1957478 - Update aom to new version 4e3595a426bacb022e8152540a32753c43822f54 from 2025-04-03 00:00:00

Status: RESOLVED FIXED

(Core :: Audio/Video: Playback, enhancement)

Description

[Update Bot]

This update covers 28 commits. Here are the overall diff statistics, and then the commit information.

---

media/libaom/moz.yaml | 4 +-
third_party/aom/CMakeLists.txt | 6 +-
third_party/aom/aom/aom_image.h | 30 +-
third_party/aom/aom/aomcx.h | 17 +-
third_party/aom/aom/internal/aom_image_internal.h | 8 +
third_party/aom/aom/src/aom_image.c | 10 +
third_party/aom/aom_dsp/x86/synonyms.h | 11 -
third_party/aom/aom_dsp/x86/synonyms_avx2.h | 15 -
third_party/aom/av1/arg_defs.c | 3 +-
third_party/aom/av1/av1_cx_iface.c | 22 +-
third_party/aom/av1/common/av1_common_int.h | 7 +
third_party/aom/av1/common/av1_loopfilter.h | 4 +
third_party/aom/av1/common/mvref_common.c | 2 +
third_party/aom/av1/common/quant_common.h | 14 +-
third_party/aom/av1/decoder/obu.c | 18 +-
third_party/aom/av1/encoder/aq_cyclicrefresh.c | 16 +-
third_party/aom/av1/encoder/av1_quantize.c | 39 +-
third_party/aom/av1/encoder/av1_quantize.h | 7 +-
third_party/aom/av1/encoder/bitstream.c | 15 +-
third_party/aom/av1/encoder/encodeframe.c | 3 +-
third_party/aom/av1/encoder/encodeframe_utils.c | 3 +-
third_party/aom/av1/encoder/encoder.c | 58 +-
third_party/aom/av1/encoder/encoder.h | 2 +
third_party/aom/av1/encoder/encoder_utils.c | 3 +-
third_party/aom/av1/encoder/firstpass.h | 2 +
third_party/aom/av1/encoder/gop_structure.c | 20 +-
third_party/aom/av1/encoder/mcomp.c | 17 +
third_party/aom/av1/encoder/nonrd_opt.h | 14 +-
third_party/aom/av1/encoder/nonrd_pickmode.c | 24 +-
third_party/aom/av1/encoder/partition_search.c | 3 +-
third_party/aom/av1/encoder/pickcdef.c | 2 +-
third_party/aom/av1/encoder/picklpf.c | 111 +++-
third_party/aom/av1/encoder/ratectrl.c | 3 +
third_party/aom/av1/encoder/rd.c | 2 +-
third_party/aom/av1/encoder/rdopt.c | 7 +
third_party/aom/av1/encoder/speed_features.c | 59 ++
third_party/aom/av1/encoder/speed_features.h | 14 +-
third_party/aom/av1/encoder/tpl_model.c | 3 +-
third_party/aom/av1/encoder/tune_butteraugli.c | 2 +-
third_party/aom/av1/encoder/txb_rdopt.c | 28 +-
third_party/aom/build/cmake/cpu.cmake | 11 -
third_party/aom/build/cmake/dist.cmake | 2 +-
third_party/aom/build/cmake/generate_aom_config_templates.cmake | 2 +-
third_party/aom/build/cmake/generate_exports.cmake | 2 +-
third_party/aom/build/cmake/pkg_config.cmake | 2 +-
third_party/aom/build/cmake/version.cmake | 2 +-
third_party/aom/docs.cmake | 2 +-
third_party/aom/test/datarate_test.cc | 78 ++
third_party/aom/test/datarate_test.h | 17 +
third_party/aom/test/encode_api_test.cc | 4 +-
third_party/aom/test/metadata_test.cc | 163 ++++-
third_party/aom/test/quantize_func_test.cc | 2 +-
third_party/aom/test/resize_test.cc | 255 ++++++--
third_party/aom/test/sharpness_test.cc | 6 +-
third_party/aom/test/svc_datarate_test.cc | 286 +++++++++-
third_party/aom/test/test.cmake | 8 +-
third_party/aom/test/wiener_test.cc | 235 ++++++++
57 files changed, 1411 insertions(+), 294 deletions(-)

---

4e3595a426bacb022e8152540a32753c43822f54 by Yunqing Wang <[e-mail]>

https://aomedia.googlesource.com/aom/commit/4e3595a426bacb022e8152540a32753c43822f54
Authored: 2025-03-26 15:42:12 -0700
Committed: 2025-03-27 13:41:21 -0700

lc-dec: adaptive use of global motion

Tested speed 1 to 3, on 2 608p sets, the decoder time reduction is
1.5 - 1.7%. The encoder time reduction for speed 2 & 3 is 0.5 - 1%.
The coding loss is 0.03% and 0.1+%.

STATS_CHANGED for low complexity decode

Change-Id: Idaaaf8aa351be083d3d38226c4ca783315ecf83b

Files Modified:

• av1/encoder/speed_features.c

---

e0170e841ed21ae84be9ed0f8328f2d7e0dd208c by Wan-Teh Chang <[e-mail]>

https://aomedia.googlesource.com/aom/commit/e0170e841ed21ae84be9ed0f8328f2d7e0dd208c
Authored: 2025-03-27 11:24:26 -0700
Committed: 2025-03-27 11:24:26 -0700

Set superres_largest_tile_ids[this_index] to 0

Set superres_largest_tile_ids[this_index] to the placeholder value 0
when we set superres_sses[this_index] and superres_rates[this_index] to
the placeholder value INT64_MAX. Otherwise,
superres_largest_tile_ids[this_index] may potentially be used
uninitialized.

Fix Coverity defect 323670 Uninitialized scalar variable.

Change-Id: Ic2c0dc73b1a4a4b2dfda8d8585eaa6e5f5e616c8

Files Modified:

• av1/encoder/encoder.c

---

2b857e1493710b403a4f00241608bf9553dc302c by Mudassir Galagnath <[e-mail]>

https://aomedia.googlesource.com/aom/commit/2b857e1493710b403a4f00241608bf9553dc302c
Authored: 2025-03-27 12:47:04 +0530
Committed: 2025-03-27 17:31:24 +0530

lc-dec: Update the resolution check to 608p

This patch updates the resolution check in
'set_good_speed_features_lc_dec_framesize_dependent()'
from 'is_480p_or_larger' to 'is_608p_or_larger'.

STATS_CHANGED for low-complexity decode, midres.

Change-Id: I657bcb0214298affde775fd5024538218d7d00d9

Files Modified:

• av1/encoder/speed_features.c

---

8d368aa8764456b46df0b9ca751cf89b90c44a04 by Marco Paniconi <[e-mail]>

https://aomedia.googlesource.com/aom/commit/8d368aa8764456b46df0b9ca751cf89b90c44a04
Authored: 2025-03-26 10:15:55 -0700
Committed: 2025-03-26 13:54:20 -0700

Fix to resize_test for realtime_only build

The OnePassGood quality mode can only be used
for non-realtime_only build.

Fixes the nightly failures.

Change-Id: I529d139264853db0181a44843816c56c7f2e1f95

Files Modified:

• test/resize_test.cc

---

f70585048a8aa101348ed9a21c1b2a83e6e54342 by Mudassir Galagnath <[e-mail]>

https://aomedia.googlesource.com/aom/commit/f70585048a8aa101348ed9a21c1b2a83e6e54342
Authored: 2025-03-08 13:04:02 +0530
Committed: 2025-03-26 10:29:40 -0700

lc-dec: Add sf skip_loop_filter_using_filt_error

This patch adds a speed feature 'skip_loop_filter_using_filt_error'
to reset luma filter levels based on the percentage improvement in
SSE between the unfiltered and filtered versions of the current
frame. This logic applies only when both luma and chroma improvement
in SSE is below a threshold. This sf is not enabled when frame type
is OVERLAY_UPDATE or INTNL_OVERLAY_UPDATE or current frame
pyramid_level <= 1.

Performance results of encoder and decoder for the streams generated
with low complexity decode enabled.

        Decode Time               BD-Rate Loss(%)

cpu-used Reduction(%) avg.psnr ssim vmaf vmaf_neg
1 4.878 0.2653 0.4171 0.0887 0.1042
2 4.913 0.3059 0.4582 0.1289 0.1261
3 5.005 0.2939 0.3940 0.0328 0.0321

STATS_CHANGED for low complexity decode

Change-Id: Ifae8dafa69721563ad810859ed66c3b17e5be900

Files Modified:

• av1/encoder/picklpf.c
• av1/encoder/speed_features.c
• av1/encoder/speed_features.h

---

0a6035a0939cfa65e3fcc83f0f6caaaa7bbc075b by Maryla <[e-mail]>

https://aomedia.googlesource.com/aom/commit/0a6035a0939cfa65e3fcc83f0f6caaaa7bbc075b
Authored: 2024-12-09 11:52:27 +0100
Committed: 2025-03-26 09:20:32 -0700

Add API for layer-specific metadata.

Bug: 377851082
Change-Id: I136ec3d41d8424830b77626b02739699cd968b42

Files Modified:

• aom/aom_image.h
• aom/internal/aom_image_internal.h
• aom/src/aom_image.c
• av1/decoder/obu.c
• av1/encoder/bitstream.c
• test/metadata_test.cc

---

8462023ca5276fbad109acd3675b2aa070d68041 by Marco Paniconi <[e-mail]>

https://aomedia.googlesource.com/aom/commit/8462023ca5276fbad109acd3675b2aa070d68041
Authored: 2025-03-18 10:49:30 -0700
Committed: 2025-03-25 09:50:18 -0700

Add external resize test with 4 threads.

To see if tsan race issue exists, as it does in vp9:
https://chromium-review.googlesource.com/c/webm/libvpx/+/6344199

Test covers real-time and good quality mode.

Change-Id: I986de33791c08d43577212c080bf3ce69cc3e2d0

Files Modified:

• test/resize_test.cc

---

7f96b230ff43bba69fb64df2066611dceadd8197 by Wan-Teh Chang <[e-mail]>

https://aomedia.googlesource.com/aom/commit/7f96b230ff43bba69fb64df2066611dceadd8197
Authored: 2025-03-20 15:45:39 -0700
Committed: 2025-03-24 10:57:18 -0700

Add a unit test for bug oss-fuzz:42537236

Convert the reproducer testcase for bug oss-fuzz:42537236 to a unit
test. The test is disabled now.

Steps to reproduce the bug:
cmake ../aom -DSANITIZE=undefined
make -j
./test_libaom --gtest_filter=12bit3 --gtest_also_run_disabled_tests

Bug: oss-fuzz:42537236
Change-Id: I7dbad3bf3e26dbbba54fa002c3d170914cbb1778

Files Modified:

• test/wiener_test.cc

---

f90abd930ae4050db0e325e7bb82bad2514d1146 by Wan-Teh Chang <[e-mail]>

https://aomedia.googlesource.com/aom/commit/f90abd930ae4050db0e325e7bb82bad2514d1146
Authored: 2025-03-20 13:54:42 -0700
Committed: 2025-03-24 10:57:18 -0700

Add a unit test for bug oss-fuzz:384759831

Convert the reproducer testcase for bug oss-fuzz:384759831 to a unit
test. The test is disabled now.

Steps to reproduce the bug:
cmake ../aom -DSANITIZE=undefined
make -j
./test_libaom --gtest_filter=12bit2 --gtest_also_run_disabled_tests

Bug: oss-fuzz:384759831
Change-Id: Ia8c796fdaff9960d6d6bcfacbcc368e9de4f9de7

Files Modified:

• test/wiener_test.cc

---

9680f2b1781fb33b9eeb52409b75c679c8a954be by Jingning Han <[e-mail]>

https://aomedia.googlesource.com/aom/commit/9680f2b1781fb33b9eeb52409b75c679c8a954be
Authored: 2025-03-22 22:07:35 -0700
Committed: 2025-03-22 22:07:35 -0700

Adjust unit test thresholds

Change-Id: I493acee7f5f654fb59889cfa6c2c10be4168abf2

Files Modified:

• test/sharpness_test.cc

---

ec0e01aa7c77be6d6845206b54f2e06edd6251f0 by Jingning Han <[e-mail]>

https://aomedia.googlesource.com/aom/commit/ec0e01aa7c77be6d6845206b54f2e06edd6251f0
Authored: 2025-03-21 13:36:36 -0700
Committed: 2025-03-21 15:54:20 -0700

Apply constraints of the coding options

BUG=b/369204758

Change-Id: I0f666a68c515414f93fb8ad8ca2c162f0f6149dc

Files Modified:

• av1/encoder/rdopt.c

---

9a439b7743ffa2d717fc3c1c60ed418f832066a4 by Jingning Han <[e-mail]>

https://aomedia.googlesource.com/aom/commit/9a439b7743ffa2d717fc3c1c60ed418f832066a4
Authored: 2025-03-19 17:01:16 -0700
Committed: 2025-03-21 15:53:41 -0700

Limit motion vector range

BUG=b/369204758

Change-Id: Ifd17015e096f6a972b55169abb0e3e0fc5b3db07

Files Modified:

• av1/encoder/mcomp.c
• av1/encoder/txb_rdopt.c

---

17f84058b9cd898bfdc68db900a13f8198dcbbe7 by Jingning Han <[e-mail]>

https://aomedia.googlesource.com/aom/commit/17f84058b9cd898bfdc68db900a13f8198dcbbe7
Authored: 2025-03-21 10:32:20 -0700
Committed: 2025-03-21 10:32:20 -0700

Further adjust the unit test thresholds

Change-Id: Idc742d53da5a09a42e97340a0b512a752173bbc1

Files Modified:

• test/sharpness_test.cc

---

a2122d5b43cca06011e93e76b1ba86aa61647a45 by Jingning Han <[e-mail]>

https://aomedia.googlesource.com/aom/commit/a2122d5b43cca06011e93e76b1ba86aa61647a45
Authored: 2025-03-20 09:39:35 -0700
Committed: 2025-03-20 09:39:35 -0700

Fix unit test failure

Change-Id: I27ab789652ccb79d8c99c93c7b6074eac1ff9d19

Files Modified:

• test/sharpness_test.cc

---

6b04fcd63890b35be5faf17642b54e2b88b64875 by Jerome Jiang <[e-mail]>

https://aomedia.googlesource.com/aom/commit/6b04fcd63890b35be5faf17642b54e2b88b64875
Authored: 2025-03-19 20:30:21 -0400
Committed: 2025-03-20 07:34:07 -0700

cmake: bump minimum version to 3.16

Quiets a warning from cmake 3.31.5

This is consistent with the change in webp
https://chromium-review.googlesource.com/c/webm/libwebp/+/6283212

Change-Id: Ic813910fc51ae84c4cfc3f004a503f813e8a5873

Files Modified:

• CMakeLists.txt
• build/cmake/dist.cmake
• build/cmake/generate_aom_config_templates.cmake
• build/cmake/generate_exports.cmake
• build/cmake/pkg_config.cmake
• build/cmake/version.cmake
• docs.cmake

---

c5e1e39e9defe95965d74ad75ac9e19aad1ee9f7 by Marco Paniconi <[e-mail]>

https://aomedia.googlesource.com/aom/commit/c5e1e39e9defe95965d74ad75ac9e19aad1ee9f7
Authored: 2025-03-19 09:08:14 -0700
Committed: 2025-03-19 12:44:41 -0700

Fix to avif_mode datarate test

Fixes the nightly failures.
-delta_mode can only be enabled for non-realtime
-enable_qm enabled only when CONFIG_QUANT_MATRIX

Change-Id: I36795c14aef0e2c52e2b721526b42bbd50bac7e7

Files Modified:

• test/datarate_test.h

---

68cf22f1cadaa8351638ba6c266cc8a3835bb0c9 by Jerome Jiang <[e-mail]>

https://aomedia.googlesource.com/aom/commit/68cf22f1cadaa8351638ba6c266cc8a3835bb0c9
Authored: 2025-03-19 08:00:31 -0400
Committed: 2025-03-19 08:00:31 -0400

Remove RTCD_ARCH_*

They are not used anywhere

Change-Id: I5735234879db90e2799d59d953f3a43f884ec22f

Files Modified:

• build/cmake/cpu.cmake

---

e8749722086cfafe543f605867bb5cb88c40d396 by Jerome Jiang <[e-mail]>

https://aomedia.googlesource.com/aom/commit/e8749722086cfafe543f605867bb5cb88c40d396
Authored: 2025-03-18 18:22:21 -0400
Committed: 2025-03-18 18:22:21 -0400

Remove RTCD_HAVE_*

Change-Id: I618d89da67aea07c96be092544c87487229a6c48

Files Modified:

• build/cmake/cpu.cmake

---

fa2a7d0c046727a61ab6b2565c9971ed118da853 by Kyle Swanson <[e-mail]>

https://aomedia.googlesource.com/aom/commit/fa2a7d0c046727a61ab6b2565c9971ed118da853
Authored: 2025-02-06 14:48:59 -0800
Committed: 2025-03-18 11:46:48 -0700

aomenc: reenable s-frames

broken since 97803fc7ec2272056b510885e55d775a1c652192

Change-Id: I83223b8239b95394f1adcaa0e7cabb6490d37e32

Files Modified:

• av1/encoder/firstpass.h
• av1/encoder/gop_structure.c

---

841f5adf5160556e4c6faea5e6166d9523d69351 by Jingning Han <[e-mail]>

https://aomedia.googlesource.com/aom/commit/841f5adf5160556e4c6faea5e6166d9523d69351
Authored: 2025-03-17 14:17:21 -0700
Committed: 2025-03-18 10:53:13 -0700

Rework sharpness controls

BUG=b/369204758

Change-Id: I5ea5a35f548182ea8b7c45aa45f7e556d5989b2c

Files Modified:

• av1/common/av1_common_int.h
• av1/encoder/av1_quantize.c
• av1/encoder/av1_quantize.h
• av1/encoder/encoder.c
• av1/encoder/encoder_utils.c
• av1/encoder/tune_butteraugli.c
• av1/encoder/txb_rdopt.c
• test/quantize_func_test.cc

---

127be6a600e56735c6b73a2f23c7bd0577a9d3f3 by Marco Paniconi <[e-mail]>

https://aomedia.googlesource.com/aom/commit/127be6a600e56735c6b73a2f23c7bd0577a9d3f3
Authored: 2025-03-06 15:11:25 -0800
Committed: 2025-03-18 09:43:11 -0700

Fix ref_mv for int_pro_motion_estimation() in nonrd_pickmode

For nonrd_pickmode (speed >= 7): fix the ref_mv input to
the int_pro_motion_estimation(), used for newmv search on
non-last reference.

The ref_mv used there should be set via av1_get_ref_mv(),
which gets the ref_mv properly (i.e., gets the mv from
mbmi_ext->ref_mv_stack[] or mbmi_ext->global_mvs[] depending
on ref_mv_idx). In this assert case it was taking ref_mv
directly from ref_mv_stack[], when it should be taking
global_mvs (which is zeromv).

No changes in stats for all rtc sets, speed >= 7.

Added a test that triggers the issue without the fix.
Since it was only triggered on high resolution, the
test upsamples a QVGA clip, using libyuv.

Bug: b:396169342
Change-Id: Ic93dc733017ab0c80db0e1d8a30cc03e3b625139

Files Modified:

• av1/encoder/nonrd_pickmode.c
• test/datarate_test.cc
• test/datarate_test.h
• test/test.cmake

---

616ba8c822fdefb7ede666962a98c1f914ddf615 by Marco Paniconi <[e-mail]>

https://aomedia.googlesource.com/aom/commit/616ba8c822fdefb7ede666962a98c1f914ddf615
Authored: 2025-03-04 22:08:26 -0800
Committed: 2025-03-18 09:41:41 -0700

rtc-svc: Fixes for external resize for svc

-Realloc and reset cyclic refresh map on resolution
change.
-Make the conditions for forced keyframe in
encoder_set_config() only for number_spatial_layers=1.
-Fix the condition for seq->max_frame_width/height, to
handle case where resolution is changed (lowered) after
init but before encoding first frame.

Added unittest for SVC with external resize for (3SL, 3TL),
which triggers failures without these fixes.
The test forces external resize several times during the
sequence, beginning with the very first frame, and sets
the layer bitrates to zero/non-zero to disable/enable
layers.

This test and some of these changes are similar to the
libvpx change:
https://chromium-review.googlesource.com/c/webm/libvpx/+/6192205

Bug: b:400885218
Change-Id: Ife56858c2cb83e16a9b45fc58f5fafe691c9dee7

Files Modified:

• av1/av1_cx_iface.c
• av1/encoder/encoder.c
• test/svc_datarate_test.cc

---

a9f70860b90b9f1e3c71e07c89832df18393ac7d by Marco Paniconi <[e-mail]>

https://aomedia.googlesource.com/aom/commit/a9f70860b90b9f1e3c71e07c89832df18393ac7d
Authored: 2025-03-10 10:48:59 -0700
Committed: 2025-03-10 10:51:20 -0700

Fix height_log2 setting in init_src_params() in pickcdef.c

For the setting of *height_log2:
mi_size_wide_log2[] shoule be mi_size_high_log2[]

Bug: b:401671154
Change-Id: Ib0065461cf501639a7f46aaba81fb77b04f43abb

Files Modified:

• av1/encoder/pickcdef.c

---

b762ef90bbe6091c87c06d4de80e964228f0918f by Marco Paniconi <[e-mail]>

https://aomedia.googlesource.com/aom/commit/b762ef90bbe6091c87c06d4de80e964228f0918f
Authored: 2025-03-09 19:39:50 -0700
Committed: 2025-03-09 20:21:23 -0700

rtc: Fix use of prev_frame for spatial layers

The usage of cm->prev_frame in cyclic_refresh
should only be for single spatial layers, and also
not used for external_ratectrl.

Also fix the update of prev_number_spatial_layers
for dropped frames, and force the usage of
cyclic_refresh_reset_resize() only for TL0, as
cyclic_refresh is only used for TL0 frames.

Bug: b:400885218
Change-Id: Id739cbae29092736e093a56ffc0defc3760e9535

Files Modified:

• av1/encoder/aq_cyclicrefresh.c
• av1/encoder/encodeframe.c
• av1/encoder/ratectrl.c

---

20f697ad51150492534c48e8a214d8192d7e150b by Julio Barba <[e-mail]>

https://aomedia.googlesource.com/aom/commit/20f697ad51150492534c48e8a214d8192d7e150b
Authored: 2025-03-06 17:04:43 -0500
Committed: 2025-03-06 19:43:00 -0500

Make --tune=iq use the all intra luma QM level formula

Using the SSIMULACRA 2 luma QM formula can result in an image that
can visually look too soft or blurry in some cases, especially at
low to mid quality levels.

--tune=ssimulacra2 will continue using the SSIMULACRA 2 formula.

Bug: aomedia:401234702
Change-Id: I60778359becf263b925468813be3de6c242c3f85

Files Modified:

• av1/av1_cx_iface.c
• av1/common/quant_common.h
• av1/encoder/av1_quantize.c

---

c745e3e07e7604735bde390302003d9b712e8f5b by Julio Barba <[e-mail]>

https://aomedia.googlesource.com/aom/commit/c745e3e07e7604735bde390302003d9b712e8f5b
Authored: 2025-03-05 19:25:18 -0500
Committed: 2025-03-06 11:18:46 -0800

(Re-)add --tune=ssimulacra2 / AOM_TUNE_SSIMULACRA2

An all intra only mode tune that focuses on maximizing SSIMULACRA 2
scores.

Currently this is a pure clone of tune IQ, but over time they'll
start diverging more and more as they each get closer to their
respective goals.

Bug: aomedia:401234702
Change-Id: I79531f2f7587a4d126b071414b8487d990a2a43a

Files Modified:

• aom/aomcx.h
• av1/arg_defs.c
• av1/av1_cx_iface.c
• av1/encoder/av1_quantize.c
• av1/encoder/encodeframe_utils.c
• av1/encoder/encoder.c
• av1/encoder/partition_search.c
• av1/encoder/rd.c
• av1/encoder/tpl_model.c
• av1/encoder/txb_rdopt.c
• test/encode_api_test.cc

---

aac1483afc4f406b13e19d55d7ce74a6ea643bb5 by Mudassir Galagnath <[e-mail]>

https://aomedia.googlesource.com/aom/commit/aac1483afc4f406b13e19d55d7ce74a6ea643bb5
Authored: 2025-03-04 16:56:29 +0530
Committed: 2025-03-06 09:31:46 -0800

lc-dec: Add sf adaptive_luma_filter_skip

This patch introduces a speed feature 'adaptive_luma_filter_skip'
to reset filter levels of luma based on the minimum filter levels
of reference frames and pyramid level of current frame.
This logic applies only when luma and chroma filter levels of
current frame are below a threshold determined by its pyramid
level. The sf is not enabled of frame type is OVERLAY_UPDATE or
INTNL_OVERLAY_UPDATE.

Performance results of encoder and decoder for the streams generated
with low complexity decode enabled.

        Decode Time               BD-Rate Loss(%)

cpu-used Reduction(%) avg.psnr ssim vmaf vmaf_neg
1 3.166 0.0645 0.0476 -0.0128 -0.0185
2 3.028 0.0918 0.0434 -0.0382 -0.0320
3 2.474 0.0538 0.0417 -0.0491 -0.0362

STATS_CHANGED for low complexity decode

Change-Id: Ic1164b27338ed0386d5f21a2f720816e710e4ff9

Files Modified:

• av1/common/av1_common_int.h
• av1/common/av1_loopfilter.h
• av1/common/mvref_common.c
• av1/encoder/encoder.c
• av1/encoder/encoder.h
• av1/encoder/picklpf.c
• av1/encoder/speed_features.c
• av1/encoder/speed_features.h

---

5c2ddf4398c5b237d03a274c651f0c3be2e27ff6 by Marco Paniconi <[e-mail]>

https://aomedia.googlesource.com/aom/commit/5c2ddf4398c5b237d03a274c651f0c3be2e27ff6
Authored: 2025-01-23 22:24:23 -0800
Committed: 2025-03-05 10:38:38 -0800

rtc-screen: Speed feature for fast encoding on scene change

Add speed features to prune the newmv and palette mode testing,
to reduce the encoding time spikes, as reported in the issue below.
Spikes can be reduced by close to ~2x in some clips.

This setting has quality loss so it's added to new speed 12, and
only enabled for resolutions >= 720p.
Only affects real-time mode for screen content.

Bug: b:391849810
Change-Id: Iacab126149690642faead4aa2c72ea3621c411a6

Files Modified:

• aom/aomcx.h
• av1/arg_defs.c
• av1/av1_cx_iface.c
• av1/encoder/nonrd_opt.h
• av1/encoder/nonrd_pickmode.c
• av1/encoder/speed_features.c
• av1/encoder/speed_features.h

---

Comment 1

[Update Bot]

I've submitted a try run for this commit: https://treeherder.mozilla.org/jobs?repo=try&revision=6e77b0d44e7eb13189a5c81de7007fe2bf17212e

Comment 2

[Update Bot]

Attachment: Bug 1957478 - Update aom to 4e3595a426bacb022e8152540a32753c43822f54

Comment 3

[Update Bot]

Attachment: Bug 1957478 - Apply mozilla patches for aom

Comment 4

[Update Bot]

The try push is done, we found jobs with unclassified failures.

Needs Investigation (Possible Intermittents):

• toolkit/components/extensions/test/mochitest/test_ext_identity.html
  • 1 of 4 failed on the same (retriggered) task
    - test-linux1804-64-qr/debug-mochitest-plain-spi-nw-15 (XyJvOnF1Tp23yW9oHcDMdQ)

Known Issues:

• dom/workers/test/test_sourcemap_header.html
  - test-linux1804-64-qr/debug-mochitest-chrome-spi-nw-1proc-3 (WWGRjhVjS9-DSQZ4IoFxiQ)

These failures could mean that the library update changed something and caused
tests to fail. You'll need to review them yourself and decide where to go from here.

In either event, I have done all I can and you will need to take it from here. If you
don't want to land my patch, you can replicate it locally for editing with
./mach vendor media/libaom/moz.yaml

When reviewing, please note that this is external code, which needs a full and
careful inspection - not a rubberstamp.

Comment 5

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/0c3af5dcd5d9
Update aom to 4e3595a426bacb022e8152540a32753c43822f54 r=padenot
https://hg.mozilla.org/integration/autoland/rev/129fec73b0b4
Apply mozilla patches for aom r=padenot

Comment 6

[Cristina Horotan [:chorotan]]

https://hg.mozilla.org/mozilla-central/rev/0c3af5dcd5d9
https://hg.mozilla.org/mozilla-central/rev/129fec73b0b4