Stockfish 13

Official release version of Stockfish 13

Bench: 3766422

-----

It is our pleasure to release Stockfish 13 to chess fans worldwide.
As usual, downloads are freely available at

https://stockfishchess.org

The Stockfish project builds on a thriving community of enthusiasts
who contribute their expertise, time, and resources to build a free
and open-source chess engine that is robust, widely available, and
very strong. We would like to thank them all!

The good news first: from now on, our users can expect more frequent
high-quality releases of Stockfish! Sadly, this decision has been
triggered by the start of sales of the Fat Fritz 2 engine by ChessBase,
which is a copy of a very recent development version of Stockfish
with minor modifications. We refer to our statement on Fat Fritz 2[1]
and a community blog[2] for further information.

This version of Stockfish is significantly stronger than any of its
predecessors. Stockfish 13 outperforms Stockfish 12 by at least
35 Elo[3]. When playing against a one-year-old Stockfish, it wins 60
times more game pairs than it loses[4]. This release features an NNUE
network retrained on billions of positions, much faster network
evaluation code, and significantly improved search heuristics, as
well as additional evaluation tweaks. In the course of its development,
this version has won the superfinals of the TCEC Season 19 and
TCEC Season 20.

Going forward, the Leela Chess Zero and Stockfish teams will join
forces to demonstrate our commitment to open source chess engines and
training tools, and open data. We are convinced that our free and
open-source chess engines serve the chess community very well.

Stay safe and enjoy chess!

The Stockfish team
[1] https://blog.stockfishchess.org/post/643239805544792064/statement-on-fat-fritz-2
[2] https://lichess.org/blog/YCvy7xMAACIA8007/fat-fritz-2-is-a-rip-off
[3] https://tests.stockfishchess.org/tests/view/602bcccf7f517a561bc49b11
[4] https://tests.stockfishchess.org/tests/view/600fbb9c735dd7f0f0352d59

.gitignore

@@ -0,0 +1,12 @@
+# Files from build
+**/*.o
+**/*.s
+src/.depend
+
+# Built binary
+src/stockfish*
+src/-lstdc++.res
+
+# Neural network for the NNUE evaluation
+**/*.nnue
+

.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

@@ -19,6 +19,7 @@ Alain Savard (Rocky640)
 Alayan Feh (Alayan-stk-2)
 Alexander Kure
 Alexander Pagel (Lolligerhans)
+Alfredo Menezes (lonfom169)
 Ali AlZhrani (Cooffe)
 Andrew Grant (AndyGrant)
 Andrey Neporada (nepal)
@@ -32,16 +33,20 @@ Bill Henry (VoyagerOne)
 Bojun Guo (noobpwnftw, Nooby)
 braich
 Brian Sheppard (SapphireBrand, briansheppard-toast)
+Bruno de Melo Costa (BM123499)
 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)
+Deshawn Mohan-Smith (GoldenRare)
+Dieter Dobbelaere (ddobbelaere)
 DiscanX
 Dominik Schlösser (domschl)
 double-beep
@@ -53,11 +58,13 @@ 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)
@@ -79,8 +86,9 @@ Jean Gauthier (OuaisBla)
 Jean-Francois Romang (jromang)
 Jekaa
 Jerry Donald Watson (jerrydonaldwatson)
+jjoshua2
 Jonathan Calovski (Mysseno)
-Jonathan Dumale (SFisGOD)
+Jonathan Buladas Dumale (SFisGOD)
 Joost VandeVondele (vondele)
 Jörg Oster (joergoster)
 Joseph Ellis (jhellis3)
@@ -92,6 +100,7 @@ Ken Takusagawa
 kinderchocolate
 Kiran Panditrao (Krgp)
 Kojirion
+Krystian Kuzniarek (kuzkry)
 Leonardo Ljubičić (ICCF World Champion)
 Leonid Pechenik (lp--)
 Linus Arver (listx)
@@ -104,8 +113,10 @@ Maciej Żenczykowski (zenczykowski)
 Malcolm Campbell (xoto10)
 Mark Tenzer (31m059)
 marotear
+Matt Ginsberg (mattginsberg)
 Matthew Lai (matthewlai)
 Matthew Sullivan (Matt14916)
+Maxim Molchanov (Maxim)
 Michael An (man)
 Michael Byrne (MichaelB7)
 Michael Chaly (Vizvezdenec)
@@ -125,6 +136,7 @@ Niklas Fiekas (niklasf)
 Nikolay Kostov (NikolayIT)
 Nguyen Pham (nguyenpham)
 Norman Schmidt (FireFather)
+notruck
 Ondrej Mosnáček (WOnder93)
 Oskar Werkelin Ahlin
 Pablo Vazquez
@@ -151,6 +163,7 @@ 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
+updatable 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
@@ -24,19 +24,14 @@ This distribution of Stockfish consists of the following files:
   * Readme.md, the file you are currently reading.
 
   * Copying.txt, a text file containing the GNU General Public License version 3.
+  
+  * AUTHORS, a text file with the list of authors for the project
 
   * 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
-```
-https://tests.stockfishchess.org/api/nn/[filename]
-```
-replacing `[filename]` as needed.
-
+  * a file with the .nnue extension, storing the neural network for the NNUE 
+    evaluation. Binary distributions will have this file embedded.
 
 ## UCI options
 
@@ -49,6 +44,9 @@ Currently, Stockfish has the following UCI options:
   * #### Hash
     The size of the hash table in MB. It is recommended to set Hash after setting Threads.
 
+  * #### Clear Hash
+    Clear the hash table.
+
   * #### Ponder
     Let Stockfish ponder its next move while the opponent is thinking.
 
@@ -58,19 +56,14 @@ 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.
@@ -109,8 +102,8 @@ Currently, Stockfish has the following UCI options:
 
   * #### SyzygyProbeDepth
     Minimum remaining search depth for which a position is probed. Set this option
-    to a higher value to probe less agressively if you experience too much slowdown
-    (in terms of nps) due to TB probing.
+    to a higher value to probe less aggressively if you experience too much slowdown
+    (in terms of nps) due to tablebase probing.
 
   * #### Syzygy50MoveRule
     Disable to let fifty-move rule draws detected by Syzygy tablebase probes count
@@ -120,6 +113,14 @@ 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.
@@ -132,42 +133,50 @@ Currently, Stockfish has the following UCI options:
     Tells the engine to use nodes searched instead of wall time to account for
     elapsed time. Useful for engine testing.
 
-  * #### Clear Hash
-    Clear the hash table.
-
   * #### Debug Log File
     Write all communication to and from the engine into a text file.
 
-## classical and NNUE evaluation
+## A note on classical evaluation versus 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
 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 (e.g. piece positions only). The network is optimized and trained
-on the evalutions of millions of positions at moderate search depth.
+on the evaluations of millions of positions at moderate search depth.
 
 The NNUE evaluation was first introduced in shogi, and ported to Stockfish afterward.
 It can be evaluated efficiently on CPUs, and exploits the fact that only parts
 of the neural network need to be updated after a typical chess move.
 [The nodchip repository](https://github.com/nodchip/Stockfish) provides additional
-tools to train and develop the NNUE networks.
+tools to train and develop the NNUE networks. On CPUs supporting modern vector instructions
+(avx2 and similar), the NNUE evaluation results in much stronger playing strength, even
+if the nodes per second computed by the engine is somewhat lower (roughly 80% of nps
+is typical).
 
-On CPUs supporting modern vector instructions (avx2 and similar), the NNUE evaluation
-results in stronger playing strength, even if the nodes per second computed by the engine
-is somewhat lower (roughly 60% of nps is typical).
+Notes:
 
-Note that the NNUE evaluation depends on the Stockfish binary and the network parameter
-file (see EvalFile). Not every parameter file is compatible with a given Stockfish binary.
-The default value of the EvalFile UCI option is the name of a network that is guaranteed
-to be compatible with that binary.
+1) the NNUE evaluation depends on the Stockfish binary and the network parameter
+file (see the EvalFile UCI option). Not every parameter file is compatible with a given
+Stockfish binary, but the default value of the EvalFile UCI option is the name of a network
+that is guaranteed to be compatible with that binary.
 
-## What to expect from Syzygybases?
+2) 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]
+```
+replacing `[filename]` as needed.
+
+## What to expect from the Syzygy tablebases?
 
 If the engine is searching a position that is not in the tablebases (e.g.
 a position with 8 pieces), it will access the tablebases during the search.
-If the engine reports a very large score (typically 153.xx), this means
-that it has found a winning line into a tablebase position.
+If the engine reports a very large score (typically 153.xx), this means 
+it has found a winning line into a tablebase position.
 
 If the engine is given a position to search that is in the tablebases, it
 will use the tablebases at the beginning of the search to preselect all
@@ -175,14 +184,14 @@ good moves, i.e. all moves that preserve the win or preserve the draw while
 taking into account the 50-move rule.
 It will then perform a search only on those moves. **The engine will not move
 immediately**, unless there is only a single good move. **The engine likely
-will not report a mate score even if the position is known to be won.**
+will not report a mate score, even if the position is known to be won.**
 
 It is therefore clear that this behaviour is not identical to what one might
 be used to with Nalimov tablebases. There are technical reasons for this
 difference, the main technical reason being that Nalimov tablebases use the
-DTM metric (distance-to-mate), while Syzygybases use a variation of the
+DTM metric (distance-to-mate), while the Syzygy tablebases use a variation of the
 DTZ metric (distance-to-zero, zero meaning any move that resets the 50-move
-counter). This special metric is one of the reasons that Syzygybases are
+counter). This special metric is one of the reasons that the Syzygy tablebases are
 more compact than Nalimov tablebases, while still storing all information
 needed for optimal play and in addition being able to take into account
 the 50-move rule.
@@ -191,8 +200,8 @@ the 50-move rule.
 
 Stockfish supports large pages on Linux and Windows. Large pages make
 the hash access more efficient, improving the engine speed, especially
-on large hash sizes. Typical increases are 5..10% in terms of nps, but
-speed increases up to 30% have been measured. The support is
+on large hash sizes. Typical increases are 5..10% in terms of nodes per
+second, but speed increases up to 30% have been measured. The support is
 automatic. Stockfish attempts to use large pages when available and
 will fall back to regular memory allocation when this is not the case.
 
@@ -200,17 +209,17 @@ will fall back to regular memory allocation when this is not the case.
 
 Large page support on Linux is obtained by the Linux kernel
 transparent huge pages functionality. Typically, transparent huge pages
-are already enabled and no configuration is needed.
+are already enabled, and no configuration is needed.
 
 ### Support on Windows
 
 The use of large pages requires "Lock Pages in Memory" privilege. See
 [Enable the Lock Pages in Memory Option (Windows)](https://docs.microsoft.com/en-us/sql/database-engine/configure-windows/enable-the-lock-pages-in-memory-option-windows)
-on how to enable this privilege. Logout/login may be needed
-afterwards. Due to memory fragmentation, it may not always be
-possible to allocate large pages even when enabled. A reboot
-might alleviate this problem. To determine whether large pages
-are in use, see the engine log.
+on how to enable this privilege, then run [RAMMap](https://docs.microsoft.com/en-us/sysinternals/downloads/rammap)
+to double-check that large pages are used. We suggest that you reboot
+your computer after you have enabled large pages, because long Windows
+sessions suffer from memory fragmentation, which may prevent Stockfish
+from getting large pages: a fresh session is better in this regard.
 
 ## Compiling Stockfish yourself from the sources
 
@@ -225,10 +234,11 @@ targets with corresponding descriptions.
 ```
     cd src
     make help
+    make net
     make build ARCH=x86-64-modern
 ```
 
-When not using the Makefile to compile (for instance with Microsoft MSVC) you
+When not using the Makefile to compile (for instance, with Microsoft MSVC) you
 need to manually set/unset some switches in the compiler command line; see
 file *types.h* for a quick reference.
 
@@ -237,8 +247,7 @@ 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
@@ -265,8 +274,9 @@ generic rather than being focused on Stockfish's precise implementation.
 Nevertheless, a helpful resource.
 
 * The latest source can always be found on [GitHub](https://github.com/official-stockfish/Stockfish).
-Discussions about Stockfish take place in the [FishCooking](https://groups.google.com/forum/#!forum/fishcooking)
-group and engine testing is done on [Fishtest](https://tests.stockfishchess.org/tests).
+Discussions about Stockfish take place these days mainly in the [FishCooking](https://groups.google.com/forum/#!forum/fishcooking)
+group and on the [Stockfish Discord channel](https://discord.gg/nv8gDtt).
+The engine testing is done on [Fishtest](https://tests.stockfishchess.org/tests).
 If you want to help improve Stockfish, please read this [guideline](https://github.com/glinscott/fishtest/wiki/Creating-my-first-test)
 first, where the basics of Stockfish development are explained.
 
@@ -274,16 +284,17 @@ first, where the basics of Stockfish development are explained.
 ## Terms of use
 
 Stockfish is free, and distributed under the **GNU General Public License version 3**
-(GPL v3). Essentially, this means that you are free to do almost exactly
+(GPL v3). Essentially, this means you are free to do almost exactly
 what you want with the program, including distributing it among your
-friends, making it available for download from your web site, selling
+friends, making it available for download from your website, selling
 it (either by itself or as part of some bigger software package), or
 using it as the starting point for a software project of your own.
 
 The only real limitation is that whenever you distribute Stockfish in
-some way, you must always include the full source code, or a pointer
-to where the source code can be found. If you make any changes to the
-source code, these changes must also be made available under the GPL.
+some way, you MUST always include the full source code, or a pointer
+to where the source code can be found, to generate the exact binary
+you are distributing. If you make any changes to the source code,
+these changes must also be made available under the GPL.
 
 For full details, read the copy of the GPL v3 found in the file named
 *Copying.txt*.

Top CPU Contributors.txt

@@ -1,154 +1,189 @@
-Contributors with >10,000 CPU hours as of January 7, 2020
+Contributors to Fishtest with >10,000 CPU hours, as of Feb 15, 2021.
 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
+Username                CPU Hours       Games played
+----------------------------------------------------
+noobpwnftw               23930906         1560559941
+dew                       1169948           70333008
+mlang                      957168           61657446
+mibere                     703840           46867607
+tvijlbrief                 517888           33379462
+JojoM                      515404           30334272
+cw                         443276           29385549
+crunchy                    427035           27344275
+grandphish2                425794           26347253
+fastgm                     414133           24519696
+gvreuls                    377843           24708884
+CSU_Dynasty                338718           23030006
+Fisherman                  326795           21820747
+TueRens                    313730           19490246
+ctoks                      298442           20052551
+velislav                   270519           17355456
+bcross                     241064           17196165
+glinscott                  217799           13780820
+nordlandia                 211692           13484886
+bking_US                   198894           11876016
+drabel                     191096           13129722
+leszek                     189170           11446821
+mgrabiak                   187153           12013300
+robal                      181389           11539242
+Thanar                     179852           12365359
+vdv                        175274            9889046
+spams                      157128           10319326
+marrco                     150292            9401741
+sqrt2                      147963            9724586
+CoffeeOne                  137086            5022516
+vdbergh                    137041            8926915
+malala                     136182            8002293
+mhoram                     132780            8398229
+xoto                       124729            8652088
+davar                      122092            7960001
+dsmith                     122059            7570238
+Data                       113305            8220352
+BrunoBanani                112960            7436849
+pemo                       109598            5036441
+Dantist                    106768            6431396
+MaZePallas                 102741            6630419
+ElbertoOne                  99028            7023771
+brabos                      92118            6186135
+linrock                     90903            6708639
+psk                         89957            5984901
+sunu                        88614            6020673
+sterni1971                  86948            5613788
+Vizvezdenec                 83761            5344740
+BRAVONE                     81239            5054681
+nssy                        76497            5259388
+cuistot                     76366            4370584
+racerschmacer               75753            5442626
+teddybaer                   75125            5407666
+Pking_cda                   73776            5293873
+0x3C33                      73133            4670293
+jromang                     72117            5054915
+solarlight                  70517            5028306
+dv8silencer                 70287            3883992
+Bobo1239                    68515            4652287
+manap                       66273            4121774
+tinker                      64321            4268390
+robnjr                      57262            4053117
+Freja                       56938            3733019
+ttruscott                   56010            3680085
+rkl                         54986            4150767
+renouve                     53811            3501516
+finfish                     51360            3370515
+eva42                       51272            3599691
+rap                         49985            3219146
+pb00067                     49727            3298270
+amicic                      49691            3042481
+ronaldjerum                 47654            3240695
+bigpen0r                    47278            3291647
+biffhero                    46564            3111352
+VoyagerOne                  45476            3452465
+eastorwest                  45033            3071805
+speedycpu                   43842            3003273
+jbwiebe                     43305            2805433
+Antihistamine               41788            2761312
+mhunt                       41735            2691355
+homyur                      39893            2850481
+gri                         39871            2515779
+oryx                        38282            2944400
+Spprtr                      38157            2470529
+SC                          37290            2731014
+csnodgrass                  36207            2688994
+jmdana                      36157            2210661
+strelock                    34716            2074055
+Garf                        33800            2747562
+skiminki                    33515            2055584
+EthanOConnor                33370            2090311
+slakovv                     32915            2021889
+yurikvelo                   32600            2255966
+Prcuvu                      30377            2170122
+manapbk                     30326            1770143
+anst                        30301            2190091
+jkiiski                     30136            1904470
+hyperbolic.tom              29840            2017394
+Pyafue                      29650            1902349
+qurashee                    27758            1509620
+OuaisBla                    27636            1578800
+chriswk                     26902            1868317
+achambord                   26582            1767323
+Fifis                       26376            1776853
+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                     23164            1591830
+cisco2015                   22895            1762069
+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
+jjoshua2                    20997            1422689
+horst.prack                 20878            1465656
+0xB00B1ES                   20590            1208666
+sphinx                      20515            1352368
+j3corre                     20405            941444
+Adrian.Schmidt123           20316            1281436
+Ente                        20017            1432602
+wei                         19973            1745989
+rstoesser                   19569            1293588
+eudhan                      19274            1283717
+jundery                     18445            1115855
+iisiraider                  18247            1101015
+ville                       17883            1384026
+chris                       17698            1487385
+purplefishies               17595            1092533
+DMBK                        17357            1279152
+DragonLord                  17014            1162790
+dju                         16515             929427
+IgorLeMasson                16064            1147232
+ako027ako                   15671            1173203
+Nikolay.IT                  15154            1068349
+Andrew Grant                15114             895539
+OssumOpossum                14857            1007129
+enedene                     14476             905279
+bpfliegel                   14298             884523
+jpulman                     13982             870599
+joster                      13794             950160
+Nesa92                      13786            1114691
+crocogoat                   13753            1114622
+Hjax                        13535             915487
+Dark_wizzie                 13422            1007152
+mpx86                       12941             693640
+mabichito                   12903             749391
+thijsk                      12886             722107
+AdrianSA                    12860             804972
+Flopzee                     12698             894821
+fatmurphy                   12547             853210
+scuzzi                      12511             845761
+Karby                       12429             735880
+SapphireBrand               12416             969604
+modolief                    12386             896470
+pgontarz                    12151             848794
+stocky                      11954             699440
+mschmidt                    11941             803401
+infinity                    11470             727027
+torbjo                      11395             729145
+Thomas A. Anderson          11372             732094
+d64                         11263             789184
+Maxim                       11129             804704
+snicolet                    11106             869170
+MooTheCow                   11008             694942
+savage84                    10965             641068
+Rudolphous                  10915             741268
+Wolfgang                    10809             580032
+rpngn                       10712             688203
+basepi                      10637             744851
+michaelrpg                  10409             735127
+dzjp                        10343             732529
+ali-al-zhrani               10324             726502
+ols                         10259             570669
+lbraesch                    10252             647825

appveyor.yml

@@ -4,7 +4,6 @@ clone_depth: 50
 branches:
   only:
     - master
-    - nnue-player-wip
 
 # Operating system (build VM template)
 os: Visual Studio 2019
@@ -62,6 +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

@@ -1,7 +1,5 @@
 # Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-# Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-# Copyright (C) 2008-2015 Marco Costalba, Joona Kiiski, Tord Romstad
-# Copyright (C) 2015-2019 Marco Costalba, Joona Kiiski, Gary Linscott, Tord Romstad
+# Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 #
 # Stockfish is free software: you can redistribute it and/or modify
 # it under the terms of the GNU General Public License as published by
@@ -67,14 +65,16 @@ 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
@@ -82,151 +82,188 @@ 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
+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
@@ -250,6 +287,8 @@ ifeq ($(ARCH),ppc-64)
 	prefetch = yes
 endif
 
+endif
+
 ### ==========================================================================
 ### Section 3. Low-level Configuration
 ### ==========================================================================
@@ -268,7 +307,7 @@ 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)
@@ -278,6 +317,10 @@ ifeq ($(COMP),gcc)
 		LDFLAGS += -m$(bits)
 	endif
 
+	ifeq ($(arch),$(filter $(arch),armv7))
+		LDFLAGS += -latomic
+	endif
+
 	ifneq ($(KERNEL),Darwin)
 	   LDFLAGS += -Wl,--no-as-needed
 	endif
@@ -321,11 +364,13 @@ ifeq ($(COMP),clang)
 
 	ifneq ($(KERNEL),Darwin)
 	ifneq ($(KERNEL),OpenBSD)
+	ifneq ($(KERNEL),FreeBSD)
 		LDFLAGS += -latomic
 	endif
 	endif
+	endif
 
-	ifeq ($(ARCH),$(filter $(ARCH),armv7 armv8))
+	ifeq ($(arch),$(filter $(arch),armv7 armv8))
 		ifeq ($(OS),Android)
 			CXXFLAGS += -m$(bits)
 			LDFLAGS += -m$(bits)
@@ -336,23 +381,40 @@ ifeq ($(COMP),clang)
 	endif
 endif
 
+ifeq ($(KERNEL),Darwin)
+	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
+	comp=clang
+	ifeq ($(arch),armv7)
+		CXX=armv7a-linux-androideabi16-clang++
+		CXXFLAGS += -mthumb -march=armv7-a -mfloat-abi=softfp -mfpu=neon
+		STRIP=arm-linux-androideabi-strip
+	endif
+	ifeq ($(arch),armv8)
+		CXX=aarch64-linux-android21-clang++
+		STRIP=aarch64-linux-android-strip
+	endif
+	LDFLAGS += -static-libstdc++ -pie -lm -latomic
+endif
+
 ifeq ($(comp),icc)
 	profile_make = icc-profile-make
 	profile_use = icc-profile-use
-else
-ifeq ($(comp),clang)
+else ifeq ($(comp),clang)
 	profile_make = clang-profile-make
 	profile_use = clang-profile-use
 else
 	profile_make = gcc-profile-make
 	profile_use = gcc-profile-use
 endif
-endif
-
-ifeq ($(KERNEL),Darwin)
-	CXXFLAGS += -arch $(arch) -mmacosx-version-min=10.15
-	LDFLAGS += -arch $(arch) -mmacosx-version-min=10.15
-endif
 
 ### Travis CI script uses COMPILER to overwrite CXX
 ifdef COMPILER
@@ -364,13 +426,26 @@ 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 ($(COMP),ndk)
+				LDFLAGS += -lpthread
+			endif
 		endif
 	endif
 endif
@@ -404,6 +479,10 @@ ifeq ($(optimize),yes)
 			CXXFLAGS += -mdynamic-no-pic
 		endif
 	endif
+
+	ifeq ($(comp),clang)
+		CXXFLAGS += -fexperimental-new-pass-manager
+	endif
 endif
 
 ### 3.4 Bits
@@ -415,7 +494,6 @@ endif
 ifeq ($(prefetch),yes)
 	ifeq ($(sse),yes)
 		CXXFLAGS += -msse
-		DEPENDFLAGS += -msse
 	endif
 else
 	CXXFLAGS += -DNO_PREFETCH
@@ -423,7 +501,7 @@ 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
@@ -432,6 +510,7 @@ ifeq ($(popcnt),yes)
 	endif
 endif
 
+
 ifeq ($(avx2),yes)
 	CXXFLAGS += -DUSE_AVX2
 	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
@@ -442,14 +521,21 @@ endif
 ifeq ($(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
 
@@ -467,19 +553,29 @@ 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
@@ -495,18 +591,40 @@ endif
 ### needs access to the optimization flags.
 ifeq ($(optimize),yes)
 ifeq ($(debug), no)
-	ifeq ($(comp),$(filter $(comp),gcc clang))
+	ifeq ($(comp),clang)
+		CXXFLAGS += -flto
+		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
 		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
@@ -523,6 +641,7 @@ endif
 ### Section 4. Public Targets
 ### ==========================================================================
 
+
 help:
 	@echo ""
 	@echo "To compile stockfish, type: "
@@ -531,31 +650,34 @@ 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"
@@ -566,27 +688,37 @@ 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 build ARCH=x86-64    (This is for 64-bit systems)"
-	@echo "make 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 build ARCH=x86-64 COMP=clang"
-	@echo "make profile-build ARCH=x86-64-bmi2 COMP=gcc COMPCXX=g++-4.8"
+	@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: net config-sanity
 	$(MAKE) ARCH=$(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)
@@ -602,23 +734,40 @@ 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:
@@ -627,7 +776,7 @@ 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:
@@ -639,7 +788,7 @@ default:
 
 all: $(EXE) .depend
 
-config-sanity:
+config-sanity: net
 	@echo ""
 	@echo "Config:"
 	@echo "debug: '$(debug)'"
@@ -651,14 +800,16 @@ config-sanity:
 	@echo "os: '$(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:"
@@ -671,25 +822,29 @@ 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) \
@@ -698,7 +853,7 @@ 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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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;
 
@@ -146,14 +148,23 @@ 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;
       }
 
+  list.emplace_back("setoption name Use NNUE value true");
+
   return list;
 }

src/bitbase.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/bitboard.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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/endgame.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -553,8 +553,8 @@ ScaleFactor Endgame<KRPPKRP>::operator()(const Position& pos) const {
   assert(verify_material(pos, strongSide, RookValueMg, 2));
   assert(verify_material(pos, weakSide,   RookValueMg, 1));
 
-  Square strongPawn1 = pos.squares<PAWN>(strongSide)[0];
-  Square strongPawn2 = pos.squares<PAWN>(strongSide)[1];
+  Square strongPawn1 = lsb(pos.pieces(strongSide, PAWN));
+  Square strongPawn2 = msb(pos.pieces(strongSide, PAWN));
   Square weakKing = pos.square<KING>(weakSide);
 
   // Does the stronger side have a passed pawn?
@@ -638,8 +638,8 @@ ScaleFactor Endgame<KBPPKB>::operator()(const Position& pos) const {
       return SCALE_FACTOR_NONE;
 
   Square weakKing = pos.square<KING>(weakSide);
-  Square strongPawn1 = pos.squares<PAWN>(strongSide)[0];
-  Square strongPawn2 = pos.squares<PAWN>(strongSide)[1];
+  Square strongPawn1 = lsb(pos.pieces(strongSide, PAWN));
+  Square strongPawn2 = msb(pos.pieces(strongSide, PAWN));
   Square blockSq1, blockSq2;
 
   if (relative_rank(strongSide, strongPawn1) > relative_rank(strongSide, strongPawn2))

src/endgame.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/evaluate.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -20,46 +20,127 @@
 #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 efficiently updatable neural network (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 Microsoft 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";
 
-  void init_NNUE() {
+  /// NNUE::init() 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 NNUE::init() {
 
     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;
+            }
+        }
   }
 
-  void verify_NNUE() {
+  /// NNUE::verify() verifies that the last net used was loaded successfully
+  void NNUE::verify() {
+
+    string eval_file = string(Options["EvalFile"]);
 
-    std::string eval_file = std::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 +188,11 @@ using namespace Trace;
 namespace {
 
   // Threshold for lazy and space evaluation
-  constexpr Value LazyThreshold1  = Value(1400);
-  constexpr Value LazyThreshold2  = Value(1300);
-  constexpr Value SpaceThreshold = Value(12222);
+  constexpr Value LazyThreshold1 =  Value(1565);
+  constexpr Value LazyThreshold2 =  Value(1102);
+  constexpr Value SpaceThreshold = Value(11551);
+  constexpr Value NNUEThreshold1 =   Value(682);
+  constexpr Value NNUEThreshold2 =   Value(176);
 
   // KingAttackWeights[PieceType] contains king attack weights by piece type
   constexpr int KingAttackWeights[PIECE_TYPE_NB] = { 0, 0, 81, 52, 44, 10 };
@@ -117,7 +200,7 @@ namespace {
   // SafeCheck[PieceType][single/multiple] contains safe check bonus by piece type,
   // higher if multiple safe checks are possible for that piece type.
   constexpr int SafeCheck[][2] = {
-      {}, {}, {792, 1283}, {645, 967}, {1084, 1897}, {772, 1119}
+      {}, {}, {803, 1292}, {639, 974}, {1087, 1878}, {759, 1132}
   };
 
 #define S(mg, eg) make_score(mg, eg)
@@ -125,19 +208,25 @@ namespace {
   // MobilityBonus[PieceType-2][attacked] contains bonuses for middle and end game,
   // indexed by piece type and number of attacked squares in the mobility area.
   constexpr Score MobilityBonus[][32] = {
-    { S(-62,-81), S(-53,-56), S(-12,-31), S( -4,-16), S(  3,  5), S( 13, 11), // Knight
-      S( 22, 17), S( 28, 20), S( 33, 25) },
-    { S(-48,-59), S(-20,-23), S( 16, -3), S( 26, 13), S( 38, 24), S( 51, 42), // Bishop
-      S( 55, 54), S( 63, 57), S( 63, 65), S( 68, 73), S( 81, 78), S( 81, 86),
-      S( 91, 88), S( 98, 97) },
-    { S(-60,-78), S(-20,-17), S(  2, 23), S(  3, 39), S(  3, 70), S( 11, 99), // Rook
-      S( 22,103), S( 31,121), S( 40,134), S( 40,139), S( 41,158), S( 48,164),
-      S( 57,168), S( 57,169), S( 62,172) },
-    { S(-30,-48), S(-12,-30), S( -8, -7), S( -9, 19), S( 20, 40), S( 23, 55), // Queen
-      S( 23, 59), S( 35, 75), S( 38, 78), S( 53, 96), S( 64, 96), S( 65,100),
-      S( 65,121), S( 66,127), S( 67,131), S( 67,133), S( 72,136), S( 72,141),
-      S( 77,147), S( 79,150), S( 93,151), S(108,168), S(108,168), S(108,171),
-      S(110,182), S(114,182), S(114,192), S(116,219) }
+    { S(-62,-79), S(-53,-57), S(-12,-31), S( -3,-17), S(  3,  7), S( 12, 13), // Knight
+      S( 21, 16), S( 28, 21), S( 37, 26) },
+    { S(-47,-59), S(-20,-25), S( 14, -8), S( 29, 12), S( 39, 21), S( 53, 40), // Bishop
+      S( 53, 56), S( 60, 58), S( 62, 65), S( 69, 72), S( 78, 78), S( 83, 87),
+      S( 91, 88), S( 96, 98) },
+    { S(-60,-82), S(-24,-15), S(  0, 17) ,S(  3, 43), S(  4, 72), S( 14,100), // Rook
+      S( 20,102), S( 30,122), S( 41,133), S(41 ,139), S( 41,153), S( 45,160),
+      S( 57,165), S( 58,170), S( 67,175) },
+    { S(-29,-49), S(-16,-29), S( -8, -8), S( -8, 17), S( 18, 39), S( 25, 54), // Queen
+      S( 23, 59), S( 37, 73), S( 41, 76), S( 54, 95), S( 65, 95) ,S( 68,101),
+      S( 69,124), S( 70,128), S( 70,132), S( 70,133) ,S( 71,136), S( 72,140),
+      S( 74,147), S( 76,149), S( 90,153), S(104,169), S(105,171), S(106,171),
+      S(112,178), S(114,185), S(114,187), S(119,221) }
+  };
+
+  // BishopPawns[distance from edge] contains a file-dependent penalty for pawns on
+  // squares of the same color as our bishop.
+  constexpr Score BishopPawns[int(FILE_NB) / 2] = {
+    S(3, 8), S(3, 9), S(2, 8), S(3, 8)
   };
 
   // KingProtector[knight/bishop] contains penalty for each distance unit to own king
@@ -145,32 +234,30 @@ 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(57, 38), S(31, 24) };
 
   // 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(7, 27), S(16, 32), S(17, 40), S(64, 71), S(170, 174), S(278, 262)
   };
 
-  // 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 RookOnClosedFile = S(10, 5);
+  constexpr Score RookOnOpenFile[] = { S(19, 6), S(47, 26) };
 
   // 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
-  constexpr Score BadOutpost          = S( -7, 36);
+  constexpr Score UncontestedOutpost  = S(  1, 10);
   constexpr Score BishopOnKingRing    = S( 24,  0);
-  constexpr Score BishopPawns         = S(  3,  7);
   constexpr Score BishopXRayPawns     = S(  4,  5);
   constexpr Score CorneredBishop      = S( 50, 50);
   constexpr Score FlankAttacks        = S(  8,  0);
@@ -180,11 +267,9 @@ 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);
-  constexpr Score RookOnQueenFile     = S(  6, 11);
   constexpr Score SliderOnQueen       = S( 60, 18);
   constexpr Score ThreatByKing        = S( 24, 89);
   constexpr Score ThreatByPawnPush    = S( 48, 39);
@@ -282,8 +367,8 @@ 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));
@@ -303,15 +388,15 @@ namespace {
     constexpr Direction Down = -pawn_push(Us);
     constexpr Bitboard OutpostRanks = (Us == WHITE ? Rank4BB | Rank5BB | Rank6BB
                                                    : Rank5BB | Rank4BB | Rank3BB);
-    const Square* pl = pos.squares<Pt>(Us);
-
+    Bitboard b1 = pos.pieces(Us, Pt);
     Bitboard b, bb;
     Score score = SCORE_ZERO;
 
     attackedBy[Us][Pt] = 0;
 
-    for (Square s = *pl; s != SQ_NONE; s = *++pl)
-    {
+    while (b1) {
+        Square s = pop_lsb(&b1);
+
         // Find attacked squares, including x-ray attacks for bishops and rooks
         b = Pt == BISHOP ? attacks_bb<BISHOP>(s, pos.pieces() ^ pos.pieces(QUEEN))
           : Pt ==   ROOK ? attacks_bb<  ROOK>(s, pos.pieces() ^ pos.pieces(QUEEN) ^ pos.pieces(Us, ROOK))
@@ -338,21 +423,21 @@ namespace {
             score += BishopOnKingRing;
 
         int mob = popcount(b & mobilityArea[Us]);
-
         mobility[Us] += MobilityBonus[Pt - 2][mob];
 
         if (Pt == BISHOP || Pt == KNIGHT)
         {
             // 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);
+            // Bonus for knights (UncontestedOutpost) if few relevant targets
+            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
                 && bb & s & ~CenterFiles // on a side outpost
                 && !(b & targets)        // no relevant attacks
                 && (!more_than_one(targets & (s & QueenSide ? QueenSide : KingSide))))
-                score += BadOutpost;
+                score += UncontestedOutpost * popcount(pos.pieces(PAWN) & (s & QueenSide ? QueenSide : KingSide));
             else if (bb & s)
                 score += Outpost[Pt == BISHOP];
             else if (Pt == KNIGHT && bb & b & ~pos.pieces(Us))
@@ -365,14 +450,14 @@ namespace {
             // Penalty if the piece is far from the king
             score -= KingProtector[Pt == BISHOP] * distance(pos.square<KING>(Us), s);
 
-            if (Pt == BISHOP)
+            if constexpr (Pt == BISHOP)
             {
                 // Penalty according to the number of our pawns on the same color square as the
                 // bishop, bigger when the center files are blocked with pawns and smaller
                 // when the bishop is outside the pawn chain.
                 Bitboard blocked = pos.pieces(Us, PAWN) & shift<Down>(pos.pieces());
 
-                score -= BishopPawns * pos.pawns_on_same_color_squares(Us, s)
+                score -= BishopPawns[edge_distance(file_of(s))] * pos.pawns_on_same_color_squares(Us, s)
                                      * (!(attackedBy[Us][PAWN] & s) + popcount(blocked & CenterFiles));
 
                 // Penalty for all enemy pawns x-rayed
@@ -397,38 +482,42 @@ namespace {
             }
         }
 
-        if (Pt == ROOK)
+        if constexpr (Pt == ROOK)
         {
-            // Bonus for rook on the same file as a queen
-            if (file_bb(s) & pos.pieces(QUEEN))
-                score += RookOnQueenFile;
-
-            // Bonus for rook on an open or semi-open file
+            // Bonuses for rook on a (semi-)open or closed file
             if (pos.is_on_semiopen_file(Us, s))
-                score += RookOnFile[pos.is_on_semiopen_file(Them, s)];
-
-            // Penalty when trapped by the king, even more if the king cannot castle
-            else if (mob <= 3)
             {
-                File kf = file_of(pos.square<KING>(Us));
-                if ((kf < FILE_E) == (file_of(s) < kf))
-                    score -= TrappedRook * (1 + !pos.castling_rights(Us));
+                score += RookOnOpenFile[pos.is_on_semiopen_file(Them, s)];
+            }
+            else
+            {
+                // If our pawn on this file is blocked, increase penalty
+                if ( pos.pieces(Us, PAWN)
+                   & shift<Down>(pos.pieces())
+                   & file_bb(s))
+                {
+                    score -= RookOnClosedFile;
+                }
+
+                // Penalty when trapped by the king, even more if the king cannot castle
+                if (mob <= 3)
+                {
+                    File kf = file_of(pos.square<KING>(Us));
+                    if ((kf < FILE_E) == (file_of(s) < kf))
+                        score -= TrappedRook * (1 + !pos.castling_rights(Us));
+                }
             }
         }
 
-        if (Pt == QUEEN)
+        if constexpr (Pt == QUEEN)
         {
             // Penalty if any relative pin or discovered attack against the queen
             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)
+    if constexpr (T)
         Trace::add(Pt, Us, score);
 
     return score;
@@ -504,18 +593,18 @@ namespace {
     int kingFlankAttack  = popcount(b1) + popcount(b2);
     int kingFlankDefense = popcount(b3);
 
-    kingDanger +=        kingAttackersCount[Them] * kingAttackersWeight[Them]
-                 + 185 * popcount(kingRing[Us] & weak)
-                 + 148 * popcount(unsafeChecks)
-                 +  98 * popcount(pos.blockers_for_king(Us))
-                 +  69 * kingAttacksCount[Them]
-                 +   3 * kingFlankAttack * kingFlankAttack / 8
-                 +       mg_value(mobility[Them] - mobility[Us])
-                 - 873 * !pos.count<QUEEN>(Them)
-                 - 100 * bool(attackedBy[Us][KNIGHT] & attackedBy[Us][KING])
-                 -   6 * mg_value(score) / 8
-                 -   4 * kingFlankDefense
-                 +  37;
+    kingDanger +=        kingAttackersCount[Them] * kingAttackersWeight[Them] // (~10 Elo)
+                 + 183 * popcount(kingRing[Us] & weak)                        // (~15 Elo)
+                 + 148 * popcount(unsafeChecks)                               // (~4 Elo)
+                 +  98 * popcount(pos.blockers_for_king(Us))                  // (~2 Elo)
+                 +  69 * kingAttacksCount[Them]                               // (~0.5 Elo)
+                 +   3 * kingFlankAttack * kingFlankAttack / 8                // (~0.5 Elo)
+                 +       mg_value(mobility[Them] - mobility[Us])              // (~0.5 Elo)
+                 - 873 * !pos.count<QUEEN>(Them)                              // (~24 Elo)
+                 - 100 * bool(attackedBy[Us][KNIGHT] & attackedBy[Us][KING])  // (~5 Elo)
+                 -   6 * mg_value(score) / 8                                  // (~8 Elo)
+                 -   4 * kingFlankDefense                                     // (~5 Elo)
+                 +  37;                                                       // (~0.5 Elo)
 
     // Transform the kingDanger units into a Score, and subtract it from the evaluation
     if (kingDanger > 100)
@@ -528,7 +617,7 @@ namespace {
     // Penalty if king flank is under attack, potentially moving toward the king
     score -= FlankAttacks * kingFlankAttack;
 
-    if (T)
+    if constexpr (T)
         Trace::add(KING, Us, score);
 
     return score;
@@ -629,7 +718,7 @@ namespace {
         score += SliderOnQueen * popcount(b & safe & attackedBy2[Us]) * (1 + queenImbalance);
     }
 
-    if (T)
+    if constexpr (T)
         Trace::add(THREAT, Us, score);
 
     return score;
@@ -683,8 +772,8 @@ 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)
@@ -699,14 +788,16 @@ namespace {
                 bb = forward_file_bb(Them, s) & pos.pieces(ROOK, QUEEN);
 
                 if (!(pos.pieces(Them) & bb))
-                    unsafeSquares &= attackedBy[Them][ALL_PIECES];
+                    unsafeSquares &= attackedBy[Them][ALL_PIECES] | pos.pieces(Them);
 
-                // If there are no enemy attacks on passed pawn span, assign a big bonus.
+                // If there are no enemy pieces or attacks on passed pawn span, assign a big bonus.
+                // Or if there is some, but they are all attacked by our pawns, assign a bit smaller bonus.
                 // Otherwise assign a smaller bonus if the path to queen is not attacked
                 // and even smaller bonus if it is attacked but block square is not.
-                int k = !unsafeSquares                    ? 35 :
-                        !(unsafeSquares & squaresToQueen) ? 20 :
-                        !(unsafeSquares & blockSq)        ?  9 :
+                int k = !unsafeSquares                    ? 36 :
+                !(unsafeSquares & ~attackedBy[Us][PAWN])  ? 30 :
+                        !(unsafeSquares & squaresToQueen) ? 17 :
+                        !(unsafeSquares & blockSq)        ?  7 :
                                                              0 ;
 
                 // Assign a larger bonus if the block square is defended
@@ -720,7 +811,7 @@ namespace {
         score += bonus - PassedFile * edge_distance(file_of(s));
     }
 
-    if (T)
+    if constexpr (T)
         Trace::add(PASSED, Us, score);
 
     return score;
@@ -728,7 +819,7 @@ 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.
 
@@ -755,11 +846,13 @@ namespace {
     behind |= shift<Down>(behind);
     behind |= shift<Down+Down>(behind);
 
+    // Compute space score based on the number of safe squares and number of our pieces
+    // increased with number of total blocked pawns in position.
     int bonus = popcount(safe) + popcount(behind & safe & ~attackedBy[Them][ALL_PIECES]);
     int weight = pos.count<ALL_PIECES>(Us) - 3 + std::min(pe->blocked_count(), 9);
     Score score = make_score(bonus * weight * weight / 16, 0);
 
-    if (T)
+    if constexpr (T)
         Trace::add(SPACE, Us, score);
 
     return score;
@@ -774,7 +867,7 @@ namespace {
   Value Evaluation<T>::winnable(Score score) const {
 
     int outflanking =  distance<File>(pos.square<KING>(WHITE), pos.square<KING>(BLACK))
-                     - distance<Rank>(pos.square<KING>(WHITE), pos.square<KING>(BLACK));
+                    + int(rank_of(pos.square<KING>(WHITE)) - rank_of(pos.square<KING>(BLACK)));
 
     bool pawnsOnBothFlanks =   (pos.pieces(PAWN) & QueenSide)
                             && (pos.pieces(PAWN) & KingSide);
@@ -801,7 +894,7 @@ 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;
@@ -816,23 +909,37 @@ namespace {
     {
         if (pos.opposite_bishops())
         {
+            // For pure opposite colored bishops endgames use scale factor
+            // based on the number of passed pawns of the strong side.
             if (   pos.non_pawn_material(WHITE) == BishopValueMg
                 && pos.non_pawn_material(BLACK) == BishopValueMg)
                 sf = 18 + 4 * popcount(pe->passed_pawns(strongSide));
+            // For every other opposite colored bishops endgames use scale factor
+            // based on the number of all pieces of the strong side.
             else
                 sf = 22 + 3 * pos.count<ALL_PIECES>(strongSide);
         }
+        // For rook endgames with strong side not having overwhelming pawn number advantage
+        // and its pawns being on one flank and weak side protecting its pieces with a king
+        // use lower scale factor.
         else if (  pos.non_pawn_material(WHITE) == RookValueMg
                 && pos.non_pawn_material(BLACK) == RookValueMg
                 && pos.count<PAWN>(strongSide) - pos.count<PAWN>(~strongSide) <= 1
                 && bool(KingSide & pos.pieces(strongSide, PAWN)) != bool(QueenSide & pos.pieces(strongSide, PAWN))
                 && (attacks_bb<KING>(pos.square<KING>(~strongSide)) & pos.pieces(~strongSide, PAWN)))
             sf = 36;
+        // For queen vs no queen endgames use scale factor
+        // based on number of minors of side that doesn't have queen.
         else if (pos.count<QUEEN>() == 1)
             sf = 37 + 3 * (pos.count<QUEEN>(WHITE) == 1 ? pos.count<BISHOP>(BLACK) + pos.count<KNIGHT>(BLACK)
                                                         : pos.count<BISHOP>(WHITE) + pos.count<KNIGHT>(WHITE));
+        // In every other case use scale factor based on
+        // the number of pawns of the strong side reduced if pawns are on a single flank.
         else
-            sf = std::min(sf, 36 + 7 * pos.count<PAWN>(strongSide));
+            sf = std::min(sf, 36 + 7 * pos.count<PAWN>(strongSide)) - 4 * !pawnsOnBothFlanks;
+
+        // Reduce scale factor in case of pawns being on a single flank
+        sf -= 4 * !pawnsOnBothFlanks;
     }
 
     // Interpolate between the middlegame and (scaled by 'sf') endgame score
@@ -840,7 +947,7 @@ namespace {
        + eg * int(PHASE_MIDGAME - me->game_phase()) * ScaleFactor(sf) / SCALE_FACTOR_NORMAL;
     v /= PHASE_MIDGAME;
 
-    if (T)
+    if constexpr (T)
     {
         Trace::add(WINNABLE, make_score(u, eg * ScaleFactor(sf) / SCALE_FACTOR_NORMAL - eg_value(score)));
         Trace::add(TOTAL, make_score(mg, eg * ScaleFactor(sf) / SCALE_FACTOR_NORMAL));
@@ -912,7 +1019,7 @@ make_v:
     Value v = winnable(score);
 
     // In case of tracing add all remaining individual evaluation terms
-    if (T)
+    if constexpr (T)
     {
         Trace::add(MATERIAL, pos.psq_score());
         Trace::add(IMBALANCE, me->imbalance());
@@ -926,9 +1033,6 @@ 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 +1044,48 @@ make_v:
 
 Value Eval::evaluate(const Position& pos) {
 
-  if (Eval::useNNUE)
-      return NNUE::evaluate(pos);
+  Value v;
+
+  if (!Eval::useNNUE)
+      v = Evaluation<NO_TRACE>(pos).value();
   else
-      return Evaluation<NO_TRACE>(pos).value();
+  {
+      // Scale and shift NNUE for compatibility with search and classical evaluation
+      auto  adjusted_NNUE = [&](){
+         int mat = pos.non_pawn_material() + 2 * PawnValueMg * pos.count<PAWN>();
+         return NNUE::evaluate(pos) * (641 + mat / 32 - 4 * pos.rule50_count()) / 1024 + Tempo;
+      };
+
+      // If there is PSQ imbalance use classical eval, with small probability if it is small
+      Value psq = Value(abs(eg_value(pos.psq_score())));
+      int   r50 = 16 + pos.rule50_count();
+      bool  largePsq = psq * 16 > (NNUEThreshold1 + pos.non_pawn_material() / 64) * r50;
+      bool  classical = largePsq || (psq > PawnValueMg / 4 && !(pos.this_thread()->nodes & 0xB));
+
+      // Use classical evaluation for really low piece endgames.
+      // The most critical case is a bishop + A/H file pawn vs naked king draw.
+      bool strongClassical = pos.non_pawn_material() < 2 * RookValueMg && pos.count<PAWN>() < 2;
+
+      v = classical || strongClassical ? Evaluation<NO_TRACE>(pos).value() : adjusted_NNUE();
+
+      // If the classical eval is small and imbalance large, use NNUE nevertheless.
+      // For the case of opposite colored bishops, switch to NNUE eval with
+      // small probability if the classical eval is less than the threshold.
+      if (   largePsq && !strongClassical
+          && (   abs(v) * 16 < NNUEThreshold2 * r50
+              || (   pos.opposite_bishops()
+                  && abs(v) * 16 < (NNUEThreshold1 + pos.non_pawn_material() / 64) * r50
+                  && !(pos.this_thread()->nodes & 0xB))))
+          v = adjusted_NNUE();
+  }
+
+  // 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 +1103,46 @@ 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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -32,15 +32,18 @@ namespace Eval {
 
   extern bool useNNUE;
   extern std::string eval_file_loaded;
-  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-62ef826d1a6d.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 name, std::istream& stream);
+    void init();
+    void verify();
 
   } // 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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,20 +21,18 @@
 #include "bitboard.h"
 #include "endgame.h"
 #include "position.h"
+#include "psqt.h"
 #include "search.h"
+#include "syzygy/tbprobe.h"
 #include "thread.h"
 #include "tt.h"
 #include "uci.h"
-#include "syzygy/tbprobe.h"
-
-namespace PSQT {
-  void init();
-}
 
 int main(int argc, char* argv[]) {
 
   std::cout << engine_info() << std::endl;
 
+  CommandLine::init(argc, argv);
   UCI::init(Options);
   Tune::init();
   PSQT::init();
@@ -44,7 +42,7 @@ int main(int argc, char* argv[]) {
   Endgames::init();
   Threads.set(size_t(Options["Threads"]));
   Search::clear(); // After threads are up
-  Eval::init_NNUE();
+  Eval::NNUE::init();
 
   UCI::loop(argc, argv);
 

src/material.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -25,31 +25,36 @@
 using namespace std;
 
 namespace {
+  #define S(mg, eg) make_score(mg, eg)
 
   // Polynomial material imbalance parameters
 
-  constexpr int QuadraticOurs[][PIECE_TYPE_NB] = {
-    //            OUR PIECES
-    // pair pawn knight bishop rook queen
-    {1438                               }, // Bishop pair
-    {  40,   38                         }, // Pawn
-    {  32,  255, -62                    }, // Knight      OUR PIECES
-    {   0,  104,   4,    0              }, // Bishop
-    { -26,   -2,  47,   105,  -208      }, // Rook
-    {-189,   24, 117,   133,  -134, -6  }  // Queen
+  // One Score parameter for each pair (our piece, another of our pieces)
+  constexpr Score QuadraticOurs[][PIECE_TYPE_NB] = {
+    // OUR PIECE 2
+    // bishop pair    pawn         knight       bishop       rook           queen
+    {S(1419, 1455)                                                                  }, // Bishop pair
+    {S( 101,   28), S( 37,  39)                                                     }, // Pawn
+    {S(  57,   64), S(249, 187), S(-49, -62)                                        }, // Knight      OUR PIECE 1
+    {S(   0,    0), S(118, 137), S( 10,  27), S(  0,   0)                           }, // Bishop
+    {S( -63,  -68), S( -5,   3), S(100,  81), S(132, 118), S(-246, -244)            }, // Rook
+    {S(-210, -211), S( 37,  14), S(147, 141), S(161, 105), S(-158, -174), S(-9,-31) }  // Queen
   };
 
-  constexpr int QuadraticTheirs[][PIECE_TYPE_NB] = {
-    //           THEIR PIECES
-    // pair pawn knight bishop rook queen
-    {                                   }, // Bishop pair
-    {  36,                              }, // Pawn
-    {   9,   63,                        }, // Knight      OUR PIECES
-    {  59,   65,  42,                   }, // Bishop
-    {  46,   39,  24,   -24,            }, // Rook
-    {  97,  100, -42,   137,  268,      }  // Queen
+  // One Score parameter for each pair (our piece, their piece)
+  constexpr Score QuadraticTheirs[][PIECE_TYPE_NB] = {
+    // THEIR PIECE
+    // bishop pair   pawn         knight       bishop       rook         queen
+    {                                                                               }, // Bishop pair
+    {S(  33,  30)                                                                   }, // Pawn
+    {S(  46,  18), S(106,  84)                                                      }, // Knight      OUR PIECE
+    {S(  75,  35), S( 59,  44), S( 60,  15)                                         }, // Bishop
+    {S(  26,  35), S(  6,  22), S( 38,  39), S(-12,  -2)                            }, // Rook
+    {S(  97,  93), S(100, 163), S(-58, -91), S(112, 192), S(276, 225)               }  // Queen
   };
 
+  #undef S
+
   // Endgame evaluation and scaling functions are accessed directly and not through
   // the function maps because they correspond to more than one material hash key.
   Endgame<KXK>    EvaluateKXK[] = { Endgame<KXK>(WHITE),    Endgame<KXK>(BLACK) };
@@ -82,11 +87,11 @@ namespace {
   /// piece type for both colors.
 
   template<Color Us>
-  int imbalance(const int pieceCount[][PIECE_TYPE_NB]) {
+  Score imbalance(const int pieceCount[][PIECE_TYPE_NB]) {
 
     constexpr Color Them = ~Us;
 
-    int bonus = 0;
+    Score bonus = SCORE_ZERO;
 
     // Second-degree polynomial material imbalance, by Tord Romstad
     for (int pt1 = NO_PIECE_TYPE; pt1 <= QUEEN; ++pt1)
@@ -130,7 +135,7 @@ 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));
@@ -213,7 +218,7 @@ Entry* probe(const Position& pos) {
   { pos.count<BISHOP>(BLACK) > 1, pos.count<PAWN>(BLACK), pos.count<KNIGHT>(BLACK),
     pos.count<BISHOP>(BLACK)    , pos.count<ROOK>(BLACK), pos.count<QUEEN >(BLACK) } };
 
-  e->value = int16_t((imbalance<WHITE>(pieceCount) - imbalance<BLACK>(pieceCount)) / 16);
+  e->score = (imbalance<WHITE>(pieceCount) - imbalance<BLACK>(pieceCount)) / 16;
   return e;
 }
 

src/material.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -37,8 +37,8 @@ namespace Material {
 
 struct Entry {
 
-  Score imbalance() const { return make_score(value, value); }
-  Phase game_phase() const { return gamePhase; }
+  Score imbalance() const { return score; }
+  Phase game_phase() const { return (Phase)gamePhase; }
   bool specialized_eval_exists() const { return evaluationFunction != nullptr; }
   Value evaluate(const Position& pos) const { return (*evaluationFunction)(pos); }
 
@@ -57,9 +57,9 @@ struct Entry {
   const EndgameBase<Value>* evaluationFunction;
   const EndgameBase<ScaleFactor>* scalingFunction[COLOR_NB]; // Could be one for each
                                                              // side (e.g. KPKP, KBPsK)
-  int16_t value;
+  Score score;
+  int16_t gamePhase;
   uint8_t factor[COLOR_NB];
-  Phase gamePhase;
 };
 
 typedef HashTable<Entry, 8192> Table;

src/misc.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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 = "13";
 
 /// 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);
@@ -340,25 +357,11 @@ void std_aligned_free(void* ptr) {
 #endif
 }
 
-/// aligned_ttmem_alloc() will return suitably aligned memory, and if possible use 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.
+/// aligned_large_pages_alloc() will return suitably aligned memory, if possible using large pages.
 
-#if defined(__linux__) && !defined(__ANDROID__)
+#if defined(_WIN32)
 
-void* aligned_ttmem_alloc(size_t allocSize, void*& mem) {
-
-  constexpr size_t alignment = 2 * 1024 * 1024; // assumed 2MB page sizes
-  size_t size = ((allocSize + alignment - 1) / alignment) * alignment; // multiple of alignment
-  if (posix_memalign(&mem, alignment, size))
-     mem = nullptr;
-  madvise(mem, allocSize, MADV_HUGEPAGE);
-  return mem;
-}
-
-#elif defined(_WIN64)
-
-static void* aligned_ttmem_alloc_large_pages(size_t allocSize) {
+static void* aligned_large_pages_alloc_win(size_t allocSize) {
 
   HANDLE hProcessToken { };
   LUID luid { };
@@ -403,23 +406,10 @@ static void* aligned_ttmem_alloc_large_pages(size_t allocSize) {
   return mem;
 }
 
-void* aligned_ttmem_alloc(size_t allocSize, void*& mem) {
-
-  static bool firstCall = true;
+void* aligned_large_pages_alloc(size_t allocSize) {
 
   // Try to allocate large pages
-  mem = aligned_ttmem_alloc_large_pages(allocSize);
-
-  // Suppress info strings on the first call. The first call occurs before 'uci'
-  // is received and in that case this output confuses some GUIs.
-  if (!firstCall)
-  {
-      if (mem)
-          sync_cout << "info string Hash table allocation: Windows large pages used." << sync_endl;
-      else
-          sync_cout << "info string Hash table allocation: Windows large pages not used." << sync_endl;
-  }
-  firstCall = false;
+  void* mem = aligned_large_pages_alloc_win(allocSize);
 
   // Fall back to regular, page aligned, allocation if necessary
   if (!mem)
@@ -430,23 +420,31 @@ void* aligned_ttmem_alloc(size_t allocSize, void*& mem) {
 
 #else
 
-void* aligned_ttmem_alloc(size_t allocSize, void*& mem) {
+void* aligned_large_pages_alloc(size_t allocSize) {
 
-  constexpr size_t alignment = 64; // assumed cache line size
-  size_t size = allocSize + alignment - 1; // allocate some extra space
-  mem = malloc(size);
-  void* ret = reinterpret_cast<void*>((uintptr_t(mem) + alignment - 1) & ~uintptr_t(alignment - 1));
-  return ret;
+#if defined(__linux__)
+  constexpr size_t alignment = 2 * 1024 * 1024; // assumed 2MB page size
+#else
+  constexpr size_t alignment = 4096; // assumed small page size
+#endif
+
+  // round up to multiples of alignment
+  size_t size = ((allocSize + alignment - 1) / alignment) * alignment;
+  void *mem = std_aligned_alloc(alignment, size);
+#if defined(MADV_HUGEPAGE)
+  madvise(mem, size, MADV_HUGEPAGE);
+#endif
+  return mem;
 }
 
 #endif
 
 
-/// aligned_ttmem_free() will free the previously allocated ttmem
+/// aligned_large_pages_free() will free the previously allocated ttmem
 
-#if defined(_WIN64)
+#if defined(_WIN32)
 
-void aligned_ttmem_free(void* mem) {
+void aligned_large_pages_free(void* mem) {
 
   if (mem && !VirtualFree(mem, 0, MEM_RELEASE))
   {
@@ -459,8 +457,8 @@ void aligned_ttmem_free(void* mem) {
 
 #else
 
-void aligned_ttmem_free(void *mem) {
-  free(mem);
+void aligned_large_pages_free(void *mem) {
+  std_aligned_free(mem);
 }
 
 #endif
@@ -571,3 +569,60 @@ 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
+
+void init(int argc, char* argv[]) {
+    (void)argc;
+    string pathSeparator;
+
+    // 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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -24,6 +24,7 @@
 #include <ostream>
 #include <string>
 #include <vector>
+#include <cstdint>
 
 #include "types.h"
 
@@ -33,8 +34,8 @@ void prefetch(void* addr);
 void start_logger(const std::string& fname);
 void* std_aligned_alloc(size_t alignment, size_t size);
 void std_aligned_free(void* ptr);
-void* aligned_ttmem_alloc(size_t size, void*& mem);
-void aligned_ttmem_free(void* mem); // nop if mem == nullptr
+void* aligned_large_pages_alloc(size_t size); // memory aligned by page size, min alignment: 4096 bytes
+void aligned_large_pages_free(void* mem); // nop if mem == nullptr
 
 void dbg_hit_on(bool b);
 void dbg_hit_on(bool c, bool b);
@@ -42,9 +43,7 @@ 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();
@@ -65,13 +64,16 @@ 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;
-}
+// `ptr` must point to an array of size at least
+// `sizeof(T) * N + alignment` bytes, where `N` is the
+// number of elements in the array.
+template <uintptr_t Alignment, typename T>
+T* align_ptr_up(T* ptr)
+{
+  static_assert(alignof(T) < Alignment);
 
+  const uintptr_t ptrint = reinterpret_cast<uintptr_t>(reinterpret_cast<char*>(ptr));
+  return reinterpret_cast<T*>(reinterpret_cast<char*>((ptrint + (Alignment - 1)) / Alignment * Alignment));
 }
 
 /// xorshift64star Pseudo-Random Number Generator
@@ -134,4 +136,11 @@ 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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -85,7 +85,7 @@ namespace {
 
             // Add pawn pushes which give discovered check. This is possible only
             // if the pawn is not on the same file as the enemy king, because we
-            // don't generate captures. Note that a possible discovery check
+            // don't generate captures. Note that a possible discovered check
             // promotion has been already generated amongst the captures.
             Bitboard dcCandidateQuiets = pos.blockers_for_king(Them) & pawnsNotOn7;
             if (dcCandidateQuiets)
@@ -134,7 +134,7 @@ namespace {
             moveList = make_promotions<Type, Up     >(moveList, pop_lsb(&b3), ksq);
     }
 
-    // Standard and en-passant captures
+    // Standard and en passant captures
     if (Type == CAPTURES || Type == EVASIONS || Type == NON_EVASIONS)
     {
         Bitboard b1 = shift<UpRight>(pawnsNotOn7) & enemies;
@@ -156,10 +156,8 @@ namespace {
         {
             assert(rank_of(pos.ep_square()) == relative_rank(Us, RANK_6));
 
-            // An en passant capture can be an evasion only if the checking piece
-            // is the double pushed pawn and so is in the target. Otherwise this
-            // is a discovery check and we are forced to do otherwise.
-            if (Type == EVASIONS && !(target & (pos.ep_square() - Up)))
+            // An en passant capture cannot resolve a discovered check.
+            if (Type == EVASIONS && (target & (pos.ep_square() + Up)))
                 return moveList;
 
             b1 = pawnsNotOn7 & pawn_attacks_bb(Them, pos.ep_square());
@@ -167,7 +165,7 @@ namespace {
             assert(b1);
 
             while (b1)
-                *moveList++ = make<ENPASSANT>(pop_lsb(&b1), pos.ep_square());
+                *moveList++ = make<EN_PASSANT>(pop_lsb(&b1), pos.ep_square());
         }
     }
 
@@ -175,29 +173,24 @@ namespace {
   }
 
 
-  template<Color Us, PieceType Pt, bool Checks>
-  ExtMove* generate_moves(const Position& pos, ExtMove* moveList, Bitboard target) {
+  template<PieceType Pt, bool Checks>
+  ExtMove* generate_moves(const Position& pos, ExtMove* moveList, Bitboard piecesToMove, Bitboard target) {
 
     static_assert(Pt != KING && Pt != PAWN, "Unsupported piece type in generate_moves()");
 
-    const Square* pl = pos.squares<Pt>(Us);
+    Bitboard bb = piecesToMove & pos.pieces(Pt);
 
-    for (Square from = *pl; from != SQ_NONE; from = *++pl)
-    {
-        if (Checks)
-        {
-            if (    (Pt == BISHOP || Pt == ROOK || Pt == QUEEN)
-                && !(attacks_bb<Pt>(from) & target & pos.check_squares(Pt)))
-                continue;
+    if (!bb)
+        return moveList;
 
-            if (pos.blockers_for_king(~Us) & from)
-                continue;
-        }
+    [[maybe_unused]] const Bitboard checkSquares = pos.check_squares(Pt);
+
+    while (bb) {
+        Square from = pop_lsb(&bb);
 
         Bitboard b = attacks_bb<Pt>(from, pos.pieces()) & target;
-
-        if (Checks)
-            b &= pos.check_squares(Pt);
+        if constexpr (Checks)
+            b &= checkSquares;
 
         while (b)
             *moveList++ = make_move(from, pop_lsb(&b));
@@ -209,8 +202,14 @@ namespace {
 
   template<Color Us, GenType Type>
   ExtMove* generate_all(const Position& pos, ExtMove* moveList) {
-    constexpr bool Checks = Type == QUIET_CHECKS; // Reduce template instantations
-    Bitboard target;
+
+    static_assert(Type != LEGAL, "Unsupported type in generate_all()");
+
+    constexpr bool Checks = Type == QUIET_CHECKS; // Reduce template instantiations
+    Bitboard target, piecesToMove = pos.pieces(Us);
+
+    if(Type == QUIET_CHECKS)
+        piecesToMove &= ~pos.blockers_for_king(~Us);
 
     switch (Type)
     {
@@ -230,15 +229,13 @@ namespace {
         case NON_EVASIONS:
             target = ~pos.pieces(Us);
             break;
-        default:
-            static_assert(true, "Unsupported type in generate_all()");
     }
 
     moveList = generate_pawn_moves<Us, Type>(pos, moveList, target);
-    moveList = generate_moves<Us, KNIGHT, Checks>(pos, moveList, target);
-    moveList = generate_moves<Us, BISHOP, Checks>(pos, moveList, target);
-    moveList = generate_moves<Us,   ROOK, Checks>(pos, moveList, target);
-    moveList = generate_moves<Us,  QUEEN, Checks>(pos, moveList, target);
+    moveList = generate_moves<KNIGHT, Checks>(pos, moveList, piecesToMove, target);
+    moveList = generate_moves<BISHOP, Checks>(pos, moveList, piecesToMove, target);
+    moveList = generate_moves<  ROOK, Checks>(pos, moveList, piecesToMove, target);
+    moveList = generate_moves< QUEEN, Checks>(pos, moveList, piecesToMove, target);
 
     if (Type != QUIET_CHECKS && Type != EVASIONS)
     {
@@ -248,7 +245,7 @@ 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));
     }
@@ -260,7 +257,7 @@ namespace {
 
 
 /// <CAPTURES>     Generates all pseudo-legal captures plus queen and checking knight promotions
-/// <QUIETS>       Generates all pseudo-legal non-captures and underpromotions(except checking knight)
+/// <QUIETS>       Generates all pseudo-legal non-captures and underpromotions (except checking knight)
 /// <NON_EVASIONS> Generates all pseudo-legal captures and non-captures
 ///
 /// Returns a pointer to the end of the move list.
@@ -283,8 +280,8 @@ template ExtMove* generate<QUIETS>(const Position&, ExtMove*);
 template ExtMove* generate<NON_EVASIONS>(const Position&, ExtMove*);
 
 
-/// generate<QUIET_CHECKS> generates all pseudo-legal non-captures.
-/// Returns a pointer to the end of the move list.
+/// generate<QUIET_CHECKS> generates all pseudo-legal non-captures giving check,
+/// except castling. Returns a pointer to the end of the move list.
 template<>
 ExtMove* generate<QUIET_CHECKS>(const Position& pos, ExtMove* moveList) {
 
@@ -357,7 +354,7 @@ ExtMove* generate<LEGAL>(const Position& pos, ExtMove* moveList) {
   moveList = pos.checkers() ? generate<EVASIONS    >(pos, moveList)
                             : generate<NON_EVASIONS>(pos, moveList);
   while (cur != moveList)
-      if (   (pinned || from_sq(*cur) == ksq || type_of(*cur) == ENPASSANT)
+      if (  ((pinned && pinned & from_sq(*cur)) || from_sq(*cur) == ksq || type_of(*cur) == EN_PASSANT)
           && !pos.legal(*cur))
           *cur = (--moveList)->move;
       else

src/movegen.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/movepick.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -73,8 +73,9 @@ MovePicker::MovePicker(const Position& p, Move ttm, Depth d, const ButterflyHist
   assert(d <= 0);
 
   stage = (pos.checkers() ? EVASION_TT : QSEARCH_TT) +
-           !(ttm && (depth > DEPTH_QS_RECAPTURES || to_sq(ttm) == recaptureSquare)
-                 && pos.pseudo_legal(ttm));
+          !(   ttm
+            && (pos.checkers() || depth > DEPTH_QS_RECAPTURES || to_sq(ttm) == recaptureSquare)
+            && pos.pseudo_legal(ttm));
 }
 
 /// MovePicker constructor for ProbCut: we generate captures with SEE greater
@@ -98,15 +99,15 @@ void MovePicker::score() {
   static_assert(Type == CAPTURES || Type == QUIETS || Type == EVASIONS, "Wrong type");
 
   for (auto& m : *this)
-      if (Type == CAPTURES)
+      if constexpr (Type == CAPTURES)
           m.value =  int(PieceValue[MG][pos.piece_on(to_sq(m))]) * 6
                    + (*captureHistory)[pos.moved_piece(m)][to_sq(m)][type_of(pos.piece_on(to_sq(m)))];
 
-      else if (Type == QUIETS)
+      else if constexpr (Type == QUIETS)
           m.value =      (*mainHistory)[pos.side_to_move()][from_to(m)]
                    + 2 * (*continuationHistory[0])[pos.moved_piece(m)][to_sq(m)]
-                   + 2 * (*continuationHistory[1])[pos.moved_piece(m)][to_sq(m)]
-                   + 2 * (*continuationHistory[3])[pos.moved_piece(m)][to_sq(m)]
+                   +     (*continuationHistory[1])[pos.moved_piece(m)][to_sq(m)]
+                   +     (*continuationHistory[3])[pos.moved_piece(m)][to_sq(m)]
                    +     (*continuationHistory[5])[pos.moved_piece(m)][to_sq(m)]
                    + (ply < MAX_LPH ? std::min(4, depth / 3) * (*lowPlyHistory)[ply][from_to(m)] : 0);
 
@@ -116,8 +117,8 @@ void MovePicker::score() {
               m.value =  PieceValue[MG][pos.piece_on(to_sq(m))]
                        - Value(type_of(pos.moved_piece(m)));
           else
-              m.value =  (*mainHistory)[pos.side_to_move()][from_to(m)]
-                       + (*continuationHistory[0])[pos.moved_piece(m)][to_sq(m)]
+              m.value =      (*mainHistory)[pos.side_to_move()][from_to(m)]
+                       + 2 * (*continuationHistory[0])[pos.moved_piece(m)][to_sq(m)]
                        - (1 << 28);
       }
 }
@@ -141,7 +142,7 @@ Move MovePicker::select(Pred filter) {
 }
 
 /// MovePicker::next_move() is the most important method of the MovePicker class. It
-/// returns a new pseudo legal move every time it is called until there are no more
+/// returns a new pseudo-legal move every time it is called until there are no more
 /// moves left, picking the move with the highest score from a list of generated moves.
 Move MovePicker::next_move(bool skipQuiets) {
 
@@ -182,7 +183,7 @@ top:
           --endMoves;
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case REFUTATION:
       if (select<Next>([&](){ return    *cur != MOVE_NONE
@@ -190,7 +191,7 @@ top:
                                     &&  pos.pseudo_legal(*cur); }))
           return *(cur - 1);
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case QUIET_INIT:
       if (!skipQuiets)
@@ -203,7 +204,7 @@ top:
       }
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case QUIET:
       if (   !skipQuiets
@@ -217,7 +218,7 @@ top:
       endMoves = endBadCaptures;
 
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case BAD_CAPTURE:
       return select<Next>([](){ return true; });
@@ -228,7 +229,7 @@ top:
 
       score<EVASIONS>();
       ++stage;
-      /* fallthrough */
+      [[fallthrough]];
 
   case EVASION:
       return select<Best>([](){ return true; });
@@ -246,14 +247,14 @@ 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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -84,11 +84,11 @@ enum StatsType { NoCaptures, Captures };
 /// unsuccessful during the current search, and is used for reduction and move
 /// ordering decisions. It uses 2 tables (one for each color) indexed by
 /// 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;
+typedef Stats<int16_t, 13365, 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;
 
@@ -108,12 +108,12 @@ typedef Stats<int16_t, 29952, PIECE_NB, SQUARE_NB> PieceToHistory;
 typedef Stats<PieceToHistory, NOT_USED, PIECE_NB, SQUARE_NB> ContinuationHistory;
 
 
-/// MovePicker class is used to pick one pseudo legal move at a time from the
+/// MovePicker class is used to pick one pseudo-legal move at a time from the
 /// current position. The most important method is next_move(), which returns a
-/// new pseudo legal move each time it is called, until there are no moves left,
-/// when MOVE_NONE is returned. In order to improve the efficiency of the alpha
-/// beta algorithm, MovePicker attempts to return the moves which are most likely
-/// to get a cut-off first.
+/// new pseudo-legal move each time it is called, until there are no moves left,
+/// when MOVE_NONE is returned. In order to improve the efficiency of the
+/// alpha-beta algorithm, MovePicker attempts to return the moves which are most
+/// likely to get a cut-off first.
 class MovePicker {
 
   enum PickType { Next, Best };

src/nnue/architectures/halfkp_256x2-32-32.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/nnue/evaluate_nnue.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -18,7 +18,6 @@
 
 // Code for calculating NNUE evaluation function
 
-#include <fstream>
 #include <iostream>
 #include <set>
 
@@ -26,35 +25,14 @@
 #include "../position.h"
 #include "../misc.h"
 #include "../uci.h"
+#include "../types.h"
 
 #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 {
 
   // Input feature converter
-  AlignedPtr<FeatureTransformer> feature_transformer;
+  LargePagePtr<FeatureTransformer> feature_transformer;
 
   // Evaluation function
   AlignedPtr<Network> network;
@@ -72,14 +50,22 @@ namespace Eval::NNUE {
     std::memset(pointer.get(), 0, sizeof(T));
   }
 
+  template <typename T>
+  void Initialize(LargePagePtr<T>& pointer) {
+
+    static_assert(alignof(T) <= 4096, "aligned_large_pages_alloc() may fail for such a big alignment requirement of T");
+    pointer.reset(reinterpret_cast<T*>(aligned_large_pages_alloc(sizeof(T))));
+    std::memset(pointer.get(), 0, sizeof(T));
+  }
+
   // Read evaluation function parameters
   template <typename T>
-  bool ReadParameters(std::istream& stream, const AlignedPtr<T>& pointer) {
+  bool ReadParameters(std::istream& stream, T& reference) {
 
     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);
+    return reference.ReadParameters(stream);
   }
 
   }  // namespace Detail
@@ -92,13 +78,13 @@ 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);
@@ -112,67 +98,47 @@ namespace Eval::NNUE {
     std::string architecture;
     if (!ReadHeader(stream, &hash_value, &architecture)) return false;
     if (hash_value != kHashValue) return false;
-    if (!Detail::ReadParameters(stream, feature_transformer)) return false;
-    if (!Detail::ReadParameters(stream, network)) return false;
+    if (!Detail::ReadParameters(stream, *feature_transformer)) return false;
+    if (!Detail::ReadParameters(stream, *network)) return false;
     return stream && stream.peek() == std::ios::traits_type::eof();
   }
 
-  // Proceed with the difference calculation if possible
-  static void UpdateAccumulatorIfPossible(const Position& pos) {
-
-    feature_transformer->UpdateAccumulatorIfPossible(pos);
-  }
-
-  // Calculate the evaluation value
-  static Value ComputeScore(const Position& pos, bool refresh) {
-
-    auto& accumulator = pos.state()->accumulator;
-    if (!refresh && accumulator.computed_score) {
-      return accumulator.score;
-    }
-
-    alignas(kCacheLineSize) TransformedFeatureType
-        transformed_features[FeatureTransformer::kBufferSize];
-    feature_transformer->Transform(pos, transformed_features, refresh);
-    alignas(kCacheLineSize) char buffer[Network::kBufferSize];
-    const auto output = network->Propagate(transformed_features, buffer);
-
-    auto score = static_cast<Value>(output[0] / FV_SCALE);
-
-    accumulator.score = score;
-    accumulator.computed_score = true;
-    return accumulator.score;
-  }
-
-  // Load the evaluation function file
-  bool load_eval_file(const std::string& evalFile) {
-
-    Initialize();
-    fileName = evalFile;
-
-    std::ifstream stream(evalFile, std::ios::binary);
-
-    const bool result = ReadParameters(stream);
-
-    return result;
-  }
-
   // 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;
+    // We manually align the arrays on the stack because with gcc < 9.3
+    // overaligning stack variables with alignas() doesn't work correctly.
+
+    constexpr uint64_t alignment = kCacheLineSize;
+
+#if defined(ALIGNAS_ON_STACK_VARIABLES_BROKEN)
+    TransformedFeatureType transformed_features_unaligned[
+      FeatureTransformer::kBufferSize + alignment / sizeof(TransformedFeatureType)];
+    char buffer_unaligned[Network::kBufferSize + alignment];
+
+    auto* transformed_features = align_ptr_up<alignment>(&transformed_features_unaligned[0]);
+    auto* buffer = align_ptr_up<alignment>(&buffer_unaligned[0]);
+#else
+    alignas(alignment)
+      TransformedFeatureType transformed_features[FeatureTransformer::kBufferSize];
+    alignas(alignment) char buffer[Network::kBufferSize];
+#endif
+
+    ASSERT_ALIGNED(transformed_features, alignment);
+    ASSERT_ALIGNED(buffer, alignment);
+
+    feature_transformer->Transform(pos, transformed_features);
+    const auto output = network->Propagate(transformed_features, buffer);
+
+    return static_cast<Value>(output[0] / FV_SCALE);
   }
 
-  // Evaluation function. Perform full calculation.
-  Value compute_eval(const Position& pos) {
-    return ComputeScore(pos, true);
-  }
+  // Load eval, from a file stream or a memory stream
+  bool load_eval(std::string name, std::istream& stream) {
 
-  // Proceed with the difference calculation if possible
-  void update_eval(const Position& pos) {
-    UpdateAccumulatorIfPossible(pos);
+    Initialize();
+    fileName = name;
+    return ReadParameters(stream);
   }
 
 } // namespace Eval::NNUE

src/nnue/evaluate_nnue.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -40,9 +40,20 @@ namespace Eval::NNUE {
     }
   };
 
+  template <typename T>
+  struct LargePageDeleter {
+    void operator()(T* ptr) const {
+      ptr->~T();
+      aligned_large_pages_free(ptr);
+    }
+  };
+
   template <typename T>
   using AlignedPtr = std::unique_ptr<T, AlignedDeleter<T>>;
 
+  template <typename T>
+  using LargePagePtr = std::unique_ptr<T, LargePageDeleter<T>>;
+
 }  // namespace Eval::NNUE
 
 #endif // #ifndef NNUE_EVALUATE_NNUE_H_INCLUDED

src/nnue/features/feature_set.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -43,48 +43,6 @@ namespace Eval::NNUE::Features {
   template <typename Derived>
   class FeatureSetBase {
 
-   public:
-    // Get a list of indices for active features
-    template <typename IndexListType>
-    static void AppendActiveIndices(
-        const Position& pos, TriggerEvent trigger, IndexListType active[2]) {
-
-      for (Color perspective : { WHITE, BLACK }) {
-        Derived::CollectActiveIndices(
-            pos, trigger, perspective, &active[perspective]);
-      }
-    }
-
-    // Get a list of indices for recently changed features
-    template <typename PositionType, typename IndexListType>
-    static void AppendChangedIndices(
-        const PositionType& pos, TriggerEvent trigger,
-        IndexListType removed[2], IndexListType added[2], bool reset[2]) {
-
-      const auto& dp = pos.state()->dirtyPiece;
-      if (dp.dirty_num == 0) return;
-
-      for (Color perspective : { WHITE, BLACK }) {
-        reset[perspective] = false;
-        switch (trigger) {
-          case TriggerEvent::kFriendKingMoved:
-            reset[perspective] =
-                dp.pieceId[0] == PIECE_ID_KING + perspective;
-            break;
-          default:
-            assert(false);
-            break;
-        }
-        if (reset[perspective]) {
-          Derived::CollectActiveIndices(
-              pos, trigger, perspective, &added[perspective]);
-        } else {
-          Derived::CollectChangedIndices(
-              pos, trigger, perspective,
-              &removed[perspective], &added[perspective]);
-        }
-      }
-    }
   };
 
   // Class template that represents the feature set
@@ -104,30 +62,6 @@ namespace Eval::NNUE::Features {
         CompileTimeList<TriggerEvent, FeatureType::kRefreshTrigger>;
     static constexpr auto kRefreshTriggers = SortedTriggerSet::kValues;
 
-   private:
-    // Get a list of indices for active features
-    static void CollectActiveIndices(
-        const Position& pos, const TriggerEvent trigger, const Color perspective,
-        IndexList* const active) {
-      if (FeatureType::kRefreshTrigger == trigger) {
-        FeatureType::AppendActiveIndices(pos, perspective, active);
-      }
-    }
-
-    // Get a list of indices for recently changed features
-    static void CollectChangedIndices(
-        const Position& pos, const TriggerEvent trigger, const Color perspective,
-        IndexList* const removed, IndexList* const added) {
-
-      if (FeatureType::kRefreshTrigger == trigger) {
-        FeatureType::AppendChangedIndices(pos, perspective, removed, added);
-      }
-    }
-
-    // Make the base class and the class template that recursively uses itself a friend
-    friend class FeatureSetBase<FeatureSet>;
-    template <typename... FeatureTypes>
-    friend class FeatureSet;
   };
 
 }  // namespace Eval::NNUE::Features

src/nnue/features/features_common.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/nnue/features/half_kp.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -23,25 +23,14 @@
 
 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
-  template <Side AssociatedKing>
-  inline void HalfKP<AssociatedKing>::GetPieces(
-      const Position& pos, Color perspective,
-      PieceSquare** pieces, Square* sq_target_k) {
-
-    *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);
+  // Index of a feature for a given king position and another piece on some square
+  inline IndexType make_index(Color perspective, Square s, Piece pc, Square ksq) {
+    return IndexType(orient(perspective, s) + kpp_board_index[perspective][pc] + PS_END * ksq);
   }
 
   // Get a list of indices for active features
@@ -49,41 +38,28 @@ 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(make_index(perspective, s, pos.piece_on(s), ksq));
     }
   }
 
   // Get a list of indices for recently changed features
   template <Side AssociatedKing>
   void HalfKP<AssociatedKing>::AppendChangedIndices(
-      const Position& pos, Color perspective,
+      const Position& pos, const DirtyPiece& dp, Color perspective,
       IndexList* removed, IndexList* added) {
 
-    PieceSquare* pieces;
-    Square sq_target_k;
-    GetPieces(pos, perspective, &pieces, &sq_target_k);
-    const auto& dp = pos.state()->dirtyPiece;
+    Square ksq = orient(perspective, pos.square<KING>(perspective));
     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(make_index(perspective, dp.from[i], pc, ksq));
+      if (dp.to[i] != SQ_NONE)
+        added->push_back(make_index(perspective, dp.to[i], pc, ksq));
     }
   }
 

src/nnue/features/half_kp.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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;
 
@@ -50,16 +50,8 @@ namespace Eval::NNUE::Features {
                                     IndexList* active);
 
     // Get a list of indices for recently changed features
-    static void AppendChangedIndices(const Position& pos, Color perspective,
+    static void AppendChangedIndices(const Position& pos, const DirtyPiece& dp, 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);
   };
 
 }  // namespace Eval::NNUE::Features

src/nnue/features/index_list.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/nnue/layers/affine_transform.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -41,6 +41,11 @@ namespace Eval::NNUE::Layers {
     static constexpr IndexType kOutputDimensions = OutputDimensions;
     static constexpr IndexType kPaddedInputDimensions =
         CeilToMultiple<IndexType>(kInputDimensions, kMaxSimdWidth);
+#if defined (USE_AVX512)
+    static constexpr const IndexType kOutputSimdWidth = kSimdWidth / 2;
+#elif defined (USE_SSSE3)
+    static constexpr const IndexType kOutputSimdWidth = kSimdWidth / 4;
+#endif
 
     // Size of forward propagation buffer used in this layer
     static constexpr std::size_t kSelfBufferSize =
@@ -62,11 +67,61 @@ 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)
+#if !defined (USE_SSSE3)
+        weights_[i] = read_little_endian<WeightType>(stream);
+#else
+        weights_[
+          (i / 4) % (kPaddedInputDimensions / 4) * kOutputDimensions * 4 +
+          i / kPaddedInputDimensions * 4 +
+          i % 4
+        ] = read_little_endian<WeightType>(stream);
+
+      // Determine if eights of weight and input products can be summed using 16bits
+      // without saturation. We assume worst case combinations of 0 and 127 for all inputs.
+      if (kOutputDimensions > 1 && !stream.fail())
+      {
+          canSaturate16.count = 0;
+#if !defined(USE_VNNI)
+          for (IndexType i = 0; i < kPaddedInputDimensions; i += 16)
+              for (IndexType j = 0; j < kOutputDimensions; ++j)
+                  for (int x = 0; x < 2; ++x)
+                  {
+                      WeightType* w = &weights_[i * kOutputDimensions + j * 4 + x * 2];
+                      int sum[2] = {0, 0};
+                      for (int k = 0; k < 8; ++k)
+                      {
+                          IndexType idx = k / 2 * kOutputDimensions * 4 + k % 2;
+                          sum[w[idx] < 0] += w[idx];
+                      }
+                      for (int sign : {-1, 1})
+                          while (sign * sum[sign == -1] > 258)
+                          {
+                              int maxK = 0, maxW = 0;
+                              for (int k = 0; k < 8; ++k)
+                              {
+                                  IndexType idx = k / 2 * kOutputDimensions * 4 + k % 2;
+                                  if (maxW < sign * w[idx])
+                                      maxK = k, maxW = sign * w[idx];
+                              }
+
+                              IndexType idx = maxK / 2 * kOutputDimensions * 4 + maxK % 2;
+                              sum[sign == -1] -= w[idx];
+                              canSaturate16.add(j, i + maxK / 2 * 4 + maxK % 2 + x * 2, w[idx]);
+                              w[idx] = 0;
+                          }
+                  }
+
+          // Non functional optimization for faster more linear access
+          std::sort(canSaturate16.ids, canSaturate16.ids + canSaturate16.count,
+                    [](const typename CanSaturate::Entry& e1, const typename CanSaturate::Entry& e2)
+                    { return e1.in == e2.in ? e1.out < e2.out : e1.in < e2.in; });
+#endif
+      }
+#endif
+
       return !stream.fail();
     }
 
@@ -75,109 +130,281 @@ namespace Eval::NNUE::Layers {
         const TransformedFeatureType* transformed_features, char* buffer) const {
       const auto input = previous_layer_.Propagate(
           transformed_features, buffer + kSelfBufferSize);
+
+#if defined (USE_AVX512)
+
+      [[maybe_unused]] const __m512i kOnes512 = _mm512_set1_epi16(1);
+
+      [[maybe_unused]] auto m512_hadd = [](__m512i sum, int bias) -> int {
+        return _mm512_reduce_add_epi32(sum) + bias;
+      };
+
+      [[maybe_unused]] auto m512_add_dpbusd_epi32 = [=](__m512i& acc, __m512i a, __m512i b) {
+#if defined (USE_VNNI)
+        acc = _mm512_dpbusd_epi32(acc, a, b);
+#else
+        __m512i product0 = _mm512_maddubs_epi16(a, b);
+        product0 = _mm512_madd_epi16(product0, kOnes512);
+        acc = _mm512_add_epi32(acc, product0);
+#endif
+      };
+
+      [[maybe_unused]] auto m512_add_dpbusd_epi32x4 = [=](__m512i& acc, __m512i a0, __m512i b0, __m512i a1, __m512i b1,
+                                                                        __m512i a2, __m512i b2, __m512i a3, __m512i b3) {
+#if defined (USE_VNNI)
+        acc = _mm512_dpbusd_epi32(acc, a0, b0);
+        acc = _mm512_dpbusd_epi32(acc, a1, b1);
+        acc = _mm512_dpbusd_epi32(acc, a2, b2);
+        acc = _mm512_dpbusd_epi32(acc, a3, b3);
+#else
+        __m512i product0 = _mm512_maddubs_epi16(a0, b0);
+        __m512i product1 = _mm512_maddubs_epi16(a1, b1);
+        __m512i product2 = _mm512_maddubs_epi16(a2, b2);
+        __m512i product3 = _mm512_maddubs_epi16(a3, b3);
+        product0 = _mm512_add_epi16(product0, product1);
+        product2 = _mm512_add_epi16(product2, product3);
+        product0 = _mm512_add_epi16(product0, product2);
+        product0 = _mm512_madd_epi16(product0, kOnes512);
+        acc = _mm512_add_epi32(acc, product0);
+#endif
+      };
+
+#endif
+#if defined (USE_AVX2)
+
+      [[maybe_unused]] const __m256i kOnes256 = _mm256_set1_epi16(1);
+
+      [[maybe_unused]] auto m256_hadd = [](__m256i sum, int bias) -> int {
+        __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));
+        return _mm_cvtsi128_si32(sum128) + bias;
+      };
+
+      [[maybe_unused]] auto m256_add_dpbusd_epi32 = [=](__m256i& acc, __m256i a, __m256i b) {
+#if defined (USE_VNNI)
+        acc = _mm256_dpbusd_epi32(acc, a, b);
+#else
+        __m256i product0 = _mm256_maddubs_epi16(a, b);
+        product0 = _mm256_madd_epi16(product0, kOnes256);
+        acc = _mm256_add_epi32(acc, product0);
+#endif
+      };
+
+      [[maybe_unused]] auto m256_add_dpbusd_epi32x4 = [=](__m256i& acc, __m256i a0, __m256i b0, __m256i a1, __m256i b1,
+                                                                        __m256i a2, __m256i b2, __m256i a3, __m256i b3) {
+#if defined (USE_VNNI)
+        acc = _mm256_dpbusd_epi32(acc, a0, b0);
+        acc = _mm256_dpbusd_epi32(acc, a1, b1);
+        acc = _mm256_dpbusd_epi32(acc, a2, b2);
+        acc = _mm256_dpbusd_epi32(acc, a3, b3);
+#else
+        __m256i product0 = _mm256_maddubs_epi16(a0, b0);
+        __m256i product1 = _mm256_maddubs_epi16(a1, b1);
+        __m256i product2 = _mm256_maddubs_epi16(a2, b2);
+        __m256i product3 = _mm256_maddubs_epi16(a3, b3);
+        product0 = _mm256_add_epi16(product0, product1);
+        product2 = _mm256_add_epi16(product2, product3);
+        product0 = _mm256_add_epi16(product0, product2);
+        product0 = _mm256_madd_epi16(product0, kOnes256);
+        acc = _mm256_add_epi32(acc, product0);
+#endif
+      };
+
+#endif
+#if defined (USE_SSSE3)
+
+      [[maybe_unused]] const __m128i kOnes128 = _mm_set1_epi16(1);
+
+      [[maybe_unused]] auto m128_hadd = [](__m128i sum, int bias) -> int {
+        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
+        return _mm_cvtsi128_si32(sum) + bias;
+      };
+
+      [[maybe_unused]] auto m128_add_dpbusd_epi32 = [=](__m128i& acc, __m128i a, __m128i b) {
+        __m128i product0 = _mm_maddubs_epi16(a, b);
+        product0 = _mm_madd_epi16(product0, kOnes128);
+        acc = _mm_add_epi32(acc, product0);
+      };
+
+      [[maybe_unused]] auto m128_add_dpbusd_epi32x4 = [=](__m128i& acc, __m128i a0, __m128i b0, __m128i a1, __m128i b1,
+                                                                        __m128i a2, __m128i b2, __m128i a3, __m128i b3) {
+        __m128i product0 = _mm_maddubs_epi16(a0, b0);
+        __m128i product1 = _mm_maddubs_epi16(a1, b1);
+        __m128i product2 = _mm_maddubs_epi16(a2, b2);
+        __m128i product3 = _mm_maddubs_epi16(a3, b3);
+        product0 = _mm_adds_epi16(product0, product1);
+        product2 = _mm_adds_epi16(product2, product3);
+        product0 = _mm_adds_epi16(product0, product2);
+        product0 = _mm_madd_epi16(product0, kOnes128);
+        acc = _mm_add_epi32(acc, product0);
+      };
+
+#endif
+
+#if defined (USE_AVX512)
+      using vec_t = __m512i;
+      #define vec_setzero _mm512_setzero_si512
+      #define vec_set_32 _mm512_set1_epi32
+      auto& vec_add_dpbusd_32 = m512_add_dpbusd_epi32;
+      auto& vec_add_dpbusd_32x4 = m512_add_dpbusd_epi32x4;
+      auto& vec_hadd = m512_hadd;
+#elif defined (USE_AVX2)
+      using vec_t = __m256i;
+      #define vec_setzero _mm256_setzero_si256
+      #define vec_set_32 _mm256_set1_epi32
+      auto& vec_add_dpbusd_32 = m256_add_dpbusd_epi32;
+      auto& vec_add_dpbusd_32x4 = m256_add_dpbusd_epi32x4;
+      auto& vec_hadd = m256_hadd;
+#elif defined (USE_SSSE3)
+      using vec_t = __m128i;
+      #define vec_setzero _mm_setzero_si128
+      #define vec_set_32 _mm_set1_epi32
+      auto& vec_add_dpbusd_32 = m128_add_dpbusd_epi32;
+      auto& vec_add_dpbusd_32x4 = m128_add_dpbusd_epi32x4;
+      auto& vec_hadd = m128_hadd;
+#endif
+
+#if defined (USE_SSSE3)
+
       const auto output = reinterpret_cast<OutputType*>(buffer);
+      const auto input_vector = reinterpret_cast<const vec_t*>(input);
 
-  #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);
+      static_assert(kOutputDimensions % kOutputSimdWidth == 0 || kOutputDimensions == 1);
 
-  #elif defined(USE_AVX2)
+      // kOutputDimensions is either 1 or a multiple of kSimdWidth
+      // because then it is also an input dimension.
+      if constexpr (kOutputDimensions % kOutputSimdWidth == 0)
+      {
+          constexpr IndexType kNumChunks = kPaddedInputDimensions / 4;
+
+          const auto input32 = reinterpret_cast<const std::int32_t*>(input);
+          vec_t* outptr = reinterpret_cast<vec_t*>(output);
+          std::memcpy(output, biases_, kOutputDimensions * sizeof(OutputType));
+
+          for (int i = 0; i < (int)kNumChunks - 3; i += 4)
+          {
+              const vec_t in0 = vec_set_32(input32[i + 0]);
+              const vec_t in1 = vec_set_32(input32[i + 1]);
+              const vec_t in2 = vec_set_32(input32[i + 2]);
+              const vec_t in3 = vec_set_32(input32[i + 3]);
+              const auto col0 = reinterpret_cast<const vec_t*>(&weights_[(i + 0) * kOutputDimensions * 4]);
+              const auto col1 = reinterpret_cast<const vec_t*>(&weights_[(i + 1) * kOutputDimensions * 4]);
+              const auto col2 = reinterpret_cast<const vec_t*>(&weights_[(i + 2) * kOutputDimensions * 4]);
+              const auto col3 = reinterpret_cast<const vec_t*>(&weights_[(i + 3) * kOutputDimensions * 4]);
+              for (int j = 0; j * kOutputSimdWidth < kOutputDimensions; ++j)
+                  vec_add_dpbusd_32x4(outptr[j], in0, col0[j], in1, col1[j], in2, col2[j], in3, col3[j]);
+          }
+          for (int i = 0; i < canSaturate16.count; ++i)
+              output[canSaturate16.ids[i].out] += input[canSaturate16.ids[i].in] * canSaturate16.ids[i].w;
+      }
+      else if constexpr (kOutputDimensions == 1)
+      {
+#if defined (USE_AVX512)
+          if constexpr (kPaddedInputDimensions % (kSimdWidth * 2) != 0)
+          {
+              constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
+              const auto input_vector256 = reinterpret_cast<const __m256i*>(input);
+
+              __m256i sum0 = _mm256_setzero_si256();
+              const auto row0 = reinterpret_cast<const __m256i*>(&weights_[0]);
+
+              for (int j = 0; j < (int)kNumChunks; ++j)
+              {
+                  const __m256i in = input_vector256[j];
+                  m256_add_dpbusd_epi32(sum0, in, row0[j]);
+              }
+              output[0] = m256_hadd(sum0, biases_[0]);
+          }
+          else
+#endif
+          {
+#if defined (USE_AVX512)
+              constexpr IndexType kNumChunks = kPaddedInputDimensions / (kSimdWidth * 2);
+#else
+              constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
+#endif
+              vec_t sum0 = vec_setzero();
+              const auto row0 = reinterpret_cast<const vec_t*>(&weights_[0]);
+
+              for (int j = 0; j < (int)kNumChunks; ++j)
+              {
+                  const vec_t in = input_vector[j];
+                  vec_add_dpbusd_32(sum0, in, row0[j]);
+              }
+              output[0] = vec_hadd(sum0, biases_[0]);
+          }
+      }
+
+#else
+
+// Use old implementation for the other architectures.
+
+      auto output = reinterpret_cast<OutputType*>(buffer);
+
+#if defined(USE_SSE2)
       constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
-      const __m256i kOnes = _mm256_set1_epi16(1);
-      const auto input_vector = reinterpret_cast<const __m256i*>(input);
-
-  #elif defined(USE_SSSE3)
-      constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
-      const __m128i kOnes = _mm_set1_epi16(1);
+      const __m128i kZeros = _mm_setzero_si128();
       const auto input_vector = reinterpret_cast<const __m128i*>(input);
 
-  #elif defined(USE_NEON)
+#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);
-  #endif
+#endif
 
       for (IndexType i = 0; i < kOutputDimensions; ++i) {
         const IndexType offset = i * kPaddedInputDimensions;
 
-  #if defined(USE_AVX512)
-        __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]));
-  #else
-            __m512i product = _mm512_maddubs_epi16(_mm512_load_si512(&input_vector[j]), _mm512_load_si512(&row[j]));
-  #endif
-
-            product = _mm512_madd_epi16(product, kOnes);
-            sum = _mm512_add_epi32(sum, product);
-        }
-        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]));
-  #else
-            __m256i sum256 = _mm256_maddubs_epi16(_mm256_load_si256(&iv_256[j]), _mm256_load_si256(&row_256[j]));
-  #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);
-        }
-
-  #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
-  #else
-            _mm256_load_si256
-  #endif
-
-            (&input_vector[j]), _mm256_load_si256(&row[j]));
-          product = _mm256_madd_epi16(product, kOnes);
-          sum = _mm256_add_epi32(sum, product);
-        }
-        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];
-
-  #elif defined(USE_SSSE3)
-        __m128i sum = _mm_cvtsi32_si128(biases_[i]);
+#if 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 product = _mm_maddubs_epi16(
-              _mm_load_si128(&input_vector[j]), _mm_load_si128(&row[j]));
-          product = _mm_madd_epi16(product, kOnes);
-          sum = _mm_add_epi32(sum, product);
+          __m128i row_j = _mm_load_si128(&row[j]);
+          __m128i input_j = _mm_load_si128(&input_vector[j]);
+          __m128i extended_row_lo = _mm_srai_epi16(_mm_unpacklo_epi8(row_j, row_j), 8);
+          __m128i extended_row_hi = _mm_srai_epi16(_mm_unpackhi_epi8(row_j, row_j), 8);
+          __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);
         }
-        sum = _mm_hadd_epi32(sum, sum);
-        sum = _mm_hadd_epi32(sum, sum);
+        __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_NEON)
+#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 extended_row_lo = _mm_srai_pi16(_mm_unpacklo_pi8(row_j, row_j), 8);
+          __m64 extended_row_hi = _mm_srai_pi16(_mm_unpackhi_pi8(row_j, row_j), 8);
+          __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]);
         for (IndexType j = 0; j < kNumChunks; ++j) {
@@ -187,15 +414,21 @@ namespace Eval::NNUE::Layers {
         }
         output[i] = sum[0] + sum[1] + sum[2] + sum[3];
 
-  #else
+#else
         OutputType sum = biases_[i];
         for (IndexType j = 0; j < kInputDimensions; ++j) {
           sum += weights_[offset + j] * input[j];
         }
         output[i] = sum;
-  #endif
+#endif
 
       }
+#if defined(USE_MMX)
+      _mm_empty();
+#endif
+
+#endif
+
       return output;
     }
 
@@ -206,8 +439,24 @@ namespace Eval::NNUE::Layers {
     PreviousLayer previous_layer_;
 
     alignas(kCacheLineSize) BiasType biases_[kOutputDimensions];
-    alignas(kCacheLineSize)
-        WeightType weights_[kOutputDimensions * kPaddedInputDimensions];
+    alignas(kCacheLineSize) WeightType weights_[kOutputDimensions * kPaddedInputDimensions];
+#if defined (USE_SSSE3)
+    struct CanSaturate {
+        int count;
+        struct Entry {
+            uint16_t out;
+            uint16_t in;
+            int8_t w;
+        } ids[kPaddedInputDimensions * kOutputDimensions * 3 / 4];
+
+        void add(int i, int j, int8_t w) {
+            ids[count].out = i;
+            ids[count].in = j;
+            ids[count].w = w;
+            ++count;
+        }
+    } canSaturate16;
+#endif
   };
 
 }  // namespace Eval::NNUE::Layers

src/nnue/layers/clipped_relu.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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_load_si256(&in[i * 4 + 0]),
+            _mm256_load_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_load_si256(&in[i * 4 + 2]),
+            _mm256_load_si256(&in[i * 4 + 3])), kWeightScaleBits);
+        _mm256_store_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/layers/input_slice.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/nnue/nnue_accumulator.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -25,13 +25,14 @@
 
 namespace Eval::NNUE {
 
+  // The accumulator of a StateInfo without parent is set to the INIT state
+  enum AccumulatorState { EMPTY, COMPUTED, INIT };
+
   // 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;
-    bool computed_accumulation;
-    bool computed_score;
+    AccumulatorState state[2];
   };
 
 }  // namespace Eval::NNUE

src/nnue/nnue_architecture.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/nnue/nnue_common.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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>
 
@@ -33,6 +36,9 @@
 #elif defined(USE_SSE2)
 #include <emmintrin.h>
 
+#elif defined(USE_MMX)
+#include <mmintrin.h>
+
 #elif defined(USE_NEON)
 #include <arm_neon.h>
 #endif
@@ -56,12 +62,43 @@ 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
+  };
+
+  constexpr uint32_t kpp_board_index[COLOR_NB][PIECE_NB] = {
+    // convention: W - us, B - them
+    // viewed from other side, W and B are reversed
+    { PS_NONE, PS_W_PAWN, PS_W_KNIGHT, PS_W_BISHOP, PS_W_ROOK, PS_W_QUEEN, PS_W_KING, PS_NONE,
+      PS_NONE, PS_B_PAWN, PS_B_KNIGHT, PS_B_BISHOP, PS_B_ROOK, PS_B_QUEEN, PS_B_KING, PS_NONE },
+    { PS_NONE, PS_B_PAWN, PS_B_KNIGHT, PS_B_BISHOP, PS_B_ROOK, PS_B_QUEEN, PS_B_KING, PS_NONE,
+      PS_NONE, PS_W_PAWN, PS_W_KNIGHT, PS_W_BISHOP, PS_W_ROOK, PS_W_QUEEN, PS_W_KING, PS_NONE }
+  };
+
   // Type of input feature after conversion
   using TransformedFeatureType = std::uint8_t;
   using IndexType = std::uint32_t;
@@ -69,7 +106,25 @@ 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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -29,6 +29,56 @@
 
 namespace Eval::NNUE {
 
+  // If vector instructions are enabled, we update and refresh the
+  // accumulator tile by tile such that each tile fits in the CPU's
+  // vector registers.
+  #define VECTOR
+
+  #ifdef USE_AVX512
+  typedef __m512i vec_t;
+  #define vec_load(a) _mm512_load_si512(a)
+  #define vec_store(a,b) _mm512_store_si512(a,b)
+  #define vec_add_16(a,b) _mm512_add_epi16(a,b)
+  #define vec_sub_16(a,b) _mm512_sub_epi16(a,b)
+  static constexpr IndexType kNumRegs = 8; // only 8 are needed
+
+  #elif USE_AVX2
+  typedef __m256i vec_t;
+  #define vec_load(a) _mm256_load_si256(a)
+  #define vec_store(a,b) _mm256_store_si256(a,b)
+  #define vec_add_16(a,b) _mm256_add_epi16(a,b)
+  #define vec_sub_16(a,b) _mm256_sub_epi16(a,b)
+  static constexpr IndexType kNumRegs = 16;
+
+  #elif USE_SSE2
+  typedef __m128i vec_t;
+  #define vec_load(a) (*(a))
+  #define vec_store(a,b) *(a)=(b)
+  #define vec_add_16(a,b) _mm_add_epi16(a,b)
+  #define vec_sub_16(a,b) _mm_sub_epi16(a,b)
+  static constexpr IndexType kNumRegs = Is64Bit ? 16 : 8;
+
+  #elif USE_MMX
+  typedef __m64 vec_t;
+  #define vec_load(a) (*(a))
+  #define vec_store(a,b) *(a)=(b)
+  #define vec_add_16(a,b) _mm_add_pi16(a,b)
+  #define vec_sub_16(a,b) _mm_sub_pi16(a,b)
+  static constexpr IndexType kNumRegs = 8;
+
+  #elif USE_NEON
+  typedef int16x8_t vec_t;
+  #define vec_load(a) (*(a))
+  #define vec_store(a,b) *(a)=(b)
+  #define vec_add_16(a,b) vaddq_s16(a,b)
+  #define vec_sub_16(a,b) vsubq_s16(a,b)
+  static constexpr IndexType kNumRegs = 16;
+
+  #else
+  #undef VECTOR
+
+  #endif
+
   // Input feature converter
   class FeatureTransformer {
 
@@ -36,6 +86,11 @@ namespace Eval::NNUE {
     // Number of output dimensions for one side
     static constexpr IndexType kHalfDimensions = kTransformedFeatureDimensions;
 
+    #ifdef VECTOR
+    static constexpr IndexType kTileHeight = kNumRegs * sizeof(vec_t) / 2;
+    static_assert(kHalfDimensions % kTileHeight == 0, "kTileHeight must divide kHalfDimensions");
+    #endif
+
    public:
     // Output type
     using OutputType = TransformedFeatureType;
@@ -50,45 +105,40 @@ namespace Eval::NNUE {
 
     // Hash value embedded in the evaluation file
     static constexpr std::uint32_t GetHashValue() {
+
       return RawFeatures::kHashValue ^ kOutputDimensions;
     }
 
     // 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();
     }
 
-    // Proceed with the difference calculation if possible
-    bool UpdateAccumulatorIfPossible(const Position& pos) const {
-      const auto now = pos.state();
-      if (now->accumulator.computed_accumulation) {
-        return true;
-      }
-      const auto prev = now->previous;
-      if (prev && prev->accumulator.computed_accumulation) {
-        UpdateAccumulator(pos);
-        return true;
-      }
-      return false;
-    }
-
     // Convert input features
-    void Transform(const Position& pos, OutputType* output, bool refresh) const {
-      if (refresh || !UpdateAccumulatorIfPossible(pos)) {
-        RefreshAccumulator(pos);
-      }
+    void Transform(const Position& pos, OutputType* output) const {
+
+      UpdateAccumulator(pos, WHITE);
+      UpdateAccumulator(pos, BLACK);
+
       const auto& accumulation = pos.state()->accumulator.accumulation;
 
-  #if defined(USE_AVX2)
+  #if defined(USE_AVX512)
+      constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth * 2);
+      static_assert(kHalfDimensions % (kSimdWidth * 2) == 0);
+      const __m512i kControl = _mm512_setr_epi64(0, 2, 4, 6, 1, 3, 5, 7);
+      const __m512i kZero = _mm512_setzero_si512();
+
+  #elif defined(USE_AVX2)
       constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
       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 +147,10 @@ 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};
@@ -106,44 +160,29 @@ namespace Eval::NNUE {
       for (IndexType p = 0; p < 2; ++p) {
         const IndexType offset = kHalfDimensions * p;
 
-  #if defined(USE_AVX2)
+  #if defined(USE_AVX512)
+        auto out = reinterpret_cast<__m512i*>(&output[offset]);
+        for (IndexType j = 0; j < kNumChunks; ++j) {
+          __m512i sum0 = _mm512_load_si512(
+              &reinterpret_cast<const __m512i*>(accumulation[perspectives[p]][0])[j * 2 + 0]);
+          __m512i sum1 = _mm512_load_si512(
+              &reinterpret_cast<const __m512i*>(accumulation[perspectives[p]][0])[j * 2 + 1]);
+          _mm512_store_si512(&out[j], _mm512_permutexvar_epi64(kControl,
+              _mm512_max_epi8(_mm512_packs_epi16(sum0, sum1), kZero)));
+        }
+
+  #elif 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_load_si256(
+              &reinterpret_cast<const __m256i*>(accumulation[perspectives[p]][0])[j * 2 + 0]);
+          __m256i sum1 = _mm256_load_si256(
+              &reinterpret_cast<const __m256i*>(accumulation[perspectives[p]][0])[j * 2 + 1]);
+          _mm256_store_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*>(
@@ -155,14 +194,25 @@ namespace Eval::NNUE {
           _mm_store_si128(&out[j],
 
   #ifdef USE_SSE41
-            _mm_max_epi8(packedbytes, kZero)
+              _mm_max_epi8(packedbytes, kZero)
   #else
-            _mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s)
+              _mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s)
   #endif
 
           );
         }
 
+  #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) {
@@ -180,166 +230,178 @@ namespace Eval::NNUE {
   #endif
 
       }
+  #if defined(USE_MMX)
+      _mm_empty();
+  #endif
     }
 
    private:
-    // Calculate cumulative value without using difference calculation
-    void RefreshAccumulator(const Position& pos) const {
-      auto& accumulator = pos.state()->accumulator;
-      IndexType i = 0;
-      Features::IndexList active_indices[2];
-      RawFeatures::AppendActiveIndices(pos, kRefreshTriggers[i],
-                                       active_indices);
-      for (Color perspective : { WHITE, BLACK }) {
-        std::memcpy(accumulator.accumulation[perspective][i], biases_,
-                   kHalfDimensions * sizeof(BiasType));
-        for (const auto index : active_indices[perspective]) {
+    void UpdateAccumulator(const Position& pos, const Color c) const {
+
+  #ifdef VECTOR
+      // Gcc-10.2 unnecessarily spills AVX2 registers if this array
+      // is defined in the VECTOR code below, once in each branch
+      vec_t acc[kNumRegs];
+  #endif
+
+      // Look for a usable accumulator of an earlier position. We keep track
+      // of the estimated gain in terms of features to be added/subtracted.
+      StateInfo *st = pos.state(), *next = nullptr;
+      int gain = pos.count<ALL_PIECES>() - 2;
+      while (st->accumulator.state[c] == EMPTY)
+      {
+        auto& dp = st->dirtyPiece;
+        // The first condition tests whether an incremental update is
+        // possible at all: if this side's king has moved, it is not possible.
+        static_assert(std::is_same_v<RawFeatures::SortedTriggerSet,
+              Features::CompileTimeList<Features::TriggerEvent, Features::TriggerEvent::kFriendKingMoved>>,
+              "Current code assumes that only kFriendlyKingMoved refresh trigger is being used.");
+        if (   dp.piece[0] == make_piece(c, KING)
+            || (gain -= dp.dirty_num + 1) < 0)
+          break;
+        next = st;
+        st = st->previous;
+      }
+
+      if (st->accumulator.state[c] == COMPUTED)
+      {
+        if (next == nullptr)
+          return;
+
+        // Update incrementally in two steps. First, we update the "next"
+        // accumulator. Then, we update the current accumulator (pos.state()).
+
+        // Gather all features to be updated. This code assumes HalfKP features
+        // only and doesn't support refresh triggers.
+        static_assert(std::is_same_v<Features::FeatureSet<Features::HalfKP<Features::Side::kFriend>>,
+                                     RawFeatures>);
+        Features::IndexList removed[2], added[2];
+        Features::HalfKP<Features::Side::kFriend>::AppendChangedIndices(pos,
+            next->dirtyPiece, c, &removed[0], &added[0]);
+        for (StateInfo *st2 = pos.state(); st2 != next; st2 = st2->previous)
+          Features::HalfKP<Features::Side::kFriend>::AppendChangedIndices(pos,
+              st2->dirtyPiece, c, &removed[1], &added[1]);
+
+        // Mark the accumulators as computed.
+        next->accumulator.state[c] = COMPUTED;
+        pos.state()->accumulator.state[c] = COMPUTED;
+
+        // Now update the accumulators listed in info[], where the last element is a sentinel.
+        StateInfo *info[3] =
+          { next, next == pos.state() ? nullptr : pos.state(), nullptr };
+  #ifdef VECTOR
+        for (IndexType j = 0; j < kHalfDimensions / kTileHeight; ++j)
+        {
+          // Load accumulator
+          auto accTile = reinterpret_cast<vec_t*>(
+            &st->accumulator.accumulation[c][0][j * kTileHeight]);
+          for (IndexType k = 0; k < kNumRegs; ++k)
+            acc[k] = vec_load(&accTile[k]);
+
+          for (IndexType i = 0; info[i]; ++i)
+          {
+            // Difference calculation for the deactivated features
+            for (const auto index : removed[i])
+            {
+              const IndexType offset = kHalfDimensions * index + j * kTileHeight;
+              auto column = reinterpret_cast<const vec_t*>(&weights_[offset]);
+              for (IndexType k = 0; k < kNumRegs; ++k)
+                acc[k] = vec_sub_16(acc[k], column[k]);
+            }
+
+            // Difference calculation for the activated features
+            for (const auto index : added[i])
+            {
+              const IndexType offset = kHalfDimensions * index + j * kTileHeight;
+              auto column = reinterpret_cast<const vec_t*>(&weights_[offset]);
+              for (IndexType k = 0; k < kNumRegs; ++k)
+                acc[k] = vec_add_16(acc[k], column[k]);
+            }
+
+            // Store accumulator
+            accTile = reinterpret_cast<vec_t*>(
+              &info[i]->accumulator.accumulation[c][0][j * kTileHeight]);
+            for (IndexType k = 0; k < kNumRegs; ++k)
+              vec_store(&accTile[k], acc[k]);
+          }
+        }
+
+  #else
+        for (IndexType i = 0; info[i]; ++i)
+        {
+          std::memcpy(info[i]->accumulator.accumulation[c][0],
+              st->accumulator.accumulation[c][0],
+              kHalfDimensions * sizeof(BiasType));
+          st = info[i];
+
+          // Difference calculation for the deactivated features
+          for (const auto index : removed[i])
+          {
+            const IndexType offset = kHalfDimensions * index;
+
+            for (IndexType j = 0; j < kHalfDimensions; ++j)
+              st->accumulator.accumulation[c][0][j] -= weights_[offset + j];
+          }
+
+          // Difference calculation for the activated features
+          for (const auto index : added[i])
+          {
+            const IndexType offset = kHalfDimensions * index;
+
+            for (IndexType j = 0; j < kHalfDimensions; ++j)
+              st->accumulator.accumulation[c][0][j] += weights_[offset + j];
+          }
+        }
+  #endif
+      }
+      else
+      {
+        // Refresh the accumulator
+        auto& accumulator = pos.state()->accumulator;
+        accumulator.state[c] = COMPUTED;
+        Features::IndexList active;
+        Features::HalfKP<Features::Side::kFriend>::AppendActiveIndices(pos, c, &active);
+
+  #ifdef VECTOR
+        for (IndexType j = 0; j < kHalfDimensions / kTileHeight; ++j)
+        {
+          auto biasesTile = reinterpret_cast<const vec_t*>(
+              &biases_[j * kTileHeight]);
+          for (IndexType k = 0; k < kNumRegs; ++k)
+            acc[k] = biasesTile[k];
+
+          for (const auto index : active)
+          {
+            const IndexType offset = kHalfDimensions * index + j * kTileHeight;
+            auto column = reinterpret_cast<const vec_t*>(&weights_[offset]);
+
+            for (unsigned k = 0; k < kNumRegs; ++k)
+              acc[k] = vec_add_16(acc[k], column[k]);
+          }
+
+          auto accTile = reinterpret_cast<vec_t*>(
+              &accumulator.accumulation[c][0][j * kTileHeight]);
+          for (unsigned k = 0; k < kNumRegs; k++)
+            vec_store(&accTile[k], acc[k]);
+        }
+
+  #else
+        std::memcpy(accumulator.accumulation[c][0], biases_,
+            kHalfDimensions * sizeof(BiasType));
+
+        for (const auto index : active)
+        {
           const IndexType offset = kHalfDimensions * index;
 
-  #if 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
-          }
-
-  #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) {
-            accumulation[j] = _mm_add_epi16(accumulation[j], column[j]);
-          }
-
-  #elif defined(USE_NEON)
-          auto accumulation = reinterpret_cast<int16x8_t*>(
-              &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) {
-            accumulation[j] = vaddq_s16(accumulation[j], column[j]);
-          }
-
-  #else
-          for (IndexType j = 0; j < kHalfDimensions; ++j) {
-            accumulator.accumulation[perspective][i][j] += weights_[offset + j];
-          }
-  #endif
-
+          for (IndexType j = 0; j < kHalfDimensions; ++j)
+            accumulator.accumulation[c][0][j] += weights_[offset + j];
         }
+  #endif
       }
 
-      accumulator.computed_accumulation = true;
-      accumulator.computed_score = false;
-    }
-
-    // Calculate cumulative value using difference calculation
-    void UpdateAccumulator(const Position& pos) const {
-      const auto prev_accumulator = pos.state()->previous->accumulator;
-      auto& accumulator = pos.state()->accumulator;
-      IndexType i = 0;
-      Features::IndexList removed_indices[2], added_indices[2];
-      bool reset[2];
-      RawFeatures::AppendChangedIndices(pos, kRefreshTriggers[i],
-                                        removed_indices, added_indices, reset);
-      for (Color perspective : { WHITE, BLACK }) {
-
-  #if defined(USE_AVX2)
-        constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-        auto accumulation = reinterpret_cast<__m256i*>(
-            &accumulator.accumulation[perspective][i][0]);
-
-  #elif defined(USE_SSE2)
-        constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-        auto accumulation = reinterpret_cast<__m128i*>(
-            &accumulator.accumulation[perspective][i][0]);
-
-  #elif defined(USE_NEON)
-        constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-        auto accumulation = reinterpret_cast<int16x8_t*>(
-            &accumulator.accumulation[perspective][i][0]);
+  #if defined(USE_MMX)
+      _mm_empty();
   #endif
-
-        if (reset[perspective]) {
-          std::memcpy(accumulator.accumulation[perspective][i], biases_,
-                      kHalfDimensions * sizeof(BiasType));
-        } else {
-          std::memcpy(accumulator.accumulation[perspective][i],
-                      prev_accumulator.accumulation[perspective][i],
-                      kHalfDimensions * sizeof(BiasType));
-          // Difference calculation for the deactivated features
-          for (const auto index : removed_indices[perspective]) {
-            const IndexType offset = kHalfDimensions * index;
-
-  #if defined(USE_AVX2)
-            auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
-              accumulation[j] = _mm256_sub_epi16(accumulation[j], column[j]);
-            }
-
-  #elif defined(USE_SSE2)
-            auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
-              accumulation[j] = _mm_sub_epi16(accumulation[j], column[j]);
-            }
-
-  #elif defined(USE_NEON)
-            auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
-              accumulation[j] = vsubq_s16(accumulation[j], column[j]);
-            }
-
-  #else
-            for (IndexType j = 0; j < kHalfDimensions; ++j) {
-              accumulator.accumulation[perspective][i][j] -=
-                  weights_[offset + j];
-            }
-  #endif
-
-          }
-        }
-        { // Difference calculation for the activated features
-          for (const auto index : added_indices[perspective]) {
-            const IndexType offset = kHalfDimensions * index;
-
-  #if defined(USE_AVX2)
-            auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
-              accumulation[j] = _mm256_add_epi16(accumulation[j], column[j]);
-            }
-
-  #elif defined(USE_SSE2)
-            auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
-              accumulation[j] = _mm_add_epi16(accumulation[j], column[j]);
-            }
-
-  #elif defined(USE_NEON)
-            auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
-              accumulation[j] = vaddq_s16(accumulation[j], column[j]);
-            }
-
-  #else
-            for (IndexType j = 0; j < kHalfDimensions; ++j) {
-              accumulator.accumulation[perspective][i][j] +=
-                  weights_[offset + j];
-            }
-  #endif
-
-          }
-        }
-      }
-
-      accumulator.computed_accumulation = true;
-      accumulator.computed_score = false;
     }
 
     using BiasType = std::int16_t;

src/pawns.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -30,29 +30,30 @@ 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( 9, 22);
+  constexpr Score Doubled       = S(13, 51);
+  constexpr Score DoubledEarly  = S(20,  7);
+  constexpr Score Isolated      = S( 3, 15);
+  constexpr Score WeakLever     = S( 4, 58);
+  constexpr Score WeakUnopposed = S(13, 24);
 
   // Bonus for blocked pawns at 5th or 6th rank
-  constexpr Score BlockedPawn[2] = { S(-11, -4), S(-3, 4) };
+  constexpr Score BlockedPawn[2] = { S(-17, -6), S(-9, 2) };
 
   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)
+    S(0, 0), S(0, 0), S(75, 78), S(-8, 16), S(-6, 10), S(-6, 6), S(0, 2)
   };
 
   // Connected pawn bonus
-  constexpr int Connected[RANK_NB] = { 0, 7, 8, 12, 29, 48, 86 };
+  constexpr int Connected[RANK_NB] = { 0, 5, 7, 11, 23, 48, 87 };
 
   // 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.
   constexpr Value ShelterStrength[int(FILE_NB) / 2][RANK_NB] = {
-    { V( -6), V( 81), V( 93), V( 58), V( 39), V( 18), V(  25) },
-    { V(-43), V( 61), V( 35), V(-49), V(-29), V(-11), V( -63) },
-    { V(-10), V( 75), V( 23), V( -2), V( 32), V(  3), V( -45) },
-    { V(-39), V(-13), V(-29), V(-52), V(-48), V(-67), V(-166) }
+    { V( -5), V( 82), V( 92), V( 54), V( 36), V( 22), V(  28) },
+    { V(-44), V( 63), V( 33), V(-50), V(-30), V(-12), V( -62) },
+    { V(-11), V( 77), V( 22), V( -6), V( 31), V(  8), V( -45) },
+    { V(-39), V(-12), V(-29), V(-50), V(-43), V(-68), V(-164) }
   };
 
   // Danger of enemy pawns moving toward our king by [distance from edge][rank].
@@ -60,12 +61,18 @@ namespace {
   // is behind our king. Note that UnblockedStorm[0][1-2] accommodate opponent pawn
   // on edge, likely blocked by our king.
   constexpr Value UnblockedStorm[int(FILE_NB) / 2][RANK_NB] = {
-    { V( 85), V(-289), V(-166), V(97), V(50), V( 45), V( 50) },
-    { V( 46), V( -25), V( 122), V(45), V(37), V(-10), V( 20) },
-    { V( -6), V(  51), V( 168), V(34), V(-2), V(-22), V(-14) },
-    { V(-15), V( -11), V( 101), V( 4), V(11), V(-15), V(-29) }
+    { V( 87), V(-288), V(-168), V( 96), V( 47), V( 44), V( 46) },
+    { V( 42), V( -25), V( 120), V( 45), V( 34), V( -9), V( 24) },
+    { V( -8), V(  51), V( 167), V( 35), V( -4), V(-16), V(-12) },
+    { V(-17), V( -13), V( 100), V(  4), V(  9), V(-16), V(-31) }
   };
 
+
+  // KingOnFile[semi-open Us][semi-open Them] contains bonuses/penalties
+  // for king when the king is on a semi-open or open file.
+  constexpr Score KingOnFile[2][2] = {{ S(-21,10), S(-7, 1)  },
+                                     {  S(  0,-3), S( 9,-4) }};
+
   #undef S
   #undef V
 
@@ -80,13 +87,14 @@ namespace {
 
     constexpr Color     Them = ~Us;
     constexpr Direction Up   = pawn_push(Us);
+    constexpr Direction Down = -Up;
 
     Bitboard neighbours, stoppers, support, phalanx, opposed;
     Bitboard lever, leverPush, blocked;
     Square s;
     bool backward, passed, doubled;
     Score score = SCORE_ZERO;
-    const Square* pl = pos.squares<PAWN>(Us);
+    Bitboard b = pos.pieces(Us, PAWN);
 
     Bitboard ourPawns   = pos.pieces(  Us, PAWN);
     Bitboard theirPawns = pos.pieces(Them, PAWN);
@@ -99,8 +107,9 @@ namespace {
     e->blockedCount += popcount(shift<Up>(ourPawns) & (theirPawns | doubleAttackThem));
 
     // Loop through all pawns of the current color and score each pawn
-    while ((s = *pl++) != SQ_NONE)
-    {
+    while (b) {
+        s = pop_lsb(&b);
+
         assert(pos.piece_on(s) == make_piece(Us, PAWN));
 
         Rank r = relative_rank(Us, s);
@@ -116,6 +125,13 @@ namespace {
         phalanx    = neighbours & rank_bb(s);
         support    = neighbours & rank_bb(s - Up);
 
+        if (doubled)
+        {
+            // Additional doubled penalty if none of their pawns is fixed
+            if (!(ourPawns & shift<Down>(theirPawns | pawn_attacks_bb<Them>(theirPawns))))
+                score -= DoubledEarly;
+        }
+
         // A pawn is backward when it is behind all pawns of the same color on
         // the adjacent files and cannot safely advance.
         backward =  !(neighbours & forward_ranks_bb(Them, s + Up))
@@ -147,7 +163,7 @@ namespace {
         if (support | phalanx)
         {
             int v =  Connected[r] * (2 + bool(phalanx) - bool(opposed))
-                   + 21 * popcount(support);
+                   + 22 * popcount(support);
 
             score += make_score(v, v * (r - 2) / 4);
         }
@@ -165,14 +181,14 @@ namespace {
 
         else if (backward)
             score -=  Backward
-                    + WeakUnopposed * !opposed;
+                    + WeakUnopposed * !opposed * bool(~(FileABB | FileHBB) & s);
 
         if (!support)
             score -=  Doubled * doubled
                     + WeakLever * more_than_one(lever);
 
-        if (blocked && r > RANK_4)
-            score += BlockedPawn[r-4];
+        if (blocked && r >= RANK_5)
+            score += BlockedPawn[r - RANK_5];
     }
 
     return score;
@@ -219,7 +235,7 @@ 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);
@@ -237,6 +253,9 @@ Score Entry::evaluate_shelter(const Position& pos, Square ksq) const {
           bonus -= make_score(UnblockedStorm[d][theirRank], 0);
   }
 
+  // King On File
+  bonus -= KingOnFile[pos.is_on_semiopen_file(Us, ksq)][pos.is_on_semiopen_file(Them, ksq)];
+
   return bonus;
 }
 

src/pawns.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/position.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -77,6 +77,8 @@ std::ostream& operator<<(std::ostream& os, const Position& pos) {
       && !pos.can_castle(ANY_CASTLING))
   {
       StateInfo st;
+      ASSERT_ALIGNED(&st, Eval::NNUE::kCacheLineSize);
+
       Position p;
       p.set(pos.fen(), pos.is_chess960(), &st, pos.this_thread());
       Tablebases::ProbeState s1, s2;
@@ -195,12 +197,8 @@ Position& Position::set(const string& fenStr, bool isChess960, StateInfo* si, Th
 
   std::memset(this, 0, sizeof(Position));
   std::memset(si, 0, sizeof(StateInfo));
-  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 +210,8 @@ 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;
       }
   }
@@ -264,6 +249,8 @@ Position& Position::set(const string& fenStr, bool isChess960, StateInfo* si, Th
       set_castling_right(c, rsq);
   }
 
+  set_state(st);
+
   // 4. En passant square.
   // Ignore if square is invalid or not on side to move relative rank 6.
   bool enpassant = false;
@@ -277,12 +264,24 @@ Position& Position::set(const string& fenStr, bool isChess960, StateInfo* si, Th
       // a) side to move have a pawn threatening epSquare
       // b) there is an enemy pawn in front of epSquare
       // c) there is no piece on epSquare or behind epSquare
+      // d) enemy pawn didn't block a check of its own color by moving forward
       enpassant = pawn_attacks_bb(~sideToMove, st->epSquare) & pieces(sideToMove, PAWN)
                && (pieces(~sideToMove, PAWN) & (st->epSquare + pawn_push(~sideToMove)))
-               && !(pieces() & (st->epSquare | (st->epSquare + pawn_push(sideToMove))));
+               && !(pieces() & (st->epSquare | (st->epSquare + pawn_push(sideToMove))))
+               && (   file_of(square<KING>(sideToMove)) == file_of(st->epSquare)
+                   || !(blockers_for_king(sideToMove) & (st->epSquare + pawn_push(~sideToMove))));
   }
 
-  if (!enpassant)
+  // It's necessary for st->previous to be intialized in this way because legality check relies on its existence
+  if (enpassant) {
+      st->previous = new StateInfo();
+      remove_piece(st->epSquare - pawn_push(sideToMove));
+      st->previous->checkersBB = attackers_to(square<KING>(~sideToMove)) & pieces(sideToMove);
+      st->previous->blockersForKing[WHITE] = slider_blockers(pieces(BLACK), square<KING>(WHITE), st->previous->pinners[BLACK]);
+      st->previous->blockersForKing[BLACK] = slider_blockers(pieces(WHITE), square<KING>(BLACK), st->previous->pinners[WHITE]);
+      put_piece(make_piece(~sideToMove, PAWN), st->epSquare - pawn_push(sideToMove));
+  }
+  else
       st->epSquare = SQ_NONE;
 
   // 5-6. Halfmove clock and fullmove number
@@ -294,7 +293,8 @@ Position& Position::set(const string& fenStr, bool isChess960, StateInfo* si, Th
 
   chess960 = isChess960;
   thisThread = th;
-  set_state(st);
+  st->accumulator.state[WHITE] = Eval::NNUE::INIT;
+  st->accumulator.state[BLACK] = Eval::NNUE::INIT;
 
   assert(pos_is_ok());
 
@@ -515,23 +515,11 @@ bool Position::legal(Move m) const {
   assert(color_of(moved_piece(m)) == us);
   assert(piece_on(square<KING>(us)) == make_piece(us, KING));
 
-  // En passant captures are a tricky special case. Because they are rather
-  // uncommon, we do it simply by testing whether the king is attacked after
-  // the move is made.
-  if (type_of(m) == ENPASSANT)
-  {
-      Square ksq = square<KING>(us);
-      Square capsq = to - pawn_push(us);
-      Bitboard occupied = (pieces() ^ from ^ capsq) | to;
-
-      assert(to == ep_square());
-      assert(moved_piece(m) == make_piece(us, PAWN));
-      assert(piece_on(capsq) == make_piece(~us, PAWN));
-      assert(piece_on(to) == NO_PIECE);
-
-      return   !(attacks_bb<  ROOK>(ksq, occupied) & pieces(~us, QUEEN, ROOK))
-            && !(attacks_bb<BISHOP>(ksq, occupied) & pieces(~us, QUEEN, BISHOP));
-  }
+  // st->previous->blockersForKing consider capsq as empty.
+  // If pinned, it has to move along the king ray.
+  if (type_of(m) == EN_PASSANT)
+      return   !(st->previous->blockersForKing[sideToMove] & from)
+            || aligned(from, to, square<KING>(us));
 
   // Castling moves generation does not check if the castling path is clear of
   // enemy attacks, it is delayed at a later time: now!
@@ -546,11 +534,9 @@ bool Position::legal(Move m) const {
           if (attackers_to(s) & pieces(~us))
               return false;
 
-      // In case of Chess960, verify that when moving the castling rook we do
-      // not discover some hidden checker.
+      // In case of Chess960, verify if the Rook blocks some checks
       // For instance an enemy queen in SQ_A1 when castling rook is in SQ_B1.
-      return   !chess960
-            || !(attacks_bb<ROOK>(to, pieces() ^ to_sq(m)) & pieces(~us, ROOK, QUEEN));
+      return !chess960 || !(blockers_for_king(us) & to_sq(m));
   }
 
   // If the moving piece is a king, check whether the destination square is
@@ -560,8 +546,8 @@ bool Position::legal(Move m) const {
 
   // A non-king move is legal if and only if it is not pinned or it
   // is moving along the ray towards or away from the king.
-  return   !(blockers_for_king(us) & from)
-        ||  aligned(from, to, square<KING>(us));
+  return !(blockers_for_king(us) & from)
+      || aligned(from, to, square<KING>(us));
 }
 
 
@@ -577,8 +563,10 @@ bool Position::pseudo_legal(const Move m) const {
   Piece pc = moved_piece(m);
 
   // Use a slower but simpler function for uncommon cases
+  // yet we skip the legality check of MoveList<LEGAL>().
   if (type_of(m) != NORMAL)
-      return MoveList<LEGAL>(*this).contains(m);
+      return checkers() ? MoveList<    EVASIONS>(*this).contains(m)
+                        : MoveList<NON_EVASIONS>(*this).contains(m);
 
   // Is not a promotion, so promotion piece must be empty
   if (promotion_type(m) - KNIGHT != NO_PIECE_TYPE)
@@ -664,31 +652,24 @@ bool Position::gives_check(Move m) const {
   case PROMOTION:
       return attacks_bb(promotion_type(m), to, pieces() ^ from) & square<KING>(~sideToMove);
 
-  // En passant capture with check? We have already handled the case
-  // of direct checks and ordinary discovered check, so the only case we
-  // need to handle is the unusual case of a discovered check through
-  // the captured pawn.
-  case ENPASSANT:
-  {
-      Square capsq = make_square(file_of(to), rank_of(from));
-      Bitboard b = (pieces() ^ from ^ capsq) | to;
+  // The double-pushed pawn blocked a check? En Passant will remove the blocker.
+  // The only discovery check that wasn't handle is through capsq and fromsq
+  // So the King must be in the same rank as fromsq to consider this possibility.
+  // st->previous->blockersForKing consider capsq as empty.
+  case EN_PASSANT:
+      return st->previous->checkersBB
+          || (   rank_of(square<KING>(~sideToMove)) == rank_of(from)
+              && st->previous->blockersForKing[~sideToMove] & from);
 
-      return  (attacks_bb<  ROOK>(square<KING>(~sideToMove), b) & pieces(sideToMove, QUEEN, ROOK))
-            | (attacks_bb<BISHOP>(square<KING>(~sideToMove), b) & pieces(sideToMove, QUEEN, BISHOP));
-  }
-  case CASTLING:
+  default: //CASTLING
   {
-      Square kfrom = from;
-      Square rfrom = to; // Castling is encoded as 'king captures the rook'
-      Square kto = relative_square(sideToMove, rfrom > kfrom ? SQ_G1 : SQ_C1);
-      Square rto = relative_square(sideToMove, rfrom > kfrom ? SQ_F1 : SQ_D1);
+      // Castling is encoded as 'king captures the rook'
+      Square ksq = square<KING>(~sideToMove);
+      Square rto = relative_square(sideToMove, to > from ? SQ_F1 : SQ_D1);
 
-      return   (attacks_bb<ROOK>(rto) & square<KING>(~sideToMove))
-            && (attacks_bb<ROOK>(rto, (pieces() ^ kfrom ^ rfrom) | rto | kto) & square<KING>(~sideToMove));
+      return   (attacks_bb<ROOK>(rto) & ksq)
+            && (attacks_bb<ROOK>(rto, pieces() ^ from ^ to) & ksq);
   }
-  default:
-      assert(false);
-      return false;
   }
 }
 
@@ -719,10 +700,8 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
   ++st->pliesFromNull;
 
   // Used by NNUE
-  st->accumulator.computed_accumulation = false;
-  st->accumulator.computed_score = false;
-  PieceId dp0 = PIECE_ID_NONE;
-  PieceId dp1 = PIECE_ID_NONE;
+  st->accumulator.state[WHITE] = Eval::NNUE::EMPTY;
+  st->accumulator.state[BLACK] = Eval::NNUE::EMPTY;
   auto& dp = st->dirtyPiece;
   dp.dirty_num = 1;
 
@@ -731,7 +710,7 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
   Square from = from_sq(m);
   Square to = to_sq(m);
   Piece pc = piece_on(from);
-  Piece captured = type_of(m) == ENPASSANT ? make_piece(them, PAWN) : piece_on(to);
+  Piece captured = type_of(m) == EN_PASSANT ? make_piece(them, PAWN) : piece_on(to);
 
   assert(color_of(pc) == us);
   assert(captured == NO_PIECE || color_of(captured) == (type_of(m) != CASTLING ? them : us));
@@ -757,7 +736,7 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
       // update non-pawn material.
       if (type_of(captured) == PAWN)
       {
-          if (type_of(m) == ENPASSANT)
+          if (type_of(m) == EN_PASSANT)
           {
               capsq -= pawn_push(us);
 
@@ -775,18 +754,16 @@ 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
       remove_piece(capsq);
 
-      if (type_of(m) == ENPASSANT)
+      if (type_of(m) == EN_PASSANT)
           board[capsq] = NO_PIECE;
 
       // Update material hash key and prefetch access to materialTable
@@ -821,11 +798,9 @@ 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);
@@ -834,7 +809,7 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
   // If the moving piece is a pawn do some special extra work
   if (type_of(pc) == PAWN)
   {
-      // Set en-passant square if the moved pawn can be captured
+      // Set en passant square if the moved pawn can be captured
       if (   (int(to) ^ int(from)) == 16
           && (pawn_attacks_bb(us, to - pawn_push(us)) & pieces(them, PAWN)))
       {
@@ -854,9 +829,12 @@ 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
@@ -950,17 +928,11 @@ 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;
 
-          if (type_of(m) == ENPASSANT)
+          if (type_of(m) == EN_PASSANT)
           {
               capsq -= pawn_push(us);
 
@@ -972,14 +944,6 @@ 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 +965,16 @@ 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
@@ -1046,17 +994,16 @@ void Position::do_null_move(StateInfo& newSt) {
   assert(!checkers());
   assert(&newSt != st);
 
-  if (Eval::useNNUE)
-  {
-      std::memcpy(&newSt, st, sizeof(StateInfo));
-      st->accumulator.computed_score = false;
-  }
-  else
-      std::memcpy(&newSt, st, offsetof(StateInfo, accumulator));
+  std::memcpy(&newSt, st, offsetof(StateInfo, accumulator));
 
   newSt.previous = st;
   st = &newSt;
 
+  st->dirtyPiece.dirty_num = 0;
+  st->dirtyPiece.piece[0] = NO_PIECE; // Avoid checks in UpdateAccumulator()
+  st->accumulator.state[WHITE] = Eval::NNUE::EMPTY;
+  st->accumulator.state[BLACK] = Eval::NNUE::EMPTY;
+
   if (st->epSquare != SQ_NONE)
   {
       st->key ^= Zobrist::enpassant[file_of(st->epSquare)];
@@ -1064,7 +1011,7 @@ void Position::do_null_move(StateInfo& newSt) {
   }
 
   st->key ^= Zobrist::side;
-  prefetch(TT.first_entry(st->key));
+  prefetch(TT.first_entry(key()));
 
   ++st->rule50;
   st->pliesFromNull = 0;
@@ -1089,7 +1036,7 @@ void Position::undo_null_move() {
 
 /// Position::key_after() computes the new hash key after the given move. Needed
 /// for speculative prefetch. It doesn't recognize special moves like castling,
-/// en-passant and promotions.
+/// en passant and promotions.
 
 Key Position::key_after(Move m) const {
 
@@ -1114,7 +1061,7 @@ bool Position::see_ge(Move m, Value threshold) const {
 
   assert(is_ok(m));
 
-  // Only deal with normal moves, assume others pass a simple see
+  // Only deal with normal moves, assume others pass a simple SEE
   if (type_of(m) != NORMAL)
       return VALUE_ZERO >= threshold;
 
@@ -1145,8 +1092,8 @@ 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;
@@ -1366,21 +1313,17 @@ bool Position::pos_is_ok() const {
               assert(0 && "pos_is_ok: Bitboards");
 
   StateInfo si = *st;
+  ASSERT_ALIGNED(&si, Eval::NNUE::kCacheLineSize);
+
   set_state(&si);
   if (std::memcmp(&si, st, sizeof(StateInfo)))
       assert(0 && "pos_is_ok: State");
 
   for (Piece pc : Pieces)
-  {
       if (   pieceCount[pc] != popcount(pieces(color_of(pc), type_of(pc)))
           || pieceCount[pc] != std::count(board, board + SQUARE_NB, pc))
           assert(0 && "pos_is_ok: Pieces");
 
-      for (int i = 0; i < pieceCount[pc]; ++i)
-          if (board[pieceList[pc][i]] != pc || index[pieceList[pc][i]] != i)
-              assert(0 && "pos_is_ok: Index");
-  }
-
   for (Color c : { WHITE, BLACK })
       for (CastlingRights cr : {c & KING_SIDE, c & QUEEN_SIDE})
       {

src/position.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -26,6 +26,7 @@
 
 #include "bitboard.h"
 #include "evaluate.h"
+#include "psqt.h"
 #include "types.h"
 
 #include "nnue/nnue_accumulator.h"
@@ -99,7 +100,6 @@ public:
   bool empty(Square s) const;
   template<PieceType Pt> int count(Color c) const;
   template<PieceType Pt> int count() const;
-  template<PieceType Pt> const Square* squares(Color c) const;
   template<PieceType Pt> Square square(Color c) const;
   bool is_on_semiopen_file(Color c, Square s) const;
 
@@ -113,7 +113,8 @@ public:
   Bitboard checkers() const;
   Bitboard blockers_for_king(Color c) const;
   Bitboard check_squares(PieceType pt) const;
-  bool is_discovery_check_on_king(Color c, Move m) const;
+  Bitboard pinners(Color c) const;
+  bool is_discovered_check_on_king(Color c, Move m) const;
 
   // Attacks to/from a given square
   Bitboard attackers_to(Square s) const;
@@ -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)
@@ -185,16 +185,11 @@ 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];
   Bitboard byColorBB[COLOR_NB];
   int pieceCount[PIECE_NB];
-  Square pieceList[PIECE_NB][16];
-  int index[SQUARE_NB];
   int castlingRightsMask[SQUARE_NB];
   Square castlingRookSquare[CASTLING_RIGHT_NB];
   Bitboard castlingPath[CASTLING_RIGHT_NB];
@@ -204,15 +199,8 @@ private:
   Thread* thisThread;
   StateInfo* st;
   bool chess960;
-
-  // List of pieces used in NNUE evaluation function
-  EvalList evalList;
 };
 
-namespace PSQT {
-  extern Score psq[PIECE_NB][SQUARE_NB];
-}
-
 extern std::ostream& operator<<(std::ostream& os, const Position& pos);
 
 inline Color Position::side_to_move() const {
@@ -260,13 +248,9 @@ template<PieceType Pt> inline int Position::count() const {
   return count<Pt>(WHITE) + count<Pt>(BLACK);
 }
 
-template<PieceType Pt> inline const Square* Position::squares(Color c) const {
-  return pieceList[make_piece(c, Pt)];
-}
-
 template<PieceType Pt> inline Square Position::square(Color c) const {
-  assert(pieceCount[make_piece(c, Pt)] == 1);
-  return squares<Pt>(c)[0];
+  assert(count<Pt>(c) == 1);
+  return lsb(pieces(c, Pt));
 }
 
 inline Square Position::ep_square() const {
@@ -309,11 +293,15 @@ 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];
 }
 
-inline bool Position::is_discovery_check_on_king(Color c, Move m) const {
+inline bool Position::is_discovered_check_on_king(Color c, Move m) const {
   return st->blockersForKing[c] & from_sq(m);
 }
 
@@ -331,7 +319,8 @@ inline int Position::pawns_on_same_color_squares(Color c, Square s) const {
 }
 
 inline Key Position::key() const {
-  return st->key;
+  return st->rule50 < 14 ? st->key
+                         : st->key ^ make_key((st->rule50 - 14) / 8);
 }
 
 inline Key Position::pawn_key() const {
@@ -380,7 +369,7 @@ inline bool Position::capture_or_promotion(Move m) const {
 inline bool Position::capture(Move m) const {
   assert(is_ok(m));
   // Castling is encoded as "king captures rook"
-  return (!empty(to_sq(m)) && type_of(m) != CASTLING) || type_of(m) == ENPASSANT;
+  return (!empty(to_sq(m)) && type_of(m) != CASTLING) || type_of(m) == EN_PASSANT;
 }
 
 inline Piece Position::captured_piece() const {
@@ -396,35 +385,25 @@ inline void Position::put_piece(Piece pc, Square s) {
   board[s] = pc;
   byTypeBB[ALL_PIECES] |= byTypeBB[type_of(pc)] |= s;
   byColorBB[color_of(pc)] |= s;
-  index[s] = pieceCount[pc]++;
-  pieceList[pc][index[s]] = s;
+  pieceCount[pc]++;
   pieceCount[make_piece(color_of(pc), ALL_PIECES)]++;
   psq += PSQT::psq[pc][s];
 }
 
 inline void Position::remove_piece(Square s) {
 
-  // WARNING: This is not a reversible operation. If we remove a piece in
-  // do_move() and then replace it in undo_move() we will put it at the end of
-  // the list and not in its original place, it means index[] and pieceList[]
-  // are not invariant to a do_move() + undo_move() sequence.
   Piece pc = board[s];
   byTypeBB[ALL_PIECES] ^= s;
   byTypeBB[type_of(pc)] ^= s;
   byColorBB[color_of(pc)] ^= s;
   /* board[s] = NO_PIECE;  Not needed, overwritten by the capturing one */
-  Square lastSquare = pieceList[pc][--pieceCount[pc]];
-  index[lastSquare] = index[s];
-  pieceList[pc][index[lastSquare]] = lastSquare;
-  pieceList[pc][pieceCount[pc]] = SQ_NONE;
+  pieceCount[pc]--;
   pieceCount[make_piece(color_of(pc), ALL_PIECES)]--;
   psq -= PSQT::psq[pc][s];
 }
 
 inline void Position::move_piece(Square from, Square to) {
 
-  // index[from] is not updated and becomes stale. This works as long as index[]
-  // is accessed just by known occupied squares.
   Piece pc = board[from];
   Bitboard fromTo = from | to;
   byTypeBB[ALL_PIECES] ^= fromTo;
@@ -432,8 +411,6 @@ inline void Position::move_piece(Square from, Square to) {
   byColorBB[color_of(pc)] ^= fromTo;
   board[from] = NO_PIECE;
   board[to] = pc;
-  index[to] = index[from];
-  pieceList[pc][index[to]] = to;
   psq += PSQT::psq[pc][to] - PSQT::psq[pc][from];
 }
 
@@ -446,20 +423,4 @@ 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/psqt.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,19 +16,22 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+
+#include "psqt.h"
+
 #include <algorithm>
 
-#include "types.h"
 #include "bitboard.h"
+#include "types.h"
 
-namespace PSQT {
 
-#define S(mg, eg) make_score(mg, eg)
+namespace
+{
 
-// Bonus[PieceType][Square / 2] contains Piece-Square scores. For each piece
-// type on a given square a (middlegame, endgame) score pair is assigned. Table
-// is defined for files A..D and white side: it is symmetric for black side and
-// second half of the files.
+auto constexpr S = make_score;
+
+// 'Bonus' contains Piece-Square parameters.
+// Scores are explicit for files A to D, implicitly mirrored for E to H.
 constexpr Score Bonus[][RANK_NB][int(FILE_NB) / 2] = {
   { },
   { },
@@ -43,14 +46,14 @@ constexpr Score Bonus[][RANK_NB][int(FILE_NB) / 2] = {
    { S(-201,-100), S(-83,-88), S(-56,-56), S(-26,-17) }
   },
   { // Bishop
-   { S(-53,-57), S( -5,-30), S( -8,-37), S(-23,-12) },
-   { S(-15,-37), S(  8,-13), S( 19,-17), S(  4,  1) },
-   { S( -7,-16), S( 21, -1), S( -5, -2), S( 17, 10) },
-   { S( -5,-20), S( 11, -6), S( 25,  0), S( 39, 17) },
-   { S(-12,-17), S( 29, -1), S( 22,-14), S( 31, 15) },
-   { S(-16,-30), S(  6,  6), S(  1,  4), S( 11,  6) },
-   { S(-17,-31), S(-14,-20), S(  5, -1), S(  0,  1) },
-   { S(-48,-46), S(  1,-42), S(-14,-37), S(-23,-24) }
+   { S(-37,-40), S(-4 ,-21), S( -6,-26), S(-16, -8) },
+   { S(-11,-26), S(  6, -9), S( 13,-12), S(  3,  1) },
+   { S(-5 ,-11), S( 15, -1), S( -4, -1), S( 12,  7) },
+   { S(-4 ,-14), S(  8, -4), S( 18,  0), S( 27, 12) },
+   { S(-8 ,-12), S( 20, -1), S( 15,-10), S( 22, 11) },
+   { S(-11,-21), S(  4,  4), S(  1,  3), S(  8,  4) },
+   { S(-12,-22), S(-10,-14), S(  4, -1), S(  0,  1) },
+   { S(-34,-32), S(  1,-29), S(-10,-26), S(-16,-17) }
   },
   { // Rook
    { S(-31, -9), S(-20,-13), S(-14,-10), S(-5, -9) },
@@ -64,13 +67,13 @@ constexpr Score Bonus[][RANK_NB][int(FILE_NB) / 2] = {
   },
   { // Queen
    { S( 3,-69), S(-5,-57), S(-5,-47), S( 4,-26) },
-   { S(-3,-55), S( 5,-31), S( 8,-22), S(12, -4) },
+   { S(-3,-54), S( 5,-31), S( 8,-22), S(12, -4) },
    { S(-3,-39), S( 6,-18), S(13, -9), S( 7,  3) },
    { S( 4,-23), S( 5, -3), S( 9, 13), S( 8, 24) },
    { S( 0,-29), S(14, -6), S(12,  9), S( 5, 21) },
-   { S(-4,-38), S(10,-18), S( 6,-12), S( 8,  1) },
+   { S(-4,-38), S(10,-18), S( 6,-11), S( 8,  1) },
    { S(-5,-50), S( 6,-27), S(10,-24), S( 8, -8) },
-   { S(-2,-75), S(-2,-52), S( 1,-43), S(-2,-36) }
+   { S(-2,-74), S(-2,-52), S( 1,-43), S(-2,-34) }
   },
   { // King
    { S(271,  1), S(327, 45), S(271, 85), S(198, 76) },
@@ -87,19 +90,22 @@ constexpr Score Bonus[][RANK_NB][int(FILE_NB) / 2] = {
 constexpr Score PBonus[RANK_NB][FILE_NB] =
   { // Pawn (asymmetric distribution)
    { },
-   { S(  3,-10), S(  3, -6), S( 10, 10), S( 19,  0), S( 16, 14), S( 19,  7), S(  7, -5), S( -5,-19) },
-   { S( -9,-10), S(-15,-10), S( 11,-10), S( 15,  4), S( 32,  4), S( 22,  3), S(  5, -6), S(-22, -4) },
-   { S( -4,  6), S(-23, -2), S(  6, -8), S( 20, -4), S( 40,-13), S( 17,-12), S(  4,-10), S( -8, -9) },
-   { S( 13, 10), S(  0,  5), S(-13,  4), S(  1, -5), S( 11, -5), S( -2, -5), S(-13, 14), S(  5,  9) },
-   { S(  5, 28), S(-12, 20), S( -7, 21), S( 22, 28), S( -8, 30), S( -5,  7), S(-15,  6), S( -8, 13) },
-   { S( -7,  0), S(  7,-11), S( -3, 12), S(-13, 21), S(  5, 25), S(-16, 19), S( 10,  4), S( -8,  7) }
+   { S(  2, -8), S(  4, -6), S( 11,  9), S( 18,  5), S( 16, 16), S( 21,  6), S(  9, -6), S( -3,-18) },
+   { S( -9, -9), S(-15, -7), S( 11,-10), S( 15,  5), S( 31,  2), S( 23,  3), S(  6, -8), S(-20, -5) },
+   { S( -3,  7), S(-20,  1), S(  8, -8), S( 19, -2), S( 39,-14), S( 17,-13), S(  2,-11), S( -5, -6) },
+   { S( 11, 12), S( -4,  6), S(-11,  2), S(  2, -6), S( 11, -5), S(  0, -4), S(-12, 14), S(  5,  9) },
+   { S(  3, 27), S(-11, 18), S( -6, 19), S( 22, 29), S( -8, 30), S( -5,  9), S(-14,  8), S(-11, 14) },
+   { S( -7, -1), S(  6,-14), S( -2, 13), S(-11, 22), S(  4, 24), S(-14, 17), S( 10,  7), S( -9,  7) }
   };
 
-#undef S
+} // namespace
+
+
+namespace PSQT
+{
 
 Score psq[PIECE_NB][SQUARE_NB];
 
-
 // PSQT::init() initializes piece-square tables: the white halves of the tables are
 // copied from Bonus[] and PBonus[], adding the piece value, then the black halves of
 // the tables are initialized by flipping and changing the sign of the white scores.
@@ -107,15 +113,15 @@ void init() {
 
   for (Piece pc : {W_PAWN, W_KNIGHT, W_BISHOP, W_ROOK, W_QUEEN, W_KING})
   {
-      Score score = make_score(PieceValue[MG][pc], PieceValue[EG][pc]);
+    Score score = make_score(PieceValue[MG][pc], PieceValue[EG][pc]);
 
-      for (Square s = SQ_A1; s <= SQ_H8; ++s)
-      {
-          File f = File(edge_distance(file_of(s)));
-          psq[ pc][s] = score + (type_of(pc) == PAWN ? PBonus[rank_of(s)][file_of(s)]
-                                                     : Bonus[pc][rank_of(s)][f]);
-          psq[~pc][flip_rank(s)] = -psq[pc][s];
-      }
+    for (Square s = SQ_A1; s <= SQ_H8; ++s)
+    {
+      File f = File(edge_distance(file_of(s)));
+      psq[ pc][s] = score + (type_of(pc) == PAWN ? PBonus[rank_of(s)][file_of(s)]
+                                                 : Bonus[pc][rank_of(s)][f]);
+      psq[~pc][flip_rank(s)] = -psq[pc][s];
+    }
   }
 }
 

src/psqt.h

@@ -0,0 +1,38 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
+
+  Stockfish is free software: you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation, either version 3 of the License, or
+  (at your option) any later version.
+
+  Stockfish is distributed in the hope that it will be useful,
+  but WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+  GNU General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+
+#ifndef PSQT_H_INCLUDED
+#define PSQT_H_INCLUDED
+
+
+#include "types.h"
+
+
+namespace PSQT
+{
+
+extern Score psq[PIECE_NB][SQUARE_NB];
+
+// Fill psqt array from a set of internally linked parameters
+extern void init();
+
+} // namespace PSQT
+
+
+#endif // PSQT_H_INCLUDED

src/search.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -48,6 +48,8 @@ struct Stack {
   int statScore;
   int moveCount;
   bool inCheck;
+  bool ttPv;
+  bool ttHit;
 };
 
 
@@ -69,7 +71,6 @@ struct RootMove {
   Value previousScore = -VALUE_INFINITE;
   int selDepth = 0;
   int tbRank = 0;
-  int bestMoveCount = 0;
   Value tbScore;
   std::vector<Move> pv;
 };

src/syzygy/tbprobe.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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
@@ -998,7 +1000,7 @@ uint8_t* set_sizes(PairsData* d, uint8_t* data) {
     // so that d->lowestSym[i] >= d->lowestSym[i+1] (when read as LittleEndian).
     // Starting from this we compute a base64[] table indexed by symbol length
     // and containing 64 bit values so that d->base64[i] >= d->base64[i+1].
-    // See http://www.eecs.harvard.edu/~michaelm/E210/huffman.pdf
+    // See https://en.wikipedia.org/wiki/Huffman_coding
     for (int i = d->base64.size() - 2; i >= 0; --i) {
         d->base64[i] = (d->base64[i + 1] + number<Sym, LittleEndian>(&d->lowestSym[i])
                                          - number<Sym, LittleEndian>(&d->lowestSym[i + 1])) / 2;
@@ -1139,7 +1141,7 @@ void* mapped(TBTable<Type>& e, const Position& pos) {
     if (e.ready.load(std::memory_order_acquire))
         return e.baseAddress; // Could be nullptr if file does not exist
 
-    std::unique_lock<std::mutex> lk(mutex);
+    std::scoped_lock<std::mutex> lk(mutex);
 
     if (e.ready.load(std::memory_order_relaxed)) // Recheck under lock
         return e.baseAddress;
@@ -1438,7 +1440,7 @@ WDLScore Tablebases::probe_wdl(Position& pos, ProbeState* result) {
 // If n = 100 immediately after a capture or pawn move, then the position
 // is also certainly a win, and during the whole phase until the next
 // capture or pawn move, the inequality to be preserved is
-// dtz + 50-movecounter <= 100.
+// dtz + 50-move-counter <= 100.
 //
 // In short, if a move is available resulting in dtz + 50-move-counter <= 99,
 // then do not accept moves leading to dtz + 50-move-counter == 100.

src/syzygy/tbprobe.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/thread.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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();
 }
 
@@ -238,16 +224,16 @@ Thread* ThreadPool::get_best_thread() const {
         votes[th->rootMoves[0].pv[0]] +=
             (th->rootMoves[0].score - minScore + 14) * int(th->completedDepth);
 
-          if (abs(bestThread->rootMoves[0].score) >= VALUE_TB_WIN_IN_MAX_PLY)
-          {
-              // Make sure we pick the shortest mate / TB conversion or stave off mate the longest
-              if (th->rootMoves[0].score > bestThread->rootMoves[0].score)
-                  bestThread = th;
-          }
-          else if (   th->rootMoves[0].score >= VALUE_TB_WIN_IN_MAX_PLY
-                   || (   th->rootMoves[0].score > VALUE_TB_LOSS_IN_MAX_PLY
-                       && votes[th->rootMoves[0].pv[0]] > votes[bestThread->rootMoves[0].pv[0]]))
-              bestThread = th;
+        if (abs(bestThread->rootMoves[0].score) >= VALUE_TB_WIN_IN_MAX_PLY)
+        {
+            // Make sure we pick the shortest mate / TB conversion or stave off mate the longest
+            if (th->rootMoves[0].score > bestThread->rootMoves[0].score)
+                bestThread = th;
+        }
+        else if (   th->rootMoves[0].score >= VALUE_TB_WIN_IN_MAX_PLY
+                 || (   th->rootMoves[0].score > VALUE_TB_LOSS_IN_MAX_PLY
+                     && votes[th->rootMoves[0].pv[0]] > votes[bestThread->rootMoves[0].pv[0]]))
+            bestThread = th;
     }
 
     return bestThread;

src/thread.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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;
@@ -73,6 +73,7 @@ public:
   CapturePieceToHistory captureHistory;
   ContinuationHistory continuationHistory[2][2];
   Score contempt;
+  int failedHighCnt;
 };
 
 

src/thread_win32_osx.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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(0.0084 + std::pow(ply + 3.0, 0.5) * 0.0042,
                            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/timeman.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/tt.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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);
   }
 }
 
@@ -61,11 +62,12 @@ void TranspositionTable::resize(size_t mbSize) {
 
   Threads.main()->wait_for_search_finished();
 
-  aligned_ttmem_free(mem);
+  aligned_large_pages_free(table);
 
   clusterCount = mbSize * 1024 * 1024 / sizeof(Cluster);
-  table = static_cast<Cluster*>(aligned_ttmem_alloc(clusterCount * sizeof(Cluster), mem));
-  if (!mem)
+
+  table = static_cast<Cluster*>(aligned_large_pages_alloc(clusterCount * sizeof(Cluster)));
+  if (!table)
   {
       std::cerr << "Failed to allocate " << mbSize
                 << "MB for transposition table." << std::endl;
@@ -119,22 +121,23 @@ 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
+          tte[i].genBound8 = uint8_t(generation8 | (tte[i].genBound8 & (GENERATION_DELTA - 1))); // 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
   TTEntry* replace = tte;
   for (int i = 1; i < ClusterSize; ++i)
       // Due to our packed storage format for generation and its cyclic
-      // nature we add 263 (256 is the modulus plus 7 to keep the unrelated
-      // lowest three bits from affecting the result) to calculate the entry
-      // age correctly even after generation8 overflows into the next cycle.
-      if (  replace->depth8 - ((263 + generation8 - replace->genBound8) & 0xF8)
-          >   tte[i].depth8 - ((263 + generation8 -   tte[i].genBound8) & 0xF8))
+      // nature we add GENERATION_CYCLE (256 is the modulus, plus what
+      // is needed to keep the unrelated lowest n bits from affecting
+      // the result) to calculate the entry age correctly even after
+      // generation8 overflows into the next cycle.
+      if (  replace->depth8 - ((GENERATION_CYCLE + generation8 - replace->genBound8) & GENERATION_MASK)
+          >   tte[i].depth8 - ((GENERATION_CYCLE + generation8 -   tte[i].genBound8) & GENERATION_MASK))
           replace = &tte[i];
 
   return found = false, replace;
@@ -149,7 +152,7 @@ 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 & GENERATION_MASK) == generation8;
 
   return cnt / ClusterSize;
 }

src/tt.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -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;
 };
 
 
@@ -72,9 +72,15 @@ class TranspositionTable {
 
   static_assert(sizeof(Cluster) == 32, "Unexpected Cluster size");
 
+  // Constants used to refresh the hash table periodically
+  static constexpr unsigned GENERATION_BITS  = 3;                                // nb of bits reserved for other things
+  static constexpr int      GENERATION_DELTA = (1 << GENERATION_BITS);           // increment for generation field
+  static constexpr int      GENERATION_CYCLE = 255 + (1 << GENERATION_BITS);     // cycle length
+  static constexpr int      GENERATION_MASK  = (0xFF << GENERATION_BITS) & 0xFF; // mask to pull out generation number
+
 public:
- ~TranspositionTable() { aligned_ttmem_free(mem); }
-  void new_search() { generation8 += 8; } // Lower 3 bits are used by PV flag and Bound
+ ~TranspositionTable() { aligned_large_pages_free(table); }
+  void new_search() { generation8 += GENERATION_DELTA; } // Lower bits are used for other things
   TTEntry* probe(const Key key, bool& found) const;
   int hashfull() const;
   void resize(size_t mbSize);
@@ -89,7 +95,6 @@ private:
 
   size_t clusterCount;
   Cluster* table;
-  void* mem;
   uint8_t generation8; // Size must be not bigger than TTEntry::genBound8
 };
 

src/tune.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/tune.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -130,9 +130,9 @@ class Tune {
     SetRange range;
   };
 
-  // Our facilty to fill the container, each Entry corresponds to a parameter to tune.
-  // We use variadic templates to deal with an unspecified number of entries, each one
-  // of a possible different type.
+  // Our facility to fill the container, each Entry corresponds to a parameter
+  // to tune. We use variadic templates to deal with an unspecified number of
+  // entries, each one of a possible different type.
   static std::string next(std::string& names, bool pop = true);
 
   int add(const SetRange&, std::string&&) { return 0; }

src/types.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -57,6 +57,12 @@
 /// _WIN32             Building on Windows (any)
 /// _WIN64             Building on Windows 64 bit
 
+#if defined(__GNUC__ ) && (__GNUC__ < 9 || (__GNUC__ == 9 && __GNUC_MINOR__ <= 2)) && defined(_WIN32) && !defined(__clang__)
+#define ALIGNAS_ON_STACK_VARIABLES_BROKEN
+#endif
+
+#define ASSERT_ALIGNED(ptr, alignment) assert(reinterpret_cast<uintptr_t>(ptr) % alignment == 0)
+
 #if defined(_WIN64) && defined(_MSC_VER) // No Makefile used
 #  include <intrin.h> // Microsoft header for _BitScanForward64()
 #  define IS_64BIT
@@ -107,7 +113,7 @@ constexpr int MAX_PLY   = 246;
 /// bit  6-11: origin square (from 0 to 63)
 /// bit 12-13: promotion piece type - 2 (from KNIGHT-2 to QUEEN-2)
 /// bit 14-15: special move flag: promotion (1), en passant (2), castling (3)
-/// NOTE: EN-PASSANT bit is set only when a pawn can be captured
+/// NOTE: en passant bit is set only when a pawn can be captured
 ///
 /// Special cases are MOVE_NONE and MOVE_NULL. We can sneak these in because in
 /// any normal move destination square is always different from origin square
@@ -121,7 +127,7 @@ enum Move : int {
 enum MoveType {
   NORMAL,
   PROMOTION = 1 << 14,
-  ENPASSANT = 2 << 14,
+  EN_PASSANT = 2 << 14,
   CASTLING  = 3 << 14
 };
 
@@ -178,7 +184,7 @@ 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,
@@ -196,27 +202,11 @@ enum PieceType {
 
 enum Piece {
   NO_PIECE,
-  W_PAWN = 1, W_KNIGHT, W_BISHOP, W_ROOK, W_QUEEN, W_KING,
-  B_PAWN = 9, B_KNIGHT, B_BISHOP, B_ROOK, B_QUEEN, B_KING,
+  W_PAWN = PAWN,     W_KNIGHT, W_BISHOP, W_ROOK, W_QUEEN, W_KING,
+  B_PAWN = PAWN + 8, B_KNIGHT, B_BISHOP, B_ROOK, B_QUEEN, B_KING,
   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 +222,8 @@ 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 +261,20 @@ 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 +324,6 @@ 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 +412,6 @@ 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 +448,6 @@ 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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -85,7 +85,7 @@ namespace {
     Position p;
     p.set(pos.fen(), Options["UCI_Chess960"], &states->back(), Threads.main());
 
-    Eval::verify_NNUE();
+    Eval::NNUE::verify();
 
     sync_cout << "\n" << Eval::trace(p) << sync_endl;
   }
@@ -170,7 +170,7 @@ namespace {
 
         if (token == "go" || token == "eval")
         {
-            cerr << "\nPosition: " << cnt++ << '/' << num << endl;
+            cerr << "\nPosition: " << cnt++ << '/' << num << " (" << pos.fen() << ")" << endl;
             if (token == "go")
             {
                go(pos, is, states);
@@ -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/uci.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by

src/ucioption.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,6 +21,7 @@
 #include <ostream>
 #include <sstream>
 
+#include "evaluate.h"
 #include "misc.h"
 #include "search.h"
 #include "thread.h"
@@ -40,8 +41,8 @@ void on_hash_size(const Option& o) { TT.resize(size_t(o)); }
 void on_logger(const Option& o) { start_logger(o); }
 void on_threads(const Option& o) { Threads.set(size_t(o)); }
 void on_tb_path(const Option& o) { Tablebases::init(o); }
-void on_use_NNUE(const Option& ) { Eval::init_NNUE(); }
-void on_eval_file(const Option& ) { Eval::init_NNUE(); }
+void on_use_NNUE(const Option& ) { Eval::NNUE::init(); }
+void on_eval_file(const Option& ) { Eval::NNUE::init(); }
 
 /// Our case insensitive less() function as required by UCI protocol
 bool CaseInsensitiveLess::operator() (const string& s1, const string& s2) const {
@@ -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

@@ -20,7 +20,7 @@ case $1 in
   --valgrind-thread)
     echo "valgrind-thread testing started"
     prefix=''
-    exeprefix='valgrind --error-exitcode=42'
+    exeprefix='valgrind --fair-sched=try --error-exitcode=42'
     postfix='1>/dev/null'
     threads="2"
   ;;
@@ -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