImageSharp.Memory;
+
+namespace SixLabors.ImageSharp.Tests.Formats.Heif.Av1;
+
+public class Av1EncoderFrameTests
+{
+    [Fact]
+    public void ExtendBordersReplicatesEveryPhysicalPlaneEdge()
+    {
+        const int visibleWidth = 5;
+        const int visibleHeight = 3;
+        const int codedWidth = 8;
+        const int codedHeight = 8;
+        const int lumaBorder = Av1EncoderFrame<byte>.LumaBorder;
+        const int chromaBorder = lumaBorder / 2;
+        MemoryAllocator allocator = Configuration.Default.MemoryAllocator;
+
+        Size lumaBufferSize = Av1EncoderFrame<byte>.GetPlaneBufferSize(visibleWidth, visibleHeight, 0, 0);
+        Size chromaBufferSize = Av1EncoderFrame<byte>.GetPlaneBufferSize(visibleWidth, visibleHeight, 1, 1);
+        using Buffer2D<byte> luma = allocator.Allocate2D<byte>(lumaBufferSize.Width, lumaBufferSize.Height);
+        using Buffer2D<byte> chromaBlue = allocator.Allocate2D<byte>(chromaBufferSize.Width, chromaBufferSize.Height);
+        using Buffer2D<byte> chromaRed = allocator.Allocate2D<byte>(chromaBufferSize.Width, chromaBufferSize.Height);
+        Buffer2DRegion<byte> lumaRegion = luma.GetRegion(lumaBorder, lumaBorder, codedWidth, codedHeight);
+        Buffer2DRegion<byte> chromaBlueRegion = chromaBlue.GetRegion(chromaBorder, chromaBorder, codedWidth / 2, codedHeight / 2);
+        Buffer2DRegion<byte> chromaRedRegion = chromaRed.GetRegion(chromaBorder, chromaBorder, codedWidth / 2, codedHeight / 2);
+
+        FillVisible(luma, lumaBorder, lumaBorder, visibleWidth, visibleHeight, 10);
+        FillVisible(chromaBlue, chromaBorder, chromaBorder, (visibleWidth + 1) / 2, (visibleHeight + 1) / 2, 80);
+        FillVisible(chromaRed, chromaBorder, chromaBorder, (visibleWidth + 1) / 2, (visibleHeight + 1) / 2, 120);
+
+        Av1EncoderFrame<byte> frame = new(
+            lumaRegion,
+            chromaBlueRegion,
+            chromaRedRegion,
+            visibleWidth,
+            visibleHeight,
+            8,
+            Av1ColorFormat.Yuv420,
+            1,
+            1);
+
+        frame.ExtendBorders();
+
+        AssertReplicatedPlane(luma, lumaBorder, lumaBorder, visibleWidth, visibleHeight, 10);
+        AssertReplicatedPlane(chromaBlue, chromaBorder, chromaBorder, (visibleWidth + 1) / 2, (visibleHeight + 1) / 2, 80);
+        AssertReplicatedPlane(chromaRed, chromaBorder, chromaBorder, (visibleWidth + 1) / 2, (visibleHeight + 1) / 2, 120);
+    }
+
+    [Theory]
+    [InlineData(5, 3, 0, 0, 160, 136)]
+    [InlineData(5, 3, 1, 0, 80, 136)]
+    [InlineData(5, 3, 1, 1, 80, 68)]
+    [InlineData(1921, 1081, 0, 0, 2080, 1216)]
+    [InlineData(1921, 1081, 1, 1, 1040, 608)]
+    public void GetPlaneBufferSizeMatchesLibaomLayout(
+        int width,
+        int height,
+        int subsamplingX,
+        int subsamplingY,
+        int expectedWidth,
+        int expectedHeight)
+    {
+        Size actual = Av1EncoderFrame<byte>.GetPlaneBufferSize(width, height, subsamplingX, subsamplingY);
+
+        Assert.Equal(new Size(expectedWidth, expectedHeight), actual);
+    }
+
+    private static void FillVisible(Buffer2D<byte> plane, int originX, int originY, int width, int height, int seed)
+    {
+        for (int y = 0; y < height; y++)
+        {
+            Span<byte> row = plane.DangerousGetRowSpan(originY + y);
+            for (int x = 0; x < width; x++)
+            {
+                row[originX + x] = (byte)(seed + (y * width) + x);
+            }
+        }
+    }
+
+    private static void AssertReplicatedPlane(Buffer2D<byte> plane, int originX, int originY, int width, int height, int seed)
+    {
+        for (int y = 0; y < plane.Height; y++)
+        {
+            ReadOnlySpan<byte> row = plane.DangerousGetRowSpan(y);
+            int sourceY = Math.Clamp(y - originY, 0, height - 1);
+            for (int x = 0; x < row.Length; x++)
+            {
+                int sourceX = Math.Clamp(x - originX, 0, width - 1);
+                Assert.Equal((byte)(seed + (sourceY * width) + sourceX), row[x]);
+            }
+        }
+    }
+}

tests/ImageSharp.Tests/Formats/Heif/Av1/Av1ForwardQuantizerTests.cs

@@ -0,0 +1,212 @@
+// Copyright (c) Six Labors.
+// Licensed under the Six Labors Split License.
+
+using SixLabors.ImageSharp.Formats.Heif.Av1;
+using SixLabors.ImageSharp.Formats.Heif.Av1.Pipeline.Quantizers;
+using SixLabors.ImageSharp.Formats.Heif.Av1.Transform;
+using SixLabors.ImageSharp.Tests.TestUtilities;
+
+namespace SixLabors.ImageSharp.Tests.Formats.Heif.Av1;
+
+/// <summary>
+/// Verifies AV1 forward quantization against current libaom's fast no-matrix arithmetic.
+/// </summary>
+[Trait("Format", "Avif")]
+public class Av1ForwardQuantizerTests
+{
+    /// <summary>
+    /// The hardware configurations covering every quantizer vector tier and the scalar fallback.
+    /// </summary>
+    private const HwIntrinsics QuantizerConfigurations =
+        HwIntrinsics.AllowAll | HwIntrinsics.DisableAVX512F | HwIntrinsics.DisableAVX | HwIntrinsics.DisableHWIntrinsic;
+
+    /// <summary>
+    /// Verifies raster quantization and scan-order EOB selection at every SIMD tier.
+    /// </summary>
+    [Fact]
+    public void FastQuantizerMatchesLibaomReferenceAcrossHardwareWidths()
+        => FeatureTestRunner.RunWithHwIntrinsicsFeature(ValidateQuantizer, QuantizerConfigurations);
+
+    /// <summary>
+    /// Verifies that repeated transform quantization uses only caller-owned buffers.
+    /// </summary>
+    [Fact]
+    public void QuantizerDoesNotAllocatePerTransform()
+    {
+        const int coefficientCount = 64;
+        int[] coefficients = new int[coefficientCount];
+        int[] quantized = new int[coefficientCount];
+        int[] dequantized = new int[coefficientCount];
+        FillCoefficients(coefficients, 73);
+
+        Av1ForwardQuantizer.QuantizeLossy(
+            coefficients,
+            quantized,
+            dequantized,
+            Av1TransformSize.Size8x8,
+            Av1TransformType.DctDct,
+            73,
+            -1,
+            3,
+            Av1BitDepth.TenBit);
+
+        long before = GC.GetAllocatedBytesForCurrentThread();
+        for (int iteration = 0; iteration < 32; iteration++)
+        {
+            Av1ForwardQuantizer.QuantizeLossy(
+                coefficients,
+                quantized,
+                dequantized,
+                Av1TransformSize.Size8x8,
+                Av1TransformType.DctDct,
+                73,
+                -1,
+                3,
+                Av1BitDepth.TenBit);
+        }
+
+        Assert.Equal(0, GC.GetAllocatedBytesForCurrentThread() - before);
+    }
+
+    /// <summary>
+    /// Exercises each transform-scale category, coded 64-point layout, quantizer range, and sample precision.
+    /// </summary>
+    private static void ValidateQuantizer()
+    {
+        ReadOnlySpan<Av1TransformSize> transformSizes =
+        [
+            Av1TransformSize.Size4x4,
+            Av1TransformSize.Size8x8,
+            Av1TransformSize.Size16x16,
+            Av1TransformSize.Size32x32,
+            Av1TransformSize.Size64x16,
+            Av1TransformSize.Size64x64,
+        ];
+
+        ReadOnlySpan<int> quantizerIndices = [1, 73, 173, 255];
+        ReadOnlySpan<Av1BitDepth> bitDepths = [Av1BitDepth.EightBit, Av1BitDepth.TenBit, Av1BitDepth.TwelveBit];
+
+        foreach (Av1TransformSize transformSize in transformSizes)
+        {
+            int coefficientCount = transformSize.GetAdjusted().GetSize2d();
+            int[] coefficients = new int[coefficientCount];
+            int[] expectedQuantized = new int[coefficientCount];
+            int[] expectedDequantized = new int[coefficientCount];
+            int[] actualQuantized = new int[coefficientCount];
+            int[] actualDequantized = new int[coefficientCount];
+
+            foreach (int qIndex in quantizerIndices)
+            {
+                FillCoefficients(coefficients, qIndex);
+
+                foreach (Av1BitDepth bitDepth in bitDepths)
+                {
+                    ushort expectedEndOfBlock = QuantizeReference(
+                        coefficients,
+                        expectedQuantized,
+                        expectedDequantized,
+                        transformSize,
+                        Av1TransformType.DctDct,
+                        qIndex,
+                        -1,
+                        3,
+                        bitDepth);
+
+                    ushort actualEndOfBlock = Av1ForwardQuantizer.QuantizeLossy(
+                        coefficients,
+                        actualQuantized,
+                        actualDequantized,
+                        transformSize,
+                        Av1TransformType.DctDct,
+                        qIndex,
+                        -1,
+                        3,
+                        bitDepth);
+
+                    Assert.Equal(expectedEndOfBlock, actualEndOfBlock);
+                    Assert.Equal(expectedQuantized, actualQuantized);
+                    Assert.Equal(expectedDequantized, actualDequantized);
+                }
+            }
+        }
+    }
+
+    /// <summary>
+    /// Fills one transform with deterministic signed values spanning threshold, rounding, and clamp behavior.
+    /// </summary>
+    private static void FillCoefficients(Span<int> coefficients, int seed)
+    {
+        for (int i = 0; i < coefficients.Length; i++)
+        {
+            coefficients[i] = (((i * 7919) + (seed * 313)) % 90001) - 45000;
+        }
+
+        coefficients[0] = 0;
+        coefficients[1] = 1;
+        coefficients[2] = -1;
+        coefficients[3] = short.MaxValue;
+        coefficients[4] = -short.MaxValue;
+    }
+
+    /// <summary>
+    /// Mirrors av1_quantize_fp_no_qmatrix from current libaom without sharing the production traversal.
+    /// </summary>
+    private static ushort QuantizeReference(
+        ReadOnlySpan<int> coefficients,
+        Span<int> quantizedCoefficients,
+        Span<int> dequantizedCoefficients,
+        Av1TransformSize transformSize,
+        Av1TransformType transformType,
+        int qIndex,
+        int dcDeltaQ,
+        int acDeltaQ,
+        Av1BitDepth bitDepth)
+    {
+        quantizedCoefficients.Clear();
+        dequantizedCoefficients.Clear();
+
+        int logScale = transformSize.GetScale();
+        int dcDequantizer = Av1QuantizationLookup.GetDcQuant(qIndex, dcDeltaQ, bitDepth);
+        int acDequantizer = Av1QuantizationLookup.GetAcQuant(qIndex, acDeltaQ, bitDepth);
+        int dcQuantizer = (1 << 16) / dcDequantizer;
+        int acQuantizer = (1 << 16) / acDequantizer;
+        int dcRounding = RoundPowerOfTwo((64 * dcDequantizer) >> 7, logScale);
+        int acRounding = RoundPowerOfTwo((64 * acDequantizer) >> 7, logScale);
+        ReadOnlySpan<short> scan = Av1ScanOrderConstants.GetScanOrder(transformSize, transformType).Scan;
+        ushort endOfBlock = 0;
+
+        for (int scanIndex = 0; scanIndex < scan.Length; scanIndex++)
+        {
+            int coefficientIndex = scan[scanIndex];
+            int coefficient = coefficients[coefficientIndex];
+            int coefficientSign = coefficient >> 31;
+            long magnitude = ((long)coefficient ^ coefficientSign) - coefficientSign;
+            int dequantizer = coefficientIndex == 0 ? dcDequantizer : acDequantizer;
+            int quantizer = coefficientIndex == 0 ? dcQuantizer : acQuantizer;
+            int rounding = coefficientIndex == 0 ? dcRounding : acRounding;
+            int quantizedMagnitude = 0;
+
+            if ((magnitude << (1 + logScale)) >= dequantizer)
+            {
+                magnitude = Math.Clamp(magnitude + rounding, short.MinValue, short.MaxValue);
+                quantizedMagnitude = (int)((magnitude * quantizer) >> (16 - logScale));
+            }
+
+            if (quantizedMagnitude != 0)
+            {
+                quantizedCoefficients[coefficientIndex] = (quantizedMagnitude ^ coefficientSign) - coefficientSign;
+                int dequantizedMagnitude = (quantizedMagnitude * dequantizer) >> logScale;
+                dequantizedCoefficients[coefficientIndex] = (dequantizedMagnitude ^ coefficientSign) - coefficientSign;
+                endOfBlock = (ushort)(scanIndex + 1);
+            }
+        }
+
+        return endOfBlock;
+    }
+
+    /// <summary>
+    /// Applies libaom's positive round-power-of-two operation.
+    /// </summary>
+    private static int RoundPowerOfTwo(int value, int shift)
+        => shift == 0 ? value : (value + (1 << (shift - 1))) >> shift;
+}

tests/ImageSharp.Tests/Formats/Heif/Av1/Av1TileDecoderStub.cs

@@ -17,6 +17,6 @@ internal class Av1TileDecoderStub : IAv1TileReader, IAv1TileWriter
     {
     }
 
-    public Span<byte> WriteTile(int tileNum)
+    public ReadOnlySpan<byte> GetTileData(int tileNum)
         => this.tileDatas[tileNum];
 }

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

@@ -162,7 +162,7 @@ public class ObuFrameHeaderTests
         // Assign 2
         Span<byte> encodedBuffer = encoded.ToArray();
         IAv1TileReader tileDecoder2 = new Av1TileDecoderStub();
-        Av1BitStreamReader reader2 = new(span);
+        Av1BitStreamReader reader2 = new(encodedBuffer);
         ObuReader obuReader2 = new();
 
         // Act 2
@@ -704,6 +704,59 @@ public class ObuFrameHeaderTests
         Assert.Equal(bitStream.Length * 8, reader.BitPosition);
     }
 
+    /// <summary>
+    /// Verifies non-uniform tile boundaries use the next stored boundary and retain the clipped final mode-info edge.
+    /// </summary>
+    [Fact]
+    public void WriteNonUniformTileBoundariesRoundTrip()
+    {
+        ObuSequenceHeader sequenceHeader = GetDefaultSequenceHeader();
+        ObuFrameHeader frameHeader = GetKeyFrameHeader();
+        ObuTileGroupHeader tileInfo = frameHeader.TilesInfo;
+        tileInfo.HasUniformTileSpacing = false;
+        tileInfo.TileColumnCount = 2;
+        tileInfo.TileRowCount = 1;
+        tileInfo.TileSizeBytes = 1;
+        tileInfo.TileColumnStartModeInfo[0] = 0;
+        tileInfo.TileColumnStartModeInfo[1] = 64;
+        tileInfo.TileColumnStartModeInfo[2] = frameHeader.ModeInfoColumnCount;
+        tileInfo.TileRowStartModeInfo[0] = 0;
+        tileInfo.TileRowStartModeInfo[1] = frameHeader.ModeInfoRowCount;
+
+        Av1TileDecoderStub sourceTiles = new();
+        sourceTiles.ReadTile([0x80], 0);
+        sourceTiles.ReadTile([0x80], 1);
+
+        using MemoryStream stream = new();
+        ObuWriter writer = new();
+        writer.WriteAll(Configuration.Default, stream, sequenceHeader, frameHeader, sourceTiles);
+        byte[] bitStream = stream.ToArray();
+        Assert.Equal([0x00, 0x80, 0x80], bitStream[^3..]);
+
+        Av1BitStreamReader reader = new(bitStream);
+        ObuReader obuReader = new();
+        Av1TileDecoderStub decodedTiles = new();
+
+        obuReader.ReadAll(ref reader, bitStream.Length, () => decodedTiles);
+
+        ObuTileGroupHeader actual = obuReader.FrameHeader.TilesInfo;
+        Assert.False(actual.HasUniformTileSpacing);
+        Assert.Equal(2, actual.TileColumnCount);
+        Assert.Equal(1, actual.TileRowCount);
+        Assert.Equal(1, actual.TileSizeBytes);
+        Assert.Equal(0, actual.TileColumnStartModeInfo[0]);
+        Assert.Equal(64, actual.TileColumnStartModeInfo[1]);
+        Assert.Equal(frameHeader.ModeInfoColumnCount, actual.TileColumnStartModeInfo[2]);
+        Assert.Equal(0, actual.TileRowStartModeInfo[0]);
+        Assert.Equal(frameHeader.ModeInfoRowCount, actual.TileRowStartModeInfo[1]);
+
+        ReadOnlySpan<byte> expectedTileData = [0x80];
+
+        Assert.True(decodedTiles.GetTileData(0).SequenceEqual(expectedTileData));
+        Assert.True(decodedTiles.GetTileData(1).SequenceEqual(expectedTileData));
+        Assert.Equal(bitStream.Length * 8, reader.BitPosition);
+    }
+
     /// <summary>
     /// Encodes one non-reduced sequence header with a single selected conformance failure.
     /// </summary>

tests/ImageSharp.Tests/Formats/Heif/HeifEncoderTests.cs

@@ -64,13 +64,19 @@ public class HeifEncoderTests
     }
 
     [Fact]
-    public void LegacyJpegRejectsZeroQuality()
+    public void LegacyJpegAcceptsZeroQuality()
     {
         using Image<Rgba32> image = new(1, 1);
+        image[0, 0] = new Rgba32(10, 20, 30);
         using MemoryStream stream = new();
         HeifEncoder encoder = new() { Quality = 0 };
 
-        Assert.Throws<NotSupportedException>(() => image.Save(stream, encoder));
+        image.Save(stream, encoder);
+
+        Assert.NotEqual(0, stream.Length);
+        stream.Position = 0;
+        using Image<Rgba32> decoded = Image.Load<Rgba32>(stream);
+        Assert.Equal(image.Size, decoded.Size);
     }
 
     [Fact]
@@ -81,6 +87,7 @@ public class HeifEncoderTests
         HeifEncoder encoder = new() { Lossless = true };
 
         Assert.Throws<NotSupportedException>(() => image.Save(stream, encoder));
+        Assert.Equal(0, stream.Length);
     }
 
     [Theory]
@@ -93,6 +100,34 @@ public class HeifEncoderTests
         HeifEncoder encoder = new() { BitDepth = bitDepth };
 
         Assert.Throws<NotSupportedException>(() => image.Save(stream, encoder));
+        Assert.Equal(0, stream.Length);
+    }
+
+    [Fact]
+    public void Av1RejectsEncodingBeforeWritingOutput()
+    {
+        using Image<Rgba32> image = new(1, 1);
+        using MemoryStream stream = new();
+        HeifEncoder encoder = new() { CompressionMethod = HeifCompressionMethod.Av1 };
+
+        Assert.Throws<NotSupportedException>(() => image.Save(stream, encoder));
+        Assert.Equal(0, stream.Length);
+    }
+
+    [Fact]
+    public void LegacyJpegEncodingDoesNotMutateSourceHeifMetadata()
+    {
+        using Image<Rgba32> image = new(1, 1);
+        image[0, 0] = new Rgba32(10, 20, 30, 255);
+        HeifMetadata metadata = image.Metadata.GetHeifMetadata();
+        metadata.CompressionMethod = HeifCompressionMethod.Av1;
+        using MemoryStream stream = new();
+        HeifEncoder encoder = new();
+
+        image.Save(stream, encoder);
+
+        Assert.Same(metadata, image.Metadata.GetHeifMetadata());
+        Assert.Equal(HeifCompressionMethod.Av1, metadata.CompressionMethod);
     }
 
     [Theory]