perf: tt in qs move ordering

[jorgenhanssen]

Summary

Makes sense to try the TT move first in QS as well.

So now it tries the TT move first (for captures and evasions) and stores the best QS move back into the TT.

Bench before:

Depth        : 14
Max seldepth : 37
Avg seldepth : 26.9
Total time   : 9637 ms
Total nodes  : 14060544
Avg NPS      : 1459016
Avg branching: 2.39
Bench hash   : 96F05167A84BEBEA

Bench after:

Depth        : 14
Max seldepth : 40
Avg seldepth : 27.0
Total time   : 9324 ms
Total nodes  : 13832192
Avg NPS      : 1483504
Avg branching: 2.38
Bench hash   : 4D286F70D4A2AEF6

[jorgenhanssen]

--------------------------------------------------
Results of grail vs grail-next (10+0.1, 1t, 16MB, UHO_Lichess_4852_v1.epd):
Elo: -2.61 +/- 3.91, nElo: -4.54 +/- 6.81
LOS: 9.55 %, DrawRatio: 47.32 %, PairsRatio: 0.96
Games: 10000, Wins: 2757, Losses: 2832, Draws: 4411, Points: 4962.5 (49.62 %)
Ptnml(0-2): [118, 1228, 2366, 1187, 101], WL/DD Ratio: 1.37
LLR: -2.92 (-99.1%) (-2.94, 2.94) [0.00, 5.00]
--------------------------------------------------

[jorgenhanssen]

--------------------------------------------------
Results of grail vs grail-next (60+0.6, 1t, 64MB, UHO_Lichess_4852_v1.epd):
Elo: 1.11 +/- 10.36, nElo: 2.06 +/- 19.23
LOS: 58.30 %, DrawRatio: 51.04 %, PairsRatio: 1.01
Games: 1254, Wins: 342, Losses: 338, Draws: 574, Points: 629.0 (50.16 %)
Ptnml(0-2): [8, 145, 320, 143, 11], WL/DD Ratio: 1.24
LLR: -0.02 (-0.8%) (-2.94, 2.94) [0.00, 5.00]
--------------------------------------------------

[jorgenhanssen]

Nah this not working on shorter TC makes sense since there is less time to build up the TT with useful entries to be worth the overhead. Gonna test a bit more on LTC and VLTC before i conclude.

[jorgenhanssen]

Re-running SPRT due to rebase

[jorgenhanssen]

STC Still about neutral

--------------------------------------------------
Results of grail vs grail-next (10+0.1, 1t, 16MB, UHO_Lichess_4852_v1.epd):
Elo: 0.35 +/- 6.05, nElo: 0.63 +/- 10.86
LOS: 54.56 %, DrawRatio: 47.43 %, PairsRatio: 1.01
Games: 3934, Wins: 1108, Losses: 1104, Draws: 1722, Points: 1969.0 (50.05 %)
Ptnml(0-2): [31, 484, 933, 488, 31], WL/DD Ratio: 1.49
LLR: -0.30 (-10.3%) (-2.94, 2.94) [0.00, 5.00]
--------------------------------------------------

[jorgenhanssen]

LTC looks good to merge. Awaiting VLTC.

--------------------------------------------------
Results of grail vs grail-next (60+0.6, 1t, 64MB, UHO_Lichess_4852_v1.epd):
Elo: 9.99 +/- 8.38, nElo: 18.92 +/- 15.86
LOS: 99.03 %, DrawRatio: 50.76 %, PairsRatio: 1.24
Games: 1844, Wins: 522, Losses: 469, Draws: 853, Points: 948.5 (51.44 %)
Ptnml(0-2): [8, 195, 468, 238, 13], WL/DD Ratio: 1.23
LLR: 1.25 (42.5%) (-2.94, 2.94) [0.00, 5.00]
--------------------------------------------------

[jorgenhanssen]

--------------------------------------------------
Results of grail vs grail-next (180+1.8, 1t, 192MB, UHO_Lichess_4852_v1.epd):
Elo: -4.47 +/- 7.96, nElo: -8.69 +/- 15.45
LOS: 13.52 %, DrawRatio: 49.43 %, PairsRatio: 0.91
Games: 1942, Wins: 512, Losses: 537, Draws: 893, Points: 958.5 (49.36 %)
Ptnml(0-2): [5, 252, 480, 231, 3], WL/DD Ratio: 1.34
LLR: -0.90 (-30.6%) (-2.94, 2.94) [0.00, 5.00]
--------------------------------------------------

[jorgenhanssen]

Would expect the effect to be better at longer TC. Looks like it is not, so I am gonna close this.