Skip to content

CHANGELOG.md

Latest commit

 

History

History
246 lines (189 loc) · 10.7 KB

CHANGELOG.md

File metadata and controls

246 lines (189 loc) · 10.7 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

QUEUED BREAKING CHANGES

[0.9.2] - 2026-04-17

Added

  • RGBX8 descriptor in zencodec dispatch — RGBX8_SRGB added to ENCODE_DESCRIPTORS alongside RGBA8_SRGB, routing through encode_rgba_with_stop since mozjpeg reads 4 bytes per pixel and ignores byte 3 regardless of whether it's alpha or padding (ed47868)
    • Handled in both the one-shot encode path and the push_rows / finish streaming path
    • Tests confirm RGBX produces byte-identical output to RGBA and to RGB when colour bytes match

Changed

  • Routed all README badges through shields.io with ?style=flat-square for consistent sizing (9f10ab2)
  • Bump zencodec to 0.1.19

Fixed

  • fast_color doctest in src/encode.rs now passes Preset::default() to Encoder::new() to match the current API signature (9f10ab2)

0.9.1 - 2026-04-14

Added

  • RGBA8 encoding — Native 4-channel input support without intermediate RGB buffer (7a55b85, 6f0ffcc)
    • Encoder::encode_rgba(), encode_rgba_with_stop(), encode_rgba_to_writer()
    • color::convert_rgba_to_ycbcr_c_compat() — direct RGBA→YCbCr, alpha ignored
    • For BGRA input, swizzle to RGBA first with garb::bytes::bgra_to_rgba_inplace
  • zencodec RGBA8 support — RGBA8_SRGB added to supported encode descriptors (d5d60ab, 06c5e4b)
    • encode() routes to native RGBA path based on pixel descriptor
    • encode_srgba8() uses native RGBA path for contiguous data, garb::rgba_to_rgb_strided for strided
    • push_rows() / finish() handle RGB8, RGBA8, and GRAY8
  • garb dependency (optional, via zencodec feature) for SIMD-optimized pixel format conversions

Changed

  • Swapped yuv crate for zenyuv in the fast-yuv path (2667a98) — internal Imazen crate, same 15-bit fixed-point BT.601 output, enables future Sharp YUV opt-in
  • Refactored encode_rgb_to_writer to share downsample+MCU+encode pipeline via encode_ycbcr_planes_to_writer helper (6f0ffcc)

[0.8.0] - 2026-02-08

Added

  • ARM NEON SIMD support — Safe SIMD implementation for aarch64 using archmage
    • Automatic runtime detection and dispatch
    • Byte-identical output to scalar implementation
    • Zero unsafe code
  • 100% safe Rust — Achieved #![forbid(unsafe_code)] with no exceptions
    • Uses archmage 0.5 for safe SIMD intrinsics
    • Uses safe_unaligned_simd 0.2.4 for safe load/store operations
  • CI coverage for ARM platforms — Tests on native arm64 runners plus QEMU for aarch64, armv7, and i686
  • WASM support — Builds successfully for wasm32-wasip1 with SIMD128

Changed

  • Updated to archmage 0.5 (from 0.4)
  • Updated to safe_unaligned_simd 0.2.4 (from 0.2.3)
  • Trellis encoding is now 6% faster than C mozjpeg (was 7%, updated measurements)
  • CI now tests 15 platform configurations including ARM and WASM

Fixed

  • Conditional compilation for forbid(unsafe_code) when mozjpeg-sys-config feature is enabled
  • wasm32 unused import warnings

[0.7.0] - 2026-02-02

Added

  • fast_color(bool) - Ergonomic API for color conversion mode
    • fast_color(true) — ~40% faster RGB→YCbCr using yuv crate (±1 rounding vs C)
    • fast_color(false) — exact C mozjpeg parity, bytewise identical output (default)
  • Byte-exact C mozjpeg parity — Baseline and Progressive modes now produce identical output to C mozjpeg (0.00% file size delta)
  • AVX2-accelerated C-compatible color conversion — 3.6 Gpix/s, exact match to C mozjpeg's jccolor.c

Changed

  • Default color conversion now produces byte-exact output matching C mozjpeg
  • kodak_dir() now returns the actual Kodak corpus directory (was aliasing cid22_dir())
  • With trellis enabled, Rust produces smaller files than C mozjpeg (-0.05% to -0.47%)
  • Trellis encoding is now 7% faster than C mozjpeg

Deprecated

  • c_compat_color(bool) — use fast_color(bool) instead (clearer semantics)

0.6.0 - 2026-02-02

Added

  • imgref feature (default) - Type-safe encoding with encode_imgref() accepting ImgRef<P> directly
    • Supports RGB<u8>, RGBA<u8> (alpha discarded), Gray<u8>, [u8; 3], [u8; 4], u8
    • Automatic stride handling for subimages
    • No dimension mix-ups (width/height baked into type)
  • Strided encoding - encode_rgb_strided() and encode_gray_strided() for:
    • Memory-aligned buffers (e.g., 64-byte aligned rows)
    • Cropping without copy (point into larger buffer)
    • GPU texture formats with padding
  • EncodeablePixel trait - Implement for custom pixel types
  • Error::InvalidStride - New error variant for stride validation

Changed

  • imgref and rgb crates are now default dependencies (can be disabled with default-features = false)

0.5.5 - 2026-02-01

Changed

  • Recommend mozjpeg crate for users who need both encoding and decoding

0.5.4 - 2026-01-13

Fixed

  • Clippy warnings in benchmarks and examples

Changed

  • Removed all unsafe from SIMD modules (now uses safe wrappers)

0.5.3 - 2026-01-10

Changed

  • Migrated from archmage 0.1 to 0.4 with safe_unaligned_simd

0.5.1 - 2026-01-08

Fixed

  • Scan optimizer parity improvements with C mozjpeg

0.5.0 - 2026-01-05

Added

  • Improved scan optimization algorithm
  • Better progressive scan selection

Fixed

  • optimize_scans parity with C mozjpeg (now within ±0.4%)

0.4.1 - 2025-01-03

Added

  • encode_ycbcr_planar_strided() - Encode strided YCbCr data without extra copies. Accepts separate stride for each plane.

0.4.0 - 2025-01-03

Added

  • encode_ycbcr_planar() - Encode pre-converted planar YCbCr data directly, bypassing RGB-to-YCbCr conversion. Supports all subsampling modes (4:4:4, 4:2:2, 4:2:0).
  • CMozjpeg::encode_ycbcr_planar() - Same capability for C mozjpeg wrapper
  • Decoder round-trip validation - CI now validates encoded JPEGs against multiple decoders

Changed

  • configure_sys() renamed to to_c_mozjpeg() - Clearer naming for C mozjpeg interop
  • sys-local is now an optional dev dependency gated by feature flag

0.3.0 - 2025-01-02

Breaking Changes

  • Encoder::new() now requires a Preset argument - Explicit preset selection replaces the previous boolean progressive parameter
  • Crate renamed from mozjpeg-oxide to mozjpeg-rs

Added

  • Preset enum with four encoding modes:
    • BaselineFastest - No optimizations, maximum speed
    • BaselineBalanced - Baseline with trellis + Huffman optimization
    • ProgressiveBalanced - Progressive with optimizations (default)
    • ProgressiveSmallest - Maximum compression with optimize_scans
  • mozjpeg-sys-config feature - Configure a C mozjpeg encoder from Rust Encoder settings via Encoder::configure_sys()
  • Smoothing filter - .smoothing(0-100) for noise reduction on dithered images
  • Robidoux quant table alias - QuantTableIdx::Robidoux (alias for MssimTuned)
  • CLI-style method names - .quant_method(), .dct_method() aliases
  • Grayscale progressive JPEG support
  • Trellis speed optimization - Adaptive search limiting for high-entropy blocks
  • EOB cross-block optimization - Opt-in via TrellisConfig::eob_optimization(true)

Fixed

  • AC refinement ZRL encoding - Fixed decoder errors on 8/24 test corpus images with successive approximation
  • Progressive encoding now works correctly for all image sizes including non-MCU-aligned dimensions

Changed

  • Public API cleanup: internal modules now hidden with #[doc(hidden)]
  • SIMD: multiversion is now the default (safe autovectorization)
  • SIMD: Hand-written AVX2 intrinsics available via simd-intrinsics feature
  • Encoder settings matrix added to README

0.2.5 - 2024-12-30

Fixed

  • Example test_refine.rs used wrong crate name

0.2.4 - 2024-12-29

Added

  • Grayscale progressive JPEG support
  • Trellis speed optimization (speed_level)

0.2.3 - 2024-12-28

Fixed

  • AC refinement ZRL loop must run for both temp>1 and temp==1 cases
  • Bytewise parity analysis documentation

0.2.2 - 2024-12-27

Fixed

  • Match C mozjpeg progressive encoding defaults exactly
  • Per-scan Huffman tables for all progressive modes

0.1.0 - 2024-12-27

Added

  • Initial release of mozjpeg-rs
  • Pure Rust JPEG encoder based on Mozilla's mozjpeg
  • Trellis quantization - Rate-distortion optimized AC and DC coefficient selection
  • Progressive JPEG encoding - Multi-scan with DC-first, AC-band progression
  • Huffman optimization - 2-pass encoding for optimal entropy coding
  • Overshoot deringing - Reduces ringing artifacts near hard edges
  • Chroma subsampling - 4:4:4, 4:2:2, 4:2:0, 4:4:0 modes
  • Quality presets - Encoder::new(), Encoder::max_compression(), Encoder::fastest()
  • Optimize scans - Tries multiple scan configurations for progressive mode
  • SIMD acceleration via multiversion crate (AVX2, SSE4.1, NEON)
  • Runtime CPU feature detection and dispatch
  • Comprehensive test suite (177+ tests)
  • FFI validation against C mozjpeg
  • Multi-platform CI (Linux, macOS, Windows on x64 and ARM64)

Performance

  • Trellis mode: Rust is ~10% faster than C mozjpeg
  • With matching settings, produces files 2% smaller than C mozjpeg
  • AVX2 intrinsics provide additional ~15% DCT speedup (opt-in feature)

Compatibility

  • MSRV: Rust 1.89.0
  • Platforms: Linux, macOS, Windows (x64 and ARM64)
  • Output compatible with all standard JPEG decoders