Julia Robson [Wed, 20 Jan 2016 12:03:16 +0000 (12:03 +0000)]
Making the forward transform consistent with high bit depth
This patch changes the code for 16bit buffers to use the same
optimisation as is used for 8bit buffers. (See change-Id:
I0452da1786d59bc8bcbe0a35fdae9f623d1d44e1 for more information
about the optimisation)
Yue Chen [Thu, 7 Jan 2016 23:13:52 +0000 (15:13 -0800)]
EXT_INTER experiment
NEW2MV is enabled, representing a new motion vector predicted from
NEARMV. It is mostly ported from nextgen, where it was named
NEW_INTER.
A few fixes are done for sub8x8 RDO to correct some misused
mv references in the original patch.
A 'bug-fix' for encoding complexity is done, reducing the additional
encoding time from 50% to 20%. In sub8x8 case, the old patch
did motion search for every interpolation filter (vp9 only
searches once). This fix also slightly improves the coding gain.
This experiment has been made compatible with REF_MV and EXT_REFS.
Coding gain (derflr/hevcmr/hevchd): 0.267%/0.542%/0.257%
ext-partition: to hold partition extensions (ex. ext-partition,
ext-coding-unit-size from nextgen)
loop-restore: to hold in-loop restoration filter (ex. loop-postfilter
from nextgen and other Wiener restoration filters)
Jingning Han [Wed, 13 Jan 2016 17:03:15 +0000 (09:03 -0800)]
Handle single ref mv pair in the candidate list for compound mode
This commit considers the case where a single reference motion
vector pair is found in the candidate list. It treats this pair
as the effective motion vector for nearestmv mode. This improves
the coding performance by 0.06% for stdhd sets.
Different quality levels are used for different regions in
the frame depending on how far they are vertically from the
center. Specifically, three segments are used based on the
mi_row index with respect number to the number of mi_rows in
the frame.
Yaowu Xu [Tue, 12 Jan 2016 17:47:07 +0000 (09:47 -0800)]
Fix encoder crashes and enc/dec mismatches
This commit adds the logic for segmentation map initialization and
disable temporal update of segmentation map when error-resilient
mode is on. It fixes the enc/dec mistmates (release build) and
assertions(debug) when both aq-mode and error-resilient are on.
Jingning Han [Tue, 12 Jan 2016 23:06:59 +0000 (15:06 -0800)]
Generate compound reference motion vector
This commit allows the codec to add motion vector pairs into
the candidate list. It further improves the compression performance
by 0.1% across derf, hevcmr, stdhd, and hevchr sets without adding
encode/decode time.
Alex Converse [Tue, 12 Jan 2016 00:16:19 +0000 (16:16 -0800)]
Compress the final ANS state.
The '110' prefix on a final byte indicates a superframe marker. Coded
data is not allowed to use this pattern on a final byte.
Code |state - l_base| little endian with the following prefix scheme:
Prefix '00': Single byte coded state.
Prefix '01': Two bytes le coded state.
Prefix '10': Three bytes le coded state.
Scott LaVarnway [Tue, 12 Jan 2016 13:09:06 +0000 (05:09 -0800)]
VP9: Eliminate unnecessary nearest/near searches
Prior to this patch, read_inter_block_mode_info() would
find the nearmv and nearestmv for all modes. Now it does not
search for ZEROMV modes and breaks out early for NEARMV and
NEWMV modes.
Jingning Han [Sat, 12 Dec 2015 06:39:40 +0000 (22:39 -0800)]
Refactor ref mv stack system
This commit re-works the reference motion vector stack process
and make it support extended context set. It unifies reference
motion vector checking process for row and column scan, as well as
for single block scan.
Yaowu Xu [Thu, 7 Jan 2016 17:43:26 +0000 (09:43 -0800)]
Enable encoder to avoid 8x4 or 4x8 partitions
This commit enables encoder to avoid 8x4 and 4x8 partitions for
scaled reference frames when libvpx is configured and built with
--enable-better-hw-compatibility
Marco [Wed, 6 Jan 2016 19:30:02 +0000 (11:30 -0800)]
vp9-svc: Fix to avoid msan unitialized value.
Move the logic for forcing zero_mode after the
(ref_frame & flag_list) check.
This was causing an memory leak under msan:
https://bugs.chromium.org/p/webrtc/issues/detail?id=5402
Yaowu Xu [Tue, 5 Jan 2016 02:24:18 +0000 (18:24 -0800)]
Assert no mv clamping for scaled references
Under --enable-better-hw-compabibility, this commit adds the asserts
that no mv clamping is applied for scaled references, so when built
with this configure option, decoder will assert if an input bitstream
triggger mv clamping for scaled reference frames.
Yaowu Xu [Thu, 3 Dec 2015 01:00:26 +0000 (17:00 -0800)]
Assert no 8x4/4x8 partition for scaled references
This commit adds a new configure option:
--enable-better-hw-compatibility
The purpose of the configure option is to provide information on known
hardware decoder implementation bugs, so encoder implementers may
choose to implement their encoders in a way to avoid triggering these
decoder bugs.
The WebM team were made aware of that a number of hardware decoders
have trouble in handling the combination of scaled frame reference
frame and 8x4 or 4x8 partitions. This commit added asserts to vp9
decoder, so when built with above configure option, the decoder can
assert if an input bitstream triggers such decoder bug.
vp10: only assume ONLY_4X4 if segmentation is disabled.
Otherwise, per-segment lossless might mean that some segments are not
lossless and they could still want to use another mode. The per-block
tx points remain uncoded on blocks where (per the segment id) the Q
value implies lossless.
Marco [Mon, 21 Dec 2015 18:35:05 +0000 (10:35 -0800)]
aq-mode=3: Don't reset segment if block is determined to be skin.
For coding block sizes <=16X16, if the block is determined to be skin,
then always allow for that block to be candidate for refresh. So if that
block happens to be on the boost segment(s), segment won't get reset to 0
and delta-q will be applied.
PSNR/SSIM metrics neutral (little/no change) on RTC clips.
Speed increase small/negligible (< 1%).
Some visual improvement on faces in a few RTC clips.
Peter de Rivaz [Mon, 14 Dec 2015 16:35:29 +0000 (16:35 +0000)]
Fix for issue 1114 compile error
In 32-bit build with --enable-shared, there is a lot of
register pressure and register src_strideq is reused.
The code needs to use the stack based version of src_stride,
but this doesn't compile when used in an lea instruction.
This patch also fixes a related segmentation fault caused by the
implementation using src_strideq even though it has been
reused.
This patch also fixes the HBD subpel variance tests that fail
when compiled without disable-optimizations.
These failures were caused by local variables in the assembler
routines colliding with the caller's stack frame.
projects / libvpx / log