Speed5 tools off

[yeqing-wu]

Do not squash and then merge. Let's do the rebase and keep all three commits so that we can keep the record of the BD rate results of each change.

1. Turn off palette intra mode at speed >= 5 for non-screen content

STATS_CHANGED
Anchor: c5df831
Speed 5 (cpu-used=5): FG16 CTC (33 frames, class A1 and A2, RA)

  Class          Y      Cb      Cr    wAvg   Enc%   Dec%
  A1         -0.03    0.19   -0.01   -0.02   98.6   100
  A2          0.05    0.22    0.08    0.06   99.0   100
  Avg w/o B2  0.03    0.21    0.05    0.04   98.9   100

2. Turn off one-sided compound at speed >= 5

STATS_CHANGED
Anchor: c5df831
Speed 5 (cpu-used=5): FG16 CTC (33 frames, class A1 and A2, RA)

  Class          Y      Cb      Cr    wAvg   Enc%   Dec%
  A1          0.05    0.11    0.16    0.05   98.6   101
  A2          0.04    0.18    0.30    0.06   98.2    99
  Avg w/o B2  0.04    0.16    0.26    0.06   98.3   100

3. Turn off TIP RefineMV/OPFL at speed >= 5
   STATS_CHANGED
   Anchor: c5df831
   Speed 5 (cpu-used=5): FG16 CTC (33 frames, class A1 and A2, RA)

  Class          Y      Cb      Cr    wAvg   Enc%   Dec%
  A1          0.71    0.66    0.45    0.70    82     86
  A2          0.74    0.41    0.58    0.72    89     86
  Avg w/o B2  0.73    0.48    0.54    0.71    87     86

[yunqingwang1]

Maybe it is better to set speed related feature in one place as other speed features? By the way, I also turned this off for LC mode in set_good_speed_features_lc_dec_framesize_independent().

[yeqing-wu]

Thanks @yunqingwang1.

At first, I thought that seq_params was set in av2_init_seq_coding_tools(), which is called before av2_set_speed_features_framesize_independent(). Then, seq_params is locked at the first key frame, preventing downstream readers from modifying it. However, after reviewing your LC mode code, I realized my understanding was incorrect. I will modify the code as you suggested.

[yeqing-wu]

Done, modified as you suggested.