platform/upstream/libvpx.git
6 years agoRemove deprecated arf_update_index from GF_GROUP
Jingning Han [Fri, 28 Sep 2018 18:28:35 +0000 (11:28 -0700)]
Remove deprecated arf_update_index from GF_GROUP

As we move to unify the GOP structure layout control, the variable
arf_update_idx and arf_ref_idx are deprecated.

Change-Id: Iadcb9e6033d419d4b2015fe747c23be59a7da787

6 years agoMerge "Fix minor bug in calculation of max arf group length."
Jingning Han [Fri, 28 Sep 2018 17:38:26 +0000 (17:38 +0000)]
Merge "Fix minor bug in calculation of max arf group length."

6 years agoMerge "Adjustment of GOP intra factor for multi-layer."
Jingning Han [Fri, 28 Sep 2018 17:38:14 +0000 (17:38 +0000)]
Merge "Adjustment of GOP intra factor for multi-layer."

6 years agoMerge "Add MID_OVERLAY_UPDATE frame type"
Jingning Han [Fri, 28 Sep 2018 17:18:54 +0000 (17:18 +0000)]
Merge "Add MID_OVERLAY_UPDATE frame type"

6 years agoMerge "Refactor gf_overlay frame type update"
Jingning Han [Fri, 28 Sep 2018 17:18:46 +0000 (17:18 +0000)]
Merge "Refactor gf_overlay frame type update"

6 years agoMerge "Revert "Merge "Adapt GOP size threshold to the allowed layer depth"""
Paul Wilkins [Fri, 28 Sep 2018 17:16:01 +0000 (17:16 +0000)]
Merge "Revert "Merge "Adapt GOP size threshold to the allowed layer depth"""

6 years agoFix minor bug in calculation of max arf group length.
Paul Wilkins [Thu, 27 Sep 2018 09:55:05 +0000 (10:55 +0100)]
Fix minor bug in calculation of max arf group length.

Their is no valid last boosted Q availably when estimating the maximum
group length for the first ARF group in a clip, so use a value based on
the current max q.

Change-Id: Ida0b4bfb7ce7433089ad808abed7f59c88527a81

6 years agoAdjustment of GOP intra factor for multi-layer.
Paul Wilkins [Fri, 28 Sep 2018 15:54:03 +0000 (16:54 +0100)]
Adjustment of GOP intra factor for multi-layer.

This provides and alternative (still to be tuned for edge cases)
approach to adjusting the gop intra factor when multi-layer coding
is in effect that does not alter single layer coding.

Change-Id: Iba86d65a6e68e86aa031b7e1f0b6a4c55761b1b8

6 years agoMerge "Fix a loophole in nonrd_pick_partition()"
Hui Su [Fri, 28 Sep 2018 16:06:00 +0000 (16:06 +0000)]
Merge "Fix a loophole in nonrd_pick_partition()"

6 years agoRevert "Merge "Adapt GOP size threshold to the allowed layer depth""
Paul Wilkins [Fri, 28 Sep 2018 12:17:31 +0000 (13:17 +0100)]
Revert "Merge "Adapt GOP size threshold to the allowed layer depth""

This reverts commit 5efde3914f1a85061e51bc8d96984216e976e90c, reversing
changes made to 3a291593726a3f0eff5e59983e84de20d0bf294c.

This is badly broken and may help somewhat for multi-layer but is hurting
massively in single layer encodes.

I ran this through this morning and while it often helps in SSIM it is badly down
for global PSNR and PSNR-HVS with some clips down by 35-40%. This is in line
with previous experiments where I have found that a bigger boost helps SSIM
but hurts PSNR and PSNR HVS.

I was also working on changes to the I factor that gave some improvements
in single layer though these were based upon the active Q mostly. I also have
looked at a bug for the first group where int_lbq is not properly defined and
will submit an interim patch for this while I look for a better solution.

In the meantime I think we should revert this.

The (Global PSNR, SSIM, PSNR-HVS) for the patch as is in my runs for
single layer vs a couple of days ago seem to be (-ve is better).

Low res 0.346, -1.475, 0.239
mid res  1.581, -1.300, 1.731 (worst result down by 30-40% in psnr)
hdres 0.665, -0.712, 1.043 (worst result down by 17-19% in psnr)
NF2k 0.927, 0.111, 1.3220 (Worst result down by 5-7% in psnr)

Change-Id: I55952b71b8cfc5a84484b3b659c5f8a530f3a755

6 years agoAdd MID_OVERLAY_UPDATE frame type
Jingning Han [Fri, 28 Sep 2018 04:01:08 +0000 (21:01 -0700)]
Add MID_OVERLAY_UPDATE frame type

Add a MID_OVERLAY_UPDATE abstract to support multi-layer
ARF-Overlay frame based approach. When setting the frame update
type to be USE_BUF_FRAME, the encoder will use show_existing_frame
to process the intermediate ARF frames. When setting the frame
update type to be MID_OVERLAY_UPDATE, the intermediate ARF frames
will go through an overlay frame for display.

Change-Id: Ia0c91452c09d39312ac22d855cdf681b7da851c5

6 years agoMerge "Remove deprecated variables from GF_GROUP structure"
Jingning Han [Fri, 28 Sep 2018 03:52:38 +0000 (03:52 +0000)]
Merge "Remove deprecated variables from GF_GROUP structure"

6 years agoMerge "Remove unused for-loop in multi-layer arf bit allocation"
Jingning Han [Fri, 28 Sep 2018 03:52:27 +0000 (03:52 +0000)]
Merge "Remove unused for-loop in multi-layer arf bit allocation"

6 years agoMerge "Adapt GOP size threshold to the allowed layer depth"
Jingning Han [Fri, 28 Sep 2018 03:52:18 +0000 (03:52 +0000)]
Merge "Adapt GOP size threshold to the allowed layer depth"

6 years agoRefactor gf_overlay frame type update
Jingning Han [Thu, 27 Sep 2018 23:27:43 +0000 (16:27 -0700)]
Refactor gf_overlay frame type update

Factor out common code.

Change-Id: Ia548842557d85ab692fe658acf97d61f008e9588

6 years agoFix a loophole in nonrd_pick_partition()
Hui Su [Thu, 27 Sep 2018 17:12:55 +0000 (10:12 -0700)]
Fix a loophole in nonrd_pick_partition()

In some rare cases, all possible paritions may be skipped during RD
search. The patch makes the encoder do rectangular partition search if
both partition-none and partition-split are not allowed.

Tested on the rtc and ytlivehr testsets with speed 5 and 7, no coding
stats changes were observed.

Change-Id: I8b6d8b62b6d2431be8e73317d113311c98f631d5

6 years agoRemove deprecated variables from GF_GROUP structure
Jingning Han [Thu, 27 Sep 2018 19:00:56 +0000 (12:00 -0700)]
Remove deprecated variables from GF_GROUP structure

Change-Id: I8c02216a369be6a51af9872f3ce05045038fc481

6 years agoRemove unused for-loop in multi-layer arf bit allocation
Jingning Han [Thu, 27 Sep 2018 17:49:44 +0000 (10:49 -0700)]
Remove unused for-loop in multi-layer arf bit allocation

The for-loop is not taking effect any more.

Change-Id: Ief2763990a6d4f487a5eb4972012d86379573d55

6 years agoAdapt GOP size threshold to the allowed layer depth
Jingning Han [Wed, 26 Sep 2018 23:34:16 +0000 (16:34 -0700)]
Adapt GOP size threshold to the allowed layer depth

Increase the total prediction error budget linearly with the
allowed ARF layer depth. This in general improves the compression
performance, but does hit corner cases on a few clips at very
low bit-rate range (corresponding to 26 - 28 dB range). To mitigate
such problem, we temporarily work around this problem by limiting
the first GOP size to be ~8 so as to not drain up the bit resource.

The overall compression performance improvements over the current
multi-layer ARF system in speed 0 are:

           overall PSNR      avg PSNR        SSIM
lowres     -0.47%            -0.13%         -1.51%
midres     -1.30%            -1.16%         -2.80%
hdres      -0.91%            -0.84%         -2.15%

Change-Id: Ia4880ab63e98e15a9db99aea6eabfd3d1da9270d

6 years agoMerge "add cfi sanitizer"
Johann Koenig [Thu, 27 Sep 2018 15:08:12 +0000 (15:08 +0000)]
Merge "add cfi sanitizer"

6 years agoadd cfi sanitizer
Johann [Wed, 26 Sep 2018 23:52:54 +0000 (16:52 -0700)]
add cfi sanitizer

Change-Id: I4262bb631c248ad188f09a37d774d1759695b0d7

6 years agoCONFIG_WEBM_IO: include webmids.h
Johann [Wed, 26 Sep 2018 23:13:01 +0000 (16:13 -0700)]
CONFIG_WEBM_IO: include webmids.h

This was previously brought in with the examples. When building
with --disable-examples and --enable-codecs-srcs, this file
gets lost.

Change-Id: Id8bd67cb78c4f06647f34e85f425dfc701c640c0

6 years agoMerge changes If96a8a1c,Iaf535fde,Icbde9880
Angie Chiang [Wed, 26 Sep 2018 19:06:29 +0000 (19:06 +0000)]
Merge changes If96a8a1c,Iaf535fde,Icbde9880

* changes:
  Add vp9_full_pixel_diamond_new
  Add vp9_refining_search_sad_new
  Add vp9_diamond_search_sad_new

6 years agoMerge "Use layer dependent gfu_boost factor"
Jingning Han [Wed, 26 Sep 2018 04:18:15 +0000 (04:18 +0000)]
Merge "Use layer dependent gfu_boost factor"

6 years agoAdd vp9_full_pixel_diamond_new
Angie Chiang [Tue, 25 Sep 2018 22:26:33 +0000 (15:26 -0700)]
Add vp9_full_pixel_diamond_new

This function will call vp9_diaomond_search_sad_new /
vp9_refining_search_sad_new accordingly.

Change-Id: If96a8a1c9c06b6b4ed3aac6d59bdb03f20c96df9

6 years agoAdd vp9_refining_search_sad_new
Angie Chiang [Tue, 25 Sep 2018 01:00:11 +0000 (18:00 -0700)]
Add vp9_refining_search_sad_new

The new version of refining search function will take into account
neighbor motion vectors' inconsistency while doing mv search

Change-Id: Iaf535fde04805de3dc7dd9a32f1695bf454e2d63

6 years agoAdd vp9_diamond_search_sad_new
Angie Chiang [Tue, 25 Sep 2018 00:37:05 +0000 (17:37 -0700)]
Add vp9_diamond_search_sad_new

This new version of diamond search function will take into account
neighbor motion vectors' inconsistency while doing mv search

Change-Id: Icbde9880305cb8aea7937d6ddcef1597bf9be018

6 years agoUse layer dependent gfu_boost factor
Jingning Han [Mon, 24 Sep 2018 20:53:53 +0000 (13:53 -0700)]
Use layer dependent gfu_boost factor

When multi-layer ARF is enabled, use the corresponding gfu_boost
factor assigned to each ARF to compute the best_quality_index
adjustment. This on average improves the coding performance by
0.2% for lowres and hdres, 0.4% for ntflx2k. It seems this change
will only affect a small group of clips, e.g., pamphlet, bowing,
mobcal_720p, etc., which tend to gain 4-5%, whereas the rest
clips remain largely identical coding statistics.

Change-Id: Ie19636a6cf32214aefd73e21ead2aea647ddbca8

6 years agoMerge "Remove redundant code"
Hui Su [Tue, 25 Sep 2018 16:53:38 +0000 (16:53 +0000)]
Merge "Remove redundant code"

6 years agoMerge "clang-format v6.0.1"
Johann Koenig [Tue, 25 Sep 2018 14:25:52 +0000 (14:25 +0000)]
Merge "clang-format v6.0.1"

6 years agoMerge "vp9,encoder: check pointers before member access"
James Zern [Tue, 25 Sep 2018 03:40:00 +0000 (03:40 +0000)]
Merge "vp9,encoder: check pointers before member access"

6 years agoclang-format v6.0.1
Johann [Tue, 25 Sep 2018 01:31:35 +0000 (18:31 -0700)]
clang-format v6.0.1

Change-Id: I83c7e64fe70f7c49aa2492ed2d640c6756b7ebaa

6 years agoMerge "sanitizer: sse2 - fix unaligned double stores"
Johann Koenig [Mon, 24 Sep 2018 23:11:49 +0000 (23:11 +0000)]
Merge "sanitizer: sse2 - fix unaligned double stores"

6 years agoMerge "segfault: fix missing alignment declaration"
Johann Koenig [Mon, 24 Sep 2018 22:30:46 +0000 (22:30 +0000)]
Merge "segfault: fix missing alignment declaration"

6 years agoMerge "fix integer overflow caused by uninitialized memory"
Johann Koenig [Mon, 24 Sep 2018 22:30:09 +0000 (22:30 +0000)]
Merge "fix integer overflow caused by uninitialized memory"

6 years agosanitizer: sse2 - fix unaligned double stores
Matthias Räncker [Fri, 21 Sep 2018 13:33:18 +0000 (15:33 +0200)]
sanitizer: sse2 - fix unaligned double stores

Signed-off-by: Matthias Räncker <theonetruecamper@gmx.de>
Change-Id: I838c8678e62f7cff13387b84d4f3ea42710a67ea

6 years agoRemove redundant code
Hui Su [Mon, 24 Sep 2018 18:27:54 +0000 (11:27 -0700)]
Remove redundant code

in set_rt_speed_feature_framesize_independent().

use_nonrd_pick_mode is already set for speed >= 5, so need to set again
for speed >= 6.

Change-Id: Idb0a4b36d21e305bd63f19e98a70f615ad76f514

6 years agoMerge "Improve subpel MV search for speed 1"
Hui Su [Mon, 24 Sep 2018 18:15:49 +0000 (18:15 +0000)]
Merge "Improve subpel MV search for speed 1"

6 years agofix integer overflow caused by uninitialized memory
Matthias Räncker [Sun, 23 Sep 2018 11:07:44 +0000 (13:07 +0200)]
fix integer overflow caused by uninitialized memory

Signed-off-by: Matthias Räncker <theonetruecamper@gmx.de>
Change-Id: I55ec2a803eff89b07376459e334d4e949bfcb2cc

6 years agosegfault: fix missing alignment declaration
Matthias Räncker [Sun, 23 Sep 2018 08:34:44 +0000 (10:34 +0200)]
segfault: fix missing alignment declaration

These variables are being fed to sse2 functions, that use aligned
loads.

Signed-off-by: Matthias Räncker <theonetruecamper@gmx.de>
Change-Id: I796c3483c6f3425d63d9262b02b19da59d536600

6 years agoRevert "Revert "Revert "Loopfilter MultiThread Optimization"""
James Zern [Sat, 22 Sep 2018 19:19:16 +0000 (12:19 -0700)]
Revert "Revert "Revert "Loopfilter MultiThread Optimization"""

This reverts commit bf6299010e815e111d7326530c249e9d99611f34.

segfaults, causes an assertion failure with corrupt input:
get_uv_tx_size: Assertion `mi->sb_type < BLOCK_8X8 ||
ss_size_lookup[mi->sb_type][pd->subsampling_x][pd->subsampling_y] !=
BLOCK_INVALID

BUG=webm:1562

Change-Id: I05a711cad3d8e7f1a8e64422b4356bdf4edb3d12

6 years agoMerge "vp8: exit with bad fragment size in decoder."
Jerome Jiang [Sat, 22 Sep 2018 07:15:21 +0000 (07:15 +0000)]
Merge "vp8: exit with bad fragment size in decoder."

6 years agoMerge "Rework is_compound_allowed logic at encoder"
Jingning Han [Sat, 22 Sep 2018 04:00:11 +0000 (04:00 +0000)]
Merge "Rework is_compound_allowed logic at encoder"

6 years agoMerge "internal stats: fix mem leak and initialize memory"
Johann Koenig [Fri, 21 Sep 2018 22:36:53 +0000 (22:36 +0000)]
Merge "internal stats: fix mem leak and initialize memory"

6 years agoMerge "better-hw-compatibility: fix out of bounds access"
Johann Koenig [Fri, 21 Sep 2018 22:30:52 +0000 (22:30 +0000)]
Merge "better-hw-compatibility: fix out of bounds access"

6 years agoMerge "Revert "third_party/googletest: update to v1.8.1""
Johann Koenig [Fri, 21 Sep 2018 22:20:29 +0000 (22:20 +0000)]
Merge "Revert "third_party/googletest: update to v1.8.1""

6 years agoRevert "third_party/googletest: update to v1.8.1"
Johann Koenig [Fri, 21 Sep 2018 22:19:14 +0000 (22:19 +0000)]
Revert "third_party/googletest: update to v1.8.1"

This reverts commit 7d777ce6136d7325a102524b6d80c19bd420e4b6.

Reason for revert: Generates build warnings on VS10/VS12

third_party\googletest\src\include\gtest/gtest-printers.h(1036): error C2770: invalid explicit template argument(s) for 'AddReference<const ::std::tr1::tuple_element<I,std::tr1::tuple<_Arg0,_Arg1>>::type>::type testing::internal::TuplePolicy<TupleT>::get(const std::tr1::tuple<_Arg0,_Arg1> &)' [C:\src\buildbot\test-libvpx\tests\i9vRsze8hQ\.build-x86-win32-vs10\test_libvpx.vcxproj]

Original change's description:
> third_party/googletest: update to v1.8.1
>
> BUG=webm:1559
>
> Change-Id: I7a0b16c7bf3f97db2d8650a190b93aae7e12a948

TBR=tomfinegan@chromium.org

Bug: webm:1559
Change-Id: Ia1a7354084c778a4c4e91b33fef6462e88986d1e

6 years agobetter-hw-compatibility: fix out of bounds access
Matthias Räncker [Thu, 20 Sep 2018 17:57:25 +0000 (19:57 +0200)]
better-hw-compatibility: fix out of bounds access

With --enable-better-hw-compatibility an access to array element -1
can be observed for VP9/ActiveMapTest.Test/0
../vp9/encoder/vp9_rdopt.c:3938:53: runtime error:
  index -1 out of bounds for type 'RefBuffer [3]'

There doesn't seem anything that would prevent ref_frame from being 0.
If there is no reference frame it can probably be assumed that it
isn't scaled.

Signed-off-by: Matthias Räncker <theonetruecamper@gmx.de>
Change-Id: I0a29cd0ffc9a19742e5e72203d5ec5d0a16eac7a

6 years agoMerge "sanitizer: fix unaligned loads"
Johann Koenig [Fri, 21 Sep 2018 18:34:02 +0000 (18:34 +0000)]
Merge "sanitizer: fix unaligned loads"

6 years agovp8: exit with bad fragment size in decoder.
Jerome Jiang [Fri, 21 Sep 2018 18:16:26 +0000 (11:16 -0700)]
vp8: exit with bad fragment size in decoder.

BUG=webm:1555
Change-Id: Ie024c9f5a21f4ed05ab6b93f1677662eeef9e6d8

6 years agoMerge "Use different corner detection score"
Angie Chiang [Fri, 21 Sep 2018 17:32:17 +0000 (17:32 +0000)]
Merge "Use different corner detection score"

6 years agoImprove subpel MV search for speed 1
Hui Su [Tue, 18 Sep 2018 17:33:23 +0000 (10:33 -0700)]
Improve subpel MV search for speed 1

Do one more subpel MV search each round. This improves coding
efficiency slightly:

lowres 0.12%
midres 0.11%
hdres  0.13%

Also renames the control flag for subpel MV search quality.

Encoding speed loss is less than 1%.

This only affects speed 1.

Change-Id: I3aecd25342f2dcacea6c143db494f7db6282cb92

6 years agoRework is_compound_allowed logic at encoder
Jingning Han [Fri, 21 Sep 2018 15:24:24 +0000 (08:24 -0700)]
Rework is_compound_allowed logic at encoder

Allow the encoder to fully utilize the decoder's capability to
handle both 1 fwd + 2 bwd case and 2 fwd + 1 bw case.

Change-Id: I3f984d52552ddb701b80b042d979f8fe09dd3a80

6 years agointernal stats: fix mem leak and initialize memory
Matthias Räncker [Fri, 21 Sep 2018 12:23:44 +0000 (14:23 +0200)]
internal stats: fix mem leak and initialize memory

Without calloc valgrind reports usuage of uninitialized data in
vpx_get_ssim_metrics.

Signed-off-by: Matthias Räncker <theonetruecamper@gmx.de>
Change-Id: I9cd38b8031ea3f22c1436894ddaf9e0ccf5a654e

6 years agoMerge "Generalize encoder comp_var_ref setting"
Jingning Han [Fri, 21 Sep 2018 03:32:44 +0000 (03:32 +0000)]
Merge "Generalize encoder comp_var_ref setting"

6 years agoMerge "Skip checking compound modes with same sign bias for sub8x8"
Jingning Han [Fri, 21 Sep 2018 03:32:08 +0000 (03:32 +0000)]
Merge "Skip checking compound modes with same sign bias for sub8x8"

6 years agoMerge "Update the comp_refs counts"
Jingning Han [Fri, 21 Sep 2018 03:32:01 +0000 (03:32 +0000)]
Merge "Update the comp_refs counts"

6 years agoMerge "Skip RD check for compound modes that have same sign bias"
Jingning Han [Fri, 21 Sep 2018 03:31:57 +0000 (03:31 +0000)]
Merge "Skip RD check for compound modes that have same sign bias"

6 years agoMerge "Sync ref frame writing with decoder"
Jingning Han [Fri, 21 Sep 2018 03:31:50 +0000 (03:31 +0000)]
Merge "Sync ref frame writing with decoder"

6 years agoMerge "Add frame_start/end to gf_group"
Jingning Han [Fri, 21 Sep 2018 03:31:45 +0000 (03:31 +0000)]
Merge "Add frame_start/end to gf_group"

6 years agosanitizer: fix unaligned loads
Matthias Räncker [Fri, 21 Sep 2018 00:16:31 +0000 (02:16 +0200)]
sanitizer: fix unaligned loads

Another instance of unaligned 4-byte loads.

Signed-off-by: Matthias Räncker <theonetruecamper@gmx.de>
Change-Id: I06afc5405bb074384eec7a8c8123e5803e522937

6 years agoUse different corner detection score
Angie Chiang [Thu, 20 Sep 2018 23:42:14 +0000 (16:42 -0700)]
Use different corner detection score

This corner detection score is better at measuring the level of
details in each block.

Change-Id: I16327a7664144ddc463c29babd11d0ca2fbb54a0

6 years agoMerge "third_party/googletest: update to v1.8.1"
Johann Koenig [Thu, 20 Sep 2018 22:08:17 +0000 (22:08 +0000)]
Merge "third_party/googletest: update to v1.8.1"

6 years agoMerge "sanitizer: fix unaligned load/stores"
Johann Koenig [Thu, 20 Sep 2018 21:56:21 +0000 (21:56 +0000)]
Merge "sanitizer: fix unaligned load/stores"

6 years agosanitizer: fix unaligned load/stores
Matthias Räncker [Thu, 20 Sep 2018 18:20:39 +0000 (20:20 +0200)]
sanitizer: fix unaligned load/stores

When built with -fsanitizer=address,undefined a number of tests,
such as ByteAlignmentTest.SwitchByteAlignment or
ByteAlignmentTest.SwitchByteAlignment produce runtime errors about
unaligned 4-byte loads/stores. While normally not really a problem,
this does technically violate the language and it is eays to fix in
a standard conforming way using memcpy which does not produce
inferior code.

Signed-off-by: Matthias Räncker <theonetruecamper@gmx.de>
Change-Id: Ie1e97ab25fe874f864df48b473569f00563181ae

6 years agothird_party/googletest: update to v1.8.1
Johann [Mon, 17 Sep 2018 20:57:30 +0000 (13:57 -0700)]
third_party/googletest: update to v1.8.1

BUG=webm:1559

Change-Id: I7a0b16c7bf3f97db2d8650a190b93aae7e12a948

6 years agoMerge "fix UB when initializing parameterized tests"
Johann Koenig [Thu, 20 Sep 2018 19:52:33 +0000 (19:52 +0000)]
Merge "fix UB  when initializing parameterized tests"

6 years agoMerge changes Ibbe7a1c1,I4333a207
Angie Chiang [Thu, 20 Sep 2018 18:15:48 +0000 (18:15 +0000)]
Merge changes Ibbe7a1c1,I4333a207

* changes:
  Add feature score for each block
  Correct mv rows/cols bug in read_frame_dpl_stats

6 years agoGeneralize encoder comp_var_ref setting
Jingning Han [Thu, 20 Sep 2018 17:03:27 +0000 (10:03 -0700)]
Generalize encoder comp_var_ref setting

Generalize the encoder comp_fixed_ref and comp_var_ref assignments.
Make it fully support 2 fwd + 1 bwd and 1 fwd + 2 bwd settings
that VP9 decoder allows.

Change-Id: Id74da9a66327189a3fdf382d447243003c431131

6 years agoSkip checking compound modes with same sign bias for sub8x8
Jingning Han [Tue, 18 Sep 2018 22:56:03 +0000 (15:56 -0700)]
Skip checking compound modes with same sign bias for sub8x8

Drop the check of compound modes where the two reference frames
share the same reference frame sign bias in sub8x8 coding blocks.

Change-Id: I47b45256582b2b5ea1372c9130d8f28cd226a29c

6 years agoUpdate the comp_refs counts
Jingning Han [Tue, 18 Sep 2018 22:26:26 +0000 (15:26 -0700)]
Update the comp_refs counts

Generalize the comp_refs counts update support the case where one
has 1 fwd and 2 bwd reference frames too.

Change-Id: I979216a95d45efef51026158f94612bef39d3c6d

6 years agofix UB when initializing parameterized tests
Matthias Räncker [Wed, 19 Sep 2018 07:47:27 +0000 (09:47 +0200)]
fix UB  when initializing parameterized tests

When running tests built with
-fsanitize=undefined and--disable-optimizations
the sanitizer will emit errors of the following general form:

runtime error: member call on address 0xxxxxxxxx which does not
point to an object of type 'WithParamInterface'
0xxxxxxxxx: note: object has invalid vptr
 00 00 00 00  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  00 ...
              ^~~~~~~~~~~~~~~~~~~~~~~
              invalid vptr

This can be traced to calls to WithParamInterface<T>::GetParam before
the object argument has been initialized. Although GetParam only
accesses static data it is a non-static member function. This causes
that call to have undefined behaviour.
The patch makes GetParam a static member function.

upstream pull request:
https://github.com/google/googletest/pull/1830

The alternative - if the pull request is denied - would be to
modify all parameterized tests to have them derive from
::libvpx_test::CodecTestWith*Params as the first base class.

Signed-off-by: Matthias Räncker <theonetruecamper@gmx.de>
Change-Id: I8e91a4fba5438c9b3e93fa398f789115ab86b521

6 years agoAdd feature score for each block
Angie Chiang [Wed, 19 Sep 2018 23:44:40 +0000 (16:44 -0700)]
Add feature score for each block

The feature score is used to indicate whether a block's mv is reliable
or not.
Now we use Harris Corner Detector method to compute the score.

Change-Id: Ibbe7a1c1f3391d0bf4b03307eaabb5cc3cfb1360

6 years agoCorrect mv rows/cols bug in read_frame_dpl_stats
Angie Chiang [Wed, 19 Sep 2018 22:37:49 +0000 (15:37 -0700)]
Correct mv rows/cols bug in read_frame_dpl_stats

When the frame size is not multiples of mv search bsize,
the fractional part will increment the mv rows/cols by 1

Change-Id: I4333a207406610c540059a9356a82084832ca85b

6 years agoSkip RD check for compound modes that have same sign bias
Jingning Han [Tue, 18 Sep 2018 22:08:53 +0000 (15:08 -0700)]
Skip RD check for compound modes that have same sign bias

The compound mode can only be run between two reference frames
with different sign bias flags. Skip the search over same sign
bias reference frames in the rate-distortion optimization.

Change-Id: I4a57feedea880883cf87200de51862beac108310

6 years agoSync ref frame writing with decoder
Jingning Han [Tue, 18 Sep 2018 19:13:18 +0000 (12:13 -0700)]
Sync ref frame writing with decoder

Enable the encoder to produce compound reference frame writing
that supports both 2 fwd + 1 bwd and 1 fwd + 2 bwd cases.

Change-Id: I63d2141435e2de7d8115d52b974fc41c2e608405

6 years agoAdd frame_start/end to gf_group
Jingning Han [Tue, 18 Sep 2018 15:41:37 +0000 (08:41 -0700)]
Add frame_start/end to gf_group

Keep the start and end frame index for each group of pictures.

Change-Id: I23c0d22e643218cf7486b238c2986101282d3fbe

6 years agoMerge "Fix buffer overrun of postproc_state.limits"
Johann Koenig [Tue, 18 Sep 2018 18:59:59 +0000 (18:59 +0000)]
Merge "Fix buffer overrun of postproc_state.limits"

6 years agoMerge "Fix stack corruption with x86 and --enable-pic"
Johann Koenig [Tue, 18 Sep 2018 18:59:49 +0000 (18:59 +0000)]
Merge "Fix stack corruption with x86 and --enable-pic"

6 years agoMerge "Rename set_arf_sign_bias() to set_ref_sign_bias"
Jingning Han [Tue, 18 Sep 2018 18:45:13 +0000 (18:45 +0000)]
Merge "Rename set_arf_sign_bias() to set_ref_sign_bias"

6 years agoMerge "Re-work set_arf_sign_bias()"
Jingning Han [Tue, 18 Sep 2018 18:45:07 +0000 (18:45 +0000)]
Merge "Re-work set_arf_sign_bias()"

6 years agoMerge "Update frame index per buffer at encoder"
Jingning Han [Tue, 18 Sep 2018 18:44:58 +0000 (18:44 +0000)]
Merge "Update frame index per buffer at encoder"

6 years agoMerge "Remove unnecessary code"
Hui Su [Tue, 18 Sep 2018 18:36:59 +0000 (18:36 +0000)]
Merge "Remove unnecessary code"

6 years agoMerge "Remove the SECOND_LEVEL_CHECKS_BEST macro"
Hui Su [Tue, 18 Sep 2018 18:36:48 +0000 (18:36 +0000)]
Merge "Remove the SECOND_LEVEL_CHECKS_BEST macro"

6 years agoFix stack corruption with x86 and --enable-pic
Matthias Räncker [Mon, 17 Sep 2018 11:57:17 +0000 (13:57 +0200)]
Fix stack corruption with x86 and --enable-pic

x86inc.asm's cglobal macro is frequently used to declare more
arguments than the function actually has. Normally, this is
done to aquire an alias to a register that would correspond to
that positional function argument if it existed. This is safe
when used in this manner.
In the case fixed here, however, the alias is used to temporarily
store adresses obtained through the GOT in memory. Because those
extra arguments don't actually exist, those stores corrupt the
callers stack frame.
SSE2/VpxHBDSubpelVarianceTest.Ref is a test that may fail as a
result.
To simply fix the space allocated to actual arguments that have
been loaded into registers already is reused.
This avoids having to allocate extra space for local variables.

Also removed duplicate code while at it.

Signed-off-by: Matthias Räncker <theonetruecamper@gmx.de>
Change-Id: I505281ecaa6be586185fe6a2d34d62bdf40c839f

6 years agoRemove unnecessary code
Hui Su [Tue, 18 Sep 2018 16:53:23 +0000 (09:53 -0700)]
Remove unnecessary code

in vp9_find_best_sub_pixel_tree().

Change-Id: I0677c05b3e402fc17dd1e7e6fae787d305e90f89

6 years agoRemove the SECOND_LEVEL_CHECKS_BEST macro
Hui Su [Tue, 18 Sep 2018 16:41:14 +0000 (09:41 -0700)]
Remove the SECOND_LEVEL_CHECKS_BEST macro

This macro is used only once and makes the code relatively harder to
read and modify.

Change-Id: I8f0344a7050758ed9770ffca211b0237fe7d8b34

6 years agoRename set_arf_sign_bias() to set_ref_sign_bias
Jingning Han [Thu, 13 Sep 2018 19:24:32 +0000 (12:24 -0700)]
Rename set_arf_sign_bias() to set_ref_sign_bias

Properly reflect its functionality that assigns the reference
frame sign bias to all the reference frames.

Change-Id: I7b597feeb06acd4c3a004cd51e4b285357315360

6 years agoRe-work set_arf_sign_bias()
Jingning Han [Mon, 17 Sep 2018 21:49:13 +0000 (14:49 -0700)]
Re-work set_arf_sign_bias()

Make it support automatic checking and assigning the reference
frame sign bias for all the reference frames.

Change-Id: Ie82f8f872e742130a652b6d5bc109039ac46ae3b

6 years agoUpdate frame index per buffer at encoder
Jingning Han [Mon, 17 Sep 2018 21:17:36 +0000 (14:17 -0700)]
Update frame index per buffer at encoder

Update the frame index counting from key frame offset for all
the processed frames at the encoder. This would allow encoder to
automatically decide frame sign bias next.

Change-Id: Ibbdc2a29b7245be27422272e1fb539596eed63d1

6 years agoMerge "Add a frame_index entry to RefCntBuffer"
Jingning Han [Tue, 18 Sep 2018 03:53:16 +0000 (03:53 +0000)]
Merge "Add a frame_index entry to RefCntBuffer"

6 years agoMerge "Assign GOP frame offset to all the coding frames"
Jingning Han [Tue, 18 Sep 2018 03:53:09 +0000 (03:53 +0000)]
Merge "Assign GOP frame offset to all the coding frames"

6 years agoMerge "cosmetics: normalize include guards"
James Zern [Tue, 18 Sep 2018 03:38:15 +0000 (03:38 +0000)]
Merge "cosmetics: normalize include guards"

6 years agovpx_mem: allow VPX_MAX_ALLOCABLE_MEMORY to be overridden
James Zern [Mon, 17 Sep 2018 19:32:58 +0000 (12:32 -0700)]
vpx_mem: allow VPX_MAX_ALLOCABLE_MEMORY to be overridden

this allows the define to be set by the build environment

Change-Id: Ib40111c5d9bae417b031b8b40a7bc135c6734044

6 years agoAdd a frame_index entry to RefCntBuffer
Jingning Han [Mon, 17 Sep 2018 18:46:17 +0000 (11:46 -0700)]
Add a frame_index entry to RefCntBuffer

This entry will only be effectively used at the encoder side.
Adding it to the RefCntBuffer data structure would help make the
associated logic a lot simpler. Its effect on the decoder side
would be explicitly sent through the bit-stream.

Change-Id: I1660dce9e0bb6e28c3315d5e0df6dc4a9298f71f

6 years agoAssign GOP frame offset to all the coding frames
Jingning Han [Mon, 17 Sep 2018 16:30:37 +0000 (09:30 -0700)]
Assign GOP frame offset to all the coding frames

Overload the use of arf_src_offset to account the relative frame
offset for all the coding frames within a GOP.

Change-Id: Ia86dede37c6a93d9f23098c15dbd936acefd75dc

6 years agoRemove multi_arf_last_grp_enabled flag.
Paul Wilkins [Mon, 17 Sep 2018 14:45:57 +0000 (15:45 +0100)]
Remove multi_arf_last_grp_enabled flag.

Delete flag and associated code.

Change-Id: I899d258a4cd7b84de9136ccfa27cf8a50108b130

6 years agoRemove multi_arf_enabled.
Paul Wilkins [Mon, 17 Sep 2018 14:41:13 +0000 (15:41 +0100)]
Remove multi_arf_enabled.

Remove deprecated multi_arf_enabled flag and associated code.

Change-Id: I73f06362a10faa5b3bd91a78eedb201a96434f18

6 years agoRemove multi_arf_allowed variable.
Paul Wilkins [Mon, 17 Sep 2018 14:28:23 +0000 (15:28 +0100)]
Remove multi_arf_allowed variable.

Removes deprecated multi_arf_allowed variable and dependent code.

Change-Id: Ic1cf341f807c38207e728c48a4c4442387db93ff

6 years agocosmetics: normalize include guards
James Zern [Wed, 22 Aug 2018 21:03:54 +0000 (14:03 -0700)]
cosmetics: normalize include guards

use the recommended format [1] of:
<PROJECT>_<PATH>_<FILE>_H_

[1] https://google.github.io/styleguide/cppguide.html#The__define_Guard
"All header files should have #define guards to prevent multiple
inclusion. The format of the symbol name should be
<PROJECT>_<PATH>_<FILE>_H_."

Change-Id: I2e8ab0b32fb23c30fa43cff5fec12d043c0d2037