Add a generic bound for u8 and u16 and convert `pred_dc`, `pred_dc_128`,
`pred_dc_left`, `pred_dc_top`, `pred_h`, `pred_v` and `pred_paeth`.

* Correctly handle NEAR mode when less than 2 entried in MV stack

If there are fewer than 2 entries in the motion vector stack, then
the NEAR motion vector is considered to be the same vector as the
global motion vector (currently 0,0), as per 7.10.2.12
"Extra search process" in the spec.

* Add support for 2nd and 3rd NEAR motion vector

* Adaptively select best MV mode

* CPU feature check for SSSE3.
* Measure rdo_cfl_alpha() speed.
* Remove redundant loop for 2x speed-up of rdo_cfl_alpha().
* Allow rdo_cfl_alpha() to return None.
* Tidy up formatting.

Move MV stack construction out of encoding function such that the
value of the nearest MV can be determined before encoding

This yields an ~4x speed-up of this function.
Update sameness test with in-range values.
Return early for zero alpha.
Elide bounds checks from SSSE3 implementation
Bounds are enforced in the safe wrapper.

* Add the chroma-from-luma predictor and a test
* Add benchmarks for CfL predictor
* Implement CfL signalling
* Plumb from predict_intra to pred_cfl
* Add struct for CfL parameters
* Compute subsampled luma AC and plumb it with CfL alpha
* Plumb CfL parameters up to RDO
* Add CfL to RDO loop, no parameter search

* Add motion vector paramater to inter prediction function

Motion vector is read from added data field in Block structure

* Add reference frame and motion vector fields/parameters

Reference frame and motion vector are added fields to the
RDOOutput structure and added parameters to the block
encoding function

* Fix inter prediction function for nonzero MVs

* Add call to encode new motion vector

* Import default cdfs for MV coding from libaom

* Fix handling of MV precision in MV coding function

* Add coding of DRL mode

* Add motion vector stack

* Add new MV counter to correctly determine context

* Use motion vectors in MV stack for prediction

* Sort MV stack according to weights

Also update weights after near search

* Fix log2() function

* Compute correct context for DRL flag

* Store MVs with 1/8 pel precision and fix MV generation

* Add some basic motion estimation

Motion estimation is full search (+/-16 pel range in 2-pel increments) and
is done independently for each block

* Fix chroma motion compensation for small blocks

Chroma motion compensation sometimes happens on 2x2 basis

Frank Bossen authored 6 years ago and Yushin Cho committed 6 years ago

* Search more intra modes in inter frames

Reuse exisiting intra mode lists to determine which intra modes
to search in an inter frame

* Remove duplicate line

Frank Bossen authored 6 years ago and Thomas Daede committed 6 years ago

* WIP motion vector reference search

* More searches

* Have a version that compiles

* Actually call MV prediction function

Fix a few bugs along the way

* Add computation of context for inter mode

* Have find_mvrefs return a mode_context value and use it

* Test but never select intra mode in inter frame

* Fixes to context derivation for inter mode

Some success with mixing intra and inter in a same frame but
still buggy

* Fixes to find MV references functions

Can encode/decode 20 frames at speed 4, but not at slower speed
settings

* Fix MV scan

Now intra/inter mixing works for all speed settings

* Fix issues that arose from rebase

* Remove lines that should not have been added

* Remove superfluous empty lines

* Supress compiler warnings

* A few cosmetic improvements

Raphaël Zumer authored 6 years ago and Thomas Daede committed 6 years ago

* Attempt to process 10-bit Y4M input

Currently compiles, encodes and decodes with desynchronization

* Use high bit depth quantization tables

* Move context::clamp() to util.rs

* Fix partition context initialization for high bit depth

* Enable and test 10-bit input

* Add 10- and 12-bit test clips to build.sh

Commented out by default, to enable as needed for local testing.

* Use the same bit depth as y4m_dec for y4m_enc

* Fix benchmark module compilation

* Fix high bit depth test encoding in 8-bit

* Fix header syntax for 12-bit 4:2:0 input

* Enable and test 12-bit input

* Reflect 12-bit support in README.md

* Keep the default C420jpeg color space in y4m_encoder

* Do not allow reconstruction output at high bit depths

* Fix reconstruction output at high bit depths

* Clean up reconstruction copy to frame buffers

Yushin Cho authored 6 years ago and Monty Montgomery committed 6 years ago

This reverts commit 8c545350.

Raphaël Zumer authored 6 years ago and Thomas Daede committed 6 years ago

* Attempt to process 10-bit Y4M input

Currently compiles, encodes and decodes with desynchronization

* Use high bit depth quantization tables

* Move context::clamp() to util.rs

* Fix partition context initialization for high bit depth

* Enable and test 10-bit input

* Add 10- and 12-bit test clips to build.sh

Commented out by default, to enable as needed for local testing.

* Use the same bit depth as y4m_dec for y4m_enc

* Fix benchmark module compilation

* Fix high bit depth test encoding in 8-bit

* Fix header syntax for 12-bit 4:2:0 input

* Enable and test 12-bit input

* Reflect 12-bit support in README.md

* Create proper structure for reference frame buffer

Also move reference frame buffer into frame invariants to
avoid having to pass an additional parameter to each function
requiring access to reference frame buffer.

* Add basic inter prediction function

Basic inter prediction function copies collocated samples
from first frame in reference frame set.

* Add mapping from reference frame index to slot in buffer

Add array that maps a reference frame index such as LAST_FRAME
into a slot of the reference frame buffer

* Use GLOBALMV instead of ZEROMV

For consistency with text specification

* Add support for writing reference frames in inter mode

* Add write_inter_mode() for NEWMV and GLOBALMV inter pred modes

Also,
- Rename ZEROMV prediction mode to GLOBALMV
- RDO search include GLOBALMV mode if inter frame
- Import CDFs required to encode NEWMV or GLOBALMV modes
- Add write_tx_tree() for tx block coding for a inter mode. This
function is also for var-tx, but currently only one tx block for a inter
mode partition since tx_mode in frame header is set as LARGEST_TX atm.

* Fix ref_frame storage and counting

* Add basic inter prediction function

Basic inter prediction function copies collocated samples
from first frame in reference frame set.

* Add mapping from reference frame index to slot in buffer

Add array that maps a reference frame index such as LAST_FRAME
into a slot of the reference frame buffer

* Use GLOBALMV instead of ZEROMV

For consistency with text specification

* Fix entropy coding issues for inter modes

* Fix issue with transform type and small blocks

Thomas Daede authored 6 years ago and Yushin Cho committed 6 years ago

This reverts commit a78777c6.

Frank Bossen authored 6 years ago and Thomas Daede committed 6 years ago

* Create proper structure for reference frame buffer

Also move reference frame buffer into frame invariants to
avoid having to pass an additional parameter to each function
requiring access to reference frame buffer.

* Add basic inter prediction function

Basic inter prediction function copies collocated samples
from first frame in reference frame set.

* Add mapping from reference frame index to slot in buffer

Add array that maps a reference frame index such as LAST_FRAME
into a slot of the reference frame buffer

* Use GLOBALMV instead of ZEROMV

For consistency with text specification

* Add support for writing reference frames in inter mode

* Add write_inter_mode() for NEWMV and GLOBALMV inter pred modes

Also,
- Rename ZEROMV prediction mode to GLOBALMV
- RDO search include GLOBALMV mode if inter frame
- Import CDFs required to encode NEWMV or GLOBALMV modes
- Add write_tx_tree() for tx block coding for a inter mode. This
function is also for var-tx, but currently only one tx block for a inter
mode partition since tx_mode in frame header is set as LARGEST_TX atm.

* Fix ref_frame storage and counting

* Add basic inter prediction function

Basic inter prediction function copies collocated samples
from first frame in reference frame set.

* Add mapping from reference frame index to slot in buffer

Add array that maps a reference frame index such as LAST_FRAME
into a slot of the reference frame buffer

* Use GLOBALMV instead of ZEROMV

For consistency with text specification

* Fix entropy coding issues for inter modes

Raphaël Zumer authored 6 years ago and Thomas Daede committed 6 years ago

Run with --features=comparative_bench
Applies to intra prediction tests and disables native function inlining
Also refactor benchmarks and reduce code duplication

Yushin Cho authored 6 years ago and Thomas Daede committed 6 years ago

This affects all intra prediction modes in rav1e except DC_PRED.

Michael Bebenita authored 6 years ago and Thomas Daede committed 6 years ago

Format.

Raphaël Zumer authored 6 years ago and Thomas Daede committed 6 years ago

* Restrict transform type RDO to speed levels 0-3

* Restrict prediction mode RDO above low speed levels

nopaeth@2018-06-21T09:29:47.514Z -> paeth@2018-06-21T09:28:52.893Z

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-0.0962 | -0.4074 | -0.3002 |   0.0372 | -0.1525 | -0.0627 |    -0.0307

* Clean the build on submodule updates

* Clean up and refactor RDO functions and data

Move RDO decision functions to rdo.rs, extract rd_cost computation function, move some data to appropriate locations.

* Remove redundant variable in partitioning decision

* Simplify top-down partitioning decision

Raphaël Zumer authored 6 years ago and Thomas Daede committed 6 years ago

Update the year and add a header to files where it was missing.

Raphaël Zumer authored 6 years ago and Thomas Daede committed 6 years ago

This also extracts the unsafe top-left pixel access from the native predictor function.

smooth@2018-06-15T19:42:21.397Z -> paeth@2018-06-16T00:12:39.612Z

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-2.2074 | -4.3757 | -5.0621 |  -0.9361 | -0.3539 | -0.1071 |    -2.5839

The most notable cause was level map always use the old
inter mode scan order as default.

Raphaël Zumer authored 6 years ago and Thomas Daede committed 6 years ago

* Add RDO decision for partitioning

master (-s=2) -> current (-s=2)

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-7.4095 |     N/A |     N/A |  -6.8990 | -9.0320 | -8.7738 |   -17.4584

master (-s=3) -> current (-s=2)

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-5.1051 |  9.5605 |     N/A |  -5.1176 | -5.3494 | -5.1293 |     1.4963

master (-s=3) -> current (-s=0)

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-7.7416 |     N/A |     N/A |  -7.1631 | -7.8223 | -7.2918 |     3.6073

* Increase RD cost comparison precision

master (-s=3) -> current (-s=0)

   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-7.5800 |     N/A |     N/A |  -7.0594 | -7.8593 | -7.3970 |     4.0382

* Avoid creating a new struct for child modes in partition decisions

* Add comments in encode_partition()

Raphaël Zumer authored 6 years ago and Thomas Daede committed 6 years ago

* Add Paeth predictor and tests

* Update submodule hash

* Fix native Paeth prediction function name in bench