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
This patch allows the encoder to skip the
un-neccessary motion search in the first pass. It
calculates the error of the zero motion vector using
the last source frame as reference and skips the
further motion search in the first pass if the error
is small.
The encoding speedup of the first pass for slideshow
videos is over 30%. Borg test shows the overall PSNR
performance remain approximately the same (derf -0.009,
hd 0.387, yt 0.021, stdhd 0.065). Individual clips may
have either PSNR gain or loss. The worst PSNR perfomance
is from yt set, with a PSNR loss of -1.1.
Change-Id: I08b2ab110b695e4689573b2567fa531b6457616e
* Only use ZEROMV, disalowing the intra modes that were previously
tested.
* Score rate and distortion as zero.
Change-Id: Ifcf99e272095725f11da1dcd26bd0f850683e680
In non frame-parallel decoding, this works the same way as
current decoding scheme. Every time after decoder finish
decoding a frame, it will swap the current mode info pointer
and previous mode info pointer if the decoded frame needs
to be shown. Both mode info pointer and previous mode info
pointer are from mode info arrays.
In frame-parallel decoding, this will become more complicated
as current frame's mode info pointer will be shared with next
frame as previous mode info pointer. But when one decoder
thread finishes decoding one frame and starts to work on next
available frame, it needs to retain the decoded frame's mode
info pointers until next frame finishes decoding. The mode info
index will serve this purpose. The decoder will use different
buffer in the mode info arrays and use the other buffer to save
previous decoded frame’s mode info.
Change-Id: If11d57d8eb0ee38c8876158e5482177fcb229428
tests failing under Win32/Win64
+ dct16x16_test: add missing avx2 functions (partially disabled)
exercises the forward transforms
no idct/iht implementations, so the c-code is used
Change-Id: I04f64a457fa0828a00f32b5c9fe4f55294f21f61
In non-rd real-time mode, choosing smaller transform size in
encoding gives better video quality and good speed gain than
choosing larger transform size. This patch set tx size search
method to ALLOW_8X8, which is better than using 4x4 or other
larger sizes.
Borg tests on rtc set at speed 6 showed significant gain on quality.
PSNR gain: 11.034% and SSIM gain: 15.466%.
The speed gain is 5% - 12% for <720p clips, and 2% - 7% for
720p clips.
Change-Id: If4dc74ed2df359346b059f47fb73b4a0193ec548
Use of stack frame variable "fps" beyond the lifetime of the function.
fps is sent as a paremeter to output_stats and stored in the
packet holding this encoded frame. This has scope beyond the
lifetime of the calling function.
This reverts commit 3f95a230c7
Change-Id: Icd8e14b3d7dd733590ada12e619b9dce95b6b0f5
The SSSE3 implementation might find a potential overflow issue in
its second 1-D transform, if all input residual pixels are close to
255. This commit fixes the issue and re-enables the unit test on
the SSSE3 version.
Change-Id: I0520478abdab7afd3ff2842516bec951111e9b3c
Right now there is just one place to check: xd->lossless and for the first
pass there is a function is_lossless_requested().
Change-Id: I949a6834e64ce51e422e2892f097f2b871b5429a
In Aq mode 2 for kf/arf/gf the segment q delta
is calculated and then applied by re-quantization without
going through the rd loop again. If the base Q != 0
but the segment Q == 0 (lossless) this can could give rise
to a situation where we have an illegal combination of
transform size and Q. (Q == 0 requires that all blocks
are coded 4x4 WHT).
Change-Id: I241a58c6494ed442e9e4630070b0cde0fb99ae45
As a side-effect, the sad unit tests for VP8 and VP9
had to be separated.
Fixes a bug in original patch:
(https://gerrit.chromium.org/gerrit/#/c/70163/8)
that was reverted due to a nightly test failure.
Change-Id: Ia2a4e9e278fd3c89d6c3c82fcc6381320ec2a8a6
In frame parallel decoding mode, there will be still several frames inside
the decoder when application stop calling vpx_codec_decode to decode frames.
The application will need to keep calling vpx_codec_get_frame to get all the
remaining decoded frames in the decoder.
Change-Id: I2ce8260a91282f045bb9a6093ff8a606b1990f14
This commit added a call to set speed feature before initializing
motion search, fixed the problem where unintialized search method
is used before its value being set.
Change-Id: I537e4612bf0d00fd6f51396fd222d4b3bd6fde58
Making this consistent with intra mode masks: you need to specify
allowed inter/intra modes to use.
Change-Id: Iaecd28bf79047259707d8e7a59a57bb7b856383e
SEG_LEVEL_SKIP requires the block size to be at least 8x8. Attempting to
use it on smaller partitions causes the decoder to reject the bitstream.
Change-Id: Ia7188cdf8ae5ac1df6bd29f3f80dbb0610e1f7b1
An overflow issue could potentially happen in the second round 1-D
transform of the SSSE3 full inverse 16x16 2D-DCT. This commit fixes
this issue.
Change-Id: Ia19e4888fda1cc929a28a5f89a5beec612d628dc
This commit enables SSSE3 implementation of the inverse 2D-DCT
with only first 10 coefficients non-zero. It reduces the runtime
of SSE2 version from 745 cycles to 538 cycles, i.e., 27% speed-up.
Change-Id: I18ba4128859b09c704a6ee361d69a86c09fe8dfe
This code dates from the ancient past and
applied an error score weighting based on pixel
brightness. This not seem to be providing any
benefit metrics wise and could be making some
visual issues in dark frames worse.
The field is left in place in the FIRSTPASS_STATS data
structure in this patch, pending changes to unit tests that
use a pre-defined first pass file.
Change-Id: Id50f04205230234858e7548ce523f11acaf3567d
The subpixel SSSE3 was fixed in this patch:
https://gerrit.chromium.org/gerrit/#/c/70283/
So the equivalent AVX2 is fixed accordingly.
Change-Id: Ieebbc1949c99d34b12b8b47692df71aca5001f3a
This commit enables the SSSE3 implementation of full inverse 16x16
2D-DCT. The unit runtime goes down from 1642 cycles to 1519 cycles,
about 7% speed-up.
Change-Id: I14d2fdf9da1fb4ed1e5db7ce24f77a1bfc8ea90d
Further changes to first pass allocation for gf/arf groups.
Three variables removed from TWO_PASS structure as only
now used locally. Dont adjust gf_group_bits in the post
encode update as this will no longer have any effect.
Change-Id: Iff89b225db923fc856f5d2aedbc899f1d7d68b55
In 8-tap filtering, to guarantee the intermediate results fit in
16 bits, the order of accumulating the products needs to be done
correctly, and the largest product should be added last. This
patch fixed the problem using the method in commit "Correct ssse3
8/16-pixel wide sub-pixel filter calculation".
Change-Id: I79d0ad60c057b15011ece84cda9648eee0809423
Restructuring to allocate the bits for each frame in
a GF group at the time the group is defined.
At the moment the allocation closely mirrors what
we had before.
Also changes the default rate adjustment method to
LONG_TERM_VBR_CORRECTION.
Change-Id: Ie5793c46c6b9c888cead5d8790792efd7d60b7c1
As mismatchs were found between the intrinsic version and c only. The
commit temporarily revert to use the matching assembly version to
allow further investigation.
Change-Id: I08436c47d4888b562c0eac8e8856d90a831442df
Use the appropriate subblock offset mode info rather than the parent
block base, when filling mbmi in the pc tree in nonrd_use_partition.
This mimics what is done in the vertical case and what is done for
both cases in nonrd_pick_partition.
This change has little practical effect at the moment since in speed 5
rt horizontal and vertical partitions are currently only used unpaired
at edges of the picture.
Change-Id: I4632f66ca84086dac56c7d36b45ddbe38a06f42a
This did the same correction as the one in commit "Correct ssse3
8/16-pixel wide sub-pixel filter calculation" to avoid saturation
during filtering.
Change-Id: Ife9aa3f62daf9114eb24fe38f7baa3c3f361b2d6
If we are already saving a lot in bits from the target (maximum)
bitrate in the constrained quality mode, allow the quantizer
to go lower than the cq level. This hopefully will solve issues
with getting too low a bitrate and consequently poor quality for
certain videos in cq mode.
Change-Id: I1c4e8b0171fcf58f95198b3add85eea5f3c8f19f
Renames all x86_64 specific assembly files to consistently
end in _x86_64.asm. This will be useful for build systems to
handle these files differently.
All new 64-bit specific assembly files should use the new
naming convention.
Change-Id: I36c89584967c82ffc4088b1b5044ac15d2bb7536
This commit changed to enable the encoder to adjust motion dection
speed threshold based on picture size. In addition, cpu-used 1 now
does a partition search every other frame instead of every third
frame for low resolution inputs.
The change has no quality/speed impact for 720p and above. Test
showed the change increase encoding time by between 3% to 6% for
cpu-used 2 encodiong of 360p sequences. It also has a compression
gain about .3%.
For cpu-used 2, the change resolved some very disturbing visual
artifacts in certain sequences when large block partitionings and
transforms are used as a result of copying the partition from a
previous frame.
Change-Id: Ic7fd22508cdb811d4ca935655adbf20109286cfa
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
The current decode_tiles decodes the frame one tile by one tile
and then loopfilter the whole frame or use another worker thread to
do loopfiltering.
|------|------|------|------|
|Tile1-|Tile2-|Tile3-|Tile4-|
|------|------|------|------|
For example, if a tile video has one row and four cols, decode_tiles
will decode the Tile1, then Tile2, then Tile3, then Tile4.
And during decode each tile, decode_tile will decode row by row in
each tile.
For frame parallel decoding, decode_tiles will decode video in row order
across the tiles. So the order will be:
"Decode 1st row of Tile1" -> "Decode 1st row of Tile2"
-> "Decode 1st row of Tile3" -> "Decode 1st row of Tile4"
-> "Decode 2nd row of Tile1" -> "Decode 2nd row of Tile2"
-> "Decode 2nd row of Tile3" -> "Decode 2nd row of Tile4"-> "loopfilter 1st row"
Change-Id: I2211f9adc6d142fbf411d491031203cb8a6dbf6b
This commit adjusts the forward 16x16 DCT computation steps to
simplify the register level operations. It fixes the corresponding
sse2 version accordingly.
Change-Id: I72a9c25b8ca9442fc5e113f47cd701ae55aa7f08
Added a skipping test in non-rd inter-mode. After interpolation
prediction step, the residuals are tested to see if they will be
quantized to 0 based on modeling between spatial domain and
frequency domain.
Set static-thresh to 800 for >=720p and 300 for <720p, rtc set
tests showed
1. Speed 5, psnr: -0.514%; ssim: -1.748%;
speedup on related clips: 5% -11%
2. Speed 6, psbr: -0.628%; ssim: -1.637%;
speedup on related clips: 4% - 9%
Change-Id: I62fbf26bc043ecd2b584f255f1a4ee5ab52bfcf3
Make all post-processor code conditionally
compilable based on the CONFIG_VP9_POSTPROC
macro.
Also, remove the vizualization code from VP9
since it is out of date and will not compile.
Change-Id: I1e9e13a09ecd43e9a3f3704c175ae8cd258ababd
vp9_block_error_sse2 can only handle 16 bytes at a time but
the function requires to handle a sequence of 32 bytes at a time
so each 16 bytes is handled in a different register.
With AVX2 optimization the 32 bytes can be handled in one register instead
of two in the SSE2
The vp9_block_error was optimized by 85%.
The user level was optimized by 1.2%
Change-Id: Ia8fffe60e61eff7432a5fbd538757894f6c319fd
