Extended the HAD calculation with SIMD to be performed in 32-bit space rather than 16-bit. This fixes #339

Refactor AlfCovariance::getClipMax loop:

This refactoring solve the issue with VS2017 in Release mode, where test
"if (E[clip_max[k]][0][k][l] != E[clip_max[k]+1][0][k][l])" was sometime true when it should have been false
(noticed with clip_max[k] == 1, i.e. during second iteration of while loop).

Some alternative tries: casting E[][][][] in long long, using std::abs(a-b) < 1,
or using std::abs(a-b) < std::numeric_limits::epsilon() failed to solve the issue.

Note: no analysis of the generated assembly have been performed to really understand the problem

Hendry authored 5 years ago and Karsten Suehring committed 5 years ago

update the nonCTC cfg files

Cache is enabled incorrectly in DMVR processing

JVET-N0805: APS for LMCS (using different APS types)

See merge request !577

Hendry authored 5 years ago and Karsten Suehring committed 5 years ago

- Fix the entry points
- Put some validation for rectangular slice
- Bug fix for N0124 signalling for number of bits to signal bottomrightidx

The per-frame decoding time (DT) value reported by the decoder
does not include time spent in the loop filters. Such time is
hereby added

- add braces as suggested
- improve indentation of "else" branch
- fix consistently weird line breaks in clipMv function calls

Reduce scope to build with #define JVET_N0242_NON_LINEAR_ALF 0

  - neighGbi array not properly initialized.

Align with specification the coding of pps_virtual_boundaries_pos_x and pps_virtual_boundaries_pos_y

This reverts merge request !580