GitLab

Support the use of 2x2 prediction and transform block size for
chroma components. This experiment builds on top of cb4x4. It
brings back the loss in performance gains due to 4x4 chroma
component constrain. The coding gains for lowres are up from 1.5%
to 2.5%.

Change-Id: I8e461e230c2d920f6df1af9d935ebe9dfa3820d9

Since gm_get_motion_vector is trying to give a motion vector for
"the frame as a whole", it makes more sense for it to calculate
the motion of the *center* of the block rather than the top-left
corner of the frame.

In theory, this change should also help the encoder make better
decisions on when to use global motion. It avoids an issue
where, early in the frame, NEARESTMV looks like a good way to use
the global motion vector *without* paying the rate cost applied
to the first few global motion blocks in each frame. This seems
to lead to a better overall result.

Change-Id: Ia5c6259ceb8b4ff3d00a5d553e1d18bdb802da59

Resolve a rare enc/dec mismatch issue triggered by the 4x4 chroma
block size in sub8x8 block coding.

To avoid using 2x2 chroma block size, it always uses the top-left
4x4 luma block coding information to predict the 4x4 chroma block.
The rest 3 4x4 luma blocks will be coded independently without
coding the collocated chroma blocks.

The compression performance gains are 1.45% for lowres set (down
from 2.4% of the original cb4x4 design).

Change-Id: Iff560fcab172ed4219434d5174c4d8dfbbb04135

Change-Id: I288890b964fda354ac58939f3a6b54a33f40666d

This CL aim at facilitating hardware implementation
The check will be turned on once we have faster motion vector
coding system

Performance drop
lowres 0.157%
midres 0.260%

Change-Id: Iedc961f44efadfe84ff9edab408f44845bffcfe9

If there no sub-pixel in both direction of MV,
the interp_filter is set to EIGHTTAP_REGULAR

Change-Id: I208c34906544f1937c57d78017999be6eddd9e87

Fixes that pack_palette_tokens() was disabled when pvq is enabled and
av1_decode_palette_tokens() was not complied.

Change-Id: I468bdc4d77ae7ddbca7fb5ed8e1da5506486a58c

Change-Id: If5d4f6d261276d9cfdd59fd7779467d92f56c395

Fix the build error when pvq and pallete experiments are both enabled.

Change-Id: I80839b0660456b81aad00ad8dd9c34b7541f7727

Change-Id: I1d599b70fb338de4b0d2076fdc7060361a46c21f

Properly scale the chroma component block size to handle the sub8x8
coding block in the encoding stage. It resolves the large performance
drop issue in cb4x4 due to minimum 4x4 block size.

Change-Id: Ia2477b43521408b72a20aca406f495faba66c033

Skip coding the intra mode for chroma component for sub8x8 blocks
that are not in the reference position.

Change-Id: Ic3daab2668ab41c3b743664faf195e34124ead33

Make the rate-distortion optimization account for the special order
for sub8x8 block chroma component coding.

Change-Id: I050abc50dda7b9e8e32f7d0959aa954df1d69398

Fix the intra prediction boundary availability check for chroma
components. Such intricacy is due to the re-schedule of sub8x8
chroma coding block to avoid 2x2 prediction and transform size.

Change-Id: I4a2dff8743cf78af8dfb90265019e8d1222ed734

Skip the 2x2 uv block rate-distortion optimization process in the
non-reference positions in key frame encoding.

Change-Id: I51112492fd79f9a1c205fba151f8658cea9ee85f

Process sub8x8 chroma component decoding in the unit of 4x4 block.

Change-Id: I02e149f6076cdb5942b8a0a0b9ac2d833b1a7e49

Change-Id: I5bc76c3edbe0792a85c995721ee93dcf517cc1ff

Change-Id: I05377f3d06b3a863e0372d100bc78ea1fbee9b82

Previously the transform type sets used for 16x32/32x16 transform
sizes corresponded to 16x16 transforms. Now they correspond to
32x32 transforms with a reduced available set.

BDRATE impact:
0.05% worse for lowres, 0.13% worse for midres

Change-Id: I01d405d497c13837f5404f9e0caf8c571f3b26cf

increases the translational part of global motion models
to the same 1/64 pel precision which is used by the warp filter.

Change-Id: I7267b18b123662416bc5c8095186b6a5eef2f7f1

The full-pixel checking is conducted in update_filter_type_count
Change-Id: I583ec4b1f3ac29d124098819aee257a8f1007ed4

Removing av1_is_interp_needed() function from it.
Use has_subpel_mv_component() only.

Change-Id: I98b674b28c276a09bb94f24d66360d24377066a8

Change-Id: I2c05ce6a96cc21826fc53aeacce91d9637d3b178

Change-Id: Ie7edfbcdd7f5f2cc3aa3aa6c2575d081616da6b6

Here we create a default set for the ext-tx experiment which
includes:

Intra: {DCT, ADST} X {DCT, ADST} + IDTX (identity in both directions)
Inter: DCT_DCT + IDTX

This set is used if indicated in the uncompressed header.

Change-Id: Ic907ada12b4e0c98cabb7c9650ab99942d0c0859

By rearranging the order of operations, we can ensure that all
intermediate values fit into 32 bits. This will help when we
vectorize the self-guided filter.

Results in the noise range.

Change-Id: Ic0c73613882bd103c4e8e57a0155b3132672ae04

Remove code related to convolve_round experiment out of
av1_convolve()

The experimental code is now implemented in av1_convolve_2d().

Change-Id: I409a390876e9d19e6ee42150f0ac20ff108244a9

Some of the conditions in 'else if' blocks were redundant. Getting
rid of them should improve speed.

Change-Id: Ib5a994c54c4913ca68c7d982abddc85c0b36db0b

Signed-integer overflow is undefined behaviour in C.

Change-Id: Ib2d7f405adacc668264a5d77aba00ca85ecbb0f7

De-duplicate implementations of 'range_check', and use precomputed pixel value
boundaries instead of computing the log2 of each pixel value.
Overall speedup: encoding with enabled-range checks goes 3.3 times
faster.

Change-Id: I2f7c97cac4ad479b07898fb69c0131e3c7e0bf98

Use the top-left 4x4 luma block's coding information for chroma
component encoding at size 4x4.

Change-Id: I4bcfbc2bf8b71f7fc30094553503468460a56f9b

The result is identical to enabling both deringing and CLPF

Change-Id: I71db5ba9e21fcaf11ad87e94841eaf80be58c0a8

The ATTRIBUTE_PACKED caused something to break in MB_MODE_INFO.
boundary_info got corrupted and moving InterpFilter around in
MB_MODE_INFO could cause a segfault or the problems to go away (for
instance by moving InterpFilter to the end of the struct).  Perhaps
something in MB_MODE_INFO is sensitive to alignment.  This should be
investigated further and get fixed or documented, but for now remove
the attribute.

Change-Id: Ic4326ab98fe784d9b7d2b47d8f086b8429cedee9

Change-Id: I3fa9d24f2a61dd1b4a234fb1cfea679af06ebdb5

Remove redundant table access.

Change-Id: Ib45f9b0bafe2a0c57449e787d60e70d35f19f6c7

When run with --enable-ec_adapt, forward updates to the tx_size_prob
 array will not be made.

objective-1-fast:

master-ec_adapt-e568e6f0@2017-02-17T15:30:27.109Z ->
 tx_size-ec_adapt@2017-02-17T15:31:49.504Z

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-0.0113 | -0.0112 | -0.0113 |  -0.0114 | -0.0114 | -0.0114 |    -0.0112

master-ec_adapt-robust-e568e6f0@2017-02-17T20:29:45.315Z ->
 tx_size-ec_adapt-robust@2017-02-17T20:33:06.405Z

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-0.0244 | -0.0243 | -0.0243 |  -0.0245 | -0.0245 | -0.0245 |    -0.0243

Change-Id: I5873c5d14e3f3868997775f85fa743d2d55d81b7

When ext-intra or filter-intra is enabled, the transform type was inferred from
prediction mode as in VP9. Should use mbmi->tx_type instead.

Change-Id: I6a8e356285a8461d67e69d38df40541d34bade04

We use a single enum instead of multiple #defines.
- Ensures better type checking
- Enum values are generated implicitly, and hard-coded #defines are not
required.
- We use ATTRIBUTE_PACKED to indicate that the enum should still use the
smallest integral type.

Change-Id: I7532428da31744d3441b363bd932a7f233ee7ab5

Replace obselete size "4" with correct one "tile_size_bytes". This
fixes a decoder crash.

BUG=aomedia:352

Change-Id: I5b1c606a4d5be946acc68b3cfe0d0603a5591c44

Replaces division with multiplication in self-guided
filter.

The guided filter requires computation of:
n^2.s^2/(n^2.s^2 + n^2.e).
This is now implemented by computation of n^2.s^2/n^2.e followed
by using a lookup table for the function f(x) = x/(x+1).
To compute n^2.s^2/n^2.e, we use an integer multiplication based
implementation which becomes feasible since n^2.e can only
take a few values and their corresponding multipliers can be
pre-computed.
There is also another divison by n, that is also integerized.

Change-Id: Id7b81bbafead0b8f04a1853ec69b9dec423bb66a