stockfish
6 weeks agoMerge remote-tracking branch 'upstream/master' into HEAD master
Steinar H. Gunderson [Sat, 8 Aug 2020 14:40:07 +0000 (16:40 +0200)]
Merge remote-tracking branch 'upstream/master' into HEAD

6 weeks agosimplified and increased threshold to switch between NNUE and classical
U-DESKTOP-3900\Mark [Fri, 7 Aug 2020 23:53:18 +0000 (19:53 -0400)]
simplified and increased threshold to switch between NNUE and classical

STC https://tests.stockfishchess.org/tests/view/5f2deb1661e3b6af6488200f
LLR: 2.96 (-2.94,2.94) {-1.50,0.50}
Total: 10376 W: 1481 L: 1359 D: 7536
Ptnml(0-2): 91, 953, 2981, 1069, 94

LTC: https://tests.stockfishchess.org/html/live_elo.html?5f2e0a0461e3b6af64882019
LLR: 2.99 (-2.94,2.94) {-1.50,0.50}
Total: 5040 W: 375 L: 315 D: 4350
Ptnml(0-2): 7, 263, 1926, 311, 13

closes https://github.com/official-stockfish/Stockfish/pull/2934

Bench: 4067325

6 weeks agoDo more futility pruning for parent nodes.
Vizvezdenec [Sat, 8 Aug 2020 02:51:26 +0000 (05:51 +0300)]
Do more futility pruning for parent nodes.

This patch increases LMRdepth threshold for futility pruning at parent nodes so it can apply more often.
With radical change to evaluation approach it seems that search is really far from optimal state, especially it parts that use static evaluation of position.

passed STC
https://tests.stockfishchess.org/tests/view/5f2da75661e3b6af64881fd0
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 8744 W: 1305 L: 1156 D: 6283
Ptnml(0-2): 75, 789, 2500, 928, 80

passed LTC
https://tests.stockfishchess.org/tests/view/5f2dcb2a61e3b6af64881ff3
LLR: 2.98 (-2.94,2.94) {0.25,1.75}
Total: 17728 W: 1256 L: 1117 D: 15355
Ptnml(0-2): 22, 961, 6774, 1070, 37

Bench: 4067325

6 weeks agoReintroduce late irreversible move extension
SFisGOD [Fri, 7 Aug 2020 19:45:08 +0000 (03:45 +0800)]
Reintroduce late irreversible move extension

Reintroduce vondele's late irreversible move extension for fortress keeping.
This was removed when we only had classical eval.
Now that we have the NNUE net, it seems that this is useful again.

STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 5352 W: 787 L: 653 D: 3912
Ptnml(0-2): 34, 451, 1579, 571, 41
https://tests.stockfishchess.org/tests/view/5f2dc8ad61e3b6af64881ff0

LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 14416 W: 1013 L: 891 D: 12512
Ptnml(0-2): 15, 722, 5623, 822, 26
https://tests.stockfishchess.org/tests/view/5f2e0e3661e3b6af6488201e

closes https://github.com/official-stockfish/Stockfish/pull/2936

Bench: 4154696

6 weeks agoExpand outposts to minors shielded by pawns
Lolligerhans [Fri, 7 Aug 2020 09:24:37 +0000 (11:24 +0200)]
Expand outposts to minors shielded by pawns

Allow any pawn in front of a minor piece to replace the pawn protection
requirement for outposts.

  +-------+  +-------+
  | . . o |  | o . . |    o  Their pawns
  | . o x |  | o . . |    x  Our pawns
  | o N . |  | x o B |  N,B  New (reachable) outpost
  | . . . |  | . _ . |    _  Reachable square behind a pawn
  +-------+  +-------+
  N outpost  B reaches
               outpost

  We want outposts to be secured by pawns against major pieces. If
a minor is shielded by any pawn from above, it is rarely at the same
time protected by our pawn attacks from below. However, the pawn shield
in itself offers some degree of protection.
  A pawn shield will now suffice to replace the pawn protection for the
outpost (and reachable outpost) bonus.

This effect stacks with the existing "minor behind pawn" bonus.

STC
https://tests.stockfishchess.org/tests/view/5f2bcd14b3ebe5cbfee85b2c
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 27248 W: 5353 L: 5119 D: 16776
Ptnml(0-2): 462, 3174, 6185, 3274, 529

LTC
https://tests.stockfishchess.org/tests/view/5f2bfef5b3ebe5cbfee85b5a
LLR: 2.96 (-2.94,2.94) {0.25,1.75}
Total: 99432 W: 12580 L: 12130 D: 74722
Ptnml(0-2): 696, 8903, 30049, 9391, 677

Closes #2935

Bench: 4143673

6 weeks agoTweak futility pruning depth.
Unai Corzo [Sat, 8 Aug 2020 06:24:20 +0000 (08:24 +0200)]
Tweak futility pruning depth.

STC https://tests.stockfishchess.org/tests/view/5f2d237161e3b6af64881f43
LLR: 2.96 (-2.94,2.94) {-0.50,1.50}
Total: 12712 W: 1823 L: 1664 D: 9225
Ptnml(0-2): 122, 1166, 3627, 1313, 128

LTC https://tests.stockfishchess.org/tests/view/5f2d473061e3b6af64881f6f
LLR: 2.96 (-2.94,2.94) {0.25,1.75}
Total: 12104 W: 912 L: 788 D: 10404
Ptnml(0-2): 13, 665, 4582, 769, 23

closes https://github.com/official-stockfish/Stockfish/pull/2930

bench: 4271421

6 weeks agoUpdate default net to nn-9931db908a9b.nnue
Sergio Vieri [Fri, 7 Aug 2020 07:15:04 +0000 (15:15 +0800)]
Update default net to nn-9931db908a9b.nnue

Net created at 20200806-1802

passed STC:
https://tests.stockfishchess.org/tests/view/5f2d00b461e3b6af64881f21
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 6672 W: 1052 L: 898 D: 4722
Ptnml(0-2): 63, 600, 1868, 730, 75

passed LTC:
https://tests.stockfishchess.org/tests/view/5f2d052a61e3b6af64881f29
LLR: 2.96 (-2.94,2.94) {0.25,1.75}
Total: 7576 W: 573 L: 463 D: 6540
Ptnml(0-2): 8, 392, 2889, 480, 19

closes https://github.com/official-stockfish/Stockfish/pull/2923

Bench: 4390086

6 weeks agoDo move legality check before pruning.
Stefan Geschwentner [Wed, 5 Aug 2020 07:29:27 +0000 (09:29 +0200)]
Do move legality check before pruning.

This alllows to simplify the code because the move counter haven't to be
decremented later if a move isn't legal. As a side effect now illegal
pruned moves doesn't included anymore in move counter. So slightly less
pruning and reductions are done.

STC:
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 111016 W: 21106 L: 21077 D: 68833
Ptnml(0-2): 1830, 13083, 25736, 12946, 1913
https://tests.stockfishchess.org/tests/view/5f28816fa5abc164f05e4c26

LTC:
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 39264 W: 4909 L: 4843 D: 29512
Ptnml(0-2): 263, 3601, 11854, 3635, 279
https://tests.stockfishchess.org/tests/view/5f297902a5abc164f05e4c8e

closes https://github.com/official-stockfish/Stockfish/pull/2906

Bench: 4390086

6 weeks agoRemove QueenInfiltration
UnaiCorzo [Tue, 4 Aug 2020 12:32:52 +0000 (14:32 +0200)]
Remove QueenInfiltration

STC https://tests.stockfishchess.org/tests/view/5f2955b1a5abc164f05e4c85
LLR: 2.96 (-2.94,2.94) {-1.50,0.50}
Total: 29216 W: 5560 L: 5416 D: 18240
Ptnml(0-2): 466, 3329, 6902, 3417, 494

LTC https://tests.stockfishchess.org/tests/view/5f299154a5abc164f05e4ca1
LLR: 2.92 (-2.94,2.94) {-1.50,0.50}
Total: 54144 W: 6635 L: 6594 D: 40915
Ptnml(0-2): 372, 4859, 16536, 4966, 339

closes https://github.com/official-stockfish/Stockfish/pull/2910

Bench: 4609008

6 weeks agoTuned pawn values
FauziAkram [Fri, 7 Aug 2020 00:37:54 +0000 (17:37 -0700)]
Tuned pawn values

Passed STC:
https://tests.stockfishchess.org/tests/view/5f2aa49fa5abc164f05e4d1b
LLR: 2.95 (-2.94,2.94) {-0.50,1.50}
Total: 40888 W: 7977 L: 7726 D: 25185
Ptnml(0-2): 665, 4806, 9333, 4893, 747

Passed LTC:
https://tests.stockfishchess.org/tests/view/5f2b1059b3ebe5cbfee85ae7
LLR: 2.98 (-2.94,2.94) {0.25,1.75}
Total: 51264 W: 6445 L: 6134 D: 38685
Ptnml(0-2): 328, 4564, 15580, 4789, 371

closes https://github.com/official-stockfish/Stockfish/pull/2920

bench: 4314943

6 weeks agoAdd tempo also to NNUE eval.
Stefan Geschwentner [Thu, 6 Aug 2020 23:08:15 +0000 (01:08 +0200)]
Add tempo also to NNUE eval.

STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 10608 W: 1507 L: 1358 D: 7743
Ptnml(0-2): 94, 945, 3074, 1100, 91
https://tests.stockfishchess.org/tests/view/5f2c5921b3ebe5cbfee85b8b

LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 7536 W: 556 L: 448 D: 6532
Ptnml(0-2): 9, 383, 2881, 481, 14
https://tests.stockfishchess.org/tests/view/5f2c6f4461e3b6af64881e95

closes https://github.com/official-stockfish/Stockfish/pull/2919

Bench: 4746616

6 weeks agoNNUE evaluation threshold
MJZ1977 [Thu, 6 Aug 2020 15:39:10 +0000 (17:39 +0200)]
NNUE evaluation threshold

The idea is to use NNUE only on quite balanced material positions. This bring a big speedup on research since NNUE eval is slower than classical eval for most of the hardwares and specially on unbalanced positions with LazyEval.

STC: https://tests.stockfishchess.org/tests/view/5f2c2680b3ebe5cbfee85b61
LLR: 2.95 (-2.94,2.94) {-0.50,1.50}
Total: 3168 W: 560 L: 400 D: 2208
Ptnml(0-2): 21, 294, 819, 404, 46

LTC: https://tests.stockfishchess.org/tests/view/5f2c2ca6b3ebe5cbfee85b69
LLR: 2.98 (-2.94,2.94) {0.25,1.75}
Total: 3200 W: 287 L: 183 D: 2730
Ptnml(0-2): 4, 149, 1191, 251, 5

closes https://github.com/official-stockfish/Stockfish/pull/2916

Bench 4746616

6 weeks agoAdd NNUE evaluation
nodchip [Wed, 5 Aug 2020 15:11:15 +0000 (17:11 +0200)]
Add NNUE evaluation

This patch ports the efficiently updatable neural network (NNUE) evaluation to Stockfish.

Both the NNUE and the classical evaluations are available, and can be used to
assign a value to a position that is later used in alpha-beta (PVS) search to find the
best move. The classical evaluation computes this value as a function of various chess
concepts, handcrafted by experts, tested and tuned using fishtest. The NNUE evaluation
computes this value with a neural network based on basic inputs. The network is optimized
and trained on the evalutions of millions of positions at moderate search depth.

The NNUE evaluation was first introduced in shogi, and ported to Stockfish afterward.
It can be evaluated efficiently on CPUs, and exploits the fact that only parts
of the neural network need to be updated after a typical chess move.
[The nodchip repository](https://github.com/nodchip/Stockfish) provides additional
tools to train and develop the NNUE networks.

This patch is the result of contributions of various authors, from various communities,
including: nodchip, ynasu87, yaneurao (initial port and NNUE authors), domschl, FireFather,
rqs, xXH4CKST3RXx, tttak, zz4032, joergoster, mstembera, nguyenpham, erbsenzaehler,
dorzechowski, and vondele.

This new evaluation needed various changes to fishtest and the corresponding infrastructure,
for which tomtor, ppigazzini, noobpwnftw, daylen, and vondele are gratefully acknowledged.

The first networks have been provided by gekkehenker and sergiovieri, with the latter
net (nn-97f742aaefcd.nnue) being the current default.

The evaluation function can be selected at run time with the `Use NNUE` (true/false) UCI option,
provided the `EvalFile` option points the the network file (depending on the GUI, with full path).

The performance of the NNUE evaluation relative to the classical evaluation depends somewhat on
the hardware, and is expected to improve quickly, but is currently on > 80 Elo on fishtest:

60000 @ 10+0.1 th 1
https://tests.stockfishchess.org/tests/view/5f28fe6ea5abc164f05e4c4c
ELO: 92.77 +-2.1 (95%) LOS: 100.0%
Total: 60000 W: 24193 L: 8543 D: 27264
Ptnml(0-2): 609, 3850, 9708, 10948, 4885

40000 @ 20+0.2 th 8
https://tests.stockfishchess.org/tests/view/5f290229a5abc164f05e4c58
ELO: 89.47 +-2.0 (95%) LOS: 100.0%
Total: 40000 W: 12756 L: 2677 D: 24567
Ptnml(0-2): 74, 1583, 8550, 7776, 2017

At the same time, the impact on the classical evaluation remains minimal, causing no significant
regression:

sprt @ 10+0.1 th 1
https://tests.stockfishchess.org/tests/view/5f2906a2a5abc164f05e4c5b
LLR: 2.94 (-2.94,2.94) {-6.00,-4.00}
Total: 34936 W: 6502 L: 6825 D: 21609
Ptnml(0-2): 571, 4082, 8434, 3861, 520

sprt @ 60+0.6 th 1
https://tests.stockfishchess.org/tests/view/5f2906cfa5abc164f05e4c5d
LLR: 2.93 (-2.94,2.94) {-6.00,-4.00}
Total: 10088 W: 1232 L: 1265 D: 7591
Ptnml(0-2): 49, 914, 3170, 843, 68

The needed networks can be found at https://tests.stockfishchess.org/nns
It is recommended to use the default one as indicated by the `EvalFile` UCI option.

Guidelines for testing new nets can be found at
https://github.com/glinscott/fishtest/wiki/Creating-my-first-test#nnue-net-tests

Integration has been discussed in various issues:
https://github.com/official-stockfish/Stockfish/issues/2823
https://github.com/official-stockfish/Stockfish/issues/2728

The integration branch will be closed after the merge:
https://github.com/official-stockfish/Stockfish/pull/2825
https://github.com/official-stockfish/Stockfish/tree/nnue-player-wip

closes https://github.com/official-stockfish/Stockfish/pull/2912

This will be an exciting time for computer chess, looking forward to seeing the evolution of
this approach.

Bench: 4746616

6 weeks agoMerge remote-tracking branch 'upstream/master' into HEAD
Steinar H. Gunderson [Mon, 3 Aug 2020 10:27:33 +0000 (12:27 +0200)]
Merge remote-tracking branch 'upstream/master' into HEAD

7 weeks agoTweak cutnode reduction
Stefan Geschwentner [Thu, 30 Jul 2020 16:56:11 +0000 (18:56 +0200)]
Tweak cutnode reduction

Less reduction for second move at non-check CUT node with depth <= 10.

STC:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 38680 W: 7490 L: 7245 D: 23945
Ptnml(0-2): 643, 4441, 8967, 4606, 683
https://tests.stockfishchess.org/tests/view/5f21e1782f7e63962b99f451

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.75}
Total: 71976 W: 9003 L: 8636 D: 54337
Ptnml(0-2): 440, 6414, 21972, 6663, 499
https://tests.stockfishchess.org/tests/view/5f2245762f7e63962b99f4bd

closes https://github.com/official-stockfish/Stockfish/pull/2868

Bench: 4746616

7 weeks agoImprove handling of queen imbalance
Stéphane Nicolet [Tue, 28 Jul 2020 08:08:09 +0000 (10:08 +0200)]
Improve handling of queen imbalance

We double the bonus for potential threats by minors and rooks against
our queen, in case of "queen vs pieces imbalance". Hopefully this will
improve a little bit the evaluation for this well-known Stockfish weakness.

passed STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 72976 W: 14003 L: 13710 D: 45263
Ptnml(0-2): 1218, 8370, 17094, 8513, 1293
https://tests.stockfishchess.org/tests/view/5efa50eb020eec13834a977d

passed LTC:
LLR: 2.93 (-2.94,2.94) {0.25,1.75}
Total: 22232 W: 2779 L: 2560 D: 16893
Ptnml(0-2): 129, 1885, 6896, 2050, 156
https://tests.stockfishchess.org/tests/view/5f1fdd2dc09435d870cb9f13

closes https://github.com/official-stockfish/Stockfish/pull/2864

Bench: 4367349

7 weeks agoRemove late irreversible move extension
UnaiCorzo [Sat, 25 Jul 2020 20:30:05 +0000 (22:30 +0200)]
Remove late irreversible move extension

We simplify away the late irreversible move extension, which
does not seem to be necessary in the current master.

STC
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 38584 W: 7464 L: 7342 D: 23778
Ptnml(0-2): 581, 4328, 9365, 4424, 594
https://tests.stockfishchess.org/tests/view/5f1c9669c09435d870cb9de9

LTC
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 27840 W: 3417 L: 3353 D: 21070
Ptnml(0-2): 120, 2315, 8994, 2363, 128
https://tests.stockfishchess.org/tests/view/5f1d2e22c09435d870cb9e21

closes https://github.com/official-stockfish/Stockfish/pull/2836

bench: 4829420

7 weeks agoRemove conditional_more_than_two().
protonspring [Sat, 25 Jul 2020 13:32:19 +0000 (07:32 -0600)]
Remove conditional_more_than_two().

This is a functional simplification that removes the conditional_more_than_two()
function, which was quite strange and kooky. Note the very minor change to the bench
value.

See this thread for relevant comments on the passing branch:
protonspring/Stockfish@d89730d...ff35b50

STC
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 59760 W: 11411 L: 11311 D: 37038
Ptnml(0-2): 992, 6863, 14044, 7015, 966
https://tests.stockfishchess.org/tests/view/5f179988c09435d870cb9b9a

LTC
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 45208 W: 5553 L: 5497 D: 34158
Ptnml(0-2): 315, 4081, 13761, 4127, 320
https://tests.stockfishchess.org/tests/view/5f184847c09435d870cb9bee

closes https://github.com/official-stockfish/Stockfish/pull/2826

Bench: 4578290

2 months agoDo not overwrite valuable TT data after probcut.
Vizvezdenec [Mon, 13 Jul 2020 17:30:58 +0000 (20:30 +0300)]
Do not overwrite valuable TT data after probcut.

This patch allows an engine to write probcut data only in case
the probcut search depth is greater than transposition table depth.

passed STC
https://tests.stockfishchess.org/tests/view/5f0b52e959f6f035328949a6
LLR: 2.97 (-2.94,2.94) {-0.50,1.50}
Total: 52544 W: 10145 L: 9880 D: 32519
Ptnml(0-2): 853, 6097, 12121, 6334, 867

passed LTC
https://tests.stockfishchess.org/tests/view/5f0bd94c59f6f035328949f3
LLR: 2.93 (-2.94,2.94) {0.25,1.75}
Total: 49576 W: 6164 L: 5863 D: 37549
Ptnml(0-2): 297, 4371, 15218, 4538, 364

closes https://github.com/official-stockfish/Stockfish/pull/2815

bench 4578298

2 months agoMultiple lazy stages.
mstembera [Sun, 12 Jul 2020 20:58:00 +0000 (13:58 -0700)]
Multiple lazy stages.

An extension of the lazy eval idea: when the score is sufficiently large
we now skip more granular parts of the eval.

Inspired by an original patch by Moez Jellouli
https://tests.stockfishchess.org/tests/view/5f03b2a159f6f03532894529
Credit to him!

STC https://tests.stockfishchess.org/tests/view/5f0a862c59f6f03532894924
LLR: 2.95 (-2.94,2.94) {-0.50,1.50}
Total: 13504 W: 2684 L: 2472 D: 8348
Ptnml(0-2): 229, 1496, 3111, 1666, 250

LTC https://tests.stockfishchess.org/tests/view/5f0ac0e159f6f0353289495b
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 31312 W: 3926 L: 3677 D: 23709
Ptnml(0-2): 185, 2773, 9509, 2986, 203

closes https://github.com/official-stockfish/Stockfish/pull/2814

bench: 4541608

2 months agoCorrects a functional change in a cleanup patch.
syzygy1 [Sat, 11 Jul 2020 14:25:34 +0000 (16:25 +0200)]
Corrects a functional change in a cleanup patch.

This corrects a functional change in
https://github.com/official-stockfish/Stockfish/commit/ddcbacd04d1c860e808202ce8c1206c8acdca627
changing evaluation of KPPvK. Bench remains unchanged at low depth

With this patch, 8/8/5k1p/8/7p/7K/8/8 b - - 1 11 is again correctly evaluated as a draw.

closes https://github.com/official-stockfish/Stockfish/pull/2807

Bench: 4366686

2 months agoIntroduce bad outpost penalty
SFisGOD [Wed, 8 Jul 2020 02:09:32 +0000 (10:09 +0800)]
Introduce bad outpost penalty

In some French games, Stockfish likes to bring the Knight to a bad outpost spot. This is evident in TCEC S18 Superfinal Game 63, where there is a Knight outpost on the queenside but is actually useless. Stockfish is effectively playing a piece down while holding ground against Leela's break on the kingside.

This patch turns the +56 mg bonus for a Knight outpost into a -7 mg penalty if it satisfies the following conditions:

* The outpost square is not on the CenterFiles (i.e. not on files C,D,E and F)
* The knight is not attacking non pawn enemies.
* The side where the outpost is located contains only few enemies, with a particular conditional_more_than_two() implementation

Thank you to apospa...@gmail.com for bringing this to our attention and for providing insights.
See https://groups.google.com/forum/?fromgroups=#!topic/fishcooking/dEXNzSIBgZU
Reference game: https://tcec-chess.com/#div=sf&game=63&season=18

Passed STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 6960 W: 1454 L: 1247 D: 4259
Ptnml(0-2): 115, 739, 1610, 856, 160
https://tests.stockfishchess.org/tests/view/5f08221059f6f0353289477e

Passed LTC:
LLR: 2.98 (-2.94,2.94) {0.25,1.75}
Total: 21440 W: 2767 L: 2543 D: 16130
Ptnml(0-2): 122, 1904, 6462, 2092, 140
https://tests.stockfishchess.org/tests/view/5f0838ed59f6f035328947a2

various related tests show strong test results, but so far no generalizations or simplifications of conditional_more_than_two() are found. See PR for details.

closes https://github.com/official-stockfish/Stockfish/pull/2803

Bench: 4366686

2 months agoMaximize usage of transposition table in probcut
Vizvezdenec [Fri, 10 Jul 2020 21:06:55 +0000 (00:06 +0300)]
Maximize usage of transposition table in probcut

Probcut is a heuristic that wasn't changed a lot in past years,
all attempts to change it using information / writing info to transposition table failed.

This patch has a number of differences that can be summarized as follows:

* For TT write/read we use depth - 3. Because probcut search is depth - 4 but we actually do the move prior to it so effectively we do depth - 3 search;
* In any case of depth of eval from transposition table being >= depth - 3 we either produce cutoff or refuse to even do probcut search, this is allowing us to write info of probcut to transposition table because we know that we wouldn't be overwriting some deeper data with our depth - 3 search - this is an important aspect of this patch;
* For some not really known reason this patch completely ignores tte->bound() - which was the case for previous patch that made probcut interact with TT, maybe 2) is the reason, although it's unproven.

A first version of this patch passed STC and LTC

passed STC
https://tests.stockfishchess.org/tests/view/5f05908a59f6f03532894613
LLR: 2.95 (-2.94,2.94) {-0.50,1.50}
Total: 95776 W: 18300 L: 17973 D: 59503
Ptnml(0-2): 1646, 10944, 22377, 11279, 1642

passed LTC
https://tests.stockfishchess.org/tests/view/5f06b54059f6f035328946bb
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 57128 W: 7266 L: 6938 D: 42924
Ptnml(0-2): 372, 5163, 17217, 5389, 423

However, an additional bugfix was needed to avoid checking a condition on ttMove if was not available. This passed non-regression bounds on top of the first version:

at STC
https://tests.stockfishchess.org/tests/view/5f080e5059f6f03532894766
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 14096 W: 2800 L: 2628 D: 8668
Ptnml(0-2): 225, 1620, 3238, 1688, 277

at LTC
https://tests.stockfishchess.org/tests/view/5f0836a559f6f0353289479c
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 25352 W: 3228 L: 3139 D: 18985
Ptnml(0-2): 175, 2350, 7549, 2415, 187

closes https://github.com/official-stockfish/Stockfish/pull/2804

Bench 4540940

2 months agoSmall cleanups
Joost VandeVondele [Thu, 9 Jul 2020 20:01:06 +0000 (22:01 +0200)]
Small cleanups

closes https://github.com/official-stockfish/Stockfish/pull/2772

No functional change

2 months agoSimplify make_promotions()
mstembera [Sun, 5 Jul 2020 22:17:04 +0000 (15:17 -0700)]
Simplify make_promotions()

Remove special case handling of QUIET_CHECKS in make_promotions()

STC https://tests.stockfishchess.org/tests/view/5f055dbb59f6f035328945fb
LLR: 2.98 (-2.94,2.94) {-1.50,0.50}
Total: 42808 W: 8177 L: 8054 D: 26577
Ptnml(0-2): 665, 4890, 10201, 4953, 695

LTC https://tests.stockfishchess.org/tests/view/5f06231a59f6f03532894661
LLR: 2.96 (-2.94,2.94) {-1.50,0.50}
Total: 9616 W: 1214 L: 1111 D: 7291
Ptnml(0-2): 53, 821, 2965, 908, 61

closes https://github.com/official-stockfish/Stockfish/pull/2800

Bench: 4576410

2 months agoConnected / blocked pawns simplification
FauziAkram [Tue, 7 Jul 2020 22:29:03 +0000 (01:29 +0300)]
Connected / blocked pawns simplification

There is no need to score blocked pawns at many places.
The idea originated from: Rocky
Tuning and testing by: Fauzi

Passed STC:
https://tests.stockfishchess.org/tests/view/5f04f8fd59f6f035328945d4
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 6352 W: 1299 L: 1118 D: 3935
Ptnml(0-2): 89, 695, 1469, 792, 131

Passed LTC:
https://tests.stockfishchess.org/tests/view/5f0527bd59f6f035328945e3
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 27648 W: 3517 L: 3433 D: 20698
Ptnml(0-2): 177, 2561, 8301, 2571, 214

closes https://github.com/official-stockfish/Stockfish/pull/2799

Bench: 4734746

2 months agoClean-up en passant processing
Alain SAVARD [Tue, 7 Jul 2020 02:43:54 +0000 (22:43 -0400)]
Clean-up en passant processing

the goal of this PR is to better document how we process
the ep square (if any) given position fen command, and to
output more meaningful (and consistent) debug fen on the "d"
command. The implementation follows
https://en.wikipedia.org/wiki/X-FEN#Encoding_en-passant
following x-fen, it is "valid" to record ep even if ep would put king en prise.

fixes #2784

closes https://github.com/official-stockfish/Stockfish/pull/2797

No functional change

2 months agoAdd a rank based bonus for blocked pawns.
Stefan Geschwentner [Mon, 6 Jul 2020 07:30:23 +0000 (09:30 +0200)]
Add a rank based bonus for blocked pawns.

Fix for overevaluated blocked pawns on the 5th and 6th rank.
This is a rewrite of the original idea that uses only two parameters.
Thanks to rocky640 for pointing this out.

STC:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 50800 W: 9707 L: 9446 D: 31647
Ptnml(0-2): 831, 5851, 11822, 6018, 878
https://tests.stockfishchess.org/tests/view/5f00b4f359f6f03532894304

LTC:
LLR: 2.93 (-2.94,2.94) {0.25,1.75}
Total: 52064 W: 6477 L: 6167 D: 39420
Ptnml(0-2): 331, 4628, 15834, 4878, 361
https://tests.stockfishchess.org/tests/view/5f0115fe59f6f03532894345

closes https://github.com/official-stockfish/Stockfish/pull/2794

Bench: 4882833

2 months agodenormalize KRKP.
protonspring [Tue, 30 Jun 2020 16:17:50 +0000 (10:17 -0600)]
denormalize KRKP.

a non-functional code style change that denormalizes the KRKP endgame,
making it somewhat easier to read.

closes https://github.com/official-stockfish/Stockfish/pull/2786

No functional change

2 months agoRemove passed pawn condition.
SFisGOD [Wed, 1 Jul 2020 16:11:23 +0000 (00:11 +0800)]
Remove passed pawn condition.

This will help scale down relatively high eval in drawish rook endgames with passed pawn like in TCEC S18 Superfinal Game 90.

Passed STC
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 50456 W: 9644 L: 9540 D: 31272
Ptnml(0-2): 760, 5637, 12332, 5737, 762
https://tests.stockfishchess.org/tests/view/5efcb76e59f6f035328940ed

Passed LTC
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 77264 W: 9518 L: 9518 D: 58228
Ptnml(0-2): 402, 6766, 24321, 6716, 427
https://tests.stockfishchess.org/tests/view/5efd2ad759f6f03532894143

closes https://github.com/official-stockfish/Stockfish/pull/2792

Bench: 4431626

2 months agoSet UCI_ShowWDL by default to false
Joost VandeVondele [Thu, 2 Jul 2020 16:58:37 +0000 (18:58 +0200)]
Set UCI_ShowWDL by default to false

UCI_ShowWDL might not be shown by GUIs that don't know the option,
but crash on the WDL output, effectively making it hard for users to
turn it off and run the engine. This sets it by default to false.

fixes https://github.com/official-stockfish/Stockfish/issues/2787

closes https://github.com/official-stockfish/Stockfish/pull/2788

No functional change.

2 months agoUse arrays
Alain SAVARD [Wed, 1 Jul 2020 06:12:59 +0000 (02:12 -0400)]
Use arrays

for safe checks, outposts and king protectors in evaluate.cpp

Tested for non regression on the safe checks
https://tests.stockfishchess.org/tests/view/5ef8b75c020eec13834a9596
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 22256 W: 4283 L: 4143 D: 13830
Ptnml(0-2): 291, 2439, 5588, 2459, 351

Tested for non regression on the safe checks, outposts and king protectors
https://tests.stockfishchess.org/tests/view/5ef8e543020eec13834a95e7
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 28400 W: 5382 L: 5253 D: 17765
Ptnml(0-2): 394, 3078, 7119, 3223, 386

closes https://github.com/official-stockfish/Stockfish/pull/2785

No functional change

2 months agoProvide WDL statistics
Joost VandeVondele [Sat, 27 Jun 2020 19:29:29 +0000 (21:29 +0200)]
Provide WDL statistics

A number of engines, GUIs and tournaments start to report WDL estimates
along or instead of scores. This patch enables reporting of those stats
in a more or less standard way (http://www.talkchess.com/forum3/viewtopic.php?t=72140)

The model this reporting uses is based on data derived from a few million fishtest LTC games,
given a score and a game ply, a win rate is provided that matches rather closely,
especially in the intermediate range [0.05, 0.95] that data. Some data is shown at
https://github.com/glinscott/fishtest/wiki/UsefulData#win-loss-draw-statistics-of-ltc-games-on-fishtest
Making the conversion game ply dependent is important for a good fit, and is in line
with experience that a +1 score in the early midgame is more likely a win than in the late endgame.

Even when enabled, the printing of the info causes no significant overhead.

Passed STC:
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 197112 W: 37226 L: 37347 D: 122539
Ptnml(0-2): 2591, 21025, 51464, 20866, 2610
https://tests.stockfishchess.org/tests/view/5ef79ef4f993893290cc146b

closes https://github.com/official-stockfish/Stockfish/pull/2778

No functional change

2 months agoTweak single queen endgame scaling.
Stefan Geschwentner [Mon, 29 Jun 2020 17:35:24 +0000 (19:35 +0200)]
Tweak single queen endgame scaling.

Increase scaling factor for each minor of the opponent side of the queen.

STC:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 14528 W: 2860 L: 2653 D: 9015
Ptnml(0-2): 217, 1632, 3408, 1741, 266
https://tests.stockfishchess.org/tests/view/5ef98384020eec13834a96a0

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.75}
Total: 34584 W: 4371 L: 4111 D: 26102
Ptnml(0-2): 205, 3080, 10501, 3262, 244
https://tests.stockfishchess.org/tests/view/5ef99972020eec13834a96c9

closes https://github.com/official-stockfish/Stockfish/pull/2782

Bench: 4523573

2 months agoScale down eval for queen imbalance
Stéphane Nicolet [Sun, 28 Jun 2020 20:24:57 +0000 (22:24 +0200)]
Scale down eval for queen imbalance

We lower the endgame value of the evaluation when we detect that there
is only one queen left on the board (more precisely, we use a scale
factor of 37/64, or about 0.58, for the endgame part of the evaluation).
Hopefully this helps a little bit for the assessment of positions with
queen imbalance, which are one of the well-known Stockfish weaknesses.

STC:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 21600 W: 4176 L: 3955 D: 13469
Ptnml(0-2): 351, 2457, 5003, 2598, 391
https://tests.stockfishchess.org/tests/view/5ef871b6020eec13834a94e8

LTC:
LLR: 2.97 (-2.94,2.94) {0.25,1.75}
Total: 248328 W: 30596 L: 29720 D: 188012
Ptnml(0-2): 1544, 22345, 75665, 22911, 1699
https://tests.stockfishchess.org/tests/view/5ef87aec020eec13834a94fe

Closes https://github.com/official-stockfish/Stockfish/pull/2781

Bench: 4441323

2 months agoScale down eval for drawish rook endgames.
Stefan Geschwentner [Sun, 28 Jun 2020 14:28:55 +0000 (16:28 +0200)]
Scale down eval for drawish rook endgames.

STC:
LLR: 2.96 (-2.94,2.94) {-0.50,1.50}
Total: 82136 W: 15694 L: 15407 D: 51035
Ptnml(0-2): 1076, 8960, 20767, 9131, 1134
https://tests.stockfishchess.org/tests/view/5ef86cf8020eec13834a94dd

LTC:
LLR: 2.93 (-2.94,2.94) {0.25,1.75}
Total: 70200 W: 8787 L: 8440 D: 52973
Ptnml(0-2): 325, 5983, 22170, 6264, 358
https://tests.stockfishchess.org/tests/view/5ef88225020eec13834a950a

closes https://github.com/official-stockfish/Stockfish/pull/2780

Bench: 4478869

2 months agoIncrease value of pawns on fifth rank
Stéphane Nicolet [Sun, 28 Jun 2020 04:00:28 +0000 (06:00 +0200)]
Increase value of pawns on fifth rank

This patch increases the endgame value of pawns on the fifth rank.
The increase is very small (+1 evaluation point, about 0.005 pawn)
for the pawns on external columns (a-b-c-f-g-h) and a bit bigger
(+7 evaluation points, about 0.033 pawn) for the pawns on d5/e5.

STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 79864 W: 15331 L: 15027 D: 49506
Ptnml(0-2): 1336, 9284, 18433, 9498, 1381
https://tests.stockfishchess.org/tests/view/5ef73e2ef993893290cc0c47

LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 47240 W: 5927 L: 5630 D: 35683
Ptnml(0-2): 320, 4133, 14440, 4384, 343
https://tests.stockfishchess.org/tests/view/5ef7c0c4f993893290cc14b7

closes https://github.com/official-stockfish/Stockfish/pull/2776

Bench: 4794633

2 months agoRemove old zobrist trick for castling rights
mstembera [Fri, 26 Jun 2020 05:08:17 +0000 (22:08 -0700)]
Remove old zobrist trick for castling rights

Removes an 8 year old micro optimization aimed at 32-bit architectures
because back then doing an xor of a Key could not be done in one instruction.
See original commit here 821e1c7

STC https://tests.stockfishchess.org/tests/view/5ef5833dde213bf647527d0c
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 162648 W: 31053 L: 31097 D: 100498
Ptnml(0-2): 2841, 18966, 37715, 19000, 2802

LTC https://tests.stockfishchess.org/tests/view/5ef7b1bbf993893290cc1489
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 62360 W: 7617 L: 7586 D: 47157
Ptnml(0-2): 423, 5662, 18994, 5663, 438

closes https://github.com/official-stockfish/Stockfish/pull/2775

bench: 4591425

2 months agoFix fragile code to use proper random 64 bit keys.
mstembera [Sat, 27 Jun 2020 00:26:46 +0000 (17:26 -0700)]
Fix fragile code to use proper random 64 bit keys.

This fixes an old issue where we want to make a position unique but only
change a small number of bits in the key instead of all 64 of them randomly.
This is fragile and can lead to non uniqueness issues in the TT.

Key make_key(uint64_t seed) takes any integer and produces a unique random 64 bit key.
It is computationally efficient and is based on a congruential pseudo random number
generator using well tested constants by Donald Knuth
(see https://en.wikipedia.org/wiki/Linear_congruential_generator)

STC https://tests.stockfishchess.org/tests/view/5ef6c78f761b685b4c724bb6
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 154320 W: 29343 L: 29376 D: 95601
Ptnml(0-2): 2543, 18170, 35891, 17889, 2667

LTC https://tests.stockfishchess.org/tests/view/5ef7d1a9020eec13834a940e
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 53488 W: 6629 L: 6584 D: 40275
Ptnml(0-2): 372, 4878, 16183, 4955, 356

closes https://github.com/official-stockfish/Stockfish/pull/2773

bench: 4626776

2 months agoRevert LTO for mingw on windows.
Joost VandeVondele [Sat, 27 Jun 2020 06:23:46 +0000 (08:23 +0200)]
Revert LTO for mingw on windows.

LTO with static linking is still only working with the latest versions of gcc,
causing problems for some devs.

on a modern mingw toolchain LTO optimizations can still be enabled as:

```
CXXFLAGS='-flto' make -j ARCH=x86-64-modern COMP=mingw profile-build
```

fixes https://github.com/official-stockfish/Stockfish/issues/2769

closes https://github.com/official-stockfish/Stockfish/pull/2774

No functional change.

2 months agosupport popcnt on armv8
Niklas Fiekas [Thu, 25 Jun 2020 10:42:25 +0000 (12:42 +0200)]
support popcnt on armv8

* Supports popcnt (thanks @daylen)
* bits = 64 is now the default

Tested with g++ (Ubuntu/Linaro 7.5.0-3ubuntu1~18.04) 7.5.0 on ThunderX CN8890,
yields about 9% speedup.

Also tested with clang version 6.0.0-1ubuntu2 (tags/RELEASE_600/final).

closes https://github.com/official-stockfish/Stockfish/pull/2770

No functional change.

2 months agoSimplify use_time_management() and allow composing
Niklas Fiekas [Wed, 24 Jun 2020 21:03:00 +0000 (23:03 +0200)]
Simplify use_time_management() and allow composing

like other `go` commands wtime/btime can now be composed.

`go depth 10 wtime 100`

will let the engine use standard time management but stop if depth 10 is reached.

fixes https://github.com/official-stockfish/Stockfish/issues/2767

closes https://github.com/official-stockfish/Stockfish/pull/2768

No functional change

2 months agoSmall cleanups
Joost VandeVondele [Wed, 24 Jun 2020 20:19:58 +0000 (22:19 +0200)]
Small cleanups

closes https://github.com/official-stockfish/Stockfish/pull/2756

No functional change

2 months agoMakefile: support lto on mingw, default to 64bits
UnaiCorzo [Tue, 23 Jun 2020 15:56:38 +0000 (17:56 +0200)]
Makefile: support lto on mingw, default to 64bits

Clean and organize uppercase and spaces

fixes https://github.com/official-stockfish/Stockfish/issues/2731

closes  https://github.com/official-stockfish/Stockfish/pull/2763

No functional change

2 months agoRemove 'Minimum Thinking Time' UCI option.
Joost VandeVondele [Wed, 24 Jun 2020 14:23:31 +0000 (16:23 +0200)]
Remove 'Minimum Thinking Time' UCI option.

the option was, since at least 2014, not correctly implemented,
ignoring all dynamic adjustments to optimum time in search.
Instead of fixing it, remove it, no need to expose an option that
will influence time management negatively.

closes https://github.com/official-stockfish/Stockfish/pull/2765

No functional change.

2 months agoSupport ARCH=armv8 in Makefile (#2355)
Niklas Fiekas [Tue, 23 Jun 2020 08:41:53 +0000 (10:41 +0200)]
Support ARCH=armv8 in Makefile (#2355)

Tested with bench run after compiling with

- g++ (Debian 6.3.0-18+deb9u1) 6.3.0 20170516
- clang version 3.8.1-24

on ThunderX CN8890.

closes https://github.com/official-stockfish/Stockfish/pull/2760

fixes https://github.com/official-stockfish/Stockfish/issues/2355

No functional change.

2 months agoDo less futility pruning for captures.
Vizvezdenec [Tue, 23 Jun 2020 11:55:52 +0000 (14:55 +0300)]
Do less futility pruning for captures.

The idea of this patch is that if capture can be described as
"less valuable piece takes more valuable piece" it's not really correct
to add only piece value of captured piece to static evaluation
since there can be more threats in other places and opponent can't really
do much but recapture our capturing piece which leaves us space for
more captures thus winning more material and increasing static eval.

passed STC
https://tests.stockfishchess.org/tests/view/5ef0167b122d6514328d760f
LLR: 2.96 (-2.94,2.94) {-0.50,1.50}
Total: 24736 W: 4838 L: 4607 D: 15291
Ptnml(0-2): 438, 2812, 5648, 3021, 449

passed LTC
https://tests.stockfishchess.org/tests/view/5ef073bc122d6514328d7693
LLR: 2.93 (-2.94,2.94) {0.25,1.75}
Total: 46152 W: 5865 L: 5567 D: 34720
Ptnml(0-2): 312, 4160, 13886, 4354, 364

closes https://github.com/official-stockfish/Stockfish/pull/2761

bench 4789930

2 months agoIntroduce bonus for queen infiltration
FauziAkram [Mon, 22 Jun 2020 09:52:31 +0000 (12:52 +0300)]
Introduce bonus for queen infiltration

Idea is that queen feels much better when it can't be kicked away now
or later by pawn moves, especially in endgame. Special thanks to
Linmiao Xu for the original idea of this patch.

passed STC:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 84008 W: 16271 L: 15958 D: 51779
Ptnml(0-2): 1476, 9688, 19420, 9887, 1533
https://tests.stockfishchess.org/tests/view/5eee7ca0447c5b640047a439

passed LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.75}
Total: 11720 W: 1522 L: 1328 D: 8870
Ptnml(0-2): 52, 1021, 3574, 1107, 106
https://tests.stockfishchess.org/tests/view/5eefc588122d6514328d75f9

closes https://github.com/official-stockfish/Stockfish/pull/2759

Bench: 4471740

2 months agoMake endgames consistent
protonspring [Wed, 17 Jun 2020 21:15:54 +0000 (15:15 -0600)]
Make endgames consistent

Changes variable names and occasionally consolidated variable declarations.
Piece squares are consistently prefixed with "weak" or "strong."

passed STC
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 29008 W: 5532 L: 5416 D: 18060
Ptnml(0-2): 355, 2983, 7723, 3077, 366
https://tests.stockfishchess.org/tests/view/5eea88d3563bc7aa75600689

closes https://github.com/official-stockfish/Stockfish/pull/2752

No functional change

2 months agosmall cleanups
Joost VandeVondele [Sun, 21 Jun 2020 13:21:46 +0000 (15:21 +0200)]
small cleanups

closes https://github.com/official-stockfish/Stockfish/pull/2695

No functional change

3 months agoMore reduction for evading pawn moves.
Vizvezdenec [Wed, 17 Jun 2020 02:36:30 +0000 (05:36 +0300)]
More reduction for evading pawn moves.

pawn moves are irreversable unlike other evading moves; pawn is the least valuable piece in the game.
So it makes a lot of sence to assume that evading pawn moves are on average not as good as other evading moves thus can be reduced more.

Passed STC
https://tests.stockfishchess.org/tests/view/5ee9602e563bc7aa756002dc
LLR: 2.95 (-2.94,2.94) {-0.50,1.50}
Total: 94176 W: 17993 L: 17668 D: 58515
Ptnml(0-2): 1634, 10742, 21989, 11111, 1612

Passed LTC
https://tests.stockfishchess.org/tests/view/5ee97342563bc7aa75600301
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 20432 W: 2572 L: 2354 D: 15506
Ptnml(0-2): 146, 1707, 6280, 1949, 134

closes https://github.com/official-stockfish/Stockfish/pull/2749

Bench: 5073064

3 months agoSmall bonus to favor thorn pawns
Stéphane Nicolet [Tue, 16 Jun 2020 11:21:24 +0000 (13:21 +0200)]
Small bonus to favor thorn pawns

We increase a little bit the midgame value of pawns on a4, h4, a6 and h6.

Original idea by Malcolm Campbell, who tried the version restricted to the
pawns on the H column a couple of weeks ago and got a patch which almost
passed LTC. The current pull request just adds the same idea for pawns on
the A column.

Possible follow-ups: maybe tweak the a5/h5 pawn values, and/or add a malus
for very low king mobility in midgame?

STC:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 33416 W: 6516 L: 6275 D: 20625
Ptnml(0-2): 575, 3847, 7659, 4016, 611
https://tests.stockfishchess.org/tests/view/5ee6c4e687586124bc2c10d4

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.75}
Total: 134368 W: 16869 L: 16319 D: 101180
Ptnml(0-2): 908, 12083, 40708, 12521, 964
https://tests.stockfishchess.org/tests/view/5ee74e60aae8aec816ab756a

closes https://github.com/official-stockfish/Stockfish/pull/2747

Bench: 5299456

3 months agoUse 128 bit multiply for TT index
mstembera [Mon, 15 Jun 2020 06:35:07 +0000 (23:35 -0700)]
Use 128 bit multiply for TT index

Remove super cluster stuff from TT and just use a 128 bit multiply.

STC https://tests.stockfishchess.org/tests/view/5ee719b3aae8aec816ab7548
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 12736 W: 2502 L: 2333 D: 7901
Ptnml(0-2): 191, 1452, 2944, 1559, 222

LTC https://tests.stockfishchess.org/tests/view/5ee732d1aae8aec816ab7556
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 27584 W: 3431 L: 3350 D: 20803
Ptnml(0-2): 173, 2500, 8400, 2511, 208

Scheme back to being derived from https://lemire.me/blog/2016/06/27/a-fast-alternative-to-the-modulo-reduction/

Also the default optimized version of the index calculation now uses fewer instructions.
https://godbolt.org/z/Tktxbv
Might benefit from mulx (requires -mbmi2)

closes https://github.com/official-stockfish/Stockfish/pull/2744

bench: 4320954

3 months agoRetuned values after eval quantize patch.
xoto10 [Sun, 14 Jun 2020 20:50:27 +0000 (21:50 +0100)]
Retuned values after eval quantize patch.

The last search tune patch was tested before the implementation of #2733 which
presumably changed the search characteristics noticeably. Another tuning run was
done, see https://tests.stockfishchess.org/tests/view/5ee5b434ca6c451633a9a08c
and the updated values passed these tests:

STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 34352 W: 6600 L: 6360 D: 21392
Ptnml(0-2): 581, 3947, 7914, 4119, 615
https://tests.stockfishchess.org/tests/view/5ee62f05ca6c451633a9a15f

LTC 60+0.6 th 1 :
LLR: 2.97 (-2.94,2.94) {0.25,1.75}
Total: 11176 W: 1499 L: 1304 D: 8373
Ptnml(0-2): 69, 933, 3403, 1100, 83
https://tests.stockfishchess.org/tests/view/5ee6205bca6c451633a9a147

SMP LTC 20+0.2 th 8 :
LLR: 2.93 (-2.94,2.94) {0.25,1.75}
Total: 54032 W: 6126 L: 5826 D: 42080
Ptnml(0-2): 278, 4454, 17280, 4698, 306
https://tests.stockfishchess.org/tests/view/5ee62f25ca6c451633a9a162

Closes https://github.com/official-stockfish/Stockfish/pull/2742

Bench 4957812

3 months agoTuned values for search constants
xoto10 [Sat, 13 Jun 2020 08:54:07 +0000 (09:54 +0100)]
Tuned values for search constants

Tuned search constants after many search patches since the last
successful tune.

1st LTC @ 60+0.6 th 1 :
LLR: 2.97 (-2.94,2.94) {0.25,1.75}
Total: 57656 W: 7369 L: 7036 D: 43251
Ptnml(0-2): 393, 5214, 17336, 5437, 448
https://tests.stockfishchess.org/tests/view/5ee1e074f29b40b0fc95af19

SMP LTC @ 20+0.2 th 8 :
LLR: 2.95 (-2.94,2.94) {0.25,1.75}
Total: 83576 W: 9731 L: 9341 D: 64504
Ptnml(0-2): 464, 7062, 26369, 7406, 487
https://tests.stockfishchess.org/tests/view/5ee35a21f29b40b0fc95b008

The changes were rebased on top of a successful patch by Viz (see #2734)
and two different ways of doing this were tested. The successful test
modified the constants in the patch by Viz in a similar manner to the
tuning run:

LTC (rebased) @ 60+0.6 th 1 :
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 193384 W: 24241 L: 23521 D: 145622
Ptnml(0-2): 1309, 17497, 58472, 17993, 1421
https://tests.stockfishchess.org/tests/view/5ee43319ca6c451633a995f9

Further work: the recent patch to quantize eval #2733 affects search quit
quite a bit, so doing another tune in, say, three months time might be a
good idea.

closes https://github.com/official-stockfish/Stockfish/pull/2735

Bench 4246971

3 months agoQuantize eval to multiples of 16
Joost VandeVondele [Wed, 10 Jun 2020 11:19:21 +0000 (13:19 +0200)]
Quantize eval to multiples of 16

Removes some excess precision, helps searchs.

Effectively reintroduces evaluation grain, with a slightly different context.
https://github.com/official-stockfish/Stockfish/commit/45dbd9cd0303d0db469670af8ec3598731a4eace

passed STC
LLR: 2.97 (-2.94,2.94) {-0.50,1.50}
Total: 197032 W: 37938 L: 37462 D: 121632
Ptnml(0-2): 3359, 22994, 45446, 23246, 3471
https://tests.stockfishchess.org/tests/view/5ee0c228f29b40b0fc95ae53

passed LTC
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 77696 W: 9970 L: 9581 D: 58145
Ptnml(0-2): 530, 7075, 23311, 7340, 592
https://tests.stockfishchess.org/tests/view/5ee21426f29b40b0fc95af43

passed LTC SMP
LLR: 2.96 (-2.94,2.94) {0.25,1.75}
Total: 64136 W: 7425 L: 7091 D: 49620
Ptnml(0-2): 345, 5416, 20228, 5718, 361
https://tests.stockfishchess.org/tests/view/5ee387bbf29b40b0fc95b04c

closes https://github.com/official-stockfish/Stockfish/pull/2733

Bench: 4939103

3 months agoAdjust history threshold for quiet moves futility pruning
Vizvezdenec [Sat, 13 Jun 2020 02:03:59 +0000 (05:03 +0300)]
Adjust history threshold for quiet moves futility pruning

This patch adjusts the threshold for futility pruning of quiet moves
using the continuation history array contHist[5], in the same way as
it is used in movepicker.

passed STC:
https://tests.stockfishchess.org/tests/view/5ee3f88bca6c451633a9959f
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 55984 W: 10822 L: 10552 D: 34610
Ptnml(0-2): 952, 6435, 12941, 6719, 945

passed LTC:
https://tests.stockfishchess.org/tests/view/5ee4186dca6c451633a995cf
LLR: 2.96 (-2.94,2.94) {0.25,1.75}
Total: 41712 W: 5402 L: 5114 D: 31196
Ptnml(0-2): 293, 3766, 12469, 4016, 312

closes https://github.com/official-stockfish/Stockfish/pull/2734

Bench: 4715960

3 months agoImprove the anti-shuffling policy
Stéphane Nicolet [Tue, 9 Jun 2020 22:10:07 +0000 (00:10 +0200)]
Improve the anti-shuffling policy

We replace the current decrease of the complexity term in initiative
when shuffling by a direct damping of the evaluation. This scheme may
have two benefits over the initiative approach:

a) the damping effect is more brutal for fortresses with heavy pieces
   on the board, because the initiative term is almost an endgame term;

b) the initiative implementation had a funny side effect, almost a bug,
   in the rare positions where mg > 0, eg < 0 and the tampered eval
   returned a positive value (ie with heavy pieces still on the board):
   sending eg to zero via shuffling would **increase** the tampered
   eval instead of decreasing it, which is somewhat illogical. This
   patch avoids this phenomenon.

STC:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 43072 W: 8373 L: 8121 D: 26578
Ptnml(0-2): 729, 4954, 9940, 5162, 751
https://tests.stockfishchess.org/tests/view/5ee008ebf29b40b0fc95ade2

LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 37376 W: 4816 L: 4543 D: 28017
Ptnml(0-2): 259, 3329, 11286, 3508, 306
https://tests.stockfishchess.org/tests/view/5ee03b06f29b40b0fc95ae0c

Closes https://github.com/official-stockfish/Stockfish/pull/2727

Bench: 4757174

3 months agoIncrease the maximum hash size by a factor of 256
Sami Kiminki [Fri, 5 Jun 2020 17:17:00 +0000 (20:17 +0300)]
Increase the maximum hash size by a factor of 256

Conceptually group hash clusters into super clusters of 256 clusters.
This scheme allows us to use hash sizes up to 32 TB
(= 2^32 super clusters = 2^40 clusters).

Use 48 bits of the Zobrist key to choose the cluster index. We use 8
extra bits to mitigate the quantization error for very large hashes when
scaling the hash key to cluster index.

The hash index computation is organized to be compatible with the existing
scheme for power-of-two hash sizes up to 128 GB.

Fixes https://github.com/official-stockfish/Stockfish/issues/1349

closes https://github.com/official-stockfish/Stockfish/pull/2722

Passed non-regression STC:
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 37976 W: 7336 L: 7211 D: 23429
Ptnml(0-2): 578, 4295, 9149, 4356, 610
https://tests.stockfishchess.org/tests/view/5edcbaaef29b40b0fc95abc5

No functional change.

3 months agoImprove Readme.md about compiling
nguyenpham [Sun, 7 Jun 2020 22:49:27 +0000 (08:49 +1000)]
Improve Readme.md about compiling

Reparagraph, add an example how to compile on Unix-like systems

closes https://github.com/official-stockfish/Stockfish/pull/2724

No functional change

3 months agoshow coordinates when displaying board
NguyenPham [Sun, 7 Jun 2020 21:48:38 +0000 (07:48 +1000)]
show coordinates when displaying board

closes https://github.com/official-stockfish/Stockfish/pull/2723

No functional change

3 months agoRefactor some threads related code.
protonspring [Mon, 1 Jun 2020 05:31:14 +0000 (23:31 -0600)]
Refactor some threads related code.

This is a code style change that moves some pure thread code into the threads class.

It is a bit more code, but it makes search.cpp cleaner and easier to read by hiding some thread specific functionality.

STC (SMP)
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 75896 W: 12073 L: 12026 D: 51797
Ptnml(0-2): 828, 8224, 19872, 8121, 903
https://tests.stockfishchess.org/tests/view/5ed492e8f29b40b0fc95a74c

closes https://github.com/official-stockfish/Stockfish/pull/2720

No functional change.

3 months agoWrap all access to LineBB and add assert
protonspring [Sun, 7 Jun 2020 03:25:32 +0000 (21:25 -0600)]
Wrap all access to LineBB and add assert

This is a non-functional code style change which provides a safe access handler for LineBB.
Also includes an assert in debug mode to verify square correctness.

closes https://github.com/official-stockfish/Stockfish/pull/2719

No functional change

3 months agojoin scale_factor, initiative and mg+eg reduction
xoto10 [Wed, 3 Jun 2020 10:05:58 +0000 (11:05 +0100)]
join scale_factor, initiative and mg+eg reduction

Merging this code into one function `winnable()`.
Should allow common concepts used to adjust the eg value,
either by addition or scaling, to be combined more effectively.

Improve trace function.

closes https://github.com/official-stockfish/Stockfish/pull/2710

No functional change.

3 months agoUse lowply-history also on low depths
pb00067 [Sat, 6 Jun 2020 10:56:38 +0000 (12:56 +0200)]
Use lowply-history also on low depths

STC:
https://tests.stockfishchess.org/tests/view/5ed75078f29b40b0fc95a8b9
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 73928 W: 14301 L: 14005 D: 45622
Ptnml(0-2): 1243, 8572, 17096, 8752, 1301

LTC:
https://tests.stockfishchess.org/tests/view/5ed895e0f29b40b0fc95a976
LLR: 2.93 (-2.94,2.94) {0.25,1.75}
Total: 154848 W: 19684 L: 19074 D: 116090
Ptnml(0-2): 1048, 14108, 46627, 14468, 1173

closes https://github.com/official-stockfish/Stockfish/pull/2718

bench: 4582693

3 months agosmall speed-up in movegen
protonspring [Thu, 4 Jun 2020 00:06:49 +0000 (18:06 -0600)]
small speed-up in movegen

pass color as a template parameter.

closes https://github.com/official-stockfish/Stockfish/pull/2715

No functional change.

3 months agoMinimal thinking time, even if only one rootMove.
Moez Jellouli [Sun, 31 May 2020 14:51:38 +0000 (16:51 +0200)]
Minimal thinking time, even if only one rootMove.

without search, the eval returned can be misleading (e.g. mate instead of draw),
leading to wrong adjudication. With a minimal search, this is avoided.
This patch leads to 1ms long searches if there is only 1 move,
similar patches all indicate a small Elo gain.

Fixes https://github.com/official-stockfish/Stockfish/issues/2707

Passed non-regression STC:
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 22312 W: 4350 L: 4204 D: 13758
Ptnml(0-2): 323, 2488, 5437, 2536, 372
https://tests.stockfishchess.org/tests/view/5ed562b0f29b40b0fc95a7d0

closes https://github.com/official-stockfish/Stockfish/pull/2709

Bench: 4733799

3 months agoSingular quiet LMR
Moez Jellouli [Tue, 2 Jun 2020 20:27:11 +0000 (22:27 +0200)]
Singular quiet LMR

If ttMove is a capture and had a singular extension, it is probably the best move.
No need to make a decrease of LMR on other moves.

STC
LLR: 2.96 (-2.94,2.94) {-0.50,1.50}
Total: 41968 W: 8170 L: 7918 D: 25880
Ptnml(0-2): 733, 4770, 9726, 5022, 733
https://tests.stockfishchess.org/tests/view/5ed6b666f29b40b0fc95a884

LTC
LLR: 2.95 (-2.94,2.94) {0.25,1.75}
Total: 71376 W: 9200 L: 8827 D: 53349
Ptnml(0-2): 486, 6544, 21342, 6743, 573
https://tests.stockfishchess.org/tests/view/5ed7578bf29b40b0fc95a8c9

closes https://github.com/official-stockfish/Stockfish/pull/2713

Bench: 4733799

3 months agoGive bonus for bishops that are alligned with enemy kingring.
Stefan Geschwentner [Sun, 31 May 2020 15:00:47 +0000 (17:00 +0200)]
Give bonus for bishops that are alligned with enemy kingring.

Inspired by the succesful patch "Give bonus for rooks that are alligned with enemy kingring" from Vizvezdenec,
this idea has been reused for bishops. Here, we only consider attacks that are not blocked by any pawn.
Also we have a 50% higher bonus than for the rooks.

STC:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 68960 W: 13495 L: 13202 D: 42263
Ptnml(0-2): 1213, 8018, 15802, 8157, 1290
https://tests.stockfishchess.org/tests/view/5ed27495042fa6d77c3553aa

LTC:
LLR: 2.96 (-2.94,2.94) {0.25,1.75}
Total: 54560 W: 7105 L: 6780 D: 40675
Ptnml(0-2): 379, 4986, 16254, 5253, 408
https://tests.stockfishchess.org/tests/view/5ed30375596e6dc1e1f97425

closes https://github.com/official-stockfish/Stockfish/pull/2708

Bench: 4860021

3 months agoIsolated pawns tweak
ElbertoOne [Sun, 31 May 2020 14:39:03 +0000 (16:39 +0200)]
Isolated pawns tweak

Give opposed doubled isolated pawns only the Doubled penalty.
The other isolated pawns get the Isolated penalty and the WeakUnopposed penalty.
The popcount condition has been replaced with an opposed check,
which is non-functional, but probably gives a speed-up.

Passed STC (https://tests.stockfishchess.org/tests/view/5ed0f0f0042fa6d77c3552f5):
LLR: 2.96 (-2.94,2.94) {-0.50,1.50}
Total: 121832 W: 23562 L: 23195 D: 75075
Ptnml(0-2): 2092, 14064, 28313, 14279, 2168

LTC: (https://tests.stockfishchess.org/tests/view/5ed22e40042fa6d77c355387)
LLR: 2.96 (-2.94,2.94) {0.25,1.75}
Total: 103368 W: 13232 L: 12768 D: 77368
Ptnml(0-2): 693, 9484, 30919, 9842, 746

closes https://github.com/official-stockfish/Stockfish/pull/2706

Bench: 4085694

3 months agoLess pruning in draw PV lines.
Joost VandeVondele [Thu, 28 May 2020 20:34:43 +0000 (22:34 +0200)]
Less pruning in draw PV lines.

no futility pruning for certain captures if the PvNode has a draw eval.

passed STC:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 59392 W: 11576 L: 11302 D: 36514
Ptnml(0-2): 977, 6816, 13920, 6922, 1061
https://tests.stockfishchess.org/tests/view/5ed0b1bb042fa6d77c355295

passed LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 64040 W: 8273 L: 7923 D: 47844
Ptnml(0-2): 424, 5842, 19220, 6028, 506
https://tests.stockfishchess.org/tests/view/5ed145e0042fa6d77c35531c

closes https://github.com/official-stockfish/Stockfish/pull/2705

Bench: 4704615

3 months agoConsolidate all attacks bitboards
protonspring [Thu, 28 May 2020 15:48:31 +0000 (09:48 -0600)]
Consolidate all attacks bitboards

This is a non-functional simplification that simplifies getting attacks bitboards.

* consolidates all attacks to attacks_bb (remove Position::attacks_from(..)).
* attacks_bb<PieceType>(square) gets pseudo attacks
* attacks_bb<PieceType>(square, bitboard) gets attacks considering occupied squares in the bitboard).
* pawn_attacks_bb(Color, Square) gets pawn attacks like other pawn attack bitboards.
* Wraps all access to PawnAttacks arrays and PseudoAttacks arrays and adds asserts as appropriate.

Passed STC
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 90208 W: 17533 L: 17482 D: 55193
Ptnml(0-2): 1412, 10232, 21798, 10217, 1445
https://tests.stockfishchess.org/tests/view/5ece996275787cc0c05d9790

closes https://github.com/official-stockfish/Stockfish/pull/2703

No functional change

3 months agoIn BlockedStorm, theirPawns includes ones attacked by us.
xoto10 [Mon, 25 May 2020 23:27:05 +0000 (00:27 +0100)]
In BlockedStorm, theirPawns includes ones attacked by us.

Pawns heading towards our king tend to be dangerous whether or not we are attacking them so remove this test.

STC:
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 91184 W: 18196 L: 18137 D: 54851
Ptnml(0-2): 1580, 10656, 21092, 10653, 1611
https://tests.stockfishchess.org/tests/view/5ecc3f7080f2c838b9655841

LTC:
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 14152 W: 2045 L: 1937 D: 10170
Ptnml(0-2): 99, 1325, 4130, 1413, 109
https://tests.stockfishchess.org/tests/view/5ecc4f3180f2c838b9655861

closes https://github.com/official-stockfish/Stockfish/pull/2702

Bench 4828973

3 months agoGive bonus for rooks that are alligned with enemy kingring
Vizvezdenec [Mon, 25 May 2020 18:14:07 +0000 (21:14 +0300)]
Give bonus for rooks that are alligned with enemy kingring

The idea of this patch is that if rooks are not directly attacking the opponent king,
they can support king attacks staying behind pawns or minor pieces and be really
deadly if position slightly opens up at enemy king ring ranks. Loosely based on
some stockfish games where it underestimated attacks on it king when enemy has one
or two rooks supporting pawn pushes towards it king.

passed STC
https://tests.stockfishchess.org/tests/view/5ecb093680f2c838b96550f9
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 53672 W: 10535 L: 10265 D: 32872
Ptnml(0-2): 952, 6210, 12258, 6448, 968

passed LTC
https://tests.stockfishchess.org/tests/view/5ecb639f80f2c838b9655117
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 62424 W: 8094 L: 7748 D: 46582
Ptnml(0-2): 426, 5734, 18565, 6042, 445

closes https://github.com/official-stockfish/Stockfish/pull/2700

Bench: 4663220

3 months agoRemove attacked pawns from storm evaluation
Moez Jellouli [Sat, 23 May 2020 23:54:37 +0000 (01:54 +0200)]
Remove attacked pawns from storm evaluation

STC:
LLR: 2.96 (-2.94,2.94) {-0.50,1.50}
Total: 54456 W: 11009 L: 10737 D: 32710
Ptnml(0-2): 929, 6326, 12523, 6444, 1006
https://tests.stockfishchess.org/tests/view/5ec962e4404591b2793008a5

LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 62448 W: 9018 L: 8664 D: 44766
Ptnml(0-2): 462, 5928, 18121, 6220, 493
https://tests.stockfishchess.org/tests/view/5ec976a8a586eee45aa2ab40

Non regression STC with "noob_3moves.epd" opening book
LLR: 3.81 (-2.94,2.94) {-1.50,0.50}
Total: 91896 W: 17770 L: 17653 D: 56473
Ptnml(0-2): 1598, 10782, 21124, 10793, 1651
https://tests.stockfishchess.org/tests/view/5ec9b83ea586eee45aa2ab96

closes https://github.com/official-stockfish/Stockfish/pull/2698

Bench 4488597

3 months agoRemove and replace DoubledIsolated penalty by Doubled
ElbertoOne [Sat, 23 May 2020 11:14:02 +0000 (13:14 +0200)]
Remove and replace DoubledIsolated penalty by Doubled

The values for both penalties were very close, so DoubledIsolated can be removed and replaced by Doubled.

Passed STC (simplification): https://tests.stockfishchess.org/tests/view/5ec7c18e2a585b485af54407
LLR: 2.97 (-2.94,2.94) {-1.50,0.50}
Total: 105360 W: 20175 L: 20136 D: 65049
Ptnml(0-2): 1803, 12230, 24572, 12275, 1800

Passed LTC:
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 15440 W: 1978 L: 1877 D: 11585
Ptnml(0-2): 92, 1405, 4667, 1422, 134

closes https://github.com/official-stockfish/Stockfish/pull/2696

Bench: 4668875

3 months agoQueen Mobility Tweak
FauziAkram [Sat, 23 May 2020 09:22:34 +0000 (12:22 +0300)]
Queen Mobility Tweak

It's ok to have low mobility values for the Queen in the middlegame, but it's absolutely not ok to have low mobility values for the Queen in the endgame.
Decrease penalty for bad mobility in MG and increase it in EG.

STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 17264 W: 3424 L: 3206 D: 10634
Ptnml(0-2): 279, 2004, 3893, 2132, 324
https://tests.stockfishchess.org/tests/view/5ec8f9c1526edcbe9091eba1

LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 175016 W: 22071 L: 21404 D: 131541
Ptnml(0-2): 1195, 15796, 52914, 16353, 1250
https://tests.stockfishchess.org/tests/view/5ec9057c404591b2793007df

closes https://github.com/official-stockfish/Stockfish/pull/2697

Bench: 4487054

3 months agosmall cleanups
Joost VandeVondele [Sat, 23 May 2020 11:26:13 +0000 (13:26 +0200)]
small cleanups

closes https://github.com/official-stockfish/Stockfish/pull/2653

No functional change

3 months agoKeep low ply history from previous move
xoto10 [Thu, 21 May 2020 20:17:21 +0000 (21:17 +0100)]
Keep low ply history from previous move

This patch keeps the low-ply history from the previous move, shifting the data down by 2 ply.

Tested with closedpos book:

STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 71584 W: 14175 L: 13891 D: 43518
Ptnml(0-2): 1069, 8228, 16993, 8354, 1148
https://tests.stockfishchess.org/tests/view/5ec0eaafe9d85f94dc429974

LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 96552 W: 13946 L: 13498 D: 69108
Ptnml(0-2): 676, 9082, 28375, 9404, 739
https://tests.stockfishchess.org/tests/view/5ec145efe9d85f94dc4299b0

closes https://github.com/official-stockfish/Stockfish/pull/2688

Bench 5148950

3 months agoImprove CI testing
Joost VandeVondele [Thu, 21 May 2020 06:25:37 +0000 (08:25 +0200)]
Improve CI testing

also enable CXXFLAGS="-D_GLIBCXX_DEBUG" in CI.

closes https://github.com/official-stockfish/Stockfish/pull/2692

No functional change.

3 months agoAdd doubled isolated pawn penalty.
Stefan Geschwentner [Fri, 22 May 2020 09:08:44 +0000 (11:08 +0200)]
Add doubled isolated pawn penalty.

This patch gives an additional penalty if a doubled isolated pawn is stopped
only by a single opponent pawn on the same file. Thanks to NKONSTANTAKIS,
who shared this idea on the forum!

https://groups.google.com/forum/?fromgroups=#!topic/fishcooking/vC4Qn-PMlS4.

STC:
LLR: 2.96 (-2.94,2.94) {-0.50,1.50}
Total: 84872 W: 16688 L: 16370 D: 51814
Ptnml(0-2): 1507, 9940, 19274, 10158, 1557
https://tests.stockfishchess.org/tests/view/5ec65bd955202b947dc5d4ac

LTC:
LLR: 2.93 (-2.94,2.94) {0.25,1.75}
Total: 58104 W: 7614 L: 7278 D: 43212
Ptnml(0-2): 411, 5369, 17196, 5625, 451
https://tests.stockfishchess.org/tests/view/5ec6e9f2c23f5b0710632b19

Closes https://github.com/official-stockfish/Stockfish/pull/2694

Bench: 5148950

3 months agoTweak knight mobility
Stefan Geschwentner [Thu, 21 May 2020 10:29:36 +0000 (12:29 +0200)]
Tweak knight mobility

New tuned values for knight mobility in endgames.

STC:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 112576 W: 21999 L: 21644 D: 68933
Ptnml(0-2): 2009, 13084, 25735, 13463, 1997
https://tests.stockfishchess.org/tests/view/5ec58379377121ac09e10272

LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.75}
Total: 125192 W: 16200 L: 15671 D: 93321
Ptnml(0-2): 891, 11584, 37182, 11983, 956
https://tests.stockfishchess.org/tests/view/5ec5c0b8377121ac09e1028b

Closes https://github.com/official-stockfish/Stockfish/pull/2693

Bench: 4778956

3 months agoA combo of parameters tweaks
Stéphane Nicolet [Wed, 20 May 2020 15:06:42 +0000 (17:06 +0200)]
A combo of parameters tweaks

This patch is a combinaison of two recent parameters tweaks which had
failed narrowly (yellow) at long time control:

• improvement in move ordering during search by softening the distinction
  between bad captures and good captures during move generation, leading
  to improved awareness of Stockfish of potential piece sacrifices (idea
  by Rahul Dsilva)
• increase in the weight of pawns in the "initiative" part of the evaluation
  function. With this change Stockfish should have more incentive to exchange
  pawns when losing, and to keep pawns when winning.

STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 10704 W: 2178 L: 1974 D: 6552
Ptnml(0-2): 168, 1185, 2464, 1345, 190
https://tests.stockfishchess.org/tests/view/5ec5553b377121ac09e1023d

LTC:
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 60592 W: 7835 L: 7494 D: 45263
Ptnml(0-2): 430, 5514, 18086, 5817, 449
https://tests.stockfishchess.org/tests/view/5ec55ca2377121ac09e10249

Closes https://github.com/official-stockfish/Stockfish/pull/2691

Bench: 4519117

3 months agoSimplify evaluation for blocked passers.
protonspring [Wed, 20 May 2020 14:33:59 +0000 (08:33 -0600)]
Simplify evaluation for blocked passers.

This is a functional simplification of the evaluation code for blocked passers.
I've also changed a few variable names for clarity.

STC
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 141984 W: 27450 L: 27466 D: 87068
Ptnml(0-2): 2414, 16511, 33175, 16461, 2431
https://tests.stockfishchess.org/tests/view/5ec4001b05aa4bc72d9759e7

LTC
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 30536 W: 3966 L: 3885 D: 22685
Ptnml(0-2): 216, 2841, 9073, 2922, 216
https://tests.stockfishchess.org/tests/view/5ec4bd0d377121ac09e101b7

Closes https://github.com/official-stockfish/Stockfish/pull/2690

Bench: 4704681

4 months agoAvoid sending info strings before 'uci' has been received
Sami Kiminki [Tue, 19 May 2020 09:08:01 +0000 (12:08 +0300)]
Avoid sending info strings before 'uci' has been received

Do not send the following info string on the first call to
aligned_ttmem_alloc() on Windows:

  info string Hash table allocation: Windows large pages [not] used.

The first call occurs before the 'uci' command has been received. This
confuses some GUIs, which expect the first engine-sent command to be
'id' as the response to the 'uci' command. (see https://github.com/official-stockfish/Stockfish/issues/2681)

closes https://github.com/official-stockfish/Stockfish/pull/2689

No functional change.

4 months agoIncrease base time use and limit max used.
xoto10 [Sun, 17 May 2020 19:46:25 +0000 (20:46 +0100)]
Increase base time use and limit max used.

This change increases the base part of optimumTime at all depths. It also reduces the size of max_scale and thus maximumTime by using a linear scale instead of pow(x, 0.3) and by limiting max_scale to no more than 7 (previously as high as 8 or 9 at very high depths).

Tested using the closedpos book:

STC 10+0.1:
LLR: 2.94 (-2.94,2.94) {-0.50,1.50}
Total: 83696 W: 16813 L: 16508 D: 50375
Ptnml(0-2): 1315, 9649, 19686, 9812, 1386
https://tests.stockfishchess.org/tests/view/5ebfa92de9d85f94dc42989b

LTC 60+0.6:
LLR: 2.95 (-2.94,2.94) {0.25,1.75}
Total: 39384 W: 5868 L: 5582 D: 27934
Ptnml(0-2): 276, 3697, 11489, 3925, 305
https://tests.stockfishchess.org/tests/view/5ec0a6dce9d85f94dc42995a

Test for non-regression:

STC Sudden Death 10+0 :
LLR: 2.94 (-2.94,2.94) {-2.00,0.00}
Total: 111976 W: 25661 L: 25768 D: 60547
Ptnml(0-2): 2567, 13420, 24118, 13319, 2564
https://tests.stockfishchess.org/tests/view/5ec23b3be9d85f94dc429a58

closes  https://github.com/official-stockfish/Stockfish/pull/2685

Bench 4395562

4 months agoDon't adjust MoveOverhead by increment
protonspring [Fri, 15 May 2020 23:23:49 +0000 (17:23 -0600)]
Don't adjust MoveOverhead by increment

This is a change to address a potential timing issue for slow networks.

Move Overhead was limited by TC increment,
which might be problematic if small increments (or sudden death)
on slow networks (needing high Move Overhead) are used.

STC, sudden death.
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 169368 W: 38023 L: 38054 D: 93291
Ptnml(0-2): 3767, 20250, 36595, 20391, 3681
https://tests.stockfishchess.org/tests/view/5ebf25efe9d85f94dc42986f

STC, 10+0.1
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 83896 W: 16092 L: 16026 D: 51778
Ptnml(0-2): 1401, 9697, 19670, 9795, 1385
https://tests.stockfishchess.org/tests/view/5ec0239de9d85f94dc42991e

closes https://github.com/official-stockfish/Stockfish/pull/2684

No functional change.

4 months agoWorkaround for older compiler
Joost VandeVondele [Fri, 15 May 2020 17:37:56 +0000 (19:37 +0200)]
Workaround for older compiler

gcc < 5 doesn't fully support the c++11 `std::is_trivially_copyable<Entry>::value`
Remove it, as it is not essential.

fixes https://github.com/official-stockfish/Stockfish/issues/2681

closes https://github.com/official-stockfish/Stockfish/pull/2682

No functional change.

4 months agoSimplify Time Management
protonspring [Wed, 13 May 2020 17:52:41 +0000 (11:52 -0600)]
Simplify Time Management

This is a functional simplification of the time management system.

With this patch, there is a simple equation for each of two distinct
time controls: basetime + increment, and x moves in y seconds (+increment).
These equations are easy to plot and understand making future modifications
or adding additional time controls much easier.

SlowMover is reset to 100 so that is has no effect unless a user changes it.

There are two scaling variables:
* Opt_scale is a scale factor (or percentage) of time to use for this current move.
* Max_scale is a scale factor to apply to the resulting optimumTime.

There seems to be some elo gain in most scenarios.
Better performance is attributable to one of two things:
* minThinkingTime was not allowing reasonable time calculations for very short games like 10+0 or 10+0.01. This is because adding almost no increment and substracting move overhead for 50 moves quickly results in almost 0 time very early in the game. Master depended on minThinkingTime to handle these short games instead of good time management. This patch addresses this issue by lowering minThinkingTime to 0 and adjusting moverOverhead if there are very low increments.
* Notice that the time distribution curves tail downward for the first 10 moves or so. This causes less time to attribute for very early moves leaving more time available for middle moves where more important decisions happen.

Here is a summary of tests for this version at different time controls:

SMP 5+0.05
LLR: 2.97 (-2.94,2.94) {-1.50,0.50}
Total: 46544 W: 7175 L: 7089 D: 32280
Ptnml(0-2): 508, 4826, 12517, 4914, 507
https://tests.stockfishchess.org/tests/user/protonspring

STC
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 20480 W: 3872 L: 3718 D: 12890
Ptnml(0-2): 295, 2364, 4824, 2406, 351
https://tests.stockfishchess.org/tests/view/5ebc343e7dd5693aad4e6873

STC, sudden death
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 7024 W: 1706 L: 1489 D: 3829
Ptnml(0-2): 149, 813, 1417, 938, 195
https://tests.stockfishchess.org/tests/view/5ebc346f7dd5693aad4e6875

STC, TCEC style
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 4192 W: 1014 L: 811 D: 2367
Ptnml(0-2): 66, 446, 912, 563, 109
https://tests.stockfishchess.org/tests/view/5ebc34857dd5693aad4e6877

40/10
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 54032 W: 10592 L: 10480 D: 32960
Ptnml(0-2): 967, 6148, 12677, 6254, 970
https://tests.stockfishchess.org/tests/view/5ebc50597dd5693aad4e688d

LTC, sudden death
LLR: 2.95 (-2.94,2.94) {-1.50,0.50}
Total: 9152 W: 1391 L: 1263 D: 6498
Ptnml(0-2): 75, 888, 2526, 1008, 79
https://tests.stockfishchess.org/tests/view/5ebc6f5c7dd5693aad4e689b

LTC
LLR: 2.98 (-2.94,2.94) {-1.50,0.50}
Total: 12344 W: 1563 L: 1459 D: 9322
Ptnml(0-2): 70, 1103, 3740, 1171, 88
https://tests.stockfishchess.org/tests/view/5ebc6f4c7dd5693aad4e6899

closes https://github.com/official-stockfish/Stockfish/pull/2678

Bench: 4395562

4 months agoFix a Windows-only crash on exit without 'quit'
Sami Kiminki [Thu, 14 May 2020 09:00:35 +0000 (12:00 +0300)]
Fix a Windows-only crash on exit without 'quit'

There was a bug in commit d4763424d2728fe2dfd0a6fe747666feb6a2fdbb
(Add support for Windows large pages) that could result in trying to
free memory allocated with VirtualAlloc incorrectly with free().

Fix this by reverting the TT.resize(0) logic in the previous commit,
and instead, just call aligned_ttmem_free() in
TranspositionTable::~TranspositionTable().

fixes https://github.com/official-stockfish/Stockfish/issues/2677

closes https://github.com/official-stockfish/Stockfish/pull/2679

No functional change

4 months agoMove 50 moves counter to initiative.
Stéphane Nicolet [Fri, 8 May 2020 08:32:52 +0000 (10:32 +0200)]
Move 50 moves counter to initiative.

simplify the usage of the 50 moves counter,
moving it frome the scale factor to initiative.

This patch was inspired by recent games where a blocked or semi-blocked position
was 'blundered', by moving a pawn, into a lost endgame. This patch improves this situation,
finding a more robust move more often.

for example (1s searches with many threads):
```
FEN 8/p3kp2/Pp2p3/1n2PpP1/5P2/1Kp5/8/R7 b - - 68 143

master:
      6 bestmove b5c7
      6 bestmove e7e8
     12 bestmove e7d8
    176 bestmove e7d7
patch:
      3 bestmove b5c7
      5 bestmove e7d8
    192 bestmove e7d7
```

fixes https://github.com/official-stockfish/Stockfish/issues/2620

the patch also tests well

passed STC
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 50168 W: 9508 L: 9392 D: 31268
Ptnml(0-2): 818, 5873, 11616, 5929, 848
https://tests.stockfishchess.org/tests/view/5ebb07287dd5693aad4e680b

passed LTC
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 7520 W: 981 L: 870 D: 5669
Ptnml(0-2): 49, 647, 2256, 760, 48
https://tests.stockfishchess.org/tests/view/5ebbff747dd5693aad4e6858

closes https://github.com/official-stockfish/Stockfish/pull/2666

Bench: 4395562

4 months agoAdd support for Windows large pages
Sami Kiminki [Mon, 4 May 2020 17:49:27 +0000 (20:49 +0300)]
Add support for Windows large pages

for users that set the needed privilige "Lock Pages in Memory"
large pages will be automatically enabled (see Readme.md).

This expert setting might improve speed, 5% - 30%, depending
on the hardware, the number of threads and hash size. More for
large hashes, large number of threads and NUMA. If the operating
system can not allocate large pages (easier after a reboot), default
allocation is used automatically. The engine log provides details.

closes https://github.com/official-stockfish/Stockfish/pull/2656

fixes https://github.com/official-stockfish/Stockfish/issues/2619

No functional change

4 months agoUse a trivially copyable struct for TBTables::Entry instead of a tuple.
Tomasz Sobczyk [Tue, 12 May 2020 19:41:55 +0000 (21:41 +0200)]
Use a trivially copyable struct for TBTables::Entry instead of a tuple.

fixes https://github.com/official-stockfish/Stockfish/issues/2673
which is a warning issued by recent gcc (10.1)

closes https://github.com/official-stockfish/Stockfish/pull/2674

No functional change

4 months agoTune pawn value
Moez Jellouli [Fri, 8 May 2020 14:59:06 +0000 (16:59 +0200)]
Tune pawn value

Small tune of PawnValue parameters -4 / -7 with "closedpos.epd" opening book.

STC:
LLR: 2.96 (-2.94,2.94) {-0.50,1.50}
Total: 58776 W: 11787 L: 11511 D: 35478
Ptnml(0-2): 975, 6876, 13443, 7086, 1008
https://tests.stockfishchess.org/tests/view/5eb5aa712326444a3b6d3e33

LTC:
LLR: 2.98 (-2.94,2.94) {0.25,1.75}
Total: 137544 W: 19687 L: 19115 D: 98742
Ptnml(0-2): 988, 13219, 39901, 13561, 1103
https://tests.stockfishchess.org/tests/view/5eb67a392326444a3b6d3e9a

Non regression STC with "noob_3moves.epd" opening book
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 98168 W: 18545 L: 18499 D: 61124
Ptnml(0-2): 1647, 11396, 22951, 11444, 1646
https://tests.stockfishchess.org/tests/view/5eb7e489e0300e8e8c896203

closes https://github.com/official-stockfish/Stockfish/pull/2670

Bench 4696646

4 months agoUse posix_memalign instead of aligned_alloc
Joost VandeVondele [Sat, 9 May 2020 17:45:07 +0000 (19:45 +0200)]
Use posix_memalign instead of aligned_alloc

should be a little more portable to older linux systems (before glibc-2.16).

fixes https://github.com/official-stockfish/Stockfish/issues/2665

closes https://github.com/official-stockfish/Stockfish/pull/2668

No functional change.

4 months agoFix syzygy dependencies issue
xoto10 [Fri, 8 May 2020 11:07:42 +0000 (12:07 +0100)]
Fix syzygy dependencies issue

fixes https://github.com/official-stockfish/Stockfish/issues/2660

The problem was caused by .depend being created with a rule for tbprobe.o not for syzygy/tbprobe.o.
This patch keeps an explicit list of sources (SRCS), generates OBJS,
and compiles all object files to the src/ directory, consistent with .depend.
VPATH is used to search the syzygy directory as needed.

joint work with @gvreuls

closes https://github.com/official-stockfish/Stockfish/pull/2664

No functional change

4 months agoPenalty for all enemy pawns xrayed by our bishop.
xoto10 [Wed, 6 May 2020 07:44:39 +0000 (08:44 +0100)]
Penalty for all enemy pawns xrayed by our bishop.

STC:
LLR: 2.93 (-2.94,2.94) {-0.50,1.50}
Total: 159760 W: 30229 L: 29813 D: 99718
Ptnml(0-2): 2659, 18309, 37534, 18713, 2665
https://tests.stockfishchess.org/tests/view/5eb1d5032326444a3b6d33ce

LTC:
LLR: 2.93 (-2.94,2.94) {0.25,1.75}
Total: 26496 W: 3908 L: 3656 D: 18932
Ptnml(0-2): 192, 2512, 7610, 2720, 214
https://tests.stockfishchess.org/tests/view/5eb1e2dd2326444a3b6d33f9

closes https://github.com/official-stockfish/Stockfish/pull/2662

Bench 5185517

4 months agoFishtest Tuning Framework
Marco Costalba [Sat, 4 Apr 2015 06:54:15 +0000 (08:54 +0200)]
Fishtest Tuning Framework

The purpose of the code is to allow developers to easily and flexibly
setup SF for a tuning session. Mainly you have just to remove 'const'
qualifiers from the variables you want to tune and flag them for
tuning, so if you have:

int myKing = 10;
Score myBonus = S(5, 15);
Value myValue[][2] = { { V(100), V(20) }, { V(7), V(78) } };

and at the end of the update you may want to call
a post update function:

void my_post_update();

If instead of default Option's min-max values,
you prefer your custom ones, returned by:

std::pair<int, int> my_range(int value)

Or you jus want to set the range directly, you can
simply add below:

TUNE(SetRange(my_range), myKing, SetRange(-200, 200), myBonus, myValue, my_post_update);

And all the magic happens :-)

At startup all the parameters are printed in a
format suitable to be copy-pasted in fishtest.

In case the post update function is slow and you have many
parameters to tune, you can add:

UPDATE_ON_LAST();

And the values update, including post update function call, will
be done only once, after the engine receives the last UCI option.
The last option is the one defined and created as the last one, so
this assumes that the GUI sends the options in the same order in
which have been defined.

closes https://github.com/official-stockfish/Stockfish/pull/2654

No functional change.

4 months agoRefine scale factor of opposite colored bishops endgames.
Vizvezdenec [Sat, 2 May 2020 13:45:20 +0000 (16:45 +0300)]
Refine scale factor of opposite colored bishops endgames.

This patch makes it dependant on the count of passed pawns of
the strong side instead of 22/64 in every case.

passed STC
https://tests.stockfishchess.org/tests/view/5ead60966ffeed51f6e32591
LLR: 2.96 (-2.94,2.94) {-0.50,1.50}
Total: 50336 W: 9473 L: 9241 D: 31622
Ptnml(0-2): 570, 5371, 13098, 5515, 614

passed LTC
https://tests.stockfishchess.org/tests/view/5ead6d3b6ffeed51f6e325b0
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 21952 W: 2810 L: 2603 D: 16539
Ptnml(0-2): 101, 1791, 7005, 1958, 121

closes https://github.com/official-stockfish/Stockfish/pull/2658

bench 4247490

4 months agoTuned safe checks and minor piece king protectors
Linmiao Xu [Sat, 25 Apr 2020 19:55:35 +0000 (15:55 -0400)]
Tuned safe checks and minor piece king protectors

A combination of terms related to king safety one tuned safe check weights,
the other tuned knight and bishop king protector weights separately with
some compensation in the high outpost bonuses given to the minor pieces.

passed STC
LLR: 2.95 (-2.94,2.94) {-0.50,1.50}
Total: 39892 W: 7594 L: 7350 D: 24948
Ptnml(0-2): 643, 4559, 9314, 4771, 659
https://tests.stockfishchess.org/tests/view/5ea49635b908f6dd28f34b82

passed LTC
LLR: 2.94 (-2.94,2.94) {0.25,1.75}
Total: 104934 W: 13300 L: 12834 D: 78800
Ptnml(0-2): 697, 9571, 31514, 9939, 746
https://tests.stockfishchess.org/tests/view/5ea4abf6b908f6dd28f34bcb

closes https://github.com/official-stockfish/Stockfish/pull/2649

Bench 4800754