Jingning Han [Fri, 12 Dec 2014 01:17:53 +0000 (17:17 -0800)]
Fix PICK_MODE_CONTEXT index in non-RD coding mode
This commit fixes a bug in the PICK_MODE_CONTEXT index for
horizontal partition case. The compression performance change
is less than 0.01% level, since most blocks are selected to
use square block size in RTC coding mode.
Peter de Rivaz [Thu, 11 Dec 2014 15:54:23 +0000 (15:54 +0000)]
Corrected optimization of 8x8 DCT code
The 8x8 DCT uses a fast version whenever possible.
There was a mistake in the checking code which
meant sometimes the fast version was used when it
was not safe to do so.
Jingning Han [Thu, 11 Dec 2014 17:29:36 +0000 (09:29 -0800)]
Refactor choose_partitioning computing scheme
This commit refactors the choose_partitioning function. It removes
redundant memset calls and makes the encoder to calculate
variance value per block only when it is needed. It reduces the
average runtime cost of choose_partitioning by 60%. Overall it
reduces speed -6 runtime by 2-5%.
JackyChen [Tue, 2 Dec 2014 20:14:47 +0000 (12:14 -0800)]
Multiframe Quality Enhancement(MFQE) in VP9.
It is the first version of MFQE in VP9. There are a few TODOs included
in this version.
Usage: Add flag --enable-vp9-postproc to config the project.
In decoder, use flag --mfqe in the command line to enable
MFQE in postproc.
Note: Need to have key frame with low quality to see the effect of this
new patch. In my experiment, I fixed the qindex to 200 in key frame.
Jingning Han [Wed, 10 Dec 2014 02:01:17 +0000 (18:01 -0800)]
Use use_prev_frame_mvs flag for ref mv search branch
Replace error_resilient flag with use_prev_frame_mvs in
vp9_pick_inter_mode reference motion vector search selection.
This effectively turns off the simplified ref mv search in the
settings of frame resizing, even if error-resilient mode is off.
hkuang [Tue, 9 Dec 2014 22:32:48 +0000 (14:32 -0800)]
Fix clang ioc warning due to NULL src_mi pointer.
The warning only happens in VP9 encoder's first pass due to src_mi
is not set up yet. But it will not fail the encoder as left_mi and
above_mi are not used in the first_pass and they will be set up again
in the second pass.
Jingning Han [Tue, 9 Dec 2014 19:31:45 +0000 (11:31 -0800)]
Make RTC coding flow support sub8x8 in key frame coding
This commit enables the use of sub8x8 blocks in RTC key frame
encoding. It requires the block size to be preset and will decide
the coding mode and encode the bit-stream.
Paul Wilkins [Thu, 27 Nov 2014 10:50:56 +0000 (10:50 +0000)]
Substantial restructuring of AQ mode 2.
The restructure moves the decision into the rd pick
modes loop and makes a decision based at the 16x16
block level instead of only the 64x64 level.
This gives finer granularity and better visual results
on the clips I have tested. Metrics results are worse
than the old AQ2 especially for PSNR and this mode
now falls between AQ0 and AQ1 in terms of visual
impact and metrics results.
Further tuning of this to follow.
It should be noted that if there are multiple iterations
of the recode loop the segment for a MB could change
in each loop if the previous loop causes a change in the
complexity / variance bin of the block. Also where a block
gets a delta Q this will alter the rd multiplier for this block
in subsequent recode iterations and frames where the
segmentation is applied.
levytamar82 [Fri, 5 Dec 2014 18:14:33 +0000 (11:14 -0700)]
SSSE3 Optimization for Atom processors using new instruction selection and ordering
The function vp9_filter_block1d16_h8_ssse3 uses the PSHUFB instruction which has a 3 cycle latency and slows execution when done in blocks of 5 or more on Atom processors.
By replacing the PSHUFB instructions with other more efficient single cycle instructions (PUNPCKLBW + PUNPCHBW + PALIGNR) performance can be improved.
In the original code, the PSHUBF uses every byte and is consecutively copied.
This is done more efficiently by PUNPCKLBW and PUNPCHBW, using PALIGNR to concatenate the intermediate result and then shift right the next consecutive 16 bytes for the final result.
This optimization improved the function performance by 23% and produced a 3% user level gain on 1080p content on Atom processors.
There was no observed performance impact on Core processors (expected).
Jim Bankoski [Sun, 7 Dec 2014 19:28:51 +0000 (11:28 -0800)]
Make the decoder Cfg available to encoder tests..
Adds decoder config as a changeable parameter to unit tests, and
changes end to end test to use commonly used parameters to enable
base test of tiles encoding and frame parallel decoding.
Jingning Han [Thu, 4 Dec 2014 22:53:36 +0000 (14:53 -0800)]
Enable conditional skip path in rd_pick_intra_sby_mode
These speed-up features for key frame coding are only turned on
in the settings of hybrid non-RD and RD mode decision. It provides
about 20% speed-up to the hybrid key frame coding at the expense
of certain compression performance loss. For vidyo1, the key frame
coding statistics are changed
9838F, 35.020 dB, 61677 us -> 9920F, 34.834 dB, 47556 us
Overall rtc set compression performance is down by -0.257%.
Jingning Han [Thu, 4 Dec 2014 17:29:24 +0000 (09:29 -0800)]
Use hybrid RD and non-RD coding flow for key frame coding
When block size is below 16x16, the encoder swap from non-RD to
RD mode for key frame coding. This largely brough back the key
frame compression performance. For vidyo1 at 1000 kbps, the key
frame coding statistics are changed
9978F, 34.183 dB, 36807 us -> 9838F, 35.020 dB, 61677 us
As compared to the full RD case
7187F, 34.930 dB, 214470 us
The overall rtc set coding performance (single key frame setting)
is improved by 1.5%.
Peter de Rivaz [Thu, 4 Dec 2014 10:51:10 +0000 (10:51 +0000)]
Use the RTC optimizations when in high bitdepth mode.
Change 72193 made the encoder behave differently
when configured with and without high bitdepth.
This change means the same algorithm is used for both.
Yunqing Wang [Tue, 2 Dec 2014 23:47:41 +0000 (15:47 -0800)]
vp9_ethread: the tile-based multi-threaded encoder
Currently, VP9 supports column-tile encoding, which allows a frame
to be encoded in multiple column tiles independently. The number of
column tiles are set by encoder option "--tile-columns". This
provides a way to encode a frame in parallel.
Based on previous set of patches, this patch implemented the tile-
based multi-threaded encoder. Each thread processes one or more
tiles.
Usage:
For HD clips:
--tile-columns=2 --threads=1/2/3/4
While using 4 threads, tests showed that the encoder achieved
2.3X - 2.5X speedup at good-quality speed 3, and 2X speedup at
realtime speed 5.
projects / libvpx / log