libvpx Venetian Duck v1.14.0-rc1

326 views
Skip to first unread message

Jerome Jiang

unread,
Jan 9, 2024, 2:16:35 PMJan 9
to
Hello,

Venetian Duck v1.14.0-rc1 has been tagged[1]. Please help testing and report issues[2].
This release is ABI incompatible with the previous release.

Thanks.

Changelog:

2024-01-02 v1.14.0 "Venetian Duck"
  This release drops support for old C compilers, such as Visual Studio 2012
  and older, that disallow mixing variable declarations and statements (a C99
  feature). It adds support for run-time CPU feature detection for Arm
  platforms, as well as support for darwin23 (macOS 14).
  - Upgrading:
    This release is ABI incompatible with the previous release.

    Various new features for rate control library for real-time: SVC parallel
    encoding, loopfilter level, support for frame dropping, and screen content.

    New callback function send_tpl_gop_stats for vp9 external rate control
    library, which can be used to transmit TPL stats for a group of pictures. A
    public header vpx_tpl.h is added for the definition of TPL stats used in
    this callback.

    libwebm is upgraded to libwebm-1.0.0.29-9-g1930e3c.

  - Enhancement:
    Improvements on Neon optimizations: VoD: 12-35% speed up for bitdepth 8,
    68%-151% speed up for high bitdepth.

    Improvements on AVX2 and SSE optimizations.
    Improvements on LSX optimizations for LoongArch.
    42-49% speedup on speed 0 VoD encoding.
    Android API level predicates.

  - Bug fixes:
    Fix to missing prototypes from the rtcd header.
    Fix to segfault when total size is enlarged but width is smaller.
    Fix to the build for arm64ec using MSVC.
    Fix to copy BLOCK_8X8's mi to PICK_MODE_CONTEXT::mic.
    Fix to -Wshadow warnings.
    Fix to heap overflow in vpx_get4x4sse_cs_neon.
    Fix to buffer overrun in highbd Neon subpel variance filters.
    Added bitexact encode test script.
    Fix to -Wl,-z,defs with Clang's sanitizers.
    Fix to decoder stability after error & continued decoding.
    Fix to mismatch of VP9 encode with NEON intrinsics with C only version.
    Fix to Arm64 MSVC compile vpx_highbd_fdct4x4_neon.
    Fix to fragments count before use.
    Fix to a case where target bandwidth is 0 for SVC.
    Fix mask in vp9_quantize_avx2,highbd_get_max_lane_eob.
    Fix to int overflow in vp9_calc_pframe_target_size_one_pass_cbr.
    Fix to integer overflow in vp8,ratectrl.c.
    Fix to interger overflow in vp9 svc.
    Fix to avg_frame_bandwidth overflow.
    Fix to per frame qp for temporal layers.
    Fix to unsigned integer overflow in sse computation.
    Fix to uninitialized mesh feature for BEST mode.
    Fix to overflow in highbd temporal_filter.
    Fix to unaligned loads w/w==4 in vpx_convolve_copy_neon.
    Skip arm64_neon.h workaround w/VS >= 2019.
    Fix to c vs avx mismatch of diamond_search_sad().
    Fix to c vs intrinsic mismatch of vpx_hadamard_32x32() function.
    Fix to a bug in vpx_hadamard_32x32_neon().
    Fix to Clang -Wunreachable-code-aggressive warnings.
    Fix to a bug in vpx_highbd_hadamard_32x32_neon().
    Fix to -Wunreachable-code in mfqe_partition.
    Force mode search on 64x64 if no mode is selected.
    Fix to ubsan failure caused by left shift of negative.
    Fix to integer overflow in calc_pframe_target_size.
    Fix to float-cast-overflow in vp8_change_config().
    Fix to a null ptr before use.
    Conditionally skip using inter frames in speed features.
    Remove invalid reference frames.
    Disable intra mode search speed features conditionally.
    Set nonrd keyframe under dynamic change of deadline for rtc.
    Fix to scaled reference offsets.
    Set skip_recode=0 in nonrd_pick_sb_modes.
    Fix to an edge case when downsizing to one.
    Fix to a bug in frame scaling.
    Fix to pred buffer stride.
    Fix to a bug in simple motion search.
    Update frame size in actual encoding.

Reply all
Reply to author
Forward
0 new messages