tree 22e1cf6bdcb714c9ccc5ea106baaf3c9eeaadca4
parent 7e1dfee837f0cb717badf5e76bd48a23155b6bd9
author Rachel Barker <rachelbarker@google.com> 1698984599 +0000
committer Rachel Barker <rachelbarker@google.com> 1699405148 +0000

Use SAD for global motion error metric

Replace custom error metric (sum of |diff|^0.7) with SAD.
This has three advantages:
1) Better results
2) SAD is faster to calculate
3) We can remove ~1300 lines of code dedicated to the custom error
   metric

Note: Despite point (2), the overall runtime is increased slightly.
This is expected - improving the accuracy of global motion leads to it
being evaluated more often, and this outweighs the time savings from
speeding up the error metric evaluations. But overall this significantly
improves the gain : encode time cost ratio of global motion.

Impact on "good" mode:

 Speed | BDRATE-PSNR | BDRATE-SSIM |   Enc time
-------+-------------+-------------+-------------
   1   |   -0.087%   |   -0.102%   |   +0.085%
   2   |   -0.075%   |   -0.074%   |   +0.331%
   3   |   -0.111%   |   -0.122%   |   +0.445%
   4   |   -0.110%   |   -0.118%   |   +0.598%

Other modes, and good mode speed 5+, are unaffected.

STATS_CHANGED

Change-Id: Ifdc52c5012d6936a97a87cff489062402e81e495