platform/upstream/libvpx.git
5 years agoRemove unused variables from VP9_COMP
Jingning Han [Thu, 13 Sep 2018 05:49:07 +0000 (22:49 -0700)]
Remove unused variables from VP9_COMP

Change-Id: I853e0925d29becb9c1f84e5c00d84649fb070a07

5 years agoMerge changes I7173f2fe,I460b6c4b,I5070657f,I2b3e1e16
Jingning Han [Wed, 12 Sep 2018 21:06:53 +0000 (21:06 +0000)]
Merge changes I7173f2fe,I460b6c4b,I5070657f,I2b3e1e16

* changes:
  Remove some deprecated FRAME_UPDATE_TYPE elements.
  Remove some deprecated constants.
  Remove unused rate control data elements
  Remove extra_arf_allowed.

5 years agoRemove some deprecated FRAME_UPDATE_TYPE elements.
Paul Wilkins [Wed, 12 Sep 2018 11:53:10 +0000 (12:53 +0100)]
Remove some deprecated FRAME_UPDATE_TYPE elements.

Removal of some frame types relating to deprecated multi-arf work.

Added a dummy value for the USE_BUF_FRAME frame type in the
declaration of the rd_frame_type_factor[FRAME_UPDATE_TYPES] structure.

Change-Id: I7173f2fe33a53117e1bde6f9621efc1a5951240b

5 years agoMerge "Fix rate control bug with recode all."
Paul Wilkins [Wed, 12 Sep 2018 11:57:02 +0000 (11:57 +0000)]
Merge "Fix rate control bug with recode all."

5 years agoRemove some deprecated constants.
Paul Wilkins [Wed, 12 Sep 2018 11:37:05 +0000 (12:37 +0100)]
Remove some deprecated constants.

Removal of some # defines relating to deprecated multi-arf work.

Change-Id: I460b6c4bee9bf0ef588eddc47329c2b17f60e5ba

5 years agoRemove unused rate control data elements
Paul Wilkins [Wed, 12 Sep 2018 11:27:04 +0000 (12:27 +0100)]
Remove unused rate control data elements

Removal of rate control structure elements related to Zoe's
deprecated multi laryer ARF work.

Change-Id: I5070657f91df7bd3f9137cf74016f737313417c8

5 years agoRemove extra_arf_allowed.
Paul Wilkins [Wed, 12 Sep 2018 11:06:36 +0000 (12:06 +0100)]
Remove extra_arf_allowed.

Removed VP9_COMP element that is no longer used.

Change-Id: I2b3e1e16244074e3510c1467b0e7532213c4ae05

5 years agoMerge "Enable rectangular partition search for speed 1"
Paul Wilkins [Wed, 12 Sep 2018 11:03:50 +0000 (11:03 +0000)]
Merge "Enable rectangular partition search for speed 1"

5 years agoMerge changes I306c582c,Id5389285
Paul Wilkins [Wed, 12 Sep 2018 11:01:38 +0000 (11:01 +0000)]
Merge changes I306c582c,Id5389285

* changes:
  Remove configure_multi_arf_buffer_updates()
  Remove update_multi_arf_ref_frames()

5 years agoMerge changes I97111413,Idcd6dcbc
Paul Wilkins [Wed, 12 Sep 2018 11:01:20 +0000 (11:01 +0000)]
Merge changes I97111413,Idcd6dcbc

* changes:
  Clean up define_gf_group()
  Clean up deprecated gop structure code

5 years agoMerge "Set GF frame layer depth to be 0"
Paul Wilkins [Wed, 12 Sep 2018 11:00:49 +0000 (11:00 +0000)]
Merge "Set GF frame layer depth to be 0"

5 years agoRemove configure_multi_arf_buffer_updates()
Jingning Han [Tue, 11 Sep 2018 21:35:42 +0000 (14:35 -0700)]
Remove configure_multi_arf_buffer_updates()

The bit-stream syntax doesn't support lst2/3/bwd reference frame
update. Remove the deprecated function that goes such assumption.

Change-Id: I306c582c2efc63928e4231adef2ee549076a987c

5 years agoRemove update_multi_arf_ref_frames()
Jingning Han [Tue, 11 Sep 2018 21:14:03 +0000 (14:14 -0700)]
Remove update_multi_arf_ref_frames()

The bit-stream syntax doesn't support the use of lst2/3 frames.
Remove the update_multi_arf_ref_frames() function that assumes
such functionality.

Change-Id: Id5389285c84fe6c578c52d210aa47ef3cb789f8e

5 years agoMerge "Simplify vp9_frame_type_qdelta()"
Jingning Han [Tue, 11 Sep 2018 23:32:34 +0000 (23:32 +0000)]
Merge "Simplify vp9_frame_type_qdelta()"

5 years agoMerge "Store mv/inter_cost/recon_error/err for ref frames"
Angie Chiang [Tue, 11 Sep 2018 22:52:18 +0000 (22:52 +0000)]
Merge "Store mv/inter_cost/recon_error/err for ref frames"

5 years agoClean up define_gf_group()
Jingning Han [Tue, 11 Sep 2018 21:12:29 +0000 (14:12 -0700)]
Clean up define_gf_group()

Remove deprecated extra_arf_allowed code.

Change-Id: I97111413e6465475e750106fddef8f344db53405

5 years agoClean up deprecated gop structure code
Jingning Han [Tue, 11 Sep 2018 21:05:31 +0000 (14:05 -0700)]
Clean up deprecated gop structure code

Gradually integrate the single-/multi-layer ARF and dual ARF
encoder control. Remove deprecated code.

Change-Id: Idcd6dcbca3f8d7597878d83dec421e16be819f55

5 years agoSimplify vp9_frame_type_qdelta()
Jingning Han [Tue, 11 Sep 2018 17:50:49 +0000 (10:50 -0700)]
Simplify vp9_frame_type_qdelta()

Make direct use of frame type in the available VP9_COMMON structure.
Eliminate the need to map through rf_level to fetch the frame type.
This change doesn't alter the coding stats. It simplifies the
vp9_frame_type_qdelta() function logic and removes unnecessary
reference to rf_level.

Change-Id: I1a7b2f5abcae39aa4a60d08a6011dde38ecf3b58

5 years agoSet GF frame layer depth to be 0
Jingning Han [Tue, 11 Sep 2018 16:19:37 +0000 (09:19 -0700)]
Set GF frame layer depth to be 0

Set the golden frame layer depth as 0 - the base layer in temporal
domain.

Change-Id: If63e1524a567fcff6162f4283811298551516be5

5 years agoRemove unused constant definition
Jingning Han [Tue, 11 Sep 2018 16:27:33 +0000 (09:27 -0700)]
Remove unused constant definition

ARF_DECAY_BREAKOUT is no long used.

Change-Id: I553f8a3087389f0343444e2551581e9de02d3427

5 years agoLocalize variable definitions
Jingning Han [Tue, 11 Sep 2018 16:21:45 +0000 (09:21 -0700)]
Localize variable definitions

Localize variable definitions in setup_frames() and
two_pass_first_group_inter().

Change-Id: I66e842791d679be6d22cef50e0b395b5aa380eac

5 years agoMerge "Rework two_pass_first_group_inter()"
Jingning Han [Tue, 11 Sep 2018 04:04:31 +0000 (04:04 +0000)]
Merge "Rework two_pass_first_group_inter()"

5 years agoMerge "Separate frame context index for GOP layers"
Jingning Han [Tue, 11 Sep 2018 04:04:14 +0000 (04:04 +0000)]
Merge "Separate frame context index for GOP layers"

5 years agoMerge "Assign layer depth for all coding frames"
Jingning Han [Tue, 11 Sep 2018 04:03:52 +0000 (04:03 +0000)]
Merge "Assign layer depth for all coding frames"

5 years agoMerge "Refactor block_rd_txfm()"
Hui Su [Tue, 11 Sep 2018 02:18:18 +0000 (02:18 +0000)]
Merge "Refactor block_rd_txfm()"

5 years agoRework two_pass_first_group_inter()
Jingning Han [Mon, 10 Sep 2018 18:55:10 +0000 (11:55 -0700)]
Rework two_pass_first_group_inter()

This function is used to in part decide if to trigger recode loop
for the first normal P frame in a GOP. Rework its design logic to
support the GOP with multi-layer ARF. Allow recode when there is
a transition from ARF/OVERLAY/USE_BUF to normal P frame.

The overall coding performance for multi-ARF gets slightly better
(less than 0.1% for show_existing_frame case). Tested on a few
clips, the encoding speed remains similar too. This change primarily
serves to help integration of multi-layer ARF and dual-ARF systems.

Change-Id: Ia44e44526b05029b1546985b3eb649e767d5444f

5 years agoRefactor block_rd_txfm()
Hui Su [Mon, 10 Sep 2018 18:53:04 +0000 (11:53 -0700)]
Refactor block_rd_txfm()

Merge two identical if branches.

Change-Id: Ie012ba9c116a30ef6fa2e7868c7a4ba886b99bc6

5 years agoStore mv/inter_cost/recon_error/err for ref frames
Angie Chiang [Wed, 5 Sep 2018 18:56:42 +0000 (11:56 -0700)]
Store mv/inter_cost/recon_error/err for ref frames

These information will help with making better mv search decision

Add functionality to dump tpl_stats for offline analysis

Change-Id: Ic2ec34368499c9bccb4d1f21a12b66453847fcf2

5 years agoSeparate frame context index for GOP layers
Jingning Han [Fri, 7 Sep 2018 23:37:42 +0000 (16:37 -0700)]
Separate frame context index for GOP layers

Use separate frame context index to code frames at different layers.
The maximum index cap is set as 3. This improves the compression
performance of multi-layer ARF by 0.15% across the test sets.

The overall coding gains from multi-layer ARF are

         avg PSNR       ovarall PSNR        SSIM
lowres   -3.9%            -3.7%            -3.2%
midres   -3.5%            -3.2%            -2.3%
nflx2k   -4.3%            -4.6%            -3.0%

Change-Id: I8a0b345fdd47823c018544a6b4748753faf89dc1

5 years agofix vp9_svc_adjust_frame_rate signature
James Zern [Sat, 8 Sep 2018 20:53:46 +0000 (13:53 -0700)]
fix vp9_svc_adjust_frame_rate signature

match the const in the header; quiets a visual studio warning.

since:
04b3d49ba vp9-svc: Allow for setting framerate per spatial layer.

Change-Id: I0a216eb8fe1a689fe6822bbfac70f7c98e9b1a70

5 years agoFix rate control bug with recode all.
Paul Wilkins [Thu, 6 Sep 2018 14:56:01 +0000 (15:56 +0100)]
Fix rate control bug with recode all.

This patch fixes a rate control bug that can manifest if the recode
loop is activated for all frame types. Specifically things go wrong when the
recode loop is used on an overlay frame that has a rate target of 0 bits.

The patch prevents adjustment of the active worst quality and repeat recode
loops for overlay frames.

The bug showed up during artificial experiments on re-distribution of bits in
ARF groups but does not activate in any current encode profile, as even best
best quality does not currently allow recodes for all frames.

Change-Id: I80872093d9ebd3350106230c42c3928e56ecb754

5 years agoMerge "Fork auto-alt-ref control"
Jingning Han [Sat, 8 Sep 2018 02:26:52 +0000 (02:26 +0000)]
Merge "Fork auto-alt-ref control"

5 years agoMerge "Extend auto-alt-ref parameter range"
Jingning Han [Sat, 8 Sep 2018 02:26:47 +0000 (02:26 +0000)]
Merge "Extend auto-alt-ref parameter range"

5 years agoMerge "Add non-greedy-mv experimental flag"
Angie Chiang [Sat, 8 Sep 2018 01:02:21 +0000 (01:02 +0000)]
Merge "Add non-greedy-mv experimental flag"

5 years agoAssign layer depth for all coding frames
Jingning Han [Fri, 7 Sep 2018 22:59:14 +0000 (15:59 -0700)]
Assign layer depth for all coding frames

Assign layer depth for the base layer ARF and the normal frames.

Change-Id: I81cbb2846c3176336622f9006701c0219652905a

5 years agoMerge "Add NORMAL_BOOST macro"
Jingning Han [Fri, 7 Sep 2018 22:27:23 +0000 (22:27 +0000)]
Merge "Add NORMAL_BOOST macro"

5 years agoEnable rectangular partition search for speed 1
Hui Su [Tue, 4 Sep 2018 21:44:02 +0000 (14:44 -0700)]
Enable rectangular partition search for speed 1

This patch enables rectangular partition search on speed 1. The encoding
speed loss is reduced thanks to recently added speed features.

This only affects speed 1 low bit-depth encoding.

Coding gains:
           avg_psnr     ovr_psnr    ssim
lowres      0.577%       0.621%    0.665%
midres      1.147%       1.215%    1.148%
hdres       0.758%       0.790%    0.769%

Tested encoding speed on 15 midres and 15 hdres clips, average speed
loss:
           QP=30       QP=40        QP=50
midres     4.43%       3.72%       -1.05%
hdres      4.41%       5.65%        3.77%

Change-Id: Ifc0712becccc69f7498796359ff12dbfa63fd7b3

5 years agoFork auto-alt-ref control
Jingning Han [Fri, 7 Sep 2018 17:25:07 +0000 (10:25 -0700)]
Fork auto-alt-ref control

Temporarily fork the auto-alt-ref control meaning. When it is set
to be 1, use single layer ARF as baseline. The value 2 would enable
dual ARF system. Any number above it would trigger automatic multi-
layer ARFs.

We would gradually refactor and integrate dual ARF and multi-layer
ARF systems next, and eventually make auto-alt-ref directly control
the layer depth.

Change-Id: I292d27111ae8a596b97444afecf4b896043e543f

5 years agoExtend auto-alt-ref parameter range
Jingning Han [Fri, 7 Sep 2018 17:20:56 +0000 (10:20 -0700)]
Extend auto-alt-ref parameter range

Extend the upper limit from 2 (dual ARFs) to maximum ARF layers.
This would later allow --auto-alt-ref to directly control the
ARF layer depth later on.

Change-Id: I6324fe980122e73dc98f81c8d7de1193a1a16e51

5 years agoAdd NORMAL_BOOST macro
Jingning Han [Thu, 6 Sep 2018 23:58:55 +0000 (16:58 -0700)]
Add NORMAL_BOOST macro

Normal frame boost factor is set to be 100 as the baseline for
ARF boost. Replace the hard coded number with a macro.

Change-Id: I81ce30138f7819844e7a2d811de9e1ccbeb85da5

5 years agoMerge "vp9-svc: Allow for setting framerate per spatial layer."
Marco Paniconi [Thu, 6 Sep 2018 16:17:50 +0000 (16:17 +0000)]
Merge "vp9-svc: Allow for setting framerate per spatial layer."

5 years agoMerge "Fix short first kf bug."
Paul Wilkins [Thu, 6 Sep 2018 14:24:05 +0000 (14:24 +0000)]
Merge "Fix short first kf bug."

5 years agoMerge "Revert "Revert "Prevent double application of min rate in two pass."""
Paul Wilkins [Thu, 6 Sep 2018 14:23:54 +0000 (14:23 +0000)]
Merge "Revert "Revert "Prevent double application of min rate in two pass."""

5 years agoMerge "Adaptive ARF factor decision"
Jingning Han [Thu, 6 Sep 2018 03:45:22 +0000 (03:45 +0000)]
Merge "Adaptive ARF factor decision"

5 years agoMerge "Recursive rate allocation for multi-layer ARF coding"
Jingning Han [Thu, 6 Sep 2018 03:45:13 +0000 (03:45 +0000)]
Merge "Recursive rate allocation for multi-layer ARF coding"

5 years agoMerge "Enable adaptive rate allocation for multi-layer ARFs"
Jingning Han [Thu, 6 Sep 2018 03:45:08 +0000 (03:45 +0000)]
Merge "Enable adaptive rate allocation for multi-layer ARFs"

5 years agoMerge "Initialize the best partition before partition RDO"
Hui Su [Wed, 5 Sep 2018 22:02:22 +0000 (22:02 +0000)]
Merge "Initialize the best partition before partition RDO"

5 years agoInitialize the best partition before partition RDO
Hui Su [Wed, 5 Sep 2018 20:00:56 +0000 (13:00 -0700)]
Initialize the best partition before partition RDO

This fixes the multi-thread encoder test failure.

Change-Id: I0c1845922068e71097a387db0969ca419accb3ed

5 years agoAdd non-greedy-mv experimental flag
Angie Chiang [Wed, 5 Sep 2018 18:46:45 +0000 (11:46 -0700)]
Add non-greedy-mv experimental flag

The experiment aims at making non-greedy mv search decision

Change-Id: I3d77048ce106771fe003f250d07b7ddf0112536f

5 years agovp9-svc: Allow for setting framerate per spatial layer.
Marco Paniconi [Wed, 29 Aug 2018 22:05:46 +0000 (15:05 -0700)]
vp9-svc: Allow for setting framerate per spatial layer.

Add duration to set_svc_ref_frame_config.

BUG=b/113346831

Change-Id: I63613aed6b1183f98d04831600a6bdd645c740df

5 years agoAdaptive ARF factor decision
Jingning Han [Fri, 31 Aug 2018 22:54:10 +0000 (15:54 -0700)]
Adaptive ARF factor decision

Re-count the factors to decide bit boost factor for the
intermediate layer ARFs. Make the gfu_boost factor assigned to
each ARF adapt to its local factors.

This and the recursive change 5bfe9eb together improves the
multi-layer ARF compression performance:

          avg_psnr      ovr_psnr     ssim
lowres    -0.39%        -0.54%       -1.6%
midres    -0.98%        -1.26%       -2.3%
hdres     -0.95%        -1.13%       -2.3%

Change-Id: I5fec3ea75cae58825787dc88dadc7e8697a041ea

5 years agoRecursive rate allocation for multi-layer ARF coding
Jingning Han [Thu, 30 Aug 2018 04:30:35 +0000 (21:30 -0700)]
Recursive rate allocation for multi-layer ARF coding

Recursively calculate the rate boost for the ARF frames at the
given layer depth from the remaining available bit resource after
the prior layer ARFs consumption.

Change-Id: I0e31bac4f87b895ca20605dc1307a8fc0d2a516d

5 years agoEnable adaptive rate allocation for multi-layer ARFs
Jingning Han [Tue, 28 Aug 2018 21:08:02 +0000 (14:08 -0700)]
Enable adaptive rate allocation for multi-layer ARFs

Increase the bit allocation for the intermediate layer ARFs. The
current strategy assigns higher offset to the lower layer ARFs.
The needed budget is borrowed from the base layer ARF allocation.

Change-Id: I16b6e9cce4dab8e73e7b097674d1a8504205026e

5 years agoMerge "Increase encoder buffer for multi-layer ARFs"
Jingning Han [Wed, 5 Sep 2018 14:30:19 +0000 (14:30 +0000)]
Merge "Increase encoder buffer for multi-layer ARFs"

5 years agoMerge "Structure the multi-layer ARF locations"
Jingning Han [Wed, 5 Sep 2018 14:30:13 +0000 (14:30 +0000)]
Merge "Structure the multi-layer ARF locations"

5 years agoMerge "Move partition search ML models to a seperate file"
Hui Su [Wed, 5 Sep 2018 04:31:58 +0000 (04:31 +0000)]
Merge "Move partition search ML models to a seperate file"

5 years agoMerge "Assign target bits for multi-layer ARF system"
Jingning Han [Tue, 4 Sep 2018 18:29:25 +0000 (18:29 +0000)]
Merge "Assign target bits for multi-layer ARF system"

5 years agoMove partition search ML models to a seperate file
Hui Su [Tue, 4 Sep 2018 17:05:15 +0000 (10:05 -0700)]
Move partition search ML models to a seperate file

Clean up vp9_encodeframe.c.

Change-Id: I4035fee94da746c74d72f71ca8334f91c5d10116

5 years agovp9-svc: Fix to first_spatial_to_encode for pattern constraint.
Marco Paniconi [Tue, 4 Sep 2018 16:08:49 +0000 (09:08 -0700)]
vp9-svc: Fix to first_spatial_to_encode for pattern constraint.

Change-Id: I876f69acf9420b3b013cb3048bbfa8ff059e2e50

5 years agoIncrease encoder buffer for multi-layer ARFs
Jingning Han [Tue, 28 Aug 2018 19:43:34 +0000 (12:43 -0700)]
Increase encoder buffer for multi-layer ARFs

When multi-layer ARF mode is enabled, increase the encoder buffer
to account for the situation where several ARFs are coded together
in a frame packet.

Change-Id: I4e53095f6b6ac5a3c8d79414411ac39880bf1523

5 years agoStructure the multi-layer ARF locations
Jingning Han [Tue, 28 Aug 2018 17:52:32 +0000 (10:52 -0700)]
Structure the multi-layer ARF locations

Fine tune the multi-layer ARF location decisions. Support deeper
layer structure.

Change-Id: I3e44cf52b6813f6267bcd7266f9aa1b7ded57f8e

5 years agoMerge "ML based rectangular partition search pruning"
Hui Su [Tue, 4 Sep 2018 16:38:12 +0000 (16:38 +0000)]
Merge "ML based rectangular partition search pruning"

5 years agoFix short first kf bug.
Paul Wilkins [Mon, 3 Sep 2018 15:48:02 +0000 (16:48 +0100)]
Fix short first kf bug.

This change is in response to quality issue in b/112953058

The quality regression observed is a result of a bug that manifested
because of a very short key frame group at the start of a chunk.
The group was so short that it was less than the minimum allowed
length of an ARF group, so the initial group was coded as a GF only
group. However, group length was not set correctly and the result
was a frame coded with a target of 0 bits.

This causes two problems:

Firstly one very poor frame that caused the issue to be raised.

Secondly that one  frame obviously overshoots its 0 target very heavily
and this has the effect moving the needle significantly in terms of the
adaptive rate control (specifically the estimate of bits per macro block
used to estimate the active Q range). Consequently there is undershoot
for most of the rest of the chunk and the overall rate ends up much lower
than the target (14Mb/s vs a target of 22Mb/s). (The sharp drop in the
overall rate is also noted in the issue.

BUG=b/112953058

Change-Id: Ide9cce57acd3dee0f9496b752902e7b4735f2c7f

5 years agoAssign target bits for multi-layer ARF system
Jingning Han [Mon, 27 Aug 2018 23:17:29 +0000 (16:17 -0700)]
Assign target bits for multi-layer ARF system

Keep the ARF and P frame rate allocation distribution. All the
intermediate ARFs are treated same as regular P frames.

Change-Id: I7807b8e6a8f19b6e1b09b9b7d119b3c88ef90b67

5 years agoMerge "Properly update the raw_src_frame for psnr calculation"
Jingning Han [Tue, 4 Sep 2018 14:10:34 +0000 (14:10 +0000)]
Merge "Properly update the raw_src_frame for psnr calculation"

5 years agoMerge "Build arf index stack"
Jingning Han [Tue, 4 Sep 2018 14:08:45 +0000 (14:08 +0000)]
Merge "Build arf index stack"

5 years agoMerge "vp9-svc: Add bypass flag to constrain inter_layer."
Marco Paniconi [Tue, 4 Sep 2018 04:07:53 +0000 (04:07 +0000)]
Merge "vp9-svc: Add bypass flag to constrain inter_layer."

5 years agovp9-svc: Add bypass flag to constrain inter_layer.
Marco Paniconi [Mon, 3 Sep 2018 17:15:38 +0000 (10:15 -0700)]
vp9-svc: Add bypass flag to constrain inter_layer.

The additional constraint imposed on inter-layer
prediction should only be done for non-bypass (fixed)
svc mode.

Change-Id: Ia22cdb7bc21684776c9a13397e177a1e1c3d55a2

5 years agoRevert "Revert "Prevent double application of min rate in two pass.""
Paul Wilkins [Mon, 3 Sep 2018 15:12:22 +0000 (16:12 +0100)]
Revert "Revert "Prevent double application of min rate in two pass.""

This rate control bug in the original patch is not the underlying cause
of the quality regression but simply unmasked a problem which stems
from applying 0 bits to the last frame in a short KF group at the start
of a chunk.

This reverts commit d10b1f233657054b1465d40f0b409c20029a540b.

Change-Id: I32c91a24a14d013853bb8e5587aa69600e6a0063

5 years agovp9-svc: Fix condition for pattern constraints
Marco Paniconi [Mon, 3 Sep 2018 05:17:32 +0000 (22:17 -0700)]
vp9-svc: Fix condition for pattern constraints

For fixed/non-flexible SVC mode: on non-key spatial
enhancement layers modify constraint on the inter-layer
prediction to include the first_spatial_layer_to_encode.

Change-Id: I6a59174976ad72d555653704dcd3b03c52e31b6f

5 years agoProperly update the raw_src_frame for psnr calculation
Jingning Han [Mon, 27 Aug 2018 22:58:41 +0000 (15:58 -0700)]
Properly update the raw_src_frame for psnr calculation

Update the raw_src_frame to be the current input source frame in
the show_existing_frame mode.

Change-Id: Ia8edf49ca948c45ffe6c60556756b36124ab092a

5 years agoBuild arf index stack
Jingning Han [Sun, 26 Aug 2018 04:39:59 +0000 (21:39 -0700)]
Build arf index stack

Stack the ARF frame indexes. Use the most recent one as the ARF
reference frame for frame coding.

Change-Id: I88a2202fa5deb2587d861b434d27ab8de0642cf7

5 years agovp9-svc: Add first_spatial_layer_to_encode per superframe
Marco Paniconi [Fri, 31 Aug 2018 22:42:19 +0000 (15:42 -0700)]
vp9-svc: Add first_spatial_layer_to_encode per superframe

VP9E_SET_SVC_LAYER_ID sets the first spatial layer to
encoder per superframe, so add this parameter to svc encoder.
This is needed, for example, to properly set is_key_frame for
spatial layers when base spatial layer is skipped encoded.

Change-Id: Ifd4ac77f539197ec021e62f4c624a6cc79d64f43

5 years agoML based rectangular partition search pruning
Hui Su [Tue, 17 Jul 2018 05:05:19 +0000 (22:05 -0700)]
ML based rectangular partition search pruning

Add a ML model to predict if rectangular partition search can be skipped
without much coding loss. This model is enabled for speed 0 low bitdepth
only.

Impact on coding performance is minor:
             avg_psnr     ovr_psnr     ssim
lowres       -0.005%       0.005%     0.017%
midres        0.100%       0.114%     0.134%
hdres         0.048%       0.083%     0.074%
jvet480p      0.035%       0.027%     0.044%
jvet720p      0.094%       0.090%     0.174%

Tested encoding speed over 20 midres and hdres clips, average speed
gain is about 8%; maximum speed gain is 23%.

Change-Id: I5d4029dec7134c53ac68ab6cf0c8077dc0b767ed

5 years agoMerge "Fix arf_src_offset calculation"
Jingning Han [Fri, 31 Aug 2018 18:19:47 +0000 (18:19 +0000)]
Merge "Fix arf_src_offset calculation"

5 years agoMerge "Set minimum frame size to be 1 byte"
Jingning Han [Fri, 31 Aug 2018 18:19:42 +0000 (18:19 +0000)]
Merge "Set minimum frame size to be 1 byte"

5 years agoMerge "Prepare multi-layer ARF coding structure"
Jingning Han [Fri, 31 Aug 2018 18:19:32 +0000 (18:19 +0000)]
Merge "Prepare multi-layer ARF coding structure"

5 years agoMerge "Build up multi-layer ARF processing order"
Jingning Han [Fri, 31 Aug 2018 18:19:13 +0000 (18:19 +0000)]
Merge "Build up multi-layer ARF processing order"

5 years agoMerge "Add element stack operations for arf index control"
Jingning Han [Fri, 31 Aug 2018 18:19:07 +0000 (18:19 +0000)]
Merge "Add element stack operations for arf index control"

5 years agoFix arf_src_offset calculation
Jingning Han [Fri, 24 Aug 2018 19:54:58 +0000 (12:54 -0700)]
Fix arf_src_offset calculation

The offset should be computed with respect to the current coding
process standing.

Change-Id: I63fc303eb062d5fd68b8d1faa3b4172cdfcce168

5 years agoSet minimum frame size to be 1 byte
Jingning Han [Thu, 23 Aug 2018 23:51:26 +0000 (16:51 -0700)]
Set minimum frame size to be 1 byte

The show_existing_frame mode still needs to be sent to the decoder.
Account for this as 1 byte. This would make the encoder properly
update its state.

Change-Id: I32a59ccb5d0e02cc6367c1a264b2de72dc1432a7

5 years agoPrepare multi-layer ARF coding structure
Jingning Han [Thu, 23 Aug 2018 17:18:53 +0000 (10:18 -0700)]
Prepare multi-layer ARF coding structure

Build the frame processing order and type queue for multi-layer
ARF coding structure.

Change-Id: I5e14c60279020dc65a883d2997ca1ca9ce739488

5 years agoBuild up multi-layer ARF processing order
Jingning Han [Wed, 22 Aug 2018 19:41:25 +0000 (12:41 -0700)]
Build up multi-layer ARF processing order

Use DFS to build the multi-layer ARF processing order.

Change-Id: Iba4b20476eb5c8a3db49a24b2b0dec325fade65b

5 years agoAdd element stack operations for arf index control
Jingning Han [Wed, 22 Aug 2018 19:03:39 +0000 (12:03 -0700)]
Add element stack operations for arf index control

Support arf index stack operation.

Change-Id: Ifcf521ffc95a520344824ffc159883b71e8fc7a0

5 years agoMerge "cosmetics,lf threading: normalize struct member names"
James Zern [Fri, 31 Aug 2018 03:22:42 +0000 (03:22 +0000)]
Merge "cosmetics,lf threading: normalize struct member names"

5 years agosilence c++ abi warning
Johann [Thu, 30 Aug 2018 21:07:02 +0000 (14:07 -0700)]
silence c++ abi warning

Linking c++ libraries built with gcc 6 and gcc 7 on arm
generates some warnings because of incompatibilities between those
compilers:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77728

libvpx does not generate a c++ library. C++ is only used for examples and tests.

Change-Id: I3d5d5ef3fb66743bff26a833d6641898975e9f71

5 years agovp9: Fix rate control stats for bypass mode in sample encoder
Marco Paniconi [Thu, 30 Aug 2018 19:59:44 +0000 (12:59 -0700)]
vp9: Fix rate control stats for bypass mode in sample encoder

Allow rate control stats to work for bypass mode
in vp9_spatial_svc_encoder.c

Change-Id: I66764a006a73b1fd13c07b4fc4e0c88b2bb2a035

5 years agoRevert "Loopfilter MultiThread Optimization"
James Zern [Thu, 30 Aug 2018 16:58:02 +0000 (09:58 -0700)]
Revert "Loopfilter MultiThread Optimization"

This reverts commit dafe064289a917977439ab6f4f002b9946496084.

Corrupted files may cause the decoder to hang as row progress in the
loopfilter is used to progress each thread.

BUG=webm:1558

Change-Id: I0674ce9af14d3fb7b2da8124e7b600616c8e734a

5 years agoMerge "rtcd: fix --required flag"
Johann Koenig [Wed, 29 Aug 2018 20:17:49 +0000 (20:17 +0000)]
Merge "rtcd: fix --required flag"

5 years agortcd: fix --required flag
Johann [Tue, 28 Aug 2018 19:56:34 +0000 (12:56 -0700)]
rtcd: fix --required flag

Always parse --required options. Previously they were only parsed for
x86_64.

Make entries passed in additive if there are existing required flags.

Mark 'neon' as required for armv8/aarch64.

BUG=chromium:876548

Change-Id: I55c6aad4536a9d8423e223e5616f3aa26d6b2941

5 years agoMerge "Skip unnecessary motion search"
Hui Su [Wed, 29 Aug 2018 15:39:31 +0000 (15:39 +0000)]
Merge "Skip unnecessary motion search"

5 years agoSkip unnecessary motion search
Hui Su [Tue, 28 Aug 2018 00:32:05 +0000 (17:32 -0700)]
Skip unnecessary motion search

If a ref frame is masked out, we do not need to do motion search for it.
It makes speed 0 a little faster.

Change-Id: I68f71255b2798b24fd1d5b28ed24a2ef87251413

5 years agoMerge "vp9: Fix ref frame update in denoiser in bypass mode."
Jerome Jiang [Tue, 28 Aug 2018 17:27:57 +0000 (17:27 +0000)]
Merge "vp9: Fix ref frame update in denoiser in bypass mode."

5 years agoMerge "Revert "Prevent double application of min rate in two pass.""
Hui Su [Tue, 28 Aug 2018 17:04:58 +0000 (17:04 +0000)]
Merge "Revert "Prevent double application of min rate in two pass.""

5 years agoRework enc/dec mismatch detection
Jingning Han [Mon, 27 Aug 2018 22:22:24 +0000 (15:22 -0700)]
Rework enc/dec mismatch detection

The previous enc/dec mismatch detection assumes the previously
reconstructed frame would always stay at frame buffer pool index
at 0. It could hence cause certain delay in enc/dec mismatch
detection when the immediate reconstruction frame is not yet
propagated to index 0 in the buffer map pool.

This change always keeps the latest decoded show frame buffer
index and directly gets the reconstructed frame from encoder and
decoder buffer pools to check for mismatch.

Change-Id: If53092cbc42ab78d55af5b83f12a489fc362f3ae

5 years agovp9: Fix ref frame update in denoiser in bypass mode.
Jerome Jiang [Mon, 27 Aug 2018 22:28:48 +0000 (15:28 -0700)]
vp9: Fix ref frame update in denoiser in bypass mode.

BUG=b/112292577
Change-Id: I8fc5711e44d0317e299aa49f781e9c438bba9d82

5 years agovp9-svc: Change default pattern for bypass mode
Marco Paniconi [Mon, 27 Aug 2018 19:03:37 +0000 (12:03 -0700)]
vp9-svc: Change default pattern for bypass mode

For sample encoder: keep default pattern for bypass
mode to example#0.

Change-Id: Icddc4600d750a23a44b26517a327b546fd8eb412

5 years agoMerge "SVC: extend api to specify temporal id for each spatial layers."
Jerome Jiang [Mon, 27 Aug 2018 17:54:52 +0000 (17:54 +0000)]
Merge "SVC: extend api to specify temporal id for each spatial layers."

5 years agoMerge "Rework the ref_frame_skip_mask feature in RDO"
Hui Su [Thu, 23 Aug 2018 22:00:50 +0000 (22:00 +0000)]
Merge "Rework the ref_frame_skip_mask feature in RDO"

5 years agoRevert "Prevent double application of min rate in two pass."
Hui Su [Thu, 23 Aug 2018 21:39:38 +0000 (21:39 +0000)]
Revert "Prevent double application of min rate in two pass."

This reverts commit 416b7051d7f610ed6d62dff18af7776ec520fd9c.

Reason for revert: it causes visual quality drop as described in b/112953058.

Original change's description:
> Prevent double application of min rate in two pass.
>
> The initial allocation of bits in the two pass code to each frame
> should be within the min max limits on the command line. However,
> when forming an ARF group the cost of the ARF is shared by frames
> in that group such that the residual bits for a frame could drop below
> the min value. This change prevents the minimum being re-applied
> after the cost of the ARF has been deducted as this may otherwise
> cause low rate sections to overshoot their target.
>
> Test runs comparing to a baseline run with min and max section pct
> 0-2000% vs one closer to the YT use case (50-150%) suggest that
> this fix not only results in better rate control but also gives a better
> rd outcome.
>
> For example the HD set vs 0-2000% baseline (opsnr, ssim).
> Old code (50-150):  +0.751, +1.099
> New code(50-150): +0.241, -0.009
>
> Change-Id: I715da7b130bf53ba8aa609532aa9e18b84f5e2ef

TBR=yaowu@google.com,paulwilkins@google.com,debargha@google.com,builds@webmproject.org

# Not skipping CQ checks because original CL landed > 1 day ago.

Change-Id: Ic9849e4e0db64e9d92bbb9df9cc923230a15c4df