The more aggressive settings should only be used when denoise_svc
condition is satisfied (which means top spatial layer).
Change-Id: Ia8e3515b27f31bf21b1976ca80a2fa826daece3a
In non-rd pickmode (speed >= 5), avoid duplication of computations in
model_rd_for_sb_y when the speed feature use_simple_block_yrd is
enabled (or for high bitdepth build under certain conditions).
QVGA, VGA and HD have 1.23%, 2.68% and 1.7% speedup on ARM for speed 8,
respectively.
Encoding results are bitexact for speed >= 5.
Change-Id: I3f9130810c21439f5ad7e159e21cb2243dcd05f1
The allow_exhaustive_searches feature improves the encoding quality
of FC_GRAPHICS_ANIMATION content a lot. For non-FC_GRAPHICS_ANIMATION
content, the quality test result is almost neutral. This patch makes
this feature to be used only for FC_GRAPHICS_ANIMATION content.
The motivation of doing that is to make this feature no longer adaptive,
which will be implemented in the following patch.
Change-Id: Ic911df6dd757402b6480789cc247801e99840369
Replace by CAST_TO_BYTEPTR/SHORTPTR.
The rule is: if a short ptr is casted to a byte ptr, any offset
operation on the byte ptr must be doubled. We do this by casting to
short ptr first, adding offset, then casting back to byte ptr.
BUG=webm:1388
Change-Id: I9e18a73ba45ddae58fc9dae470c0ff34951fe248
The scaling filter with zero shift will give sub-sampling for
2x downsampling. Allow for a phase shift to get an averaging filter.
Usage is for source scaling in 1 pass SVC mode for 1:2 downscale.
Reduces aliasing in downsampled image.
Keep the phase to 0/off for now.
Change-Id: Ic547ea0748d151b675f877527e656407fcf4d51e
This condiiton is not needed as key_frame should set the refresh
of the reference frames, but good to have for clarity in condition.
Change-Id: Icf9838e7e4f0ff5cf0a9562ae3b5d6c7e6f78702
This reverts commit 863f860bfc.
This causes encoder / decoder mismatches in various
VP9/DatarateTestVP9Large.BasicRateTargeting3TemporalLayers tests
BUG=webm:1408
Change-Id: Ic200c39d7ed9c0b0247ef562f5d6f7b2625f7e14
For low resolutions (<= CIF): use quarter-pixel and simple_block_yrd.
~5% gain on RTC_derf.
~6-7% slowdown on ARM.
Change-Id: I4439ebd1116b9decac04786503f978840b68a60c
Fix to avoid getting stuck at very low Q even
though content is changing, which can happen for --min-q=0.
Fix is to more aggressively increase active_worst_quality
when detecting significant rate_deviation at very low Q.
Change will only affect 1 pass VBR for --min-q < 4, so no
change in ytlive metrics for --min-q >= 4.
Change-Id: I4dd77dd7c08a30a4390da0ff2c8bda6fccfa76d7
Useful for SVC, where the top layer enhancement frames may
not update any reference buffers, as is the case for the
patterns in the 1 pass CBR SVC when #temporal_layers > 1.
~3% encoder speedup for SVC patterns with temporal layers
in 1 pass CBR mode.
Updated the SVC datarate tests for the mismatch frames.
Set the frame-dropper off in some tests with #temporal_layers > 1
so we can correctly set #mismatch frames. Adjusted rate target
threshold for tests where frame-dropper was turned off.
Change-Id: Ia0c142f02100be0fed61cd2049691be9c59d6793
The MV unit test revealed an integer overflow issue in vp9_mcomp.c.
This was caused if the MV was very large. In mv_err_cost(), when
mv->row = 8184, mv->col = 8184 and ref_mv is 0, mv_cost = 34363
and error_per_bit = 132412, causing the overflow.
BUG=webm:1406
Change-Id: I35f8299f22f9bee39cd9153d7b00d0993838845e
Set adaptive_rd_thresh to 2 when simple block yrd is not used.
Fix regression caused by computing y sad without
int_pro_motion_estimation on low res motion clips.
Overall 0.07% quality loss on rtc_derf.
Change only affects low res on speed 8.
Change-Id: Ic6a188a56529f1034d6431005fb4b0e24e8a7e27
For speed 5, 1 pass CBR: Don't use the nonrd_pick_partition
on the segment, rather use choose_partitioning followed by
nonrd_select_partition (as is done on base segment).
Little/no quality loss on RTC and RTC_derf (< 0.3%),
speedup of at least 5%.
Change-Id: I5273d5f950e60adf5e437b4ca8c4f63964641e83
If the noise estimation is avoided due to large motion,
the last_source for denoising should still be updated.
Change-Id: I67155ea7dbe9ac2785978e64a27bdafd7d57aac0
To reduce refresh on partial super-blocks on boundary,
for noisy input. Reduces some artifacts on noisy input.
Change-Id: I10b5808a296874e08c7f378b3df58466591d8dbe
Edit
Move the condition for effectively disabling the denoising
for speed 5 into the vp9_denoiser_denoise().
This is cleaner, and also moving the condition into vp9_denoiser_denoise
will keep the denoiser buffer updated with the current source.
This allows for more consistent behavior if speed is changed midstream.
Change-Id: Ia001f591c56e454bf724c3ae73c024badb183ef8
To prevent the motion vector out of range bug, added a motion vector unit
test in VP9. In the 4k video encoding, always forced to use extreme motion
vectors and also encouraged to use INTER modes. In the decoding, checked if
the motion vector was valid, and also checked the encoder/decoder mismatch.
The tests showed that this unit test could reveal the issue we saw before.
Change-Id: I0a880bd847dad8a13f7fd2012faf6868b02fa3b4
Temporal denoiser runs in non-rd pickmode, so it is only used
for speed >= 5. Regression exists for speed 5, due to use of
reference_partition (which use non-rd pickmode for partitioning).
Avoid denoising for now at speed 5.
Change-Id: I74a74d2e1404d7cfd33dcf4ec06dd2e503256cf0
Base the low_content_frame metric on the motion vectors,
and adjust the logic for preventing golden update.
Small change in behavior: small positive gain (~0.2-1%) on clips
with high activity.
Change-Id: I0b861c8e9666cd82b45cde5ee57ee8a1e5ab453c
Code cleanup: merged two functions that were doing postencode
update for cylic refresh, remove some unused code and fix comments.
No change in behavior.
Change-Id: I9be0d7e346d34dec29bf4e5bb380a7bf81c8480a
BUG=webm:1397
(yunqingwang)
To verify that this patch wouldn't cause much performance change,
the Borg tests were run. Here was the result:
avg_psnr overall_psnr ssim
hdres: -0.002 0.006 0.013
midres: 0 0 0
lowres: 0 0 0
Change-Id: Iae395ae7b741e0513cf5bab9dcace110b792a67d
The row mt sync read uses sync_range = 1, and wouldn't work if we want
to use a sync_range that is greater than 1. To make it work, this sync
read code is modified. Pass in col instead of col - 1 to make it
consistent with other row mt code in VP9, and then add 1 in "while"
codition.
Change-Id: I4a0e487190ac5d47b8216368da12d80fec779c1a
Issue/bug happens for denoising with spatial layers, where
the golden (spatial) reference is used in pickmode, but
denoising is only done wrt to last (temporal).
Fix is to make sure set_ref_ptrs is set before build predictors
in denoiser.
Change-Id: I793cf441341edf7c4a88b8ab1e1b22b3cb0eb508
Temporary override to condition for disallowing intra-search in SVC,
since golden (spatial) reference is currently suppressed due to
artifact issue.
Change-Id: I28ed7fdddc9fcdbcc0a4175a247a3ecc94c11767
For non-rd variance partition, avoid the chrome check
unless y_sad is below some threshold.
Small decrease in avgPSNR (~0.3) on RTC set.
Small/negligible decrease on RTC_derf.
Change-Id: I7af44235af514058ccf9a4f10bb737da9d720866
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
Make the source_sad feature work properly for cases of VBR or
screen_content with SVC.
Added unittest for SVC with screen-content on.
Change-Id: Iba5254fd8833fb11da521e00cc1317ec81d3f89b
Since y_sad is not computed yet (on the early exit due to source_sad),
no need to check for setting color_sensitiviy.
Only affects speed >=8. No change in behavior.
Change-Id: I3a6f2d20fed38d8b8ec51b75bcacf9a21f2db916
Allow for simple_block_rd for VGA resoln, and reduce
adaptive_rd_thresh to 1.
On average no loss on RTC set, ~4% speedup on mac.
Change-Id: Ib549c4061c853776062b5e34040f839d470fbebc
Change it to row based array to avoid the slow down cause by sync.
row-mt on, speed 8, 2 threads: ~4% speedup for VGA on ARM benefited
from adaptive_rd_threshold.
Change-Id: I887e65a53af20a6c4f48d293daaee09dab3512cf
Add additional condition to split to 16x16, for resolutions <= 360p,
reduces dragging artifact near moving boundary.
Small/no change on RTC metrics.
Change-Id: I314694f2166435d918f74e7ab42f002b07f40dae
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
Add routine vp9_model_rd_from_var_lapndz_vec and call it from model_rd_for_sb
to model the rate and distortion for MAX_MB_PLANE Laplacian sources in
parallel. The caller ensures that all sources have non-zero variance.
Measured a 18% to 25% reduction in retired instructions, and 17% to 24%
reduction in instruction execution cost with different compilers for the
Laplacian modeling.
No change in behavior.
TEST=Verified that encoded files match bit for bit, with and without this
change.
BUG=b/33678225
Change-Id: I6b76947f21c659a349adb896e13e99f6e3f951e6
Don't denoise spatial layer frames whose base layer is a key frame.
Disallow golden reference for SVC with denoising on frames
that will be denoised (highest layer), as this removes bad artifact.
Will re-enable when issue is resolved.
Change-Id: I87a6597812330500966458172acfce54af65f70f
Fix the update of the denoiser buffer when the base
spatial layer is a key frame. And allow for better/lower
QP on high spatial layers when their base layer is key frame.
Change-Id: I96b2426f1eaa43b8b8d4c31a68b0c6d68c3024a2
Reduce it from 5 to 4, small/no change in metrics or speed.
Small reduction in dragging artifact near moving head.
Change-Id: Ic3bc5ca67c70bf0c89fc2ed14454840a28ae5b6a
This patch was based on Yang Xian's intern project code. Further modifications
were done.
1. Moved machine-learning related parameters into the context structure.
2. Corrected the calculation of sum_eobs.
3. Removed unused parameters and calculations.
4. Made it work with multiple tiles.
5. Added a speed feature for the machine-learning based partition search
early termination.
6. Re-organized the code.
The patch was rebased to the top-of-tree.
Borg test BDRATE result:
4k set: PSNR: +0.144%; SSIM: +0.043%;
hdres set: PSNR: +0.149%; SSIM: +0.269%;
midres set: PSNR: +0.127%; SSIM: +0.257%;
Average speed gain result:
4k clips: 22%;
hd clips: 23%;
midres clips: 15%.
Change-Id: I0220e93a8277e6a7ea4b2c34b605966e3b1584ac
Fixes an issue when the LAST and golden is not used as a reference,
in which case its possible no encoding mode is set (since intra may be
skipped under certain codtions). Fix is to make sure intra is searched
if no inter mode is checked.
Issue can happen for temporal layer pattern#7 in vpx_temporal_svc_encoder.c
Change-Id: I5ab4999b2f9dbd739044888e0916b5ec491d966b
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
Enable row-mt for SVC for real-time mode, speed >=5.
Add the controls to the sample encoders, but keep it off for now.
Add the control and enable it for the 1 pass CBR unittests.
For speed 7, 3 layer SVC, 2 threads, row-mt enabled gives about ~5% speedup.
Change-Id: Ie8e77323c17263e3e7a7b9858aec12a3a93ec0c1
this is similar to the x86 configuration and helps mitigate an issue
with a circular dependency between this function and the ssse3 variant
causing an outsized increase in binary size (~300K for chrome)
chrome.dll:
.text 255B000 -> 252B000
.data 7B000 -> 75000
-221184 bytes
BUG=chromium:697956
Change-Id: Ic95b142ecd62dd4f1795788aa27dd8fab59b708c
The 2 thresholds(i.e. partition_search_breakout_dist_thr and
partition_search_breakout_rate_thr) are used as the partition search
early termination speed feature. This refactoring patch made this
feature to be frame size dependent consistently throughout the code.
Change-Id: Idaa0bd8400badaa0f8e2091e3f41ed2544e71be9
Fix the conditon for getting last_source when denoising is on.
This avoids unneeded scaling in the case of SVC.
No change in quality.
Change-Id: I32c1c2c9085104da51af8535716bcc4d55fb0f42
Reduce the level from 4 to 2.
This gives ~1-2% quality gain on RTC set, with small decreaee in speed (~1-2% on mac).
Change-Id: I7d959731badcee3d45b2f4a08efe378765016a13
This reverts commit d3db846cc5.
This change causes a large drop in psnr (4-5db) on low framerate
difficult content (tested at 360/480p)
BUG=b/35804225
Change-Id: I8e90012d3b9c8a0cddb062ba93b01b36c0e0c0a0
From commit:
https://chromium-review.googlesource.com/c/441393/
On non-segment the set_vbp_thresholds() should be called
again to adjust thresholds based on content_state of superblock.
This was the intended behavior from 441393.
Small change in RTC metrics and speed.
Change-Id: I45e5fbdc4af74db76b3cb4f13074fcae0eb2219e
new_mt is a very generic name that will get obsolete soon enough.
Since this is exposed as a codec control, renaming it to row_mt to
signify row level paralellism. Also renaming the ETHREAD_BIT_MATCH
codec control to ROW_MT_BIT_EXACT.
Change-Id: Ic7872d78bb3b12fb4cf92ba028ec8e08eb3a9558
vp9_highbd_block_error_8bit_c was a very simple wrapper around
vp9_block_error_c. The SSE2 implemention was practically identical to
the non-HBD one. It was missing some minor improvements which only
went into the original version.
In quick speed tests, the AVX implementation showed minimal
improvement over SSE2 when it does not detect overflow. However, when
overflow is detected the function is run a second time. The
OperationCheck test seems to trigger this case and reverses any
speed benefits by running ~60% slower. AVX2 on the other hand is
always 30-40% faster.
Change-Id: I9fcb9afbcb560f234c7ae1b13ddb69eca3988ba1
Only works for bitdepth = 8 when compiled with high bitdepth flag.
4x speed ups for handling 1:2 down/upsampling.
Validated manually for:
1) Dynamic resize for a single layer encoding
2) SVC encoding with 3 spatial layers
Results are bitexact with the patch and the speed gain (~4x) in the
scaling was verified.
BUG=webm:1371
Change-Id: I1bdb5f4d4bd0df67763fc271b6aa355e60f34712
The reduction showed improvement on RTC when aq-mode=3 is on.
Add that (cyclic refresh enabled) to the condition.
Only affects 1 pass CBR.
Change-Id: I5d0843002d8e31d7c165098a62e7a71146b08664
For speed 8 only.
3% speed up for QVGA and 6.3% for VGA on Nexus 6.
~3% avgPSNR decrease on rtc_derf and 2.9% on rtc.
Disabled for now.
Change-Id: I70133f1f6c804d663d594df437bfe7fdb0030d6a
Increase the variance partition thresholds for superblocks that
have low sum-diff (from source analysis prior to encoding frame).
Use it for now only for speed >= 7 or for denoising on.
Small change on metrics for rtc set: less than ~0.1 avgPNSR decrease
on RTC set, for both speed 7 and 8.
Change-Id: I38325046ebd5f371f51d6e91233d68ff73561af1
Use the simple block_yrd under certain conditions.
The optimization code is completed but the speed is still slower
(~6% on 720p) than the low-bitdepth build.
For now, use the more complex block_yrd under certain conditions
(always use it for speed <= 5, otherwise use it on key frames and for
bsize >= 32x32).
This gives about ~2-3% gain in quality for speed 7 on RTC set
(over high bitdepth build), with about the same encoder fps as the
low bitdepth build.
Change-Id: Ibe92a1945d0bd635f880befb4c815727df62d754
Modified the code to facilitate bit-match tests in first pass
Added unit-tests to test the row based multi-threading behavior for bit-exactness
Change-Id: Ieaf6a8f935bb1075597e0a3b52d9989c8546d7df
This change subtracts out low complexity intra regions that are also low
error in the inter domain, in the calculation of the frame prediction decay.
The rationale here his that low complexity regions (such as sky) do not imply
high prediction decay in the same way as high error intra or neutral blocks.
The effect of this is small in most clips but in a few clips it can be > 10%.
(E.g. In to tree)
Change-Id: If67ac23d17fca14285cad2defa464c61c9ea861c
vp9[_highbd]_quantize]_fp[_32x32] and vp9_fdct8x8_quant do not make use
of these parameters.
scan is used for C code and iscan is used for SIMD implementations.
Change-Id: I908a0ff7d3febac33da97e0596e040ec7bc18ca5
* changes:
quantize_fp_32x32 highbd ssse3: enable existing function
quantize_fp highbd ssse3: use tran_low_t for coeff
quantize_fp highbd sse2: use tran_low_t for coeff
The previous implementation confused bit/bytes/elements. It was using
'32' as the multiplier but that was mistakenly adopted because a 32x32
transform embedded the stride.
Change-Id: Ieeb867a332416b9a40580b5e7c9b20088e9e691a
The weight segment needs to only be computed once per frame,
so remove it from the funciton vp9_cyclic_refresh_rc_bits_per_mb(),
which is called within a loop inside vp9_rc_regulate_q.
Change-Id: Ia0e18b89abb97e42c466d4dbc47700d7f76555db
vp9_compute_qdelta_by_rate has almost 2% overhead in profiling on Nexus 6.
Reduce the calling of that function in speed 8 by estimating the delta-q.
Both rtc and rtc_derf show little/no change in avg psnr/ssim.
Encoding speed is 2~3% faster on Nexus 6.
Change-Id: If25933715783f31104a18a5092ea347b1221b5f5
This small change replaces the frame boost check in the arf group
length break out clause with a test against a prediction decay value.
The boost value is in fact partly dependent on the decay value but
this change means that the per frame boost calculation can be adjusted
without influencing the group length calculation.
The value chosen gives a close match on all the test sets with the previous
code (on average) but it was noted that a lower threshold was slightly better
for 1080P and up and a slightly higher value for small image sizes.
Change-Id: I4d5b9f67d5b17b0d99ea3f796d3d6202fd61ee0c
The function scale_sse_threshold() returns a threshold scaled
if necessary for use with 10 and 12 bit from an 8 bit baseline.
SSE error values would be expected to rise for the 10 and 12
bit cases where there are more bits of precision.
Hence the threshold used for the test should also be scaled up.
Change-Id: I4009c98b6eecd1bf64c3c38aaa56598e0136b03d
(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
