Fix mixup of GLOBALMV and NEAR0MV
Fixes #2297
Please rebase.
Coverage increased (+1.6%) to 82.011% when pulling 2f02ed48b9583a264d8194ab3a1da64070f16554 on shreevari:mixup into 99114995e8771dd923a146e5616f7474e9b33eb7 on xiph:master.
This is what I get when running these commits. Unfortunately, it seems to give a loss. That doesn't necessarily mean that this shouldn't be commited.
https://beta.arewecompressedyet.com/?job=ref_near_mv_global_mv_s6%402020-06-01T17%3A17%3A56.780Z&job=fix_near_mv_global_mv_s6_correct%402020-06-01T17%3A43%3A10.635Z
https://beta.arewecompressedyet.com/?job=ref_near_mv_global_mv_s3%402020-06-01T16%3A52%3A53.098Z&job=fix_near_mv_global_mv_s3%402020-06-01T16%3A43%3A45.501Z
Average difference between the reference and the fixes in comparison on AWCY are as follows:
| Commit | PSNR | PSNR HVS | SSIM | CIEDE 2000 | APSNR | MS SSIM | VMAF |
|---|---|---|---|---|---|---|---|
| Simple swap vs ref | 0.06 | 0.06 | 0.07 | 0.13 | 0.06 | 0.04 | 0.14 |
| Push GLOBALMV unconditionally vs ref | 0.07 | 0.11 | 0.09 | 0.10 | 0.07 | 0.12 | 0.22 |
Pushing GLOBALMV unconditionally seems to perform worse than otherwise