v4.1.0-cqp-extended to v4.2.0 · Alliance for Open Media / SVT-AV1 · GitLab

Commits on Source 100+

635f0b23 Arm: Add Neon DotProd svt_vmaf_compute_avg_mad
Jun 12, 2026
affd26ed Arm: Add Neon I8MM svt_vmaf_compute_avg_mad
Jun 12, 2026
3603c1a2 Arm: Add Neon svt_vmaf_apply_unsharp_row
Jun 12, 2026
36b0b175 Arm: Add SVE2 svt_vmaf_apply_unsharp_row
Jun 12, 2026
3880b001 Arm: Add Neon svt_vmaf_vpass_row
Jun 12, 2026
07b00b3d Add Arm Neon single- and 4-reference SAD kernels
Jun 12, 2026
166a84dd Add dedicated Arm Neon 4-reference SAD (sadWxHx4d) kernels
Jun 13, 2026
4ad26c17 Arm Neon: add trailing-zero pre-scan to highbd quantize_fp
Jun 13, 2026
3e063122 Arm Neon: zero the skipped quantizer tail in-loop instead of memset
Jun 13, 2026
6ab12f91 Fix RESIZE_DYNAMIC under --rtc and align with libaom
Jun 14, 2026
ccb0fbb6 Add an RD-gated trailing coefficient zeroing pass after the existing cheap EOB...
Jun 15, 2026
b7c9e4d0 Fix RTC M13
Jun 15, 2026
f54bc752 Reduce VBV test initial buffer to trigger overflow without recode
Jun 15, 2026
963c1e36 Few simplifications in quantize_b*_neon functions
Jun 15, 2026
af65f979 Exploit the fact that quant_shift is power of 2
Jun 15, 2026
4911b2ea Add hardware CRC-32C kernels for hash-based motion estimation
Jun 16, 2026
a2a2f211 Share one process-wide CRC-32C table instead of copying it per block
Jun 16, 2026
57b923d8 Remove the CRC-32C calculator context entirely
Jun 16, 2026
858929ec Encoder API: clean-reject rejected on-the-fly setting changes instead of hanging
Jun 16, 2026
85587223 Add API regression test: rejected on-the-fly change must not deadlock
Jun 16, 2026
16d5b327 Added circuitry to change MG size on the fly for LD
Jun 17, 2026
da172fd0 Use frame type for settings instead of temporal layer
Jun 17, 2026
00b0b4ec Change refresh mask toggle to be updated before refresh mask
Jun 17, 2026
de98a6c0 Add settings validation for on the fly MG size changes
Jun 17, 2026
73322c85 On-the-fly MG size change takes effect immediately
Jun 17, 2026
e5ed78c3 Remove RTC checks in non-RTC RC files
Jun 17, 2026
d93c87de Detect Visual Studio 2026 in svt_av1_print_version
Jun 17, 2026
3735fc4b FwdTxfm2dAsmTest: report absolute Opt ns/call in speed test
Jun 17, 2026
436dddef lowbd forward 4x4: int16 NEON kernel for all tx_types (8-bit)
Jun 17, 2026
81cc5723 lowbd forward 8x8: int16 NEON kernel for all tx_types (8-bit)
Jun 17, 2026
b184a053 lowbd forward 16x16: int16 NEON kernel for all tx_types (8-bit)
Jun 17, 2026
329bdf18 lowbd forward 32x32: int16 NEON kernel (DCT/IDTX)
Jun 17, 2026
a397541e test: worst-case dynamic-range check for lowbd forward (4x4..32x32)
Jun 17, 2026
e59f7efa lowbd forward: guard square HBD int32 bodies with CONFIG_ENABLE_HIGH_BIT_DEPTH
Jun 17, 2026
1474a68e lowbd forward 8x16: int16 NEON kernel for all tx_types (8-bit)
Jun 17, 2026
de6e6e31 lowbd forward 16x8: int16 NEON kernel for all tx_types (8-bit)
Jun 17, 2026
73a01695 lowbd forward 8x32 + 32x8: int16 NEON kernels (8-bit)
Jun 17, 2026
c0dcccd0 lowbd forward 16x32 + 32x16: int16 NEON kernels (8-bit)
Jun 17, 2026
b4a8a55e lowbd forward: extend HBD guard to ported rectangular sizes
Jun 17, 2026
cbc7fecf lowbd forward 4-wide (4x8/8x4/4x16/16x4): int16 NEON kernels (8-bit)
Jun 17, 2026
d5f04b32 fast all-int16 forward (CONFIG_ENABLE_FAST_LBD_TXFM): per-product vqrdmulh
Jun 17, 2026
7b41a3fc lowbd forward: guard now-unused int32 HBD kernels in RTC builds
Jun 17, 2026
07375540 Fix RTC M10 candidate-count heap-buffer-overflow
Jun 18, 2026
e4e76ade cleanup: unify RTC M9/M10 max_can_count branch per review comment
Jun 19, 2026
8fd0bd35 follow-up: keep only commented-out M9 block per review
Jun 19, 2026
9afeefd4 Always init ME/HME data before the static-b64 ME bypass
Jun 19, 2026
da23da62 Add RTC large-resolution encode regression test
Jun 19, 2026
b8a8d13b Drop redundant p_sb_best_mv memset in static-b64 ME bypass
Jun 19, 2026
8e7043bd test: use BSD-3-Clause-Clear header in RTC resolution API test
Jun 19, 2026
29c1a434 Optimize TUNE-VMAF preprocessing
Jun 19, 2026
5a60234d Rate control: fix Debug crash on superres recode (assert before scs is set)
Jun 21, 2026
8c215cbe RTC: make the static-b64 ME bypass deterministic (skip farther refs)
Jun 22, 2026
4858ad29 Cleanup of regular pd0 path and using the light path as the default across...
Jun 22, 2026
7bc48941 Add 8-bit CDEF NEON path (RTCD-dispatched)
Jun 23, 2026
6404a1bc CDEF: cache per-fb dlist/count in search and reuse in apply (SB=64); simplify...
Jun 23, 2026
bdbfbfd6 CDEF: make finish_cdef_search RDO scratch and svt_av1_cdef_frame line/col/row...
Jun 23, 2026
7e44982d CDEF: all-8-bit boundary-aware filter — refactor border handling and drop the...
Jun 23, 2026
9b04591d CDEF: width-specialized gather copy (apply + native-8 search), fold the top...
Jun 23, 2026
aec1e7b8 RTC LBD micro-opts: outline convolve_2d_sr im_block into a NOINLINE helper...
Jun 23, 2026
8de07d9f test: benchmarking/encode.py tweak
Jun 23, 2026
04101600 CDEF: specialize svt_cdef_filter_fb_lbd into luma (BLOCK_8X8) and chroma...
Jun 23, 2026
7d6cf891 CDEF: HBD cdef_copy.h cleanup — use RTCD svt_aom_copy_rect8_8bit_to_16bit...
Jun 23, 2026
5c5aa715 CDEF: fix x86 link error — move svt_cdef_filter_block_8bit_bounded_c (and its...
Jun 23, 2026
ec17f838 Add LPD1 score-based early luma-TX skip decision with grayscale-aware gating
Jun 23, 2026
aca2d898 Do not spawn idle TPL-dispatch processes when TPL is disabled
Jun 26, 2026
d46985f9 gitignore: ignore build-*/ out-of-tree build directories
Jun 27, 2026
79082cea Improve luma-dominant content detection and QP-aware skip scoring
Jun 28, 2026
07737643 App: write coded-resolution recon for resized streams (honor FRAME_SIZE metadata)
Jun 29, 2026
bce8ad54 TF Bug Fixes
Jun 30, 2026
8b3e3dc9 verify_settings: only index based on hl if it's valid
Jun 30, 2026
2f0ca98b Benchmark: fast path for evaluating and tuning SVT with PSNR extracted from stats
Jun 30, 2026
9be5faad Remove unused fields from ConvolveParams
Jun 30, 2026
6d81aeac Clean up in svt_estimate_noise_fp16_neon()
Jun 30, 2026
b1cdf75b Fix --hbd-mds value-0 doc comment and config-dump print guard
Jun 30, 2026
19094a19 Mv: drop pack(1) so int16 x/y stay naturally aligned
Jul 01, 2026
e63a288f Avoid left shift of negative value in MD / ME / inter-prediction
Jul 01, 2026
70b15c6d NEON: avoid unaligned uint32 access in pack / 2b-compress kernels
Jul 01, 2026
673e663b Fix leak of prediction structure group on repeated set_parameter
Jul 01, 2026
bc91d3ee Add ENABLE_STACK_PROTECTOR option to allow disabling -fstack-protector-strong
Jul 01, 2026
7a88adeb Add NEON kernels (gradient-coherence, detail-count, h-pass) + AArch64 support...
Jul 02, 2026
a8cb8696 consolidate memory allocations into large pools
Jul 02, 2026
a5b12b3d Reduce output bitstream buffer size (with dynamic growth if needed)
Jul 02, 2026
3469ade4 Consolidate SB data allocations
Jul 02, 2026
d3c4cb39 Reduce allocations for single thread mode in sys_resource_manager
Jul 02, 2026
c5ab9ea5 ASM_NEON: add Neon convert_8bit_to_16bit and convert_16bit_to_8bit
Jul 05, 2026
d440a65b ASM_NEON: use vmovn_high_u16 in convert_16bit_to_8bit
Jul 05, 2026
8f3dfc2f ASM_NEON: fix clang-format in convert_8bit_to_16bit and convert_16bit_to_8bit
Jul 05, 2026
f4989125 ASM_NEON: fold convert 8/16-bit width==8 path into the main loop
Jul 05, 2026
1b4566ec Avoid forming NULL + offset in temporal filtering / inter-prediction
Jul 06, 2026
2acdf460 Avoid divide-by-zero -> inf -> int-cast UB in TPL boost from r0
Jul 06, 2026
30c42932 Fix lossless uninit-read: gate lossless WHT on TX_4X4
Jul 06, 2026
df776789 Move OBMC convolve scratch buffers out of stack
Jul 07, 2026
0369db3b Unify the FLAT and 3L settings, M10 3L changes, and M7/M8 FLAT changes
Jul 08, 2026
beef4a39 RTC MD tuning: variance-adaptive LPD1 + HME@M12, periodic...
Jul 08, 2026
98bcbfa4 Re-tune m13 for RTC
Jul 09, 2026
627d95c7 Fix memory leak with resize, remove assert
Jul 10, 2026
7ee2bd97 Macros Cleanup
Jul 13, 2026
adedb68f Bump version to 4.2.0-rc and update changelog
Jul 13, 2026
155b7d8e cl: update and reword for 4.2.0
Jul 13, 2026
d0dd3c53 cmake: remove rc tag
Jul 13, 2026

Additional commits have been omitted to prevent performance issues.