Monty authored Sep 20, 2018 and Monty Montgomery committed Sep 22, 2018

No reason for the top level deblocking to be chunking things by SB in
the final filter.  There are fewer checks and corner-cases needed to
simply make full-frame passes.

This removes dependency on libaom for most of the CDFs.

Monty authored Sep 20, 2018 and Monty Montgomery committed Sep 20, 2018

Fix a subtraction that accidentally preceeded its guard statement.

Monty authored Sep 19, 2018 and Monty Montgomery committed Sep 19, 2018

Get the V and H passes closer together in memory to further reduce
memory/cache pressure during deblocking.  This is also a 'get the
pattern right' setup for the heavier RDO.

Monty authored Sep 18, 2018 and Monty Montgomery committed Sep 19, 2018

Testing indicates that the sharpness parameter has negligable effect,
and almost all of it is negative.  Increasing the value from zero
causes regression on all metrics.  As a result, nail it to 0 like in
libaom.

Monty authored Sep 18, 2018 and Monty Montgomery committed Sep 19, 2018

Dn't perform all the block-level delta math when block-level deltas aren't enabled.

* Add encoding of order_hint

For now a single bit is used for encoding order_hint

* Add intermediate variable

Add intermediate variable from which frame_type and order_hint are
derived

Don't test these modes at speed settings higher than 2

Fixes the extra silly indentation of this file.

Signed-off-by: Luca Barbato <lu_zero@gentoo.org>

And use a conditional slice outside the chroma loop.

Monty authored Sep 13, 2018 and Monty Montgomery committed Sep 13, 2018

Improve cache behavior by running vertical and horizontal deblocking in one pass.  Horizontal lags one SB behind vertical.

ME use predicted search center for topdown partition search

Yushin Cho authored Sep 12, 2018 and Yushin Cho committed Sep 12, 2018

In ME search range setup, rename x_min, x_max, y_min and y_max variables
as mvx_min, mvx_max, ... since they are limits of mv allowed to code thus needed
to search.

* 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

Luca Barbato authored Sep 12, 2018 and Thomas Daede committed Sep 12, 2018

Store the frame_type in the packet and implement Display for it.

See section 7.10.2.14 context and clamping process

Fix an off-by-one

And use it in rav1e to output the same stats it did before.

Monty authored Sep 11, 2018 and Monty Montgomery committed Sep 11, 2018

Inplement an ~ direct copy of libaom's LPF_PICK_FROM_Q deblocking
filter configuration.  This is not real RDO, but should be superior to
the current flat configuration.

The libaom code also mentions the magic numbers almost certainly need a
new regression fit.

Monty authored Sep 11, 2018 and Monty Montgomery committed Sep 11, 2018

The frame itself isn't welded onto the uncompressed header packets
until after both are complete, so jog the order such that deblocking
filter RDO can get stuck before the uncompressed header gets written
out.