When the superblock partition is based on the nonrd-pickmode,
we need to avoid the denoising. Current condition was based on
the speed level. This change is to make the condition at the
superblock level, as the switch in partitioning may be done at
sb level based on source_sad (e.g., in speed 6).
Change-Id: I12ece4f60b93ed34ee65ff2d6cdce1213c36de04
When int_pro_motion_estimation is done for superblock in
choose_partitioning, use it to avoid the full_pixel_search
for NEWMV mode, if bsize is >= 32X32.
For speed > 7.
Small/neutral change on RTC metrics.
~1-2% speedup on arm on high motion clip.
Change-Id: I3cfe6833ff4bf75d4afa83eaf058ad45729de85b
Add a low-variance high-sumdiff to the superblock content state
and use it to limit the mv and bias some decisions in non-rd pickmode.
Only affects speed >= 6.
Reduces artifact for lighting changes.
Small/no difference in metrics on RTC set.
Change-Id: Ic84b2379fe0ae3fa71ae826ee6bae3eaf551a25b
A previous patch turned on allow_exhaustive_searches feature only for
FC_GRAPHICS_ANIMATION content. This patch further modified the feature
by removing the exhaustive search limit, and made it no longer adaptive.
As a result, the 2 counts that recorded the number of motion searches
were removed, which helped achieve the determinism in the row based
multi-threading encoding. Tests showed that this patch didn't cause
the encoder much slower.
Used exhaustive_searches_thresh for this speed feature, and removed
allow_exhaustive_searches. Also, refactored the speed feature code
to follow the general speed feature setting style.
Change-Id: Ib96b182c4c8dfff4c1ab91d2497cc42bb9e5a4aa
Refactor to split the 1 passs source sad computation into scene
detection (currently used for VBR and screen-content mode), and
superblock based source sad computation (used in non-rd CBR mode).
This allows the source sad computation for CBR mode to be
multi-threaded.
No change in compression.
Change-Id: I112f2918613ccbd37c1771d852606d3af18c1388
For each superblock, keep track of how far from current frame
was the last significant content change, and use that (along
with GF distance), to turnoff GF search in non-rd pickmode.
Only enabled for speed >= 8.
avgPNSR on RTC/RTC_derf down by ~0.9/1.2.
Speedup on mac: ~3-5%.
Speedup on arm: 3.6% for VGA and 4.4% for HD.
Change-Id: Ic3f3d6a2af650aca6ba0064d2b1db8d48c035ac7
The sum of tx bloxk eobs is needed in the machine learning based partition
early termination. The eobs are first accumulated during tx search, and
then the value associated with the best tx_size is copied to ctx for later
use.
After the sum of eobs are calculated correctly, re-enabled
ml_partition_search_early_termination speed feature.
Re-did the quality/speed test to check the impact of the fix.
1. Borg test BDRATE result:
4k set: PSNR: +0.183%; SSIM: +0.100%;
hdres set: PSNR: +0.168%; SSIM: +0.256%;
midres set: PSNR: +0.186%; SSIM: +0.326%;
2.Average speed gain result:
4k clips: 21%;
hd clips: 26%;
midres clips: 15%.
The result is in line with the original result.
Change-Id: I4209a95c89be03b4cbfb6a95b16885f89feddbda
shift the bsse[] member of the macroblock struct to the front to avoid
an incorrect offset (0) to the upper half of bsse[0] which leads to a
negative resulting in a crash. restrict this to visual studio versions
before 2015 (the bug was observed with 2013, fixed in 2015) to avoid any
potential cache impact on other platforms.
https://connect.microsoft.com/VisualStudio/feedback/details/2396360/bad-structure-offset-in-32-bit-code
BUG=webm:1054
Change-Id: I40f68a1d421ccc503cc712192263bab4f7dde076
(Yunqing Wang)
This patch implements the row-based multi-threading within tiles in
the encoding pass, and substantially speeds up the multi-threaded
encoder in VP9.
Speed tests at speed 1 on STDHD(using 4 tiles) set show that the
average speedups of the encoding pass(second pass in the 2-pass
encoding) is 7% while using 2 threads, 16% while using 4 threads,
85% while using 8 threads, and 116% while using 16 threads.
Change-Id: I12e41dbc171951958af9e6d098efd6e2c82827de
(yunqingwang)
1. Rebased the patch. Incorporated recent first pass changes.
2. Turned on the first pass unit test.
Change-Id: Ia2f7ba8152d0b6dd6bf8efb9dfaf505ba7d8edee
Only for speed >= 7, and affects skipping of intra modes.
Threshold is set low for now, needs to be tuned.
Small/no difference in metrics on rtc clips.
Change-Id: If9bdbd43f08d1f80407cdd2e9e5e96780dcd2424
Previously Tx domain rd was used in all cases above speed 0.
Coefficient optimization was only enabled for best and speed 0.
This patch selectively sets these features at other speed settings
based on block complexity.
For the Netflix and HD sets in particular the quality gains are
large compared to the speed hit. At speed 1 the average psnr
gain in the NF set is > 2.5% with one clip coming in at 18%
and some points almost 30%. Average gains for the lower
resolution test sets are around 1%.
The gains are biggest at low Q so some further optimization
may be possible.
Change-Id: I340376c7b2a78e5389a34b7ebdc41072808d0576
1. Skip golden non-zeromv and newmv-last for bsize >= 16x16 if the
temporal variance obtained from choose_partitioning is very low.
2. Skip horz and vert INTRA mode for speed 8.
This change works best on the clips with little noise and with some
motion (e.g. gips_motion which has > 5% speed up). PSNR drop is 1.78%
on rtc test set, no obvious visual quality regression found.
Change-Id: Ib43b5b20e67809d03c5a6890818ddff59e1fc94a
Skip intra-mode and some inter-modes (newmv, nearmv, nearestmv) for
golden frame if the variance got from choose_partitioning is very low.
Only for 1 pass real-time CBR mode and bsize >= 32x32, it has ~2.5%
speed up with less than 0.1% PSNR drop for rtc test set. Don't see
visual regression.
Change-Id: I70efbc95a1007231ae36f02c5b2fbf6cd35077ad
The bit to error transformation got doubled as a result of going from
8-bit to 9-bit costs (change d13385c).
Use defines to derive the scale numbers and comment some of the fields.
derf: -0.023 BDRATE
hevcmr: +0.067 BDRATE
stdhd: +0.098 BDRATE
(These are substantially smaller than than the original gains from 8 to
9 bit costing.)
Change-Id: I6a2b3b029b2f1415e4f90a05709b2333ec0eea9b
If a superblock contains alot of "skin" then force split
of 64x64 partition, and make some adjustments in mode selection.
This helps to reduce artifacts on moving face/skin areas at low bitrates.
Little/no change in metrics: avgPSNR/SSIM down by ~0.12%.
Small encoding time increase < 1%.
Change-Id: Ic57f52148c3716f391419fab0530d916e4c1d186
This change alters the nature and use of exhaustive motion search.
Firstly any exhaustive search is preceded by a normal step search.
The exhaustive search is only carried out if the distortion resulting
from the step search is above a threshold value.
Secondly the simple +/- 64 exhaustive search is replaced by a
multi stage mesh based search where each stage has a range
and step/interval size. Subsequent stages use the best position from
the previous stage as the center of the search but use a reduced range
and interval size.
For example:
stage 1: Range +/- 64 interval 4
stage 2: Range +/- 32 interval 2
stage 3: Range +/- 15 interval 1
This process, especially when it follows on from a normal step
search, has shown itself to be almost as effective as a full range
exhaustive search with step 1 but greatly lowers the computational
complexity such that it can be used in some cases for speeds 0-2.
This patch also removes a double exhaustive search for sub 8x8 blocks
which also contained a bug (the two searches used different distortion
metrics).
For best quality in my test animation sequence this patch has almost
no impact on quality but improves encode speed by more than 5X.
Restricted use in good quality speeds 0-2 yields significant quality gains
on the animation test of 0.2 - 0.5 db with only a small impact on encode
speed. On most clips though the quality gain and speed impact are small.
Change-Id: Id22967a840e996e1db273f6ac4ff03f4f52d49aa
to MB_MODE_INFO_EXT. This saves 36 bytes per 8x8 area for
both the decoder and encoder. (encoder has two MODE_INFO
buffers)
Change-Id: If006abb2224acaf326df3c2be09e77e967662107
This commit allows the encoder to account for additional chroma
plane costs in the mode decision process, if the current block
potentially contains significant color change. It improves the
visual quality at very low bit-rates.
The compression performance of dark720p is improved by 12.39% in
speed 6. For jimred at 150 kbps, the PSNR of V component (red)
increased by 0.2 dB, at the expense of about 5% increase in
encoding time. Note that for sequences where the chroma components
are fairly consistent, the encoding time increase is negligible.
On average the rtc set compression performance is improved by
1.172% in PSNR and 1.920% in SSIM.
Change-Id: Ia55b24ef23a25304f7ec9958fbf07fd6e658505c
The max_partition_size and max_partition_size are set at the
beginning while setting speed features, and then adjusted at
SB level. Moving them to mb struct ensures there is a local
copy for each thread.
Change-Id: I7dd08dc918d9f772fcd718bbd6533e0787720ad4
This commit removes the cyclic aq mode dependency on
in_static_area and reworks the corresponding cut-off thresholds.
It improves the compression performance of speed -5 by 1.47% in
PSNR and 2.07% in SSIM, and the compression performance of speed
-6 by 3.10% in PSNR and 5.25% in SSIM. Speed wise, about 1% faster
in both settings at high bit-rates.
Change-Id: I1ffc775afdc047964448d9dff5751491ba4ff4a9
This commit removes unused header file vp9_onyxc_int.h and repeatedly
included file vpx_ports/mem.h from vp9_block.h
Change-Id: I400b210bd1da48f1880bd50a8f4a6e2c690e15a1
Adds various high bitdepth transform functions and tests.
Much of the changes are related to using typedefs tran_low_t
and tran_high_t for the final transform cofficients and intermediate
stages of the transform computation respectively rather than fixed
types int16_t/int. When vp9_highbitdepth configure flag is off,
these map tp int16_t/int32_t, but when the flag is on, they map
to int32_t/int64_t to make space for needed extra precision.
Change-Id: I3c56de79e15b904d6f655b62ffae170729befdd8
This commit allows the encoder to store outcomes of single reference
frame modes and compares them to decide if the inter prediction
filter, forward transform, and quantization can be skipped.
The compression performance of speed 3 is down
derf -0.364%
stdhd -0.198%
For test sequences, the speed 3 runtime is reduced
highway CIF 100 kbps, 51976 ms -> 45033 ms, 13% speed-up
stockholm 720p 1000 kbps, 71826 ms -> 67838 ms, 5.5% speed-up
pedestrian 1080p 2000 kbps, 154924 ms -> 150702 ms, 2.6% speed-up
Change-Id: I5aa26f918d2b4b5197a2c0afa2779319f1c88e44
This commit extends the sse and forward transform computation flag
to support the case 64x64 blocks where there are 4 32x32 2D-DCT
blocks.
Change-Id: I86a3e805dfaa0f3abd812f590520c71aa0e40473
This commit enables encoder to select fast forward transform and
quantization path according to the prediction residual sse/variance,
in the rate-distortion optimization scheme.
Change-Id: Ief9fc3844fd4107166d401970e800c6e5ce2b5fe
This commit re-designs the quantization process for transform
coefficient blocks of size 4x4 to 16x16. It improves compression
performance for speed 7 by 3.85%. The SSSE3 version for the
new quantization process is included.
The average runtime of the 8x8 block quantization is reduced
from 285 cycles -> 255 cycles, i.e., over 10% faster.
Change-Id: I61278aa02efc70599b962d3314671db5b0446a50
Before encoding a frame, calculate and store each 16x16 block's
variance of source difference between last and current frame.
Find partitioning threshold T for the frame from its variance
histogram, and then use T to make partition decisions.
Comparing with fixed 16x16 partitioning, rtc set test showed an
overall psnr gain of 3.242%, and ssim gain of 3.751%. The best
psnr gain is 8.653%.
The overall encoding speed didn't change much. It got faster for
some clips(for example, 12% speedup for vidyo1), and a little
slower for others.
Also, a minor modification was made in datarate unit test.
Change-Id: Ie290743aa3814e83607b93831b667a2a49d0932c
This commit enables a fast path computational flow for forward
transformation. It checks the sse and variance of prediction
residuals and decides if the quantized coefficients are all
zero, dc only, or more. It then selects the corresponding coding
path in the forward transformation and quantization stage.
It is currently enabled in rtc coding mode. Will do it for rd
coding mode next.
In speed -6, the runtime for pedestrian_area 1080p at 1000 kbps
goes down from 14234 ms to 13704 ms, i.e., about 4% speed-up.
Overall coding performance for rtc set is changed by -0.18%.
Change-Id: I0452da1786d59bc8bcbe0a35fdae9f623d1d44e1
The final goal is eventually to get rid of both itxm_add and fwd_txm4x4.
This patch does it in the decoder.
Change-Id: Ibb3db57efbcbb1ac387c6742538a9fcf2c6f24a5
