This prevents ill-defined behaviour, such as setting x->skip for a mode
that is excluded because of frame-level flags (e.g. filter selection,
compound prediction selection), then not breaking out of the RD loop
because the mode is not allowed, but keeping the flag on. Whatever mode
is iterated through next in the RD loop will then carry this flag, and
all sort of bad stuff happens, such as x->skip being set on intra pred
modes.
Change-Id: I5bec46b36e38292174acb1c564b3caf00a9b4b9a
Previous commit does not build cleanly on Jenkins with the DWT/DCT
hybrid experiment enabled (--enable-dwtdcthybrid).
Change-Id: Ia67e8f59d17ef2d5200ec6b90dfe6711ed6835a5
Fixes some scaling issues. Adds an option to only compute the
dct on the low-low subband for 32x32 and 64x64 blocks using
only a single 16x16 dct after 1 and 2 wavelet decomposition
levels respectively. Also adds an option to use a 8x8 dct
as building block.
Currenlty with the 2/6 filter and with a single 16x16 dct on
the low low band, the reuslts compared to full 32x32 dct is
as follows:
derf: -0.15%
yt: -0.29%
std-hd: -0.18%
hd: -0.6%
These are my current recommended settings, since the 2/6 filter
is very simple.
Results with 8x8 dct are about 0.3% worse.
Change-Id: I00100cdc96e32deced591985785ef0d06f325e44
and vp9_mb_lpf_vertical_edge_w_sse2. This was quickly done so we can
run some tests over the weekend. Future commits will optimize/refactor these
functions further.
The decoder performance improved by ~17% for the clip used.
Change-Id: I612687cd5a7670ee840a0cbc3c68dc2b84d4af76
Updated the rtcd_defs and used the sse2 uv version
of the loopfilter. The performance improved by ~8%
for the test clip used.
Change-Id: I5a0bca3b6674198d40ca4a77b8cc722ddde79c36
The commit changed to not to use wider lpf within a superblock when
32x32 transform is used for the block.
The commit also changed to use the shorter version of loop filtering:
for UV planes.
Change-Id: I344c1fb9a3be9d1200782a788bcb0b001fedcff8
This patch removes the old pred-filter experiment and replaces it
with one that is implemented using the switchable filter framework.
If the pred-filter experiment is enabled, three interopolation
filters are tested during mode selection; the standard 8-tap
interpolation filter, a sharp 8-tap filter and a (new) 8-tap
smoothing filter.
The 6-tap filter code has been preserved for now and if the
enable-6tap experiment is enabled (in addition to the pred-filter
experiment) the original 6-tap filter replaces the new 8-tap smooth
filter in the switchable mode.
The new experiment applies the prediction filter in cases of a
fractional-pel motion vector. Future patches will apply the filter
where the mv is pel-aligned and also to intra predicted blocks.
Change-Id: I08e8cba978f2bbf3019f8413f376b8e2cd85eba4
This is to add to the 64x64 transform experiment as an alternative to
a 64x64 DCT.
Two levels of wavelet decomposition is used on a 64x64 block, followed
by 16x16 DCT on the four lowest subbands. The highest three subbands
are left untransformed after the first level DWT.
Change-Id: I3d48d5800468d655191933894df6b46e15adca56
This is to fix a decoder crash when decoder skips a number of frame to
continue decoding from a later key frame.
Change-Id: I3ba116eba6c3440e0528a21f53745f694302e4ad
This commit did a couple of minor cleanup/refactoring to prepare for
futher loop filter experiments. It merged y_only version of loop filter
function into the regular one, which makes sure that same logic is used
for functions for picking level and for actual loop filtering.
Change-Id: Id10c94dccd45f58e5310bacfdf6ee63cbb60b86f
Mode selection for SBs could enter an infinite loop because
the interpolation filter mode index was not being reset
correctly.
Change-Id: I4bbe726f29ef5b6836e94884067c46084713cc11
This experimental change reorders the search so
that all possible references that match the target
reference frame are tested first and these in order
of distance from the current block. These will usually
be the highest scoring candidates.
If we do not find enough good candidates this way
we try non matching cases. These will usually be lower
scoring candidates.
The change in order together with breakouts when
we have found enough candidates should reduce
the computational cost and especially reduce the number
of sort operations.
Quality Results:
Std Hd +0.228%, Hd +0.074%, YT +0.046%, derf +0.137%
This effect is probably due to the fact that more distant
weak candidates are now less likely to get "promoted" over
near candidates even if they are repeated.
Change-Id: Iec37e77d88a48ad0ee1f315b14327a95d63f81f6
Old Scheme:
When SWITCHABLE filter selection is enabled the encoder
evaluates the use of each interpolation filter type and
selects the best one to use at the MB level. A frame-
level flag can be set to force the use of a particular
filter type for all MBs in a frame if it is more efficient
to encode that way. The logic here involved a Q dependent
threshold that assumed that the second 8-tap filter was
a high-pass filter. However, this requires a trip around
the recode loop. If the frame-level flag indicates use
of a particular filter, the other filters are not
evaluated in the pick_mode loop.
New Scheme:
Each filter type is evaluated at the MB level and a record
of the best filter is kept, irrespective of what filter
is signaled at the frame-level. Once all MBs have been
encoded, a decision is made as to what frame-level mode
to set for the *next* frame. If one filter is used by 80%
or more of the MBs, then this filter is forced since it
is assumed that this will be more efficient if the
next frame has similar characteristics. i.e. there is a
one-frame lag between measuring the filter selection and
setting the frame-level mode to use.
Change-Id: I6a7e7ced8f27e120fafb99db2dc9c6293f8d20f7
Remove special case function cost_coeffs_2x2() and change function
cost_coeffs() to handle 2nd order haar block as it is handle all
other block types already.
Change-Id: I2aac6f81ee0ae9e03d6a8da4f8681d69b79ce41f
Read mode before calling vp9_find_best_ref_mvs(). If the mode is
ZEROMV, the best ref_mvs are not needed. Then, we can skip calling
vp9_find_best_ref_mvs().
Change-Id: I5baa3658dd3f1c7107211cbbbcf919b4584be2e2
The 2-D inverse transform X = M1*Z*Transposed_M2 was calculated
in 2 steps from left to right:
1. Vertical transform: Y = M1*Z
2. Horizontal transform: X= Y*Transposed_M2
In SIMD, a transpose is needed in vertical transform.
Here, switched the calculation order to do it from right to left.
In this way, we could eliminate that transpose by writing the
intermediate results out to their transposed positions.
Change-Id: I34dfe5eb01292f6e363712420d99475e2e81e12c
Various fixups to resolve issues when building vp9-preview under the more stringent
checks placed on the experimental branch.
Change-Id: I21749de83552e1e75c799003f849e6a0f1a35b07
Adds an experiment to derive the previous context of a coefficient
not just from the previous coefficient in the scan order but from a
combination of several neighboring coefficients previously encountered
in scan order. A precomputed table of neighbors for each location
for each scan type and block size is used. Currently 5 neighbors are
used.
Results are about 0.2% positive using a strategy where the max coef
magnitude from the 5 neigbors is used to derive the context.
Change-Id: Ie708b54d8e1898af742846ce2d1e2b0d89fd4ad5
This patch changes the token packing to call the bool encoder API rather
than inlining it into the token packing function, and similarly removes
a special get_signed case from the detokenizer. This allows easier
experimentation with changing the bool coder as a whole.
Change-Id: I52c3625bbe4960b68cfb873b0e39ade0c82f9e91
For coefficients, use int16_t (instead of short); for pixel values in
16-bit intermediates, use uint16_t (instead of unsigned short); for all
others, use uint8_t (instead of unsigned char).
Change-Id: I3619cd9abf106c3742eccc2e2f5e89a62774f7da
In some cases intra modes in inter frames give
an over smoothed appearance. Especially with
noisy but flat content.
Also in some cases there were problems with key
frame sizing again with very flat but noisy content.
These are temporary changes to help alleviate the
visual problems but will almost certainly hurt metric
results especially at the very low data rate end.
Change-Id: I11549179a19277ffc283d9788bc70168f2a8bdc9
This is to fix a decoder crash when decoder skips a number of frame to
continue decoding from a later key frame.
Change-Id: I3ba116eba6c3440e0528a21f53745f694302e4ad
MSVC 2012 (_MSC_VER=1600) introduced the definition, this commit
prevents the redefinition of the macro
Change-Id: I7de92e7e9e865a342f2bcc4b071f8d3c9b2a508c
The mismatch was caused by an improper merge of cleanup code around
tokenize_b() and stuff_b() with TX32X32 experiment.
Change-Id: I225ae62f015983751f017386548d9c988c30664c
Modifies the scanning pattern and uses a floating point 16x16
dct implementation for now to handle scaling better.
Also experiments are in progress with 2/6 and 9/7 wavelets.
Results have improved to within ~0.25% of 32x32 dct for std-hd
and about 0.03% for derf. This difference can probably be bridged by
re-optimizing the entropy stats for these transforms. Currently
the stats used are common between 32x32 dct and dwt/dct.
Experiments are in progress with various scan pattern - wavelet
combinations.
Ideally the subbands should be tokenized separately, and an
experiment will be condcuted next on that.
Change-Id: Ia9cbfc2d63cb7a47e562b2cd9341caf962bcc110
Gives 0.5-0.6% improvement on derf and stdhd, and 1.1% on hd. The
old tables basically derive from times that we had only 4x4 or
only 4x4 and 8x8 DCTs.
Note that some values are filled with 128, because e.g. ADST ever
only occurs as Y-with-DC, as does 32x32; 16x16 ever only occurs
as Y-with-DC or as UV (as complement of 32x32 Y); and 8x8 Y2 ever
only has 4 coefficients max. If preferred, I can add values of
other tables in their place (e.g. use 4x4 2nd order high-frequency
probabilities for 8x8 2nd order), so that they make at least some
sense if we ever implement a larger 2nd order transform for the
8x8 DCT (etc.), please let me know
Change-Id: I917db356f2aff8865f528eb873c56ef43aa5ce22
As suggested by Yaowu, we can use eob to reduce the complexity
of the vp9_ihtllm_c function. For the 1080p test clip used, the decoder
performance improved by 17%.
Change-Id: I32486f2f06f9b8f60467d2a574209aa3a3daa435
Add a function clip_pixel() to clip a pixel value to the [0,255] range
of allowed values, and use this where-ever appropriate (e.g. prediction,
reconstruction). Likewise, consistently use the recently added function
clip_prob(), which calculates a binary probability in the [1,255] range.
If possible, try to use get_prob() or its sister get_binary_prob() to
calculate binary probabilities, for consistency.
Since in some places, this means that binary probability calculations
are changed (we use {255,256}*count0/(total) in a range of places,
and all of these are now changed to use 256*count0+(total>>1)/total),
this changes the encoding result, so this patch warrants some extensive
testing.
Change-Id: Ibeeff8d886496839b8e0c0ace9ccc552351f7628
Some further changes and refactoring of mv
reference code and selection of center point for
searches. Mainly relates to not passing so many
different local copies of things around.
Some place holder comments.
Change-Id: I309f10ffe9a9cde7663e7eae19eb594371c8d055
This commit changed the ENTROPY_CONTEXT conversion between MBs that
have different transform sizes.
In additioin, this commit also did a number of cleanup/bug fix:
1. removed duplicate function vp9_fix_contexts() and changed to use
vp8_reset_mb_token_contexts() for both encoder and decoder
2. fixed a bug in stuff_mb_16x16 where wrong context was used for
the UV.
3. changed reset all context to 0 if a MB is skipped to simplify the
logic.
Change-Id: I7bc57a5fb6dbf1f85eac1543daaeb3a61633275c
Don't use vp9_decode_coefs_4x4() for 2nd order DC or luma blocks. The
code introduces some overhead which is unnecessary for these cases.
Also, remove variable declarations that are only used once, remove
magic offsets into the coefficient buffer (use xd->block[i].qcoeff
instead of xd->qcoeff + magic_offset), and fix a few Google Style
Guide violations.
Change-Id: I0ae653fd80ca7f1e4bccd87ecef95ddfff8f28b4
Use these, instead of the 4/5-dimensional arrays, to hold statistics,
counts, accumulations and probabilities for coefficient tokens. This
commit also re-allows ENTROPY_STATS to compile.
Change-Id: If441ffac936f52a3af91d8f2922ea8a0ceabdaa5
This adds Debargha's DCT/DWT hybrid and a regular 32x32 DCT, and adds
code all over the place to wrap that in the bitstream/encoder/decoder/RD.
Some implementation notes (these probably need careful review):
- token range is extended by 1 bit, since the value range out of this
transform is [-16384,16383].
- the coefficients coming out of the FDCT are manually scaled back by
1 bit, or else they won't fit in int16_t (they are 17 bits). Because
of this, the RD error scoring does not right-shift the MSE score by
two (unlike for 4x4/8x8/16x16).
- to compensate for this loss in precision, the quantizer is halved
also. This is currently a little hacky.
- FDCT and IDCT is double-only right now. Needs a fixed-point impl.
- There are no default probabilities for the 32x32 transform yet; I'm
simply using the 16x16 luma ones. A future commit will add newly
generated probabilities for all transforms.
- No ADST version. I don't think we'll add one for this level; if an
ADST is desired, transform-size selection can scale back to 16x16
or lower, and use an ADST at that level.
Additional notes specific to Debargha's DWT/DCT hybrid:
- coefficient scale is different for the top/left 16x16 (DCT-over-DWT)
block than for the rest (DWT pixel differences) of the block. Therefore,
RD error scoring isn't easily scalable between coefficient and pixel
domain. Thus, unfortunately, we need to compute the RD distortion in
the pixel domain until we figure out how to scale these appropriately.
Change-Id: I00386f20f35d7fabb19aba94c8162f8aee64ef2b
Only declare the functions in vpx_scale RTCD and include the relevant
header.
Remove unused files and functions in vpx_scale to avoid wasting time
renaming. vpx_scale/win32/scaleopt.c contains functions which have not
been called in a long time but are potentially optimized.
The 'vp8' functions have not been renamed yet. That is for after the
cleanup.
Change-Id: I2c325a101d60fa9d27e7dfcd5b52a864b4a1e09c
This patch reduces the cpu cost of the MV ref
search by only allowing insert for candidates
that would be in the current top 4.
This could alter the outcome and slightly favors
near candidates which are tested first but also
limits the worst case loop count to 4 and means in
many cases it will drop out and not happen.
Change-Id: Idd795a825f9fd681f30f4fcd550c34c38939e113
1. remove the dependency on non existing "vp9_temporal_filter_x86.h"
2. prefix filenames with vp9_ in obj_int_extract.bat to reflect the
change of the actual filenames.
Change-Id: Ib1b4d96ac41788f76917764a6722d8461c857302
The commit changed the trellis quantization function optimize_b() to
work for MBs using all transform sizes, and eliminated the function
for MB using 16x16 transform only, optimize_b_16x16.
Change-Id: I3fa650587ab5198ed16315b38754783a72b33ba2
Only declare the functions in vpx_scale RTCD and include the relevant
header.
Remove unused files and functions in vpx_scale to avoid wasting time
renaming. vpx_scale/win32/scaleopt.c contains functions which have not
been called in a long time but are potentially optimized.
The 'vp8' functions have not been renamed yet. That is for after the
cleanup.
Change-Id: I2c325a101d60fa9d27e7dfcd5b52a864b4a1e09c
- vpx_calloc is called on arf_not_zz above.
- Note The removed vpx_memset call had an issue with sizeof.
Change-Id: I86fd7a167d0a042e581e613e2a6c0b5e63073fc6
Adds support for compound inter-intra prediction with superblocks.
Also, fixes a bug that disabled intra modes for superblocks.
Change-Id: I4d711317e1bc19df8c2f32dc645429f7fff31036
Allows switchbale filters to be used without mismatch when the
superblock experiment is on.
Also removes a spurious clamping code in decodemv.c which causes
rare encode/decode mismatches.
Change-Id: I809d9ee0b2859552b613500b539a615515b863ae
Previously, the "!=" check is logically incorrect when eob is at 0 and
effective coefficient starting position is 1. This commit should have
no effect on bitstream.
Change-Id: I6ce3a847c7e72bfbe4f7c74f88e3310c6b9b6d30
This patch allows use of 8x8 and 4x4 ADST correctly for Intra
16x16 modes and Intra 8x8 modes when the block size selected
is smaller than the prediction mode. Also includes some cleanups
and refactoring.
Rebase.
Change-Id: Ie3257bdf07bdb9c6e9476915e3a80183c8fa005a
This commit makes sure Y2 entropy coding context is always updated on
every macroblock even there is no Y2 block.
Change-Id: Ie307cfc46526efe55613be39f9f178d2531b56ba
This change included:
1. Aligned reads in vp9_mbloop_filter_vertical_edge function.
Since we actually read 16 bytes, we can align the reads to read
starting at (s - 8) instead of (s - 5).
2. Combined u, v loop filters.
3. Added 8x16 transpose.
This gave 2% decoder performance gain (tulip clip).
Change-Id: Ib14c2f1645c4a3436df17fe2f24789506bf0bb58
This commit removed a couple of redundant data structures in frame
coding contextsm, mode_context and mode_context_a, and changed to
use vp9_mode_contexts only. The switch of the context for different
frame type now relies on the switch of frame coding context between
lfc and lfc_a. This commit also removed a number of memcpy among
these redundant data structure.
Change-Id: I42e8174bd60f466b0860afc44c1263896471b0f3
Not all segment feature data elements are full-range powers of two, so
there are values that can be encoded that are invalid. Add a new function
to clamp values to the maximum allowed.
Change-Id: Ie47cb80ef2d54292e6b8db9f699c57214a915bc4
Support for gyp which doesn't support multiple objects in the same
static library having the same basename.
Change-Id: Ib947eefbaf68f8b177a796d23f875ccdfa6bc9dc
Modified the mv_pred() fuunction that chooses a centre
point from which to start step searches to use the top
candidate vectors chosen previously.
Some gains (mainly on HD and tested with SB off).
Std_hd 0.874%, YT-hd 0.174%, YT 0.05%, Derf 0.036%
Change-Id: Ie232284f561838b8ecee0e28dcbb07a9cd46cf56
Vp9_sad3x16_sse2() is heavily called in decoder, in which the
unaligned reads consume lots of cpu cycles. When CONFIG_SUBPELREFMV
is off, the unaligned offset is 1. In this situation,
we can adjust the src_ptr to be 4-byte aligned, and then do the
aligned reads. This reduced the reading time significantly. Tests
on 1080p clip showed over 2% decoder performance gain with
CONFIG_SUBPELREFM off.
Change-Id: I953afe3ac5406107933ef49d0b695eafba9a6507
Add a new function vp9_decode_mb_tokens() that handles the switch
between different per-tx-size detokenize functions. Make actual
implementations (vp9_decode_mb_tokens_NxN()) static.
Change-Id: I9e0c4ef410bfa90128a02b472c079a955776816d
Don't declare variables if they only ever have a single value and are
used only as argument to another function call; instead, just hardcode
the value in the function call directly. Split out UV and Y coefficient
loops for clarity. Use xd->block[].qcoeff instead of xd->qcoeff + magic
to remove use of magic offset variables.
Change-Id: I5b17eda1bb666c69c2b7ea957d5525cd78192e33
Don't declare variables if they only ever have a single value and are
used only as argument to another function call; instead, just hardcode
the value in the function call directly. Also remove unneeded brackets
around a code block, and remove the magic offsets 64 and 256 for chroma
values in the coefficient memory block.
Change-Id: I14fc14120a81ea1d6fb862674e8bf8cf6ba3d114
This prevents the relatively expensive token-from-coefficient lookup
function get_token(), plus a duplicate loop..
Change-Id: Ibecd407b2a91d3593d439ec4646e43fa26d2ff91
Just like for all other block modes, b_pred tokens can be read together
before starting macroblock reconstruction. This removes special cases
for b_pred in decode_macroblock() and allows to make decode_coefs_4x4()
static in detokenize.c.
While at it, remove the redundant handling and checking of plane_type
and block_index (i) in decode_coefs_4x4(). Since the function is static,
and is called only from decode_mb_tokens_4x4(), we don't need to worry
that the arguments ever go out of sync.
Change-Id: I2d415da0b51b89d0490a6b9e24cc86363c2090f7
Experiments with a larger set of contexts and some
clean up to replace magic numbers regarding the
number of contexts.
The starting values and rate of backwards adaption
are still suspect and based on a small set of tests.
Added forwards adjustment of probabilities.
The net result of adding the new context and forward
update is small compared to the old context from the
legacy find_near function. (down a little on derf but
up by a similar amount for HD)
HOWEVER.... with the new context and forward update
the impact of disabling the reverse update (which may be
necessary in some use cases to facilitate parallel decoding)
is hugely reduced.
For the old context without forward update, the impact of
turning off reverse update (Experiment was with SB off) was
Derf - 0.9, Yt -1.89, ythd -2.75 and sthd -8.35. The impact was
mainly at low data rates.
With the new context and forward update enabled the impact
for all the test sets was no more than 0.5-1% (again most at
the low end).
Change-Id: Ic751b414c8ce7f7f3ebc6f19a741d774d2b4b556
