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

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.

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.

The Deblocking state must be mutable, and now that we're finally
trying to make the FrameInvariants actually invariant over a frame,
it's time to move the DeblockingState out.

* Plumb SuperBlockOffset and use superblock dimensions

Fix for #554

Roll in typo fixes and comment corrections from Yushin's review.
No functional change.

Finishes formatting of me.rs

Kyle Siefring authored 6 years ago and Thomas Daede committed 6 years ago

Add transforms up to x16 except for flipadst

Fix a type mismatch in deblocking

No RDO as yet, but the deblocking filter and parameter infrastructure are
full implemented, functional and tested.

* Modify plane data structure to allow for padded border

* Pad reconstructed frame

* Have ME consider motion vectors pointing outside frame boundary

* Remove clamping inside motion compensation loop

Clamping is now done outside the loop and relies on the frame
being padded.

* Reduce size of intermediate buffer in motion compensation

* Update encode/decode tests

* 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.

Gideon Mayhak authored 6 years ago and Thomas Daede committed 6 years ago

Added information about AppVeyor Windows builds, and other minor edits.

This reduces the time for cdef_filter_frame by ~33%.

Change MV mode if NEAREST isn't NEAREST anymore instead of doing repeated RDO optimizations. Improves run time by about 20-25%.

Add assert statement to check that nearest MV is indeed the same
as the MV passed as argument to encode_block_b()

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

* Move motion compensation to separate function

* Motion compensate for every iteration of tx search

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

Also does pulling tx_type_rdo() out of chroma RDO, as requested in #521.