Common/Helpers
- Add InterleaveLower and InterleaveUpper to Vector128_ and Vector256_
- Add unit test for InterleaveLower and InterleaveUpper (specifically for Vector256_)
- Add Average to Numerics.cs
Common
- Add 32 and 33 to the InlineArray.tt text template
Formats/Jxl/IO/Metadata
- Remove unnecessary System.Runtime.CompilerServices using directive from JxlCustomTransformData and JxlOpsinInvreseMatrix
Formats/Jxl/Processing/Decoder
- Remove unncessary using SixLabors.ImageSharp.Formats.Jxl.IO
Formats/Jxl/Processing/Encoder
- Add partial Fast Lossless Encoder work (+enc_fast_lossless.cc; largest file in libjxl source)
- Add linear algebra (+enc_linalg.cc, +enc_linalg.h)
Formats/Jxl/Processing/Jpeg
- Work that would later become JXL<->JPEG lossless coding mode
Formats/Jxl/Processing/Modular/Encoding/ContextPrediction
- Finish context prediction (+context_predict.h)
Formats/Jxl/Processing/Modular/Transforms
- Finish Reversible Color Transform (+rct.cc, +rct.h, +enc_rct.cc, +enc_rct.h)
- Finish Palette/Indexed coding (+palette.cc, +palette.h, +enc_palette.cc, enc_palette.h)
- Finish Squeeze transform (+squeeze.cc, +squeeze.h, +enc_squeeze.cc, +enc_squeeze.h)
Formats/Jxl/Processing/RenderPipeline
- Incomplete render pipeline abstractions with EPF (Edge Preserving Filter) 0 stage (+render_pipeline_stage.cc, +render_pipeline_stage.h, +stage_epf.cc, +stage_epf.h)
Formats/Jxl/Processing/Splines
- Remove unnecessary System.Runtime.CompilerServices using directive
Formats/Jxl/Processing
- Add dequantizer matrices
- Remove JxlEndianness (prefer ByteOrder from ImageSharp/Common)
- Add missing constant to JxlLoopFilter
- Remove unnecessary using SixLabors.ImageSharp.Common.Helpers from JxlMath
- Replace JxlPixelFormat to use ByteOrder
- Update quantizers to use dequantizer matrices and quantizer weights
- Add quantizer encoding and constants
- Add SIMD utilities
- Remove System.Runtime.CompilerServices using from JxlWeightsSeparable5
- Remove InlineArray3, InlineArray36 and InlineArray15 from InlineArrays (3 and 15 already exist in System.Runtime.CompilerServices; 36 already exists in InlineArray.tt from ImageSharp/Common)
NEXT STEPS
The current focus would be applying refactors and optimizations from reviews, followed by completing the JPEG XL modular.
Remove the `Numerics.Normalize` wrapper and call `TensorPrimitives.Divide` directly when normalizing resize kernels. This also drops the now-unused normalization test coverage and the redundant benchmark-only `System.Numerics.Tensors` package reference for net10.0.
Replaced ImageSharp’s custom TensorPrimitives_ and several Vector128/256/512 helper fallbacks with direct System.Numerics and System.Numerics.Tensors APIs across color conversion, codecs, filtering, blending, and convolution paths. Removed the obsolete compatibility helper files, updated package references (including System.Numerics.Tensors and newer System.IO.Hashing), and aligned tests with runtime behavior changes such as inverted float clamp bounds now throwing ArgumentException.
Replace the custom zlib deflate stack with `System.IO.Compression.ZLibStream`, and add `ChunkedWriteStream` to keep PNG/TIFF/EXR output framed in fixed-size segments. Update the affected compressors and tests to match the new streaming path.
Move the main target framework from net8.0 to net10.0 and the preview lane from net10.0 to net11.0, adding a global.json floor for SDK 10+ and updating CI coverage and test projects to match.
The change also adjusts hardware-intrinsic feature tests for .NET 11's x86-64-v2 baseline, updates Magick.NET, adds PDF files to LFS, and converts many internal helper and codec APIs from Span<T> to ReadOnlySpan<T> where inputs are not mutated.
Use tile width (instead of full image width) when creating T4/T6/Modified Huffman decompressors for tiled TIFF decoding, preventing row-size mismatches and false overrun failures. Also standardize the CCITT overrun error message and add regression coverage for both strip-based and tiled reporter samples that must throw when decoded pixels exceed image width.
Adds explicit bounds checks before writing decoded CCITT runs in both T4 and Modified Huffman TIFF decompressors, preventing oversized runs from writing past the current row. Error messages were also standardized/capitalized for CCITT parsing failures. Includes new regression tests using compressed TIFF PoCs to verify decoding now throws ImageFormatException with the expected width-overrun message.
The allocation limits on MemoryAllocator were internal with no public
way to raise the 1 GB single-buffer default, and ApplyOptions was not
callable from external subclasses, which blocked custom allocators.
- Expose MemoryGroupAllocationLimitBytes, SingleBufferAllocationLimitBytes,
and AccumulativeAllocationLimitBytes with validated protected setters.
- Make ApplyOptions protected so derived allocators can accept options.
- Add MemoryAllocatorOptions.SingleBufferAllocationLimitMegabytes to raise
the contiguous buffer cap up to 2047 MB without a custom allocator.
- Add public-api tests that implement a complete external allocator.
- Add behavior tests for the new option and its validation.
The binary PGM and PPM formats store 16-bit samples most significant
byte first. The decoder and encoder read and wrote them in native
little-endian order, so every wide sample was byte swapped. This
corrects the swizzled colors reported for 16-bit PPM images.
- Reverse the sample byte order in BinaryDecoder and BinaryEncoder.
- Fix integer division in the max pixel value upscale factor.
- Add exact-value, reference-decoder, upscale, and round-trip tests.
- Add a 16-bit PPM test image and its Magick-anchored reference PNG.
- Document the binary decoder and encoder in full.
Introduce the `Plus` alpha composition mode and add support for ColorDodge, ColorBurn, SoftLight, Difference, Exclusion, Hue, Saturation, Color, and Luminosity across straight-alpha and associated-alpha pixel blenders. The generated blender/function tables and selector logic were updated to expose the new combinations, and tests were expanded to cover mode mapping, expected blend results, and parity between straight and premultiplied alpha paths.
Drops the `AntialiasThreshold` field/property from `GraphicsOptions` and updates related XML docs to reflect the simplified antialiasing behavior. Clone logic and test coverage were adjusted accordingly, including removal of threshold assertions and comparer checks.
Route BlendWithCoverage calls through AssociatedAlphaPorterDuffFunctions when TOperator.IsAssociatedAlpha is true, across all SIMD widths (scalar, Vector256, Vector512) and all BlendWithCoverageFunction overloads.
Tightened icon frame bounds validation to treat `ImageOffset == available` as invalid, preventing non-empty resources from starting at the exclusive end of the icon stream. Added a regression test that mutates an ICO directory entry to this boundary value and verifies both `Decode` and `Identify` throw `InvalidImageContentException` with the expected message.
Fixes EXIF serialization when `ExifLong8Array` downgrades to `ExifDataType.Long` but still stores values as `ulong`. `ExifWriter.WriteValue` now handles `ulong` inputs in the `Long` branch by writing 32-bit values, and a BigTIFF metadata test verifies mixed in-range values are written correctly as two little-endian `uint`s.
Update TIFF metadata creation to build ImageMetadata after IFD tag processing and copy root-frame EXIF/ICC/IPTC/XMP profiles to image scope via deep clones. This keeps image-level metadata in sync for downstream single-frame encoders while preserving independent mutability between frame and image metadata. Tests now cover preserved ICC handling and verify image-level profile presence, equality, and non-shared instances for EXIF/XMP/IPTC, including ignore-metadata behavior.
Handle CgBI chunks consistently when they appear after IHDR by routing both identify/decode paths through a new `ReadCgbiChunk()` method that immediately validates already-read headers. This prevents invalid CgBI images from bypassing compatibility checks when chunk order is unusual. Added PNG tests that reorder the first two chunks to confirm incompatible CgBI headers now throw in both identify and decode flows. Also renames `ZlibInflateStream` to `ZlibInflateReader` and updates EXR/PNG/TIFF call sites for clearer intent.
Use exact zero determinant checks in `TransformUtilities.IsDegenerate` for `Matrix3x2` and `Matrix4x4` instead of epsilon-based near-zero detection, preventing valid transforms from being treated as degenerate during extreme downscales. Added a regression test for issue #3156 to verify resizing large square images (1024/2048) to 1x1 succeeds and preserves the expected pixel.
Slice the destination span to source length in `Vector4Converters.RgbaCompatible.FromVector4` so SIMD/scalar tails never process spare destination capacity. Expanded pixel-format tests to run across hardware intrinsic configurations, added dedicated RGB24/BGR24 bulk conversion coverage with sentinel checks for out-of-range writes, and relaxed associated-alpha half packing assertions to treat NaN payload differences as valid IEEE behavior.
Introduces `RgbaHalf` and `RgbaHalfP` plus bulk pixel-operation support, updates AOT seeding, and routes EXR decoding through the new scaled conversions. Also fixes scaled/native range handling for half, SNORM, and signed integer pixel formats, and expands tests for the new layouts and edge cases.
Introduce explicit associated/unassociated pixel conversion APIs across `IPixel` and `PixelOperations`, and rework associated-alpha operations to use representation-aware bulk hooks. Update pixel implementations, converters, blenders, and generated code to consistently apply `Premultiply`/`UnPremultiply` semantics, plus SIMD-backed conversion paths for half and signed formats.
Fix codec and processing behavior where alpha handling was incorrect or lossy (notably TIFF/EXR decode/encode, convolution/resize/filter/normalization/dithering paths), and add targeted tests for these cases. Expand AOT compiler roots (including EXR and additional pixel/processor coverage), add a new public API test project for pixel-operations extensibility, and add broad alpha-representation and quantization regression coverage.
Refactors `Color` to track both exposed and stored alpha representations, adds `ToScaledVector4(PixelAlphaRepresentation)`, and adjusts pixel conversion paths so associated formats preserve canonical values without unnecessary unpremultiply/reassociate loss. This also updates equality/hash behavior to compare canonical scaled values.
SIMD helpers are split into `MultiplyAddEstimate` vs `FusedMultiplyAdd`, with byte-to-float normalization updated to match scalar rounding exactly across vector widths. JPEG converters, resize kernels, and Porter-Duff/associated-alpha blending paths were updated to use the appropriate helper for either fast estimate or strict fused rounding semantics. Tests were expanded substantially (including exhaustive component/alpha cases and fused-order checks), incidental test cleanup was applied, benchmark comment snapshots were refreshed, and one black/white reference output image was updated.
Introduce Rgba32P, Bgra32P, Argb32P, Abgr32P, NormalizedByte4P, and HalfVector4P.
Add representation-aware scalar and bulk conversions, optimized pixel operations, and associated-alpha blending across all Porter-Duff modes. Include comprehensive conversion, layout, blending, and performance coverage.
KernelSamplingMap.CorrectBorder sliced the offset span by
radius * kernelSize, so bounds narrower than the kernel radius threw
ArgumentOutOfRangeException (GaussianBlur of a 130x8 region with
sigma 10, for example), and bounds exactly equal to the radius let
Bounce offsets escape the sampling range by folding an overshoot of a
full extent to min - 1.
Route bounds no larger than the radius through an exact per-mode
correction that folds each offset into range: clamp for Repeat,
periodic reflection for Mirror and Bounce, and true modulo for Wrap.
The single-pass corrections are unchanged for larger bounds.
winscripter
Stefan Nikolei
James Jackson-South
MoonsvnLyn
Erik White