libaom-3.14.0

[Douglas R. Reno]

New minor version

[Joe Locash]

2026-05-12 v3.14.0
  This release is ABI compatible with the last release.

  This release includes significant compression efficiency and
  perceptual quality improvements for layered image encoding, in
  addition to bug fixes.

  The `build` directory in the source tree was removed and the original
  build/cmake directory was moved up one level. If you use the CMake
  toolchain files in the original build/cmake/toolchains directory,
  they are now in the cmake/toolchains directory.

  The unsupported aom_img_fmt_t enum constants AOM_IMG_FMT_AOMYV12 and
  AOM_IMG_FMT_AOMI420 are deprecated and will be removed in a future
  release. Do not use them.

  The new codec controls AV1E_SET_EXTERNAL_RATE_CONTROL (including the
  new aom_rc_funcs_t struct and the new "aom/aom_ext_ratectrl.h" and
  "aom/aom_tpl.h" headers) and AV1E_GET_GOP_INFO (including the new
  aom_gop_info_t struct) have the experimental API status and are NOT
  part of the stable API. Applications that need the backward
  compatibility of the stable API must not use
  AV1E_SET_EXTERNAL_RATE_CONTROL and AV1E_GET_GOP_INFO.

  - New Features
    * Tuning modes AOM_TUNE_IQ and AOM_TUNE_SSIMULACRA2 now work with
      inter-frame encoding modes (good-quality and realtime), enabling
      efficient layered image encoding.
    * Variance Boost (deltaq-mode 6) support has been extended to
      include good-quality and realtime modes.
    * New value 2 for use_fixed_qp_offsets: disallow the application of
      per-frame QP offsets, enabling full control of each frame's QP, by
      adjusting rc_cfg.cq_level between each encoded frame.
    * Low complexity decode mode (the encoder optimizations that produce
      encoded bitstreams that can be decoded at low computational
      complexity) now supports good-quality encoding (speed 1 to 3) for
      not only vertical videos (608p to 1080p) but also horizontal
      videos (720p to 1080p).
    * Realtime encoding with non-zero lookahead is supported, for
      !REALTME_ONLY_CONFIG.
    * New AOM_EFLAG_FREEZE_INTERNAL_STATE flag for aom_codec_encode() to
      encode a frame without updating encoder state such as reference
      buffers, CDF tables, and rate control state. Useful for
      speculative encoding.
    * New aom_matrix_coefficients_t enum constants:
      - AOM_CICP_MC_IPT_C2 (15): IPT-C2
      - AOM_CICP_MC_YCGCO_RE (16): YCgCo-Re
      - AOM_CICP_MC_YCGCO_RO (17): YCgCo-Ro
    * New aom_metadata_insert_flags_t enum constants for adding
      layer-specific metadata OBUs:
      - AOM_MIF_NON_KEY_FRAME_LAYER_SPECIFIC: Adds layer-specific
        metadata if it's not a keyframe
      - AOM_MIF_KEY_FRAME_LAYER_SPECIFIC: Adds layer-specific metadata
        only if it's a keyframe
      - AOM_MIF_ANY_FRAME_LAYER_SPECIFIC: Adds layer-specific metadata
        to any type of frame
    * Experimental: New codec control AV1E_SET_EXTERNAL_RATE_CONTROL to
      enable an external rate control library. Not part of the stable
      API.
    * Experimental: New codec control AV1E_GET_GOP_INFO to get the GOP
      structure from the encoder. Not part of the stable API.
    * New codec control AOME_SET_VALIDATE_HBD_INPUT to validate high
      bitdepth (HBD) input and ensure that every pixel is within the
      valid range. The corresponding option for aom_codec_set_option()
      is "validate-hbd-input".
    * New codec control AOMD_SET_FRAME_SIZE_LIMIT to set the maximum
      frame size for a decoder

  - Compression Efficiency Improvements
    * When encoding 2-layered images with AOM_TUNE_IQ vs. AOM_TUNE_SSIM:
      - Good-quality mode: up to 15% gains at similar SSIMULACRA 2
        scores.
      - Realtime mode: up to 30% gains at similar SSIMULACRA 2 scores.
    * Enable screen detection mode 2 (anti-aliased text and graphics
      aware) in all-intra mode.
    * Some minor quantization matrix formula and delta chroma q tweaks
      for the AOM_TUNE_SSIMULACRA2 tuning mode, with up to 0.2%
      SSIMULACRA 2 score efficiency gains.
    * Re-tune and re-work encoder features, achieving coding gains at
      each speed level and better coding efficiency vs complexity
      tradeoff (an estimated 20-30% encoder time reduction with about
      1-5% vmaf gains).
    * RTC screen: improvements to quality and scroll detection for high
      resolutions.

  - Perceptual Quality Improvements
    * AOM_TUNE_IQ improves the visual quality of layered image encoding
      compared to AOM_TUNE_SSIM.
    * Adaptive sharpness: tweak loop filter sharpness threshold to
      improve visual quality at QPs 29 and 30.
    * Improve encoder algorithms in loop restoration, CDEF, temporal
      filter and rate control, delivering better visual quality.

  - Speedups
    * Make Adaptive CDEF more decoder friendly by limiting CDEF
      application to only regions where high filter strengths are
      needed.
    * Encoder complexity is reduced significantly, especially at speed
      1 to 3.
    * Further AArch64 SIMD optimization of filtering algorithms,
      particularly using Armv8.6 Neon I8MM.
    * Additional tweaks to AArch64 Neon paths for quantization, SAD,
      subpel variance and intra-predictors.

  - Bug Fixes
    * Commit 4cc0867f: fix multithreading crashes when loop restoration
      filtering is used by the encoder.
    * Commit 74aaa1ef: fix AOM_SCALING_MODE crashes when compound mask
      prediction modes are used by the encoder.
    * b:502030569: Avoid assertion failure in `read_uncompressed_header`
      triggered by fuzzed input
    * b:503197490: Fix an integer overflow in target bits calculation
    * b:503691210: Avoid reference to uninitialized member variables for
      multi workers.
    * b:504317456: Handle buffer pointer in LAP mode to avoid overflow
    * b:502133197: Avoid a bitread assertion failure in
      get_av1config_from_obu()

Fixed at bd748206c1.

[zeckma]

This is a security update as disclosed in #23342. I misread the timeline and how this was in the book already, and that ticket was for libaom 3.14.0 -> 3.14.1. Both this and that ticket are security updates however, and that ticket goes over all the security information. So all disclosure can happen over there. Marking as high, still.