[x265] [ANN] New Release v4.3

[Kirithika Kalirathnam]

Hello everyone,
We are pleased to announce the release of x265 v4.3. This version includes new features, optimizations and critical bug fixes.
Repository link:https://github.com/Multicorewareinc/x265.
Please find the release notes below.
Version 4.3
===========

Release date - 31st July 2026

New feature
-----------
1. LoongArch64 architecture support
2. Foveated encoding - spatially varies quality around a gaze point, spending fewer bits in the viewer's periphery.
3. Selective (noise-based) MCSTF filtering avoiding unnecessary filtering cost/quality impact on low-noise content.

Enhancements to existing features
----------------------------------
1. MCSTF : Reworked Multi-threaded motion estimation and bilateral filtering, HME decoupled from the b-adapt flow, lowres planes generated with HM-equivalent averaging, TestBench harness support and profiling stats.
2. Visual Studio 2026 (vc18) build scripts.
3. Migrated continuous integration from Bitbucket Pipelines to GitHub Actions, with code-quality checks and smoke tests on pull requests.

API changes
-----------
1. Foveated encoding command-line options: :option:`--fovea-gaze`, :option:`--fovea-delta`, :option:`--fovea-sigma`, and :option:`--fovea-gaze-file`.
2. MCSTF command-line options: :option:`--selective-mcstf` and :option:`--mcstf-ref-range`.
3. Removed the MCSTF param-only options: temporalFilterStrength and searchRangeForLayers. Callers referencing the removed fields must be updated.

Optimizations
-------------
1. LoongArch64 SIMD optimizations across the loopfilter, ipfilter, DCT, motion compensation, block copy, pixel processing and intra-prediction modules.
2. AArch64 SIMD optimizations: improved Neon and SVE DCT16/DCT32, stride-aware DCT, new SVE2 implementations of psyCost, sa8d and satd, and faster Neon sa8d, satd and psyCost (including 12-bit sa8d).
3. AVX2 implementations for the MCSTF kernels resulted in speed up of 2x when combined with multithreading.
4. Threaded Motion Estimation: Early exit diamond search

Bug fixes
---------
1. Resolve ASAN/UBSAN reported out-of-bounds accesses and undefined behavior.
2. Fix TestBench segfault at high bit depth caused by alias trampolines.
3. Resolve TSAN data races in weight-analysis chroma extension, rate-control row stats, threaded-ME, and shared encoder state.
4. Fix MCSTF segfault and fix threaded-me crashes and multi-pool subscription issue
5. Fix scenecut not being active when :option:`--bframes` is 0.
6. Fix incorrect thread-pool size and regression in :option:`--pools` behaviour
7. Fix bitstream error when VUI timing info is disabled.
8. Fix i686 build failure (non-PC-relative PLT references) and PowerPC64 build with Clang.
9. Fix AArch64 filter-prim-sve test failures.
10. CPU feature detection fixes for OpenBSD/FreeBSD (elf_aux_info) and PowerPC.
11. Silence Clang -Wnontrivial-memcall and -Wundef warnings; assorted build-warning cleanups.

Known issues
------------
1. Performance regression observed with the threaded-me feature on high-frequency systems and higher resolutions (4k).
2. Inconsistent hash mismatch with abr-ladder feature

Thanks,
Kirithika
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20260731/41da4f0c/attachment.htm>