2022-08-31 v3.5.0
This release is ABI compatible with the last one, including speedup and memory
optimizations, and new APIs and features.
- New Features
* Support for frame parallel encode for larger number of threads. --fp-mt
flag is available for all build configurations.
* New codec control AV1E_GET_NUM_OPERATING_POINTS
- Speedup and Memory Optimizations
* Speed-up multithreaded encoding for good quality mode for larger number of
threads through frame parallel encoding:
o 30-34% encode time reduction for 1080p, 16 threads, 1x1 tile
configuration (tile_rows x tile_columns)
o 18-28% encode time reduction for 1080p, 16 threads, 2x4 tile
configuration
o 18-20% encode time reduction for 2160p, 32 threads, 2x4 tile
configuration
* 16-20% speed-up for speed=6 to 8 in still-picture encoding mode
* 5-6% heap memory reduction for speed=6 to 10 in real-time encoding mode
* Improvements to the speed for speed=7, 8 in real-time encoding mode
* Improvements to the speed for speed=9, 10 in real-time screen encoding
mode
* Optimizations to improve multi-thread efficiency in real-time encoding
mode
* 10-15% speed up for SVC with temporal layers
* SIMD optimizations:
o Improve av1_quantize_fp_32x32_neon() 1.05x to 1.24x faster
o Add aom_highbd_quantize_b{,_32x32,_64x64}_adaptive_neon() 3.15x to 5.6x
faster than "C"
o Improve av1_quantize_fp_64x64_neon() 1.17x to 1.66x faster
o Add aom_quantize_b_avx2() 1.4x to 1.7x faster than aom_quantize_b_avx()
o Add aom_quantize_b_32x32_avx2() 1.4x to 2.3x faster than
aom_quantize_b_32x32_avx()
o Add aom_quantize_b_64x64_avx2() 2.0x to 2.4x faster than
aom_quantize_b_64x64_ssse3()
o Add aom_highbd_quantize_b_32x32_avx2() 9.0x to 10.5x faster than
aom_highbd_quantize_b_32x32_c()
o Add aom_highbd_quantize_b_64x64_avx2() 7.3x to 9.7x faster than
aom_highbd_quantize_b_64x64_c()
o Improve aom_highbd_quantize_b_avx2() 1.07x to 1.20x faster
o Improve av1_quantize_fp_avx2() 1.13x to 1.49x faster
o Improve av1_quantize_fp_32x32_avx2() 1.07x to 1.54x faster
o Improve av1_quantize_fp_64x64_avx2() 1.03x to 1.25x faster
o Improve av1_quantize_lp_avx2() 1.07x to 1.16x faster
- Bug fixes including but not limited to
* aomedia:3206 Assert that skip_width > 0 for deconvolve function
* aomedia:3278 row_mt enc: Delay top-right sync when intraBC is enabled
* aomedia:3282 blend_a64_*_neon: fix bus error in armv7
* aomedia:3283 FRAME_PARALLEL: Propagate border size to all cpis
* aomedia:3283 RESIZE_MODE: Fix incorrect strides being used for motion
search
* aomedia:3286 rtc-svc: Fix to dynamic_enable spatial layers
* aomedia:3289 rtc-screen: Fix to skipping inter-mode test in nonrd
* aomedia:3289 rtc-screen: Fix for skip newmv on flat blocks
* aomedia:3299 Fix build failure with CONFIG_TUNE_VMAF=1
* aomedia:3296 Fix the conflict --enable-tx-size-search=0 with nonrd mode
--enable-tx-size-search will be ignored in non-rd pick mode
* aomedia:3304 Fix off-by-one error of max w/h in validate_config
* aomedia:3306 Do not use pthread_setname_np on GNU/Hurd
* aomedia:3325 row-multithreading produces invalid bitstream in some cases
* chromium:1346938, chromium:1338114
* compiler_flags.cmake: fix flag detection w/cmake 3.17-3.18.2
* tools/*.py: update to python3
* aom_configure.cmake: detect PIE and set CONFIG_PIC
* test/simd_cmp_impl: use explicit types w/CompareSimd*
* rtc: Fix to disable segm for aq-mode=3
* rtc: Fix to color_sensitivity in variance partition
* rtc-screen: Fix bsize in model rd computation for intra chroma
* Fixes to ensure the correct behavior of the encoder algorithms (like
segmentation, computation of statistics, etc.)