It would be good to cherry-pick the bug fixes and other improvements from https://aomedia.googlesource.com/aom/ to AVM. Large parts of the repositories are similar enough. Because studying each AOM commit is a lot of work that is unlikely to happen, a minimal effort would consist in copying each commit that does not conflict with AVM and that builds successfully. <details><summary>Script to generate a list of AOM commits that can be applied incrementally to AVM without conflict or build failure</summary> ```sh #!/bin/bash set -eu if [[ "$#" -ge 1 ]]; then DIR="$(eval echo "${1}")" else DIR="$(mktemp -d)" echo "Using temporary directory ${DIR}, pass it as argument to use it again." fi DIR=$(realpath "${DIR}") AOM="${DIR}/aom" AVM="${DIR}/avm" PATCH="${DIR}/patch" # AOM if [[ ! -e "${AOM}" ]]; then echo "Cloning AOM" git clone https://aomedia.googlesource.com/aom "${AOM}" &> /dev/null # To find the commit where AVM was forked: # cd aom # git remote add avm https://gitlab.com/AOMediaCodec/avm.git # git fetch avm main # git checkout -b avm_main avm/main # git merge-base main avm_main fi pushd "${AOM}" > /dev/null COMMITS=$(git rev-list --reverse 50cbe975eaea1edbee693330d0f3c2f33ede2c73..HEAD) popd > /dev/null NUM_COMMITS=$(echo "$COMMITS" | wc -l) # AVM echo "Cloning AVM" rm -rf "${AVM}" git clone --depth=1 https://gitlab.com/AOMediaCodec/avm.git "${AVM}" &> /dev/null pushd "${AVM}" > /dev/null git config --local core.filemode false git config --local commit.gpgsign false echo "build" > ".git/info/exclude" echo "Configuring AVM" cmake -S . -B build &> /dev/null echo "Building AVM" cmake --build build --parallel &> /dev/null popd > /dev/null # Try each AOM commit in AVM. echo "Trying ${NUM_COMMITS} commits" for COMMIT in $COMMITS; do pushd "${AOM}" > /dev/null git format-patch -1 ${COMMIT} --output="${PATCH}" popd > /dev/null pushd "${AVM}" > /dev/null if git apply --check "${PATCH}" 2> /dev/null ; then git apply --whitespace=nowarn "${PATCH}" > /dev/null # Remove untracked files. git clean --force -d --exclude=build --quiet # If there is at least one remaining change: if [[ $(git diff) ]]; then pushd "${AOM}" > /dev/null SHORT_COMMIT=$(git log -1 --pretty=format:"%h" ${COMMIT}) SUBJECT=$(git log -1 --pretty=format:"%s" ${COMMIT}) popd > /dev/null if ! cmake --build build --parallel &> /dev/null ; then echo "Does not build. Discarding ${SHORT_COMMIT} ${SUBJECT}" git reset --hard > /dev/null else echo "${SHORT_COMMIT} ${SUBJECT}" # Commit the changes in case incrementally cherry-picking helps. git add -A git commit -a -m "${SUBJECT}" -m "Cherry-picked from aom ${COMMIT}" > /dev/null fi fi fi popd > /dev/null done ``` </details> <details><summary>List of AOM commits that can be applied incrementally to AVM without conflict or build failure</summary> ``` 142c04a2a7 Add supporting functions to optical flow. 766ec4cf82 Aggressively skip txfm search based on skip rd for speed 5 and 6 2ce375ef1f Add Lucas-Kanade algorithm to optical flow. b56d4cb53f Fix tpl stats calculation ea5d0890f1 In main_loop(), free 'argv' before returning error 9101c0e9d0 Add general description for coefficient coding bd163435c7 Fix enum mixing in pass2_strategy.c 7df3c1fe38 Disable GoldenFrameIntervalTest 6979397bb8 Make is_obu_in_current_operating_point match spec 02937b125d read_obu_header set temporal/spatial_layer_id to 0 35fa4e8674 Aggressively skip txfm search based on skip rd for speed 4 08e6af14d3 Disable StillPictureEncodePresenceTest 9f172eb724 Improve sf prune_inter_modes_based_on_tpl 3100305f99 Implement the frame_header_copy check in the spec. e2bf7e9c91 Document adaptive quantization modes 1, 2, 3 45ea103279 Fix comment typo: aq => deltaq a77319b4af Add ref buffer management state enum c4cb348496 Align rate allocation strategy within a GOP 846447448e Temporarily disable SFrame unit test e53999055e Set allow_screen_content_tools to 0 in rt mode 29051c3934 Improve pre-processing parameter prediction (tune=vmaf_neg) 20e6ae7a8c Remove src alt frame check in show_existing_frame 96579adcc5 Increase max gf interval by 1 de2a658387 Remove commented code in post_encode_update 23a79039bd Add an early return in select_tx_size_and_type() f88a15172d Added links to AV1 technical overview document eb11c30d48 Tautological "this flag is forced to 0" comments 6e464f3a05 Simplify update_mask_best_mv in compound mode search cb3b65b986 Fix a few TODOs in the encoder guide. cbe65bf995 Deleted unused rate control fields. cfaec3d199 Avoid memset in filter_intra_predictor module 267e9b9dba Use Linux man page formatting in arg_show_usage() 85fef1dd02 Revert "Use Linux man page formatting in arg_show_usage()" b52ee6d44a Update CHANGELOG for libaom v2.0.1 21f223da38 Do not truncate long options in arg_show_usage() ac2c30baa0 Skip update kf count when reaching the end of kf group 854b2766a8 Replace non-ASCII characters 20efbde178 Document new value for keyframe filtering option. 4ce7d13e14 Use Round2() in filter_intra_predictor functions 27619e3e0d Rework tpl model layout ca463b1563 Clean-up auto_min_partition_based_on_simple_motion speed feature 30be9a9171 Exclude load_buffer_4x16 from realtime only build 845b203fe3 Fix compilation for CONFIG_LPF_MASK 90aa0dfdf3 Allocate alt-ref buffer conditionally e6afebb285 Add av1_exponential_entropy & av1_laplace_entropy 461d767d5e Fix compile error of collecting partition stats af47cec994 Remove designated initializers from y4m_test 7926ca6cd1 AV1 RT: Introduce CDF update every other frame c075e5578a Remove TODO comment on cpi check in encoder_encode a524b84f9a Add film mode to tune content 9421ac871a The max height of GF group pyramid structure is 5 d2f082e349 Fix tune vmaf build a5ac501c9d Change key frame detection criteria 05c26176a3 Document valid values for aomenc 'usage' option 628c89a2ed Remove ctx->config.enc null check in encoder_init 129879d1a4 Remove unused speed feature setting in all intra mode 140c945e14 Optimize bit stream buffer allocation size ac3c657a14 Do not skip motion search even if zero mv error is small 828a4ab931 Revert "Optimize bit stream buffer allocation size" 508aff4f11 Update model_intra_yrd_and_prune() comment 5767064698 Fix an error 84d9ba8543 Adjustment to the optical flow API. 4597808eb8 Update AUTHORS and mailmap c5d62470d2 Compile aom_clear_system_state to nothing on x86 builds 28f9e96f09 Set ctx->cx_data_sz to 0 if ctx->cx_data is NULL e50deac4e3 Bump up encoder ABI for change to aom_codec_cx_pkt 0f2d37729b Optimize compressed data buffer allocation size 8b6eaa4d37 Document the default of AOME_SET_TUNING is PSNR f034976bbc Bump up codec ABI version for control IDs re-arrangement e591232917 dump_obu: Fix spatial_layer_id output. 8d7db2365f Remove pad in aom_codec_cx_pkt_t data 0cb03d98d4 Optimize loops related to offsetr and offsetc 169849bea5 Guard out of boundary write in rate allocation 82e76bdd6d aom_dsp: Fix aom_noise_strength_lut_init with negative size c1f542fc55 Fix forced keyframe detection 0a5da45c7f Remove txb_coeff_cost_map from RD_STATS 4076c10a32 Remove extra parameters in update_txb_coeff_cost() d2a6356c57 aom_dsp: Asan fix in aom_film_grain_table_lookup d0cac70b54 Fix a free on invalid ptr when img allocation fails 5c9bc41810 Convert input frames to monochrome before encoding ae2a5780e4 Add CONFIG_FRAME_PARALLEL_ENCODE config macro e7acade5ae Fix NULL pointer access when multi-threading is enabled 63b692deb7 Clean-up initialization in MonochromeTest e15483f6ed Use NULL instead of 0 for null pointers 12adc723ac aom_dsp: Fix aom_noise_strength_lut_init with zero size d3fe717f06 Don't include the "common/tools_common.h" header 06279bbcaf Add -z defs linker option when building shared lib acbeeae6e1 aom_codec_internal.h,cosmetics: rm duplicate 'interface' 62e8d9c9da Test operating_point_idc[op_num] == 0 first 927505150f Fix kf counters in fwd kf 41cc545b4b CMake: don't use -z defs on windows 12287adee9 Revert "Fix kf counters in fwd kf" 3ffdc1ab75 Add a SB QP offset option for perceptual quality tuning def0fae453 Add comment for deltaq-mode in encoder helper 028d298a17 Correctly set delta_lf 9b460f0cd2 Document aom_img_set_rect() more precisely 754f6461f3 Rename av1_check_grain_params_equiv() 29dd4e20cd aom_install: don't exclude msvc from install f3ef3f2eab test/best_encode.sh: update aomenc flags 0607f6bef8 image2yuvconfig() should calculate uv_crop_width 2a81195723 Remove unnecessary includes 2deaf2a0e8 ivfdec: replace uses of warn with fprintf(stderr) 0714565d89 encoder.h: add doxygen comment for component_timer a786b2d355 aom/aom_encoder.h: remove configure option reference c694eaa3db Improve aom_satd SSE2 optimization 09b158f6e8 rtc: Add zero bitrate exit to encoder ff103cac8a Fix SIMD inline optimization for building with gcc 67f9fb7ca9 Only search cur_frame in ref_frame_map if non-NULL 07aa6c51c0 Improve code comment about not forcing inlining 5ac45eccdd update libwebm to libwebm-1.0.0.28-19-gba0e436 e98a5fb1a4 Correct the comment for mi_alloc_bsize e6808edffa Fix for v256_shr_n_byte(a, n) when n > 16 faf048e8ec update tools/cpplint.py 4c3263bd65 aom_codec_internal.h: add LIBAOM_FORMAT_PRINTF 34b3136e16 update libwebm to libwebm-1.0.0.28-20-g206d268 a3cf693bb1 highbd_sad_sse2.asm: fix xmm count for sad_skip be8acaf606 test/DummyVideoSource::ReallocImage: check img_ alloc a7795776a4 Fix decoder model initial presentation delay bug. 376cacd1f2 Fix DRY_RUN_COSTCOEFFS rate estimate 1f49a6c0ea aom/aomdx.h: remove configure option references 2e6b85e5d7 grain_table.c: check fscanf returns for Y/Cr/Cb headers 0890faa241 cdef: Pad cdef_directions ae9863d702 cdef(8bpp): Improve max calculations. 32559a4eeb cdef(highbd): Remove unnecessary loop code. e3a94dc55c Allintra: Avoid screen content detection in speed=9 a8facc34df Fix the typo "use_txform_domain_distortion" 6195b8d0d7 Count sign bit cost in tpl rate cost f03b45ec92 Document that the ctrl_id argument must be nonzero 491889ee15 Assert av1_ctrl_args, av1_arg_ctrl_map same size 96147b2166 Simplify InvalidParams, add InvalidControlId 66d4420e86 Set the error string for aom_codec_control() 89de3d5e26 Fix the comment in ratectrl.h d4dfe02d97 update .mailmap dc7efaa56c libwebm: update to libwebm-1.0.0.28-28-gee0bab5 109d8cf6d9 Add missing type-checked ids f563bf5dde aomdx.h: Add last missing control 5f85698e8c av1_lowbd_fwd_txfm_: Remove (fwd_txfm2d_func == NULL) check. 12a8cf8e46 Assert that CNN multithreading has valid workers 82654c4f08 noise_model.c: add a few missing allocation checks 1191d85a15 optical_flow.c: remove unused get_subpixels() baa38c9dae aom_noise_data_validate: add missing alloc checks 3dd1c5ac45 pickrst,update_[ab]_sep_sym: clamp intermediates c1eb8ae394 grain_table.c: add missing alloc checks a12071aea1 [arm]: 8bpp neon quantize cleanup f42f3f2c6b [arm]: Improve av1_quantize_fp_neon(). d05224fc83 aom_configure.cmake: fix flag propagation w/Android NDK <r23 c434716cb6 aom_film_grain_table_lookup: clarify unchecked return 2ee5c86514 av1_pick_filter_restoration: add missing alloc check 0f99d9cb47 Add 10 and 12b ranges to aom_color_range_t comment 1dce6c71cd tools/lint-hunks.py: add text=True to more Subprocess calls bd8392a552 {y4m,film_grain_table}_test: check temp file ptr 2226b71687 GetTempOutFile(): use testing::TempDir() 905abbb423 av1/arg_defs: use a valid target level index in help msg f4ce589a8a test/video_source.h: add missing gtest.h include 1c69d568ca compiler_flags.cmake: fix flag detection w/cmake 3.17-3.18.2 7235e65746 aom_configure.cmake: detect PIE and set CONFIG_PIC a8a5df980e .gitignore: add android studio folder 2b142a105d Make the encoder ABI check allow v3.0.0 and v3.1.x 1df26009da aom_configure: only override CONFIG_PIC if not set on cmd line 6e18489831 Fix off-by-one error of max w/h in validate_config c7bb33c78f Remove obsolete comments in av1_encode_strategy() 27b29a98f6 Remove memset of mbmi_ext_info->frame_base buffer 04737b7a8e Do not use pthread_setname_np on GNU/Hurd 151b52fa45 webmdec,WebmInputContext: make timestamp_ns signed ea12902134 wedge_utils_{avx2,sse2}: fix implicit conversion warnings c947dbef09 AOME_SET_SCALEMODE applies to the next frame only eb8db9d65f md5_utils.c: add AOM_NO_UNSIGNED_SHIFT_CHECK 7d9347a17e sad_test,ShortSrc: break on error a0fc19ba89 av1_convolve_scale_sse4,highbd_vfilter8: merge result vars d510b7a581 obudec,file_is_obu: fix leak on failure f35e491258 obudec,file_is_obu: rewind file on error 818489af43 Add an assertion to av1_set_ssim_rdmult() 6bb6341914 Add a comment for valid_ref_frame_size() d3c97c15b7 Assert that initial_display_delay >= 1 d0b9639ad0 Fix logical expression error in av1_frame_scaled b00034a467 AVX2: Forced inline sadMxNx4d e63dc124f2 Minor optimization for avx2 sadx4d c0239a23c2 av1_resize_and_extend_frame_c should extend frame c73a30defc Fix tile config related comments 296e0507be aom_integer.h: set AOM_FORCE_INLINE to __inline for Win32 6ab16d8612 Update comment on the SCALABILITY_STRUCTURES enum 32ed2d8ac6 Change param of AOME_SET_SPATIAL_LAYER_ID to int 07f66e9a4d av1_fwd_txfm2d_neon: fix -Wmaybe-uninitialized warnings 582d2fd1e9 Don't read 8 bytes from a 3-byte array ccc171c1a7 Simplify rate computation in rate_estimator() 002e191faa Remove common subexpr in aom_flat_block_finder_run f770aef0b5 Modify av1_foreach_transformed_block_in_plane() 7395a3832c Remove an assertion condition in convolve_c function ac7396a7d2 Misc. cleanup for av1_round_shift_array_sse4_1 83ed5ab7db Move some var decls in aom_flat_block_finder_run() 99fdc35efa Set _img->bit_depth in y4m_input_fetch_frame() 8a38501688 thread_loop: don't lock around execute() call 726eeb6887 thread_loop: signal only when status changes to OK 7be63cef99 thread_loop: lock/unlock mutex outside thread loop 21b9f742ba highbd_inv_txfm_neon: simplify round_shift_8x8 d5e4936d1c highbd_inv_txfm_neon: fix undefined behavior 2b9f6b384e set_fwd_txfm_non_scale_range: suppress -Wstringop-overflow 9545a2079b set_fwd_txfm_non_scale_range: convert checks to asserts b7ebfe8828 av1_fwd_txfm2d: remove unused table 8e54fcf69e tools_common,CodecInfo: fix interface fn ptr proto ec6d7b22e7 tools_common,CodecInfo: add missing const c84115434e test.cmake: Exclude _RC_ from libaom_test_srcs.txt 9bf6e76925 Call glibc feenableexcept() to trap division by 0 11360fc44b Check for feenableexcept() failure e6f555bcfb Print info message after feenableexcept() failure 7904274483 Fix signed integer overflow. c8ce570992 temporal_filter_test.cc: Remove unused using decls e8bcc97952 ethread.c: remove redundant returns 96158b0907 util.cmake: Fix typo AOM_CONFIG_VARS in comment 365a968684 Fix comment typos (likely copy-and-paste errors) 1990370876 Remove duplicated function in aomenc.sh ac28610f0e Add comments about aom_codec_enc_init_ver failure 1366075393 Call feenableexcept() after InitGoogleTest() aa2c60d71c Have aom_codec_error take const aom_codec_ctx_t * 4db345f322 Minor improvement to the default get_msb function 17c580e642 Remove stage_idx variable in av1_fadst4_new_neon() ddedbb1ddd Remove the feenableexcept() call in main() 055657d55d Setter codec controls return early if same value ada77c87b5 aom_timer.h: Don't include aom/aom_integer.h e16208f07d update libwebm to libwebm-1.0.0.29-9-g1930e3c 00fb4e0fd2 cmake: include tools_common.h in usage_exit.c{,c} c743cceaa1 y4menc: make colorspace8() static 4b80045c7c tools_common.h: fix get_aom_{decoder,encoder}_count protos 938fe87604 lightfield_bitstream_parsing: make process_tile_list static 5a87816f28 arm_cpudetect: fix aom_arm_cpu_caps w/Win Arm64 3dbb8cabfb aom_thread: Fix leaking min/max macros 8ad11952c3 fft_test: fix operator<< output ca11f82d39 simd,arm: use intrinsic functions where appropriate b6cfc3cf45 fix redefinition of WIN32_LEAN_AND_MEAN 9ee85431dd v{64,128}_intrinsics_arm.h: add some missing casts 6b1ffc3256 Don't use -Wl,-z,defs with Clang's sanitizers 2ac866cbb4 Explicitly specify /std:c++14 for MSVC 4f0dafa112 third_party/libyuv: remove unused asm files d1319e8acb Fix update of skip_build_pred flag in interp filter search de85f58ede v64_intrinsics_arm.h: simplify u32_store_unaligned 15026ee7e5 Use 64-bit multiplication in does_level_match() 40120fca72 Replace include(FindFoo) with find_package(Foo) 6062c09628 Destroy codec context in case of an error 0f3d8c8daa transform_test_base.h: add missing includes 99ed4b7501 Update a comment in av1_write_obu_header() 7c79bd3397 fft*: add missing prototypes 3cbf29248b lightfield_tile_list_decoder: use fatal() for OOM c3bf067159 tools_common,die_codec(): output to stderr b1c8367d29 Fix a spelling error: Unqeual -> Unequal 596d2d03ab iadst4x4_neon: fix conversion warning d482168a39 Fix -Wcast-qual warnings in yy_loadu2_128 6a1d482399 Fix a misspelling of "maximum" 63caf2eaaf test/examples.sh: restore working directory b/w tests 189c74d56c Optimize the implementation of av1_cnn_activate_c() fbed8eca54 Cosmetic: Remove extra parentheses bf894dafb8 psnr.h,cosmetics: fix a typo (PNSR -> PSNR) c00aec9969 fwd_txfm_neon.c: Remove unused aom_fdct8x8_1_neon 4a238a3309 bitops.h: enable MSVC intrinsics for Arm targets 5d1040a5ab Fix potential deadlock issue in multithreaded modules e1b78c57e4 Include stdint.h for UINT64_MAX. a5825fc2e1 Clarify the comment for update_error_state() c0be7636df Add -flax-vector-conversions=none to Clang builds 036e166964 Restructure the control flow in decoder_set_fb_fn ab4257591a Fix potential memory leaks in av1_loop_restoration_alloc() e2ba9f09f2 Fix over reads in aom_convolve_copy_neon() 90222123fa require_cxx_flag should say Checking C++ compiler 4a55828395 Update third_party/libwebm to commit affd7f4 8a3dfd5395 Pass the aligned width/height in av1_lookahead_push 461d7a5ddd Remove stale comment in av1_lookahead_push() fce89190af Forward-port stack alignment change from libvpx. 8db7d78930 Improve fuzzer coverage cae8fbe572 Change license to BSD 2. 01ee589ad6 Add link to MemorySanitizer doc about -Wl,-z,defs 14010c6f0f Add const to two input pointer parameters 57586292f7 Add documentation on SIMD usage 87a1e9eaa6 Fix a clang-tidy misc-include-cleaner warning 874b7ca3ad disflow_avx2.c: Make compatible with gcc <= 9 b53a630746 Increase filter buffer size to silence warnings d3d5e945a1 Adding comment to gaussian filter buffer alloc 6b4246129d Simplify conditions in aom_img_plane_width/height f4eaf8b55e common/tools_common.h: port f{seek,tell}o fix from libvpx 753413f8a1 av1_dec_fuzzer: add aom_codec_peek_stream_info coverage 283aaef609 av1_dec_fuzzer: Initialize stream_info.is_annexb 4637f5d7bb set_encoder_config,cosmetics: collapse some blank lines 249c4c48d3 synonyms.h: s/_mm_loadu_si64/_mm_loadl_epi64/ 4aefb9325a Detect an invalid row offset get_ls_tile_buffer() 03eb247ac4 tools/obu_parser.cc: make some functions static b930f29247 Include "config/aom_config.h" aaccabe092 Define pthread_attr_getstacksize/setstacksize 0f151b1b80 external_frame_buffer_test.cc: update copyright 817973695d rtcd.pl: add license header to generated files 9759ad181d Fix unsigned int overflow in init_rate_histogram() 49e5deed0c List standard C headers in alphabetical order e756a62b37 Fix interger overflow in _calc_iframe_target_size_one_pass_cbr 0a5353aac2 Update third_party/libwebm to commit f4b07ec e448dcd13a update libwebm to libwebm-1.0.0.31-10-g3b63004 7b24d44e6d Document sframe_mode setting is not implemented a3962ef9a6 debug_util.h: fix prototype typo fa82bedb8d bitwriter_buffer.c: make aom_wb_overwrite_bit() static 172161d87f cdef_block.c: make cdef_directions_padded[] static 17dfba0fbc Add #ifndef header guard to config/aom_version.h 9ed60cc5ff Add macro name as comment for header guard #endif 599199beb6 Clamp the calculation of sb64_target_rate to INT_MAX e736ba87af Remove two unnecessary/incorrect uint32_t casts dba2aaf00c Add the experimental feature policy 09c7db8939 Some cleanup for av1_convert_sect5obus_to_annexb() fece42acda Revert "Export film grain funcs again" a42ea198de av1/exports_test: rm av1_fwd_txfm2d_test 591fcbabf3 av1/exports_dec: rm av1_add_film_grain b8f3821562 aom/exports_com: rm aom_img_metadata_array_{alloc,free} 8d389aa4f3 metadata_test.cc: remove unneeded bitstream.h include ed5a2445b8 Do not use aom_write_bit_buffer in write_av1config 61046cf6d4 misc_model_weights.h: make tables static 00e3da1954 fast_9.c: make aom_fast9_corner_score static e88161627a av1_encoder.dox: fix doxygen 1.9.8 warnings 58017cc719 args_helper.c: make arg_init() static 95a2b89739 av1_fwd_txfm1d_sse4.c: rm unused av1_fadst4_sse4_1 77846f5e9a Ignore .DS_Store ecad807691 timing: make arrays const 278c089117 Use the `img` variable in Encoder::InitEncoder() 4afbf1b464 cmake: add float-cast-overflow w/ubsan ``` </details> This is a rather small subset of the 5883 AOM commits merged since the AVM fork 50cbe97. Not everything is relevant in this small subset but there may be a few bug fixes and general improvements worth to import.