libvpx-1.16.0

[Bruce Dubbs]

New minor version.

[Joe Locash]

2026-01-06 v1.16.0 "Xenonetta Duck"
  This release includes Arm SVE2 and Neon optimizations for 12-tap filters,
  AVX512 implementations for SAD, support for per-frame and per-spatial-layer
  PSNR calculation, and numerous bug fixes.

  - Upgrading:
    This release is ABI incompatible with the previous release.

    Unit tests require C++17 to build.

    Support for 32-bit iOS targets (armv7, armv7s, and i386) has been removed.

  - Enhancement:
    Optimized Arm SVE2 and Neon implementations for 12-tap convolution filters.
    Optimized Neon High Bitdepth (HBD) SAD and sad_avg functions.
    Added Arm Neon DotProd and I8MM implementations for vpx_convolve12.
    Added AVX512 implementations for SAD64 and sad_skip functions.
    Added SSSE3 and AVX2 implementations for 12-tap temporal filter prediction.
    Added support for per-frame and per-spatial-layer PSNR calculation.

    Adjusted temporal filter strength to improve visual quality and reduce block
    artifacts.

    Added support for darwin24 (macOS 15) and darwin25 (macOS 26).
    libwebm is upgraded to commit b4f01ea.

  - Bug fixes:
    Fix to heap buffer overflow in vp9_deblock, vp9_post_proc_frame, and
    vp9_pack_bitstream.

    Fix to integer overflow in vp9_highbd_post_proc, vp9_rc_regulate_q,
    tiny_ssim, and vp9_calc_pframe_target_size_one_pass_cbr.

    Fix to use-of-uninitialized-value in vp9_highbd_post_proc, mfqe, and
    vp8_datarate_test.

    Fix to out-of-bounds in log_tile_cols_from_picsize_level.
    Fix to double free on initialization failure in vpx_codec_enc_init_multi.
    Fix to division-by-zero crash in vpxenc with 0 FPS numerator input.

    Fix to various build failures for Arm/SVE2, macOS cross-compilation, and
    Xcode 16.

[Joe Locash]

Fixed at a91fbf220d.

[Bruce Dubbs]

Milestone renamed