Another scaling revert

Revert recent passer because it seems to not scale for longer time
controls. Adjust comments accordingly.

Passed VVLTC SPRT with STC bounds:
https://tests.stockfishchess.org/tests/view/680ddff73629b02d74b15b3f
LLR: 2.95 (-2.94,2.94) <0.00,2.00>
Total: 198316 W: 51317 L: 50846 D: 96153
Ptnml(0-2): 17, 18459, 61737, 18926, 19

Passed VVLTC SPRT with LTC bounds:
https://tests.stockfishchess.org/tests/view/680d5b7e3629b02d74b15a21
LLR: 2.95 (-2.94,2.94) <0.50,2.50>
Total: 123274 W: 31738 L: 31283 D: 60253
Ptnml(0-2): 7, 11444, 38282, 11895, 9

closes https://github.com/official-stockfish/Stockfish/pull/6038

Bench: 2173845
This commit is contained in:
Michael Chaly
2025-04-28 02:48:31 +03:00
committed by Disservin
parent 81cc004060
commit b73c8982df

View File

@@ -1244,6 +1244,8 @@ moves_loop: // When in check, search starts here
ss->reduction = 0;
// Do a full-depth search when reduced LMR search fails high
// (*Scaler) Usually doing more shallower searches
// doesn't scale well to longer TCs
if (value > alpha && d < newDepth)
{
// Adjust full-depth search based on LMR results - if the result was
@@ -1260,11 +1262,7 @@ moves_loop: // When in check, search starts here
update_continuation_histories(ss, movedPiece, move.to_sq(), 1508);
}
else if (value > alpha && value < bestValue + 9)
{
newDepth--;
if (value < bestValue + 4)
newDepth--;
}
}
// Step 18. Full-depth search when LMR is skipped