Fast two-pass: additional dry-pass simplifications

[yeqing-wu]

Three simplifications to the dry pass of fast two-pass partitions.

1. Skip deblock-aware distortion. The dry pass ranks shapes only, so reconstructing and filtering to nudge distortion by the deblocking delta is wasted precision.

2. Reuse the dry pass's simple motion searches in the wet pass. MVs, sse/var, rate/dist/rdcost and start MVs depend only on source, references and geometry, so recomputing them is redundant. A new preserve_sms_cache flag drives av2_reset_sms_cross_pass_state(), which clears only what must not cross the pass boundary.

3. Cap two transform-side inner loops in search_tx_type, select_tx_partition_type and choose_tx_size_type_from_rd on x->apply_dry_pass_shortcuts: force the (set_idx=0, stx=0) IST/stx baseline, and restrict the tx-partition search to NONE and SPLIT (2 of 8 variants).

STATS_CHANGED

Anchor: commit c5df831
Speed 4 (cpu-used=4): FG16 CTC (33 frames, class A1 and A2, RA)
Speed 5 (cpu-used=5): FG16 CTC (33 frames, class A1 and A2, RA)

1) Speed 4 (cpu-used=4)
  +------------+------+-------+-------+------+------+------+
  | Class      |    Y |    Cb |    Cr | wAvg | Enc% | Dec% |
  +------------+------+-------+-------+------+------+------+
  | A1         | 0.27 | -0.23 | -0.21 | 0.23 |   95 |  100 |
  | A2         | 0.31 | -0.42 | -0.46 | 0.25 |   95 |  100 |
  | Avg w/o B2 | 0.30 | -0.36 | -0.39 | 0.25 |   95 |  100 |
  +------------+------+-------+-------+------+------+------+

2) Speed 5 (cpu-used=5)
  +------------+------+-------+-------+------+------+------+
  | Class      |    Y |    Cb |    Cr | wAvg | Enc% | Dec% |
  +------------+------+-------+-------+------+------+------+
  | A1         | 0.39 |  0.01 |  0.13 | 0.37 |   94 |  100 |
  | A2         | 0.26 | -0.39 | -0.57 | 0.20 |   94 |  100 |
  | Avg w/o B2 | 0.30 | -0.27 | -0.37 | 0.25 |   94 |  100 |
  +------------+------+-------+-------+------+------+------+

[leolzhao]

What's the run-time savings for cpu-used = 6? @yeqing-wu

[yeqing-wu]

What's the run-time savings for cpu-used = 6? @yeqing-wu

I have not tested that since I saw both speed 4 and 5 are within the allowed ratio.