Stockfish 12

Official release version of Stockfish 12

Bench: 3624569

-----------------------

It is our pleasure to release Stockfish 12 to users world-wide

Downloads will be freely available at

https://stockfishchess.org/download/

This version 12 of Stockfish plays significantly stronger than
any of its predecessors. In a match against Stockfish 11,
Stockfish 12 will typically win at least ten times more game pairs
than it loses.

This jump in strength, visible in regular progression tests during
development[1], results from the introduction of an efficiently
updatable neural network (NNUE) for the evaluation in Stockfish[2],
and associated tuning of the engine as a whole. The concept of the
NNUE evaluation was first introduced in shogi, and ported to
Stockfish afterward. Stockfish remains a CPU-only engine, since the
NNUE networks can be very efficiently evaluated on CPUs. The
recommended parameters of the NNUE network are embedded in
distributed binaries, and Stockfish will use NNUE by default.

Both the NNUE and the classical evaluations are available, and
can be used to assign values to positions that are 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
evaluations of millions of positions.

The Stockfish project builds on a thriving community of enthusiasts
that contribute their expertise, time, and resources to build a free
and open source chess engine that is robust, widely available, and
very strong. We invite chess fans to join the fishtest testing
framework and programmers to contribute on github[3].

Stay safe and enjoy chess!

The Stockfish team

[1] https://github.com/glinscott/fishtest/wiki/Regression-Tests
[2] https://github.com/official-stockfish/Stockfish/commit/84f3e867903f62480c33243dd0ecbffd342796fc
[3] https://stockfishchess.org/get-involved/

.travis.yml

@@ -43,26 +43,47 @@ before_script:
   - cd src
 
 script:
+  # Download net
+  - make net
+
   # Obtain bench reference from git log
   - git log HEAD | grep "\b[Bb]ench[ :]\+[0-9]\{7\}" | head -n 1 | sed "s/[^0-9]*\([0-9]*\).*/\1/g" > git_sig
   - export benchref=$(cat git_sig)
   - echo "Reference bench:" $benchref
 
-  #
   # Compiler version string
   - $COMPILER -v
 
-  #
+  # test help target
+  - make help
+
   # Verify bench number against various builds
   - export CXXFLAGS="-Werror -D_GLIBCXX_DEBUG"
-  - make clean && make -j2 ARCH=x86-64 optimize=no debug=yes build && ../tests/signature.sh $benchref
+  - make clean && make -j2 ARCH=x86-64-modern optimize=no debug=yes build && ../tests/signature.sh $benchref
+  - export CXXFLAGS="-Werror"
+  - make clean && make -j2 ARCH=x86-64-modern build && ../tests/signature.sh $benchref
+  - make clean && make -j2 ARCH=x86-64-ssse3 build && ../tests/signature.sh $benchref
+  - make clean && make -j2 ARCH=x86-64-sse3-popcnt build && ../tests/signature.sh $benchref
+  - make clean && make -j2 ARCH=x86-64 build && ../tests/signature.sh $benchref
+  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=general-64 build && ../tests/signature.sh $benchref; fi
   - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=x86-32 optimize=no debug=yes build && ../tests/signature.sh $benchref; fi
+  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=x86-32-sse41-popcnt build && ../tests/signature.sh $benchref; fi
+  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=x86-32-sse2 build && ../tests/signature.sh $benchref; fi
   - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=x86-32 build && ../tests/signature.sh $benchref; fi
+  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=general-32 build && ../tests/signature.sh $benchref; fi
+  # workaround: exclude a custom version of llvm+clang, which doesn't find llvm-profdata on ubuntu
+  - if [[ "$TRAVIS_OS_NAME" != "linux" || "$COMP" == "gcc" ]]; then make clean && make -j2 ARCH=x86-64-modern profile-build && ../tests/signature.sh $benchref; fi
+
+  # compile only for some more advanced architectures (might not run in travis)
+  - make clean && make -j2 ARCH=x86-64-avx2 build
+  - make clean && make -j2 ARCH=x86-64-bmi2 build
+  - make clean && make -j2 ARCH=x86-64-avx512 build
+  - make clean && make -j2 ARCH=x86-64-vnni512 build
+  - make clean && make -j2 ARCH=x86-64-vnni256 build
 
   #
   # Check perft and reproducible search
-  - export CXXFLAGS="-Werror"
-  - make clean && make -j2 ARCH=x86-64 build
+  - make clean && make -j2 ARCH=x86-64-modern build
   - ../tests/perft.sh
   - ../tests/reprosearch.sh
 
@@ -70,11 +91,11 @@ script:
   # Valgrind
   #
   - export CXXFLAGS="-O1 -fno-inline"
-  - if [ -x "$(command -v valgrind )" ]; then make clean && make -j2 ARCH=x86-64 debug=yes optimize=no build > /dev/null && ../tests/instrumented.sh --valgrind; fi
+  - if [ -x "$(command -v valgrind )" ]; then make clean && make -j2 ARCH=x86-64-modern debug=yes optimize=no build > /dev/null && ../tests/instrumented.sh --valgrind; fi
   - if [ -x "$(command -v valgrind )" ]; then ../tests/instrumented.sh --valgrind-thread; fi
 
   #
   # Sanitizer
   #
-  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=x86-64 sanitize=undefined optimize=no debug=yes build > /dev/null && ../tests/instrumented.sh --sanitizer-undefined; fi
-  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=x86-64 sanitize=thread    optimize=no debug=yes build > /dev/null && ../tests/instrumented.sh --sanitizer-thread; fi
+  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=x86-64-modern sanitize=undefined optimize=no debug=yes build > /dev/null && ../tests/instrumented.sh --sanitizer-undefined; fi
+  - if [[ "$TRAVIS_OS_NAME" == "linux" ]]; then make clean && make -j2 ARCH=x86-64-modern sanitize=thread    optimize=no debug=yes build > /dev/null && ../tests/instrumented.sh --sanitizer-thread; fi

AUTHORS

@@ -36,10 +36,11 @@ Bryan Cross (crossbr)
 candirufish
 Chess13234
 Chris Cain (ceebo)
+Dale Weiler (graphitemaster)
 Dan Schmidt (dfannius)
 Daniel Axtens (daxtens)
 Daniel Dugovic (ddugovic)
-Dariusz Orzechowski
+Dariusz Orzechowski (dorzechowski)
 David Zar
 Daylen Yang (daylen)
 DiscanX
@@ -53,11 +54,13 @@ Ernesto Gatti
 Linmiao Xu (linrock)
 Fabian Beuke (madnight)
 Fabian Fichter (ianfab)
+Fanael Linithien (Fanael)
 fanon
 Fauzi Akram Dabat (FauziAkram)
 Felix Wittmann
 gamander
 Gary Heckman (gheckman)
+George Sobala (gsobala)
 gguliash
 Gian-Carlo Pascutto (gcp)
 Gontran Lemaire (gonlem)
@@ -79,6 +82,7 @@ Jean Gauthier (OuaisBla)
 Jean-Francois Romang (jromang)
 Jekaa
 Jerry Donald Watson (jerrydonaldwatson)
+jjoshua2
 Jonathan Calovski (Mysseno)
 Jonathan Dumale (SFisGOD)
 Joost VandeVondele (vondele)
@@ -125,6 +129,7 @@ Niklas Fiekas (niklasf)
 Nikolay Kostov (NikolayIT)
 Nguyen Pham (nguyenpham)
 Norman Schmidt (FireFather)
+notruck
 Ondrej Mosnáček (WOnder93)
 Oskar Werkelin Ahlin
 Pablo Vazquez
@@ -151,6 +156,7 @@ Sami Kiminki (skiminki)
 Sebastian Buchwald (UniQP)
 Sergei Antonov (saproj)
 Sergei Ivanov (svivanov72)
+Sergio Vieri (sergiovieri)
 sf-x
 Shane Booth (shane31)
 Shawn Varghese (xXH4CKST3RXx)

README.md

@@ -4,17 +4,17 @@
 [![Build Status](https://ci.appveyor.com/api/projects/status/github/official-stockfish/Stockfish?branch=master&svg=true)](https://ci.appveyor.com/project/mcostalba/stockfish/branch/master)
 
 [Stockfish](https://stockfishchess.org) is a free, powerful UCI chess engine
-derived from Glaurung 2.1. It features two evaluation functions, the classical
-evaluation based on handcrafted terms, and the NNUE evaluation based on
-efficiently updateable neural networks. The classical evaluation runs efficiently
-on most 64bit CPU architectures, while the NNUE evaluation benefits strongly from the
-vector intrinsics available on modern CPUs (avx2 or similar).
+derived from Glaurung 2.1. Stockfish is not a complete chess program and requires a
+UCI-compatible graphical user interface (GUI) (e.g. XBoard with PolyGlot, Scid,
+Cute Chess, eboard, Arena, Sigma Chess, Shredder, Chess Partner or Fritz) in order
+to be used comfortably. Read the documentation for your GUI of choice for information
+about how to use Stockfish with it.
 
-Stockfish is not a complete chess program and requires a
-UCI-compatible GUI (e.g. XBoard with PolyGlot, Scid, Cute Chess, eboard, Arena,
-Sigma Chess, Shredder, Chess Partner or Fritz) in order to be used comfortably.
-Read the documentation for your GUI of choice for information about how to use
-Stockfish with it.
+The Stockfish engine features two evaluation functions for chess, the classical
+evaluation based on handcrafted terms, and the NNUE evaluation based on efficiently
+updateable neural networks. The classical evaluation runs efficiently on almost all
+CPU architectures, while the NNUE evaluation benefits from the vector
+intrinsics available on most CPUs (sse2, avx2, neon, or similar).
 
 
 ## Files
@@ -28,10 +28,14 @@ This distribution of Stockfish consists of the following files:
   * src, a subdirectory containing the full source code, including a Makefile
     that can be used to compile Stockfish on Unix-like systems.
 
-To use the NNUE evaluation an additional data file with neural network parameters
-needs to be downloaded. The filename for the default set can be found as the default
-value of the `EvalFile` UCI option, with the format
-`nn-[SHA256 first 12 digits].nnue` (e.g. nn-c157e0a5755b.nnue). This file can be downloaded from
+  * a file with the .nnue extension, storing the neural network for the NNUE 
+    evaluation. Binary distributions will have this file embedded.
+
+Note: to use the NNUE evaluation, the additional data file with neural network parameters
+needs to be available. Normally, this file is already embedded in the binary or it can be downloaded.
+The filename for the default (recommended) net can be found as the default
+value of the `EvalFile` UCI option, with the format `nn-[SHA256 first 12 digits].nnue`
+(for instance, `nn-c157e0a5755b.nnue`). This file can be downloaded from
 ```
 https://tests.stockfishchess.org/api/nn/[filename]
 ```
@@ -58,19 +62,14 @@ Currently, Stockfish has the following UCI options:
 
   * #### Use NNUE
     Toggle between the NNUE and classical evaluation functions. If set to "true",
-    the network parameters must be availabe to load from file (see also EvalFile).
+    the network parameters must be available to load from file (see also EvalFile),
+    if they are not embedded in the binary.
 
   * #### EvalFile
     The name of the file of the NNUE evaluation parameters. Depending on the GUI the
-    filename should include the full path to the folder/directory that contains the file.
-
-  * #### Contempt
-    A positive value for contempt favors middle game positions and avoids draws,
-    effective for the classical evaluation only.
-
-  * #### Analysis Contempt
-    By default, contempt is set to prefer the side to move. Set this option to "White"
-    or "Black" to analyse with contempt for that side, or "Off" to disable contempt.
+    filename might have to include the full path to the folder/directory that contains the file.
+    Other locations, such as the directory that contains the binary and the working directory,
+    are also searched.
 
   * #### UCI_AnalyseMode
     An option handled by your GUI.
@@ -120,6 +119,14 @@ Currently, Stockfish has the following UCI options:
     Limit Syzygy tablebase probing to positions with at most this many pieces left
     (including kings and pawns).
 
+  * #### Contempt
+    A positive value for contempt favors middle game positions and avoids draws,
+    effective for the classical evaluation only.
+
+  * #### Analysis Contempt
+    By default, contempt is set to prefer the side to move. Set this option to "White"
+    or "Black" to analyse with contempt for that side, or "Off" to disable contempt.
+
   * #### Move Overhead
     Assume a time delay of x ms due to network and GUI overheads. This is useful to
     avoid losses on time in those cases.
@@ -138,7 +145,7 @@ Currently, Stockfish has the following UCI options:
   * #### Debug Log File
     Write all communication to and from the engine into a text file.
 
-## classical and NNUE evaluation
+## A note on classical and NNUE evaluation
 
 Both approaches assign a value to a position that is used in alpha-beta (PVS) search
 to find the best move. The classical evaluation computes this value as a function
@@ -226,6 +233,7 @@ targets with corresponding descriptions.
     cd src
     make help
     make build ARCH=x86-64-modern
+    make net
 ```
 
 When not using the Makefile to compile (for instance with Microsoft MSVC) you
@@ -237,8 +245,7 @@ compiler you used to create your executable. These informations can
 be found by typing the following commands in a console:
 
 ```
-    ./stockfish
-    compiler
+    ./stockfish compiler
 ```
 
 ## Understanding the code base and participating in the project

Top CPU Contributors.txt

@@ -1,154 +1,173 @@
-Contributors with >10,000 CPU hours as of January 7, 2020
+Contributors with >10,000 CPU hours as of Sept 2, 2020
 Thank you!
 
 Username                  CPU Hours   Games played
 --------------------------------------------------
-noobpwnftw                  9305707      695548021
-mlang                        780050       61648867
-dew                          621626       43921547
-mibere                       524702       42238645
-crunchy                      354587       27344275
-cw                           354495       27274181
-fastgm                       332801       22804359
-JojoM                        295750       20437451
-CSU_Dynasty                  262015       21828122
-Fisherman                    232181       18939229
-ctoks                        218866       17622052
-glinscott                    201989       13780820
-tvijlbrief                   201204       15337115
-velislav                     188630       14348485
-gvreuls                      187164       15149976
-bking_US                     180289       11876016
-nordlandia                   172076       13467830
-leszek                       157152       11443978
-Thanar                       148021       12365359
-spams                        141975       10319326
-drabel                       138073       11121749
-vdv                          137850        9394330
-mgrabiak                     133578       10454324
-TueRens                      132485       10878471
-bcross                       129683       11557084
-marrco                       126078        9356740
-sqrt2                        125830        9724586
-robal                        122873        9593418
-vdbergh                      120766        8926915
-malala                       115926        8002293
-CoffeeOne                    114241        5004100
-dsmith                       113189        7570238
-BrunoBanani                  104644        7436849
-Data                          92328        8220352
-mhoram                        89333        6695109
-davar                         87924        7009424
-xoto                          81094        6869316
-ElbertoOne                    80899        7023771
-grandphish2                   78067        6160199
-brabos                        77212        6186135
-psk                           75733        5984901
-BRAVONE                       73875        5054681
-sunu                          70771        5597972
-sterni1971                    70605        5590573
-MaZePallas                    66886        5188978
-Vizvezdenec                   63708        4967313
-nssy                          63462        5259388
-jromang                       61634        4940891
-teddybaer                     61231        5407666
-Pking_cda                     60099        5293873
-solarlight                    57469        5028306
-dv8silencer                   56913        3883992
-tinker                        54936        4086118
-renouve                       49732        3501516
-Freja                         49543        3733019
-robnjr                        46972        4053117
-rap                           46563        3219146
-Bobo1239                      46036        3817196
-ttruscott                     45304        3649765
-racerschmacer                 44881        3975413
-finfish                       44764        3370515
-eva42                         41783        3599691
-biffhero                      40263        3111352
-bigpen0r                      39817        3291647
-mhunt                         38871        2691355
-ronaldjerum                   38820        3240695
-Antihistamine                 38785        2761312
-pb00067                       38038        3086320
-speedycpu                     37591        3003273
-rkl                           37207        3289580
-VoyagerOne                    37050        3441673
-jbwiebe                       35320        2805433
-cuistot                       34191        2146279
-homyur                        33927        2850481
-manap                         32873        2327384
-gri                           32538        2515779
-oryx                          31267        2899051
-EthanOConnor                  30959        2090311
-SC                            30832        2730764
-csnodgrass                    29505        2688994
-jmdana                        29458        2205261
-strelock                      28219        2067805
-jkiiski                       27832        1904470
-Pyafue                        27533        1902349
-Garf                          27515        2747562
-eastorwest                    27421        2317535
-slakovv                       26903        2021889
-Prcuvu                        24835        2170122
-anst                          24714        2190091
-hyperbolic.tom                24319        2017394
-Patrick_G                     23687        1801617
-Sharaf_DG                     22896        1786697
-nabildanial                   22195        1519409
-chriswk                       21931        1868317
-achambord                     21665        1767323
-Zirie                         20887        1472937
-team-oh                       20217        1636708
-Isidor                        20096        1680691
-ncfish1                       19931        1520927
-nesoneg                       19875        1463031
-Spprtr                        19853        1548165
-JanErik                       19849        1703875
-agg177                        19478        1395014
-SFTUser                       19231        1567999
-xor12                         19017        1680165
-sg4032                        18431        1641865
-rstoesser                     18118        1293588
-MazeOfGalious                 17917        1629593
-j3corre                       17743         941444
-cisco2015                     17725        1690126
-ianh2105                      17706        1632562
-dex                           17678        1467203
-jundery                       17194        1115855
-iisiraider                    17019        1101015
-horst.prack                   17012        1465656
-Adrian.Schmidt123             16563        1281436
-purplefishies                 16342        1092533
-wei                           16274        1745989
-ville                         16144        1384026
-eudhan                        15712        1283717
-OuaisBla                      15581         972000
-DragonLord                    15559        1162790
-dju                           14716         875569
-chris                         14479        1487385
-0xB00B1ES                     14079        1001120
-OssumOpossum                  13776        1007129
-enedene                       13460         905279
-bpfliegel                     13346         884523
-Ente                          13198        1156722
-IgorLeMasson                  13087        1147232
-jpulman                       13000         870599
-ako027ako                     12775        1173203
-Nikolay.IT                    12352        1068349
-Andrew Grant                  12327         895539
-joster                        12008         950160
-AdrianSA                      11996         804972
-Nesa92                        11455        1111993
-fatmurphy                     11345         853210
-Dark_wizzie                   11108        1007152
-modolief                      10869         896470
-mschmidt                      10757         803401
-infinity                      10594         727027
-mabichito                     10524         749391
-Thomas A. Anderson            10474         732094
-thijsk                        10431         719357
-Flopzee                       10339         894821
-crocogoat                     10104        1013854
-SapphireBrand                 10104         969604
-stocky                        10017         699440
+noobpwnftw                 19352969     1231459677
+mlang                        957168       61657446
+dew                          949885       56893432
+mibere                       703817       46865007
+crunchy                      427035       27344275
+cw                           416006       27521077
+JojoM                        415904       24479564
+fastgm                       404873       23953472
+CSU_Dynasty                  335774       22850550
+tvijlbrief                   335199       21871270
+Fisherman                    325053       21786603
+gvreuls                      311480       20751516
+ctoks                        275877       18710423
+velislav                     241267       15596372
+glinscott                    217799       13780820
+nordlandia                   211692       13484886
+bcross                       206213       14934233
+bking_US                     198894       11876016
+leszek                       189170       11446821
+mgrabiak                     183896       11778092
+drabel                       181408       12489478
+TueRens                      181349       12192000
+Thanar                       179852       12365359
+vdv                          175171        9881246
+robal                        166948       10702862
+spams                        157128       10319326
+marrco                       149947        9376421
+sqrt2                        147963        9724586
+vdbergh                      137041        8926915
+CoffeeOne                    136294        5004100
+malala                       136182        8002293
+mhoram                       128934        8177193
+davar                        122092        7960001
+dsmith                       122059        7570238
+xoto                         119696        8222144
+grandphish2                  116481        7582197
+Data                         113305        8220352
+BrunoBanani                  112960        7436849
+ElbertoOne                    99028        7023771
+MaZePallas                    98571        6362619
+brabos                        92118        6186135
+psk                           89957        5984901
+sunu                          88463        6007033
+sterni1971                    86948        5613788
+Vizvezdenec                   83752        5343724
+BRAVONE                       81239        5054681
+nssy                          76497        5259388
+teddybaer                     75125        5407666
+Pking_cda                     73776        5293873
+jromang                       70695        4940891
+solarlight                    70517        5028306
+dv8silencer                   70287        3883992
+Bobo1239                      68515        4652287
+racerschmacer                 67468        4935996
+manap                         66273        4121774
+tinker                        63458        4213726
+linrock                       59082        4516053
+robnjr                        57262        4053117
+Freja                         56938        3733019
+ttruscott                     56005        3679485
+renouve                       53811        3501516
+cuistot                       52532        3014920
+finfish                       51360        3370515
+eva42                         51272        3599691
+rkl                           50759        3840947
+rap                           49985        3219146
+pb00067                       49727        3298270
+ronaldjerum                   47654        3240695
+bigpen0r                      47278        3291647
+biffhero                      46564        3111352
+VoyagerOne                    45386        3445881
+speedycpu                     43842        3003273
+jbwiebe                       43305        2805433
+Antihistamine                 41788        2761312
+mhunt                         41735        2691355
+eastorwest                    40387        2812173
+homyur                        39893        2850481
+gri                           39871        2515779
+oryx                          38228        2941656
+0x3C33                        37773        2529097
+SC                            37290        2731014
+csnodgrass                    36207        2688994
+jmdana                        36108        2205261
+strelock                      34716        2074055
+Garf                          33800        2747562
+EthanOConnor                  33370        2090311
+slakovv                       32915        2021889
+Spprtr                        32591        2139601
+Prcuvu                        30377        2170122
+anst                          30301        2190091
+jkiiski                       30136        1904470
+hyperbolic.tom                29840        2017394
+Pyafue                        29650        1902349
+OuaisBla                      27629        1578000
+chriswk                       26902        1868317
+achambord                     26582        1767323
+Patrick_G                     26276        1801617
+yorkman                       26193        1992080
+SFTUser                       25182        1675689
+nabildanial                   24942        1519409
+Sharaf_DG                     24765        1786697
+ncfish1                       24411        1520927
+agg177                        23890        1395014
+JanErik                       23408        1703875
+Isidor                        23388        1680691
+Norabor                       22976        1587862
+cisco2015                     22880        1759669
+Zirie                         22542        1472937
+team-oh                       22272        1636708
+MazeOfGalious                 21978        1629593
+sg4032                        21945        1643065
+ianh2105                      21725        1632562
+xor12                         21628        1680365
+dex                           21612        1467203
+nesoneg                       21494        1463031
+horst.prack                   20878        1465656
+0xB00B1ES                     20590        1208666
+j3corre                       20405         941444
+Adrian.Schmidt123             20316        1281436
+wei                           19973        1745989
+rstoesser                     19569        1293588
+eudhan                        19274        1283717
+Ente                          19070        1373058
+jundery                       18445        1115855
+iisiraider                    18247        1101015
+ville                         17883        1384026
+chris                         17698        1487385
+purplefishies                 17595        1092533
+DragonLord                    17014        1162790
+dju                           16515         929427
+IgorLeMasson                  16064        1147232
+ako027ako                     15671        1173203
+Nikolay.IT                    15154        1068349
+Andrew Grant                  15114         895539
+yurikvelo                     15027        1165616
+OssumOpossum                  14857        1007129
+enedene                       14476         905279
+bpfliegel                     14298         884523
+jpulman                       13982         870599
+joster                        13794         950160
+Nesa92                        13786        1114691
+Dark_wizzie                   13422        1007152
+Hjax                          13350         900887
+Fifis                         13313         965473
+mabichito                     12903         749391
+thijsk                        12886         722107
+crocogoat                     12876        1048802
+AdrianSA                      12860         804972
+Flopzee                       12698         894821
+fatmurphy                     12547         853210
+SapphireBrand                 12416         969604
+modolief                      12386         896470
+scuzzi                        12362         833465
+pgontarz                      12151         848794
+stocky                        11954         699440
+mschmidt                      11941         803401
+infinity                      11470         727027
+torbjo                        11387         728873
+Thomas A. Anderson            11372         732094
+snicolet                      11106         869170
+amicic                        10779         733593
+rpngn                         10712         688203
+d64                           10680         771144
+basepi                        10637         744851
+jjoshua2                      10559         670905
+dzjp                          10343         732529
+ols                           10259         570669
+lbraesch                      10252         647825

appveyor.yml

@@ -61,6 +61,20 @@ before_build:
 
 build_script:
   - cmake --build . --config %CONFIGURATION% -- /verbosity:minimal
+  - ps: |
+      # Download default NNUE net from fishtest
+      $nnuenet = Get-Content -Path src\evaluate.h | Select-String -CaseSensitive -Pattern "EvalFileDefaultName" | Select-String -CaseSensitive -Pattern "nn-[a-z0-9]{12}.nnue"
+      $dummy = $nnuenet -match "(?<nnuenet>nn-[a-z0-9]{12}.nnue)"
+      $nnuenet = $Matches.nnuenet
+      Write-Host "Default net:" $nnuenet
+      $nnuedownloadurl = "https://tests.stockfishchess.org/api/nn/$nnuenet"
+      $nnuefilepath = "src\${env:CONFIGURATION}\$nnuenet"
+      if (Test-Path -Path $nnuefilepath) {
+            Write-Host "Already available."
+      } else {
+            Write-Host "Downloading $nnuedownloadurl to $nnuefilepath"
+            Invoke-WebRequest -Uri $nnuedownloadurl -OutFile $nnuefilepath
+      }
 
 before_test:
   - cd src/%CONFIGURATION%

src/Makefile

@@ -67,14 +67,16 @@ endif
 # bits = 64/32        --- -DIS_64BIT       --- 64-/32-bit operating system
 # prefetch = yes/no   --- -DUSE_PREFETCH   --- Use prefetch asm-instruction
 # popcnt = yes/no     --- -DUSE_POPCNT     --- Use popcnt asm-instruction
+# pext = yes/no       --- -DUSE_PEXT       --- Use pext x86_64 asm-instruction
 # sse = yes/no        --- -msse            --- Use Intel Streaming SIMD Extensions
-# sse3 = yes/no       --- -msse3           --- Use Intel Streaming SIMD Extensions 3
+# mmx = yes/no        --- -mmmx            --- Use Intel MMX instructions
+# sse2 = yes/no       --- -msse2           --- Use Intel Streaming SIMD Extensions 2
 # ssse3 = yes/no      --- -mssse3          --- Use Intel Supplemental Streaming SIMD Extensions 3
 # sse41 = yes/no      --- -msse4.1         --- Use Intel Streaming SIMD Extensions 4.1
-# sse42 = yes/no      --- -msse4.2         --- Use Intel Streaming SIMD Extensions 4.2
 # avx2 = yes/no       --- -mavx2           --- Use Intel Advanced Vector Extensions 2
-# pext = yes/no       --- -DUSE_PEXT       --- Use pext x86_64 asm-instruction
 # avx512 = yes/no     --- -mavx512bw       --- Use Intel Advanced Vector Extensions 512
+# vnni256 = yes/no    --- -mavx512vnni     --- Use Intel Vector Neural Network Instructions 256
+# vnni512 = yes/no    --- -mavx512vnni     --- Use Intel Vector Neural Network Instructions 512
 # neon = yes/no       --- -DUSE_NEON       --- Use ARM SIMD architecture
 #
 # Note that Makefile is space sensitive, so when adding new architectures
@@ -82,152 +84,188 @@ endif
 # at the end of the line for flag values.
 
 ### 2.1. General and architecture defaults
+
+ifeq ($(ARCH),)
+   ARCH = x86-64-modern
+   help_skip_sanity = yes
+endif
+# explicitly check for the list of supported architectures (as listed with make help),
+# the user can override with `make ARCH=x86-32-vnni256 SUPPORTED_ARCH=true`
+ifeq ($(ARCH), $(filter $(ARCH), \
+                 x86-64-vnni512 x86-64-vnni256 x86-64-avx512 x86-64-bmi2 x86-64-avx2 \
+                 x86-64-sse41-popcnt x86-64-modern x86-64-ssse3 x86-64-sse3-popcnt \
+                 x86-64 x86-32-sse41-popcnt x86-32-sse2 x86-32 ppc-64 ppc-32 \
+                 armv7 armv7-neon armv8 apple-silicon general-64 general-32))
+   SUPPORTED_ARCH=true
+else
+   SUPPORTED_ARCH=false
+endif
+
 optimize = yes
 debug = no
 sanitize = no
 bits = 64
 prefetch = no
 popcnt = no
+pext = no
 sse = no
-sse3 = no
+mmx = no
+sse2 = no
 ssse3 = no
 sse41 = no
-sse42 = no
 avx2 = no
-pext = no
 avx512 = no
+vnni256 = no
+vnni512 = no
 neon = no
-ARCH = x86-64-modern
+STRIP = strip
 
 ### 2.2 Architecture specific
+
+ifeq ($(findstring x86,$(ARCH)),x86)
+
+# x86-32/64
+
+ifeq ($(findstring x86-32,$(ARCH)),x86-32)
+	arch = i386
+	bits = 32
+	sse = yes
+	mmx = yes
+else
+	arch = x86_64
+	sse = yes
+	sse2 = yes
+endif
+
+ifeq ($(findstring -sse,$(ARCH)),-sse)
+	sse = yes
+endif
+
+ifeq ($(findstring -popcnt,$(ARCH)),-popcnt)
+	popcnt = yes
+endif
+
+ifeq ($(findstring -mmx,$(ARCH)),-mmx)
+	mmx = yes
+endif
+
+ifeq ($(findstring -sse2,$(ARCH)),-sse2)
+	sse = yes
+	sse2 = yes
+endif
+
+ifeq ($(findstring -ssse3,$(ARCH)),-ssse3)
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+endif
+
+ifeq ($(findstring -sse41,$(ARCH)),-sse41)
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+endif
+
+ifeq ($(findstring -modern,$(ARCH)),-modern)
+	popcnt = yes
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+endif
+
+ifeq ($(findstring -avx2,$(ARCH)),-avx2)
+	popcnt = yes
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+	avx2 = yes
+endif
+
+ifeq ($(findstring -bmi2,$(ARCH)),-bmi2)
+	popcnt = yes
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+	avx2 = yes
+	pext = yes
+endif
+
+ifeq ($(findstring -avx512,$(ARCH)),-avx512)
+	popcnt = yes
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+	avx2 = yes
+	pext = yes
+	avx512 = yes
+endif
+
+ifeq ($(findstring -vnni256,$(ARCH)),-vnni256)
+	popcnt = yes
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+	avx2 = yes
+	pext = yes
+	vnni256 = yes
+endif
+
+ifeq ($(findstring -vnni512,$(ARCH)),-vnni512)
+	popcnt = yes
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+	avx2 = yes
+	pext = yes
+	avx512 = yes
+	vnni512 = yes
+endif
+
+ifeq ($(sse),yes)
+	prefetch = yes
+endif
+
+# 64-bit pext is not available on x86-32
+ifeq ($(bits),32)
+	pext = no
+endif
+
+else
+
+# all other architectures
+
 ifeq ($(ARCH),general-32)
 	arch = any
 	bits = 32
 endif
 
-ifeq ($(ARCH),x86-32-old)
-	arch = i386
-	bits = 32
-endif
-
-ifeq ($(ARCH),x86-32)
-	arch = i386
-	bits = 32
-	prefetch = yes
-	sse = yes
-endif
-
 ifeq ($(ARCH),general-64)
 	arch = any
 endif
 
-ifeq ($(ARCH),x86-64)
-	arch = x86_64
-	prefetch = yes
-	sse = yes
-endif
-
-ifeq ($(ARCH),x86-64-sse3)
-	arch = x86_64
-	prefetch = yes
-	sse = yes
-	sse3 = yes
-endif
-
-ifeq ($(ARCH),x86-64-sse3-popcnt)
-	arch = x86_64
-	prefetch = yes
-	sse = yes
-	sse3 = yes
-	popcnt = yes
-endif
-
-ifeq ($(ARCH),x86-64-ssse3)
-	arch = x86_64
-	prefetch = yes
-	sse = yes
-	sse3 = yes
-	ssse3 = yes
-endif
-
-ifeq ($(ARCH),x86-64-sse41)
-	arch = x86_64
-	prefetch = yes
-	popcnt = yes
-	sse = yes
-	sse3 = yes
-	ssse3 = yes
-	sse41 = yes
-endif
-
-ifeq ($(ARCH),x86-64-modern)
-	arch = x86_64
-	prefetch = yes
-	popcnt = yes
-	sse = yes
-	sse3 = yes
-	ssse3 = yes
-	sse41 = yes
-endif
-
-ifeq ($(ARCH),x86-64-sse42)
-	arch = x86_64
-	prefetch = yes
-	popcnt = yes
-	sse = yes
-	sse3 = yes
-	ssse3 = yes
-	sse41 = yes
-	sse42 = yes
-endif
-
-ifeq ($(ARCH),x86-64-avx2)
-	arch = x86_64
-	prefetch = yes
-	popcnt = yes
-	sse = yes
-	sse3 = yes
-	ssse3 = yes
-	sse41 = yes
-	sse42 = yes
-	avx2 = yes
-endif
-
-ifeq ($(ARCH),x86-64-bmi2)
-	arch = x86_64
-	prefetch = yes
-	popcnt = yes
-	sse = yes
-	sse3 = yes
-	ssse3 = yes
-	sse41 = yes
-	sse42 = yes
-	avx2 = yes
-	pext = yes
-endif
-
-ifeq ($(ARCH),x86-64-avx512)
-	arch = x86_64
-	prefetch = yes
-	popcnt = yes
-	sse = yes
-	sse3 = yes
-	ssse3 = yes
-	sse41 = yes
-	sse42 = yes
-	avx2 = yes
-	pext = yes
-	avx512 = yes
-endif
-
 ifeq ($(ARCH),armv7)
 	arch = armv7
 	prefetch = yes
 	bits = 32
 endif
 
+ifeq ($(ARCH),armv7-neon)
+	arch = armv7
+	prefetch = yes
+	popcnt = yes
+	neon = yes
+	bits = 32
+endif
+
 ifeq ($(ARCH),armv8)
-	arch = armv8-a
+	arch = armv8
 	prefetch = yes
 	popcnt = yes
 	neon = yes
@@ -251,6 +289,8 @@ ifeq ($(ARCH),ppc-64)
 	prefetch = yes
 endif
 
+endif
+
 ### ==========================================================================
 ### Section 3. Low-level Configuration
 ### ==========================================================================
@@ -269,7 +309,7 @@ ifeq ($(COMP),gcc)
 	CXX=g++
 	CXXFLAGS += -pedantic -Wextra -Wshadow
 
-	ifeq ($(ARCH),$(filter $(ARCH),armv7 armv8))
+	ifeq ($(arch),$(filter $(arch),armv7 armv8))
 		ifeq ($(OS),Android)
 			CXXFLAGS += -m$(bits)
 			LDFLAGS += -m$(bits)
@@ -279,6 +319,10 @@ ifeq ($(COMP),gcc)
 		LDFLAGS += -m$(bits)
 	endif
 
+	ifeq ($(arch),$(filter $(arch),armv7))
+		LDFLAGS += -latomic
+	endif
+
 	ifneq ($(KERNEL),Darwin)
 	   LDFLAGS += -Wl,--no-as-needed
 	endif
@@ -326,7 +370,7 @@ ifeq ($(COMP),clang)
 	endif
 	endif
 
-	ifeq ($(ARCH),$(filter $(ARCH),armv7 armv8))
+	ifeq ($(arch),$(filter $(arch),armv7 armv8))
 		ifeq ($(OS),Android)
 			CXXFLAGS += -m$(bits)
 			LDFLAGS += -m$(bits)
@@ -351,8 +395,28 @@ endif
 endif
 
 ifeq ($(KERNEL),Darwin)
-	CXXFLAGS += -arch $(arch) -mmacosx-version-min=10.15
-	LDFLAGS += -arch $(arch) -mmacosx-version-min=10.15
+	CXXFLAGS += -arch $(arch) -mmacosx-version-min=10.14
+	LDFLAGS += -arch $(arch) -mmacosx-version-min=10.14
+	XCRUN = xcrun
+endif
+
+# To cross-compile for Android, NDK version r21 or later is recommended.
+# In earlier NDK versions, you'll need to pass -fno-addrsig if using GNU binutils.
+# Currently we don't know how to make PGO builds with the NDK yet.
+ifeq ($(COMP),ndk)
+	CXXFLAGS += -stdlib=libc++ -fPIE
+	ifeq ($(arch),armv7)
+		comp=armv7a-linux-androideabi16-clang
+		CXX=armv7a-linux-androideabi16-clang++
+		CXXFLAGS += -mthumb -march=armv7-a -mfloat-abi=softfp -mfpu=neon
+		STRIP=arm-linux-androideabi-strip
+	endif
+	ifeq ($(arch),armv8)
+		comp=aarch64-linux-android21-clang
+		CXX=aarch64-linux-android21-clang++
+		STRIP=aarch64-linux-android-strip
+	endif
+	LDFLAGS += -static-libstdc++ -pie -lm -latomic
 endif
 
 ### Travis CI script uses COMPILER to overwrite CXX
@@ -365,13 +429,26 @@ ifdef COMPCXX
 	CXX=$(COMPCXX)
 endif
 
+### Sometimes gcc is really clang
+ifeq ($(COMP),gcc)
+	gccversion = $(shell $(CXX) --version)
+	gccisclang = $(findstring clang,$(gccversion))
+	ifneq ($(gccisclang),)
+		profile_make = clang-profile-make
+		profile_use = clang-profile-use
+	endif
+endif
+
 ### On mingw use Windows threads, otherwise POSIX
 ifneq ($(comp),mingw)
+	CXXFLAGS += -DUSE_PTHREADS
 	# On Android Bionic's C library comes with its own pthread implementation bundled in
 	ifneq ($(OS),Android)
 		# Haiku has pthreads in its libroot, so only link it in on other platforms
 		ifneq ($(KERNEL),Haiku)
-			LDFLAGS += -lpthread
+			ifneq ($(COMP),ndk)
+				LDFLAGS += -lpthread
+			endif
 		endif
 	endif
 endif
@@ -416,7 +493,6 @@ endif
 ifeq ($(prefetch),yes)
 	ifeq ($(sse),yes)
 		CXXFLAGS += -msse
-		DEPENDFLAGS += -msse
 	endif
 else
 	CXXFLAGS += -DNO_PREFETCH
@@ -424,7 +500,7 @@ endif
 
 ### 3.6 popcnt
 ifeq ($(popcnt),yes)
-	ifeq ($(arch),$(filter $(arch),ppc64 armv8-a arm64))
+	ifeq ($(arch),$(filter $(arch),ppc64 armv7 armv8 arm64))
 		CXXFLAGS += -DUSE_POPCNT
 	else ifeq ($(comp),icc)
 		CXXFLAGS += -msse3 -DUSE_POPCNT
@@ -433,6 +509,7 @@ ifeq ($(popcnt),yes)
 	endif
 endif
 
+
 ifeq ($(avx2),yes)
 	CXXFLAGS += -DUSE_AVX2
 	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
@@ -443,14 +520,21 @@ endif
 ifeq ($(avx512),yes)
 	CXXFLAGS += -DUSE_AVX512
 	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
-		CXXFLAGS += -mavx512bw
+		CXXFLAGS += -mavx512f -mavx512bw
 	endif
 endif
 
-ifeq ($(sse42),yes)
-	CXXFLAGS += -DUSE_SSE42
+ifeq ($(vnni256),yes)
+	CXXFLAGS += -DUSE_VNNI
 	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
-		CXXFLAGS += -msse4.2
+		CXXFLAGS += -mavx512f -mavx512bw -mavx512vnni -mavx512dq -mavx512vl -mprefer-vector-width=256
+	endif
+endif
+
+ifeq ($(vnni512),yes)
+	CXXFLAGS += -DUSE_VNNI
+	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
+		CXXFLAGS += -mavx512vnni -mavx512dq -mavx512vl
 	endif
 endif
 
@@ -468,19 +552,29 @@ ifeq ($(ssse3),yes)
 	endif
 endif
 
-ifeq ($(sse3),yes)
-	CXXFLAGS += -DUSE_SSE3
+ifeq ($(sse2),yes)
+	CXXFLAGS += -DUSE_SSE2
 	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
-		CXXFLAGS += -msse3
+		CXXFLAGS += -msse2
+	endif
+endif
+
+ifeq ($(mmx),yes)
+	CXXFLAGS += -DUSE_MMX
+	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
+		CXXFLAGS += -mmmx
 	endif
 endif
 
 ifeq ($(neon),yes)
 	CXXFLAGS += -DUSE_NEON
-endif
-
-ifeq ($(arch),x86_64)
-	CXXFLAGS += -DUSE_SSE2
+	ifeq ($(KERNEL),Linux)
+	ifneq ($(COMP),ndk)
+	ifneq ($(arch),armv8)
+		CXXFLAGS += -mfpu=neon
+	endif
+	endif
+	endif
 endif
 
 ### 3.7 pext
@@ -496,18 +590,43 @@ endif
 ### needs access to the optimization flags.
 ifeq ($(optimize),yes)
 ifeq ($(debug), no)
-	ifeq ($(comp),$(filter $(comp),gcc clang))
+	ifeq ($(COMP),ndk)
+		CXXFLAGS += -flto=thin
+		LDFLAGS += $(CXXFLAGS)
+	else ifeq ($(comp),clang)
+		CXXFLAGS += -flto=thin
+		ifneq ($(findstring MINGW,$(KERNEL)),)
+			CXXFLAGS += -fuse-ld=lld
+		else ifneq ($(findstring MSYS,$(KERNEL)),)
+			CXXFLAGS += -fuse-ld=lld
+		endif
+		LDFLAGS += $(CXXFLAGS)
+
+# GCC and CLANG use different methods for parallelizing LTO and CLANG pretends to be
+# GCC on some systems.
+	else ifeq ($(comp),gcc)
+	ifeq ($(gccisclang),)
 		CXXFLAGS += -flto
+		LDFLAGS += $(CXXFLAGS) -flto=jobserver
+		ifneq ($(findstring MINGW,$(KERNEL)),)
+			LDFLAGS += -save-temps
+		else ifneq ($(findstring MSYS,$(KERNEL)),)
+			LDFLAGS += -save-temps
+		endif
+	else
+		CXXFLAGS += -flto=thin
 		LDFLAGS += $(CXXFLAGS)
 	endif
 
 # To use LTO and static linking on windows, the tool chain requires a recent gcc:
-# gcc version 10.1 in msys2 or TDM-GCC version 9.2 are know to work, older might not.
+# gcc version 10.1 in msys2 or TDM-GCC version 9.2 are known to work, older might not.
 # So, only enable it for a cross from Linux by default.
-	ifeq ($(comp),mingw)
+	else ifeq ($(comp),mingw)
 	ifeq ($(KERNEL),Linux)
+	ifneq ($(arch),i386)
 		CXXFLAGS += -flto
-		LDFLAGS += $(CXXFLAGS)
+		LDFLAGS += $(CXXFLAGS) -flto=jobserver
+	endif
 	endif
 	endif
 endif
@@ -524,6 +643,7 @@ endif
 ### Section 4. Public Targets
 ### ==========================================================================
 
+
 help:
 	@echo ""
 	@echo "To compile stockfish, type: "
@@ -532,31 +652,34 @@ help:
 	@echo ""
 	@echo "Supported targets:"
 	@echo ""
+	@echo "help                    > Display architecture details"
 	@echo "build                   > Standard build"
-	@echo "profile-build           > Standard build with PGO"
+	@echo "net                     > Download the default nnue net"
+	@echo "profile-build           > Faster build (with profile-guided optimization)"
 	@echo "strip                   > Strip executable"
 	@echo "install                 > Install executable"
 	@echo "clean                   > Clean up"
-	@echo "net                     > Download the default nnue net"
 	@echo ""
 	@echo "Supported archs:"
 	@echo ""
+	@echo "x86-64-vnni512          > x86 64-bit with vnni support 512bit wide"
+	@echo "x86-64-vnni256          > x86 64-bit with vnni support 256bit wide"
 	@echo "x86-64-avx512           > x86 64-bit with avx512 support"
 	@echo "x86-64-bmi2             > x86 64-bit with bmi2 support"
 	@echo "x86-64-avx2             > x86 64-bit with avx2 support"
-	@echo "x86-64-sse42            > x86 64-bit with sse42 support"
-	@echo "x86-64-modern           > x86 64-bit with sse41 support (x86-64-sse41)"
-	@echo "x86-64-sse41            > x86 64-bit with sse41 support"
+	@echo "x86-64-sse41-popcnt     > x86 64-bit with sse41 and popcnt support"
+	@echo "x86-64-modern           > common modern CPU, currently x86-64-sse41-popcnt"
 	@echo "x86-64-ssse3            > x86 64-bit with ssse3 support"
 	@echo "x86-64-sse3-popcnt      > x86 64-bit with sse3 and popcnt support"
-	@echo "x86-64-sse3             > x86 64-bit with sse3 support"
-	@echo "x86-64                  > x86 64-bit generic"
-	@echo "x86-32                  > x86 32-bit (also enables SSE)"
-	@echo "x86-32-old              > x86 32-bit fall back for old hardware"
+	@echo "x86-64                  > x86 64-bit generic (with sse2 support)"
+	@echo "x86-32-sse41-popcnt     > x86 32-bit with sse41 and popcnt support"
+	@echo "x86-32-sse2             > x86 32-bit with sse2 support"
+	@echo "x86-32                  > x86 32-bit generic (with mmx and sse support)"
 	@echo "ppc-64                  > PPC 64-bit"
 	@echo "ppc-32                  > PPC 32-bit"
 	@echo "armv7                   > ARMv7 32-bit"
-	@echo "armv8                   > ARMv8 64-bit"
+	@echo "armv7-neon              > ARMv7 32-bit with popcnt and neon"
+	@echo "armv8                   > ARMv8 64-bit with popcnt and neon"
 	@echo "apple-silicon           > Apple silicon ARM64"
 	@echo "general-64              > unspecified 64-bit"
 	@echo "general-32              > unspecified 32-bit"
@@ -567,30 +690,37 @@ help:
 	@echo "mingw                   > Gnu compiler with MinGW under Windows"
 	@echo "clang                   > LLVM Clang compiler"
 	@echo "icc                     > Intel compiler"
+	@echo "ndk                     > Google NDK to cross-compile for Android"
 	@echo ""
 	@echo "Simple examples. If you don't know what to do, you likely want to run: "
 	@echo ""
-	@echo "make -j build ARCH=x86-64    (This is for 64-bit systems)"
-	@echo "make -j build ARCH=x86-32    (This is for 32-bit systems)"
+	@echo "make -j build ARCH=x86-64  (A portable, slow compile for 64-bit systems)"
+	@echo "make -j build ARCH=x86-32  (A portable, slow compile for 32-bit systems)"
 	@echo ""
-	@echo "Advanced examples, for experienced users: "
+	@echo "Advanced examples, for experienced users looking for performance: "
 	@echo ""
-	@echo "make -j build ARCH=x86-64-modern COMP=clang"
-	@echo "make -j profile-build ARCH=x86-64-bmi2 COMP=gcc COMPCXX=g++-4.8"
-	@echo ""
-	@echo "The selected architecture $(ARCH) enables the following configuration: "
+	@echo "make    help  ARCH=x86-64-bmi2"
+	@echo "make -j profile-build ARCH=x86-64-bmi2 COMP=gcc COMPCXX=g++-9.0"
+	@echo "make -j build ARCH=x86-64-ssse3 COMP=clang"
 	@echo ""
+	@echo "-------------------------------"
+ifeq ($(SUPPORTED_ARCH)$(help_skip_sanity), true)
+	@echo "The selected architecture $(ARCH) will enable the following configuration: "
 	@$(MAKE) ARCH=$(ARCH) COMP=$(COMP) config-sanity
+else
+	@echo "Specify a supported architecture with the ARCH option for more details"
+	@echo ""
+endif
 
 
 .PHONY: help build profile-build strip install clean net objclean profileclean \
         config-sanity icc-profile-use icc-profile-make gcc-profile-use gcc-profile-make \
         clang-profile-use clang-profile-make
 
-build: config-sanity
+build: config-sanity net
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) all
 
-profile-build: config-sanity objclean profileclean
+profile-build: net config-sanity objclean profileclean
 	@echo ""
 	@echo "Step 1/4. Building instrumented executable ..."
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) $(profile_make)
@@ -606,23 +736,40 @@ profile-build: config-sanity objclean profileclean
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) profileclean
 
 strip:
-	strip $(EXE)
+	$(STRIP) $(EXE)
 
 install:
 	-mkdir -p -m 755 $(BINDIR)
 	-cp $(EXE) $(BINDIR)
 	-strip $(BINDIR)/$(EXE)
 
-#clean all
+# clean all
 clean: objclean profileclean
 	@rm -f .depend *~ core
 
+# evaluation network (nnue)
 net:
-	$(eval nnuenet := $(shell grep EvalFile ucioption.cpp | grep Option | sed 's/.*\(nn-[a-z0-9]\{12\}.nnue\).*/\1/'))
+	$(eval nnuenet := $(shell grep EvalFileDefaultName evaluate.h | grep define | sed 's/.*\(nn-[a-z0-9]\{12\}.nnue\).*/\1/'))
 	@echo "Default net: $(nnuenet)"
 	$(eval nnuedownloadurl := https://tests.stockfishchess.org/api/nn/$(nnuenet))
-	$(eval curl_or_wget := $(shell if hash curl 2>/dev/null; then echo "curl -sL"; elif hash wget 2>/dev/null; then echo "wget -qO-"; fi))
-	@if test -f "$(nnuenet)"; then echo "Already available."; else echo "Downloading $(nnuedownloadurl)"; $(curl_or_wget) $(nnuedownloadurl) > $(nnuenet); fi
+	$(eval curl_or_wget := $(shell if hash curl 2>/dev/null; then echo "curl -skL"; elif hash wget 2>/dev/null; then echo "wget -qO-"; fi))
+	@if test -f "$(nnuenet)"; then \
+            echo "Already available."; \
+         else \
+            if [ "x$(curl_or_wget)" = "x" ]; then \
+               echo "Automatic download failed: neither curl nor wget is installed. Install one of these tools or download the net manually"; exit 1; \
+            else \
+               echo "Downloading $(nnuedownloadurl)"; $(curl_or_wget) $(nnuedownloadurl) > $(nnuenet);\
+            fi; \
+        fi;
+	$(eval shasum_command := $(shell if hash shasum 2>/dev/null; then echo "shasum -a 256 "; elif hash sha256sum 2>/dev/null; then echo "sha256sum "; fi))
+	@if [ "x$(shasum_command)" != "x" ]; then \
+	    if [ "$(nnuenet)" != "nn-"`$(shasum_command) $(nnuenet) | cut -c1-12`".nnue" ]; then \
+                echo "Failed download or $(nnuenet) corrupted, please delete!"; exit 1; \
+            fi \
+         else \
+            echo "shasum / sha256sum not found, skipping net validation"; \
+        fi
 
 # clean binaries and objects
 objclean:
@@ -631,7 +778,7 @@ objclean:
 # clean auxiliary profiling files
 profileclean:
 	@rm -rf profdir
-	@rm -f bench.txt *.gcda *.gcno ./syzygy/*.gcda ./nnue/*.gcda ./nnue/features/*.gcda
+	@rm -f bench.txt *.gcda *.gcno ./syzygy/*.gcda ./nnue/*.gcda ./nnue/features/*.gcda *.s
 	@rm -f stockfish.profdata *.profraw
 
 default:
@@ -655,14 +802,16 @@ config-sanity:
 	@echo "os: '$(OS)'"
 	@echo "prefetch: '$(prefetch)'"
 	@echo "popcnt: '$(popcnt)'"
+	@echo "pext: '$(pext)'"
 	@echo "sse: '$(sse)'"
-	@echo "sse3: '$(sse3)'"
+	@echo "mmx: '$(mmx)'"
+	@echo "sse2: '$(sse2)'"
 	@echo "ssse3: '$(ssse3)'"
 	@echo "sse41: '$(sse41)'"
-	@echo "sse42: '$(sse42)'"
 	@echo "avx2: '$(avx2)'"
-	@echo "pext: '$(pext)'"
 	@echo "avx512: '$(avx512)'"
+	@echo "vnni256: '$(vnni256)'"
+	@echo "vnni512: '$(vnni512)'"
 	@echo "neon: '$(neon)'"
 	@echo ""
 	@echo "Flags:"
@@ -675,25 +824,29 @@ config-sanity:
 	@test "$(debug)" = "yes" || test "$(debug)" = "no"
 	@test "$(sanitize)" = "undefined" || test "$(sanitize)" = "thread" || test "$(sanitize)" = "address" || test "$(sanitize)" = "no"
 	@test "$(optimize)" = "yes" || test "$(optimize)" = "no"
+	@test "$(SUPPORTED_ARCH)" = "true"
 	@test "$(arch)" = "any" || test "$(arch)" = "x86_64" || test "$(arch)" = "i386" || \
 	 test "$(arch)" = "ppc64" || test "$(arch)" = "ppc" || \
-	 test "$(arch)" = "armv7" || test "$(arch)" = "armv8-a" || test "$(arch)" = "arm64"
+	 test "$(arch)" = "armv7" || test "$(arch)" = "armv8" || test "$(arch)" = "arm64"
 	@test "$(bits)" = "32" || test "$(bits)" = "64"
 	@test "$(prefetch)" = "yes" || test "$(prefetch)" = "no"
 	@test "$(popcnt)" = "yes" || test "$(popcnt)" = "no"
+	@test "$(pext)" = "yes" || test "$(pext)" = "no"
 	@test "$(sse)" = "yes" || test "$(sse)" = "no"
-	@test "$(sse3)" = "yes" || test "$(sse3)" = "no"
+	@test "$(mmx)" = "yes" || test "$(mmx)" = "no"
+	@test "$(sse2)" = "yes" || test "$(sse2)" = "no"
 	@test "$(ssse3)" = "yes" || test "$(ssse3)" = "no"
 	@test "$(sse41)" = "yes" || test "$(sse41)" = "no"
-	@test "$(sse42)" = "yes" || test "$(sse42)" = "no"
 	@test "$(avx2)" = "yes" || test "$(avx2)" = "no"
-	@test "$(pext)" = "yes" || test "$(pext)" = "no"
 	@test "$(avx512)" = "yes" || test "$(avx512)" = "no"
+	@test "$(vnni256)" = "yes" || test "$(vnni256)" = "no"
+	@test "$(vnni512)" = "yes" || test "$(vnni512)" = "no"
 	@test "$(neon)" = "yes" || test "$(neon)" = "no"
-	@test "$(comp)" = "gcc" || test "$(comp)" = "icc" || test "$(comp)" = "mingw" || test "$(comp)" = "clang"
+	@test "$(comp)" = "gcc" || test "$(comp)" = "icc" || test "$(comp)" = "mingw" || test "$(comp)" = "clang" \
+	|| test "$(comp)" = "armv7a-linux-androideabi16-clang"  || test "$(comp)" = "aarch64-linux-android21-clang"
 
 $(EXE): $(OBJS)
-	$(CXX) -o $@ $(OBJS) $(LDFLAGS)
+	+$(CXX) -o $@ $(OBJS) $(LDFLAGS)
 
 clang-profile-make:
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) \
@@ -702,7 +855,7 @@ clang-profile-make:
 	all
 
 clang-profile-use:
-	llvm-profdata merge -output=stockfish.profdata *.profraw
+	$(XCRUN) llvm-profdata merge -output=stockfish.profdata *.profraw
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) \
 	EXTRACXXFLAGS='-fprofile-instr-use=stockfish.profdata' \
 	EXTRALDFLAGS='-fprofile-use ' \

src/benchmark.cpp

@@ -95,8 +95,9 @@ const vector<string> Defaults = {
 /// setup_bench() builds a list of UCI commands to be run by bench. There
 /// are five parameters: TT size in MB, number of search threads that
 /// should be used, the limit value spent for each position, a file name
-/// where to look for positions in FEN format and the type of the limit:
-/// depth, perft, nodes and movetime (in millisecs).
+/// where to look for positions in FEN format, the type of the limit:
+/// depth, perft, nodes and movetime (in millisecs), and evaluation type
+/// mixed (default), classical, NNUE.
 ///
 /// bench -> search default positions up to depth 13
 /// bench 64 1 15 -> search default positions up to depth 15 (TT = 64MB)
@@ -115,6 +116,7 @@ vector<string> setup_bench(const Position& current, istream& is) {
   string limit     = (is >> token) ? token : "13";
   string fenFile   = (is >> token) ? token : "default";
   string limitType = (is >> token) ? token : "depth";
+  string evalType  = (is >> token) ? token : "mixed";
 
   go = limitType == "eval" ? "eval" : "go " + limitType + " " + limit;
 
@@ -146,13 +148,20 @@ vector<string> setup_bench(const Position& current, istream& is) {
   list.emplace_back("setoption name Hash value " + ttSize);
   list.emplace_back("ucinewgame");
 
+  size_t posCounter = 0;
+
   for (const string& fen : fens)
       if (fen.find("setoption") != string::npos)
           list.emplace_back(fen);
       else
       {
+          if (evalType == "classical" || (evalType == "mixed" && posCounter % 2 == 0))
+              list.emplace_back("setoption name Use NNUE value false");
+          else if (evalType == "NNUE" || (evalType == "mixed" && posCounter % 2 != 0))
+              list.emplace_back("setoption name Use NNUE value true");
           list.emplace_back("position fen " + fen);
           list.emplace_back(go);
+          ++posCounter;
       }
 
   return list;

src/bitboard.cpp

@@ -39,6 +39,16 @@ namespace {
   Bitboard BishopTable[0x1480]; // To store bishop attacks
 
   void init_magics(PieceType pt, Bitboard table[], Magic magics[]);
+
+}
+
+
+/// safe_destination() returns the bitboard of target square for the given step
+/// from the given square. If the step is off the board, returns empty bitboard.
+
+inline Bitboard safe_destination(Square s, int step) {
+    Square to = Square(s + step);
+    return is_ok(to) && distance(s, to) <= 2 ? square_bb(to) : Bitboard(0);
 }
 
 
@@ -110,7 +120,7 @@ namespace {
     Direction   RookDirections[4] = {NORTH, SOUTH, EAST, WEST};
     Direction BishopDirections[4] = {NORTH_EAST, SOUTH_EAST, SOUTH_WEST, NORTH_WEST};
 
-    for(Direction d : (pt == ROOK ? RookDirections : BishopDirections))
+    for (Direction d : (pt == ROOK ? RookDirections : BishopDirections))
     {
         Square s = sq;
         while(safe_destination(s, d) && !(occupied & s))

src/bitboard.h

@@ -279,16 +279,6 @@ inline int edge_distance(File f) { return std::min(f, File(FILE_H - f)); }
 inline int edge_distance(Rank r) { return std::min(r, Rank(RANK_8 - r)); }
 
 
-/// safe_destination() returns the bitboard of target square for the given step
-/// from the given square. If the step is off the board, returns empty bitboard.
-
-inline Bitboard safe_destination(Square s, int step)
-{
-    Square to = Square(s + step);
-    return is_ok(to) && distance(s, to) <= 2 ? square_bb(to) : Bitboard(0);
-}
-
-
 /// attacks_bb(Square) returns the pseudo attacks of the give piece type
 /// assuming an empty board.
 

src/evaluate.cpp

@@ -20,46 +20,126 @@
 #include <cassert>
 #include <cstdlib>
 #include <cstring>   // For std::memset
+#include <fstream>
 #include <iomanip>
 #include <sstream>
 #include <iostream>
+#include <streambuf>
+#include <vector>
 
 #include "bitboard.h"
 #include "evaluate.h"
 #include "material.h"
+#include "misc.h"
 #include "pawns.h"
 #include "thread.h"
 #include "uci.h"
+#include "incbin/incbin.h"
+
+
+// Macro to embed the default NNUE file data in the engine binary (using incbin.h, by Dale Weiler).
+// This macro invocation will declare the following three variables
+//     const unsigned char        gEmbeddedNNUEData[];  // a pointer to the embedded data
+//     const unsigned char *const gEmbeddedNNUEEnd;     // a marker to the end
+//     const unsigned int         gEmbeddedNNUESize;    // the size of the embedded file
+// Note that this does not work in Microsof Visual Studio.
+#if !defined(_MSC_VER) && !defined(NNUE_EMBEDDING_OFF)
+  INCBIN(EmbeddedNNUE, EvalFileDefaultName);
+#else
+  const unsigned char        gEmbeddedNNUEData[1] = {0x0};
+  const unsigned char *const gEmbeddedNNUEEnd = &gEmbeddedNNUEData[1];
+  const unsigned int         gEmbeddedNNUESize = 1;
+#endif
+
+
+using namespace std;
+using namespace Eval::NNUE;
 
 namespace Eval {
 
   bool useNNUE;
-  std::string eval_file_loaded="None";
+  string eval_file_loaded = "None";
+
+  /// init_NNUE() tries to load a nnue network at startup time, or when the engine
+  /// receives a UCI command "setoption name EvalFile value nn-[a-z0-9]{12}.nnue"
+  /// The name of the nnue network is always retrieved from the EvalFile option.
+  /// We search the given network in three locations: internally (the default
+  /// network may be embedded in the binary), in the active working directory and
+  /// in the engine directory. Distro packagers may define the DEFAULT_NNUE_DIRECTORY
+  /// variable to have the engine search in a special directory in their distro.
 
   void init_NNUE() {
 
     useNNUE = Options["Use NNUE"];
-    std::string eval_file = std::string(Options["EvalFile"]);
-    if (useNNUE && eval_file_loaded != eval_file)
-        if (Eval::NNUE::load_eval_file(eval_file))
-            eval_file_loaded = eval_file;
+    if (!useNNUE)
+        return;
+
+    string eval_file = string(Options["EvalFile"]);
+
+    #if defined(DEFAULT_NNUE_DIRECTORY)
+    #define stringify2(x) #x
+    #define stringify(x) stringify2(x)
+    vector<string> dirs = { "<internal>" , "" , CommandLine::binaryDirectory , stringify(DEFAULT_NNUE_DIRECTORY) };
+    #else
+    vector<string> dirs = { "<internal>" , "" , CommandLine::binaryDirectory };
+    #endif
+
+    for (string directory : dirs)
+        if (eval_file_loaded != eval_file)
+        {
+            if (directory != "<internal>")
+            {
+                ifstream stream(directory + eval_file, ios::binary);
+                if (load_eval(eval_file, stream))
+                    eval_file_loaded = eval_file;
+            }
+
+            if (directory == "<internal>" && eval_file == EvalFileDefaultName)
+            {
+                // C++ way to prepare a buffer for a memory stream
+                class MemoryBuffer : public basic_streambuf<char> {
+                    public: MemoryBuffer(char* p, size_t n) { setg(p, p, p + n); setp(p, p + n); }
+                };
+
+                MemoryBuffer buffer(const_cast<char*>(reinterpret_cast<const char*>(gEmbeddedNNUEData)),
+                                    size_t(gEmbeddedNNUESize));
+
+                istream stream(&buffer);
+                if (load_eval(eval_file, stream))
+                    eval_file_loaded = eval_file;
+            }
+        }
   }
 
+  /// verify_NNUE() verifies that the last net used was loaded successfully
   void verify_NNUE() {
 
-    std::string eval_file = std::string(Options["EvalFile"]);
+    string eval_file = string(Options["EvalFile"]);
+
     if (useNNUE && eval_file_loaded != eval_file)
     {
-        std::cerr << "Use of NNUE evaluation, but the file " << eval_file << " was not loaded successfully. "
-                  << "These network evaluation parameters must be available, compatible with this version of the code. "
-                  << "The UCI option EvalFile might need to specify the full path, including the directory/folder name, to the file." << std::endl;
-        std::exit(EXIT_FAILURE);
+        UCI::OptionsMap defaults;
+        UCI::init(defaults);
+
+        string msg1 = "If the UCI option \"Use NNUE\" is set to true, network evaluation parameters compatible with the engine must be available.";
+        string msg2 = "The option is set to true, but the network file " + eval_file + " was not loaded successfully.";
+        string msg3 = "The UCI option EvalFile might need to specify the full path, including the directory name, to the network file.";
+        string msg4 = "The default net can be downloaded from: https://tests.stockfishchess.org/api/nn/" + string(defaults["EvalFile"]);
+        string msg5 = "The engine will be terminated now.";
+
+        sync_cout << "info string ERROR: " << msg1 << sync_endl;
+        sync_cout << "info string ERROR: " << msg2 << sync_endl;
+        sync_cout << "info string ERROR: " << msg3 << sync_endl;
+        sync_cout << "info string ERROR: " << msg4 << sync_endl;
+        sync_cout << "info string ERROR: " << msg5 << sync_endl;
+
+        exit(EXIT_FAILURE);
     }
 
     if (useNNUE)
-        sync_cout << "info string NNUE evaluation using " << eval_file << " enabled." << sync_endl;
+        sync_cout << "info string NNUE evaluation using " << eval_file << " enabled" << sync_endl;
     else
-        sync_cout << "info string classical evaluation enabled." << sync_endl;
+        sync_cout << "info string classical evaluation enabled" << sync_endl;
   }
 }
 
@@ -107,9 +187,11 @@ using namespace Trace;
 namespace {
 
   // Threshold for lazy and space evaluation
-  constexpr Value LazyThreshold1  = Value(1400);
-  constexpr Value LazyThreshold2  = Value(1300);
+  constexpr Value LazyThreshold1 =  Value(1400);
+  constexpr Value LazyThreshold2 =  Value(1300);
   constexpr Value SpaceThreshold = Value(12222);
+  constexpr Value NNUEThreshold1 =   Value(550);
+  constexpr Value NNUEThreshold2 =   Value(150);
 
   // KingAttackWeights[PieceType] contains king attack weights by piece type
   constexpr int KingAttackWeights[PIECE_TYPE_NB] = { 0, 0, 81, 52, 44, 10 };
@@ -145,26 +227,26 @@ namespace {
 
   // Outpost[knight/bishop] contains bonuses for each knight or bishop occupying a
   // pawn protected square on rank 4 to 6 which is also safe from a pawn attack.
-  constexpr Score Outpost[] = { S(56, 36), S(30, 23) };
+  constexpr Score Outpost[] = { S(56, 34), S(31, 23) };
 
   // PassedRank[Rank] contains a bonus according to the rank of a passed pawn
   constexpr Score PassedRank[RANK_NB] = {
-    S(0, 0), S(10, 28), S(17, 33), S(15, 41), S(62, 72), S(168, 177), S(276, 260)
+    S(0, 0), S(9, 28), S(15, 31), S(17, 39), S(64, 70), S(171, 177), S(277, 260)
   };
 
   // RookOnFile[semiopen/open] contains bonuses for each rook when there is
   // no (friendly) pawn on the rook file.
-  constexpr Score RookOnFile[] = { S(19, 7), S(48, 29) };
+  constexpr Score RookOnFile[] = { S(19, 7), S(48, 27) };
 
   // ThreatByMinor/ByRook[attacked PieceType] contains bonuses according to
   // which piece type attacks which one. Attacks on lesser pieces which are
   // pawn-defended are not considered.
   constexpr Score ThreatByMinor[PIECE_TYPE_NB] = {
-    S(0, 0), S(5, 32), S(57, 41), S(77, 56), S(88, 119), S(79, 161)
+    S(0, 0), S(5, 32), S(55, 41), S(77, 56), S(89, 119), S(79, 162)
   };
 
   constexpr Score ThreatByRook[PIECE_TYPE_NB] = {
-    S(0, 0), S(3, 46), S(37, 68), S(42, 60), S(0, 38), S(58, 41)
+    S(0, 0), S(3, 44), S(37, 68), S(42, 60), S(0, 39), S(58, 43)
   };
 
   // Assorted bonuses and penalties
@@ -180,7 +262,6 @@ namespace {
   constexpr Score MinorBehindPawn     = S( 18,  3);
   constexpr Score PassedFile          = S( 11,  8);
   constexpr Score PawnlessFlank       = S( 17, 95);
-  constexpr Score QueenInfiltration   = S( -2, 14);
   constexpr Score ReachableOutpost    = S( 31, 22);
   constexpr Score RestrictedPiece     = S(  7,  7);
   constexpr Score RookOnKingRing      = S( 16,  0);
@@ -282,8 +363,8 @@ namespace {
     attackedBy2[Us] = dblAttackByPawn | (attackedBy[Us][KING] & attackedBy[Us][PAWN]);
 
     // Init our king safety tables
-    Square s = make_square(Utility::clamp(file_of(ksq), FILE_B, FILE_G),
-                           Utility::clamp(rank_of(ksq), RANK_2, RANK_7));
+    Square s = make_square(std::clamp(file_of(ksq), FILE_B, FILE_G),
+                           std::clamp(rank_of(ksq), RANK_2, RANK_7));
     kingRing[Us] = attacks_bb<KING>(s) | s;
 
     kingAttackersCount[Them] = popcount(kingRing[Us] & pe->pawn_attacks(Them));
@@ -345,7 +426,8 @@ namespace {
         {
             // Bonus if the piece is on an outpost square or can reach one
             // Reduced bonus for knights (BadOutpost) if few relevant targets
-            bb = OutpostRanks & attackedBy[Us][PAWN] & ~pe->pawn_attacks_span(Them);
+            bb = OutpostRanks & (attackedBy[Us][PAWN] | shift<Down>(pos.pieces(PAWN)))
+                              & ~pe->pawn_attacks_span(Them);
             Bitboard targets = pos.pieces(Them) & ~pos.pieces(PAWN);
 
             if (   Pt == KNIGHT
@@ -422,10 +504,6 @@ namespace {
             Bitboard queenPinners;
             if (pos.slider_blockers(pos.pieces(Them, ROOK, BISHOP), s, queenPinners))
                 score -= WeakQueen;
-
-            // Bonus for queen on weak square in enemy camp
-            if (relative_rank(Us, s) > RANK_4 && (~pe->pawn_attacks_span(Them) & s))
-                score += QueenInfiltration;
         }
     }
     if (T)
@@ -683,8 +761,8 @@ namespace {
             Square blockSq = s + Up;
 
             // Adjust bonus based on the king's proximity
-            bonus += make_score(0, (  (king_proximity(Them, blockSq) * 19) / 4
-                                     - king_proximity(Us,   blockSq) *  2) * w);
+            bonus += make_score(0, (  king_proximity(Them, blockSq) * 19 / 4
+                                    - king_proximity(Us,   blockSq) *  2) * w);
 
             // If blockSq is not the queening square then consider also a second push
             if (r != RANK_7)
@@ -728,7 +806,7 @@ namespace {
 
 
   // Evaluation::space() computes a space evaluation for a given side, aiming to improve game
-  // play in the opening. It is based on the number of safe squares on the 4 central files
+  // play in the opening. It is based on the number of safe squares on the four central files
   // on ranks 2 to 4. Completely safe squares behind a friendly pawn are counted twice.
   // Finally, the space bonus is multiplied by a weight which decreases according to occupancy.
 
@@ -801,7 +879,7 @@ namespace {
     // Now apply the bonus: note that we find the attacking side by extracting the
     // sign of the midgame or endgame values, and that we carefully cap the bonus
     // so that the midgame and endgame scores do not change sign after the bonus.
-    int u = ((mg > 0) - (mg < 0)) * Utility::clamp(complexity + 50, -abs(mg), 0);
+    int u = ((mg > 0) - (mg < 0)) * std::clamp(complexity + 50, -abs(mg), 0);
     int v = ((eg > 0) - (eg < 0)) * std::max(complexity, -abs(eg));
 
     mg += u;
@@ -926,9 +1004,6 @@ make_v:
     // Side to move point of view
     v = (pos.side_to_move() == WHITE ? v : -v) + Tempo;
 
-    // Damp down the evaluation linearly when shuffling
-    v = v * (100 - pos.rule50_count()) / 100;
-
     return v;
   }
 
@@ -940,10 +1015,21 @@ make_v:
 
 Value Eval::evaluate(const Position& pos) {
 
-  if (Eval::useNNUE)
-      return NNUE::evaluate(pos);
-  else
-      return Evaluation<NO_TRACE>(pos).value();
+  bool classical = !Eval::useNNUE
+                ||  abs(eg_value(pos.psq_score())) * 16 > NNUEThreshold1 * (16 + pos.rule50_count());
+  Value v = classical ? Evaluation<NO_TRACE>(pos).value()
+                      : NNUE::evaluate(pos) * 5 / 4 + Tempo;
+
+  if (classical && Eval::useNNUE && abs(v) * 16 < NNUEThreshold2 * (16 + pos.rule50_count()))
+      v = NNUE::evaluate(pos) * 5 / 4 + Tempo;
+
+  // Damp down the evaluation linearly when shuffling
+  v = v * (100 - pos.rule50_count()) / 100;
+
+  // Guarantee evaluation does not hit the tablebase range
+  v = std::clamp(v, VALUE_TB_LOSS_IN_MAX_PLY + 1, VALUE_TB_WIN_IN_MAX_PLY - 1);
+
+  return v;
 }
 
 /// trace() is like evaluate(), but instead of returning a value, it returns
@@ -961,42 +1047,46 @@ std::string Eval::trace(const Position& pos) {
 
   Value v;
 
-  if (Eval::useNNUE)
-  {
-      v = NNUE::evaluate(pos);
-  }
-  else
-  {
-      std::memset(scores, 0, sizeof(scores));
+  std::memset(scores, 0, sizeof(scores));
 
-      pos.this_thread()->contempt = SCORE_ZERO; // Reset any dynamic contempt
+  pos.this_thread()->contempt = SCORE_ZERO; // Reset any dynamic contempt
 
-      v = Evaluation<TRACE>(pos).value();
+  v = Evaluation<TRACE>(pos).value();
 
-      ss << std::showpoint << std::noshowpos << std::fixed << std::setprecision(2)
-         << "     Term    |    White    |    Black    |    Total   \n"
-         << "             |   MG    EG  |   MG    EG  |   MG    EG \n"
-         << " ------------+-------------+-------------+------------\n"
-         << "    Material | " << Term(MATERIAL)
-         << "   Imbalance | " << Term(IMBALANCE)
-         << "       Pawns | " << Term(PAWN)
-         << "     Knights | " << Term(KNIGHT)
-         << "     Bishops | " << Term(BISHOP)
-         << "       Rooks | " << Term(ROOK)
-         << "      Queens | " << Term(QUEEN)
-         << "    Mobility | " << Term(MOBILITY)
-         << " King safety | " << Term(KING)
-         << "     Threats | " << Term(THREAT)
-         << "      Passed | " << Term(PASSED)
-         << "       Space | " << Term(SPACE)
-         << "    Winnable | " << Term(WINNABLE)
-         << " ------------+-------------+-------------+------------\n"
-         << "       Total | " << Term(TOTAL);
-  }
+  ss << std::showpoint << std::noshowpos << std::fixed << std::setprecision(2)
+     << "     Term    |    White    |    Black    |    Total   \n"
+     << "             |   MG    EG  |   MG    EG  |   MG    EG \n"
+     << " ------------+-------------+-------------+------------\n"
+     << "    Material | " << Term(MATERIAL)
+     << "   Imbalance | " << Term(IMBALANCE)
+     << "       Pawns | " << Term(PAWN)
+     << "     Knights | " << Term(KNIGHT)
+     << "     Bishops | " << Term(BISHOP)
+     << "       Rooks | " << Term(ROOK)
+     << "      Queens | " << Term(QUEEN)
+     << "    Mobility | " << Term(MOBILITY)
+     << " King safety | " << Term(KING)
+     << "     Threats | " << Term(THREAT)
+     << "      Passed | " << Term(PASSED)
+     << "       Space | " << Term(SPACE)
+     << "    Winnable | " << Term(WINNABLE)
+     << " ------------+-------------+-------------+------------\n"
+     << "       Total | " << Term(TOTAL);
 
   v = pos.side_to_move() == WHITE ? v : -v;
 
-  ss << "\nFinal evaluation: " << to_cp(v) << " (white side)\n";
+  ss << "\nClassical evaluation: " << to_cp(v) << " (white side)\n";
+
+  if (Eval::useNNUE)
+  {
+      v = NNUE::evaluate(pos);
+      v = pos.side_to_move() == WHITE ? v : -v;
+      ss << "\nNNUE evaluation:      " << to_cp(v) << " (white side)\n";
+  }
+
+  v = evaluate(pos);
+  v = pos.side_to_move() == WHITE ? v : -v;
+  ss << "\nFinal evaluation:     " << to_cp(v) << " (white side)\n";
 
   return ss.str();
 }

src/evaluate.h

@@ -35,12 +35,17 @@ namespace Eval {
   void init_NNUE();
   void verify_NNUE();
 
+  // The default net name MUST follow the format nn-[SHA256 first 12 digits].nnue
+  // for the build process (profile-build and fishtest) to work. Do not change the
+  // name of the macro, as it is used in the Makefile.
+  #define EvalFileDefaultName   "nn-82215d0fd0df.nnue"
+
   namespace NNUE {
 
     Value evaluate(const Position& pos);
     Value compute_eval(const Position& pos);
     void  update_eval(const Position& pos);
-    bool  load_eval_file(const std::string& evalFile);
+    bool  load_eval(std::string streamName, std::istream& stream);
 
   } // namespace NNUE
 

src/incbin/UNLICENCE

@@ -0,0 +1,26 @@
+The file "incbin.h" is free and unencumbered software released into
+the public domain by Dale Weiler, see:
+   <https://github.com/graphitemaster/incbin>
+
+Anyone is free to copy, modify, publish, use, compile, sell, or
+distribute this software, either in source code form or as a compiled
+binary, for any purpose, commercial or non-commercial, and by any
+means.
+
+In jurisdictions that recognize copyright laws, the author or authors
+of this software dedicate any and all copyright interest in the
+software to the public domain. We make this dedication for the benefit
+of the public at large and to the detriment of our heirs and
+successors. We intend this dedication to be an overt act of
+relinquishment in perpetuity of all present and future rights to this
+software under copyright law.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+OTHER DEALINGS IN THE SOFTWARE.
+
+For more information, please refer to <http://unlicense.org/>

src/incbin/incbin.h

@@ -0,0 +1,368 @@
+/**
+ * @file incbin.h
+ * @author Dale Weiler
+ * @brief Utility for including binary files
+ *
+ * Facilities for including binary files into the current translation unit and
+ * making use from them externally in other translation units.
+ */
+#ifndef INCBIN_HDR
+#define INCBIN_HDR
+#include <limits.h>
+#if   defined(__AVX512BW__) || \
+      defined(__AVX512CD__) || \
+      defined(__AVX512DQ__) || \
+      defined(__AVX512ER__) || \
+      defined(__AVX512PF__) || \
+      defined(__AVX512VL__) || \
+      defined(__AVX512F__)
+# define INCBIN_ALIGNMENT_INDEX 6
+#elif defined(__AVX__)      || \
+      defined(__AVX2__)
+# define INCBIN_ALIGNMENT_INDEX 5
+#elif defined(__SSE__)      || \
+      defined(__SSE2__)     || \
+      defined(__SSE3__)     || \
+      defined(__SSSE3__)    || \
+      defined(__SSE4_1__)   || \
+      defined(__SSE4_2__)   || \
+      defined(__neon__)
+# define INCBIN_ALIGNMENT_INDEX 4
+#elif ULONG_MAX != 0xffffffffu
+# define INCBIN_ALIGNMENT_INDEX 3
+# else
+# define INCBIN_ALIGNMENT_INDEX 2
+#endif
+
+/* Lookup table of (1 << n) where `n' is `INCBIN_ALIGNMENT_INDEX' */
+#define INCBIN_ALIGN_SHIFT_0 1
+#define INCBIN_ALIGN_SHIFT_1 2
+#define INCBIN_ALIGN_SHIFT_2 4
+#define INCBIN_ALIGN_SHIFT_3 8
+#define INCBIN_ALIGN_SHIFT_4 16
+#define INCBIN_ALIGN_SHIFT_5 32
+#define INCBIN_ALIGN_SHIFT_6 64
+
+/* Actual alignment value */
+#define INCBIN_ALIGNMENT \
+    INCBIN_CONCATENATE( \
+        INCBIN_CONCATENATE(INCBIN_ALIGN_SHIFT, _), \
+        INCBIN_ALIGNMENT_INDEX)
+
+/* Stringize */
+#define INCBIN_STR(X) \
+    #X
+#define INCBIN_STRINGIZE(X) \
+    INCBIN_STR(X)
+/* Concatenate */
+#define INCBIN_CAT(X, Y) \
+    X ## Y
+#define INCBIN_CONCATENATE(X, Y) \
+    INCBIN_CAT(X, Y)
+/* Deferred macro expansion */
+#define INCBIN_EVAL(X) \
+    X
+#define INCBIN_INVOKE(N, ...) \
+    INCBIN_EVAL(N(__VA_ARGS__))
+
+/* Green Hills uses a different directive for including binary data */
+#if defined(__ghs__)
+#  if (__ghs_asm == 2)
+#    define INCBIN_MACRO ".file"
+/* Or consider the ".myrawdata" entry in the ld file */
+#  else
+#    define INCBIN_MACRO "\tINCBIN"
+#  endif
+#else
+#  define INCBIN_MACRO ".incbin"
+#endif
+
+#ifndef _MSC_VER
+#  define INCBIN_ALIGN \
+    __attribute__((aligned(INCBIN_ALIGNMENT)))
+#else
+#  define INCBIN_ALIGN __declspec(align(INCBIN_ALIGNMENT))
+#endif
+
+#if defined(__arm__) || /* GNU C and RealView */ \
+    defined(__arm) || /* Diab */ \
+    defined(_ARM) /* ImageCraft */
+#  define INCBIN_ARM
+#endif
+
+#ifdef __GNUC__
+/* Utilize .balign where supported */
+#  define INCBIN_ALIGN_HOST ".balign " INCBIN_STRINGIZE(INCBIN_ALIGNMENT) "\n"
+#  define INCBIN_ALIGN_BYTE ".balign 1\n"
+#elif defined(INCBIN_ARM)
+/*
+ * On arm assemblers, the alignment value is calculated as (1 << n) where `n' is
+ * the shift count. This is the value passed to `.align'
+ */
+#  define INCBIN_ALIGN_HOST ".align " INCBIN_STRINGIZE(INCBIN_ALIGNMENT_INDEX) "\n"
+#  define INCBIN_ALIGN_BYTE ".align 0\n"
+#else
+/* We assume other inline assembler's treat `.align' as `.balign' */
+#  define INCBIN_ALIGN_HOST ".align " INCBIN_STRINGIZE(INCBIN_ALIGNMENT) "\n"
+#  define INCBIN_ALIGN_BYTE ".align 1\n"
+#endif
+
+/* INCBIN_CONST is used by incbin.c generated files */
+#if defined(__cplusplus)
+#  define INCBIN_EXTERNAL extern "C"
+#  define INCBIN_CONST    extern const
+#else
+#  define INCBIN_EXTERNAL extern
+#  define INCBIN_CONST    const
+#endif
+
+/**
+ * @brief Optionally override the linker section into which data is emitted.
+ *
+ * @warning If you use this facility, you'll have to deal with platform-specific linker output
+ * section naming on your own
+ *
+ * Overriding the default linker output section, e.g for esp8266/Arduino:
+ * @code
+ * #define INCBIN_OUTPUT_SECTION ".irom.text"
+ * #include "incbin.h"
+ * INCBIN(Foo, "foo.txt");
+ * // Data is emitted into program memory that never gets copied to RAM
+ * @endcode
+ */
+#if !defined(INCBIN_OUTPUT_SECTION)
+#  if defined(__APPLE__)
+#    define INCBIN_OUTPUT_SECTION         ".const_data"
+#  else
+#    define INCBIN_OUTPUT_SECTION         ".rodata"
+#  endif
+#endif
+
+#if defined(__APPLE__)
+/* The directives are different for Apple branded compilers */
+#  define INCBIN_SECTION         INCBIN_OUTPUT_SECTION "\n"
+#  define INCBIN_GLOBAL(NAME)    ".globl " INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME "\n"
+#  define INCBIN_INT             ".long "
+#  define INCBIN_MANGLE          "_"
+#  define INCBIN_BYTE            ".byte "
+#  define INCBIN_TYPE(...)
+#else
+#  define INCBIN_SECTION         ".section " INCBIN_OUTPUT_SECTION "\n"
+#  define INCBIN_GLOBAL(NAME)    ".global " INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME "\n"
+#  if defined(__ghs__)
+#    define INCBIN_INT           ".word "
+#  else
+#    define INCBIN_INT           ".int "
+#  endif
+#  if defined(__USER_LABEL_PREFIX__)
+#    define INCBIN_MANGLE        INCBIN_STRINGIZE(__USER_LABEL_PREFIX__)
+#  else
+#    define INCBIN_MANGLE        ""
+#  endif
+#  if defined(INCBIN_ARM)
+/* On arm assemblers, `@' is used as a line comment token */
+#    define INCBIN_TYPE(NAME)    ".type " INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME ", %object\n"
+#  elif defined(__MINGW32__) || defined(__MINGW64__)
+/* Mingw doesn't support this directive either */
+#    define INCBIN_TYPE(NAME)
+#  else
+/* It's safe to use `@' on other architectures */
+#    define INCBIN_TYPE(NAME)    ".type " INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME ", @object\n"
+#  endif
+#  define INCBIN_BYTE            ".byte "
+#endif
+
+/* List of style types used for symbol names */
+#define INCBIN_STYLE_CAMEL 0
+#define INCBIN_STYLE_SNAKE 1
+
+/**
+ * @brief Specify the prefix to use for symbol names.
+ *
+ * By default this is `g', producing symbols of the form:
+ * @code
+ * #include "incbin.h"
+ * INCBIN(Foo, "foo.txt");
+ *
+ * // Now you have the following symbols:
+ * // const unsigned char gFooData[];
+ * // const unsigned char *const gFooEnd;
+ * // const unsigned int gFooSize;
+ * @endcode
+ *
+ * If however you specify a prefix before including: e.g:
+ * @code
+ * #define INCBIN_PREFIX incbin
+ * #include "incbin.h"
+ * INCBIN(Foo, "foo.txt");
+ *
+ * // Now you have the following symbols instead:
+ * // const unsigned char incbinFooData[];
+ * // const unsigned char *const incbinFooEnd;
+ * // const unsigned int incbinFooSize;
+ * @endcode
+ */
+#if !defined(INCBIN_PREFIX)
+#  define INCBIN_PREFIX g
+#endif
+
+/**
+ * @brief Specify the style used for symbol names.
+ *
+ * Possible options are
+ * - INCBIN_STYLE_CAMEL "CamelCase"
+ * - INCBIN_STYLE_SNAKE "snake_case"
+ *
+ * Default option is *INCBIN_STYLE_CAMEL* producing symbols of the form:
+ * @code
+ * #include "incbin.h"
+ * INCBIN(Foo, "foo.txt");
+ *
+ * // Now you have the following symbols:
+ * // const unsigned char <prefix>FooData[];
+ * // const unsigned char *const <prefix>FooEnd;
+ * // const unsigned int <prefix>FooSize;
+ * @endcode
+ *
+ * If however you specify a style before including: e.g:
+ * @code
+ * #define INCBIN_STYLE INCBIN_STYLE_SNAKE
+ * #include "incbin.h"
+ * INCBIN(foo, "foo.txt");
+ *
+ * // Now you have the following symbols:
+ * // const unsigned char <prefix>foo_data[];
+ * // const unsigned char *const <prefix>foo_end;
+ * // const unsigned int <prefix>foo_size;
+ * @endcode
+ */
+#if !defined(INCBIN_STYLE)
+#  define INCBIN_STYLE INCBIN_STYLE_CAMEL
+#endif
+
+/* Style lookup tables */
+#define INCBIN_STYLE_0_DATA Data
+#define INCBIN_STYLE_0_END End
+#define INCBIN_STYLE_0_SIZE Size
+#define INCBIN_STYLE_1_DATA _data
+#define INCBIN_STYLE_1_END _end
+#define INCBIN_STYLE_1_SIZE _size
+
+/* Style lookup: returning identifier */
+#define INCBIN_STYLE_IDENT(TYPE) \
+    INCBIN_CONCATENATE( \
+        INCBIN_STYLE_, \
+        INCBIN_CONCATENATE( \
+            INCBIN_EVAL(INCBIN_STYLE), \
+            INCBIN_CONCATENATE(_, TYPE)))
+
+/* Style lookup: returning string literal */
+#define INCBIN_STYLE_STRING(TYPE) \
+    INCBIN_STRINGIZE( \
+        INCBIN_STYLE_IDENT(TYPE)) \
+
+/* Generate the global labels by indirectly invoking the macro with our style
+ * type and concatenating the name against them. */
+#define INCBIN_GLOBAL_LABELS(NAME, TYPE) \
+    INCBIN_INVOKE( \
+        INCBIN_GLOBAL, \
+        INCBIN_CONCATENATE( \
+            NAME, \
+            INCBIN_INVOKE( \
+                INCBIN_STYLE_IDENT, \
+                TYPE))) \
+    INCBIN_INVOKE( \
+        INCBIN_TYPE, \
+        INCBIN_CONCATENATE( \
+            NAME, \
+            INCBIN_INVOKE( \
+                INCBIN_STYLE_IDENT, \
+                TYPE)))
+
+/**
+ * @brief Externally reference binary data included in another translation unit.
+ *
+ * Produces three external symbols that reference the binary data included in
+ * another translation unit.
+ *
+ * The symbol names are a concatenation of `INCBIN_PREFIX' before *NAME*; with
+ * "Data", as well as "End" and "Size" after. An example is provided below.
+ *
+ * @param NAME The name given for the binary data
+ *
+ * @code
+ * INCBIN_EXTERN(Foo);
+ *
+ * // Now you have the following symbols:
+ * // extern const unsigned char <prefix>FooData[];
+ * // extern const unsigned char *const <prefix>FooEnd;
+ * // extern const unsigned int <prefix>FooSize;
+ * @endcode
+ */
+#define INCBIN_EXTERN(NAME) \
+    INCBIN_EXTERNAL const INCBIN_ALIGN unsigned char \
+        INCBIN_CONCATENATE( \
+            INCBIN_CONCATENATE(INCBIN_PREFIX, NAME), \
+            INCBIN_STYLE_IDENT(DATA))[]; \
+    INCBIN_EXTERNAL const INCBIN_ALIGN unsigned char *const \
+    INCBIN_CONCATENATE( \
+        INCBIN_CONCATENATE(INCBIN_PREFIX, NAME), \
+        INCBIN_STYLE_IDENT(END)); \
+    INCBIN_EXTERNAL const unsigned int \
+        INCBIN_CONCATENATE( \
+            INCBIN_CONCATENATE(INCBIN_PREFIX, NAME), \
+            INCBIN_STYLE_IDENT(SIZE))
+
+/**
+ * @brief Include a binary file into the current translation unit.
+ *
+ * Includes a binary file into the current translation unit, producing three symbols
+ * for objects that encode the data and size respectively.
+ *
+ * The symbol names are a concatenation of `INCBIN_PREFIX' before *NAME*; with
+ * "Data", as well as "End" and "Size" after. An example is provided below.
+ *
+ * @param NAME The name to associate with this binary data (as an identifier.)
+ * @param FILENAME The file to include (as a string literal.)
+ *
+ * @code
+ * INCBIN(Icon, "icon.png");
+ *
+ * // Now you have the following symbols:
+ * // const unsigned char <prefix>IconData[];
+ * // const unsigned char *const <prefix>IconEnd;
+ * // const unsigned int <prefix>IconSize;
+ * @endcode
+ *
+ * @warning This must be used in global scope
+ * @warning The identifiers may be different if INCBIN_STYLE is not default
+ *
+ * To externally reference the data included by this in another translation unit
+ * please @see INCBIN_EXTERN.
+ */
+#ifdef _MSC_VER
+#define INCBIN(NAME, FILENAME) \
+    INCBIN_EXTERN(NAME)
+#else
+#define INCBIN(NAME, FILENAME) \
+    __asm__(INCBIN_SECTION \
+            INCBIN_GLOBAL_LABELS(NAME, DATA) \
+            INCBIN_ALIGN_HOST \
+            INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(DATA) ":\n" \
+            INCBIN_MACRO " \"" FILENAME "\"\n" \
+            INCBIN_GLOBAL_LABELS(NAME, END) \
+            INCBIN_ALIGN_BYTE \
+            INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(END) ":\n" \
+                INCBIN_BYTE "1\n" \
+            INCBIN_GLOBAL_LABELS(NAME, SIZE) \
+            INCBIN_ALIGN_HOST \
+            INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(SIZE) ":\n" \
+                INCBIN_INT INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(END) " - " \
+                           INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(DATA) "\n" \
+            INCBIN_ALIGN_HOST \
+            ".text\n" \
+    ); \
+    INCBIN_EXTERN(NAME)
+
+#endif
+#endif

src/main.cpp

@@ -35,6 +35,7 @@ int main(int argc, char* argv[]) {
 
   std::cout << engine_info() << std::endl;
 
+  CommandLine::init(argc, argv);
   UCI::init(Options);
   Tune::init();
   PSQT::init();

src/material.cpp

@@ -130,7 +130,7 @@ Entry* probe(const Position& pos) {
 
   Value npm_w = pos.non_pawn_material(WHITE);
   Value npm_b = pos.non_pawn_material(BLACK);
-  Value npm   = Utility::clamp(npm_w + npm_b, EndgameLimit, MidgameLimit);
+  Value npm   = std::clamp(npm_w + npm_b, EndgameLimit, MidgameLimit);
 
   // Map total non-pawn material into [PHASE_ENDGAME, PHASE_MIDGAME]
   e->gamePhase = Phase(((npm - EndgameLimit) * PHASE_MIDGAME) / (MidgameLimit - EndgameLimit));

src/misc.cpp

@@ -51,6 +51,11 @@ typedef bool(*fun3_t)(HANDLE, CONST GROUP_AFFINITY*, PGROUP_AFFINITY);
 #include <sys/mman.h>
 #endif
 
+#if defined(__APPLE__) || defined(__ANDROID__) || defined(__OpenBSD__) || (defined(__GLIBCXX__) && !defined(_GLIBCXX_HAVE_ALIGNED_ALLOC) && !defined(_WIN32))
+#define POSIXALIGNEDALLOC
+#include <stdlib.h>
+#endif
+
 #include "misc.h"
 #include "thread.h"
 
@@ -60,7 +65,7 @@ namespace {
 
 /// Version number. If Version is left empty, then compile date in the format
 /// DD-MM-YY and show in engine_info.
-const string Version = "";
+const string Version = "12";
 
 /// Our fancy logging facility. The trick here is to replace cin.rdbuf() and
 /// cout.rdbuf() with two Tie objects that tie cin and cout to a file stream. We
@@ -127,6 +132,7 @@ public:
 
 } // namespace
 
+
 /// engine_info() returns the full name of the current Stockfish version. This
 /// will be either "Stockfish <Tag> DD-MM-YY" (where DD-MM-YY is the date when
 /// the program was compiled) or "Stockfish <Version>", depending on whether
@@ -214,26 +220,33 @@ const std::string compiler_info() {
 
   compiler += "\nCompilation settings include: ";
   compiler += (Is64Bit ? " 64bit" : " 32bit");
+  #if defined(USE_VNNI)
+    compiler += " VNNI";
+  #endif
   #if defined(USE_AVX512)
     compiler += " AVX512";
   #endif
+  compiler += (HasPext ? " BMI2" : "");
   #if defined(USE_AVX2)
     compiler += " AVX2";
   #endif
-  #if defined(USE_SSE42)
-    compiler += " SSE42";
-  #endif
   #if defined(USE_SSE41)
     compiler += " SSE41";
   #endif
   #if defined(USE_SSSE3)
     compiler += " SSSE3";
   #endif
-  #if defined(USE_SSE3)
-    compiler += " SSE3";
+  #if defined(USE_SSE2)
+    compiler += " SSE2";
   #endif
-    compiler += (HasPext ? " BMI2" : "");
-    compiler += (HasPopCnt ? " POPCNT" : "");
+  compiler += (HasPopCnt ? " POPCNT" : "");
+  #if defined(USE_MMX)
+    compiler += " MMX";
+  #endif
+  #if defined(USE_NEON)
+    compiler += " NEON";
+  #endif
+
   #if !defined(NDEBUG)
     compiler += " DEBUG";
   #endif
@@ -316,13 +329,16 @@ void prefetch(void* addr) {
 
 #endif
 
-/// Wrappers for systems where the c++17 implementation doesn't guarantee the availability of aligned_alloc.
-/// Memory allocated with std_aligned_alloc must be freed with std_aligned_free.
-///
+
+/// std_aligned_alloc() is our wrapper for systems where the c++17 implementation
+/// does not guarantee the availability of aligned_alloc(). Memory allocated with
+/// std_aligned_alloc() must be freed with std_aligned_free().
 
 void* std_aligned_alloc(size_t alignment, size_t size) {
-#if defined(__APPLE__)
-  return aligned_alloc(alignment, size);
+
+#if defined(POSIXALIGNEDALLOC)
+  void *mem;
+  return posix_memalign(&mem, alignment, size) ? nullptr : mem;
 #elif defined(_WIN32)
   return _mm_malloc(size, alignment);
 #else
@@ -331,7 +347,8 @@ void* std_aligned_alloc(size_t alignment, size_t size) {
 }
 
 void std_aligned_free(void* ptr) {
-#if defined(__APPLE__)
+
+#if defined(POSIXALIGNEDALLOC)
   free(ptr);
 #elif defined(_WIN32)
   _mm_free(ptr);
@@ -340,7 +357,7 @@ void std_aligned_free(void* ptr) {
 #endif
 }
 
-/// aligned_ttmem_alloc() will return suitably aligned memory, and if possible use large pages.
+/// aligned_ttmem_alloc() will return suitably aligned memory, if possible using large pages.
 /// The returned pointer is the aligned one, while the mem argument is the one that needs
 /// to be passed to free. With c++17 some of this functionality could be simplified.
 
@@ -352,7 +369,9 @@ void* aligned_ttmem_alloc(size_t allocSize, void*& mem) {
   size_t size = ((allocSize + alignment - 1) / alignment) * alignment; // multiple of alignment
   if (posix_memalign(&mem, alignment, size))
      mem = nullptr;
+#if defined(MADV_HUGEPAGE)
   madvise(mem, allocSize, MADV_HUGEPAGE);
+#endif
   return mem;
 }
 
@@ -571,3 +590,61 @@ void bindThisThread(size_t idx) {
 #endif
 
 } // namespace WinProcGroup
+
+#ifdef _WIN32
+#include <direct.h>
+#define GETCWD _getcwd
+#else
+#include <unistd.h>
+#define GETCWD getcwd
+#endif
+
+namespace CommandLine {
+
+string argv0;            // path+name of the executable binary, as given by argv[0]
+string binaryDirectory;  // path of the executable directory
+string workingDirectory; // path of the working directory
+string pathSeparator;    // Separator for our current OS
+
+void init(int argc, char* argv[]) {
+    (void)argc;
+    string separator;
+
+    // extract the path+name of the executable binary
+    argv0 = argv[0];
+
+#ifdef _WIN32
+    pathSeparator = "\\";
+  #ifdef _MSC_VER
+    // Under windows argv[0] may not have the extension. Also _get_pgmptr() had
+    // issues in some windows 10 versions, so check returned values carefully.
+    char* pgmptr = nullptr;
+    if (!_get_pgmptr(&pgmptr) && pgmptr != nullptr && *pgmptr)
+        argv0 = pgmptr;
+  #endif
+#else
+    pathSeparator = "/";
+#endif
+
+    // extract the working directory
+    workingDirectory = "";
+    char buff[40000];
+    char* cwd = GETCWD(buff, 40000);
+    if (cwd)
+        workingDirectory = cwd;
+
+    // extract the binary directory path from argv0
+    binaryDirectory = argv0;
+    size_t pos = binaryDirectory.find_last_of("\\/");
+    if (pos == std::string::npos)
+        binaryDirectory = "." + pathSeparator;
+    else
+        binaryDirectory.resize(pos + 1);
+
+    // pattern replacement: "./" at the start of path is replaced by the working directory
+    if (binaryDirectory.find("." + pathSeparator) == 0)
+        binaryDirectory.replace(0, 1, workingDirectory);
+}
+
+
+} // namespace CommandLine

src/misc.h

@@ -42,9 +42,7 @@ void dbg_mean_of(int v);
 void dbg_print();
 
 typedef std::chrono::milliseconds::rep TimePoint; // A value in milliseconds
-
 static_assert(sizeof(TimePoint) == sizeof(int64_t), "TimePoint should be 64 bits");
-
 inline TimePoint now() {
   return std::chrono::duration_cast<std::chrono::milliseconds>
         (std::chrono::steady_clock::now().time_since_epoch()).count();
@@ -65,14 +63,6 @@ std::ostream& operator<<(std::ostream&, SyncCout);
 #define sync_cout std::cout << IO_LOCK
 #define sync_endl std::endl << IO_UNLOCK
 
-namespace Utility {
-
-/// Clamp a value between lo and hi. Available in c++17.
-template<class T> constexpr const T& clamp(const T& v, const T& lo, const T& hi) {
-  return v < lo ? lo : v > hi ? hi : v;
-}
-
-}
 
 /// xorshift64star Pseudo-Random Number Generator
 /// This class is based on original code written and dedicated
@@ -134,4 +124,11 @@ namespace WinProcGroup {
   void bindThisThread(size_t idx);
 }
 
+namespace CommandLine {
+  void init(int argc, char* argv[]);
+
+  extern std::string binaryDirectory;  // path of the executable directory
+  extern std::string workingDirectory; // path of the working directory
+}
+
 #endif // #ifndef MISC_H_INCLUDED

src/movegen.cpp

@@ -248,7 +248,7 @@ namespace {
             *moveList++ = make_move(ksq, pop_lsb(&b));
 
         if ((Type != CAPTURES) && pos.can_castle(Us & ANY_CASTLING))
-            for(CastlingRights cr : { Us & KING_SIDE, Us & QUEEN_SIDE } )
+            for (CastlingRights cr : { Us & KING_SIDE, Us & QUEEN_SIDE } )
                 if (!pos.castling_impeded(cr) && pos.can_castle(cr))
                     *moveList++ = make<CASTLING>(ksq, pos.castling_rook_square(cr));
     }

src/movepick.cpp

@@ -182,7 +182,7 @@ top:
           --endMoves;
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case REFUTATION:
       if (select<Next>([&](){ return    *cur != MOVE_NONE
@@ -190,7 +190,7 @@ top:
                                     &&  pos.pseudo_legal(*cur); }))
           return *(cur - 1);
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case QUIET_INIT:
       if (!skipQuiets)
@@ -203,7 +203,7 @@ top:
       }
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case QUIET:
       if (   !skipQuiets
@@ -217,7 +217,7 @@ top:
       endMoves = endBadCaptures;
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case BAD_CAPTURE:
       return select<Next>([](){ return true; });
@@ -228,7 +228,7 @@ top:
 
       score<EVASIONS>();
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case EVASION:
       return select<Best>([](){ return true; });
@@ -246,14 +246,14 @@ top:
           return MOVE_NONE;
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case QCHECK_INIT:
       cur = moves;
       endMoves = generate<QUIET_CHECKS>(pos, cur);
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case QCHECK:
       return select<Next>([](){ return true; });

src/movepick.h

@@ -86,9 +86,9 @@ enum StatsType { NoCaptures, Captures };
 /// the move's from and to squares, see www.chessprogramming.org/Butterfly_Boards
 typedef Stats<int16_t, 10692, COLOR_NB, int(SQUARE_NB) * int(SQUARE_NB)> ButterflyHistory;
 
-/// At higher depths LowPlyHistory records successful quiet moves near the root and quiet
-/// moves which are/were in the PV (ttPv)
-/// It is cleared with each new search and filled during iterative deepening
+/// At higher depths LowPlyHistory records successful quiet moves near the root
+/// and quiet moves which are/were in the PV (ttPv). It is cleared with each new
+/// search and filled during iterative deepening.
 constexpr int MAX_LPH = 4;
 typedef Stats<int16_t, 10692, MAX_LPH, int(SQUARE_NB) * int(SQUARE_NB)> LowPlyHistory;
 

src/nnue/evaluate_nnue.cpp

@@ -18,7 +18,6 @@
 
 // Code for calculating NNUE evaluation function
 
-#include <fstream>
 #include <iostream>
 #include <set>
 
@@ -29,30 +28,29 @@
 
 #include "evaluate_nnue.h"
 
-ExtPieceSquare kpp_board_index[PIECE_NB] = {
- // convention: W - us, B - them
- // viewed from other side, W and B are reversed
-    { PS_NONE,     PS_NONE     },
-    { PS_W_PAWN,   PS_B_PAWN   },
-    { PS_W_KNIGHT, PS_B_KNIGHT },
-    { PS_W_BISHOP, PS_B_BISHOP },
-    { PS_W_ROOK,   PS_B_ROOK   },
-    { PS_W_QUEEN,  PS_B_QUEEN  },
-    { PS_W_KING,   PS_B_KING   },
-    { PS_NONE,     PS_NONE     },
-    { PS_NONE,     PS_NONE     },
-    { PS_B_PAWN,   PS_W_PAWN   },
-    { PS_B_KNIGHT, PS_W_KNIGHT },
-    { PS_B_BISHOP, PS_W_BISHOP },
-    { PS_B_ROOK,   PS_W_ROOK   },
-    { PS_B_QUEEN,  PS_W_QUEEN  },
-    { PS_B_KING,   PS_W_KING   },
-    { PS_NONE,     PS_NONE     }
-};
-
-
 namespace Eval::NNUE {
 
+  uint32_t kpp_board_index[PIECE_NB][COLOR_NB] = {
+   // convention: W - us, B - them
+   // viewed from other side, W and B are reversed
+      { PS_NONE,     PS_NONE     },
+      { PS_W_PAWN,   PS_B_PAWN   },
+      { PS_W_KNIGHT, PS_B_KNIGHT },
+      { PS_W_BISHOP, PS_B_BISHOP },
+      { PS_W_ROOK,   PS_B_ROOK   },
+      { PS_W_QUEEN,  PS_B_QUEEN  },
+      { PS_W_KING,   PS_B_KING   },
+      { PS_NONE,     PS_NONE     },
+      { PS_NONE,     PS_NONE     },
+      { PS_B_PAWN,   PS_W_PAWN   },
+      { PS_B_KNIGHT, PS_W_KNIGHT },
+      { PS_B_BISHOP, PS_W_BISHOP },
+      { PS_B_ROOK,   PS_W_ROOK   },
+      { PS_B_QUEEN,  PS_W_QUEEN  },
+      { PS_B_KING,   PS_W_KING   },
+      { PS_NONE,     PS_NONE     }
+  };
+
   // Input feature converter
   AlignedPtr<FeatureTransformer> feature_transformer;
 
@@ -77,7 +75,7 @@ namespace Eval::NNUE {
   bool ReadParameters(std::istream& stream, const AlignedPtr<T>& pointer) {
 
     std::uint32_t header;
-    stream.read(reinterpret_cast<char*>(&header), sizeof(header));
+    header = read_little_endian<std::uint32_t>(stream);
     if (!stream || header != T::GetHashValue()) return false;
     return pointer->ReadParameters(stream);
   }
@@ -92,13 +90,13 @@ namespace Eval::NNUE {
   }
 
   // Read network header
-  bool ReadHeader(std::istream& stream,
-    std::uint32_t* hash_value, std::string* architecture) {
-
+  bool ReadHeader(std::istream& stream, std::uint32_t* hash_value, std::string* architecture)
+  {
     std::uint32_t version, size;
-    stream.read(reinterpret_cast<char*>(&version), sizeof(version));
-    stream.read(reinterpret_cast<char*>(hash_value), sizeof(*hash_value));
-    stream.read(reinterpret_cast<char*>(&size), sizeof(size));
+
+    version     = read_little_endian<std::uint32_t>(stream);
+    *hash_value = read_little_endian<std::uint32_t>(stream);
+    size        = read_little_endian<std::uint32_t>(stream);
     if (!stream || version != kVersion) return false;
     architecture->resize(size);
     stream.read(&(*architecture)[0], size);
@@ -144,25 +142,17 @@ namespace Eval::NNUE {
     return accumulator.score;
   }
 
-  // Load the evaluation function file
-  bool load_eval_file(const std::string& evalFile) {
+  // Load eval, from a file stream or a memory stream
+  bool load_eval(std::string streamName, std::istream& stream) {
 
     Initialize();
-    fileName = evalFile;
-
-    std::ifstream stream(evalFile, std::ios::binary);
-
-    const bool result = ReadParameters(stream);
-
-    return result;
+    fileName = streamName;
+    return ReadParameters(stream);
   }
 
   // Evaluation function. Perform differential calculation.
   Value evaluate(const Position& pos) {
-    Value v = ComputeScore(pos, false);
-    v = Utility::clamp(v, VALUE_TB_LOSS_IN_MAX_PLY + 1, VALUE_TB_WIN_IN_MAX_PLY - 1);
-
-    return v;
+    return ComputeScore(pos, false);
   }
 
   // Evaluation function. Perform full calculation.

src/nnue/features/feature_set.h

@@ -68,8 +68,7 @@ namespace Eval::NNUE::Features {
         reset[perspective] = false;
         switch (trigger) {
           case TriggerEvent::kFriendKingMoved:
-            reset[perspective] =
-                dp.pieceId[0] == PIECE_ID_KING + perspective;
+            reset[perspective] = dp.piece[0] == make_piece(perspective, KING);
             break;
           default:
             assert(false);

src/nnue/features/half_kp.cpp

@@ -23,25 +23,17 @@
 
 namespace Eval::NNUE::Features {
 
-  // Find the index of the feature quantity from the king position and PieceSquare
-  template <Side AssociatedKing>
-  inline IndexType HalfKP<AssociatedKing>::MakeIndex(Square sq_k, PieceSquare p) {
-    return static_cast<IndexType>(PS_END) * static_cast<IndexType>(sq_k) + p;
+  // Orient a square according to perspective (rotates by 180 for black)
+  inline Square orient(Color perspective, Square s) {
+    return Square(int(s) ^ (bool(perspective) * 63));
   }
 
-  // Get pieces information
+  // Find the index of the feature quantity from the king position and PieceSquare
   template <Side AssociatedKing>
-  inline void HalfKP<AssociatedKing>::GetPieces(
-      const Position& pos, Color perspective,
-      PieceSquare** pieces, Square* sq_target_k) {
+  inline IndexType HalfKP<AssociatedKing>::MakeIndex(
+      Color perspective, Square s, Piece pc, Square ksq) {
 
-    *pieces = (perspective == BLACK) ?
-        pos.eval_list()->piece_list_fb() :
-        pos.eval_list()->piece_list_fw();
-    const PieceId target = (AssociatedKing == Side::kFriend) ?
-        static_cast<PieceId>(PIECE_ID_KING + perspective) :
-        static_cast<PieceId>(PIECE_ID_KING + ~perspective);
-    *sq_target_k = static_cast<Square>(((*pieces)[target] - PS_W_KING) % SQUARE_NB);
+    return IndexType(orient(perspective, s) + kpp_board_index[pc][perspective] + PS_END * ksq);
   }
 
   // Get a list of indices for active features
@@ -49,16 +41,11 @@ namespace Eval::NNUE::Features {
   void HalfKP<AssociatedKing>::AppendActiveIndices(
       const Position& pos, Color perspective, IndexList* active) {
 
-    // Do nothing if array size is small to avoid compiler warning
-    if (RawFeatures::kMaxActiveDimensions < kMaxActiveDimensions) return;
-
-    PieceSquare* pieces;
-    Square sq_target_k;
-    GetPieces(pos, perspective, &pieces, &sq_target_k);
-    for (PieceId i = PIECE_ID_ZERO; i < PIECE_ID_KING; ++i) {
-      if (pieces[i] != PS_NONE) {
-        active->push_back(MakeIndex(sq_target_k, pieces[i]));
-      }
+    Square ksq = orient(perspective, pos.square<KING>(perspective));
+    Bitboard bb = pos.pieces() & ~pos.pieces(KING);
+    while (bb) {
+      Square s = pop_lsb(&bb);
+      active->push_back(MakeIndex(perspective, s, pos.piece_on(s), ksq));
     }
   }
 
@@ -68,22 +55,15 @@ namespace Eval::NNUE::Features {
       const Position& pos, Color perspective,
       IndexList* removed, IndexList* added) {
 
-    PieceSquare* pieces;
-    Square sq_target_k;
-    GetPieces(pos, perspective, &pieces, &sq_target_k);
+    Square ksq = orient(perspective, pos.square<KING>(perspective));
     const auto& dp = pos.state()->dirtyPiece;
     for (int i = 0; i < dp.dirty_num; ++i) {
-      if (dp.pieceId[i] >= PIECE_ID_KING) continue;
-      const auto old_p = static_cast<PieceSquare>(
-          dp.old_piece[i].from[perspective]);
-      if (old_p != PS_NONE) {
-        removed->push_back(MakeIndex(sq_target_k, old_p));
-      }
-      const auto new_p = static_cast<PieceSquare>(
-          dp.new_piece[i].from[perspective]);
-      if (new_p != PS_NONE) {
-        added->push_back(MakeIndex(sq_target_k, new_p));
-      }
+      Piece pc = dp.piece[i];
+      if (type_of(pc) == KING) continue;
+      if (dp.from[i] != SQ_NONE)
+        removed->push_back(MakeIndex(perspective, dp.from[i], pc, ksq));
+      if (dp.to[i] != SQ_NONE)
+        added->push_back(MakeIndex(perspective, dp.to[i], pc, ksq));
     }
   }
 

src/nnue/features/half_kp.h

@@ -41,7 +41,7 @@ namespace Eval::NNUE::Features {
     static constexpr IndexType kDimensions =
         static_cast<IndexType>(SQUARE_NB) * static_cast<IndexType>(PS_END);
     // Maximum number of simultaneously active features
-    static constexpr IndexType kMaxActiveDimensions = PIECE_ID_KING;
+    static constexpr IndexType kMaxActiveDimensions = 30; // Kings don't count
     // Trigger for full calculation instead of difference calculation
     static constexpr TriggerEvent kRefreshTrigger = TriggerEvent::kFriendKingMoved;
 
@@ -53,13 +53,9 @@ namespace Eval::NNUE::Features {
     static void AppendChangedIndices(const Position& pos, Color perspective,
                                      IndexList* removed, IndexList* added);
 
-    // Index of a feature for a given king position and another piece on some square
-    static IndexType MakeIndex(Square sq_k, PieceSquare p);
-
    private:
-    // Get pieces information
-    static void GetPieces(const Position& pos, Color perspective,
-                          PieceSquare** pieces, Square* sq_target_k);
+    // Index of a feature for a given king position and another piece on some square
+    static IndexType MakeIndex(Color perspective, Square s, Piece pc, Square sq_k);
   };
 
 }  // namespace Eval::NNUE::Features

src/nnue/layers/affine_transform.h

@@ -62,11 +62,10 @@ namespace Eval::NNUE::Layers {
    // Read network parameters
     bool ReadParameters(std::istream& stream) {
       if (!previous_layer_.ReadParameters(stream)) return false;
-      stream.read(reinterpret_cast<char*>(biases_),
-                  kOutputDimensions * sizeof(BiasType));
-      stream.read(reinterpret_cast<char*>(weights_),
-                  kOutputDimensions * kPaddedInputDimensions *
-                  sizeof(WeightType));
+      for (std::size_t i = 0; i < kOutputDimensions; ++i)
+        biases_[i] = read_little_endian<BiasType>(stream);
+      for (std::size_t i = 0; i < kOutputDimensions * kPaddedInputDimensions; ++i)
+        weights_[i] = read_little_endian<WeightType>(stream);
       return !stream.fail();
     }
 
@@ -79,19 +78,32 @@ namespace Eval::NNUE::Layers {
 
   #if defined(USE_AVX512)
       constexpr IndexType kNumChunks = kPaddedInputDimensions / (kSimdWidth * 2);
-      const __m512i kOnes = _mm512_set1_epi16(1);
       const auto input_vector = reinterpret_cast<const __m512i*>(input);
+  #if !defined(USE_VNNI)
+      const __m512i kOnes = _mm512_set1_epi16(1);
+  #endif
 
   #elif defined(USE_AVX2)
       constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
-      const __m256i kOnes = _mm256_set1_epi16(1);
       const auto input_vector = reinterpret_cast<const __m256i*>(input);
+  #if !defined(USE_VNNI)
+      const __m256i kOnes = _mm256_set1_epi16(1);
+  #endif
 
-  #elif defined(USE_SSSE3)
+  #elif defined(USE_SSE2)
       constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
+  #ifndef USE_SSSE3
+      const __m128i kZeros = _mm_setzero_si128();
+  #else
       const __m128i kOnes = _mm_set1_epi16(1);
+  #endif
       const auto input_vector = reinterpret_cast<const __m128i*>(input);
 
+  #elif defined(USE_MMX)
+      constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
+      const __m64 kZeros = _mm_setzero_si64();
+      const auto input_vector = reinterpret_cast<const __m64*>(input);
+
   #elif defined(USE_NEON)
       constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
       const auto input_vector = reinterpret_cast<const int8x8_t*>(input);
@@ -104,79 +116,115 @@ namespace Eval::NNUE::Layers {
         __m512i sum = _mm512_setzero_si512();
         const auto row = reinterpret_cast<const __m512i*>(&weights_[offset]);
         for (IndexType j = 0; j < kNumChunks; ++j) {
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-            __m512i product = _mm512_maddubs_epi16(_mm512_loadu_si512(&input_vector[j]), _mm512_load_si512(&row[j]));
+  #if defined(USE_VNNI)
+            sum = _mm512_dpbusd_epi32(sum, _mm512_loadA_si512(&input_vector[j]), _mm512_load_si512(&row[j]));
   #else
-            __m512i product = _mm512_maddubs_epi16(_mm512_load_si512(&input_vector[j]), _mm512_load_si512(&row[j]));
-  #endif
-
+            __m512i product = _mm512_maddubs_epi16(_mm512_loadA_si512(&input_vector[j]), _mm512_load_si512(&row[j]));
             product = _mm512_madd_epi16(product, kOnes);
             sum = _mm512_add_epi32(sum, product);
+  #endif
         }
-        output[i] = _mm512_reduce_add_epi32(sum) + biases_[i];
 
         // Note: Changing kMaxSimdWidth from 32 to 64 breaks loading existing networks.
         // As a result kPaddedInputDimensions may not be an even multiple of 64(512bit)
         // and we have to do one more 256bit chunk.
         if (kPaddedInputDimensions != kNumChunks * kSimdWidth * 2)
         {
-            const auto iv_256  = reinterpret_cast<const __m256i*>(input);
-            const auto row_256 = reinterpret_cast<const __m256i*>(&weights_[offset]);
-            int j = kNumChunks * 2;
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)  // See HACK comment below in AVX2.
-            __m256i sum256 = _mm256_maddubs_epi16(_mm256_loadu_si256(&iv_256[j]), _mm256_load_si256(&row_256[j]));
+            const auto iv256  = reinterpret_cast<const __m256i*>(&input_vector[kNumChunks]);
+            const auto row256 = reinterpret_cast<const __m256i*>(&row[kNumChunks]);
+  #if defined(USE_VNNI)
+            __m256i product256 = _mm256_dpbusd_epi32(
+                _mm512_castsi512_si256(sum), _mm256_loadA_si256(&iv256[0]), _mm256_load_si256(&row256[0]));
+            sum = _mm512_inserti32x8(sum, product256, 0);
   #else
-            __m256i sum256 = _mm256_maddubs_epi16(_mm256_load_si256(&iv_256[j]), _mm256_load_si256(&row_256[j]));
+            __m256i product256 = _mm256_maddubs_epi16(_mm256_loadA_si256(&iv256[0]), _mm256_load_si256(&row256[0]));
+            sum = _mm512_add_epi32(sum, _mm512_cvtepi16_epi32(product256));
   #endif
-
-            sum256 = _mm256_madd_epi16(sum256, _mm256_set1_epi16(1));
-            sum256 = _mm256_hadd_epi32(sum256, sum256);
-            sum256 = _mm256_hadd_epi32(sum256, sum256);
-            const __m128i lo = _mm256_extracti128_si256(sum256, 0);
-            const __m128i hi = _mm256_extracti128_si256(sum256, 1);
-            output[i] += _mm_cvtsi128_si32(lo) + _mm_cvtsi128_si32(hi);
         }
+        output[i] = _mm512_reduce_add_epi32(sum) + biases_[i];
 
   #elif defined(USE_AVX2)
         __m256i sum = _mm256_setzero_si256();
         const auto row = reinterpret_cast<const __m256i*>(&weights_[offset]);
         for (IndexType j = 0; j < kNumChunks; ++j) {
-          __m256i product = _mm256_maddubs_epi16(
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-            // HACK: Use _mm256_loadu_si256() instead of _mm256_load_si256. Because the binary
-            //       compiled with g++ in MSYS2 crashes here because the output memory is not aligned
-            //       even though alignas is specified.
-            _mm256_loadu_si256
+  #if defined(USE_VNNI)
+          sum = _mm256_dpbusd_epi32(sum, _mm256_loadA_si256(&input_vector[j]), _mm256_load_si256(&row[j]));
   #else
-            _mm256_load_si256
-  #endif
-
-            (&input_vector[j]), _mm256_load_si256(&row[j]));
+          __m256i product = _mm256_maddubs_epi16(_mm256_loadA_si256(&input_vector[j]), _mm256_load_si256(&row[j]));
           product = _mm256_madd_epi16(product, kOnes);
           sum = _mm256_add_epi32(sum, product);
+  #endif
         }
-        sum = _mm256_hadd_epi32(sum, sum);
-        sum = _mm256_hadd_epi32(sum, sum);
-        const __m128i lo = _mm256_extracti128_si256(sum, 0);
-        const __m128i hi = _mm256_extracti128_si256(sum, 1);
-        output[i] = _mm_cvtsi128_si32(lo) + _mm_cvtsi128_si32(hi) + biases_[i];
+        __m128i sum128 = _mm_add_epi32(_mm256_castsi256_si128(sum), _mm256_extracti128_si256(sum, 1));
+        sum128 = _mm_add_epi32(sum128, _mm_shuffle_epi32(sum128, _MM_PERM_BADC));
+        sum128 = _mm_add_epi32(sum128, _mm_shuffle_epi32(sum128, _MM_PERM_CDAB));
+        output[i] = _mm_cvtsi128_si32(sum128) + biases_[i];
 
   #elif defined(USE_SSSE3)
-        __m128i sum = _mm_cvtsi32_si128(biases_[i]);
+        __m128i sum = _mm_setzero_si128();
         const auto row = reinterpret_cast<const __m128i*>(&weights_[offset]);
-        for (IndexType j = 0; j < kNumChunks; ++j) {
-          __m128i product = _mm_maddubs_epi16(
-              _mm_load_si128(&input_vector[j]), _mm_load_si128(&row[j]));
+        for (int j = 0; j < (int)kNumChunks - 1; j += 2) {
+          __m128i product0 = _mm_maddubs_epi16(_mm_load_si128(&input_vector[j]), _mm_load_si128(&row[j]));
+          product0 = _mm_madd_epi16(product0, kOnes);
+          sum = _mm_add_epi32(sum, product0);
+          __m128i product1 = _mm_maddubs_epi16(_mm_load_si128(&input_vector[j+1]), _mm_load_si128(&row[j+1]));
+          product1 = _mm_madd_epi16(product1, kOnes);
+          sum = _mm_add_epi32(sum, product1);
+        }
+        if (kNumChunks & 0x1) {
+          __m128i product = _mm_maddubs_epi16(_mm_load_si128(&input_vector[kNumChunks-1]), _mm_load_si128(&row[kNumChunks-1]));
           product = _mm_madd_epi16(product, kOnes);
           sum = _mm_add_epi32(sum, product);
         }
-        sum = _mm_hadd_epi32(sum, sum);
-        sum = _mm_hadd_epi32(sum, sum);
+        sum = _mm_add_epi32(sum, _mm_shuffle_epi32(sum, 0x4E)); //_MM_PERM_BADC
+        sum = _mm_add_epi32(sum, _mm_shuffle_epi32(sum, 0xB1)); //_MM_PERM_CDAB
+        output[i] = _mm_cvtsi128_si32(sum) + biases_[i];
+
+  #elif defined(USE_SSE2)
+        __m128i sum_lo = _mm_cvtsi32_si128(biases_[i]);
+        __m128i sum_hi = kZeros;
+        const auto row = reinterpret_cast<const __m128i*>(&weights_[offset]);
+        for (IndexType j = 0; j < kNumChunks; ++j) {
+          __m128i row_j = _mm_load_si128(&row[j]);
+          __m128i input_j = _mm_load_si128(&input_vector[j]);
+          __m128i row_signs = _mm_cmpgt_epi8(kZeros, row_j);
+          __m128i extended_row_lo = _mm_unpacklo_epi8(row_j, row_signs);
+          __m128i extended_row_hi = _mm_unpackhi_epi8(row_j, row_signs);
+          __m128i extended_input_lo = _mm_unpacklo_epi8(input_j, kZeros);
+          __m128i extended_input_hi = _mm_unpackhi_epi8(input_j, kZeros);
+          __m128i product_lo = _mm_madd_epi16(extended_row_lo, extended_input_lo);
+          __m128i product_hi = _mm_madd_epi16(extended_row_hi, extended_input_hi);
+          sum_lo = _mm_add_epi32(sum_lo, product_lo);
+          sum_hi = _mm_add_epi32(sum_hi, product_hi);
+        }
+        __m128i sum = _mm_add_epi32(sum_lo, sum_hi);
+        __m128i sum_high_64 = _mm_shuffle_epi32(sum, _MM_SHUFFLE(1, 0, 3, 2));
+        sum = _mm_add_epi32(sum, sum_high_64);
+        __m128i sum_second_32 = _mm_shufflelo_epi16(sum, _MM_SHUFFLE(1, 0, 3, 2));
+        sum = _mm_add_epi32(sum, sum_second_32);
         output[i] = _mm_cvtsi128_si32(sum);
 
+  #elif defined(USE_MMX)
+        __m64 sum_lo = _mm_cvtsi32_si64(biases_[i]);
+        __m64 sum_hi = kZeros;
+        const auto row = reinterpret_cast<const __m64*>(&weights_[offset]);
+        for (IndexType j = 0; j < kNumChunks; ++j) {
+          __m64 row_j = row[j];
+          __m64 input_j = input_vector[j];
+          __m64 row_signs = _mm_cmpgt_pi8(kZeros, row_j);
+          __m64 extended_row_lo = _mm_unpacklo_pi8(row_j, row_signs);
+          __m64 extended_row_hi = _mm_unpackhi_pi8(row_j, row_signs);
+          __m64 extended_input_lo = _mm_unpacklo_pi8(input_j, kZeros);
+          __m64 extended_input_hi = _mm_unpackhi_pi8(input_j, kZeros);
+          __m64 product_lo = _mm_madd_pi16(extended_row_lo, extended_input_lo);
+          __m64 product_hi = _mm_madd_pi16(extended_row_hi, extended_input_hi);
+          sum_lo = _mm_add_pi32(sum_lo, product_lo);
+          sum_hi = _mm_add_pi32(sum_hi, product_hi);
+        }
+        __m64 sum = _mm_add_pi32(sum_lo, sum_hi);
+        sum = _mm_add_pi32(sum, _mm_unpackhi_pi32(sum, sum));
+        output[i] = _mm_cvtsi64_si32(sum);
+
   #elif defined(USE_NEON)
         int32x4_t sum = {biases_[i]};
         const auto row = reinterpret_cast<const int8x8_t*>(&weights_[offset]);
@@ -196,6 +244,9 @@ namespace Eval::NNUE::Layers {
   #endif
 
       }
+  #if defined(USE_MMX)
+      _mm_empty();
+  #endif
       return output;
     }
 

src/nnue/layers/clipped_relu.h

@@ -74,55 +74,17 @@ namespace Eval::NNUE::Layers {
       const auto out = reinterpret_cast<__m256i*>(output);
       for (IndexType i = 0; i < kNumChunks; ++i) {
         const __m256i words0 = _mm256_srai_epi16(_mm256_packs_epi32(
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-          // HACK: Use _mm256_loadu_si256() instead of _mm256_load_si256. Because the binary
-          //       compiled with g++ in MSYS2 crashes here because the output memory is not aligned
-          //       even though alignas is specified.
-          _mm256_loadu_si256
-  #else
-          _mm256_load_si256
-  #endif
-
-          (&in[i * 4 + 0]),
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-          _mm256_loadu_si256
-  #else
-          _mm256_load_si256
-  #endif
-
-          (&in[i * 4 + 1])), kWeightScaleBits);
+            _mm256_loadA_si256(&in[i * 4 + 0]),
+            _mm256_loadA_si256(&in[i * 4 + 1])), kWeightScaleBits);
         const __m256i words1 = _mm256_srai_epi16(_mm256_packs_epi32(
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-          _mm256_loadu_si256
-  #else
-          _mm256_load_si256
-  #endif
-
-          (&in[i * 4 + 2]),
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-          _mm256_loadu_si256
-  #else
-          _mm256_load_si256
-  #endif
-
-          (&in[i * 4 + 3])), kWeightScaleBits);
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-        _mm256_storeu_si256
-  #else
-        _mm256_store_si256
-  #endif
-
-          (&out[i], _mm256_permutevar8x32_epi32(_mm256_max_epi8(
+            _mm256_loadA_si256(&in[i * 4 + 2]),
+            _mm256_loadA_si256(&in[i * 4 + 3])), kWeightScaleBits);
+        _mm256_storeA_si256(&out[i], _mm256_permutevar8x32_epi32(_mm256_max_epi8(
             _mm256_packs_epi16(words0, words1), kZero), kOffsets));
       }
       constexpr IndexType kStart = kNumChunks * kSimdWidth;
 
-  #elif defined(USE_SSSE3)
+  #elif defined(USE_SSE2)
       constexpr IndexType kNumChunks = kInputDimensions / kSimdWidth;
 
   #ifdef USE_SSE41
@@ -153,6 +115,24 @@ namespace Eval::NNUE::Layers {
       }
       constexpr IndexType kStart = kNumChunks * kSimdWidth;
 
+  #elif defined(USE_MMX)
+      constexpr IndexType kNumChunks = kInputDimensions / kSimdWidth;
+      const __m64 k0x80s = _mm_set1_pi8(-128);
+      const auto in = reinterpret_cast<const __m64*>(input);
+      const auto out = reinterpret_cast<__m64*>(output);
+      for (IndexType i = 0; i < kNumChunks; ++i) {
+        const __m64 words0 = _mm_srai_pi16(
+            _mm_packs_pi32(in[i * 4 + 0], in[i * 4 + 1]),
+            kWeightScaleBits);
+        const __m64 words1 = _mm_srai_pi16(
+            _mm_packs_pi32(in[i * 4 + 2], in[i * 4 + 3]),
+            kWeightScaleBits);
+        const __m64 packedbytes = _mm_packs_pi16(words0, words1);
+        out[i] = _mm_subs_pi8(_mm_adds_pi8(packedbytes, k0x80s), k0x80s);
+      }
+      _mm_empty();
+      constexpr IndexType kStart = kNumChunks * kSimdWidth;
+
   #elif defined(USE_NEON)
       constexpr IndexType kNumChunks = kInputDimensions / (kSimdWidth / 2);
       const int8x8_t kZero = {0};

src/nnue/nnue_accumulator.h

@@ -26,7 +26,7 @@
 namespace Eval::NNUE {
 
   // Class that holds the result of affine transformation of input features
-  struct alignas(32) Accumulator {
+  struct alignas(kCacheLineSize) Accumulator {
     std::int16_t
         accumulation[2][kRefreshTriggers.size()][kTransformedFeatureDimensions];
     Value score;

src/nnue/nnue_common.h

@@ -21,6 +21,9 @@
 #ifndef NNUE_COMMON_H_INCLUDED
 #define NNUE_COMMON_H_INCLUDED
 
+#include <cstring>
+#include <iostream>
+
 #if defined(USE_AVX2)
 #include <immintrin.h>
 
@@ -33,10 +36,36 @@
 #elif defined(USE_SSE2)
 #include <emmintrin.h>
 
+#elif defined(USE_MMX)
+#include <mmintrin.h>
+
 #elif defined(USE_NEON)
 #include <arm_neon.h>
 #endif
 
+// HACK: Use _mm256_loadu_si256() instead of _mm256_load_si256. Otherwise a binary
+//       compiled with older g++ crashes because the output memory is not aligned
+//       even though alignas is specified.
+#if defined(USE_AVX2)
+#if defined(__GNUC__ ) && (__GNUC__ < 9) && defined(_WIN32) && !defined(__clang__)
+#define _mm256_loadA_si256  _mm256_loadu_si256
+#define _mm256_storeA_si256 _mm256_storeu_si256
+#else
+#define _mm256_loadA_si256  _mm256_load_si256
+#define _mm256_storeA_si256 _mm256_store_si256
+#endif
+#endif
+
+#if defined(USE_AVX512)
+#if defined(__GNUC__ ) && (__GNUC__ < 9) && defined(_WIN32) && !defined(__clang__)
+#define _mm512_loadA_si512   _mm512_loadu_si512
+#define _mm512_storeA_si512  _mm512_storeu_si512
+#else
+#define _mm512_loadA_si512   _mm512_load_si512
+#define _mm512_storeA_si512  _mm512_store_si512
+#endif
+#endif
+
 namespace Eval::NNUE {
 
   // Version of the evaluation file
@@ -56,12 +85,36 @@ namespace Eval::NNUE {
   #elif defined(USE_SSE2)
   constexpr std::size_t kSimdWidth = 16;
 
+  #elif defined(USE_MMX)
+  constexpr std::size_t kSimdWidth = 8;
+
   #elif defined(USE_NEON)
   constexpr std::size_t kSimdWidth = 16;
   #endif
 
   constexpr std::size_t kMaxSimdWidth = 32;
 
+  // unique number for each piece type on each square
+  enum {
+    PS_NONE     =  0,
+    PS_W_PAWN   =  1,
+    PS_B_PAWN   =  1 * SQUARE_NB + 1,
+    PS_W_KNIGHT =  2 * SQUARE_NB + 1,
+    PS_B_KNIGHT =  3 * SQUARE_NB + 1,
+    PS_W_BISHOP =  4 * SQUARE_NB + 1,
+    PS_B_BISHOP =  5 * SQUARE_NB + 1,
+    PS_W_ROOK   =  6 * SQUARE_NB + 1,
+    PS_B_ROOK   =  7 * SQUARE_NB + 1,
+    PS_W_QUEEN  =  8 * SQUARE_NB + 1,
+    PS_B_QUEEN  =  9 * SQUARE_NB + 1,
+    PS_W_KING   = 10 * SQUARE_NB + 1,
+    PS_END      = PS_W_KING, // pieces without kings (pawns included)
+    PS_B_KING   = 11 * SQUARE_NB + 1,
+    PS_END2     = 12 * SQUARE_NB + 1
+  };
+
+  extern uint32_t kpp_board_index[PIECE_NB][COLOR_NB];
+
   // Type of input feature after conversion
   using TransformedFeatureType = std::uint8_t;
   using IndexType = std::uint32_t;
@@ -69,7 +122,25 @@ namespace Eval::NNUE {
   // Round n up to be a multiple of base
   template <typename IntType>
   constexpr IntType CeilToMultiple(IntType n, IntType base) {
-    return (n + base - 1) / base * base;
+      return (n + base - 1) / base * base;
+  }
+
+  // read_little_endian() is our utility to read an integer (signed or unsigned, any size)
+  // from a stream in little-endian order. We swap the byte order after the read if
+  // necessary to return a result with the byte ordering of the compiling machine.
+  template <typename IntType>
+  inline IntType read_little_endian(std::istream& stream) {
+
+      IntType result;
+      std::uint8_t u[sizeof(IntType)];
+      typename std::make_unsigned<IntType>::type v = 0;
+
+      stream.read(reinterpret_cast<char*>(u), sizeof(IntType));
+      for (std::size_t i = 0; i < sizeof(IntType); ++i)
+          v = (v << 8) | u[sizeof(IntType) - i - 1];
+
+      std::memcpy(&result, &v, sizeof(IntType));
+      return result;
   }
 
 }  // namespace Eval::NNUE

src/nnue/nnue_feature_transformer.h

@@ -55,10 +55,10 @@ namespace Eval::NNUE {
 
     // Read network parameters
     bool ReadParameters(std::istream& stream) {
-      stream.read(reinterpret_cast<char*>(biases_),
-                  kHalfDimensions * sizeof(BiasType));
-      stream.read(reinterpret_cast<char*>(weights_),
-                  kHalfDimensions * kInputDimensions * sizeof(WeightType));
+      for (std::size_t i = 0; i < kHalfDimensions; ++i)
+        biases_[i] = read_little_endian<BiasType>(stream);
+      for (std::size_t i = 0; i < kHalfDimensions * kInputDimensions; ++i)
+        weights_[i] = read_little_endian<WeightType>(stream);
       return !stream.fail();
     }
 
@@ -88,7 +88,7 @@ namespace Eval::NNUE {
       constexpr int kControl = 0b11011000;
       const __m256i kZero = _mm256_setzero_si256();
 
-  #elif defined(USE_SSSE3)
+  #elif defined(USE_SSE2)
       constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
 
   #ifdef USE_SSE41
@@ -97,6 +97,10 @@ namespace Eval::NNUE {
       const __m128i k0x80s = _mm_set1_epi8(-128);
   #endif
 
+  #elif defined(USE_MMX)
+      constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
+      const __m64 k0x80s = _mm_set1_pi8(-128);
+
   #elif defined(USE_NEON)
       constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
       const int8x8_t kZero = {0};
@@ -109,41 +113,15 @@ namespace Eval::NNUE {
   #if defined(USE_AVX2)
         auto out = reinterpret_cast<__m256i*>(&output[offset]);
         for (IndexType j = 0; j < kNumChunks; ++j) {
-          __m256i sum0 =
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-            // HACK: Use _mm256_loadu_si256() instead of _mm256_load_si256. Because the binary
-            //       compiled with g++ in MSYS2 crashes here because the output memory is not aligned
-            //       even though alignas is specified.
-            _mm256_loadu_si256
-  #else
-            _mm256_load_si256
-  #endif
-
-            (&reinterpret_cast<const __m256i*>(
-              accumulation[perspectives[p]][0])[j * 2 + 0]);
-          __m256i sum1 =
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-            _mm256_loadu_si256
-  #else
-            _mm256_load_si256
-  #endif
-
-            (&reinterpret_cast<const __m256i*>(
-              accumulation[perspectives[p]][0])[j * 2 + 1]);
-
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-          _mm256_storeu_si256
-  #else
-          _mm256_store_si256
-  #endif
-
-          (&out[j], _mm256_permute4x64_epi64(_mm256_max_epi8(
+          __m256i sum0 = _mm256_loadA_si256(
+              &reinterpret_cast<const __m256i*>(accumulation[perspectives[p]][0])[j * 2 + 0]);
+          __m256i sum1 = _mm256_loadA_si256(
+            &reinterpret_cast<const __m256i*>(accumulation[perspectives[p]][0])[j * 2 + 1]);
+          _mm256_storeA_si256(&out[j], _mm256_permute4x64_epi64(_mm256_max_epi8(
               _mm256_packs_epi16(sum0, sum1), kZero), kControl));
         }
 
-  #elif defined(USE_SSSE3)
+  #elif defined(USE_SSE2)
         auto out = reinterpret_cast<__m128i*>(&output[offset]);
         for (IndexType j = 0; j < kNumChunks; ++j) {
           __m128i sum0 = _mm_load_si128(&reinterpret_cast<const __m128i*>(
@@ -163,6 +141,17 @@ namespace Eval::NNUE {
           );
         }
 
+  #elif defined(USE_MMX)
+        auto out = reinterpret_cast<__m64*>(&output[offset]);
+        for (IndexType j = 0; j < kNumChunks; ++j) {
+          __m64 sum0 = *(&reinterpret_cast<const __m64*>(
+              accumulation[perspectives[p]][0])[j * 2 + 0]);
+          __m64 sum1 = *(&reinterpret_cast<const __m64*>(
+              accumulation[perspectives[p]][0])[j * 2 + 1]);
+          const __m64 packedbytes = _mm_packs_pi16(sum0, sum1);
+          out[j] = _mm_subs_pi8(_mm_adds_pi8(packedbytes, k0x80s), k0x80s);
+        }
+
   #elif defined(USE_NEON)
         const auto out = reinterpret_cast<int8x8_t*>(&output[offset]);
         for (IndexType j = 0; j < kNumChunks; ++j) {
@@ -180,6 +169,9 @@ namespace Eval::NNUE {
   #endif
 
       }
+  #if defined(USE_MMX)
+      _mm_empty();
+  #endif
     }
 
    private:
@@ -195,27 +187,37 @@ namespace Eval::NNUE {
                    kHalfDimensions * sizeof(BiasType));
         for (const auto index : active_indices[perspective]) {
           const IndexType offset = kHalfDimensions * index;
+  #if defined(USE_AVX512)
+          auto accumulation = reinterpret_cast<__m512i*>(
+              &accumulator.accumulation[perspective][i][0]);
+          auto column = reinterpret_cast<const __m512i*>(&weights_[offset]);
+          constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
+          for (IndexType j = 0; j < kNumChunks; ++j)
+            _mm512_storeA_si512(&accumulation[j], _mm512_add_epi16(_mm512_loadA_si512(&accumulation[j]), column[j]));
 
-  #if defined(USE_AVX2)
+  #elif defined(USE_AVX2)
           auto accumulation = reinterpret_cast<__m256i*>(
               &accumulator.accumulation[perspective][i][0]);
           auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
           constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-          for (IndexType j = 0; j < kNumChunks; ++j) {
-  #if defined(__MINGW32__) || defined(__MINGW64__)
-            _mm256_storeu_si256(&accumulation[j], _mm256_add_epi16(_mm256_loadu_si256(&accumulation[j]), column[j]));
-  #else
-            accumulation[j] = _mm256_add_epi16(accumulation[j], column[j]);
-  #endif
-          }
+          for (IndexType j = 0; j < kNumChunks; ++j)
+            _mm256_storeA_si256(&accumulation[j], _mm256_add_epi16(_mm256_loadA_si256(&accumulation[j]), column[j]));
 
   #elif defined(USE_SSE2)
           auto accumulation = reinterpret_cast<__m128i*>(
               &accumulator.accumulation[perspective][i][0]);
           auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
           constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-          for (IndexType j = 0; j < kNumChunks; ++j) {
+          for (IndexType j = 0; j < kNumChunks; ++j)
             accumulation[j] = _mm_add_epi16(accumulation[j], column[j]);
+
+  #elif defined(USE_MMX)
+          auto accumulation = reinterpret_cast<__m64*>(
+              &accumulator.accumulation[perspective][i][0]);
+          auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
+          constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
+          for (IndexType j = 0; j < kNumChunks; ++j) {
+            accumulation[j] = _mm_add_pi16(accumulation[j], column[j]);
           }
 
   #elif defined(USE_NEON)
@@ -223,18 +225,19 @@ namespace Eval::NNUE {
               &accumulator.accumulation[perspective][i][0]);
           auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
           constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-          for (IndexType j = 0; j < kNumChunks; ++j) {
+          for (IndexType j = 0; j < kNumChunks; ++j)
             accumulation[j] = vaddq_s16(accumulation[j], column[j]);
-          }
 
   #else
-          for (IndexType j = 0; j < kHalfDimensions; ++j) {
+          for (IndexType j = 0; j < kHalfDimensions; ++j)
             accumulator.accumulation[perspective][i][j] += weights_[offset + j];
-          }
   #endif
 
         }
       }
+  #if defined(USE_MMX)
+      _mm_empty();
+  #endif
 
       accumulator.computed_accumulation = true;
       accumulator.computed_score = false;
@@ -261,6 +264,11 @@ namespace Eval::NNUE {
         auto accumulation = reinterpret_cast<__m128i*>(
             &accumulator.accumulation[perspective][i][0]);
 
+  #elif defined(USE_MMX)
+        constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
+        auto accumulation = reinterpret_cast<__m64*>(
+            &accumulator.accumulation[perspective][i][0]);
+
   #elif defined(USE_NEON)
         constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
         auto accumulation = reinterpret_cast<int16x8_t*>(
@@ -290,6 +298,12 @@ namespace Eval::NNUE {
               accumulation[j] = _mm_sub_epi16(accumulation[j], column[j]);
             }
 
+  #elif defined(USE_MMX)
+            auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
+            for (IndexType j = 0; j < kNumChunks; ++j) {
+              accumulation[j] = _mm_sub_pi16(accumulation[j], column[j]);
+            }
+
   #elif defined(USE_NEON)
             auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
             for (IndexType j = 0; j < kNumChunks; ++j) {
@@ -321,6 +335,12 @@ namespace Eval::NNUE {
               accumulation[j] = _mm_add_epi16(accumulation[j], column[j]);
             }
 
+  #elif defined(USE_MMX)
+            auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
+            for (IndexType j = 0; j < kNumChunks; ++j) {
+              accumulation[j] = _mm_add_pi16(accumulation[j], column[j]);
+            }
+
   #elif defined(USE_NEON)
             auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
             for (IndexType j = 0; j < kNumChunks; ++j) {
@@ -337,6 +357,9 @@ namespace Eval::NNUE {
           }
         }
       }
+  #if defined(USE_MMX)
+      _mm_empty();
+  #endif
 
       accumulator.computed_accumulation = true;
       accumulator.computed_score = false;

src/pawns.cpp

@@ -30,21 +30,21 @@ namespace {
   #define S(mg, eg) make_score(mg, eg)
 
   // Pawn penalties
-  constexpr Score Backward      = S( 9, 24);
-  constexpr Score Doubled       = S(11, 56);
-  constexpr Score Isolated      = S( 5, 15);
-  constexpr Score WeakLever     = S( 0, 56);
-  constexpr Score WeakUnopposed = S(13, 27);
+  constexpr Score Backward      = S( 8, 27);
+  constexpr Score Doubled       = S(11, 55);
+  constexpr Score Isolated      = S( 5, 17);
+  constexpr Score WeakLever     = S( 2, 54);
+  constexpr Score WeakUnopposed = S(15, 25);
 
   // Bonus for blocked pawns at 5th or 6th rank
-  constexpr Score BlockedPawn[2] = { S(-11, -4), S(-3, 4) };
+  constexpr Score BlockedPawn[2] = { S(-13, -4), S(-4, 3) };
 
   constexpr Score BlockedStorm[RANK_NB] = {
     S(0, 0), S(0, 0), S(76, 78), S(-10, 15), S(-7, 10), S(-4, 6), S(-1, 2)
   };
 
   // Connected pawn bonus
-  constexpr int Connected[RANK_NB] = { 0, 7, 8, 12, 29, 48, 86 };
+  constexpr int Connected[RANK_NB] = { 0, 7, 8, 11, 24, 45, 85 };
 
   // Strength of pawn shelter for our king by [distance from edge][rank].
   // RANK_1 = 0 is used for files where we have no pawn, or pawn is behind our king.
@@ -219,7 +219,7 @@ Score Entry::evaluate_shelter(const Position& pos, Square ksq) const {
 
   Score bonus = make_score(5, 5);
 
-  File center = Utility::clamp(file_of(ksq), FILE_B, FILE_G);
+  File center = std::clamp(file_of(ksq), FILE_B, FILE_G);
   for (File f = File(center - 1); f <= File(center + 1); ++f)
   {
       b = ourPawns & file_bb(f);

src/position.cpp

@@ -198,9 +198,6 @@ Position& Position::set(const string& fenStr, bool isChess960, StateInfo* si, Th
   std::fill_n(&pieceList[0][0], sizeof(pieceList) / sizeof(Square), SQ_NONE);
   st = si;
 
-  // Each piece on board gets a unique ID used to track the piece later
-  PieceId piece_id, next_piece_id = PIECE_ID_ZERO;
-
   ss >> std::noskipws;
 
   // 1. Piece placement
@@ -212,21 +209,8 @@ Position& Position::set(const string& fenStr, bool isChess960, StateInfo* si, Th
       else if (token == '/')
           sq += 2 * SOUTH;
 
-      else if ((idx = PieceToChar.find(token)) != string::npos)
-      {
-          auto pc = Piece(idx);
-          put_piece(pc, sq);
-
-          if (Eval::useNNUE)
-          {
-              // Kings get a fixed ID, other pieces get ID in order of placement
-              piece_id =
-                (idx == W_KING) ? PIECE_ID_WKING :
-                (idx == B_KING) ? PIECE_ID_BKING :
-                next_piece_id++;
-              evalList.put_piece(piece_id, sq, pc);
-          }
-
+      else if ((idx = PieceToChar.find(token)) != string::npos) {
+          put_piece(Piece(idx), sq);
           ++sq;
       }
   }
@@ -721,8 +705,6 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
   // Used by NNUE
   st->accumulator.computed_accumulation = false;
   st->accumulator.computed_score = false;
-  PieceId dp0 = PIECE_ID_NONE;
-  PieceId dp1 = PIECE_ID_NONE;
   auto& dp = st->dirtyPiece;
   dp.dirty_num = 1;
 
@@ -775,12 +757,10 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
 
       if (Eval::useNNUE)
       {
-          dp.dirty_num = 2; // 2 pieces moved
-          dp1 = piece_id_on(capsq);
-          dp.pieceId[1] = dp1;
-          dp.old_piece[1] = evalList.piece_with_id(dp1);
-          evalList.put_piece(dp1, capsq, NO_PIECE);
-          dp.new_piece[1] = evalList.piece_with_id(dp1);
+          dp.dirty_num = 2;  // 1 piece moved, 1 piece captured
+          dp.piece[1] = captured;
+          dp.from[1] = capsq;
+          dp.to[1] = SQ_NONE;
       }
 
       // Update board and piece lists
@@ -821,11 +801,9 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
   {
       if (Eval::useNNUE)
       {
-          dp0 = piece_id_on(from);
-          dp.pieceId[0] = dp0;
-          dp.old_piece[0] = evalList.piece_with_id(dp0);
-          evalList.put_piece(dp0, to, pc);
-          dp.new_piece[0] = evalList.piece_with_id(dp0);
+          dp.piece[0] = pc;
+          dp.from[0] = from;
+          dp.to[0] = to;
       }
 
       move_piece(from, to);
@@ -854,9 +832,12 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
 
           if (Eval::useNNUE)
           {
-              dp0 = piece_id_on(to);
-              evalList.put_piece(dp0, to, promotion);
-              dp.new_piece[0] = evalList.piece_with_id(dp0);
+              // Promoting pawn to SQ_NONE, promoted piece from SQ_NONE
+              dp.to[0] = SQ_NONE;
+              dp.piece[dp.dirty_num] = promotion;
+              dp.from[dp.dirty_num] = SQ_NONE;
+              dp.to[dp.dirty_num] = to;
+              dp.dirty_num++;
           }
 
           // Update hash keys
@@ -950,12 +931,6 @@ void Position::undo_move(Move m) {
   {
       move_piece(to, from); // Put the piece back at the source square
 
-      if (Eval::useNNUE)
-      {
-          PieceId dp0 = st->dirtyPiece.pieceId[0];
-          evalList.put_piece(dp0, from, pc);
-      }
-
       if (st->capturedPiece)
       {
           Square capsq = to;
@@ -972,14 +947,6 @@ void Position::undo_move(Move m) {
           }
 
           put_piece(st->capturedPiece, capsq); // Restore the captured piece
-
-          if (Eval::useNNUE)
-          {
-              PieceId dp1 = st->dirtyPiece.pieceId[1];
-              assert(evalList.piece_with_id(dp1).from[WHITE] == PS_NONE);
-              assert(evalList.piece_with_id(dp1).from[BLACK] == PS_NONE);
-              evalList.put_piece(dp1, capsq, st->capturedPiece);
-          }
       }
   }
 
@@ -1001,32 +968,16 @@ void Position::do_castling(Color us, Square from, Square& to, Square& rfrom, Squ
   rto = relative_square(us, kingSide ? SQ_F1 : SQ_D1);
   to = relative_square(us, kingSide ? SQ_G1 : SQ_C1);
 
-  if (Eval::useNNUE)
+  if (Do && Eval::useNNUE)
   {
-      PieceId dp0, dp1;
       auto& dp = st->dirtyPiece;
-      dp.dirty_num = 2; // 2 pieces moved
-
-      if (Do)
-      {
-          dp0 = piece_id_on(from);
-          dp1 = piece_id_on(rfrom);
-          dp.pieceId[0] = dp0;
-          dp.old_piece[0] = evalList.piece_with_id(dp0);
-          evalList.put_piece(dp0, to, make_piece(us, KING));
-          dp.new_piece[0] = evalList.piece_with_id(dp0);
-          dp.pieceId[1] = dp1;
-          dp.old_piece[1] = evalList.piece_with_id(dp1);
-          evalList.put_piece(dp1, rto, make_piece(us, ROOK));
-          dp.new_piece[1] = evalList.piece_with_id(dp1);
-      }
-      else
-      {
-          dp0 = piece_id_on(to);
-          dp1 = piece_id_on(rto);
-          evalList.put_piece(dp0, from, make_piece(us, KING));
-          evalList.put_piece(dp1, rfrom, make_piece(us, ROOK));
-      }
+      dp.piece[0] = make_piece(us, KING);
+      dp.from[0] = from;
+      dp.to[0] = to;
+      dp.piece[1] = make_piece(us, ROOK);
+      dp.from[1] = rfrom;
+      dp.to[1] = rto;
+      dp.dirty_num = 2;
   }
 
   // Remove both pieces first since squares could overlap in Chess960
@@ -1145,8 +1096,8 @@ bool Position::see_ge(Move m, Value threshold) const {
 
       // Don't allow pinned pieces to attack (except the king) as long as
       // there are pinners on their original square.
-      if (st->pinners[~stm] & occupied)
-          stmAttackers &= ~st->blockersForKing[stm];
+      if (pinners(~stm) & occupied)
+          stmAttackers &= ~blockers_for_king(stm);
 
       if (!stmAttackers)
           break;

src/position.h

@@ -113,6 +113,7 @@ public:
   Bitboard checkers() const;
   Bitboard blockers_for_king(Color c) const;
   Bitboard check_squares(PieceType pt) const;
+  Bitboard pinners(Color c) const;
   bool is_discovery_check_on_king(Color c, Move m) const;
 
   // Attacks to/from a given square
@@ -170,7 +171,6 @@ public:
 
   // Used by NNUE
   StateInfo* state() const;
-  const EvalList* eval_list() const;
 
 private:
   // Initialization helpers (used while setting up a position)
@@ -185,9 +185,6 @@ private:
   template<bool Do>
   void do_castling(Color us, Square from, Square& to, Square& rfrom, Square& rto);
 
-  // ID of a piece on a given square
-  PieceId piece_id_on(Square sq) const;
-
   // Data members
   Piece board[SQUARE_NB];
   Bitboard byTypeBB[PIECE_TYPE_NB];
@@ -204,9 +201,6 @@ private:
   Thread* thisThread;
   StateInfo* st;
   bool chess960;
-
-  // List of pieces used in NNUE evaluation function
-  EvalList evalList;
 };
 
 namespace PSQT {
@@ -309,6 +303,10 @@ inline Bitboard Position::blockers_for_king(Color c) const {
   return st->blockersForKing[c];
 }
 
+inline Bitboard Position::pinners(Color c) const {
+  return st->pinners[c];
+}
+
 inline Bitboard Position::check_squares(PieceType pt) const {
   return st->checkSquares[pt];
 }
@@ -446,20 +444,4 @@ inline StateInfo* Position::state() const {
   return st;
 }
 
-inline const EvalList* Position::eval_list() const {
-
-  return &evalList;
-}
-
-inline PieceId Position::piece_id_on(Square sq) const
-{
-
-  assert(piece_on(sq) != NO_PIECE);
-
-  PieceId pid = evalList.piece_id_list[sq];
-  assert(is_ok(pid));
-
-  return pid;
-}
-
 #endif // #ifndef POSITION_H_INCLUDED

src/search.cpp

@@ -63,9 +63,9 @@ namespace {
   constexpr uint64_t TtHitAverageResolution = 1024;
 
   // Razor and futility margins
-  constexpr int RazorMargin = 527;
+  constexpr int RazorMargin = 510;
   Value futility_margin(Depth d, bool improving) {
-    return Value(227 * (d - improving));
+    return Value(223 * (d - improving));
   }
 
   // Reductions lookup table, initialized at startup
@@ -73,7 +73,7 @@ namespace {
 
   Depth reduction(bool i, Depth d, int mn) {
     int r = Reductions[d] * Reductions[mn];
-    return (r + 570) / 1024 + (!i && r > 1018);
+    return (r + 509) / 1024 + (!i && r > 894);
   }
 
   constexpr int futility_move_count(bool improving, Depth depth) {
@@ -82,7 +82,7 @@ namespace {
 
   // History and stats update bonus, based on depth
   int stat_bonus(Depth d) {
-    return d > 15 ? 27 : 17 * d * d + 133 * d - 134;
+    return d > 13 ? 29 : 17 * d * d + 134 * d - 134;
   }
 
   // Add a small random component to draw evaluations to avoid 3fold-blindness
@@ -192,7 +192,7 @@ namespace {
 void Search::init() {
 
   for (int i = 1; i < MAX_MOVES; ++i)
-      Reductions[i] = int((24.8 + std::log(Threads.size())) * std::log(i));
+      Reductions[i] = int((22.0 + std::log(Threads.size())) * std::log(i));
 }
 
 
@@ -335,7 +335,7 @@ void Thread::search() {
   // for match (TC 60+0.6) results spanning a wide range of k values.
   PRNG rng(now());
   double floatLevel = Options["UCI_LimitStrength"] ?
-                      Utility::clamp(std::pow((Options["UCI_Elo"] - 1346.6) / 143.4, 1 / 0.806), 0.0, 20.0) :
+                      std::clamp(std::pow((Options["UCI_Elo"] - 1346.6) / 143.4, 1 / 0.806), 0.0, 20.0) :
                         double(Options["Skill Level"]);
   int intLevel = int(floatLevel) +
                  ((floatLevel - int(floatLevel)) * 1024 > rng.rand<unsigned>() % 1024  ? 1 : 0);
@@ -403,12 +403,12 @@ void Thread::search() {
           if (rootDepth >= 4)
           {
               Value prev = rootMoves[pvIdx].previousScore;
-              delta = Value(19);
+              delta = Value(17);
               alpha = std::max(prev - delta,-VALUE_INFINITE);
               beta  = std::min(prev + delta, VALUE_INFINITE);
 
               // Adjust contempt based on root move's previousScore (dynamic contempt)
-              int dct = ct + (110 - ct / 2) * prev / (abs(prev) + 140);
+              int dct = ct + (105 - ct / 2) * prev / (abs(prev) + 149);
 
               contempt = (us == WHITE ?  make_score(dct, dct / 2)
                                       : -make_score(dct, dct / 2));
@@ -506,13 +506,13 @@ void Thread::search() {
           && !Threads.stop
           && !mainThread->stopOnPonderhit)
       {
-          double fallingEval = (296 + 6 * (mainThread->bestPreviousScore - bestValue)
-                                    + 6 * (mainThread->iterValue[iterIdx] - bestValue)) / 725.0;
-          fallingEval = Utility::clamp(fallingEval, 0.5, 1.5);
+          double fallingEval = (318 + 6 * (mainThread->bestPreviousScore - bestValue)
+                                    + 6 * (mainThread->iterValue[iterIdx] - bestValue)) / 825.0;
+          fallingEval = std::clamp(fallingEval, 0.5, 1.5);
 
           // If the bestMove is stable over several iterations, reduce time accordingly
-          timeReduction = lastBestMoveDepth + 10 < completedDepth ? 1.92 : 0.95;
-          double reduction = (1.47 + mainThread->previousTimeReduction) / (2.22 * timeReduction);
+          timeReduction = lastBestMoveDepth + 9 < completedDepth ? 1.92 : 0.95;
+          double reduction = (1.47 + mainThread->previousTimeReduction) / (2.32 * timeReduction);
 
           // Use part of the gained time from a previous stable move for the current move
           for (Thread* th : Threads)
@@ -537,7 +537,7 @@ void Thread::search() {
           }
           else if (   Threads.increaseDepth
                    && !mainThread->ponder
-                   && Time.elapsed() > totalTime * 0.56)
+                   && Time.elapsed() > totalTime * 0.58)
                    Threads.increaseDepth = false;
           else
                    Threads.increaseDepth = true;
@@ -596,8 +596,8 @@ namespace {
     Key posKey;
     Move ttMove, move, excludedMove, bestMove;
     Depth extension, newDepth;
-    Value bestValue, value, ttValue, eval, maxValue, probcutBeta;
-    bool ttHit, ttPv, formerPv, givesCheck, improving, didLMR, priorCapture;
+    Value bestValue, value, ttValue, eval, maxValue, probCutBeta;
+    bool ttHit, formerPv, givesCheck, improving, didLMR, priorCapture;
     bool captureOrPromotion, doFullDepthSearch, moveCountPruning,
          ttCapture, singularQuietLMR;
     Piece movedPiece;
@@ -644,6 +644,7 @@ namespace {
     assert(0 <= ss->ply && ss->ply < MAX_PLY);
 
     (ss+1)->ply = ss->ply + 1;
+    (ss+1)->ttPv = false;
     (ss+1)->excludedMove = bestMove = MOVE_NONE;
     (ss+2)->killers[0] = (ss+2)->killers[1] = MOVE_NONE;
     Square prevSq = to_sq((ss-1)->currentMove);
@@ -667,10 +668,11 @@ namespace {
     ttValue = ttHit ? value_from_tt(tte->value(), ss->ply, pos.rule50_count()) : VALUE_NONE;
     ttMove =  rootNode ? thisThread->rootMoves[thisThread->pvIdx].pv[0]
             : ttHit    ? tte->move() : MOVE_NONE;
-    ttPv = PvNode || (ttHit && tte->is_pv());
-    formerPv = ttPv && !PvNode;
+    if (!excludedMove)
+        ss->ttPv = PvNode || (ttHit && tte->is_pv());
+    formerPv = ss->ttPv && !PvNode;
 
-    if (   ttPv
+    if (   ss->ttPv
         && depth > 12
         && ss->ply - 1 < MAX_LPH
         && !priorCapture
@@ -748,7 +750,7 @@ namespace {
                 if (    b == BOUND_EXACT
                     || (b == BOUND_LOWER ? value >= beta : value <= alpha))
                 {
-                    tte->save(posKey, value_to_tt(value, ss->ply), ttPv, b,
+                    tte->save(posKey, value_to_tt(value, ss->ply), ss->ttPv, b,
                               std::min(MAX_PLY - 1, depth + 6),
                               MOVE_NONE, VALUE_NONE);
 
@@ -794,15 +796,11 @@ namespace {
     else
     {
         if ((ss-1)->currentMove != MOVE_NULL)
-        {
-            int bonus = -(ss-1)->statScore / 512;
-
-            ss->staticEval = eval = evaluate(pos) + bonus;
-        }
+            ss->staticEval = eval = evaluate(pos);
         else
             ss->staticEval = eval = -(ss-1)->staticEval + 2 * Tempo;
 
-        tte->save(posKey, VALUE_NONE, ttPv, BOUND_NONE, DEPTH_NONE, MOVE_NONE, eval);
+        tte->save(posKey, VALUE_NONE, ss->ttPv, BOUND_NONE, DEPTH_NONE, MOVE_NONE, eval);
     }
 
     // Step 7. Razoring (~1 Elo)
@@ -811,12 +809,13 @@ namespace {
         &&  eval <= alpha - RazorMargin)
         return qsearch<NT>(pos, ss, alpha, beta);
 
-    improving =  (ss-2)->staticEval == VALUE_NONE ? (ss->staticEval > (ss-4)->staticEval
-              || (ss-4)->staticEval == VALUE_NONE) : ss->staticEval > (ss-2)->staticEval;
+    improving =  (ss-2)->staticEval == VALUE_NONE
+               ? ss->staticEval > (ss-4)->staticEval || (ss-4)->staticEval == VALUE_NONE
+               : ss->staticEval > (ss-2)->staticEval;
 
     // Step 8. Futility pruning: child node (~50 Elo)
     if (   !PvNode
-        &&  depth < 6
+        &&  depth < 8
         &&  eval - futility_margin(depth, improving) >= beta
         &&  eval < VALUE_KNOWN_WIN) // Do not return unproven wins
         return eval;
@@ -824,10 +823,10 @@ namespace {
     // Step 9. Null move search with verification search (~40 Elo)
     if (   !PvNode
         && (ss-1)->currentMove != MOVE_NULL
-        && (ss-1)->statScore < 23824
+        && (ss-1)->statScore < 22977
         &&  eval >= beta
         &&  eval >= ss->staticEval
-        &&  ss->staticEval >= beta - 33 * depth - 33 * improving + 112 * ttPv + 311
+        &&  ss->staticEval >= beta - 30 * depth - 28 * improving + 84 * ss->ttPv + 182
         && !excludedMove
         &&  pos.non_pawn_material(us)
         && (ss->ply >= thisThread->nmpMinPly || us != thisThread->nmpColor))
@@ -835,7 +834,7 @@ namespace {
         assert(eval - beta >= 0);
 
         // Null move dynamic reduction based on depth and value
-        Depth R = (737 + 77 * depth) / 246 + std::min(int(eval - beta) / 192, 3);
+        Depth R = (817 + 71 * depth) / 213 + std::min(int(eval - beta) / 192, 3);
 
         ss->currentMove = MOVE_NULL;
         ss->continuationHistory = &thisThread->continuationHistory[0][0][NO_PIECE][0];
@@ -871,7 +870,7 @@ namespace {
         }
     }
 
-    probcutBeta = beta + 176 - 49 * improving;
+    probCutBeta = beta + 176 - 49 * improving;
 
     // Step 10. ProbCut (~10 Elo)
     // If we have a good enough capture and a reduced search returns a value
@@ -879,22 +878,30 @@ namespace {
     if (   !PvNode
         &&  depth > 4
         &&  abs(beta) < VALUE_TB_WIN_IN_MAX_PLY
+        // if value from transposition table is lower than probCutBeta, don't attempt probCut
+        // there and in further interactions with transposition table cutoff depth is set to depth - 3
+        // because probCut search has depth set to depth - 4 but we also do a move before it
+        // so effective depth is equal to depth - 3
         && !(   ttHit
              && tte->depth() >= depth - 3
              && ttValue != VALUE_NONE
-             && ttValue < probcutBeta))
+             && ttValue < probCutBeta))
     {
+        // if ttMove is a capture and value from transposition table is good enough produce probCut
+        // cutoff without digging into actual probCut search
         if (   ttHit
             && tte->depth() >= depth - 3
             && ttValue != VALUE_NONE
-            && ttValue >= probcutBeta
+            && ttValue >= probCutBeta
             && ttMove
             && pos.capture_or_promotion(ttMove))
-            return probcutBeta;
+            return probCutBeta;
 
-        assert(probcutBeta < VALUE_INFINITE);
-        MovePicker mp(pos, ttMove, probcutBeta - ss->staticEval, &captureHistory);
+        assert(probCutBeta < VALUE_INFINITE);
+        MovePicker mp(pos, ttMove, probCutBeta - ss->staticEval, &captureHistory);
         int probCutCount = 0;
+        bool ttPv = ss->ttPv;
+        ss->ttPv = false;
 
         while (   (move = mp.next_move()) != MOVE_NONE
                && probCutCount < 2 + 2 * cutNode)
@@ -915,16 +922,17 @@ namespace {
                 pos.do_move(move, st);
 
                 // Perform a preliminary qsearch to verify that the move holds
-                value = -qsearch<NonPV>(pos, ss+1, -probcutBeta, -probcutBeta+1);
+                value = -qsearch<NonPV>(pos, ss+1, -probCutBeta, -probCutBeta+1);
 
                 // If the qsearch held, perform the regular search
-                if (value >= probcutBeta)
-                    value = -search<NonPV>(pos, ss+1, -probcutBeta, -probcutBeta+1, depth - 4, !cutNode);
+                if (value >= probCutBeta)
+                    value = -search<NonPV>(pos, ss+1, -probCutBeta, -probCutBeta+1, depth - 4, !cutNode);
 
                 pos.undo_move(move);
 
-                if (value >= probcutBeta)
+                if (value >= probCutBeta)
                 {
+                    // if transposition table doesn't have equal or more deep info write probCut data into it
                     if ( !(ttHit
                        && tte->depth() >= depth - 3
                        && ttValue != VALUE_NONE))
@@ -934,17 +942,14 @@ namespace {
                     return value;
                 }
             }
+         ss->ttPv = ttPv;
     }
 
-    // Step 11. Internal iterative deepening (~1 Elo)
-    if (depth >= 7 && !ttMove)
-    {
-        search<NT>(pos, ss, alpha, beta, depth - 7, cutNode);
-
-        tte = TT.probe(posKey, ttHit);
-        ttValue = ttHit ? value_from_tt(tte->value(), ss->ply, pos.rule50_count()) : VALUE_NONE;
-        ttMove = ttHit ? tte->move() : MOVE_NONE;
-    }
+    // Step 11. If the position is not in TT, decrease depth by 2
+    if (   PvNode
+        && depth >= 6
+        && !ttMove)
+        depth -= 2;
 
 moves_loop: // When in check, search starts from here
 
@@ -986,6 +991,10 @@ moves_loop: // When in check, search starts from here
                                   thisThread->rootMoves.begin() + thisThread->pvLast, move))
           continue;
 
+      // Check for legality
+      if (!rootNode && !pos.legal(move))
+          continue;
+
       ss->moveCount = ++moveCount;
 
       if (rootNode && thisThread == Threads.main() && Time.elapsed() > 3000)
@@ -1024,17 +1033,17 @@ moves_loop: // When in check, search starts from here
                   continue;
 
               // Futility pruning: parent node (~5 Elo)
-              if (   lmrDepth < 6
+              if (   lmrDepth < 7
                   && !ss->inCheck
-                  && ss->staticEval + 284 + 188 * lmrDepth <= alpha
+                  && ss->staticEval + 283 + 170 * lmrDepth <= alpha
                   &&  (*contHist[0])[movedPiece][to_sq(move)]
                     + (*contHist[1])[movedPiece][to_sq(move)]
                     + (*contHist[3])[movedPiece][to_sq(move)]
-                    + (*contHist[5])[movedPiece][to_sq(move)] / 2 < 28388)
+                    + (*contHist[5])[movedPiece][to_sq(move)] / 2 < 27376)
                   continue;
 
               // Prune moves with negative SEE (~20 Elo)
-              if (!pos.see_ge(move, Value(-(29 - std::min(lmrDepth, 17)) * lmrDepth * lmrDepth)))
+              if (!pos.see_ge(move, Value(-(29 - std::min(lmrDepth, 18)) * lmrDepth * lmrDepth)))
                   continue;
           }
           else
@@ -1051,12 +1060,12 @@ moves_loop: // When in check, search starts from here
                   && !(PvNode && abs(bestValue) < 2)
                   && PieceValue[MG][type_of(movedPiece)] >= PieceValue[MG][type_of(pos.piece_on(to_sq(move)))]
                   && !ss->inCheck
-                  && ss->staticEval + 267 + 391 * lmrDepth
+                  && ss->staticEval + 169 + 244 * lmrDepth
                      + PieceValue[MG][type_of(pos.piece_on(to_sq(move)))] <= alpha)
                   continue;
 
               // See based pruning
-              if (!pos.see_ge(move, Value(-202) * depth)) // (~25 Elo)
+              if (!pos.see_ge(move, Value(-221) * depth)) // (~25 Elo)
                   continue;
           }
       }
@@ -1068,15 +1077,14 @@ moves_loop: // When in check, search starts from here
       // then that move is singular and should be extended. To verify this we do
       // a reduced search on all the other moves but the ttMove and if the
       // result is lower than ttValue minus a margin, then we will extend the ttMove.
-      if (    depth >= 6
+      if (    depth >= 7
           &&  move == ttMove
           && !rootNode
           && !excludedMove // Avoid recursive singular search
        /* &&  ttValue != VALUE_NONE Already implicit in the next condition */
           &&  abs(ttValue) < VALUE_KNOWN_WIN
           && (tte->bound() & BOUND_LOWER)
-          &&  tte->depth() >= depth - 3
-          &&  pos.legal(move))
+          &&  tte->depth() >= depth - 3)
       {
           Value singularBeta = ttValue - ((formerPv + 4) * depth) / 2;
           Depth singularDepth = (depth - 1 + 3 * formerPv) / 2;
@@ -1116,34 +1124,28 @@ moves_loop: // When in check, search starts from here
                && (pos.is_discovery_check_on_king(~us, move) || pos.see_ge(move)))
           extension = 1;
 
-      // Passed pawn extension
-      else if (   move == ss->killers[0]
-               && pos.advanced_pawn_push(move)
-               && pos.pawn_passed(us, to_sq(move)))
-          extension = 1;
-
       // Last captures extension
       else if (   PieceValue[EG][pos.captured_piece()] > PawnValueEg
                && pos.non_pawn_material() <= 2 * RookValueMg)
           extension = 1;
 
       // Castling extension
-      if (type_of(move) == CASTLING)
+      if (   type_of(move) == CASTLING
+          && popcount(pos.pieces(us) & ~pos.pieces(PAWN) & (to_sq(move) & KingSide ? KingSide : QueenSide)) <= 2)
           extension = 1;
 
+      // Late irreversible move extension
+      if (   move == ttMove
+          && pos.rule50_count() > 80
+          && (captureOrPromotion || type_of(movedPiece) == PAWN))
+          extension = 2;
+
       // Add extension to new depth
       newDepth += extension;
 
       // Speculative prefetch as early as possible
       prefetch(TT.first_entry(pos.key_after(move)));
 
-      // Check for legality just before making the move
-      if (!rootNode && !pos.legal(move))
-      {
-          ss->moveCount = --moveCount;
-          continue;
-      }
-
       // Update the current move (this must be done after singular extension search)
       ss->currentMove = move;
       ss->continuationHistory = &thisThread->continuationHistory[ss->inCheck]
@@ -1157,13 +1159,12 @@ moves_loop: // When in check, search starts from here
       // Step 16. Reduced depth search (LMR, ~200 Elo). If the move fails high it will be
       // re-searched at full depth.
       if (    depth >= 3
-          &&  moveCount > 1 + 2 * rootNode
-          && (!rootNode || thisThread->best_move_count(move) == 0)
+          &&  moveCount > 1 + 2 * rootNode + 2 * (PvNode && abs(bestValue) < 2)
           && (  !captureOrPromotion
               || moveCountPruning
               || ss->staticEval + PieceValue[EG][pos.captured_piece()] <= alpha
               || cutNode
-              || thisThread->ttHitAverage < 415 * TtHitAverageResolution * TtHitAverageWindow / 1024))
+              || thisThread->ttHitAverage < 427 * TtHitAverageResolution * TtHitAverageWindow / 1024))
       {
           Depth r = reduction(improving, depth, moveCount);
 
@@ -1175,7 +1176,7 @@ moves_loop: // When in check, search starts from here
               r--;
 
           // Decrease reduction if the ttHit running average is large
-          if (thisThread->ttHitAverage > 473 * TtHitAverageResolution * TtHitAverageWindow / 1024)
+          if (thisThread->ttHitAverage > 509 * TtHitAverageResolution * TtHitAverageWindow / 1024)
               r--;
 
           // Reduction if other threads are searching this position
@@ -1183,7 +1184,7 @@ moves_loop: // When in check, search starts from here
               r++;
 
           // Decrease reduction if position is or has been on the PV (~10 Elo)
-          if (ttPv)
+          if (ss->ttPv)
               r -= 2;
 
           if (moveCountPruning && !formerPv)
@@ -1212,23 +1213,23 @@ moves_loop: // When in check, search starts from here
               // hence break make_move(). (~2 Elo)
               else if (    type_of(move) == NORMAL
                        && !pos.see_ge(reverse_move(move)))
-                  r -= 2 + ttPv - (type_of(movedPiece) == PAWN);
+                  r -= 2 + ss->ttPv - (type_of(movedPiece) == PAWN);
 
               ss->statScore =  thisThread->mainHistory[us][from_to(move)]
                              + (*contHist[0])[movedPiece][to_sq(move)]
                              + (*contHist[1])[movedPiece][to_sq(move)]
                              + (*contHist[3])[movedPiece][to_sq(move)]
-                             - 4826;
+                             - 5287;
 
               // Decrease/increase reduction by comparing opponent's stat score (~10 Elo)
-              if (ss->statScore >= -100 && (ss-1)->statScore < -112)
+              if (ss->statScore >= -106 && (ss-1)->statScore < -104)
                   r--;
 
-              else if ((ss-1)->statScore >= -125 && ss->statScore < -138)
+              else if ((ss-1)->statScore >= -119 && ss->statScore < -140)
                   r++;
 
               // Decrease/increase reduction for moves with a good/bad history (~30 Elo)
-              r -= ss->statScore / 14615;
+              r -= ss->statScore / 14884;
           }
           else
           {
@@ -1238,11 +1239,11 @@ moves_loop: // When in check, search starts from here
 
             // Unless giving check, this capture is likely bad
             if (   !givesCheck
-                && ss->staticEval + PieceValue[EG][pos.captured_piece()] + 211 * depth <= alpha)
+                && ss->staticEval + PieceValue[EG][pos.captured_piece()] + 213 * depth <= alpha)
                 r++;
           }
 
-          Depth d = Utility::clamp(newDepth - r, 1, newDepth);
+          Depth d = std::clamp(newDepth - r, 1, newDepth);
 
           value = -search<NonPV>(pos, ss+1, -(alpha+1), -alpha, d, true);
 
@@ -1390,8 +1391,17 @@ moves_loop: // When in check, search starts from here
     if (PvNode)
         bestValue = std::min(bestValue, maxValue);
 
+    // If no good move is found and the previous position was ttPv, then the previous
+    // opponent move is probably good and the new position is added to the search tree.
+    if (bestValue <= alpha)
+        ss->ttPv = ss->ttPv || ((ss-1)->ttPv && depth > 3);
+    // Otherwise, a counter move has been found and if the position is the last leaf
+    // in the search tree, remove the position from the search tree.
+    else if (depth > 3)
+        ss->ttPv = ss->ttPv && (ss+1)->ttPv;
+
     if (!excludedMove && !(rootNode && thisThread->pvIdx))
-        tte->save(posKey, value_to_tt(bestValue, ss->ply), ttPv,
+        tte->save(posKey, value_to_tt(bestValue, ss->ply), ss->ttPv,
                   bestValue >= beta ? BOUND_LOWER :
                   PvNode && bestMove ? BOUND_EXACT : BOUND_UPPER,
                   depth, bestMove, ss->staticEval);
@@ -1500,7 +1510,7 @@ moves_loop: // When in check, search starts from here
         if (PvNode && bestValue > alpha)
             alpha = bestValue;
 
-        futilityBase = bestValue + 141;
+        futilityBase = bestValue + 145;
     }
 
     const PieceToHistory* contHist[] = { (ss-1)->continuationHistory, (ss-2)->continuationHistory,
@@ -1534,6 +1544,10 @@ moves_loop: // When in check, search starts from here
       {
           assert(type_of(move) != ENPASSANT); // Due to !pos.advanced_pawn_push
 
+          // moveCount pruning
+          if (moveCount > 2)
+              continue;
+
           futilityValue = futilityBase + PieceValue[EG][pos.piece_on(to_sq(move))];
 
           if (futilityValue <= alpha)
@@ -1550,7 +1564,7 @@ moves_loop: // When in check, search starts from here
       }
 
       // Do not search moves with negative SEE values
-      if (  !ss->inCheck && !pos.see_ge(move))
+      if (!ss->inCheck && !pos.see_ge(move))
           continue;
 
       // Speculative prefetch as early as possible
@@ -1569,6 +1583,12 @@ moves_loop: // When in check, search starts from here
                                                                 [pos.moved_piece(move)]
                                                                 [to_sq(move)];
 
+      if (  !captureOrPromotion
+          && moveCount
+          && (*contHist[0])[pos.moved_piece(move)][to_sq(move)] < CounterMovePruneThreshold
+          && (*contHist[1])[pos.moved_piece(move)][to_sq(move)] < CounterMovePruneThreshold)
+          continue;
+
       // Make and search the move
       pos.do_move(move, st, givesCheck);
       value = -qsearch<NT>(pos, ss+1, -beta, -alpha, depth - 1);
@@ -1751,7 +1771,7 @@ moves_loop: // When in check, search starts from here
     }
 
     if (depth > 11 && ss->ply < MAX_LPH)
-        thisThread->lowPlyHistory[ss->ply][from_to(move)] << stat_bonus(depth - 6);
+        thisThread->lowPlyHistory[ss->ply][from_to(move)] << stat_bonus(depth - 7);
   }
 
   // When playing with strength handicap, choose best move among a set of RootMoves
@@ -1840,12 +1860,15 @@ string UCI::pv(const Position& pos, Depth depth, Value alpha, Value beta) {
   {
       bool updated = rootMoves[i].score != -VALUE_INFINITE;
 
-      if (depth == 1 && !updated)
+      if (depth == 1 && !updated && i > 0)
           continue;
 
-      Depth d = updated ? depth : depth - 1;
+      Depth d = updated ? depth : std::max(1, depth - 1);
       Value v = updated ? rootMoves[i].score : rootMoves[i].previousScore;
 
+      if (v == -VALUE_INFINITE)
+          v = VALUE_ZERO;
+
       bool tb = TB::RootInTB && abs(v) < VALUE_MATE_IN_MAX_PLY;
       v = tb ? rootMoves[i].tbScore : v;
 
@@ -1943,7 +1966,7 @@ void Tablebases::rank_root_moves(Position& pos, Search::RootMoves& rootMoves) {
     if (RootInTB)
     {
         // Sort moves according to TB rank
-        std::sort(rootMoves.begin(), rootMoves.end(),
+        std::stable_sort(rootMoves.begin(), rootMoves.end(),
                   [](const RootMove &a, const RootMove &b) { return a.tbRank > b.tbRank; } );
 
         // Probe during search only if DTZ is not available and we are winning

src/search.h

@@ -48,6 +48,7 @@ struct Stack {
   int statScore;
   int moveCount;
   bool inCheck;
+  bool ttPv;
 };
 
 

src/syzygy/tbprobe.cpp

@@ -223,7 +223,9 @@ public:
 
         *mapping = statbuf.st_size;
         *baseAddress = mmap(nullptr, statbuf.st_size, PROT_READ, MAP_SHARED, fd, 0);
+#if defined(MADV_RANDOM)
         madvise(*baseAddress, statbuf.st_size, MADV_RANDOM);
+#endif
         ::close(fd);
 
         if (*baseAddress == MAP_FAILED)
@@ -758,7 +760,7 @@ Ret do_probe_table(const Position& pos, T* entry, WDLScore wdl, ProbeState* resu
     if (entry->hasPawns) {
         idx = LeadPawnIdx[leadPawnsCnt][squares[0]];
 
-        std::sort(squares + 1, squares + leadPawnsCnt, pawns_comp);
+        std::stable_sort(squares + 1, squares + leadPawnsCnt, pawns_comp);
 
         for (int i = 1; i < leadPawnsCnt; ++i)
             idx += Binomial[i][MapPawns[squares[i]]];
@@ -859,7 +861,7 @@ encode_remaining:
 
     while (d->groupLen[++next])
     {
-        std::sort(groupSq, groupSq + d->groupLen[next]);
+        std::stable_sort(groupSq, groupSq + d->groupLen[next]);
         uint64_t n = 0;
 
         // Map down a square if "comes later" than a square in the previous

src/thread.cpp

@@ -51,17 +51,6 @@ Thread::~Thread() {
 }
 
 
-/// Thread::bestMoveCount(Move move) return best move counter for the given root move
-
-int Thread::best_move_count(Move move) const {
-
-  auto rm = std::find(rootMoves.begin() + pvIdx,
-                      rootMoves.begin() + pvLast, move);
-
-  return rm != rootMoves.begin() + pvLast ? rm->bestMoveCount : 0;
-}
-
-
 /// Thread::clear() reset histories, usually before a new game
 
 void Thread::clear() {
@@ -204,21 +193,18 @@ void ThreadPool::start_thinking(Position& pos, StateListPtr& states,
 
   // We use Position::set() to set root position across threads. But there are
   // some StateInfo fields (previous, pliesFromNull, capturedPiece) that cannot
-  // be deduced from a fen string, so set() clears them and to not lose the info
-  // we need to backup and later restore setupStates->back(). Note that setupStates
-  // is shared by threads but is accessed in read-only mode.
-  StateInfo tmp = setupStates->back();
-
+  // be deduced from a fen string, so set() clears them and they are set from
+  // setupStates->back() later. The rootState is per thread, earlier states are shared
+  // since they are read-only.
   for (Thread* th : *this)
   {
       th->nodes = th->tbHits = th->nmpMinPly = th->bestMoveChanges = 0;
       th->rootDepth = th->completedDepth = 0;
       th->rootMoves = rootMoves;
-      th->rootPos.set(pos.fen(), pos.is_chess960(), &setupStates->back(), th);
+      th->rootPos.set(pos.fen(), pos.is_chess960(), &th->rootState, th);
+      th->rootState = setupStates->back();
   }
 
-  setupStates->back() = tmp;
-
   main()->start_searching();
 }
 

src/thread.h

@@ -54,7 +54,6 @@ public:
   void idle_loop();
   void start_searching();
   void wait_for_search_finished();
-  int best_move_count(Move move) const;
 
   Pawns::Table pawnsTable;
   Material::Table materialTable;
@@ -65,6 +64,7 @@ public:
   std::atomic<uint64_t> nodes, tbHits, bestMoveChanges;
 
   Position rootPos;
+  StateInfo rootState;
   Search::RootMoves rootMoves;
   Depth rootDepth, completedDepth;
   CounterMoveHistory counterMoves;

src/thread_win32_osx.h

@@ -27,7 +27,7 @@
 /// The implementation calls pthread_create() with the stack size parameter
 /// equal to the linux 8MB default, on platforms that support it.
 
-#if defined(__APPLE__) || defined(__MINGW32__) || defined(__MINGW64__)
+#if defined(__APPLE__) || defined(__MINGW32__) || defined(__MINGW64__) || defined(USE_PTHREADS)
 
 #include <pthread.h>
 

src/timeman.cpp

@@ -38,9 +38,9 @@ void TimeManagement::init(Search::LimitsType& limits, Color us, int ply) {
   TimePoint slowMover       = TimePoint(Options["Slow Mover"]);
   TimePoint npmsec          = TimePoint(Options["nodestime"]);
 
-  // opt_scale is a percentage of available time to use for the current move.
-  // max_scale is a multiplier applied to optimumTime.
-  double opt_scale, max_scale;
+  // optScale is a percentage of available time to use for the current move.
+  // maxScale is a multiplier applied to optimumTime.
+  double optScale, maxScale;
 
   // If we have to play in 'nodes as time' mode, then convert from time
   // to nodes, and use resulting values in time management formulas.
@@ -75,22 +75,22 @@ void TimeManagement::init(Search::LimitsType& limits, Color us, int ply) {
   // game time for the current move, so also cap to 20% of available game time.
   if (limits.movestogo == 0)
   {
-      opt_scale = std::min(0.008 + std::pow(ply + 3.0, 0.5) / 250.0,
+      optScale = std::min(0.008 + std::pow(ply + 3.0, 0.5) / 250.0,
                            0.2 * limits.time[us] / double(timeLeft));
-      max_scale = std::min(7.0, 4.0 + ply / 12.0);
+      maxScale = std::min(7.0, 4.0 + ply / 12.0);
   }
 
   // x moves in y seconds (+ z increment)
   else
   {
-      opt_scale = std::min((0.8 + ply / 128.0) / mtg,
+      optScale = std::min((0.8 + ply / 128.0) / mtg,
                             0.8 * limits.time[us] / double(timeLeft));
-      max_scale = std::min(6.3, 1.5 + 0.11 * mtg);
+      maxScale = std::min(6.3, 1.5 + 0.11 * mtg);
   }
 
   // Never use more than 80% of the available time for this move
-  optimumTime = TimePoint(opt_scale * timeLeft);
-  maximumTime = TimePoint(std::min(0.8 * limits.time[us] - moveOverhead, max_scale * optimumTime));
+  optimumTime = TimePoint(optScale * timeLeft);
+  maximumTime = TimePoint(std::min(0.8 * limits.time[us] - moveOverhead, maxScale * optimumTime));
 
   if (Options["Ponder"])
       optimumTime += optimumTime / 4;

src/tt.cpp

@@ -37,18 +37,19 @@ void TTEntry::save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev)
   if (m || (uint16_t)k != key16)
       move16 = (uint16_t)m;
 
-  // Overwrite less valuable entries
-  if ((uint16_t)k != key16
-      || d - DEPTH_OFFSET > depth8 - 4
-      || b == BOUND_EXACT)
+  // Overwrite less valuable entries (cheapest checks first)
+  if (b == BOUND_EXACT
+      || (uint16_t)k != key16
+      || d - DEPTH_OFFSET > depth8 - 4)
   {
-      assert(d >= DEPTH_OFFSET);
+      assert(d > DEPTH_OFFSET);
+      assert(d < 256 + DEPTH_OFFSET);
 
       key16     = (uint16_t)k;
+      depth8    = (uint8_t)(d - DEPTH_OFFSET);
+      genBound8 = (uint8_t)(TT.generation8 | uint8_t(pv) << 2 | b);
       value16   = (int16_t)v;
       eval16    = (int16_t)ev;
-      genBound8 = (uint8_t)(TT.generation8 | uint8_t(pv) << 2 | b);
-      depth8    = (uint8_t)(d - DEPTH_OFFSET);
   }
 }
 
@@ -119,11 +120,11 @@ TTEntry* TranspositionTable::probe(const Key key, bool& found) const {
   const uint16_t key16 = (uint16_t)key;  // Use the low 16 bits as key inside the cluster
 
   for (int i = 0; i < ClusterSize; ++i)
-      if (!tte[i].key16 || tte[i].key16 == key16)
+      if (tte[i].key16 == key16 || !tte[i].depth8)
       {
           tte[i].genBound8 = uint8_t(generation8 | (tte[i].genBound8 & 0x7)); // Refresh
 
-          return found = (bool)tte[i].key16, &tte[i];
+          return found = (bool)tte[i].depth8, &tte[i];
       }
 
   // Find an entry to be replaced according to the replacement strategy
@@ -149,7 +150,7 @@ int TranspositionTable::hashfull() const {
   int cnt = 0;
   for (int i = 0; i < 1000; ++i)
       for (int j = 0; j < ClusterSize; ++j)
-          cnt += (table[i].entry[j].genBound8 & 0xF8) == generation8;
+          cnt += table[i].entry[j].depth8 && (table[i].entry[j].genBound8 & 0xF8) == generation8;
 
   return cnt / ClusterSize;
 }

src/tt.h

@@ -25,13 +25,13 @@
 /// TTEntry struct is the 10 bytes transposition table entry, defined as below:
 ///
 /// key        16 bit
-/// move       16 bit
-/// value      16 bit
-/// eval value 16 bit
+/// depth       8 bit
 /// generation  5 bit
 /// pv node     1 bit
 /// bound type  2 bit
-/// depth       8 bit
+/// move       16 bit
+/// value      16 bit
+/// eval value 16 bit
 
 struct TTEntry {
 
@@ -47,11 +47,11 @@ private:
   friend class TranspositionTable;
 
   uint16_t key16;
+  uint8_t  depth8;
+  uint8_t  genBound8;
   uint16_t move16;
   int16_t  value16;
   int16_t  eval16;
-  uint8_t  genBound8;
-  uint8_t  depth8;
 };
 
 

src/types.h

@@ -178,7 +178,7 @@ enum Value : int {
   VALUE_MATE_IN_MAX_PLY  =  VALUE_MATE - MAX_PLY,
   VALUE_MATED_IN_MAX_PLY = -VALUE_MATE_IN_MAX_PLY,
 
-  PawnValueMg   = 124,   PawnValueEg   = 206,
+  PawnValueMg   = 126,   PawnValueEg   = 208,
   KnightValueMg = 781,   KnightValueEg = 854,
   BishopValueMg = 825,   BishopValueEg = 915,
   RookValueMg   = 1276,  RookValueEg   = 1380,
@@ -201,22 +201,6 @@ enum Piece {
   PIECE_NB = 16
 };
 
-// An ID used to track the pieces. Max. 32 pieces on board.
-enum PieceId {
-  PIECE_ID_ZERO   = 0,
-  PIECE_ID_KING   = 30,
-  PIECE_ID_WKING  = 30,
-  PIECE_ID_BKING  = 31,
-  PIECE_ID_NONE   = 32
-};
-
-inline PieceId operator++(PieceId& d, int) {
-
-  PieceId x = d;
-  d = PieceId(int(d) + 1);
-  return x;
-}
-
 constexpr Value PieceValue[PHASE_NB][PIECE_NB] = {
   { VALUE_ZERO, PawnValueMg, KnightValueMg, BishopValueMg, RookValueMg, QueenValueMg, VALUE_ZERO, VALUE_ZERO,
     VALUE_ZERO, PawnValueMg, KnightValueMg, BishopValueMg, RookValueMg, QueenValueMg, VALUE_ZERO, VALUE_ZERO },
@@ -232,7 +216,8 @@ enum : int {
   DEPTH_QS_RECAPTURES = -5,
 
   DEPTH_NONE   = -6,
-  DEPTH_OFFSET = DEPTH_NONE
+
+  DEPTH_OFFSET = -7 // value used only for TT entry occupancy check
 };
 
 enum Square : int {
@@ -270,93 +255,20 @@ enum Rank : int {
   RANK_1, RANK_2, RANK_3, RANK_4, RANK_5, RANK_6, RANK_7, RANK_8, RANK_NB
 };
 
-// unique number for each piece type on each square
-enum PieceSquare : uint32_t {
-  PS_NONE     =  0,
-  PS_W_PAWN   =  1,
-  PS_B_PAWN   =  1 * SQUARE_NB + 1,
-  PS_W_KNIGHT =  2 * SQUARE_NB + 1,
-  PS_B_KNIGHT =  3 * SQUARE_NB + 1,
-  PS_W_BISHOP =  4 * SQUARE_NB + 1,
-  PS_B_BISHOP =  5 * SQUARE_NB + 1,
-  PS_W_ROOK   =  6 * SQUARE_NB + 1,
-  PS_B_ROOK   =  7 * SQUARE_NB + 1,
-  PS_W_QUEEN  =  8 * SQUARE_NB + 1,
-  PS_B_QUEEN  =  9 * SQUARE_NB + 1,
-  PS_W_KING   = 10 * SQUARE_NB + 1,
-  PS_END      = PS_W_KING, // pieces without kings (pawns included)
-  PS_B_KING   = 11 * SQUARE_NB + 1,
-  PS_END2     = 12 * SQUARE_NB + 1
-};
-
-struct ExtPieceSquare {
-  PieceSquare from[COLOR_NB];
-};
-
-// Array for finding the PieceSquare corresponding to the piece on the board
-extern ExtPieceSquare kpp_board_index[PIECE_NB];
-
-constexpr bool is_ok(PieceId pid);
-constexpr Square rotate180(Square sq);
-
-// Structure holding which tracked piece (PieceId) is where (PieceSquare)
-class EvalList {
-
-public:
-  // Max. number of pieces without kings is 30 but must be a multiple of 4 in AVX2
-  static const int MAX_LENGTH = 32;
-
-  // Array that holds the piece id for the pieces on the board
-  PieceId piece_id_list[SQUARE_NB];
-
-  // List of pieces, separate from White and Black POV
-  PieceSquare* piece_list_fw() const { return const_cast<PieceSquare*>(pieceListFw); }
-  PieceSquare* piece_list_fb() const { return const_cast<PieceSquare*>(pieceListFb); }
-
-  // Place the piece pc with piece_id on the square sq on the board
-  void put_piece(PieceId piece_id, Square sq, Piece pc)
-  {
-      assert(is_ok(piece_id));
-      if (pc != NO_PIECE)
-      {
-          pieceListFw[piece_id] = PieceSquare(kpp_board_index[pc].from[WHITE] + sq);
-          pieceListFb[piece_id] = PieceSquare(kpp_board_index[pc].from[BLACK] + rotate180(sq));
-          piece_id_list[sq] = piece_id;
-      }
-      else
-      {
-          pieceListFw[piece_id] = PS_NONE;
-          pieceListFb[piece_id] = PS_NONE;
-          piece_id_list[sq] = piece_id;
-      }
-  }
-
-  // Convert the specified piece_id piece to ExtPieceSquare type and return it
-  ExtPieceSquare piece_with_id(PieceId piece_id) const
-  {
-      ExtPieceSquare eps;
-      eps.from[WHITE] = pieceListFw[piece_id];
-      eps.from[BLACK] = pieceListFb[piece_id];
-      return eps;
-  }
-
-private:
-  PieceSquare pieceListFw[MAX_LENGTH];
-  PieceSquare pieceListFb[MAX_LENGTH];
-};
-
-// For differential evaluation of pieces that changed since last turn
+// Keep track of what a move changes on the board (used by NNUE)
 struct DirtyPiece {
 
   // Number of changed pieces
   int dirty_num;
 
-  // The ids of changed pieces, max. 2 pieces can change in one move
-  PieceId pieceId[2];
+  // Max 3 pieces can change in one move. A promotion with capture moves
+  // both the pawn and the captured piece to SQ_NONE and the piece promoted
+  // to from SQ_NONE to the capture square.
+  Piece piece[3];
 
-  // What changed from the piece with that piece number
-  ExtPieceSquare old_piece[2];
-  ExtPieceSquare new_piece[2];
+  // From and to squares, which may be SQ_NONE
+  Square from[3];
+  Square to[3];
 };
 
 /// Score enum stores a middlegame and an endgame value in a single integer (enum).
@@ -406,8 +318,6 @@ ENABLE_FULL_OPERATORS_ON(Value)
 ENABLE_FULL_OPERATORS_ON(Direction)
 
 ENABLE_INCR_OPERATORS_ON(Piece)
-ENABLE_INCR_OPERATORS_ON(PieceSquare)
-ENABLE_INCR_OPERATORS_ON(PieceId)
 ENABLE_INCR_OPERATORS_ON(PieceType)
 ENABLE_INCR_OPERATORS_ON(Square)
 ENABLE_INCR_OPERATORS_ON(File)
@@ -496,10 +406,6 @@ inline Color color_of(Piece pc) {
   return Color(pc >> 3);
 }
 
-constexpr bool is_ok(PieceId pid) {
-  return pid < PIECE_ID_NONE;
-}
-
 constexpr bool is_ok(Square s) {
   return s >= SQ_A1 && s <= SQ_H8;
 }
@@ -536,11 +442,6 @@ constexpr Square to_sq(Move m) {
   return Square(m & 0x3F);
 }
 
-// Return relative square when turning the board 180 degrees
-constexpr Square rotate180(Square sq) {
-  return (Square)(sq ^ 0x3F);
-}
-
 constexpr int from_to(Move m) {
  return m & 0xFFF;
 }

src/uci.cpp

@@ -211,7 +211,7 @@ namespace {
      double b = (((bs[0] * m + bs[1]) * m + bs[2]) * m) + bs[3];
 
      // Transform eval to centipawns with limited range
-     double x = Utility::clamp(double(100 * v) / PawnValueEg, -1000.0, 1000.0);
+     double x = std::clamp(double(100 * v) / PawnValueEg, -1000.0, 1000.0);
 
      // Return win rate in per mille (rounded to nearest)
      return int(0.5 + 1000 / (1 + std::exp((a - x) / b)));

src/ucioption.cpp

@@ -21,6 +21,7 @@
 #include <ostream>
 #include <sstream>
 
+#include "evaluate.h"
 #include "misc.h"
 #include "search.h"
 #include "thread.h"
@@ -78,8 +79,8 @@ void init(OptionsMap& o) {
   o["SyzygyProbeDepth"]      << Option(1, 1, 100);
   o["Syzygy50MoveRule"]      << Option(true);
   o["SyzygyProbeLimit"]      << Option(7, 0, 7);
-  o["Use NNUE"]              << Option(false, on_use_NNUE);
-  o["EvalFile"]              << Option("nn-97f742aaefcd.nnue", on_eval_file);
+  o["Use NNUE"]              << Option(true, on_use_NNUE);
+  o["EvalFile"]              << Option(EvalFileDefaultName, on_eval_file);
 }
 
 

tests/instrumented.sh

@@ -70,7 +70,7 @@ for args in "eval" \
             "go depth 10" \
             "go movetime 1000" \
             "go wtime 8000 btime 8000 winc 500 binc 500" \
-            "bench 128 $threads 10 default depth"
+            "bench 128 $threads 8 default depth"
 do
 
    echo "$prefix $exeprefix ./stockfish $args $postfix"
@@ -80,7 +80,7 @@ done
 
 # more general testing, following an uci protocol exchange
 cat << EOF > game.exp
- set timeout 10
+ set timeout 240
  spawn $exeprefix ./stockfish
 
  send "uci\n"
@@ -98,7 +98,7 @@ cat << EOF > game.exp
  expect "bestmove"
 
  send "position fen 5rk1/1K4p1/8/8/3B4/8/8/8 b - - 0 1\n"
- send "go depth 30\n"
+ send "go depth 20\n"
  expect "bestmove"
 
  send "quit\n"
@@ -121,7 +121,7 @@ cat << EOF > syzygy.exp
  send "uci\n"
  send "setoption name SyzygyPath value ../tests/syzygy/\n"
  expect "info string Found 35 tablebases" {} timeout {exit 1}
- send "bench 128 1 10 default depth\n"
+ send "bench 128 1 8 default depth\n"
  send "quit\n"
  expect eof