Stockfish 14

Official release version of Stockfish 14

Bench: 4770936

---

Today, we have the pleasure to announce Stockfish 14.

As usual, downloads will be freely available at https://stockfishchess.org

The engine is now significantly stronger than just a few months ago,
and wins four times more game pairs than it loses against the previous
release version [0]. Stockfish 14 is now at least 400 Elo ahead of
Stockfish 7, a top engine in 2016 [1]. During the last five years,
Stockfish has thus gained about 80 Elo per year.

Stockfish 14 evaluates positions more accurately than Stockfish 13 as
a result of two major steps forward in defining and training the
efficiently updatable neural network (NNUE) that provides the evaluation
for positions.

First, the collaboration with the Leela Chess Zero team - announced
previously [2] - has come to fruition. The LCZero team has provided a
collection of billions of positions evaluated by Leela that we have
combined with billions of positions evaluated by Stockfish to train the
NNUE net that powers Stockfish 14. The fact that we could use and combine
these datasets freely was essential for the progress made and demonstrates
the power of open source and open data [3].

Second, the architecture of the NNUE network was significantly updated:
the new network is not only larger, but more importantly, it deals better
with large material imbalances and can specialize for multiple phases of
the game [4]. A new project, kick-started by Gary Linscott and
Tomasz Sobczyk, led to a GPU accelerated net trainer written in
pytorch.[5] This tool allows for training high-quality nets in a couple
of hours.

Finally, this release features some search refinements, minor bug
fixes and additional improvements. For example, Stockfish is now about
90 Elo stronger for chess960 (Fischer random chess) at short time control.

The Stockfish project builds on a thriving community of enthusiasts
(thanks everybody!) that contribute their expertise, time, and resources
to build a free and open-source chess engine that is robust, widely
available, and very strong. We invite our chess fans to join the fishtest
testing framework and programmers to contribute to the project on
github [6].

Stay safe and enjoy chess!

The Stockfish team

[0] https://tests.stockfishchess.org/tests/view/60dae5363beab81350aca077
[1] https://nextchessmove.com/dev-builds
[2] https://stockfishchess.org/blog/2021/stockfish-13/
[3] https://lczero.org/blog/2021/06/the-importance-of-open-data/
[4] https://github.com/official-stockfish/Stockfish/commit/e8d64af1
[5] https://github.com/glinscott/nnue-pytorch/
[6] https://stockfishchess.org/get-involved/

.github/workflows/stockfish.yml

@@ -0,0 +1,201 @@
+name: Stockfish
+on:
+  push:
+    branches:
+      - master
+      - tools
+      - github_ci
+  pull_request:
+    branches:
+      - master
+      - tools
+jobs:
+  Stockfish:
+    name: ${{ matrix.config.name }}
+    runs-on: ${{ matrix.config.os }}
+    env:
+      COMPILER: ${{ matrix.config.compiler }}
+      COMP: ${{ matrix.config.comp }}
+      CXXFLAGS: "-Werror"
+    strategy:
+      matrix:
+        config:
+          - {
+              name: "Ubuntu 20.04 GCC",
+              os: ubuntu-20.04,
+              compiler: g++,
+              comp: gcc,
+              run_expensive_tests: true
+            }
+          - {
+              name: "Ubuntu 20.04 Clang",
+              os: ubuntu-20.04,
+              compiler: clang++,
+              comp: clang,
+              run_expensive_tests: false
+            }
+
+    defaults:
+      run:
+        working-directory: src
+    steps:
+      - uses: actions/checkout@v2
+        with:
+          fetch-depth: 0
+
+      - name: Download required packages
+        run: |
+          sudo apt update
+          sudo apt install expect valgrind g++-multilib
+
+      - name: Download the used network from the fishtest framework
+        run: |
+          make net
+
+      - name: Extract the bench number from the commit history
+        run: |
+          git log HEAD | grep "\b[Bb]ench[ :]\+[0-9]\{7\}" | head -n 1 | sed "s/[^0-9]*\([0-9]*\).*/\1/g" > git_sig
+          [ -s git_sig ] && echo "benchref=$(cat git_sig)" >> $GITHUB_ENV && echo "Reference bench:" $(cat git_sig) || echo "No bench found"
+
+      - name: Check compiler
+        run: |
+          $COMPILER -v
+
+      - name: Test help target
+        run: |
+          make help
+
+      # x86-32 tests
+
+      - name: Test debug x86-32 build
+        run: |
+          export CXXFLAGS="-Werror -D_GLIBCXX_DEBUG"
+          make clean
+          make -j2 ARCH=x86-32 optimize=no debug=yes build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-32 build
+        run: |
+          make clean
+          make -j2 ARCH=x86-32 build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-32-sse41-popcnt build
+        run: |
+          make clean
+          make -j2 ARCH=x86-32-sse41-popcnt build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-32-sse2 build
+        run: |
+          make clean
+          make -j2 ARCH=x86-32-sse2 build
+          ../tests/signature.sh $benchref
+
+      - name: Test general-32 build
+        run: |
+          make clean
+          make -j2 ARCH=general-32 build
+          ../tests/signature.sh $benchref
+
+      # x86-64 tests
+
+      - name: Test debug x86-64-modern build
+        run: |
+          export CXXFLAGS="-Werror -D_GLIBCXX_DEBUG"
+          make clean
+          make -j2 ARCH=x86-64-modern optimize=no debug=yes build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64-modern build
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-modern build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64-ssse3 build
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-ssse3 build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64-sse3-popcnt build
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-sse3-popcnt build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64 build
+        run: |
+          make clean
+          make -j2 ARCH=x86-64 build
+          ../tests/signature.sh $benchref
+
+      - name: Test general-64 build
+        run: |
+          make clean
+          make -j2 ARCH=general-64 build
+          ../tests/signature.sh $benchref
+
+      # x86-64 with newer extensions tests
+
+      - name: Compile x86-64-avx2 build
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-avx2 build
+
+      - name: Compile x86-64-bmi2 build
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-bmi2 build
+
+      - name: Compile x86-64-avx512 build
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-avx512 build
+
+      - name: Compile x86-64-vnni512 build
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-vnni512 build
+
+      - name: Compile x86-64-vnni256 build
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-vnni256 build
+
+      # Other tests
+
+      - name: Check perft and search reproducibility
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-modern build
+          ../tests/perft.sh
+          ../tests/reprosearch.sh
+
+      # Sanitizers
+
+      - name: Run under valgrind
+        if: ${{ matrix.config.run_expensive_tests }}
+        run: |
+          export CXXFLAGS="-O1 -fno-inline"
+          make clean
+          make -j2 ARCH=x86-64-modern debug=yes optimize=no build > /dev/null
+          ../tests/instrumented.sh --valgrind
+          ../tests/instrumented.sh --valgrind-thread
+
+      - name: Run with UB sanitizer
+        if: ${{ matrix.config.run_expensive_tests }}
+        run: |
+          export CXXFLAGS="-O1 -fno-inline"
+          make clean
+          make -j2 ARCH=x86-64-modern sanitize=undefined optimize=no debug=yes build > /dev/null
+          ../tests/instrumented.sh --sanitizer-undefined
+
+      - name: Run with thread sanitizer
+        if: ${{ matrix.config.run_expensive_tests }}
+        run: |
+          export CXXFLAGS="-O1 -fno-inline"
+          make clean
+          make -j2 ARCH=x86-64-modern sanitize=thread optimize=no debug=yes build > /dev/null
+          ../tests/instrumented.sh --sanitizer-thread

.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

@@ -1,101 +0,0 @@
-language: cpp
-dist: bionic
-
-matrix:
-  include:
-    - os: linux
-      compiler: gcc
-      addons:
-        apt:
-          packages: ['g++-8', 'g++-8-multilib', 'g++-multilib', 'valgrind', 'expect', 'curl']
-      env:
-        - COMPILER=g++-8
-        - COMP=gcc
-
-    - os: linux
-      compiler: clang
-      addons:
-        apt:
-          packages: ['clang-10', 'llvm-10-dev', 'g++-multilib', 'valgrind', 'expect', 'curl']
-      env:
-        - COMPILER=clang++-10
-        - COMP=clang
-
-    - os: osx
-      osx_image: xcode12
-      compiler: gcc
-      env:
-        - COMPILER=g++
-        - COMP=gcc
-
-    - os: osx
-      osx_image: xcode12
-      compiler: clang
-      env:
-        - COMPILER=clang++
-        - COMP=clang
-
-branches:
-  only:
-   - master
-
-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-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 $benchref; fi
-
-  # compile only for some more advanced architectures (might not run in travis)
-  - make clean && make -j2 ARCH=x86-64-avx2 build
-  - make clean && make -j2 ARCH=x86-64-bmi2 build
-  - make clean && make -j2 ARCH=x86-64-avx512 build
-  - make clean && make -j2 ARCH=x86-64-vnni512 build
-  - make clean && make -j2 ARCH=x86-64-vnni256 build
-
-  #
-  # Check perft and reproducible search
-  - make clean && make -j2 ARCH=x86-64-modern build
-  - ../tests/perft.sh
-  - ../tests/reprosearch.sh
-
-  #
-  # Valgrind
-  #
-  - export CXXFLAGS="-O1 -fno-inline"
-  - if [ -x "$(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-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

@@ -1,4 +1,4 @@
-# List of authors for Stockfish, as of August 4, 2020
+# List of authors for Stockfish, as of June 14, 2021
 
 # Founders of the Stockfish project and fishtest infrastructure
 Tord Romstad (romstad)
@@ -19,12 +19,15 @@ 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)
 Andy Duplain
+Antoine Champion (antoinechampion)
 Aram Tumanian (atumanian)
 Arjun Temurnikar
+Artem Solopiy (EntityFX)
 Auguste Pop
 Balint Pfliegel
 Ben Koshy (BKSpurgeon)
@@ -32,6 +35,7 @@ Bill Henry (VoyagerOne)
 Bojun Guo (noobpwnftw, Nooby)
 braich
 Brian Sheppard (SapphireBrand, briansheppard-toast)
+Bruno de Melo Costa (BM123499)
 Bryan Cross (crossbr)
 candirufish
 Chess13234
@@ -43,9 +47,12 @@ Daniel Dugovic (ddugovic)
 Dariusz Orzechowski (dorzechowski)
 David Zar
 Daylen Yang (daylen)
+Deshawn Mohan-Smith (GoldenRare)
+Dieter Dobbelaere (ddobbelaere)
 DiscanX
 Dominik Schlösser (domschl)
 double-beep
+Douglas Matos Gomes (dsmsgms)
 Eduardo Cáceres (eduherminio)
 Eelco de Groot (KingDefender)
 Elvin Liu (solarlight2)
@@ -84,11 +91,12 @@ 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)
 Joseph R. Prostko
+Julian Willemer (NightlyKing)
 jundery
 Justin Blanchard (UncombedCoconut)
 Kelly Wilson
@@ -96,6 +104,7 @@ Ken Takusagawa
 kinderchocolate
 Kiran Panditrao (Krgp)
 Kojirion
+Krystian Kuzniarek (kuzkry)
 Leonardo Ljubičić (ICCF World Champion)
 Leonid Pechenik (lp--)
 Linus Arver (listx)
@@ -108,8 +117,10 @@ 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)
@@ -160,10 +171,12 @@ Sergio Vieri (sergiovieri)
 sf-x
 Shane Booth (shane31)
 Shawn Varghese (xXH4CKST3RXx)
+Siad Daboul (Topologist)
 Stefan Geschwentner (locutus2)
 Stefano Cardanobile (Stefano80)
 Steinar Gunderson (sesse)
 Stéphane Nicolet (snicolet)
+Prokop Randáček (ProkopRandacek)
 Thanar2
 thaspel
 theo77186

README.md

@@ -1,6 +1,6 @@
 ## Overview
 
-[![Build Status](https://travis-ci.org/official-stockfish/Stockfish.svg?branch=master)](https://travis-ci.org/official-stockfish/Stockfish)
+[![Build Status](https://github.com/official-stockfish/Stockfish/actions/workflows/stockfish.yml/badge.svg)](https://github.com/official-stockfish/Stockfish/actions)
 [![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
@@ -12,7 +12,7 @@ about how to use Stockfish with it.
 
 The Stockfish engine features two evaluation functions for chess, the classical
 evaluation based on handcrafted terms, and the NNUE evaluation based on efficiently
-updateable neural networks. The classical evaluation runs efficiently on almost all
+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).
 
@@ -21,30 +21,28 @@ intrinsics available on most CPUs (sse2, avx2, neon, or similar).
 
 This distribution of Stockfish consists of the following files:
 
-  * Readme.md, the file you are currently reading.
+  * [Readme.md](https://github.com/official-stockfish/Stockfish/blob/master/README.md), the file you are currently reading.
 
-  * Copying.txt, a text file containing the GNU General Public License version 3.
+  * [Copying.txt](https://github.com/official-stockfish/Stockfish/blob/master/Copying.txt), a text file containing the GNU General Public License version 3.
 
-  * src, a subdirectory containing the full source code, including a Makefile
+  * [AUTHORS](https://github.com/official-stockfish/Stockfish/blob/master/AUTHORS), a text file with the list of authors for the project
+
+  * [src](https://github.com/official-stockfish/Stockfish/tree/master/src), a subdirectory containing the full source code, including a Makefile
     that can be used to compile Stockfish on Unix-like systems.
 
   * a file with the .nnue extension, storing the neural network for the NNUE
     evaluation. Binary distributions will have this file embedded.
 
-Note: to use the NNUE evaluation, the additional data file with neural network parameters
-needs to be available. Normally, this file is already embedded in the binary or it can be downloaded.
-The filename for the default (recommended) net can be found as the default
-value of the `EvalFile` UCI option, with the format `nn-[SHA256 first 12 digits].nnue`
-(for instance, `nn-c157e0a5755b.nnue`). This file can be downloaded from
-```
-https://tests.stockfishchess.org/api/nn/[filename]
-```
-replacing `[filename]` as needed.
+## The UCI protocol and available options
 
+The Universal Chess Interface (UCI) is a standard protocol used to communicate with
+a chess engine, and is the recommended way to do so for typical graphical user interfaces
+(GUI) or chess tools. Stockfish implements the majority of it options as described
+in [the UCI protocol](https://www.shredderchess.com/download/div/uci.zip).
 
-## UCI options
-
-Currently, Stockfish has the following UCI options:
+Developers can see the default values for UCI options available in Stockfish by typing
+`./stockfish uci` in a terminal, but the majority of users will typically see them and
+change them via a chess GUI. This is a list of available UCI options in Stockfish:
 
   * #### Threads
     The number of CPU threads used for searching a position. For best performance, set
@@ -53,6 +51,9 @@ 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.
 
@@ -108,8 +109,8 @@ 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
@@ -119,14 +120,6 @@ 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.
@@ -139,42 +132,78 @@ 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.
 
-## A note on classical and NNUE evaluation
+For developers the following non-standard commands might be of interest, mainly useful for debugging:
+
+  * #### bench *ttSize threads limit fenFile limitType evalType*
+    Performs a standard benchmark using various options. The signature of a version (standard node
+    count) is obtained using all defaults. `bench` is currently `bench 16 1 13 default depth mixed`.
+
+  * #### compiler
+    Give information about the compiler and environment used for building a binary.
+
+  * #### d
+    Display the current position, with ascii art and fen.
+
+  * #### eval
+    Return the evaluation of the current position.
+
+  * #### export_net [filename]
+    Exports the currently loaded network to a file.
+    If the currently loaded network is the embedded network and the filename
+    is not specified then the network is saved to the file matching the name
+    of the embedded network, as defined in evaluate.h.
+    If the currently loaded network is not the embedded network (some net set
+    through the UCI setoption) then the filename parameter is required and the
+    network is saved into that file.
+
+  * #### flip
+    Flips the side to move.
+
+
+## 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.
+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
@@ -182,14 +211,14 @@ 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.
@@ -198,8 +227,8 @@ 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.
 
@@ -207,17 +236,17 @@ 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
 
@@ -232,17 +261,17 @@ targets with corresponding descriptions.
 ```
     cd src
     make help
-    make build ARCH=x86-64-modern
     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.
 
-When reporting an issue or a bug, please tell us which version and
-compiler you used to create your executable. These informations can
-be found by typing the following commands in a console:
+When reporting an issue or a bug, please tell us which Stockfish version
+and which compiler you used to create your executable. This information
+can be found by typing the following command in a console:
 
 ```
     ./stockfish compiler
@@ -250,8 +279,8 @@ be found by typing the following commands in a console:
 
 ## Understanding the code base and participating in the project
 
-Stockfish's improvement over the last couple of years has been a great
-community effort. There are a few ways to help contribute to its growth.
+Stockfish's improvement over the last decade has been a great community
+effort. There are a few ways to help contribute to its growth.
 
 ### Donating hardware
 
@@ -272,8 +301,9 @@ 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.
 
@@ -281,16 +311,17 @@ 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*.
+[*Copying.txt*](https://github.com/official-stockfish/Stockfish/blob/master/Copying.txt).

Top CPU Contributors.txt

@@ -1,173 +1,205 @@
-Contributors with >10,000 CPU hours as of Sept 2, 2020
+Contributors to Fishtest with >10,000 CPU hours, as of Jun 29, 2021.
 Thank you!
 
-Username                  CPU Hours   Games played
---------------------------------------------------
-noobpwnftw                 19352969     1231459677
-mlang                        957168       61657446
-dew                          949885       56893432
-mibere                       703817       46865007
-crunchy                      427035       27344275
-cw                           416006       27521077
-JojoM                        415904       24479564
-fastgm                       404873       23953472
-CSU_Dynasty                  335774       22850550
-tvijlbrief                   335199       21871270
-Fisherman                    325053       21786603
-gvreuls                      311480       20751516
-ctoks                        275877       18710423
-velislav                     241267       15596372
-glinscott                    217799       13780820
-nordlandia                   211692       13484886
-bcross                       206213       14934233
-bking_US                     198894       11876016
-leszek                       189170       11446821
-mgrabiak                     183896       11778092
-drabel                       181408       12489478
-TueRens                      181349       12192000
-Thanar                       179852       12365359
-vdv                          175171        9881246
-robal                        166948       10702862
-spams                        157128       10319326
-marrco                       149947        9376421
-sqrt2                        147963        9724586
-vdbergh                      137041        8926915
-CoffeeOne                    136294        5004100
-malala                       136182        8002293
-mhoram                       128934        8177193
-davar                        122092        7960001
-dsmith                       122059        7570238
-xoto                         119696        8222144
-grandphish2                  116481        7582197
-Data                         113305        8220352
-BrunoBanani                  112960        7436849
-ElbertoOne                    99028        7023771
-MaZePallas                    98571        6362619
-brabos                        92118        6186135
-psk                           89957        5984901
-sunu                          88463        6007033
-sterni1971                    86948        5613788
-Vizvezdenec                   83752        5343724
-BRAVONE                       81239        5054681
-nssy                          76497        5259388
-teddybaer                     75125        5407666
-Pking_cda                     73776        5293873
-jromang                       70695        4940891
-solarlight                    70517        5028306
-dv8silencer                   70287        3883992
-Bobo1239                      68515        4652287
-racerschmacer                 67468        4935996
-manap                         66273        4121774
-tinker                        63458        4213726
-linrock                       59082        4516053
-robnjr                        57262        4053117
-Freja                         56938        3733019
-ttruscott                     56005        3679485
-renouve                       53811        3501516
-cuistot                       52532        3014920
-finfish                       51360        3370515
-eva42                         51272        3599691
-rkl                           50759        3840947
-rap                           49985        3219146
-pb00067                       49727        3298270
-ronaldjerum                   47654        3240695
-bigpen0r                      47278        3291647
-biffhero                      46564        3111352
-VoyagerOne                    45386        3445881
-speedycpu                     43842        3003273
-jbwiebe                       43305        2805433
-Antihistamine                 41788        2761312
-mhunt                         41735        2691355
-eastorwest                    40387        2812173
-homyur                        39893        2850481
-gri                           39871        2515779
-oryx                          38228        2941656
-0x3C33                        37773        2529097
-SC                            37290        2731014
-csnodgrass                    36207        2688994
-jmdana                        36108        2205261
-strelock                      34716        2074055
-Garf                          33800        2747562
-EthanOConnor                  33370        2090311
-slakovv                       32915        2021889
-Spprtr                        32591        2139601
-Prcuvu                        30377        2170122
-anst                          30301        2190091
-jkiiski                       30136        1904470
-hyperbolic.tom                29840        2017394
-Pyafue                        29650        1902349
-OuaisBla                      27629        1578000
-chriswk                       26902        1868317
-achambord                     26582        1767323
-Patrick_G                     26276        1801617
-yorkman                       26193        1992080
-SFTUser                       25182        1675689
-nabildanial                   24942        1519409
-Sharaf_DG                     24765        1786697
-ncfish1                       24411        1520927
-agg177                        23890        1395014
-JanErik                       23408        1703875
-Isidor                        23388        1680691
-Norabor                       22976        1587862
-cisco2015                     22880        1759669
-Zirie                         22542        1472937
-team-oh                       22272        1636708
-MazeOfGalious                 21978        1629593
-sg4032                        21945        1643065
-ianh2105                      21725        1632562
-xor12                         21628        1680365
-dex                           21612        1467203
-nesoneg                       21494        1463031
-horst.prack                   20878        1465656
-0xB00B1ES                     20590        1208666
-j3corre                       20405         941444
-Adrian.Schmidt123             20316        1281436
-wei                           19973        1745989
-rstoesser                     19569        1293588
-eudhan                        19274        1283717
-Ente                          19070        1373058
-jundery                       18445        1115855
-iisiraider                    18247        1101015
-ville                         17883        1384026
-chris                         17698        1487385
-purplefishies                 17595        1092533
-DragonLord                    17014        1162790
-dju                           16515         929427
-IgorLeMasson                  16064        1147232
-ako027ako                     15671        1173203
-Nikolay.IT                    15154        1068349
-Andrew Grant                  15114         895539
-yurikvelo                     15027        1165616
-OssumOpossum                  14857        1007129
-enedene                       14476         905279
-bpfliegel                     14298         884523
-jpulman                       13982         870599
-joster                        13794         950160
-Nesa92                        13786        1114691
-Dark_wizzie                   13422        1007152
-Hjax                          13350         900887
-Fifis                         13313         965473
-mabichito                     12903         749391
-thijsk                        12886         722107
-crocogoat                     12876        1048802
-AdrianSA                      12860         804972
-Flopzee                       12698         894821
-fatmurphy                     12547         853210
-SapphireBrand                 12416         969604
-modolief                      12386         896470
-scuzzi                        12362         833465
-pgontarz                      12151         848794
-stocky                        11954         699440
-mschmidt                      11941         803401
-infinity                      11470         727027
-torbjo                        11387         728873
-Thomas A. Anderson            11372         732094
-snicolet                      11106         869170
-amicic                        10779         733593
-rpngn                         10712         688203
-d64                           10680         771144
-basepi                        10637         744851
-jjoshua2                      10559         670905
-dzjp                          10343         732529
-ols                           10259         570669
-lbraesch                      10252         647825
+Username                 CPU Hours       Games played
+-----------------------------------------------------
+noobpwnftw                27649494         1834734733
+mlang                      1426107           89454622
+dew                        1380910           82831648
+mibere                      703840           46867607
+grandphish2                 692707           41737913
+tvijlbrief                  669642           42371594
+JojoM                       597778           35297180
+TueRens                     519226           31823562
+cw                          458421           30307421
+fastgm                      439667           25950040
+gvreuls                     436599           28177460
+crunchy                     427035           27344275
+CSU_Dynasty                 374765           25106278
+Fisherman                   326901           21822979
+ctoks                       325477           21767943
+velislav                    295343           18844324
+linrock                     292789           10624427
+bcross                      278584           19488961
+okrout                      262818           13803272
+pemo                        245982           11376085
+glinscott                   217799           13780820
+leszek                      212346           12959025
+nordlandia                  211692           13484886
+bking_US                    198894           11876016
+drabel                      196463           13450602
+robal                       195473           12375650
+mgrabiak                    187226           12016564
+Dantist                     183202           10990484
+Thanar                      179852           12365359
+vdv                         175274            9889046
+spams                       157128           10319326
+marrco                      150295            9402141
+sqrt2                       147963            9724586
+mhoram                      141278            8901241
+CoffeeOne                   137100            5024116
+vdbergh                     137041            8926915
+malala                      136182            8002293
+xoto                        133702            9156676
+davar                       122092            7960001
+dsmith                      122059            7570238
+Data                        113305            8220352
+BrunoBanani                 112960            7436849
+MaZePallas                  102823            6633619
+sterni1971                  100532            5880772
+ElbertoOne                   99028            7023771
+brabos                       92118            6186135
+oz                           92100            6486640
+psk                          89957            5984901
+amicic                       89156            5392305
+sunu                         88851            6028873
+Vizvezdenec                  83761            5344740
+0x3C33                       82614            5271253
+BRAVONE                      81239            5054681
+racerschmacer                80899            5759262
+cuistot                      80300            4606144
+nssy                         76497            5259388
+teddybaer                    75125            5407666
+Pking_cda                    73776            5293873
+jromang                      72192            5057715
+solarlight                   70517            5028306
+dv8silencer                  70287            3883992
+Bobo1239                     68515            4652287
+manap                        66273            4121774
+skiminki                     65088            4023328
+tinker                       64333            4268790
+sschnee                      60767            3500800
+qurashee                     57344            3168264
+robnjr                       57262            4053117
+Freja                        56938            3733019
+ttruscott                    56010            3680085
+rkl                          55132            4164467
+renouve                      53811            3501516
+finfish                      51360            3370515
+eva42                        51272            3599691
+rap                          49985            3219146
+pb00067                      49727            3298270
+ronaldjerum                  47654            3240695
+bigpen0r                     47653            3335327
+eastorwest                   47585            3221629
+biffhero                     46564            3111352
+VoyagerOne                   45476            3452465
+yurikvelo                    44834            3034550
+speedycpu                    43842            3003273
+jbwiebe                      43305            2805433
+Spprtr                       42279            2680153
+DesolatedDodo                42007            2447516
+Antihistamine                41788            2761312
+mhunt                        41735            2691355
+homyur                       39893            2850481
+gri                          39871            2515779
+Fifis                        38776            2529121
+oryx                         38724            2966648
+SC                           37290            2731014
+csnodgrass                   36207            2688994
+jmdana                       36157            2210661
+strelock                     34716            2074055
+rpngn                        33951            2057395
+Garf                         33922            2751802
+EthanOConnor                 33370            2090311
+slakovv                      32915            2021889
+manapbk                      30987            1810399
+Prcuvu                       30377            2170122
+anst                         30301            2190091
+jkiiski                      30136            1904470
+hyperbolic.tom               29840            2017394
+Pyafue                       29650            1902349
+Wolfgang                     29260            1658936
+zeryl                        28156            1579911
+OuaisBla                     27636            1578800
+DMBK                         27051            1999456
+chriswk                      26902            1868317
+achambord                    26582            1767323
+Patrick_G                    26276            1801617
+yorkman                      26193            1992080
+SFTUser                      25182            1675689
+nabildanial                  24942            1519409
+Sharaf_DG                    24765            1786697
+ncfish1                      24411            1520927
+rodneyc                      24227            1409514
+agg177                       23890            1395014
+JanErik                      23408            1703875
+Isidor                       23388            1680691
+Norabor                      23164            1591830
+cisco2015                    22897            1762669
+Zirie                        22542            1472937
+team-oh                      22272            1636708
+MazeOfGalious                21978            1629593
+sg4032                       21947            1643265
+ianh2105                     21725            1632562
+xor12                        21628            1680365
+dex                          21612            1467203
+nesoneg                      21494            1463031
+sphinx                       21211            1384728
+jjoshua2                     21001            1423089
+horst.prack                  20878            1465656
+Ente                         20865            1477066
+0xB00B1ES                    20590            1208666
+j3corre                      20405             941444
+Adrian.Schmidt123            20316            1281436
+wei                          19973            1745989
+MaxKlaxxMiner                19850            1009176
+rstoesser                    19569            1293588
+gopeto                       19491            1174952
+eudhan                       19274            1283717
+jundery                      18445            1115855
+megaman7de                   18377            1067540
+iisiraider                   18247            1101015
+ville                        17883            1384026
+chris                        17698            1487385
+purplefishies                17595            1092533
+dju                          17353             978595
+DragonLord                   17014            1162790
+IgorLeMasson                 16064            1147232
+ako027ako                    15671            1173203
+chuckstablers                15289             891576
+Nikolay.IT                   15154            1068349
+Andrew Grant                 15114             895539
+OssumOpossum                 14857            1007129
+Karby                        14808             867120
+enedene                      14476             905279
+bpfliegel                    14298             884523
+mpx86                        14019             759568
+jpulman                      13982             870599
+crocogoat                    13803            1117422
+joster                       13794             950160
+Nesa92                       13786            1114691
+Hjax                         13535             915487
+jsys14                       13459             785000
+Dark_wizzie                  13422            1007152
+mabichito                    12903             749391
+thijsk                       12886             722107
+AdrianSA                     12860             804972
+Flopzee                      12698             894821
+fatmurphy                    12547             853210
+Rudolphous                   12520             832340
+scuzzi                       12511             845761
+SapphireBrand                12416             969604
+modolief                     12386             896470
+Machariel                    12335             810784
+pgontarz                     12151             848794
+stocky                       11954             699440
+mschmidt                     11941             803401
+Maxim                        11543             836024
+infinity                     11470             727027
+torbjo                       11395             729145
+Thomas A. Anderson           11372             732094
+savage84                     11358             670860
+d64                          11263             789184
+MooTheCow                    11237             720174
+snicolet                     11106             869170
+ali-al-zhrani                11086             767926
+AndreasKrug                  10875             887457
+pirt                         10806             836519
+basepi                       10637             744851
+michaelrpg                   10508             739039
+dzjp                         10343             732529
+aga                          10302             622975
+ols                          10259             570669
+lbraesch                     10252             647825
+FormazChar                   10059             757283

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
@@ -33,13 +31,17 @@ PREFIX = /usr/local
 BINDIR = $(PREFIX)/bin
 
 ### Built-in benchmark for pgo-builds
-PGOBENCH = ./$(EXE) bench
+ifeq ($(SDE_PATH),)
+	PGOBENCH = ./$(EXE) bench
+else
+	PGOBENCH = $(SDE_PATH) -- ./$(EXE) bench
+endif
 
 ### Source and object files
 SRCS = benchmark.cpp bitbase.cpp bitboard.cpp endgame.cpp evaluate.cpp main.cpp \
 	material.cpp misc.cpp movegen.cpp movepick.cpp pawns.cpp position.cpp psqt.cpp \
 	search.cpp thread.cpp timeman.cpp tt.cpp uci.cpp ucioption.cpp tune.cpp syzygy/tbprobe.cpp \
-	nnue/evaluate_nnue.cpp nnue/features/half_kp.cpp
+	nnue/evaluate_nnue.cpp nnue/features/half_ka_v2.cpp
 
 OBJS = $(notdir $(SRCS:.cpp=.o))
 
@@ -59,9 +61,11 @@ endif
 # ----------------------------------------------------------------------------
 #
 # debug = yes/no      --- -DNDEBUG         --- Enable/Disable debug mode
-# sanitize = undefined/thread/no (-fsanitize )
+# sanitize = none/<sanitizer> ... (-fsanitize )
 #                     --- ( undefined )    --- enable undefined behavior checks
-#                     --- ( thread    )    --- enable threading error  checks
+#                     --- ( thread    )    --- enable threading error checks
+#                     --- ( address   )    --- enable memory access checks
+#                     --- ...etc...        --- see compiler documentation for supported sanitizers
 # optimize = yes/no   --- (-O3/-fast etc.) --- Enable/Disable optimizations
 # arch = (name)       --- (-arch)          --- Target architecture
 # bits = 64/32        --- -DIS_64BIT       --- 64-/32-bit operating system
@@ -82,6 +86,10 @@ endif
 # Note that Makefile is space sensitive, so when adding new architectures
 # or modifying existing flags, you have to make sure there are no extra spaces
 # at the end of the line for flag values.
+#
+# Example of use for these flags:
+# make build ARCH=x86-64-avx512 debug=on sanitize="address undefined"
+
 
 ### 2.1. General and architecture defaults
 
@@ -94,7 +102,7 @@ endif
 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 \
+                 x86-64 x86-32-sse41-popcnt x86-32-sse2 x86-32 ppc-64 ppc-32 e2k \
                  armv7 armv7-neon armv8 apple-silicon general-64 general-32))
    SUPPORTED_ARCH=true
 else
@@ -103,7 +111,7 @@ endif
 
 optimize = yes
 debug = no
-sanitize = no
+sanitize = none
 bits = 64
 prefetch = no
 popcnt = no
@@ -289,6 +297,17 @@ ifeq ($(ARCH),ppc-64)
 	prefetch = yes
 endif
 
+ifeq ($(findstring e2k,$(ARCH)),e2k)
+	arch = e2k
+	mmx = yes
+	bits = 64
+	sse = yes
+	sse2 = yes
+	ssse3 = yes
+	sse41 = yes
+	popcnt = yes
+endif
+
 endif
 
 ### ==========================================================================
@@ -366,9 +385,11 @@ ifeq ($(COMP),clang)
 
 	ifneq ($(KERNEL),Darwin)
 	ifneq ($(KERNEL),OpenBSD)
+	ifneq ($(KERNEL),FreeBSD)
 		LDFLAGS += -latomic
 	endif
 	endif
+	endif
 
 	ifeq ($(arch),$(filter $(arch),armv7 armv8))
 		ifeq ($(OS),Android)
@@ -381,19 +402,6 @@ ifeq ($(COMP),clang)
 	endif
 endif
 
-ifeq ($(comp),icc)
-	profile_make = icc-profile-make
-	profile_use = icc-profile-use
-else
-ifeq ($(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=10.14
 	LDFLAGS += -arch $(arch) -mmacosx-version-min=10.14
@@ -405,20 +413,30 @@ endif
 # Currently we don't know how to make PGO builds with the NDK yet.
 ifeq ($(COMP),ndk)
 	CXXFLAGS += -stdlib=libc++ -fPIE
+	comp=clang
 	ifeq ($(arch),armv7)
-		comp=armv7a-linux-androideabi16-clang
 		CXX=armv7a-linux-androideabi16-clang++
 		CXXFLAGS += -mthumb -march=armv7-a -mfloat-abi=softfp -mfpu=neon
 		STRIP=arm-linux-androideabi-strip
 	endif
 	ifeq ($(arch),armv8)
-		comp=aarch64-linux-android21-clang
 		CXX=aarch64-linux-android21-clang++
 		STRIP=aarch64-linux-android-strip
 	endif
 	LDFLAGS += -static-libstdc++ -pie -lm -latomic
 endif
 
+ifeq ($(comp),icc)
+	profile_make = icc-profile-make
+	profile_use = icc-profile-use
+else ifeq ($(comp),clang)
+	profile_make = clang-profile-make
+	profile_use = clang-profile-use
+else
+	profile_make = gcc-profile-make
+	profile_use = gcc-profile-use
+endif
+
 ### Travis CI script uses COMPILER to overwrite CXX
 ifdef COMPILER
 	COMPCXX=$(COMPILER)
@@ -461,9 +479,9 @@ else
 endif
 
 ### 3.2.2 Debugging with undefined behavior sanitizers
-ifneq ($(sanitize),no)
-        CXXFLAGS += -g3 -fsanitize=$(sanitize)
-        LDFLAGS += -fsanitize=$(sanitize)
+ifneq ($(sanitize),none)
+        CXXFLAGS += -g3 $(addprefix -fsanitize=,$(sanitize))
+        LDFLAGS += $(addprefix -fsanitize=,$(sanitize))
 endif
 
 ### 3.3 Optimization
@@ -482,6 +500,10 @@ ifeq ($(optimize),yes)
 			CXXFLAGS += -mdynamic-no-pic
 		endif
 	endif
+
+	ifeq ($(comp),clang)
+		CXXFLAGS += -fexperimental-new-pass-manager
+	endif
 endif
 
 ### 3.4 Bits
@@ -509,7 +531,6 @@ ifeq ($(popcnt),yes)
 	endif
 endif
 
-
 ifeq ($(avx2),yes)
 	CXXFLAGS += -DUSE_AVX2
 	ifeq ($(comp),$(filter $(comp),gcc clang mingw))
@@ -590,11 +611,8 @@ endif
 ### needs access to the optimization flags.
 ifeq ($(optimize),yes)
 ifeq ($(debug), no)
-	ifeq ($(COMP),ndk)
-		CXXFLAGS += -flto=thin
-		LDFLAGS += $(CXXFLAGS)
-	else ifeq ($(comp),clang)
-		CXXFLAGS += -flto=thin
+	ifeq ($(comp),clang)
+		CXXFLAGS += -flto
 		ifneq ($(findstring MINGW,$(KERNEL)),)
 			CXXFLAGS += -fuse-ld=lld
 		else ifneq ($(findstring MSYS,$(KERNEL)),)
@@ -614,7 +632,7 @@ ifeq ($(debug), no)
 			LDFLAGS += -save-temps
 		endif
 	else
-		CXXFLAGS += -flto=thin
+		CXXFLAGS += -flto
 		LDFLAGS += $(CXXFLAGS)
 	endif
 
@@ -680,6 +698,7 @@ help:
 	@echo "armv7                   > ARMv7 32-bit"
 	@echo "armv7-neon              > ARMv7 32-bit with popcnt and neon"
 	@echo "armv8                   > ARMv8 64-bit with popcnt and neon"
+	@echo "e2k                     > Elbrus 2000"
 	@echo "apple-silicon           > Apple silicon ARM64"
 	@echo "general-64              > unspecified 64-bit"
 	@echo "general-32              > unspecified 32-bit"
@@ -717,7 +736,7 @@ endif
         config-sanity icc-profile-use icc-profile-make gcc-profile-use gcc-profile-make \
         clang-profile-use clang-profile-make
 
-build: config-sanity net
+build: net config-sanity
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) all
 
 profile-build: net config-sanity objclean profileclean
@@ -780,6 +799,9 @@ profileclean:
 	@rm -rf profdir
 	@rm -f bench.txt *.gcda *.gcno ./syzygy/*.gcda ./nnue/*.gcda ./nnue/features/*.gcda *.s
 	@rm -f stockfish.profdata *.profraw
+	@rm -f stockfish.exe.lto_wrapper_args
+	@rm -f stockfish.exe.ltrans.out
+	@rm -f ./-lstdc++.res
 
 default:
 	help
@@ -790,7 +812,7 @@ default:
 
 all: $(EXE) .depend
 
-config-sanity:
+config-sanity: net
 	@echo ""
 	@echo "Config:"
 	@echo "debug: '$(debug)'"
@@ -822,11 +844,10 @@ config-sanity:
 	@echo "Testing config sanity. If this fails, try 'make help' ..."
 	@echo ""
 	@test "$(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)" = "ppc64" || test "$(arch)" = "ppc" || test "$(arch)" = "e2k" || \
 	 test "$(arch)" = "armv7" || test "$(arch)" = "armv8" || test "$(arch)" = "arm64"
 	@test "$(bits)" = "32" || test "$(bits)" = "64"
 	@test "$(prefetch)" = "yes" || test "$(prefetch)" = "no"
@@ -862,14 +883,15 @@ clang-profile-use:
 	all
 
 gcc-profile-make:
+	@mkdir -p profdir
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) \
-	EXTRACXXFLAGS='-fprofile-generate' \
+	EXTRACXXFLAGS='-fprofile-generate=profdir' \
 	EXTRALDFLAGS='-lgcov' \
 	all
 
 gcc-profile-use:
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) \
-	EXTRACXXFLAGS='-fprofile-use -fno-peel-loops -fno-tracer' \
+	EXTRACXXFLAGS='-fprofile-use=profdir -fno-peel-loops -fno-tracer' \
 	EXTRALDFLAGS='-lgcov' \
 	all
 

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
@@ -92,6 +92,8 @@ const vector<string> Defaults = {
 
 } // namespace
 
+namespace Stockfish {
+
 /// 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
@@ -164,5 +166,9 @@ vector<string> setup_bench(const Position& current, istream& is) {
           ++posCounter;
       }
 
+  list.emplace_back("setoption name Use NNUE value true");
+
   return list;
 }
+
+} // namespace Stockfish

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
@@ -23,6 +23,8 @@
 #include "bitboard.h"
 #include "types.h"
 
+namespace Stockfish {
+
 namespace {
 
   // There are 24 possible pawn squares: files A to D and ranks from 2 to 7.
@@ -66,7 +68,6 @@ namespace {
 
 } // namespace
 
-
 bool Bitbases::probe(Square wksq, Square wpsq, Square bksq, Color stm) {
 
   assert(file_of(wpsq) <= FILE_D);
@@ -96,7 +97,6 @@ void Bitbases::init() {
           KPKBitbase.set(idx);
 }
 
-
 namespace {
 
   KPKPosition::KPKPosition(unsigned idx) {
@@ -150,8 +150,8 @@ namespace {
     Bitboard b = attacks_bb<KING>(ksq[stm]);
 
     while (b)
-        r |= stm == WHITE ? db[index(BLACK, ksq[BLACK] , pop_lsb(&b), psq)]
-                          : db[index(WHITE, pop_lsb(&b),  ksq[WHITE], psq)];
+        r |= stm == WHITE ? db[index(BLACK, ksq[BLACK], pop_lsb(b), psq)]
+                          : db[index(WHITE, pop_lsb(b), ksq[WHITE], psq)];
 
     if (stm == WHITE)
     {
@@ -168,3 +168,5 @@ namespace {
   }
 
 } // namespace
+
+} // namespace Stockfish

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
@@ -22,11 +22,14 @@
 #include "bitboard.h"
 #include "misc.h"
 
+namespace Stockfish {
+
 uint8_t PopCnt16[1 << 16];
 uint8_t SquareDistance[SQUARE_NB][SQUARE_NB];
 
 Bitboard SquareBB[SQUARE_NB];
 Bitboard LineBB[SQUARE_NB][SQUARE_NB];
+Bitboard BetweenBB[SQUARE_NB][SQUARE_NB];
 Bitboard PseudoAttacks[PIECE_TYPE_NB][SQUARE_NB];
 Bitboard PawnAttacks[COLOR_NB][SQUARE_NB];
 
@@ -42,7 +45,6 @@ namespace {
 
 }
 
-
 /// 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.
 
@@ -55,7 +57,7 @@ inline Bitboard safe_destination(Square s, int step) {
 /// Bitboards::pretty() returns an ASCII representation of a bitboard suitable
 /// to be printed to standard output. Useful for debugging.
 
-const std::string Bitboards::pretty(Bitboard b) {
+std::string Bitboards::pretty(Bitboard b) {
 
   std::string s = "+---+---+---+---+---+---+---+---+\n";
 
@@ -106,12 +108,17 @@ void Bitboards::init() {
 
       for (PieceType pt : { BISHOP, ROOK })
           for (Square s2 = SQ_A1; s2 <= SQ_H8; ++s2)
+          {
               if (PseudoAttacks[pt][s1] & s2)
-                  LineBB[s1][s2] = (attacks_bb(pt, s1, 0) & attacks_bb(pt, s2, 0)) | s1 | s2;
+              {
+                  LineBB[s1][s2]    = (attacks_bb(pt, s1, 0) & attacks_bb(pt, s2, 0)) | s1 | s2;
+                  BetweenBB[s1][s2] = (attacks_bb(pt, s1, square_bb(s2)) & attacks_bb(pt, s2, square_bb(s1)));
+              }
+              BetweenBB[s1][s2] |= s2;
+          }
   }
 }
 
-
 namespace {
 
   Bitboard sliding_attack(PieceType pt, Square sq, Bitboard occupied) {
@@ -123,7 +130,7 @@ namespace {
     for (Direction d : (pt == ROOK ? RookDirections : BishopDirections))
     {
         Square s = sq;
-        while(safe_destination(s, d) && !(occupied & s))
+        while (safe_destination(s, d) && !(occupied & s))
             attacks |= (s += d);
     }
 
@@ -211,3 +218,5 @@ namespace {
     }
   }
 }
+
+} // namespace Stockfish

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
@@ -23,19 +23,21 @@
 
 #include "types.h"
 
+namespace Stockfish {
+
 namespace Bitbases {
 
 void init();
 bool probe(Square wksq, Square wpsq, Square bksq, Color us);
 
-}
+} // namespace Stockfish::Bitbases
 
 namespace Bitboards {
 
 void init();
-const std::string pretty(Bitboard b);
+std::string pretty(Bitboard b);
 
-}
+} // namespace Stockfish::Bitboards
 
 constexpr Bitboard AllSquares = ~Bitboard(0);
 constexpr Bitboard DarkSquares = 0xAA55AA55AA55AA55ULL;
@@ -73,6 +75,7 @@ extern uint8_t PopCnt16[1 << 16];
 extern uint8_t SquareDistance[SQUARE_NB][SQUARE_NB];
 
 extern Bitboard SquareBB[SQUARE_NB];
+extern Bitboard BetweenBB[SQUARE_NB][SQUARE_NB];
 extern Bitboard LineBB[SQUARE_NB][SQUARE_NB];
 extern Bitboard PseudoAttacks[PIECE_TYPE_NB][SQUARE_NB];
 extern Bitboard PawnAttacks[COLOR_NB][SQUARE_NB];
@@ -209,23 +212,29 @@ constexpr Bitboard adjacent_files_bb(Square s) {
 inline Bitboard line_bb(Square s1, Square s2) {
 
   assert(is_ok(s1) && is_ok(s2));
+
   return LineBB[s1][s2];
 }
 
 
-/// between_bb() returns a bitboard representing squares that are linearly
-/// between the two given squares (excluding the given squares). If the given
-/// squares are not on a same file/rank/diagonal, we return 0. For instance,
-/// between_bb(SQ_C4, SQ_F7) will return a bitboard with squares D5 and E6.
+/// between_bb(s1, s2) returns a bitboard representing the squares in the semi-open
+/// segment between the squares s1 and s2 (excluding s1 but including s2). If the
+/// given squares are not on a same file/rank/diagonal, it returns s2. For instance,
+/// between_bb(SQ_C4, SQ_F7) will return a bitboard with squares D5, E6 and F7, but
+/// between_bb(SQ_E6, SQ_F8) will return a bitboard with the square F8. This trick
+/// allows to generate non-king evasion moves faster: the defending piece must either
+/// interpose itself to cover the check or capture the checking piece.
 
 inline Bitboard between_bb(Square s1, Square s2) {
-  Bitboard b = line_bb(s1, s2) & ((AllSquares << s1) ^ (AllSquares << s2));
-  return b & (b - 1); //exclude lsb
+
+  assert(is_ok(s1) && is_ok(s2));
+
+  return BetweenBB[s1][s2];
 }
 
 
-/// forward_ranks_bb() returns a bitboard representing the squares on the ranks
-/// in front of the given one, from the point of view of the given color. For instance,
+/// forward_ranks_bb() returns a bitboard representing the squares on the ranks in
+/// front of the given one, from the point of view of the given color. For instance,
 /// forward_ranks_bb(BLACK, SQ_D3) will return the 16 squares on ranks 1 and 2.
 
 constexpr Bitboard forward_ranks_bb(Color c, Square s) {
@@ -412,13 +421,20 @@ inline Square msb(Bitboard b) {
 
 #endif
 
+/// least_significant_square_bb() returns the bitboard of the least significant
+/// square of a non-zero bitboard. It is equivalent to square_bb(lsb(bb)).
+
+inline Bitboard least_significant_square_bb(Bitboard b) {
+  assert(b);
+  return b & -b;
+}
 
 /// pop_lsb() finds and clears the least significant bit in a non-zero bitboard
 
-inline Square pop_lsb(Bitboard* b) {
-  assert(*b);
-  const Square s = lsb(*b);
-  *b &= *b - 1;
+inline Square pop_lsb(Bitboard& b) {
+  assert(b);
+  const Square s = lsb(b);
+  b &= b - 1;
   return s;
 }
 
@@ -430,4 +446,6 @@ inline Square frontmost_sq(Color c, Bitboard b) {
   return c == WHITE ? msb(b) : lsb(b);
 }
 
+} // namespace Stockfish
+
 #endif // #ifndef BITBOARD_H_INCLUDED

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
@@ -22,6 +22,8 @@
 #include "endgame.h"
 #include "movegen.h"
 
+namespace Stockfish {
+
 namespace {
 
   // Used to drive the king towards the edge of the board
@@ -553,8 +555,8 @@ 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 +640,8 @@ 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))
@@ -741,3 +743,5 @@ ScaleFactor Endgame<KPKP>::operator()(const Position& pos) const {
   // it's probably at least a draw even with the pawn.
   return Bitbases::probe(strongKing, strongPawn, weakKing, us) ? SCALE_FACTOR_NONE : SCALE_FACTOR_DRAW;
 }
+
+} // namespace Stockfish

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
@@ -28,6 +28,7 @@
 #include "position.h"
 #include "types.h"
 
+namespace Stockfish {
 
 /// EndgameCode lists all supported endgame functions by corresponding codes
 
@@ -120,4 +121,6 @@ namespace Endgames {
   }
 }
 
+} // namespace Stockfish
+
 #endif // #ifndef ENDGAME_H_INCLUDED

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
@@ -33,16 +33,18 @@
 #include "misc.h"
 #include "pawns.h"
 #include "thread.h"
+#include "timeman.h"
 #include "uci.h"
 #include "incbin/incbin.h"
 
 
-// Macro to embed the default NNUE file data in the engine binary (using incbin.h, by Dale Weiler).
+// 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 Microsof Visual Studio.
+// Note that this does not work in Microsoft Visual Studio.
 #if !defined(_MSC_VER) && !defined(NNUE_EMBEDDING_OFF)
   INCBIN(EmbeddedNNUE, EvalFileDefaultName);
 #else
@@ -53,22 +55,23 @@
 
 
 using namespace std;
-using namespace Eval::NNUE;
+
+namespace Stockfish {
 
 namespace Eval {
 
   bool useNNUE;
   string eval_file_loaded = "None";
 
-  /// init_NNUE() tries to load a nnue network at startup time, or when the engine
+  /// 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.
+  /// The name of the NNUE network is always retrieved from the EvalFile option.
   /// We search the given network in three locations: internally (the default
   /// network may be embedded in the binary), in the active working directory and
   /// in the engine directory. Distro packagers may define the DEFAULT_NNUE_DIRECTORY
   /// variable to have the engine search in a special directory in their distro.
 
-  void init_NNUE() {
+  void NNUE::init() {
 
     useNNUE = Options["Use NNUE"];
     if (!useNNUE)
@@ -111,8 +114,8 @@ namespace Eval {
         }
   }
 
-  /// verify_NNUE() verifies that the last net used was loaded successfully
-  void verify_NNUE() {
+  /// NNUE::verify() verifies that the last net used was loaded successfully
+  void NNUE::verify() {
 
     string eval_file = string(Options["EvalFile"]);
 
@@ -177,7 +180,7 @@ namespace Trace {
     else
         os << scores[t][WHITE] << " | " << scores[t][BLACK];
 
-    os << " | " << scores[t][WHITE] - scores[t][BLACK] << "\n";
+    os << " | " << scores[t][WHITE] - scores[t][BLACK] << " |\n";
     return os;
   }
 }
@@ -187,11 +190,9 @@ 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 NNUEThreshold1 =   Value(550);
-  constexpr Value NNUEThreshold2 =   Value(150);
+  constexpr Value LazyThreshold1    =  Value(1565);
+  constexpr Value LazyThreshold2    =  Value(1102);
+  constexpr Value SpaceThreshold    =  Value(11551);
 
   // KingAttackWeights[PieceType] contains king attack weights by piece type
   constexpr int KingAttackWeights[PIECE_TYPE_NB] = { 0, 0, 81, 52, 44, 10 };
@@ -199,7 +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)
@@ -207,19 +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
@@ -227,16 +234,15 @@ 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, 34), S(31, 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(9, 28), S(15, 31), S(17, 39), S(64, 70), S(171, 177), S(277, 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, 27) };
+  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
@@ -249,12 +255,12 @@ namespace {
     S(0, 0), S(3, 44), S(37, 68), S(42, 60), S(0, 39), S(58, 43)
   };
 
+  constexpr Value CorneredBishop = Value(50);
+
   // 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);
   constexpr Score Hanging             = S( 69, 36);
   constexpr Score KnightOnQueen       = S( 16, 11);
@@ -265,7 +271,6 @@ namespace {
   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);
@@ -384,15 +389,16 @@ 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))
@@ -419,13 +425,12 @@ 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
+            // 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);
@@ -434,7 +439,7 @@ namespace {
                 && 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))
@@ -447,14 +452,14 @@ 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
@@ -472,33 +477,40 @@ namespace {
                 {
                     Direction d = pawn_push(Us) + (file_of(s) == FILE_A ? EAST : WEST);
                     if (pos.piece_on(s + d) == make_piece(Us, PAWN))
-                        score -= !pos.empty(s + d + pawn_push(Us))                ? CorneredBishop * 4
-                                : pos.piece_on(s + d + d) == make_piece(Us, PAWN) ? CorneredBishop * 2
-                                                                                  : CorneredBishop;
+                        score -= !pos.empty(s + d + pawn_push(Us)) ? 4 * make_score(CorneredBishop, CorneredBishop)
+                                                                   : 3 * make_score(CorneredBishop, CorneredBishop);
                 }
             }
         }
 
-        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;
@@ -506,7 +518,7 @@ namespace {
                 score -= WeakQueen;
         }
     }
-    if (T)
+    if constexpr (T)
         Trace::add(Pt, Us, score);
 
     return score;
@@ -582,18 +594,18 @@ 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)
@@ -606,7 +618,7 @@ 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;
@@ -645,11 +657,11 @@ namespace {
     {
         b = (defended | weak) & (attackedBy[Us][KNIGHT] | attackedBy[Us][BISHOP]);
         while (b)
-            score += ThreatByMinor[type_of(pos.piece_on(pop_lsb(&b)))];
+            score += ThreatByMinor[type_of(pos.piece_on(pop_lsb(b)))];
 
         b = weak & attackedBy[Us][ROOK];
         while (b)
-            score += ThreatByRook[type_of(pos.piece_on(pop_lsb(&b)))];
+            score += ThreatByRook[type_of(pos.piece_on(pop_lsb(b)))];
 
         if (weak & attackedBy[Us][KING])
             score += ThreatByKing;
@@ -707,7 +719,7 @@ namespace {
         score += SliderOnQueen * popcount(b & safe & attackedBy2[Us]) * (1 + queenImbalance);
     }
 
-    if (T)
+    if constexpr (T)
         Trace::add(THREAT, Us, score);
 
     return score;
@@ -747,7 +759,7 @@ namespace {
 
     while (b)
     {
-        Square s = pop_lsb(&b);
+        Square s = pop_lsb(b);
 
         assert(!(pos.pieces(Them, PAWN) & forward_file_bb(Us, s + Up)));
 
@@ -777,14 +789,16 @@ 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
@@ -798,7 +812,7 @@ namespace {
         score += bonus - PassedFile * edge_distance(file_of(s));
     }
 
-    if (T)
+    if constexpr (T)
         Trace::add(PASSED, Us, score);
 
     return score;
@@ -833,11 +847,13 @@ 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;
@@ -852,7 +868,7 @@ 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);
@@ -889,28 +905,42 @@ namespace {
     Color strongSide = eg > VALUE_DRAW ? WHITE : BLACK;
     int sf = me->scale_factor(pos, strongSide);
 
-    // If scale factor is not already specific, scale down via general heuristics
+    // If scale factor is not already specific, scale up/down via general heuristics
     if (sf == SCALE_FACTOR_NORMAL)
     {
         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
@@ -918,7 +948,7 @@ 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));
@@ -948,7 +978,7 @@ namespace {
     // Initialize score by reading the incrementally updated scores included in
     // the position object (material + piece square tables) and the material
     // imbalance. Score is computed internally from the white point of view.
-    Score score = pos.psq_score() + me->imbalance() + pos.this_thread()->contempt;
+    Score score = pos.psq_score() + me->imbalance() + pos.this_thread()->trend;
 
     // Probe the pawn hash table
     pe = Pawns::probe(pos);
@@ -990,7 +1020,7 @@ 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());
@@ -1002,12 +1032,48 @@ make_v:
     v = (v / 16) * 16;
 
     // Side to move point of view
-    v = (pos.side_to_move() == WHITE ? v : -v) + Tempo;
+    v = (pos.side_to_move() == WHITE ? v : -v);
 
     return v;
   }
 
-} // namespace
+
+  /// Fisher Random Chess: correction for cornered bishops, to fix chess960 play with NNUE
+
+  Value fix_FRC(const Position& pos) {
+
+    constexpr Bitboard Corners =  1ULL << SQ_A1 | 1ULL << SQ_H1 | 1ULL << SQ_A8 | 1ULL << SQ_H8;
+
+    if (!(pos.pieces(BISHOP) & Corners))
+        return VALUE_ZERO;
+
+    int correction = 0;
+
+    if (   pos.piece_on(SQ_A1) == W_BISHOP
+        && pos.piece_on(SQ_B2) == W_PAWN)
+        correction += !pos.empty(SQ_B3) ? -CorneredBishop * 4
+                                        : -CorneredBishop * 3;
+
+    if (   pos.piece_on(SQ_H1) == W_BISHOP
+        && pos.piece_on(SQ_G2) == W_PAWN)
+        correction += !pos.empty(SQ_G3) ? -CorneredBishop * 4
+                                        : -CorneredBishop * 3;
+
+    if (   pos.piece_on(SQ_A8) == B_BISHOP
+        && pos.piece_on(SQ_B7) == B_PAWN)
+        correction += !pos.empty(SQ_B6) ? CorneredBishop * 4
+                                        : CorneredBishop * 3;
+
+    if (   pos.piece_on(SQ_H8) == B_BISHOP
+        && pos.piece_on(SQ_G7) == B_PAWN)
+        correction += !pos.empty(SQ_G6) ? CorneredBishop * 4
+                                        : CorneredBishop * 3;
+
+    return pos.side_to_move() == WHITE ?  Value(correction)
+                                       : -Value(correction);
+  }
+
+} // namespace Eval
 
 
 /// evaluate() is the evaluator for the outer world. It returns a static
@@ -1015,13 +1081,36 @@ make_v:
 
 Value Eval::evaluate(const Position& pos) {
 
-  bool classical = !Eval::useNNUE
-                ||  abs(eg_value(pos.psq_score())) * 16 > NNUEThreshold1 * (16 + pos.rule50_count());
-  Value v = classical ? Evaluation<NO_TRACE>(pos).value()
-                      : NNUE::evaluate(pos) * 5 / 4 + Tempo;
+  Value v;
 
-  if (classical && Eval::useNNUE && abs(v) * 16 < NNUEThreshold2 * (16 + pos.rule50_count()))
-      v = NNUE::evaluate(pos) * 5 / 4 + Tempo;
+  if (!Eval::useNNUE)
+      v = Evaluation<NO_TRACE>(pos).value();
+  else
+  {
+      // Scale and shift NNUE for compatibility with search and classical evaluation
+      auto  adjusted_NNUE = [&]()
+      {
+         int scale =   903
+                     + 32 * pos.count<PAWN>()
+                     + 32 * pos.non_pawn_material() / 1024;
+
+         Value nnue = NNUE::evaluate(pos, true) * scale / 1024;
+
+         if (pos.is_chess960())
+             nnue += fix_FRC(pos);
+
+         return nnue;
+      };
+
+      // If there is PSQ imbalance we use the classical eval, but we switch to
+      // NNUE eval faster when shuffling or if the material on the board is high.
+      int r50 = pos.rule50_count();
+      Value psq = Value(abs(eg_value(pos.psq_score())));
+      bool classical = psq * 5 > (750 + pos.non_pawn_material() / 64) * (5 + r50);
+
+      v = classical ? Evaluation<NO_TRACE>(pos).value()  // classical
+                    : adjusted_NNUE();                   // NNUE
+  }
 
   // Damp down the evaluation linearly when shuffling
   v = v * (100 - pos.rule50_count()) / 100;
@@ -1037,7 +1126,7 @@ Value Eval::evaluate(const Position& pos) {
 /// descriptions and values of each evaluation term. Useful for debugging.
 /// Trace scores are from white's point of view
 
-std::string Eval::trace(const Position& pos) {
+std::string Eval::trace(Position& pos) {
 
   if (pos.checkers())
       return "Final evaluation: none (in check)";
@@ -1049,44 +1138,55 @@ std::string Eval::trace(const Position& pos) {
 
   std::memset(scores, 0, sizeof(scores));
 
-  pos.this_thread()->contempt = SCORE_ZERO; // Reset any dynamic contempt
+  pos.this_thread()->trend = SCORE_ZERO; // Reset any dynamic contempt
 
   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);
-
-  v = pos.side_to_move() == WHITE ? v : -v;
-
-  ss << "\nClassical evaluation: " << to_cp(v) << " (white side)\n";
+     << " Contributing terms for the classical eval:\n"
+     << "+------------+-------------+-------------+-------------+\n"
+     << "|    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)
+     << "+------------+-------------+-------------+-------------+\n";
 
+  if (Eval::useNNUE)
+      ss << '\n' << NNUE::trace(pos) << '\n';
+
+  ss << std::showpoint << std::showpos << std::fixed << std::setprecision(2) << std::setw(15);
+
+  v = pos.side_to_move() == WHITE ? v : -v;
+  ss << "\nClassical evaluation   " << to_cp(v) << " (white side)\n";
   if (Eval::useNNUE)
   {
-      v = NNUE::evaluate(pos);
+      v = NNUE::evaluate(pos, false);
       v = pos.side_to_move() == WHITE ? v : -v;
-      ss << "\nNNUE evaluation:      " << to_cp(v) << " (white side)\n";
+      ss << "NNUE 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";
+  ss << "Final evaluation       " << to_cp(v) << " (white side)";
+  if (Eval::useNNUE)
+     ss << " [with scaled NNUE, hybrid, ...]";
+  ss << "\n";
 
   return ss.str();
 }
+
+} // namespace Stockfish

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
@@ -20,35 +20,43 @@
 #define EVALUATE_H_INCLUDED
 
 #include <string>
+#include <optional>
 
 #include "types.h"
 
+namespace Stockfish {
+
 class Position;
 
 namespace Eval {
 
-  std::string trace(const Position& pos);
+  std::string trace(Position& pos);
   Value evaluate(const Position& pos);
 
   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-82215d0fd0df.nnue"
+  #define EvalFileDefaultName   "nn-3475407dc199.nnue"
 
   namespace NNUE {
 
-    Value evaluate(const Position& pos);
-    Value compute_eval(const Position& pos);
-    void  update_eval(const Position& pos);
-    bool  load_eval(std::string streamName, std::istream& stream);
+    std::string trace(Position& pos);
+    Value evaluate(const Position& pos, bool adjusted = false);
+
+    void init();
+    void verify();
+
+    bool load_eval(std::string name, std::istream& stream);
+    bool save_eval(std::ostream& stream);
+    bool save_eval(const std::optional<std::string>& filename);
 
   } // namespace NNUE
 
 } // namespace Eval
 
+} // namespace Stockfish
+
 #endif // #ifndef EVALUATE_H_INCLUDED

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
@@ -21,15 +21,14 @@
 #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();
-}
+using namespace Stockfish;
 
 int main(int argc, char* argv[]) {
 
@@ -45,7 +44,7 @@ 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
@@ -24,32 +24,39 @@
 
 using namespace std;
 
+namespace Stockfish {
+
 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) };
@@ -67,7 +74,7 @@ namespace {
 
   bool is_KBPsK(const Position& pos, Color us) {
     return   pos.non_pawn_material(us) == BishopValueMg
-          && pos.count<PAWN  >(us) >= 1;
+          && pos.count<PAWN>(us) >= 1;
   }
 
   bool is_KQKRPs(const Position& pos, Color us) {
@@ -82,11 +89,11 @@ 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)
@@ -213,8 +220,10 @@ 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;
 }
 
 } // namespace Material
+
+} // namespace Stockfish

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
@@ -24,7 +24,7 @@
 #include "position.h"
 #include "types.h"
 
-namespace Material {
+namespace Stockfish::Material {
 
 /// Material::Entry contains various information about a material configuration.
 /// It contains a material imbalance evaluation, a function pointer to a special
@@ -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); }
 
@@ -57,15 +57,15 @@ 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;
 
 Entry* probe(const Position& pos);
 
-} // namespace Material
+} // namespace Stockfish::Material
 
 #endif // #ifndef MATERIAL_H_INCLUDED

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
@@ -51,7 +51,7 @@ 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))
+#if defined(__APPLE__) || defined(__ANDROID__) || defined(__OpenBSD__) || (defined(__GLIBCXX__) && !defined(_GLIBCXX_HAVE_ALIGNED_ALLOC) && !defined(_WIN32)) || defined(__e2k__)
 #define POSIXALIGNEDALLOC
 #include <stdlib.h>
 #endif
@@ -61,11 +61,13 @@ typedef bool(*fun3_t)(HANDLE, CONST GROUP_AFFINITY*, PGROUP_AFFINITY);
 
 using namespace std;
 
+namespace Stockfish {
+
 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 = "12";
+const string Version = "14";
 
 /// 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
@@ -138,7 +140,7 @@ public:
 /// the program was compiled) or "Stockfish <Version>", depending on whether
 /// Version is empty.
 
-const string engine_info(bool to_uci) {
+string engine_info(bool to_uci) {
 
   const string months("Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec");
   string month, day, year;
@@ -161,7 +163,7 @@ const string engine_info(bool to_uci) {
 
 /// compiler_info() returns a string trying to describe the compiler we use
 
-const std::string compiler_info() {
+std::string compiler_info() {
 
   #define stringify2(x) #x
   #define stringify(x) stringify2(x)
@@ -190,6 +192,18 @@ const std::string compiler_info() {
      compiler += "(version ";
      compiler += stringify(_MSC_FULL_VER) "." stringify(_MSC_BUILD);
      compiler += ")";
+  #elif defined(__e2k__) && defined(__LCC__)
+    #define dot_ver2(n) \
+      compiler += (char)'.'; \
+      compiler += (char)('0' + (n) / 10); \
+      compiler += (char)('0' + (n) % 10);
+
+     compiler += "MCST LCC ";
+     compiler += "(version ";
+     compiler += std::to_string(__LCC__ / 100);
+     dot_ver2(__LCC__ % 100)
+     dot_ver2(__LCC_MINOR__)
+     compiler += ")";
   #elif __GNUC__
      compiler += "g++ (GNUC) ";
      compiler += make_version_string(__GNUC__, __GNUC_MINOR__, __GNUC_PATCHLEVEL__);
@@ -357,27 +371,15 @@ void std_aligned_free(void* ptr) {
 #endif
 }
 
-/// aligned_ttmem_alloc() will return suitably aligned memory, if possible using large pages.
-/// The returned pointer is the aligned one, while the mem argument is the one that needs
-/// to be passed to free. With c++17 some of this functionality could be simplified.
+/// 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) {
+static void* aligned_large_pages_alloc_windows(size_t allocSize) {
 
-  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;
-#if defined(MADV_HUGEPAGE)
-  madvise(mem, allocSize, MADV_HUGEPAGE);
-#endif
-  return mem;
-}
-
-#elif defined(_WIN64)
-
-static void* aligned_ttmem_alloc_large_pages(size_t allocSize) {
+  #if !defined(_WIN64)
+    return nullptr;
+  #else
 
   HANDLE hProcessToken { };
   LUID luid { };
@@ -420,25 +422,14 @@ static void* aligned_ttmem_alloc_large_pages(size_t allocSize) {
   CloseHandle(hProcessToken);
 
   return mem;
+
+  #endif
 }
 
-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_windows(allocSize);
 
   // Fall back to regular, page aligned, allocation if necessary
   if (!mem)
@@ -449,37 +440,46 @@ 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))
   {
       DWORD err = GetLastError();
-      std::cerr << "Failed to free transposition table. Error code: 0x" <<
-          std::hex << err << std::dec << std::endl;
+      std::cerr << "Failed to free large page memory. Error code: 0x"
+                << std::hex << err
+                << std::dec << std::endl;
       exit(EXIT_FAILURE);
   }
 }
 
 #else
 
-void aligned_ttmem_free(void *mem) {
-  free(mem);
+void aligned_large_pages_free(void *mem) {
+  std_aligned_free(mem);
 }
 
 #endif
@@ -604,11 +604,10 @@ namespace CommandLine {
 string argv0;            // path+name of the executable binary, as given by argv[0]
 string binaryDirectory;  // path of the executable directory
 string workingDirectory; // path of the working directory
-string pathSeparator;    // Separator for our current OS
 
 void init(int argc, char* argv[]) {
     (void)argc;
-    string separator;
+    string pathSeparator;
 
     // extract the path+name of the executable binary
     argv0 = argv[0];
@@ -648,3 +647,5 @@ void init(int argc, char* argv[]) {
 
 
 } // namespace CommandLine
+
+} // namespace Stockfish

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
@@ -24,17 +24,20 @@
 #include <ostream>
 #include <string>
 #include <vector>
+#include <cstdint>
 
 #include "types.h"
 
-const std::string engine_info(bool to_uci = false);
-const std::string compiler_info();
+namespace Stockfish {
+
+std::string engine_info(bool to_uci = false);
+std::string compiler_info();
 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);
@@ -64,6 +67,67 @@ std::ostream& operator<<(std::ostream&, SyncCout);
 #define sync_endl std::endl << IO_UNLOCK
 
 
+// align_ptr_up() : get the first aligned element of an array.
+// 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));
+}
+
+
+// IsLittleEndian : true if and only if the binary is compiled on a little endian machine
+static inline const union { uint32_t i; char c[4]; } Le = { 0x01020304 };
+static inline const bool IsLittleEndian = (Le.c[0] == 4);
+
+
+template <typename T>
+class ValueListInserter {
+public:
+  ValueListInserter(T* v, std::size_t& s) :
+    values(v),
+    size(&s)
+  {
+  }
+
+  void push_back(const T& value) { values[(*size)++] = value; }
+private:
+  T* values;
+  std::size_t* size;
+};
+
+template <typename T, std::size_t MaxSize>
+class ValueList {
+
+public:
+  std::size_t size() const { return size_; }
+  void resize(std::size_t newSize) { size_ = newSize; }
+  void push_back(const T& value) { values_[size_++] = value; }
+  T& operator[](std::size_t index) { return values_[index]; }
+  T* begin() { return values_; }
+  T* end() { return values_ + size_; }
+  const T& operator[](std::size_t index) const { return values_[index]; }
+  const T* begin() const { return values_; }
+  const T* end() const { return values_ + size_; }
+  operator ValueListInserter<T>() { return ValueListInserter(values_, size_); }
+
+  void swap(ValueList& other) {
+    const std::size_t maxSize = std::max(size_, other.size_);
+    for (std::size_t i = 0; i < maxSize; ++i) {
+      std::swap(values_[i], other.values_[i]);
+    }
+    std::swap(size_, other.size_);
+  }
+
+private:
+  T values_[MaxSize];
+  std::size_t size_ = 0;
+};
+
 /// xorshift64star Pseudo-Random Number Generator
 /// This class is based on original code written and dedicated
 /// to the public domain by Sebastiano Vigna (2014).
@@ -131,4 +195,6 @@ namespace CommandLine {
   extern std::string workingDirectory; // path of the working directory
 }
 
+} // namespace Stockfish
+
 #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
@@ -21,24 +21,21 @@
 #include "movegen.h"
 #include "position.h"
 
+namespace Stockfish {
+
 namespace {
 
   template<GenType Type, Direction D>
-  ExtMove* make_promotions(ExtMove* moveList, Square to, Square ksq) {
+  ExtMove* make_promotions(ExtMove* moveList, Square to) {
 
     if (Type == CAPTURES || Type == EVASIONS || Type == NON_EVASIONS)
-    {
         *moveList++ = make<PROMOTION>(to - D, to, QUEEN);
-        if (attacks_bb<KNIGHT>(to) & ksq)
-            *moveList++ = make<PROMOTION>(to - D, to, KNIGHT);
-    }
 
     if (Type == QUIETS || Type == EVASIONS || Type == NON_EVASIONS)
     {
         *moveList++ = make<PROMOTION>(to - D, to, ROOK);
         *moveList++ = make<PROMOTION>(to - D, to, BISHOP);
-        if (!(attacks_bb<KNIGHT>(to) & ksq))
-            *moveList++ = make<PROMOTION>(to - D, to, KNIGHT);
+        *moveList++ = make<PROMOTION>(to - D, to, KNIGHT);
     }
 
     return moveList;
@@ -55,20 +52,16 @@ namespace {
     constexpr Direction UpRight  = (Us == WHITE ? NORTH_EAST : SOUTH_WEST);
     constexpr Direction UpLeft   = (Us == WHITE ? NORTH_WEST : SOUTH_EAST);
 
-    const Square ksq = pos.square<KING>(Them);
-    Bitboard emptySquares;
+    const Bitboard emptySquares = Type == QUIETS || Type == QUIET_CHECKS ? target : ~pos.pieces();
+    const Bitboard enemies      = Type == EVASIONS ? pos.checkers()
+                                : Type == CAPTURES ? target : pos.pieces(Them);
 
     Bitboard pawnsOn7    = pos.pieces(Us, PAWN) &  TRank7BB;
     Bitboard pawnsNotOn7 = pos.pieces(Us, PAWN) & ~TRank7BB;
 
-    Bitboard enemies = (Type == EVASIONS ? pos.pieces(Them) & target:
-                        Type == CAPTURES ? target : pos.pieces(Them));
-
     // Single and double pawn pushes, no promotions
     if (Type != CAPTURES)
     {
-        emptySquares = (Type == QUIETS || Type == QUIET_CHECKS ? target : ~pos.pieces());
-
         Bitboard b1 = shift<Up>(pawnsNotOn7)   & emptySquares;
         Bitboard b2 = shift<Up>(b1 & TRank3BB) & emptySquares;
 
@@ -80,33 +73,24 @@ namespace {
 
         if (Type == QUIET_CHECKS)
         {
-            b1 &= pawn_attacks_bb(Them, ksq);
-            b2 &= pawn_attacks_bb(Them, ksq);
-
-            // 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
-            // promotion has been already generated amongst the captures.
-            Bitboard dcCandidateQuiets = pos.blockers_for_king(Them) & pawnsNotOn7;
-            if (dcCandidateQuiets)
-            {
-                Bitboard dc1 = shift<Up>(dcCandidateQuiets) & emptySquares & ~file_bb(ksq);
-                Bitboard dc2 = shift<Up>(dc1 & TRank3BB) & emptySquares;
-
-                b1 |= dc1;
-                b2 |= dc2;
-            }
+            // To make a quiet check, you either make a direct check by pushing a pawn
+            // or push a blocker pawn that is not on the same file as the enemy king.
+            // Discovered check promotion has been already generated amongst the captures.
+            Square ksq = pos.square<KING>(Them);
+            Bitboard dcCandidatePawns = pos.blockers_for_king(Them) & ~file_bb(ksq);
+            b1 &= pawn_attacks_bb(Them, ksq) | shift<   Up>(dcCandidatePawns);
+            b2 &= pawn_attacks_bb(Them, ksq) | shift<Up+Up>(dcCandidatePawns);
         }
 
         while (b1)
         {
-            Square to = pop_lsb(&b1);
+            Square to = pop_lsb(b1);
             *moveList++ = make_move(to - Up, to);
         }
 
         while (b2)
         {
-            Square to = pop_lsb(&b2);
+            Square to = pop_lsb(b2);
             *moveList++ = make_move(to - Up - Up, to);
         }
     }
@@ -114,27 +98,24 @@ namespace {
     // Promotions and underpromotions
     if (pawnsOn7)
     {
-        if (Type == CAPTURES)
-            emptySquares = ~pos.pieces();
-
-        if (Type == EVASIONS)
-            emptySquares &= target;
-
         Bitboard b1 = shift<UpRight>(pawnsOn7) & enemies;
         Bitboard b2 = shift<UpLeft >(pawnsOn7) & enemies;
         Bitboard b3 = shift<Up     >(pawnsOn7) & emptySquares;
 
+        if (Type == EVASIONS)
+            b3 &= target;
+
         while (b1)
-            moveList = make_promotions<Type, UpRight>(moveList, pop_lsb(&b1), ksq);
+            moveList = make_promotions<Type, UpRight>(moveList, pop_lsb(b1));
 
         while (b2)
-            moveList = make_promotions<Type, UpLeft >(moveList, pop_lsb(&b2), ksq);
+            moveList = make_promotions<Type, UpLeft >(moveList, pop_lsb(b2));
 
         while (b3)
-            moveList = make_promotions<Type, Up     >(moveList, pop_lsb(&b3), ksq);
+            moveList = make_promotions<Type, Up     >(moveList, pop_lsb(b3));
     }
 
-    // Standard and en-passant captures
+    // Standard and en passant captures
     if (Type == CAPTURES || Type == EVASIONS || Type == NON_EVASIONS)
     {
         Bitboard b1 = shift<UpRight>(pawnsNotOn7) & enemies;
@@ -142,13 +123,13 @@ namespace {
 
         while (b1)
         {
-            Square to = pop_lsb(&b1);
+            Square to = pop_lsb(b1);
             *moveList++ = make_move(to - UpRight, to);
         }
 
         while (b2)
         {
-            Square to = pop_lsb(&b2);
+            Square to = pop_lsb(b2);
             *moveList++ = make_move(to - UpLeft, to);
         }
 
@@ -156,10 +137,8 @@ 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 +146,7 @@ namespace {
             assert(b1);
 
             while (b1)
-                *moveList++ = make<ENPASSANT>(pop_lsb(&b1), pos.ep_square());
+                *moveList++ = make<EN_PASSANT>(pop_lsb(b1), pos.ep_square());
         }
     }
 
@@ -180,27 +159,19 @@ namespace {
 
     static_assert(Pt != KING && Pt != PAWN, "Unsupported piece type in generate_moves()");
 
-    const Square* pl = pos.squares<Pt>(Us);
+    Bitboard bb = pos.pieces(Us, Pt);
 
-    for (Square from = *pl; from != SQ_NONE; from = *++pl)
+    while (bb)
     {
-        if (Checks)
-        {
-            if (    (Pt == BISHOP || Pt == ROOK || Pt == QUEEN)
-                && !(attacks_bb<Pt>(from) & target & pos.check_squares(Pt)))
-                continue;
-
-            if (pos.blockers_for_king(~Us) & from)
-                continue;
-        }
-
+        Square from = pop_lsb(bb);
         Bitboard b = attacks_bb<Pt>(from, pos.pieces()) & target;
 
-        if (Checks)
+        // To check, you either move freely a blocker or make a direct check.
+        if (Checks && (Pt == QUEEN || !(pos.blockers_for_king(~Us) & from)))
             b &= pos.check_squares(Pt);
 
         while (b)
-            *moveList++ = make_move(from, pop_lsb(&b));
+            *moveList++ = make_move(from, pop_lsb(b));
     }
 
     return moveList;
@@ -209,45 +180,38 @@ namespace {
 
   template<Color Us, GenType Type>
   ExtMove* generate_all(const Position& pos, ExtMove* moveList) {
-    constexpr bool Checks = Type == QUIET_CHECKS; // Reduce template instantations
+
+    static_assert(Type != LEGAL, "Unsupported type in generate_all()");
+
+    constexpr bool Checks = Type == QUIET_CHECKS; // Reduce template instantiations
+    const Square ksq = pos.square<KING>(Us);
     Bitboard target;
 
-    switch (Type)
+    // Skip generating non-king moves when in double check
+    if (Type != EVASIONS || !more_than_one(pos.checkers()))
     {
-        case CAPTURES:
-            target =  pos.pieces(~Us);
-            break;
-        case QUIETS:
-        case QUIET_CHECKS:
-            target = ~pos.pieces();
-            break;
-        case EVASIONS:
-        {
-            Square checksq = lsb(pos.checkers());
-            target = between_bb(pos.square<KING>(Us), checksq) | checksq;
-            break;
-        }
-        case NON_EVASIONS:
-            target = ~pos.pieces(Us);
-            break;
-        default:
-            static_assert(true, "Unsupported type in generate_all()");
+        target = Type == EVASIONS     ?  between_bb(ksq, lsb(pos.checkers()))
+               : Type == NON_EVASIONS ? ~pos.pieces( Us)
+               : Type == CAPTURES     ?  pos.pieces(~Us)
+                                      : ~pos.pieces(   ); // QUIETS || QUIET_CHECKS
+
+        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_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);
-
-    if (Type != QUIET_CHECKS && Type != EVASIONS)
+    if (!Checks || pos.blockers_for_king(~Us) & ksq)
     {
-        Square ksq = pos.square<KING>(Us);
-        Bitboard b = attacks_bb<KING>(ksq) & target;
-        while (b)
-            *moveList++ = make_move(ksq, pop_lsb(&b));
+        Bitboard b = attacks_bb<KING>(ksq) & (Type == EVASIONS ? ~pos.pieces(Us) : target);
+        if (Checks)
+            b &= ~attacks_bb<QUEEN>(pos.square<KING>(~Us));
 
-        if ((Type != CAPTURES) && pos.can_castle(Us & ANY_CASTLING))
+        while (b)
+            *moveList++ = make_move(ksq, pop_lsb(b));
+
+        if ((Type == QUIETS || Type == NON_EVASIONS) && pos.can_castle(Us & ANY_CASTLING))
             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));
@@ -259,8 +223,10 @@ namespace {
 } // 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)
+/// <CAPTURES>     Generates all pseudo-legal captures plus queen promotions
+/// <QUIETS>       Generates all pseudo-legal non-captures and underpromotions
+/// <EVASIONS>     Generates all pseudo-legal check evasions when the side to move is in check
+/// <QUIET_CHECKS> Generates all pseudo-legal non-captures giving check, except castling and promotions
 /// <NON_EVASIONS> Generates all pseudo-legal captures and non-captures
 ///
 /// Returns a pointer to the end of the move list.
@@ -268,8 +234,8 @@ namespace {
 template<GenType Type>
 ExtMove* generate(const Position& pos, ExtMove* moveList) {
 
-  static_assert(Type == CAPTURES || Type == QUIETS || Type == NON_EVASIONS, "Unsupported type in generate()");
-  assert(!pos.checkers());
+  static_assert(Type != LEGAL, "Unsupported type in generate()");
+  assert((Type == EVASIONS) == (bool)pos.checkers());
 
   Color us = pos.side_to_move();
 
@@ -280,70 +246,11 @@ ExtMove* generate(const Position& pos, ExtMove* moveList) {
 // Explicit template instantiations
 template ExtMove* generate<CAPTURES>(const Position&, ExtMove*);
 template ExtMove* generate<QUIETS>(const Position&, ExtMove*);
+template ExtMove* generate<EVASIONS>(const Position&, ExtMove*);
+template ExtMove* generate<QUIET_CHECKS>(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.
-template<>
-ExtMove* generate<QUIET_CHECKS>(const Position& pos, ExtMove* moveList) {
-
-  assert(!pos.checkers());
-
-  Color us = pos.side_to_move();
-  Bitboard dc = pos.blockers_for_king(~us) & pos.pieces(us) & ~pos.pieces(PAWN);
-
-  while (dc)
-  {
-     Square from = pop_lsb(&dc);
-     PieceType pt = type_of(pos.piece_on(from));
-
-     Bitboard b = attacks_bb(pt, from, pos.pieces()) & ~pos.pieces();
-
-     if (pt == KING)
-         b &= ~attacks_bb<QUEEN>(pos.square<KING>(~us));
-
-     while (b)
-         *moveList++ = make_move(from, pop_lsb(&b));
-  }
-
-  return us == WHITE ? generate_all<WHITE, QUIET_CHECKS>(pos, moveList)
-                     : generate_all<BLACK, QUIET_CHECKS>(pos, moveList);
-}
-
-
-/// generate<EVASIONS> generates all pseudo-legal check evasions when the side
-/// to move is in check. Returns a pointer to the end of the move list.
-template<>
-ExtMove* generate<EVASIONS>(const Position& pos, ExtMove* moveList) {
-
-  assert(pos.checkers());
-
-  Color us = pos.side_to_move();
-  Square ksq = pos.square<KING>(us);
-  Bitboard sliderAttacks = 0;
-  Bitboard sliders = pos.checkers() & ~pos.pieces(KNIGHT, PAWN);
-
-  // Find all the squares attacked by slider checkers. We will remove them from
-  // the king evasions in order to skip known illegal moves, which avoids any
-  // useless legality checks later on.
-  while (sliders)
-      sliderAttacks |= line_bb(ksq, pop_lsb(&sliders)) & ~pos.checkers();
-
-  // Generate evasions for king, capture and non capture moves
-  Bitboard b = attacks_bb<KING>(ksq) & ~pos.pieces(us) & ~sliderAttacks;
-  while (b)
-      *moveList++ = make_move(ksq, pop_lsb(&b));
-
-  if (more_than_one(pos.checkers()))
-      return moveList; // Double check, only a king move can save the day
-
-  // Generate blocking evasions or captures of the checking piece
-  return us == WHITE ? generate_all<WHITE, EVASIONS>(pos, moveList)
-                     : generate_all<BLACK, EVASIONS>(pos, moveList);
-}
-
-
 /// generate<LEGAL> generates all the legal moves in the given position
 
 template<>
@@ -357,7 +264,7 @@ 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
@@ -365,3 +272,5 @@ ExtMove* generate<LEGAL>(const Position& pos, ExtMove* moveList) {
 
   return moveList;
 }
+
+} // namespace Stockfish

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
@@ -23,6 +23,8 @@
 
 #include "types.h"
 
+namespace Stockfish {
+
 class Position;
 
 enum GenType {
@@ -70,4 +72,6 @@ private:
   ExtMove moveList[MAX_MOVES], *last;
 };
 
+} // namespace Stockfish
+
 #endif // #ifndef MOVEGEN_H_INCLUDED

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
@@ -20,6 +20,8 @@
 
 #include "movepick.h"
 
+namespace Stockfish {
+
 namespace {
 
   enum Stages {
@@ -73,8 +75,9 @@ 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 +101,15 @@ 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 +119,8 @@ 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 +144,7 @@ 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) {
 
@@ -262,3 +265,5 @@ top:
   assert(false);
   return MOVE_NONE; // Silence warning
 }
+
+} // namespace Stockfish

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
@@ -27,6 +27,8 @@
 #include "position.h"
 #include "types.h"
 
+namespace Stockfish {
+
 /// StatsEntry stores the stat table value. It is usually a number but could
 /// be a move or even a nested history. We use a class instead of naked value
 /// to directly call history update operator<<() on the entry so to use stats
@@ -84,7 +86,7 @@ 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
@@ -108,12 +110,12 @@ 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 };
@@ -156,4 +158,6 @@ private:
   ExtMove moves[MAX_MOVES];
 };
 
+} // namespace Stockfish
+
 #endif // #ifndef MOVEPICK_H_INCLUDED

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

@@ -1,54 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 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/>.
-*/
-
-// Definition of input features and network structure used in NNUE evaluation function
-
-#ifndef NNUE_HALFKP_256X2_32_32_H_INCLUDED
-#define NNUE_HALFKP_256X2_32_32_H_INCLUDED
-
-#include "../features/feature_set.h"
-#include "../features/half_kp.h"
-
-#include "../layers/input_slice.h"
-#include "../layers/affine_transform.h"
-#include "../layers/clipped_relu.h"
-
-namespace Eval::NNUE {
-
-// Input features used in evaluation function
-using RawFeatures = Features::FeatureSet<
-    Features::HalfKP<Features::Side::kFriend>>;
-
-// Number of input feature dimensions after conversion
-constexpr IndexType kTransformedFeatureDimensions = 256;
-
-namespace Layers {
-
-// Define network structure
-using InputLayer = InputSlice<kTransformedFeatureDimensions * 2>;
-using HiddenLayer1 = ClippedReLU<AffineTransform<InputLayer, 32>>;
-using HiddenLayer2 = ClippedReLU<AffineTransform<HiddenLayer1, 32>>;
-using OutputLayer = AffineTransform<HiddenLayer2, 1>;
-
-}  // namespace Layers
-
-using Network = Layers::OutputLayer;
-
-}  // namespace Eval::NNUE
-
-#endif // #ifndef NNUE_HALFKP_256X2_32_32_H_INCLUDED

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
@@ -20,149 +20,424 @@
 
 #include <iostream>
 #include <set>
+#include <sstream>
+#include <iomanip>
+#include <fstream>
 
 #include "../evaluate.h"
 #include "../position.h"
 #include "../misc.h"
 #include "../uci.h"
+#include "../types.h"
 
 #include "evaluate_nnue.h"
 
-namespace Eval::NNUE {
-
-  uint32_t kpp_board_index[PIECE_NB][COLOR_NB] = {
-   // convention: W - us, B - them
-   // viewed from other side, W and B are reversed
-      { PS_NONE,     PS_NONE     },
-      { PS_W_PAWN,   PS_B_PAWN   },
-      { PS_W_KNIGHT, PS_B_KNIGHT },
-      { PS_W_BISHOP, PS_B_BISHOP },
-      { PS_W_ROOK,   PS_B_ROOK   },
-      { PS_W_QUEEN,  PS_B_QUEEN  },
-      { PS_W_KING,   PS_B_KING   },
-      { PS_NONE,     PS_NONE     },
-      { PS_NONE,     PS_NONE     },
-      { PS_B_PAWN,   PS_W_PAWN   },
-      { PS_B_KNIGHT, PS_W_KNIGHT },
-      { PS_B_BISHOP, PS_W_BISHOP },
-      { PS_B_ROOK,   PS_W_ROOK   },
-      { PS_B_QUEEN,  PS_W_QUEEN  },
-      { PS_B_KING,   PS_W_KING   },
-      { PS_NONE,     PS_NONE     }
-  };
+namespace Stockfish::Eval::NNUE {
 
   // Input feature converter
-  AlignedPtr<FeatureTransformer> feature_transformer;
+  LargePagePtr<FeatureTransformer> featureTransformer;
 
   // Evaluation function
-  AlignedPtr<Network> network;
+  AlignedPtr<Network> network[LayerStacks];
 
   // Evaluation function file name
   std::string fileName;
+  std::string netDescription;
 
   namespace Detail {
 
   // Initialize the evaluation function parameters
   template <typename T>
-  void Initialize(AlignedPtr<T>& pointer) {
+  void initialize(AlignedPtr<T>& pointer) {
 
     pointer.reset(reinterpret_cast<T*>(std_aligned_alloc(alignof(T), sizeof(T))));
     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 read_parameters(std::istream& stream, T& reference) {
 
     std::uint32_t header;
     header = read_little_endian<std::uint32_t>(stream);
-    if (!stream || header != T::GetHashValue()) return false;
-    return pointer->ReadParameters(stream);
+    if (!stream || header != T::get_hash_value()) return false;
+    return reference.read_parameters(stream);
+  }
+
+  // Write evaluation function parameters
+  template <typename T>
+  bool write_parameters(std::ostream& stream, const T& reference) {
+
+    write_little_endian<std::uint32_t>(stream, T::get_hash_value());
+    return reference.write_parameters(stream);
   }
 
   }  // namespace Detail
 
   // Initialize the evaluation function parameters
-  void Initialize() {
+  void initialize() {
 
-    Detail::Initialize(feature_transformer);
-    Detail::Initialize(network);
+    Detail::initialize(featureTransformer);
+    for (std::size_t i = 0; i < LayerStacks; ++i)
+      Detail::initialize(network[i]);
   }
 
   // Read network header
-  bool ReadHeader(std::istream& stream, std::uint32_t* hash_value, std::string* architecture)
+  bool read_header(std::istream& stream, std::uint32_t* hashValue, std::string* desc)
   {
     std::uint32_t version, size;
 
     version     = read_little_endian<std::uint32_t>(stream);
-    *hash_value = read_little_endian<std::uint32_t>(stream);
+    *hashValue  = 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);
+    if (!stream || version != Version) return false;
+    desc->resize(size);
+    stream.read(&(*desc)[0], size);
+    return !stream.fail();
+  }
+
+  // Write network header
+  bool write_header(std::ostream& stream, std::uint32_t hashValue, const std::string& desc)
+  {
+    write_little_endian<std::uint32_t>(stream, Version);
+    write_little_endian<std::uint32_t>(stream, hashValue);
+    write_little_endian<std::uint32_t>(stream, desc.size());
+    stream.write(&desc[0], desc.size());
     return !stream.fail();
   }
 
   // Read network parameters
-  bool ReadParameters(std::istream& stream) {
+  bool read_parameters(std::istream& stream) {
 
-    std::uint32_t hash_value;
-    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;
+    std::uint32_t hashValue;
+    if (!read_header(stream, &hashValue, &netDescription)) return false;
+    if (hashValue != HashValue) return false;
+    if (!Detail::read_parameters(stream, *featureTransformer)) return false;
+    for (std::size_t i = 0; i < LayerStacks; ++i)
+      if (!Detail::read_parameters(stream, *(network[i]))) return false;
     return stream && stream.peek() == std::ios::traits_type::eof();
   }
 
-  // Proceed with the difference calculation if possible
-  static void UpdateAccumulatorIfPossible(const Position& pos) {
+  // Write network parameters
+  bool write_parameters(std::ostream& stream) {
 
-    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 eval, from a file stream or a memory stream
-  bool load_eval(std::string streamName, std::istream& stream) {
-
-    Initialize();
-    fileName = streamName;
-    return ReadParameters(stream);
+    if (!write_header(stream, HashValue, netDescription)) return false;
+    if (!Detail::write_parameters(stream, *featureTransformer)) return false;
+    for (std::size_t i = 0; i < LayerStacks; ++i)
+      if (!Detail::write_parameters(stream, *(network[i]))) return false;
+    return (bool)stream;
   }
 
   // Evaluation function. Perform differential calculation.
-  Value evaluate(const Position& pos) {
-    return ComputeScore(pos, false);
+  Value evaluate(const Position& pos, bool adjusted) {
+
+    // 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 = CacheLineSize;
+
+#if defined(ALIGNAS_ON_STACK_VARIABLES_BROKEN)
+    TransformedFeatureType transformedFeaturesUnaligned[
+      FeatureTransformer::BufferSize + alignment / sizeof(TransformedFeatureType)];
+    char bufferUnaligned[Network::BufferSize + alignment];
+
+    auto* transformedFeatures = align_ptr_up<alignment>(&transformedFeaturesUnaligned[0]);
+    auto* buffer = align_ptr_up<alignment>(&bufferUnaligned[0]);
+#else
+    alignas(alignment)
+      TransformedFeatureType transformedFeatures[FeatureTransformer::BufferSize];
+    alignas(alignment) char buffer[Network::BufferSize];
+#endif
+
+    ASSERT_ALIGNED(transformedFeatures, alignment);
+    ASSERT_ALIGNED(buffer, alignment);
+
+    const std::size_t bucket = (pos.count<ALL_PIECES>() - 1) / 4;
+    const auto psqt = featureTransformer->transform(pos, transformedFeatures, bucket);
+    const auto output = network[bucket]->propagate(transformedFeatures, buffer);
+
+    int materialist = psqt;
+    int positional  = output[0];
+
+    int delta_npm = abs(pos.non_pawn_material(WHITE) - pos.non_pawn_material(BLACK));
+    int entertainment = (adjusted && delta_npm <= BishopValueMg - KnightValueMg ? 7 : 0);
+
+    int A = 128 - entertainment;
+    int B = 128 + entertainment;
+
+    int sum = (A * materialist + B * positional) / 128;
+
+    return static_cast<Value>( sum / OutputScale );
   }
 
-  // Evaluation function. Perform full calculation.
-  Value compute_eval(const Position& pos) {
-    return ComputeScore(pos, true);
+  struct NnueEvalTrace {
+    static_assert(LayerStacks == PSQTBuckets);
+
+    Value psqt[LayerStacks];
+    Value positional[LayerStacks];
+    std::size_t correctBucket;
+  };
+
+  static NnueEvalTrace trace_evaluate(const Position& pos) {
+
+    // 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 = CacheLineSize;
+
+#if defined(ALIGNAS_ON_STACK_VARIABLES_BROKEN)
+    TransformedFeatureType transformedFeaturesUnaligned[
+      FeatureTransformer::BufferSize + alignment / sizeof(TransformedFeatureType)];
+    char bufferUnaligned[Network::BufferSize + alignment];
+
+    auto* transformedFeatures = align_ptr_up<alignment>(&transformedFeaturesUnaligned[0]);
+    auto* buffer = align_ptr_up<alignment>(&bufferUnaligned[0]);
+#else
+    alignas(alignment)
+      TransformedFeatureType transformedFeatures[FeatureTransformer::BufferSize];
+    alignas(alignment) char buffer[Network::BufferSize];
+#endif
+
+    ASSERT_ALIGNED(transformedFeatures, alignment);
+    ASSERT_ALIGNED(buffer, alignment);
+
+    NnueEvalTrace t{};
+    t.correctBucket = (pos.count<ALL_PIECES>() - 1) / 4;
+    for (std::size_t bucket = 0; bucket < LayerStacks; ++bucket) {
+      const auto psqt = featureTransformer->transform(pos, transformedFeatures, bucket);
+      const auto output = network[bucket]->propagate(transformedFeatures, buffer);
+
+      int materialist = psqt;
+      int positional  = output[0];
+
+      t.psqt[bucket] = static_cast<Value>( materialist / OutputScale );
+      t.positional[bucket] = static_cast<Value>( positional / OutputScale );
+    }
+
+    return t;
   }
 
-  // Proceed with the difference calculation if possible
-  void update_eval(const Position& pos) {
-    UpdateAccumulatorIfPossible(pos);
+  static const std::string PieceToChar(" PNBRQK  pnbrqk");
+
+  // Requires the buffer to have capacity for at least 5 values
+  static void format_cp_compact(Value v, char* buffer) {
+
+    buffer[0] = (v < 0 ? '-' : v > 0 ? '+' : ' ');
+
+    int cp = std::abs(100 * v / PawnValueEg);
+
+    if (cp >= 10000)
+    {
+      buffer[1] = '0' + cp / 10000; cp %= 10000;
+      buffer[2] = '0' + cp / 1000; cp %= 1000;
+      buffer[3] = '0' + cp / 100; cp %= 100;
+      buffer[4] = ' ';
+    }
+    else if (cp >= 1000)
+    {
+      buffer[1] = '0' + cp / 1000; cp %= 1000;
+      buffer[2] = '0' + cp / 100; cp %= 100;
+      buffer[3] = '.';
+      buffer[4] = '0' + cp / 10;
+    }
+    else
+    {
+      buffer[1] = '0' + cp / 100; cp %= 100;
+      buffer[2] = '.';
+      buffer[3] = '0' + cp / 10; cp %= 10;
+      buffer[4] = '0' + cp / 1;
+    }
   }
 
-} // namespace Eval::NNUE
+  // Requires the buffer to have capacity for at least 7 values
+  static void format_cp_aligned_dot(Value v, char* buffer) {
+    buffer[0] = (v < 0 ? '-' : v > 0 ? '+' : ' ');
+
+    int cp = std::abs(100 * v / PawnValueEg);
+
+    if (cp >= 10000)
+    {
+      buffer[1] = '0' + cp / 10000; cp %= 10000;
+      buffer[2] = '0' + cp / 1000; cp %= 1000;
+      buffer[3] = '0' + cp / 100; cp %= 100;
+      buffer[4] = '.';
+      buffer[5] = '0' + cp / 10; cp %= 10;
+      buffer[6] = '0' + cp;
+    }
+    else if (cp >= 1000)
+    {
+      buffer[1] = ' ';
+      buffer[2] = '0' + cp / 1000; cp %= 1000;
+      buffer[3] = '0' + cp / 100; cp %= 100;
+      buffer[4] = '.';
+      buffer[5] = '0' + cp / 10; cp %= 10;
+      buffer[6] = '0' + cp;
+    }
+    else
+    {
+      buffer[1] = ' ';
+      buffer[2] = ' ';
+      buffer[3] = '0' + cp / 100; cp %= 100;
+      buffer[4] = '.';
+      buffer[5] = '0' + cp / 10; cp %= 10;
+      buffer[6] = '0' + cp / 1;
+    }
+  }
+
+
+  // trace() returns a string with the value of each piece on a board,
+  // and a table for (PSQT, Layers) values bucket by bucket.
+
+  std::string trace(Position& pos) {
+
+    std::stringstream ss;
+
+    char board[3*8+1][8*8+2];
+    std::memset(board, ' ', sizeof(board));
+    for (int row = 0; row < 3*8+1; ++row)
+      board[row][8*8+1] = '\0';
+
+    // A lambda to output one box of the board
+    auto writeSquare = [&board](File file, Rank rank, Piece pc, Value value) {
+
+      const int x = ((int)file) * 8;
+      const int y = (7 - (int)rank) * 3;
+      for (int i = 1; i < 8; ++i)
+         board[y][x+i] = board[y+3][x+i] = '-';
+      for (int i = 1; i < 3; ++i)
+         board[y+i][x] = board[y+i][x+8] = '|';
+      board[y][x] = board[y][x+8] = board[y+3][x+8] = board[y+3][x] = '+';
+      if (pc != NO_PIECE)
+        board[y+1][x+4] = PieceToChar[pc];
+      if (value != VALUE_NONE)
+        format_cp_compact(value, &board[y+2][x+2]);
+    };
+
+    // We estimate the value of each piece by doing a differential evaluation from
+    // the current base eval, simulating the removal of the piece from its square.
+    Value base = evaluate(pos);
+    base = pos.side_to_move() == WHITE ? base : -base;
+
+    for (File f = FILE_A; f <= FILE_H; ++f)
+      for (Rank r = RANK_1; r <= RANK_8; ++r)
+      {
+        Square sq = make_square(f, r);
+        Piece pc = pos.piece_on(sq);
+        Value v = VALUE_NONE;
+
+        if (pc != NO_PIECE && type_of(pc) != KING)
+        {
+          auto st = pos.state();
+
+          pos.remove_piece(sq);
+          st->accumulator.computed[WHITE] = false;
+          st->accumulator.computed[BLACK] = false;
+
+          Value eval = evaluate(pos);
+          eval = pos.side_to_move() == WHITE ? eval : -eval;
+          v = base - eval;
+
+          pos.put_piece(pc, sq);
+          st->accumulator.computed[WHITE] = false;
+          st->accumulator.computed[BLACK] = false;
+        }
+
+        writeSquare(f, r, pc, v);
+      }
+
+    ss << " NNUE derived piece values:\n";
+    for (int row = 0; row < 3*8+1; ++row)
+        ss << board[row] << '\n';
+    ss << '\n';
+
+    auto t = trace_evaluate(pos);
+
+    ss << " NNUE network contributions "
+       << (pos.side_to_move() == WHITE ? "(White to move)" : "(Black to move)") << std::endl
+       << "+------------+------------+------------+------------+\n"
+       << "|   Bucket   |  Material  | Positional |   Total    |\n"
+       << "|            |   (PSQT)   |  (Layers)  |            |\n"
+       << "+------------+------------+------------+------------+\n";
+
+    for (std::size_t bucket = 0; bucket < LayerStacks; ++bucket)
+    {
+      char buffer[3][8];
+      std::memset(buffer, '\0', sizeof(buffer));
+
+      format_cp_aligned_dot(t.psqt[bucket], buffer[0]);
+      format_cp_aligned_dot(t.positional[bucket], buffer[1]);
+      format_cp_aligned_dot(t.psqt[bucket] + t.positional[bucket], buffer[2]);
+
+      ss <<  "|  " << bucket    << "        "
+         << " |  " << buffer[0] << "  "
+         << " |  " << buffer[1] << "  "
+         << " |  " << buffer[2] << "  "
+         << " |";
+      if (bucket == t.correctBucket)
+          ss << " <-- this bucket is used";
+      ss << '\n';
+    }
+
+    ss << "+------------+------------+------------+------------+\n";
+
+    return ss.str();
+  }
+
+
+  // Load eval, from a file stream or a memory stream
+  bool load_eval(std::string name, std::istream& stream) {
+
+    initialize();
+    fileName = name;
+    return read_parameters(stream);
+  }
+
+  // Save eval, to a file stream or a memory stream
+  bool save_eval(std::ostream& stream) {
+
+    if (fileName.empty())
+      return false;
+
+    return write_parameters(stream);
+  }
+
+  /// Save eval, to a file given by its name
+  bool save_eval(const std::optional<std::string>& filename) {
+
+    std::string actualFilename;
+    std::string msg;
+
+    if (filename.has_value())
+        actualFilename = filename.value();
+    else
+    {
+        if (eval_file_loaded != EvalFileDefaultName)
+        {
+             msg = "Failed to export a net. A non-embedded net can only be saved if the filename is specified";
+
+             sync_cout << msg << sync_endl;
+             return false;
+        }
+        actualFilename = EvalFileDefaultName;
+    }
+
+    std::ofstream stream(actualFilename, std::ios_base::binary);
+    bool saved = save_eval(stream);
+
+    msg = saved ? "Network saved successfully to " + actualFilename
+                : "Failed to export a net";
+
+    sync_cout << msg << sync_endl;
+    return saved;
+  }
+
+
+} // namespace Stockfish::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
@@ -25,11 +25,11 @@
 
 #include <memory>
 
-namespace Eval::NNUE {
+namespace Stockfish::Eval::NNUE {
 
   // Hash value of evaluation function structure
-  constexpr std::uint32_t kHashValue =
-      FeatureTransformer::GetHashValue() ^ Network::GetHashValue();
+  constexpr std::uint32_t HashValue =
+      FeatureTransformer::get_hash_value() ^ Network::get_hash_value();
 
   // Deleter for automating release of memory area
   template <typename T>
@@ -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>>;
 
-}  // namespace Eval::NNUE
+  template <typename T>
+  using LargePagePtr = std::unique_ptr<T, LargePageDeleter<T>>;
+
+}  // namespace Stockfish::Eval::NNUE
 
 #endif // #ifndef NNUE_EVALUATE_NNUE_H_INCLUDED

src/nnue/features/feature_set.h

@@ -1,134 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 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/>.
-*/
-
-// A class template that represents the input feature set of the NNUE evaluation function
-
-#ifndef NNUE_FEATURE_SET_H_INCLUDED
-#define NNUE_FEATURE_SET_H_INCLUDED
-
-#include "features_common.h"
-#include <array>
-
-namespace Eval::NNUE::Features {
-
-  // Class template that represents a list of values
-  template <typename T, T... Values>
-  struct CompileTimeList;
-
-  template <typename T, T First, T... Remaining>
-  struct CompileTimeList<T, First, Remaining...> {
-    static constexpr bool Contains(T value) {
-      return value == First || CompileTimeList<T, Remaining...>::Contains(value);
-    }
-    static constexpr std::array<T, sizeof...(Remaining) + 1>
-        kValues = {{First, Remaining...}};
-  };
-
-  // Base class of feature set
-  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.piece[0] == make_piece(perspective, KING);
-            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
-  template <typename FeatureType>
-  class FeatureSet<FeatureType> : public FeatureSetBase<FeatureSet<FeatureType>> {
-
-   public:
-    // Hash value embedded in the evaluation file
-    static constexpr std::uint32_t kHashValue = FeatureType::kHashValue;
-    // Number of feature dimensions
-    static constexpr IndexType kDimensions = FeatureType::kDimensions;
-    // Maximum number of simultaneously active features
-    static constexpr IndexType kMaxActiveDimensions =
-        FeatureType::kMaxActiveDimensions;
-    // Trigger for full calculation instead of difference calculation
-    using SortedTriggerSet =
-        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
-
-#endif // #ifndef NNUE_FEATURE_SET_H_INCLUDED

src/nnue/features/half_ka_v2.cpp

@@ -0,0 +1,85 @@
+/*
+  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/>.
+*/
+
+//Definition of input features HalfKAv2 of NNUE evaluation function
+
+#include "half_ka_v2.h"
+
+#include "../../position.h"
+
+namespace Stockfish::Eval::NNUE::Features {
+
+  // Orient a square according to perspective (rotates by 180 for black)
+  inline Square HalfKAv2::orient(Color perspective, Square s) {
+    return Square(int(s) ^ (bool(perspective) * 56));
+  }
+
+  // Index of a feature for a given king position and another piece on some square
+  inline IndexType HalfKAv2::make_index(Color perspective, Square s, Piece pc, Square ksq) {
+    return IndexType(orient(perspective, s) + PieceSquareIndex[perspective][pc] + PS_NB * ksq);
+  }
+
+  // Get a list of indices for active features
+  void HalfKAv2::append_active_indices(
+    const Position& pos,
+    Color perspective,
+    ValueListInserter<IndexType> active
+  ) {
+    Square ksq = orient(perspective, pos.square<KING>(perspective));
+    Bitboard bb = pos.pieces();
+    while (bb)
+    {
+      Square s = pop_lsb(bb);
+      active.push_back(make_index(perspective, s, pos.piece_on(s), ksq));
+    }
+  }
+
+
+  // append_changed_indices() : get a list of indices for recently changed features
+
+  void HalfKAv2::append_changed_indices(
+    Square ksq,
+    StateInfo* st,
+    Color perspective,
+    ValueListInserter<IndexType> removed,
+    ValueListInserter<IndexType> added
+  ) {
+    const auto& dp = st->dirtyPiece;
+    Square oriented_ksq = orient(perspective, ksq);
+    for (int i = 0; i < dp.dirty_num; ++i) {
+      Piece pc = dp.piece[i];
+      if (dp.from[i] != SQ_NONE)
+        removed.push_back(make_index(perspective, dp.from[i], pc, oriented_ksq));
+      if (dp.to[i] != SQ_NONE)
+        added.push_back(make_index(perspective, dp.to[i], pc, oriented_ksq));
+    }
+  }
+
+  int HalfKAv2::update_cost(StateInfo* st) {
+    return st->dirtyPiece.dirty_num;
+  }
+
+  int HalfKAv2::refresh_cost(const Position& pos) {
+    return pos.count<ALL_PIECES>();
+  }
+
+  bool HalfKAv2::requires_refresh(StateInfo* st, Color perspective) {
+    return st->dirtyPiece.piece[0] == make_piece(perspective, KING);
+  }
+
+}  // namespace Stockfish::Eval::NNUE::Features

src/nnue/features/half_ka_v2.h

@@ -0,0 +1,111 @@
+/*
+  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/>.
+*/
+
+//Definition of input features HalfKP of NNUE evaluation function
+
+#ifndef NNUE_FEATURES_HALF_KA_V2_H_INCLUDED
+#define NNUE_FEATURES_HALF_KA_V2_H_INCLUDED
+
+#include "../nnue_common.h"
+
+#include "../../evaluate.h"
+#include "../../misc.h"
+
+namespace Stockfish {
+  struct StateInfo;
+}
+
+namespace Stockfish::Eval::NNUE::Features {
+
+  // Feature HalfKAv2: Combination of the position of own king
+  // and the position of pieces
+  class HalfKAv2 {
+
+    // unique number for each piece type on each square
+    enum {
+      PS_NONE     =  0,
+      PS_W_PAWN   =  0,
+      PS_B_PAWN   =  1 * SQUARE_NB,
+      PS_W_KNIGHT =  2 * SQUARE_NB,
+      PS_B_KNIGHT =  3 * SQUARE_NB,
+      PS_W_BISHOP =  4 * SQUARE_NB,
+      PS_B_BISHOP =  5 * SQUARE_NB,
+      PS_W_ROOK   =  6 * SQUARE_NB,
+      PS_B_ROOK   =  7 * SQUARE_NB,
+      PS_W_QUEEN  =  8 * SQUARE_NB,
+      PS_B_QUEEN  =  9 * SQUARE_NB,
+      PS_KING     =  10 * SQUARE_NB,
+      PS_NB = 11 * SQUARE_NB
+    };
+
+    static constexpr IndexType PieceSquareIndex[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_KING, PS_NONE,
+        PS_NONE, PS_B_PAWN, PS_B_KNIGHT, PS_B_BISHOP, PS_B_ROOK, PS_B_QUEEN, PS_KING, PS_NONE },
+      { PS_NONE, PS_B_PAWN, PS_B_KNIGHT, PS_B_BISHOP, PS_B_ROOK, PS_B_QUEEN, PS_KING, PS_NONE,
+        PS_NONE, PS_W_PAWN, PS_W_KNIGHT, PS_W_BISHOP, PS_W_ROOK, PS_W_QUEEN, PS_KING, PS_NONE }
+    };
+
+    // Orient a square according to perspective (rotates by 180 for black)
+    static Square orient(Color perspective, Square s);
+
+    // Index of a feature for a given king position and another piece on some square
+    static IndexType make_index(Color perspective, Square s, Piece pc, Square ksq);
+
+   public:
+    // Feature name
+    static constexpr const char* Name = "HalfKAv2(Friend)";
+
+    // Hash value embedded in the evaluation file
+    static constexpr std::uint32_t HashValue = 0x5f234cb8u;
+
+    // Number of feature dimensions
+    static constexpr IndexType Dimensions =
+        static_cast<IndexType>(SQUARE_NB) * static_cast<IndexType>(PS_NB);
+
+    // Maximum number of simultaneously active features.
+    static constexpr IndexType MaxActiveDimensions = 32;
+
+    // Get a list of indices for active features
+    static void append_active_indices(
+      const Position& pos,
+      Color perspective,
+      ValueListInserter<IndexType> active);
+
+    // Get a list of indices for recently changed features
+    static void append_changed_indices(
+      Square ksq,
+      StateInfo* st,
+      Color perspective,
+      ValueListInserter<IndexType> removed,
+      ValueListInserter<IndexType> added);
+
+    // Returns the cost of updating one perspective, the most costly one.
+    // Assumes no refresh needed.
+    static int update_cost(StateInfo* st);
+    static int refresh_cost(const Position& pos);
+
+    // Returns whether the change stored in this StateInfo means that
+    // a full accumulator refresh is required.
+    static bool requires_refresh(StateInfo* st, Color perspective);
+  };
+
+}  // namespace Stockfish::Eval::NNUE::Features
+
+#endif // #ifndef NNUE_FEATURES_HALF_KA_V2_H_INCLUDED

src/nnue/features/half_kp.cpp

@@ -1,72 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 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/>.
-*/
-
-//Definition of input features HalfKP of NNUE evaluation function
-
-#include "half_kp.h"
-#include "index_list.h"
-
-namespace Eval::NNUE::Features {
-
-  // 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));
-  }
-
-  // Find the index of the feature quantity from the king position and PieceSquare
-  template <Side AssociatedKing>
-  inline IndexType HalfKP<AssociatedKing>::MakeIndex(
-      Color perspective, Square s, Piece pc, Square ksq) {
-
-    return IndexType(orient(perspective, s) + kpp_board_index[pc][perspective] + PS_END * ksq);
-  }
-
-  // Get a list of indices for active features
-  template <Side AssociatedKing>
-  void HalfKP<AssociatedKing>::AppendActiveIndices(
-      const Position& pos, Color perspective, IndexList* active) {
-
-    Square ksq = orient(perspective, pos.square<KING>(perspective));
-    Bitboard bb = pos.pieces() & ~pos.pieces(KING);
-    while (bb) {
-      Square s = pop_lsb(&bb);
-      active->push_back(MakeIndex(perspective, s, pos.piece_on(s), ksq));
-    }
-  }
-
-  // Get a list of indices for recently changed features
-  template <Side AssociatedKing>
-  void HalfKP<AssociatedKing>::AppendChangedIndices(
-      const Position& pos, Color perspective,
-      IndexList* removed, IndexList* added) {
-
-    Square ksq = orient(perspective, pos.square<KING>(perspective));
-    const auto& dp = pos.state()->dirtyPiece;
-    for (int i = 0; i < dp.dirty_num; ++i) {
-      Piece pc = dp.piece[i];
-      if (type_of(pc) == KING) continue;
-      if (dp.from[i] != SQ_NONE)
-        removed->push_back(MakeIndex(perspective, dp.from[i], pc, ksq));
-      if (dp.to[i] != SQ_NONE)
-        added->push_back(MakeIndex(perspective, dp.to[i], pc, ksq));
-    }
-  }
-
-  template class HalfKP<Side::kFriend>;
-
-}  // namespace Eval::NNUE::Features

src/nnue/features/half_kp.h

@@ -1,63 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 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/>.
-*/
-
-//Definition of input features HalfKP of NNUE evaluation function
-
-#ifndef NNUE_FEATURES_HALF_KP_H_INCLUDED
-#define NNUE_FEATURES_HALF_KP_H_INCLUDED
-
-#include "../../evaluate.h"
-#include "features_common.h"
-
-namespace Eval::NNUE::Features {
-
-  // Feature HalfKP: Combination of the position of own king
-  // and the position of pieces other than kings
-  template <Side AssociatedKing>
-  class HalfKP {
-
-   public:
-    // Feature name
-    static constexpr const char* kName = "HalfKP(Friend)";
-    // Hash value embedded in the evaluation file
-    static constexpr std::uint32_t kHashValue =
-        0x5D69D5B9u ^ (AssociatedKing == Side::kFriend);
-    // Number of feature dimensions
-    static constexpr IndexType kDimensions =
-        static_cast<IndexType>(SQUARE_NB) * static_cast<IndexType>(PS_END);
-    // Maximum number of simultaneously active features
-    static constexpr IndexType kMaxActiveDimensions = 30; // Kings don't count
-    // Trigger for full calculation instead of difference calculation
-    static constexpr TriggerEvent kRefreshTrigger = TriggerEvent::kFriendKingMoved;
-
-    // Get a list of indices for active features
-    static void AppendActiveIndices(const Position& pos, Color perspective,
-                                    IndexList* active);
-
-    // Get a list of indices for recently changed features
-    static void AppendChangedIndices(const Position& pos, Color perspective,
-                                     IndexList* removed, IndexList* added);
-
-   private:
-    // Index of a feature for a given king position and another piece on some square
-    static IndexType MakeIndex(Color perspective, Square s, Piece pc, Square sq_k);
-  };
-
-}  // namespace Eval::NNUE::Features
-
-#endif // #ifndef NNUE_FEATURES_HALF_KP_H_INCLUDED

src/nnue/features/index_list.h

@@ -1,64 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2020 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/>.
-*/
-
-// Definition of index list of input features
-
-#ifndef NNUE_FEATURES_INDEX_LIST_H_INCLUDED
-#define NNUE_FEATURES_INDEX_LIST_H_INCLUDED
-
-#include "../../position.h"
-#include "../nnue_architecture.h"
-
-namespace Eval::NNUE::Features {
-
-  // Class template used for feature index list
-  template <typename T, std::size_t MaxSize>
-  class ValueList {
-
-   public:
-    std::size_t size() const { return size_; }
-    void resize(std::size_t size) { size_ = size; }
-    void push_back(const T& value) { values_[size_++] = value; }
-    T& operator[](std::size_t index) { return values_[index]; }
-    T* begin() { return values_; }
-    T* end() { return values_ + size_; }
-    const T& operator[](std::size_t index) const { return values_[index]; }
-    const T* begin() const { return values_; }
-    const T* end() const { return values_ + size_; }
-
-    void swap(ValueList& other) {
-      const std::size_t max_size = std::max(size_, other.size_);
-      for (std::size_t i = 0; i < max_size; ++i) {
-        std::swap(values_[i], other.values_[i]);
-      }
-      std::swap(size_, other.size_);
-    }
-
-   private:
-    T values_[MaxSize];
-    std::size_t size_ = 0;
-  };
-
-  //Type of feature index list
-  class IndexList
-      : public ValueList<IndexType, RawFeatures::kMaxActiveDimensions> {
-  };
-
-}  // namespace Eval::NNUE::Features
-
-#endif // NNUE_FEATURES_INDEX_LIST_H_INCLUDED

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
@@ -24,10 +24,10 @@
 #include <iostream>
 #include "../nnue_common.h"
 
-namespace Eval::NNUE::Layers {
+namespace Stockfish::Eval::NNUE::Layers {
 
   // Affine transformation layer
-  template <typename PreviousLayer, IndexType OutputDimensions>
+  template <typename PreviousLayer, IndexType OutDims>
   class AffineTransform {
    public:
     // Input/output type
@@ -36,217 +36,387 @@ namespace Eval::NNUE::Layers {
     static_assert(std::is_same<InputType, std::uint8_t>::value, "");
 
     // Number of input/output dimensions
-    static constexpr IndexType kInputDimensions =
-        PreviousLayer::kOutputDimensions;
-    static constexpr IndexType kOutputDimensions = OutputDimensions;
-    static constexpr IndexType kPaddedInputDimensions =
-        CeilToMultiple<IndexType>(kInputDimensions, kMaxSimdWidth);
+    static constexpr IndexType InputDimensions =
+        PreviousLayer::OutputDimensions;
+    static constexpr IndexType OutputDimensions = OutDims;
+    static constexpr IndexType PaddedInputDimensions =
+        ceil_to_multiple<IndexType>(InputDimensions, MaxSimdWidth);
+#if defined (USE_AVX512)
+    static constexpr const IndexType OutputSimdWidth = SimdWidth / 2;
+#elif defined (USE_SSSE3)
+    static constexpr const IndexType OutputSimdWidth = SimdWidth / 4;
+#endif
 
     // Size of forward propagation buffer used in this layer
-    static constexpr std::size_t kSelfBufferSize =
-        CeilToMultiple(kOutputDimensions * sizeof(OutputType), kCacheLineSize);
+    static constexpr std::size_t SelfBufferSize =
+        ceil_to_multiple(OutputDimensions * sizeof(OutputType), CacheLineSize);
 
     // Size of the forward propagation buffer used from the input layer to this layer
-    static constexpr std::size_t kBufferSize =
-        PreviousLayer::kBufferSize + kSelfBufferSize;
+    static constexpr std::size_t BufferSize =
+        PreviousLayer::BufferSize + SelfBufferSize;
 
     // Hash value embedded in the evaluation file
-    static constexpr std::uint32_t GetHashValue() {
-      std::uint32_t hash_value = 0xCC03DAE4u;
-      hash_value += kOutputDimensions;
-      hash_value ^= PreviousLayer::GetHashValue() >> 1;
-      hash_value ^= PreviousLayer::GetHashValue() << 31;
-      return hash_value;
+    static constexpr std::uint32_t get_hash_value() {
+      std::uint32_t hashValue = 0xCC03DAE4u;
+      hashValue += OutputDimensions;
+      hashValue ^= PreviousLayer::get_hash_value() >> 1;
+      hashValue ^= PreviousLayer::get_hash_value() << 31;
+      return hashValue;
     }
 
-   // Read network parameters
-    bool ReadParameters(std::istream& stream) {
-      if (!previous_layer_.ReadParameters(stream)) return false;
-      for (std::size_t i = 0; i < kOutputDimensions; ++i)
-        biases_[i] = read_little_endian<BiasType>(stream);
-      for (std::size_t i = 0; i < kOutputDimensions * kPaddedInputDimensions; ++i)
-        weights_[i] = read_little_endian<WeightType>(stream);
+    // Read network parameters
+    bool read_parameters(std::istream& stream) {
+      if (!previousLayer.read_parameters(stream)) return false;
+      for (std::size_t i = 0; i < OutputDimensions; ++i)
+        biases[i] = read_little_endian<BiasType>(stream);
+      for (std::size_t i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+#if !defined (USE_SSSE3)
+        weights[i] = read_little_endian<WeightType>(stream);
+#else
+        weights[
+          (i / 4) % (PaddedInputDimensions / 4) * OutputDimensions * 4 +
+          i / PaddedInputDimensions * 4 +
+          i % 4
+        ] = read_little_endian<WeightType>(stream);
+#endif
+
+      return !stream.fail();
+    }
+
+    // Write network parameters
+    bool write_parameters(std::ostream& stream) const {
+      if (!previousLayer.write_parameters(stream)) return false;
+      for (std::size_t i = 0; i < OutputDimensions; ++i)
+          write_little_endian<BiasType>(stream, biases[i]);
+#if !defined (USE_SSSE3)
+      for (std::size_t i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+          write_little_endian<WeightType>(stream, weights[i]);
+#else
+      std::unique_ptr<WeightType[]> unscrambledWeights = std::make_unique<WeightType[]>(OutputDimensions * PaddedInputDimensions);
+      for (std::size_t i = 0; i < OutputDimensions * PaddedInputDimensions; ++i) {
+          unscrambledWeights[i] =
+              weights[
+                (i / 4) % (PaddedInputDimensions / 4) * OutputDimensions * 4 +
+                i / PaddedInputDimensions * 4 +
+                i % 4
+              ];
+      }
+
+      for (std::size_t i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+          write_little_endian<WeightType>(stream, unscrambledWeights[i]);
+#endif
+
       return !stream.fail();
     }
 
     // Forward propagation
-    const OutputType* Propagate(
-        const TransformedFeatureType* transformed_features, char* buffer) const {
-      const auto input = previous_layer_.Propagate(
-          transformed_features, buffer + kSelfBufferSize);
-      const auto output = reinterpret_cast<OutputType*>(buffer);
+    const OutputType* propagate(
+        const TransformedFeatureType* transformedFeatures, char* buffer) const {
+      const auto input = previousLayer.propagate(
+          transformedFeatures, buffer + SelfBufferSize);
 
-  #if defined(USE_AVX512)
-      constexpr IndexType kNumChunks = kPaddedInputDimensions / (kSimdWidth * 2);
-      const auto input_vector = reinterpret_cast<const __m512i*>(input);
-  #if !defined(USE_VNNI)
-      const __m512i kOnes = _mm512_set1_epi16(1);
-  #endif
+#if defined (USE_AVX512)
 
-  #elif defined(USE_AVX2)
-      constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
-      const auto input_vector = reinterpret_cast<const __m256i*>(input);
-  #if !defined(USE_VNNI)
-      const __m256i kOnes = _mm256_set1_epi16(1);
-  #endif
+      [[maybe_unused]] const __m512i Ones512 = _mm512_set1_epi16(1);
 
-  #elif defined(USE_SSE2)
-      constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
-  #ifndef USE_SSSE3
-      const __m128i kZeros = _mm_setzero_si128();
-  #else
-      const __m128i kOnes = _mm_set1_epi16(1);
-  #endif
-      const auto input_vector = reinterpret_cast<const __m128i*>(input);
+      [[maybe_unused]] auto m512_hadd = [](__m512i sum, int bias) -> int {
+        return _mm512_reduce_add_epi32(sum) + bias;
+      };
 
-  #elif defined(USE_MMX)
-      constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
-      const __m64 kZeros = _mm_setzero_si64();
-      const auto input_vector = reinterpret_cast<const __m64*>(input);
+      [[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, Ones512);
+        acc = _mm512_add_epi32(acc, product0);
+#endif
+      };
 
-  #elif defined(USE_NEON)
-      constexpr IndexType kNumChunks = kPaddedInputDimensions / kSimdWidth;
-      const auto input_vector = reinterpret_cast<const int8x8_t*>(input);
-  #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_adds_epi16(product0, product1);
+        product0 = _mm512_madd_epi16(product0, Ones512);
+        product2 = _mm512_adds_epi16(product2, product3);
+        product2 = _mm512_madd_epi16(product2, Ones512);
+        acc = _mm512_add_epi32(acc, _mm512_add_epi32(product0, product2));
+#endif
+      };
 
-      for (IndexType i = 0; i < kOutputDimensions; ++i) {
-        const IndexType offset = i * kPaddedInputDimensions;
+#endif
+#if defined (USE_AVX2)
 
-  #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(USE_VNNI)
-            sum = _mm512_dpbusd_epi32(sum, _mm512_loadA_si512(&input_vector[j]), _mm512_load_si512(&row[j]));
-  #else
-            __m512i product = _mm512_maddubs_epi16(_mm512_loadA_si512(&input_vector[j]), _mm512_load_si512(&row[j]));
-            product = _mm512_madd_epi16(product, kOnes);
-            sum = _mm512_add_epi32(sum, product);
-  #endif
-        }
+      [[maybe_unused]] const __m256i Ones256 = _mm256_set1_epi16(1);
 
-        // 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 iv256  = reinterpret_cast<const __m256i*>(&input_vector[kNumChunks]);
-            const auto row256 = reinterpret_cast<const __m256i*>(&row[kNumChunks]);
-  #if defined(USE_VNNI)
-            __m256i product256 = _mm256_dpbusd_epi32(
-                _mm512_castsi512_si256(sum), _mm256_loadA_si256(&iv256[0]), _mm256_load_si256(&row256[0]));
-            sum = _mm512_inserti32x8(sum, product256, 0);
-  #else
-            __m256i product256 = _mm256_maddubs_epi16(_mm256_loadA_si256(&iv256[0]), _mm256_load_si256(&row256[0]));
-            sum = _mm512_add_epi32(sum, _mm512_cvtepi16_epi32(product256));
-  #endif
-        }
-        output[i] = _mm512_reduce_add_epi32(sum) + biases_[i];
-
-  #elif defined(USE_AVX2)
-        __m256i sum = _mm256_setzero_si256();
-        const auto row = reinterpret_cast<const __m256i*>(&weights_[offset]);
-        for (IndexType j = 0; j < kNumChunks; ++j) {
-  #if defined(USE_VNNI)
-          sum = _mm256_dpbusd_epi32(sum, _mm256_loadA_si256(&input_vector[j]), _mm256_load_si256(&row[j]));
-  #else
-          __m256i product = _mm256_maddubs_epi16(_mm256_loadA_si256(&input_vector[j]), _mm256_load_si256(&row[j]));
-          product = _mm256_madd_epi16(product, kOnes);
-          sum = _mm256_add_epi32(sum, product);
-  #endif
-        }
+      [[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));
-        output[i] = _mm_cvtsi128_si32(sum128) + biases_[i];
+        return _mm_cvtsi128_si32(sum128) + bias;
+      };
 
-  #elif defined(USE_SSSE3)
-        __m128i sum = _mm_setzero_si128();
-        const auto row = reinterpret_cast<const __m128i*>(&weights_[offset]);
-        for (int j = 0; j < (int)kNumChunks - 1; j += 2) {
-          __m128i product0 = _mm_maddubs_epi16(_mm_load_si128(&input_vector[j]), _mm_load_si128(&row[j]));
-          product0 = _mm_madd_epi16(product0, kOnes);
-          sum = _mm_add_epi32(sum, product0);
-          __m128i product1 = _mm_maddubs_epi16(_mm_load_si128(&input_vector[j+1]), _mm_load_si128(&row[j+1]));
-          product1 = _mm_madd_epi16(product1, kOnes);
-          sum = _mm_add_epi32(sum, product1);
-        }
-        if (kNumChunks & 0x1) {
-          __m128i product = _mm_maddubs_epi16(_mm_load_si128(&input_vector[kNumChunks-1]), _mm_load_si128(&row[kNumChunks-1]));
-          product = _mm_madd_epi16(product, kOnes);
-          sum = _mm_add_epi32(sum, product);
-        }
+      [[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, Ones256);
+        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_adds_epi16(product0, product1);
+        product0 = _mm256_madd_epi16(product0, Ones256);
+        product2 = _mm256_adds_epi16(product2, product3);
+        product2 = _mm256_madd_epi16(product2, Ones256);
+        acc = _mm256_add_epi32(acc, _mm256_add_epi32(product0, product2));
+#endif
+      };
+
+#endif
+#if defined (USE_SSSE3)
+
+      [[maybe_unused]] const __m128i Ones128 = _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
-        output[i] = _mm_cvtsi128_si32(sum) + biases_[i];
+        return _mm_cvtsi128_si32(sum) + bias;
+      };
 
-  #elif defined(USE_SSE2)
-        __m128i sum_lo = _mm_cvtsi32_si128(biases_[i]);
-        __m128i sum_hi = kZeros;
-        const auto row = reinterpret_cast<const __m128i*>(&weights_[offset]);
-        for (IndexType j = 0; j < kNumChunks; ++j) {
+      [[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, Ones128);
+        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);
+        product0 = _mm_madd_epi16(product0, Ones128);
+        product2 = _mm_adds_epi16(product2, product3);
+        product2 = _mm_madd_epi16(product2, Ones128);
+        acc = _mm_add_epi32(acc, _mm_add_epi32(product0, product2));
+      };
+
+#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)
+      // Different layout, we process 4 inputs at a time, always.
+      static_assert(InputDimensions % 4 == 0);
+
+      const auto output = reinterpret_cast<OutputType*>(buffer);
+      const auto inputVector = reinterpret_cast<const vec_t*>(input);
+
+      static_assert(OutputDimensions % OutputSimdWidth == 0 || OutputDimensions == 1);
+
+      // OutputDimensions is either 1 or a multiple of SimdWidth
+      // because then it is also an input dimension.
+      if constexpr (OutputDimensions % OutputSimdWidth == 0)
+      {
+          constexpr IndexType NumChunks = InputDimensions / 4;
+
+          const auto input32 = reinterpret_cast<const std::int32_t*>(input);
+          vec_t* outptr = reinterpret_cast<vec_t*>(output);
+          std::memcpy(output, biases, OutputDimensions * sizeof(OutputType));
+
+          for (int i = 0; i < (int)NumChunks - 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) * OutputDimensions * 4]);
+              const auto col1 = reinterpret_cast<const vec_t*>(&weights[(i + 1) * OutputDimensions * 4]);
+              const auto col2 = reinterpret_cast<const vec_t*>(&weights[(i + 2) * OutputDimensions * 4]);
+              const auto col3 = reinterpret_cast<const vec_t*>(&weights[(i + 3) * OutputDimensions * 4]);
+              for (int j = 0; j * OutputSimdWidth < OutputDimensions; ++j)
+                  vec_add_dpbusd_32x4(outptr[j], in0, col0[j], in1, col1[j], in2, col2[j], in3, col3[j]);
+          }
+      }
+      else if constexpr (OutputDimensions == 1)
+      {
+#if defined (USE_AVX512)
+          if constexpr (PaddedInputDimensions % (SimdWidth * 2) != 0)
+          {
+              constexpr IndexType NumChunks = PaddedInputDimensions / SimdWidth;
+              const auto inputVector256 = 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)NumChunks; ++j)
+              {
+                  const __m256i in = inputVector256[j];
+                  m256_add_dpbusd_epi32(sum0, in, row0[j]);
+              }
+              output[0] = m256_hadd(sum0, biases[0]);
+          }
+          else
+#endif
+          {
+#if defined (USE_AVX512)
+              constexpr IndexType NumChunks = PaddedInputDimensions / (SimdWidth * 2);
+#else
+              constexpr IndexType NumChunks = PaddedInputDimensions / SimdWidth;
+#endif
+              vec_t sum0 = vec_setzero();
+              const auto row0 = reinterpret_cast<const vec_t*>(&weights[0]);
+
+              for (int j = 0; j < (int)NumChunks; ++j)
+              {
+                  const vec_t in = inputVector[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)
+      // At least a multiple of 16, with SSE2.
+      static_assert(InputDimensions % SimdWidth == 0);
+      constexpr IndexType NumChunks = InputDimensions / SimdWidth;
+      const __m128i Zeros = _mm_setzero_si128();
+      const auto inputVector = reinterpret_cast<const __m128i*>(input);
+
+#elif defined(USE_MMX)
+      static_assert(InputDimensions % SimdWidth == 0);
+      constexpr IndexType NumChunks = InputDimensions / SimdWidth;
+      const __m64 Zeros = _mm_setzero_si64();
+      const auto inputVector = reinterpret_cast<const __m64*>(input);
+
+#elif defined(USE_NEON)
+      static_assert(InputDimensions % SimdWidth == 0);
+      constexpr IndexType NumChunks = InputDimensions / SimdWidth;
+      const auto inputVector = reinterpret_cast<const int8x8_t*>(input);
+#endif
+
+      for (IndexType i = 0; i < OutputDimensions; ++i) {
+        const IndexType offset = i * PaddedInputDimensions;
+
+#if defined(USE_SSE2)
+        __m128i sumLo = _mm_cvtsi32_si128(biases[i]);
+        __m128i sumHi = Zeros;
+        const auto row = reinterpret_cast<const __m128i*>(&weights[offset]);
+        for (IndexType j = 0; j < NumChunks; ++j) {
           __m128i row_j = _mm_load_si128(&row[j]);
-          __m128i input_j = _mm_load_si128(&input_vector[j]);
-          __m128i row_signs = _mm_cmpgt_epi8(kZeros, row_j);
-          __m128i extended_row_lo = _mm_unpacklo_epi8(row_j, row_signs);
-          __m128i extended_row_hi = _mm_unpackhi_epi8(row_j, row_signs);
-          __m128i extended_input_lo = _mm_unpacklo_epi8(input_j, kZeros);
-          __m128i extended_input_hi = _mm_unpackhi_epi8(input_j, kZeros);
-          __m128i product_lo = _mm_madd_epi16(extended_row_lo, extended_input_lo);
-          __m128i product_hi = _mm_madd_epi16(extended_row_hi, extended_input_hi);
-          sum_lo = _mm_add_epi32(sum_lo, product_lo);
-          sum_hi = _mm_add_epi32(sum_hi, product_hi);
+          __m128i input_j = _mm_load_si128(&inputVector[j]);
+          __m128i extendedRowLo = _mm_srai_epi16(_mm_unpacklo_epi8(row_j, row_j), 8);
+          __m128i extendedRowHi = _mm_srai_epi16(_mm_unpackhi_epi8(row_j, row_j), 8);
+          __m128i extendedInputLo = _mm_unpacklo_epi8(input_j, Zeros);
+          __m128i extendedInputHi = _mm_unpackhi_epi8(input_j, Zeros);
+          __m128i productLo = _mm_madd_epi16(extendedRowLo, extendedInputLo);
+          __m128i productHi = _mm_madd_epi16(extendedRowHi, extendedInputHi);
+          sumLo = _mm_add_epi32(sumLo, productLo);
+          sumHi = _mm_add_epi32(sumHi, productHi);
         }
-        __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 = _mm_add_epi32(sumLo, sumHi);
+        __m128i sumHigh_64 = _mm_shuffle_epi32(sum, _MM_SHUFFLE(1, 0, 3, 2));
+        sum = _mm_add_epi32(sum, sumHigh_64);
         __m128i sum_second_32 = _mm_shufflelo_epi16(sum, _MM_SHUFFLE(1, 0, 3, 2));
         sum = _mm_add_epi32(sum, sum_second_32);
         output[i] = _mm_cvtsi128_si32(sum);
 
-  #elif defined(USE_MMX)
-        __m64 sum_lo = _mm_cvtsi32_si64(biases_[i]);
-        __m64 sum_hi = kZeros;
-        const auto row = reinterpret_cast<const __m64*>(&weights_[offset]);
-        for (IndexType j = 0; j < kNumChunks; ++j) {
+#elif defined(USE_MMX)
+        __m64 sumLo = _mm_cvtsi32_si64(biases[i]);
+        __m64 sumHi = Zeros;
+        const auto row = reinterpret_cast<const __m64*>(&weights[offset]);
+        for (IndexType j = 0; j < NumChunks; ++j) {
           __m64 row_j = row[j];
-          __m64 input_j = input_vector[j];
-          __m64 row_signs = _mm_cmpgt_pi8(kZeros, row_j);
-          __m64 extended_row_lo = _mm_unpacklo_pi8(row_j, row_signs);
-          __m64 extended_row_hi = _mm_unpackhi_pi8(row_j, row_signs);
-          __m64 extended_input_lo = _mm_unpacklo_pi8(input_j, kZeros);
-          __m64 extended_input_hi = _mm_unpackhi_pi8(input_j, kZeros);
-          __m64 product_lo = _mm_madd_pi16(extended_row_lo, extended_input_lo);
-          __m64 product_hi = _mm_madd_pi16(extended_row_hi, extended_input_hi);
-          sum_lo = _mm_add_pi32(sum_lo, product_lo);
-          sum_hi = _mm_add_pi32(sum_hi, product_hi);
+          __m64 input_j = inputVector[j];
+          __m64 extendedRowLo = _mm_srai_pi16(_mm_unpacklo_pi8(row_j, row_j), 8);
+          __m64 extendedRowHi = _mm_srai_pi16(_mm_unpackhi_pi8(row_j, row_j), 8);
+          __m64 extendedInputLo = _mm_unpacklo_pi8(input_j, Zeros);
+          __m64 extendedInputHi = _mm_unpackhi_pi8(input_j, Zeros);
+          __m64 productLo = _mm_madd_pi16(extendedRowLo, extendedInputLo);
+          __m64 productHi = _mm_madd_pi16(extendedRowHi, extendedInputHi);
+          sumLo = _mm_add_pi32(sumLo, productLo);
+          sumHi = _mm_add_pi32(sumHi, productHi);
         }
-        __m64 sum = _mm_add_pi32(sum_lo, sum_hi);
+        __m64 sum = _mm_add_pi32(sumLo, sumHi);
         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) {
-          int16x8_t product = vmull_s8(input_vector[j * 2], row[j * 2]);
-          product = vmlal_s8(product, input_vector[j * 2 + 1], row[j * 2 + 1]);
+#elif defined(USE_NEON)
+        int32x4_t sum = {biases[i]};
+        const auto row = reinterpret_cast<const int8x8_t*>(&weights[offset]);
+        for (IndexType j = 0; j < NumChunks; ++j) {
+          int16x8_t product = vmull_s8(inputVector[j * 2], row[j * 2]);
+          product = vmlal_s8(product, inputVector[j * 2 + 1], row[j * 2 + 1]);
           sum = vpadalq_s16(sum, product);
         }
         output[i] = sum[0] + sum[1] + sum[2] + sum[3];
 
-  #else
-        OutputType sum = biases_[i];
-        for (IndexType j = 0; j < kInputDimensions; ++j) {
-          sum += weights_[offset + j] * input[j];
+#else
+        OutputType sum = biases[i];
+        for (IndexType j = 0; j < InputDimensions; ++j) {
+          sum += weights[offset + j] * input[j];
         }
         output[i] = sum;
-  #endif
+#endif
 
       }
-  #if defined(USE_MMX)
+#if defined(USE_MMX)
       _mm_empty();
-  #endif
+#endif
+
+#endif
+
       return output;
     }
 
@@ -254,13 +424,12 @@ namespace Eval::NNUE::Layers {
     using BiasType = OutputType;
     using WeightType = std::int8_t;
 
-    PreviousLayer previous_layer_;
+    PreviousLayer previousLayer;
 
-    alignas(kCacheLineSize) BiasType biases_[kOutputDimensions];
-    alignas(kCacheLineSize)
-        WeightType weights_[kOutputDimensions * kPaddedInputDimensions];
+    alignas(CacheLineSize) BiasType biases[OutputDimensions];
+    alignas(CacheLineSize) WeightType weights[OutputDimensions * PaddedInputDimensions];
   };
 
-}  // namespace Eval::NNUE::Layers
+}  // namespace Stockfish::Eval::NNUE::Layers
 
 #endif // #ifndef NNUE_LAYERS_AFFINE_TRANSFORM_H_INCLUDED

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
@@ -23,7 +23,7 @@
 
 #include "../nnue_common.h"
 
-namespace Eval::NNUE::Layers {
+namespace Stockfish::Eval::NNUE::Layers {
 
   // Clipped ReLU
   template <typename PreviousLayer>
@@ -35,132 +35,157 @@ namespace Eval::NNUE::Layers {
     static_assert(std::is_same<InputType, std::int32_t>::value, "");
 
     // Number of input/output dimensions
-    static constexpr IndexType kInputDimensions =
-        PreviousLayer::kOutputDimensions;
-    static constexpr IndexType kOutputDimensions = kInputDimensions;
+    static constexpr IndexType InputDimensions =
+        PreviousLayer::OutputDimensions;
+    static constexpr IndexType OutputDimensions = InputDimensions;
 
     // Size of forward propagation buffer used in this layer
-    static constexpr std::size_t kSelfBufferSize =
-        CeilToMultiple(kOutputDimensions * sizeof(OutputType), kCacheLineSize);
+    static constexpr std::size_t SelfBufferSize =
+        ceil_to_multiple(OutputDimensions * sizeof(OutputType), CacheLineSize);
 
     // Size of the forward propagation buffer used from the input layer to this layer
-    static constexpr std::size_t kBufferSize =
-        PreviousLayer::kBufferSize + kSelfBufferSize;
+    static constexpr std::size_t BufferSize =
+        PreviousLayer::BufferSize + SelfBufferSize;
 
     // Hash value embedded in the evaluation file
-    static constexpr std::uint32_t GetHashValue() {
-      std::uint32_t hash_value = 0x538D24C7u;
-      hash_value += PreviousLayer::GetHashValue();
-      return hash_value;
+    static constexpr std::uint32_t get_hash_value() {
+      std::uint32_t hashValue = 0x538D24C7u;
+      hashValue += PreviousLayer::get_hash_value();
+      return hashValue;
     }
 
     // Read network parameters
-    bool ReadParameters(std::istream& stream) {
-      return previous_layer_.ReadParameters(stream);
+    bool read_parameters(std::istream& stream) {
+      return previousLayer.read_parameters(stream);
+    }
+
+    // Write network parameters
+    bool write_parameters(std::ostream& stream) const {
+      return previousLayer.write_parameters(stream);
     }
 
     // Forward propagation
-    const OutputType* Propagate(
-        const TransformedFeatureType* transformed_features, char* buffer) const {
-      const auto input = previous_layer_.Propagate(
-          transformed_features, buffer + kSelfBufferSize);
+    const OutputType* propagate(
+        const TransformedFeatureType* transformedFeatures, char* buffer) const {
+      const auto input = previousLayer.propagate(
+          transformedFeatures, buffer + SelfBufferSize);
       const auto output = reinterpret_cast<OutputType*>(buffer);
 
   #if defined(USE_AVX2)
-      constexpr IndexType kNumChunks = kInputDimensions / kSimdWidth;
-      const __m256i kZero = _mm256_setzero_si256();
-      const __m256i kOffsets = _mm256_set_epi32(7, 3, 6, 2, 5, 1, 4, 0);
-      const auto in = reinterpret_cast<const __m256i*>(input);
-      const auto out = reinterpret_cast<__m256i*>(output);
-      for (IndexType i = 0; i < kNumChunks; ++i) {
-        const __m256i words0 = _mm256_srai_epi16(_mm256_packs_epi32(
-            _mm256_loadA_si256(&in[i * 4 + 0]),
-            _mm256_loadA_si256(&in[i * 4 + 1])), kWeightScaleBits);
-        const __m256i words1 = _mm256_srai_epi16(_mm256_packs_epi32(
-            _mm256_loadA_si256(&in[i * 4 + 2]),
-            _mm256_loadA_si256(&in[i * 4 + 3])), kWeightScaleBits);
-        _mm256_storeA_si256(&out[i], _mm256_permutevar8x32_epi32(_mm256_max_epi8(
-            _mm256_packs_epi16(words0, words1), kZero), kOffsets));
+      if constexpr (InputDimensions % SimdWidth == 0) {
+        constexpr IndexType NumChunks = InputDimensions / SimdWidth;
+        const __m256i Zero = _mm256_setzero_si256();
+        const __m256i Offsets = _mm256_set_epi32(7, 3, 6, 2, 5, 1, 4, 0);
+        const auto in = reinterpret_cast<const __m256i*>(input);
+        const auto out = reinterpret_cast<__m256i*>(output);
+        for (IndexType i = 0; i < NumChunks; ++i) {
+          const __m256i words0 = _mm256_srai_epi16(_mm256_packs_epi32(
+              _mm256_load_si256(&in[i * 4 + 0]),
+              _mm256_load_si256(&in[i * 4 + 1])), WeightScaleBits);
+          const __m256i words1 = _mm256_srai_epi16(_mm256_packs_epi32(
+              _mm256_load_si256(&in[i * 4 + 2]),
+              _mm256_load_si256(&in[i * 4 + 3])), WeightScaleBits);
+          _mm256_store_si256(&out[i], _mm256_permutevar8x32_epi32(_mm256_max_epi8(
+              _mm256_packs_epi16(words0, words1), Zero), Offsets));
+        }
+      } else {
+        constexpr IndexType NumChunks = InputDimensions / (SimdWidth / 2);
+        const __m128i Zero = _mm_setzero_si128();
+        const auto in = reinterpret_cast<const __m128i*>(input);
+        const auto out = reinterpret_cast<__m128i*>(output);
+        for (IndexType i = 0; i < NumChunks; ++i) {
+          const __m128i words0 = _mm_srai_epi16(_mm_packs_epi32(
+              _mm_load_si128(&in[i * 4 + 0]),
+              _mm_load_si128(&in[i * 4 + 1])), WeightScaleBits);
+          const __m128i words1 = _mm_srai_epi16(_mm_packs_epi32(
+              _mm_load_si128(&in[i * 4 + 2]),
+              _mm_load_si128(&in[i * 4 + 3])), WeightScaleBits);
+          const __m128i packedbytes = _mm_packs_epi16(words0, words1);
+          _mm_store_si128(&out[i], _mm_max_epi8(packedbytes, Zero));
+        }
       }
-      constexpr IndexType kStart = kNumChunks * kSimdWidth;
+      constexpr IndexType Start =
+        InputDimensions % SimdWidth == 0
+        ? InputDimensions / SimdWidth * SimdWidth
+        : InputDimensions / (SimdWidth / 2) * (SimdWidth / 2);
 
   #elif defined(USE_SSE2)
-      constexpr IndexType kNumChunks = kInputDimensions / kSimdWidth;
+      constexpr IndexType NumChunks = InputDimensions / SimdWidth;
 
   #ifdef USE_SSE41
-      const __m128i kZero = _mm_setzero_si128();
+      const __m128i Zero = _mm_setzero_si128();
   #else
       const __m128i k0x80s = _mm_set1_epi8(-128);
   #endif
 
       const auto in = reinterpret_cast<const __m128i*>(input);
       const auto out = reinterpret_cast<__m128i*>(output);
-      for (IndexType i = 0; i < kNumChunks; ++i) {
+      for (IndexType i = 0; i < NumChunks; ++i) {
         const __m128i words0 = _mm_srai_epi16(_mm_packs_epi32(
             _mm_load_si128(&in[i * 4 + 0]),
-            _mm_load_si128(&in[i * 4 + 1])), kWeightScaleBits);
+            _mm_load_si128(&in[i * 4 + 1])), WeightScaleBits);
         const __m128i words1 = _mm_srai_epi16(_mm_packs_epi32(
             _mm_load_si128(&in[i * 4 + 2]),
-            _mm_load_si128(&in[i * 4 + 3])), kWeightScaleBits);
+            _mm_load_si128(&in[i * 4 + 3])), WeightScaleBits);
         const __m128i packedbytes = _mm_packs_epi16(words0, words1);
         _mm_store_si128(&out[i],
 
   #ifdef USE_SSE41
-          _mm_max_epi8(packedbytes, kZero)
+          _mm_max_epi8(packedbytes, Zero)
   #else
           _mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s)
   #endif
 
         );
       }
-      constexpr IndexType kStart = kNumChunks * kSimdWidth;
+      constexpr IndexType Start = NumChunks * SimdWidth;
 
   #elif defined(USE_MMX)
-      constexpr IndexType kNumChunks = kInputDimensions / kSimdWidth;
+      constexpr IndexType NumChunks = InputDimensions / SimdWidth;
       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) {
+      for (IndexType i = 0; i < NumChunks; ++i) {
         const __m64 words0 = _mm_srai_pi16(
             _mm_packs_pi32(in[i * 4 + 0], in[i * 4 + 1]),
-            kWeightScaleBits);
+            WeightScaleBits);
         const __m64 words1 = _mm_srai_pi16(
             _mm_packs_pi32(in[i * 4 + 2], in[i * 4 + 3]),
-            kWeightScaleBits);
+            WeightScaleBits);
         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;
+      constexpr IndexType Start = NumChunks * SimdWidth;
 
   #elif defined(USE_NEON)
-      constexpr IndexType kNumChunks = kInputDimensions / (kSimdWidth / 2);
-      const int8x8_t kZero = {0};
+      constexpr IndexType NumChunks = InputDimensions / (SimdWidth / 2);
+      const int8x8_t Zero = {0};
       const auto in = reinterpret_cast<const int32x4_t*>(input);
       const auto out = reinterpret_cast<int8x8_t*>(output);
-      for (IndexType i = 0; i < kNumChunks; ++i) {
+      for (IndexType i = 0; i < NumChunks; ++i) {
         int16x8_t shifted;
         const auto pack = reinterpret_cast<int16x4_t*>(&shifted);
-        pack[0] = vqshrn_n_s32(in[i * 2 + 0], kWeightScaleBits);
-        pack[1] = vqshrn_n_s32(in[i * 2 + 1], kWeightScaleBits);
-        out[i] = vmax_s8(vqmovn_s16(shifted), kZero);
+        pack[0] = vqshrn_n_s32(in[i * 2 + 0], WeightScaleBits);
+        pack[1] = vqshrn_n_s32(in[i * 2 + 1], WeightScaleBits);
+        out[i] = vmax_s8(vqmovn_s16(shifted), Zero);
       }
-      constexpr IndexType kStart = kNumChunks * (kSimdWidth / 2);
+      constexpr IndexType Start = NumChunks * (SimdWidth / 2);
   #else
-      constexpr IndexType kStart = 0;
+      constexpr IndexType Start = 0;
   #endif
 
-      for (IndexType i = kStart; i < kInputDimensions; ++i) {
+      for (IndexType i = Start; i < InputDimensions; ++i) {
         output[i] = static_cast<OutputType>(
-            std::max(0, std::min(127, input[i] >> kWeightScaleBits)));
+            std::max(0, std::min(127, input[i] >> WeightScaleBits)));
       }
       return output;
     }
 
    private:
-    PreviousLayer previous_layer_;
+    PreviousLayer previousLayer;
   };
 
-}  // namespace Eval::NNUE::Layers
+}  // namespace Stockfish::Eval::NNUE::Layers
 
 #endif // NNUE_LAYERS_CLIPPED_RELU_H_INCLUDED

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
@@ -23,46 +23,51 @@
 
 #include "../nnue_common.h"
 
-namespace Eval::NNUE::Layers {
+namespace Stockfish::Eval::NNUE::Layers {
 
 // Input layer
-template <IndexType OutputDimensions, IndexType Offset = 0>
+template <IndexType OutDims, IndexType Offset = 0>
 class InputSlice {
  public:
   // Need to maintain alignment
-  static_assert(Offset % kMaxSimdWidth == 0, "");
+  static_assert(Offset % MaxSimdWidth == 0, "");
 
   // Output type
   using OutputType = TransformedFeatureType;
 
   // Output dimensionality
-  static constexpr IndexType kOutputDimensions = OutputDimensions;
+  static constexpr IndexType OutputDimensions = OutDims;
 
   // Size of forward propagation buffer used from the input layer to this layer
-  static constexpr std::size_t kBufferSize = 0;
+  static constexpr std::size_t BufferSize = 0;
 
   // Hash value embedded in the evaluation file
-  static constexpr std::uint32_t GetHashValue() {
-    std::uint32_t hash_value = 0xEC42E90Du;
-    hash_value ^= kOutputDimensions ^ (Offset << 10);
-    return hash_value;
+  static constexpr std::uint32_t get_hash_value() {
+    std::uint32_t hashValue = 0xEC42E90Du;
+    hashValue ^= OutputDimensions ^ (Offset << 10);
+    return hashValue;
   }
 
   // Read network parameters
-  bool ReadParameters(std::istream& /*stream*/) {
+  bool read_parameters(std::istream& /*stream*/) {
+    return true;
+  }
+
+  // Write network parameters
+  bool write_parameters(std::ostream& /*stream*/) const {
     return true;
   }
 
   // Forward propagation
-  const OutputType* Propagate(
-      const TransformedFeatureType* transformed_features,
+  const OutputType* propagate(
+      const TransformedFeatureType* transformedFeatures,
       char* /*buffer*/) const {
-    return transformed_features + Offset;
+    return transformedFeatures + Offset;
   }
 
  private:
 };
 
-}  // namespace Layers
+}  // namespace Stockfish::Eval::NNUE::Layers
 
 #endif // #ifndef NNUE_LAYERS_INPUT_SLICE_H_INCLUDED

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
@@ -23,17 +23,15 @@
 
 #include "nnue_architecture.h"
 
-namespace Eval::NNUE {
+namespace Stockfish::Eval::NNUE {
 
   // Class that holds the result of affine transformation of input features
-  struct alignas(kCacheLineSize) Accumulator {
-    std::int16_t
-        accumulation[2][kRefreshTriggers.size()][kTransformedFeatureDimensions];
-    Value score;
-    bool computed_accumulation;
-    bool computed_score;
+  struct alignas(CacheLineSize) Accumulator {
+    std::int16_t accumulation[2][TransformedFeatureDimensions];
+    std::int32_t psqtAccumulation[2][PSQTBuckets];
+    bool computed[2];
   };
 
-}  // namespace Eval::NNUE
+}  // namespace Stockfish::Eval::NNUE
 
 #endif // NNUE_ACCUMULATOR_H_INCLUDED

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
@@ -21,18 +21,40 @@
 #ifndef NNUE_ARCHITECTURE_H_INCLUDED
 #define NNUE_ARCHITECTURE_H_INCLUDED
 
-// Defines the network structure
-#include "architectures/halfkp_256x2-32-32.h"
+#include "nnue_common.h"
 
-namespace Eval::NNUE {
+#include "features/half_ka_v2.h"
 
-  static_assert(kTransformedFeatureDimensions % kMaxSimdWidth == 0, "");
-  static_assert(Network::kOutputDimensions == 1, "");
+#include "layers/input_slice.h"
+#include "layers/affine_transform.h"
+#include "layers/clipped_relu.h"
+
+namespace Stockfish::Eval::NNUE {
+
+  // Input features used in evaluation function
+  using FeatureSet = Features::HalfKAv2;
+
+  // Number of input feature dimensions after conversion
+  constexpr IndexType TransformedFeatureDimensions = 512;
+  constexpr IndexType PSQTBuckets = 8;
+  constexpr IndexType LayerStacks = 8;
+
+  namespace Layers {
+
+    // Define network structure
+    using InputLayer = InputSlice<TransformedFeatureDimensions * 2>;
+    using HiddenLayer1 = ClippedReLU<AffineTransform<InputLayer, 16>>;
+    using HiddenLayer2 = ClippedReLU<AffineTransform<HiddenLayer1, 32>>;
+    using OutputLayer = AffineTransform<HiddenLayer2, 1>;
+
+  }  // namespace Layers
+
+  using Network = Layers::OutputLayer;
+
+  static_assert(TransformedFeatureDimensions % MaxSimdWidth == 0, "");
+  static_assert(Network::OutputDimensions == 1, "");
   static_assert(std::is_same<Network::OutputType, std::int32_t>::value, "");
 
-  // Trigger for full calculation instead of difference calculation
-  constexpr auto kRefreshTriggers = RawFeatures::kRefreshTriggers;
-
-}  // namespace Eval::NNUE
+}  // namespace Stockfish::Eval::NNUE
 
 #endif // #ifndef NNUE_ARCHITECTURE_H_INCLUDED

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
@@ -24,6 +24,8 @@
 #include <cstring>
 #include <iostream>
 
+#include "../misc.h"  // for IsLittleEndian
+
 #if defined(USE_AVX2)
 #include <immintrin.h>
 
@@ -43,77 +45,33 @@
 #include <arm_neon.h>
 #endif
 
-// HACK: Use _mm256_loadu_si256() instead of _mm256_load_si256. Otherwise a binary
-//       compiled with older g++ crashes because the output memory is not aligned
-//       even though alignas is specified.
-#if defined(USE_AVX2)
-#if defined(__GNUC__ ) && (__GNUC__ < 9) && defined(_WIN32) && !defined(__clang__)
-#define _mm256_loadA_si256  _mm256_loadu_si256
-#define _mm256_storeA_si256 _mm256_storeu_si256
-#else
-#define _mm256_loadA_si256  _mm256_load_si256
-#define _mm256_storeA_si256 _mm256_store_si256
-#endif
-#endif
-
-#if defined(USE_AVX512)
-#if defined(__GNUC__ ) && (__GNUC__ < 9) && defined(_WIN32) && !defined(__clang__)
-#define _mm512_loadA_si512   _mm512_loadu_si512
-#define _mm512_storeA_si512  _mm512_storeu_si512
-#else
-#define _mm512_loadA_si512   _mm512_load_si512
-#define _mm512_storeA_si512  _mm512_store_si512
-#endif
-#endif
-
-namespace Eval::NNUE {
+namespace Stockfish::Eval::NNUE {
 
   // Version of the evaluation file
-  constexpr std::uint32_t kVersion = 0x7AF32F16u;
+  constexpr std::uint32_t Version = 0x7AF32F20u;
 
   // Constant used in evaluation value calculation
-  constexpr int FV_SCALE = 16;
-  constexpr int kWeightScaleBits = 6;
+  constexpr int OutputScale = 16;
+  constexpr int WeightScaleBits = 6;
 
   // Size of cache line (in bytes)
-  constexpr std::size_t kCacheLineSize = 64;
+  constexpr std::size_t CacheLineSize = 64;
 
   // SIMD width (in bytes)
   #if defined(USE_AVX2)
-  constexpr std::size_t kSimdWidth = 32;
+  constexpr std::size_t SimdWidth = 32;
 
   #elif defined(USE_SSE2)
-  constexpr std::size_t kSimdWidth = 16;
+  constexpr std::size_t SimdWidth = 16;
 
   #elif defined(USE_MMX)
-  constexpr std::size_t kSimdWidth = 8;
+  constexpr std::size_t SimdWidth = 8;
 
   #elif defined(USE_NEON)
-  constexpr std::size_t kSimdWidth = 16;
+  constexpr std::size_t SimdWidth = 16;
   #endif
 
-  constexpr std::size_t kMaxSimdWidth = 32;
-
-  // unique number for each piece type on each square
-  enum {
-    PS_NONE     =  0,
-    PS_W_PAWN   =  1,
-    PS_B_PAWN   =  1 * SQUARE_NB + 1,
-    PS_W_KNIGHT =  2 * SQUARE_NB + 1,
-    PS_B_KNIGHT =  3 * SQUARE_NB + 1,
-    PS_W_BISHOP =  4 * SQUARE_NB + 1,
-    PS_B_BISHOP =  5 * SQUARE_NB + 1,
-    PS_W_ROOK   =  6 * SQUARE_NB + 1,
-    PS_B_ROOK   =  7 * SQUARE_NB + 1,
-    PS_W_QUEEN  =  8 * SQUARE_NB + 1,
-    PS_B_QUEEN  =  9 * SQUARE_NB + 1,
-    PS_W_KING   = 10 * SQUARE_NB + 1,
-    PS_END      = PS_W_KING, // pieces without kings (pawns included)
-    PS_B_KING   = 11 * SQUARE_NB + 1,
-    PS_END2     = 12 * SQUARE_NB + 1
-  };
-
-  extern uint32_t kpp_board_index[PIECE_NB][COLOR_NB];
+  constexpr std::size_t MaxSimdWidth = 32;
 
   // Type of input feature after conversion
   using TransformedFeatureType = std::uint8_t;
@@ -121,7 +79,7 @@ namespace Eval::NNUE {
 
   // Round n up to be a multiple of base
   template <typename IntType>
-  constexpr IntType CeilToMultiple(IntType n, IntType base) {
+  constexpr IntType ceil_to_multiple(IntType n, IntType base) {
       return (n + base - 1) / base * base;
   }
 
@@ -130,19 +88,77 @@ namespace Eval::NNUE {
   // 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];
+      if (IsLittleEndian)
+          stream.read(reinterpret_cast<char*>(&result), sizeof(IntType));
+      else
+      {
+          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));
+      }
 
-      std::memcpy(&result, &v, sizeof(IntType));
       return result;
   }
 
-}  // namespace Eval::NNUE
+  // write_little_endian() is our utility to write an integer (signed or unsigned, any size)
+  // to a stream in little-endian order. We swap the byte order before the write if
+  // necessary to always write in little endian order, independantly of the byte
+  // ordering of the compiling machine.
+  template <typename IntType>
+  inline void write_little_endian(std::ostream& stream, IntType value) {
+
+      if (IsLittleEndian)
+          stream.write(reinterpret_cast<const char*>(&value), sizeof(IntType));
+      else
+      {
+          std::uint8_t u[sizeof(IntType)];
+          typename std::make_unsigned<IntType>::type v = value;
+
+          std::size_t i = 0;
+          // if constexpr to silence the warning about shift by 8
+          if constexpr (sizeof(IntType) > 1)
+          {
+            for (; i + 1 < sizeof(IntType); ++i)
+            {
+                u[i] = v;
+                v >>= 8;
+            }
+          }
+          u[i] = v;
+
+          stream.write(reinterpret_cast<char*>(u), sizeof(IntType));
+      }
+  }
+
+  // read_little_endian(s, out, N) : read integers in bulk from a little indian stream.
+  // This reads N integers from stream s and put them in array out.
+  template <typename IntType>
+  inline void read_little_endian(std::istream& stream, IntType* out, std::size_t count) {
+      if (IsLittleEndian)
+          stream.read(reinterpret_cast<char*>(out), sizeof(IntType) * count);
+      else
+          for (std::size_t i = 0; i < count; ++i)
+              out[i] = read_little_endian<IntType>(stream);
+  }
+
+  // write_little_endian(s, values, N) : write integers in bulk to a little indian stream.
+  // This takes N integers from array values and writes them on stream s.
+  template <typename IntType>
+  inline void write_little_endian(std::ostream& stream, const IntType* values, std::size_t count) {
+      if (IsLittleEndian)
+          stream.write(reinterpret_cast<const char*>(values), sizeof(IntType) * count);
+      else
+          for (std::size_t i = 0; i < count; ++i)
+              write_little_endian<IntType>(stream, values[i]);
+  }
+
+}  // namespace Stockfish::Eval::NNUE
 
 #endif // #ifndef NNUE_COMMON_H_INCLUDED

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
@@ -23,356 +23,593 @@
 
 #include "nnue_common.h"
 #include "nnue_architecture.h"
-#include "features/index_list.h"
 
 #include <cstring> // std::memset()
 
-namespace Eval::NNUE {
+namespace Stockfish::Eval::NNUE {
+
+  using BiasType       = std::int16_t;
+  using WeightType     = std::int16_t;
+  using PSQTWeightType = std::int32_t;
+
+  // 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
+
+  static_assert(PSQTBuckets % 8 == 0,
+    "Per feature PSQT values cannot be processed at granularity lower than 8 at a time.");
+
+  #ifdef USE_AVX512
+  typedef __m512i vec_t;
+  typedef __m256i psqt_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)
+  #define vec_load_psqt(a) _mm256_load_si256(a)
+  #define vec_store_psqt(a,b) _mm256_store_si256(a,b)
+  #define vec_add_psqt_32(a,b) _mm256_add_epi32(a,b)
+  #define vec_sub_psqt_32(a,b) _mm256_sub_epi32(a,b)
+  #define vec_zero_psqt() _mm256_setzero_si256()
+  #define NumRegistersSIMD 32
+
+  #elif USE_AVX2
+  typedef __m256i vec_t;
+  typedef __m256i psqt_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)
+  #define vec_load_psqt(a) _mm256_load_si256(a)
+  #define vec_store_psqt(a,b) _mm256_store_si256(a,b)
+  #define vec_add_psqt_32(a,b) _mm256_add_epi32(a,b)
+  #define vec_sub_psqt_32(a,b) _mm256_sub_epi32(a,b)
+  #define vec_zero_psqt() _mm256_setzero_si256()
+  #define NumRegistersSIMD 16
+
+  #elif USE_SSE2
+  typedef __m128i vec_t;
+  typedef __m128i psqt_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)
+  #define vec_load_psqt(a) (*(a))
+  #define vec_store_psqt(a,b) *(a)=(b)
+  #define vec_add_psqt_32(a,b) _mm_add_epi32(a,b)
+  #define vec_sub_psqt_32(a,b) _mm_sub_epi32(a,b)
+  #define vec_zero_psqt() _mm_setzero_si128()
+  #define NumRegistersSIMD (Is64Bit ? 16 : 8)
+
+  #elif USE_MMX
+  typedef __m64 vec_t;
+  typedef __m64 psqt_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)
+  #define vec_load_psqt(a) (*(a))
+  #define vec_store_psqt(a,b) *(a)=(b)
+  #define vec_add_psqt_32(a,b) _mm_add_pi32(a,b)
+  #define vec_sub_psqt_32(a,b) _mm_sub_pi32(a,b)
+  #define vec_zero_psqt() _mm_setzero_si64()
+  #define NumRegistersSIMD 8
+
+  #elif USE_NEON
+  typedef int16x8_t vec_t;
+  typedef int32x4_t psqt_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)
+  #define vec_load_psqt(a) (*(a))
+  #define vec_store_psqt(a,b) *(a)=(b)
+  #define vec_add_psqt_32(a,b) vaddq_s32(a,b)
+  #define vec_sub_psqt_32(a,b) vsubq_s32(a,b)
+  #define vec_zero_psqt() psqt_vec_t{0}
+  #define NumRegistersSIMD 16
+
+  #else
+  #undef VECTOR
+
+  #endif
+
+
+  #ifdef VECTOR
+
+      // Compute optimal SIMD register count for feature transformer accumulation.
+
+      // We use __m* types as template arguments, which causes GCC to emit warnings
+      // about losing some attribute information. This is irrelevant to us as we
+      // only take their size, so the following pragma are harmless.
+      #pragma GCC diagnostic push
+      #pragma GCC diagnostic ignored "-Wignored-attributes"
+
+      template <typename SIMDRegisterType,
+                typename LaneType,
+                int      NumLanes,
+                int      MaxRegisters>
+      static constexpr int BestRegisterCount()
+      {
+          #define RegisterSize  sizeof(SIMDRegisterType)
+          #define LaneSize      sizeof(LaneType)
+
+          static_assert(RegisterSize >= LaneSize);
+          static_assert(MaxRegisters <= NumRegistersSIMD);
+          static_assert(MaxRegisters > 0);
+          static_assert(NumRegistersSIMD > 0);
+          static_assert(RegisterSize % LaneSize == 0);
+          static_assert((NumLanes * LaneSize) % RegisterSize == 0);
+
+          const int ideal = (NumLanes * LaneSize) / RegisterSize;
+          if (ideal <= MaxRegisters)
+            return ideal;
+
+          // Look for the largest divisor of the ideal register count that is smaller than MaxRegisters
+          for (int divisor = MaxRegisters; divisor > 1; --divisor)
+            if (ideal % divisor == 0)
+              return divisor;
+
+          return 1;
+      }
+
+      static constexpr int NumRegs     = BestRegisterCount<vec_t, WeightType, TransformedFeatureDimensions, NumRegistersSIMD>();
+      static constexpr int NumPsqtRegs = BestRegisterCount<psqt_vec_t, PSQTWeightType, PSQTBuckets, NumRegistersSIMD>();
+
+      #pragma GCC diagnostic pop
+
+  #endif
+
+
 
   // Input feature converter
   class FeatureTransformer {
 
    private:
     // Number of output dimensions for one side
-    static constexpr IndexType kHalfDimensions = kTransformedFeatureDimensions;
+    static constexpr IndexType HalfDimensions = TransformedFeatureDimensions;
+
+    #ifdef VECTOR
+    static constexpr IndexType TileHeight = NumRegs * sizeof(vec_t) / 2;
+    static constexpr IndexType PsqtTileHeight = NumPsqtRegs * sizeof(psqt_vec_t) / 4;
+    static_assert(HalfDimensions % TileHeight == 0, "TileHeight must divide HalfDimensions");
+    static_assert(PSQTBuckets % PsqtTileHeight == 0, "PsqtTileHeight must divide PSQTBuckets");
+    #endif
 
    public:
     // Output type
     using OutputType = TransformedFeatureType;
 
     // Number of input/output dimensions
-    static constexpr IndexType kInputDimensions = RawFeatures::kDimensions;
-    static constexpr IndexType kOutputDimensions = kHalfDimensions * 2;
+    static constexpr IndexType InputDimensions = FeatureSet::Dimensions;
+    static constexpr IndexType OutputDimensions = HalfDimensions * 2;
 
     // Size of forward propagation buffer
-    static constexpr std::size_t kBufferSize =
-        kOutputDimensions * sizeof(OutputType);
+    static constexpr std::size_t BufferSize =
+        OutputDimensions * sizeof(OutputType);
 
     // Hash value embedded in the evaluation file
-    static constexpr std::uint32_t GetHashValue() {
-      return RawFeatures::kHashValue ^ kOutputDimensions;
+    static constexpr std::uint32_t get_hash_value() {
+      return FeatureSet::HashValue ^ OutputDimensions;
     }
 
     // Read network parameters
-    bool ReadParameters(std::istream& stream) {
-      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);
+    bool read_parameters(std::istream& stream) {
+
+      read_little_endian<BiasType      >(stream, biases     , HalfDimensions                  );
+      read_little_endian<WeightType    >(stream, weights    , HalfDimensions * InputDimensions);
+      read_little_endian<PSQTWeightType>(stream, psqtWeights, PSQTBuckets    * InputDimensions);
+
       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;
+    // Write network parameters
+    bool write_parameters(std::ostream& stream) const {
+
+      write_little_endian<BiasType      >(stream, biases     , HalfDimensions                  );
+      write_little_endian<WeightType    >(stream, weights    , HalfDimensions * InputDimensions);
+      write_little_endian<PSQTWeightType>(stream, psqtWeights, PSQTBuckets    * InputDimensions);
+
+      return !stream.fail();
     }
 
     // Convert input features
-    void Transform(const Position& pos, OutputType* output, bool refresh) const {
-      if (refresh || !UpdateAccumulatorIfPossible(pos)) {
-        RefreshAccumulator(pos);
-      }
-      const auto& accumulation = pos.state()->accumulator.accumulation;
-
-  #if defined(USE_AVX2)
-      constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
-      constexpr int kControl = 0b11011000;
-      const __m256i kZero = _mm256_setzero_si256();
-
-  #elif defined(USE_SSE2)
-      constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
-
-  #ifdef USE_SSE41
-      const __m128i kZero = _mm_setzero_si128();
-  #else
-      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};
-  #endif
+    std::int32_t transform(const Position& pos, OutputType* output, int bucket) const {
+      update_accumulator(pos, WHITE);
+      update_accumulator(pos, BLACK);
 
       const Color perspectives[2] = {pos.side_to_move(), ~pos.side_to_move()};
-      for (IndexType p = 0; p < 2; ++p) {
-        const IndexType offset = kHalfDimensions * p;
+      const auto& accumulation = pos.state()->accumulator.accumulation;
+      const auto& psqtAccumulation = pos.state()->accumulator.psqtAccumulation;
 
-  #if defined(USE_AVX2)
-        auto out = reinterpret_cast<__m256i*>(&output[offset]);
-        for (IndexType j = 0; j < kNumChunks; ++j) {
-          __m256i sum0 = _mm256_loadA_si256(
-              &reinterpret_cast<const __m256i*>(accumulation[perspectives[p]][0])[j * 2 + 0]);
-          __m256i sum1 = _mm256_loadA_si256(
-            &reinterpret_cast<const __m256i*>(accumulation[perspectives[p]][0])[j * 2 + 1]);
-          _mm256_storeA_si256(&out[j], _mm256_permute4x64_epi64(_mm256_max_epi8(
-              _mm256_packs_epi16(sum0, sum1), kZero), kControl));
-        }
+      const auto psqt = (
+            psqtAccumulation[perspectives[0]][bucket]
+          - psqtAccumulation[perspectives[1]][bucket]
+        ) / 2;
 
-  #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*>(
-              accumulation[perspectives[p]][0])[j * 2 + 0]);
-          __m128i sum1 = _mm_load_si128(&reinterpret_cast<const __m128i*>(
-              accumulation[perspectives[p]][0])[j * 2 + 1]);
-      const __m128i packedbytes = _mm_packs_epi16(sum0, sum1);
 
-          _mm_store_si128(&out[j],
-
-  #ifdef USE_SSE41
-            _mm_max_epi8(packedbytes, kZero)
-  #else
-            _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) {
-          int16x8_t sum = reinterpret_cast<const int16x8_t*>(
-              accumulation[perspectives[p]][0])[j];
-          out[j] = vmax_s8(vqmovn_s16(sum), kZero);
-        }
-
-  #else
-        for (IndexType j = 0; j < kHalfDimensions; ++j) {
-          BiasType sum = accumulation[static_cast<int>(perspectives[p])][0][j];
-          output[offset + j] = static_cast<OutputType>(
-              std::max<int>(0, std::min<int>(127, sum)));
-        }
-  #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]) {
-          const IndexType offset = kHalfDimensions * index;
   #if defined(USE_AVX512)
-          auto accumulation = reinterpret_cast<__m512i*>(
-              &accumulator.accumulation[perspective][i][0]);
-          auto column = reinterpret_cast<const __m512i*>(&weights_[offset]);
-          constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
-          for (IndexType j = 0; j < kNumChunks; ++j)
-            _mm512_storeA_si512(&accumulation[j], _mm512_add_epi16(_mm512_loadA_si512(&accumulation[j]), column[j]));
+
+      constexpr IndexType NumChunks = HalfDimensions / (SimdWidth * 2);
+      static_assert(HalfDimensions % (SimdWidth * 2) == 0);
+      const __m512i Control = _mm512_setr_epi64(0, 2, 4, 6, 1, 3, 5, 7);
+      const __m512i Zero = _mm512_setzero_si512();
+
+      for (IndexType p = 0; p < 2; ++p)
+      {
+          const IndexType offset = HalfDimensions * p;
+          auto out = reinterpret_cast<__m512i*>(&output[offset]);
+          for (IndexType j = 0; j < NumChunks; ++j)
+          {
+              __m512i sum0 = _mm512_load_si512(&reinterpret_cast<const __m512i*>
+                                              (accumulation[perspectives[p]])[j * 2 + 0]);
+              __m512i sum1 = _mm512_load_si512(&reinterpret_cast<const __m512i*>
+                                              (accumulation[perspectives[p]])[j * 2 + 1]);
+
+              _mm512_store_si512(&out[j], _mm512_permutexvar_epi64(Control,
+                                 _mm512_max_epi8(_mm512_packs_epi16(sum0, sum1), Zero)));
+          }
+      }
+      return psqt;
 
   #elif defined(USE_AVX2)
-          auto accumulation = reinterpret_cast<__m256i*>(
-              &accumulator.accumulation[perspective][i][0]);
-          auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
-          constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-          for (IndexType j = 0; j < kNumChunks; ++j)
-            _mm256_storeA_si256(&accumulation[j], _mm256_add_epi16(_mm256_loadA_si256(&accumulation[j]), column[j]));
+
+      constexpr IndexType NumChunks = HalfDimensions / SimdWidth;
+      constexpr int Control = 0b11011000;
+      const __m256i Zero = _mm256_setzero_si256();
+
+      for (IndexType p = 0; p < 2; ++p)
+      {
+          const IndexType offset = HalfDimensions * p;
+          auto out = reinterpret_cast<__m256i*>(&output[offset]);
+          for (IndexType j = 0; j < NumChunks; ++j)
+          {
+              __m256i sum0 = _mm256_load_si256(&reinterpret_cast<const __m256i*>
+                                              (accumulation[perspectives[p]])[j * 2 + 0]);
+              __m256i sum1 = _mm256_load_si256(&reinterpret_cast<const __m256i*>
+                                              (accumulation[perspectives[p]])[j * 2 + 1]);
+
+              _mm256_store_si256(&out[j], _mm256_permute4x64_epi64(
+                                 _mm256_max_epi8(_mm256_packs_epi16(sum0, sum1), Zero), Control));
+          }
+      }
+      return psqt;
 
   #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]);
+
+      #ifdef USE_SSE41
+      constexpr IndexType NumChunks = HalfDimensions / SimdWidth;
+      const __m128i Zero = _mm_setzero_si128();
+      #else
+      constexpr IndexType NumChunks = HalfDimensions / SimdWidth;
+      const __m128i k0x80s = _mm_set1_epi8(-128);
+      #endif
+
+      for (IndexType p = 0; p < 2; ++p)
+      {
+          const IndexType offset = HalfDimensions * p;
+          auto out = reinterpret_cast<__m128i*>(&output[offset]);
+          for (IndexType j = 0; j < NumChunks; ++j)
+          {
+              __m128i sum0 = _mm_load_si128(&reinterpret_cast<const __m128i*>
+                                           (accumulation[perspectives[p]])[j * 2 + 0]);
+              __m128i sum1 = _mm_load_si128(&reinterpret_cast<const __m128i*>
+                                           (accumulation[perspectives[p]])[j * 2 + 1]);
+              const __m128i packedbytes = _mm_packs_epi16(sum0, sum1);
+
+              #ifdef USE_SSE41
+              _mm_store_si128(&out[j], _mm_max_epi8(packedbytes, Zero));
+              #else
+              _mm_store_si128(&out[j], _mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s));
+              #endif
+          }
+      }
+      return psqt;
 
   #elif defined(USE_MMX)
-          auto accumulation = reinterpret_cast<__m64*>(
-              &accumulator.accumulation[perspective][i][0]);
-          auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
-          constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-          for (IndexType j = 0; j < kNumChunks; ++j) {
-            accumulation[j] = _mm_add_pi16(accumulation[j], column[j]);
+
+      constexpr IndexType NumChunks = HalfDimensions / SimdWidth;
+      const __m64 k0x80s = _mm_set1_pi8(-128);
+
+      for (IndexType p = 0; p < 2; ++p)
+      {
+          const IndexType offset = HalfDimensions * p;
+          auto out = reinterpret_cast<__m64*>(&output[offset]);
+          for (IndexType j = 0; j < NumChunks; ++j)
+          {
+              __m64 sum0 = *(&reinterpret_cast<const __m64*>(accumulation[perspectives[p]])[j * 2 + 0]);
+              __m64 sum1 = *(&reinterpret_cast<const __m64*>(accumulation[perspectives[p]])[j * 2 + 1]);
+              const __m64 packedbytes = _mm_packs_pi16(sum0, sum1);
+              out[j] = _mm_subs_pi8(_mm_adds_pi8(packedbytes, k0x80s), k0x80s);
           }
+      }
+      _mm_empty();
+      return psqt;
 
   #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]);
+
+      constexpr IndexType NumChunks = HalfDimensions / (SimdWidth / 2);
+      const int8x8_t Zero = {0};
+
+      for (IndexType p = 0; p < 2; ++p)
+      {
+          const IndexType offset = HalfDimensions * p;
+          const auto out = reinterpret_cast<int8x8_t*>(&output[offset]);
+          for (IndexType j = 0; j < NumChunks; ++j)
+          {
+              int16x8_t sum = reinterpret_cast<const int16x8_t*>(accumulation[perspectives[p]])[j];
+              out[j] = vmax_s8(vqmovn_s16(sum), Zero);
+          }
+      }
+      return psqt;
 
   #else
-          for (IndexType j = 0; j < kHalfDimensions; ++j)
-            accumulator.accumulation[perspective][i][j] += weights_[offset + j];
-  #endif
 
-        }
+      for (IndexType p = 0; p < 2; ++p)
+      {
+          const IndexType offset = HalfDimensions * p;
+          for (IndexType j = 0; j < HalfDimensions; ++j)
+          {
+              BiasType sum = accumulation[perspectives[p]][j];
+              output[offset + j] = static_cast<OutputType>(std::max<int>(0, std::min<int>(127, sum)));
+          }
       }
-  #if defined(USE_MMX)
-      _mm_empty();
+      return psqt;
+
   #endif
 
-      accumulator.computed_accumulation = true;
-      accumulator.computed_score = false;
-    }
+   } // end of function transform()
 
-    // 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]);
+   private:
+    void update_accumulator(const Position& pos, const Color perspective) const {
 
-  #elif defined(USE_MMX)
-        constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-        auto accumulation = reinterpret_cast<__m64*>(
-            &accumulator.accumulation[perspective][i][0]);
+      // The size must be enough to contain the largest possible update.
+      // That might depend on the feature set and generally relies on the
+      // feature set's update cost calculation to be correct and never
+      // allow updates with more added/removed features than MaxActiveDimensions.
+      using IndexList = ValueList<IndexType, FeatureSet::MaxActiveDimensions>;
 
-  #elif defined(USE_NEON)
-        constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
-        auto accumulation = reinterpret_cast<int16x8_t*>(
-            &accumulator.accumulation[perspective][i][0]);
+  #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[NumRegs];
+      psqt_vec_t psqt[NumPsqtRegs];
   #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));
+      // 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 = FeatureSet::refresh_cost(pos);
+      while (st->previous && !st->accumulator.computed[perspective])
+      {
+        // This governs when a full feature refresh is needed and how many
+        // updates are better than just one full refresh.
+        if (   FeatureSet::requires_refresh(st, perspective)
+            || (gain -= FeatureSet::update_cost(st) + 1) < 0)
+          break;
+        next = st;
+        st = st->previous;
+      }
+
+      if (st->accumulator.computed[perspective])
+      {
+        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.
+        const Square ksq = pos.square<KING>(perspective);
+        IndexList removed[2], added[2];
+        FeatureSet::append_changed_indices(
+          ksq, next, perspective, removed[0], added[0]);
+        for (StateInfo *st2 = pos.state(); st2 != next; st2 = st2->previous)
+          FeatureSet::append_changed_indices(
+            ksq, st2, perspective, removed[1], added[1]);
+
+        // Mark the accumulators as computed.
+        next->accumulator.computed[perspective] = true;
+        pos.state()->accumulator.computed[perspective] = true;
+
+        // Now update the accumulators listed in states_to_update[], where the last element is a sentinel.
+        StateInfo *states_to_update[3] =
+          { next, next == pos.state() ? nullptr : pos.state(), nullptr };
+  #ifdef VECTOR
+        for (IndexType j = 0; j < HalfDimensions / TileHeight; ++j)
+        {
+          // Load accumulator
+          auto accTile = reinterpret_cast<vec_t*>(
+            &st->accumulator.accumulation[perspective][j * TileHeight]);
+          for (IndexType k = 0; k < NumRegs; ++k)
+            acc[k] = vec_load(&accTile[k]);
+
+          for (IndexType i = 0; states_to_update[i]; ++i)
+          {
+            // Difference calculation for the deactivated features
+            for (const auto index : removed[i])
+            {
+              const IndexType offset = HalfDimensions * index + j * TileHeight;
+              auto column = reinterpret_cast<const vec_t*>(&weights[offset]);
+              for (IndexType k = 0; k < NumRegs; ++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 = HalfDimensions * index + j * TileHeight;
+              auto column = reinterpret_cast<const vec_t*>(&weights[offset]);
+              for (IndexType k = 0; k < NumRegs; ++k)
+                acc[k] = vec_add_16(acc[k], column[k]);
+            }
+
+            // Store accumulator
+            accTile = reinterpret_cast<vec_t*>(
+              &states_to_update[i]->accumulator.accumulation[perspective][j * TileHeight]);
+            for (IndexType k = 0; k < NumRegs; ++k)
+              vec_store(&accTile[k], acc[k]);
+          }
+        }
+
+        for (IndexType j = 0; j < PSQTBuckets / PsqtTileHeight; ++j)
+        {
+          // Load accumulator
+          auto accTilePsqt = reinterpret_cast<psqt_vec_t*>(
+            &st->accumulator.psqtAccumulation[perspective][j * PsqtTileHeight]);
+          for (std::size_t k = 0; k < NumPsqtRegs; ++k)
+            psqt[k] = vec_load_psqt(&accTilePsqt[k]);
+
+          for (IndexType i = 0; states_to_update[i]; ++i)
+          {
+            // Difference calculation for the deactivated features
+            for (const auto index : removed[i])
+            {
+              const IndexType offset = PSQTBuckets * index + j * PsqtTileHeight;
+              auto columnPsqt = reinterpret_cast<const psqt_vec_t*>(&psqtWeights[offset]);
+              for (std::size_t k = 0; k < NumPsqtRegs; ++k)
+                psqt[k] = vec_sub_psqt_32(psqt[k], columnPsqt[k]);
+            }
+
+            // Difference calculation for the activated features
+            for (const auto index : added[i])
+            {
+              const IndexType offset = PSQTBuckets * index + j * PsqtTileHeight;
+              auto columnPsqt = reinterpret_cast<const psqt_vec_t*>(&psqtWeights[offset]);
+              for (std::size_t k = 0; k < NumPsqtRegs; ++k)
+                psqt[k] = vec_add_psqt_32(psqt[k], columnPsqt[k]);
+            }
+
+            // Store accumulator
+            accTilePsqt = reinterpret_cast<psqt_vec_t*>(
+              &states_to_update[i]->accumulator.psqtAccumulation[perspective][j * PsqtTileHeight]);
+            for (std::size_t k = 0; k < NumPsqtRegs; ++k)
+              vec_store_psqt(&accTilePsqt[k], psqt[k]);
+          }
+        }
+
+  #else
+        for (IndexType i = 0; states_to_update[i]; ++i)
+        {
+          std::memcpy(states_to_update[i]->accumulator.accumulation[perspective],
+              st->accumulator.accumulation[perspective],
+              HalfDimensions * sizeof(BiasType));
+
+          for (std::size_t k = 0; k < PSQTBuckets; ++k)
+            states_to_update[i]->accumulator.psqtAccumulation[perspective][k] = st->accumulator.psqtAccumulation[perspective][k];
+
+          st = states_to_update[i];
+
           // Difference calculation for the deactivated features
-          for (const auto index : removed_indices[perspective]) {
-            const IndexType offset = kHalfDimensions * index;
+          for (const auto index : removed[i])
+          {
+            const IndexType offset = HalfDimensions * 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]);
-            }
+            for (IndexType j = 0; j < HalfDimensions; ++j)
+              st->accumulator.accumulation[perspective][j] -= weights[offset + 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]);
-            }
+            for (std::size_t k = 0; k < PSQTBuckets; ++k)
+              st->accumulator.psqtAccumulation[perspective][k] -= psqtWeights[index * PSQTBuckets + k];
+          }
 
-  #elif defined(USE_MMX)
-            auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
-              accumulation[j] = _mm_sub_pi16(accumulation[j], column[j]);
-            }
+          // Difference calculation for the activated features
+          for (const auto index : added[i])
+          {
+            const IndexType offset = HalfDimensions * index;
 
-  #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
+            for (IndexType j = 0; j < HalfDimensions; ++j)
+              st->accumulator.accumulation[perspective][j] += weights[offset + j];
 
+            for (std::size_t k = 0; k < PSQTBuckets; ++k)
+              st->accumulator.psqtAccumulation[perspective][k] += psqtWeights[index * PSQTBuckets + k];
           }
         }
-        { // 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_MMX)
-            auto column = reinterpret_cast<const __m64*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
-              accumulation[j] = _mm_add_pi16(accumulation[j], column[j]);
-            }
-
-  #elif defined(USE_NEON)
-            auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
-            for (IndexType j = 0; j < kNumChunks; ++j) {
-              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
-
-          }
-        }
       }
+      else
+      {
+        // Refresh the accumulator
+        auto& accumulator = pos.state()->accumulator;
+        accumulator.computed[perspective] = true;
+        IndexList active;
+        FeatureSet::append_active_indices(pos, perspective, active);
+
+  #ifdef VECTOR
+        for (IndexType j = 0; j < HalfDimensions / TileHeight; ++j)
+        {
+          auto biasesTile = reinterpret_cast<const vec_t*>(
+              &biases[j * TileHeight]);
+          for (IndexType k = 0; k < NumRegs; ++k)
+            acc[k] = biasesTile[k];
+
+          for (const auto index : active)
+          {
+            const IndexType offset = HalfDimensions * index + j * TileHeight;
+            auto column = reinterpret_cast<const vec_t*>(&weights[offset]);
+
+            for (unsigned k = 0; k < NumRegs; ++k)
+              acc[k] = vec_add_16(acc[k], column[k]);
+          }
+
+          auto accTile = reinterpret_cast<vec_t*>(
+              &accumulator.accumulation[perspective][j * TileHeight]);
+          for (unsigned k = 0; k < NumRegs; k++)
+            vec_store(&accTile[k], acc[k]);
+        }
+
+        for (IndexType j = 0; j < PSQTBuckets / PsqtTileHeight; ++j)
+        {
+          for (std::size_t k = 0; k < NumPsqtRegs; ++k)
+            psqt[k] = vec_zero_psqt();
+
+          for (const auto index : active)
+          {
+            const IndexType offset = PSQTBuckets * index + j * PsqtTileHeight;
+            auto columnPsqt = reinterpret_cast<const psqt_vec_t*>(&psqtWeights[offset]);
+
+            for (std::size_t k = 0; k < NumPsqtRegs; ++k)
+              psqt[k] = vec_add_psqt_32(psqt[k], columnPsqt[k]);
+          }
+
+          auto accTilePsqt = reinterpret_cast<psqt_vec_t*>(
+            &accumulator.psqtAccumulation[perspective][j * PsqtTileHeight]);
+          for (std::size_t k = 0; k < NumPsqtRegs; ++k)
+            vec_store_psqt(&accTilePsqt[k], psqt[k]);
+        }
+
+  #else
+        std::memcpy(accumulator.accumulation[perspective], biases,
+            HalfDimensions * sizeof(BiasType));
+
+        for (std::size_t k = 0; k < PSQTBuckets; ++k)
+          accumulator.psqtAccumulation[perspective][k] = 0;
+
+        for (const auto index : active)
+        {
+          const IndexType offset = HalfDimensions * index;
+
+          for (IndexType j = 0; j < HalfDimensions; ++j)
+            accumulator.accumulation[perspective][j] += weights[offset + j];
+
+          for (std::size_t k = 0; k < PSQTBuckets; ++k)
+            accumulator.psqtAccumulation[perspective][k] += psqtWeights[index * PSQTBuckets + k];
+        }
+  #endif
+      }
+
   #if defined(USE_MMX)
       _mm_empty();
   #endif
-
-      accumulator.computed_accumulation = true;
-      accumulator.computed_score = false;
     }
 
-    using BiasType = std::int16_t;
-    using WeightType = std::int16_t;
-
-    alignas(kCacheLineSize) BiasType biases_[kHalfDimensions];
-    alignas(kCacheLineSize)
-        WeightType weights_[kHalfDimensions * kInputDimensions];
+    alignas(CacheLineSize) BiasType biases[HalfDimensions];
+    alignas(CacheLineSize) WeightType weights[HalfDimensions * InputDimensions];
+    alignas(CacheLineSize) PSQTWeightType psqtWeights[InputDimensions * PSQTBuckets];
   };
 
-}  // namespace Eval::NNUE
+}  // namespace Stockfish::Eval::NNUE
 
 #endif // #ifndef NNUE_FEATURE_TRANSFORMER_H_INCLUDED

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
@@ -24,35 +24,38 @@
 #include "position.h"
 #include "thread.h"
 
+namespace Stockfish {
+
 namespace {
 
   #define V Value
   #define S(mg, eg) make_score(mg, eg)
 
   // Pawn penalties
-  constexpr Score Backward      = S( 8, 27);
-  constexpr Score Doubled       = S(11, 55);
-  constexpr Score Isolated      = S( 5, 17);
-  constexpr Score WeakLever     = S( 2, 54);
-  constexpr Score WeakUnopposed = S(15, 25);
+  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(-13, -4), S(-4, 3) };
+  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, 11, 24, 45, 85 };
+  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 +63,18 @@ 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 +89,14 @@ 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 +109,10 @@ 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 +128,13 @@ 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 +166,7 @@ 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 +184,14 @@ 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;
@@ -237,6 +256,9 @@ 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;
 }
 
@@ -269,7 +291,7 @@ Score Entry::do_king_safety(const Position& pos) {
   if (pawns & attacks_bb<KING>(ksq))
       minPawnDist = 1;
   else while (pawns)
-      minPawnDist = std::min(minPawnDist, distance(ksq, pop_lsb(&pawns)));
+      minPawnDist = std::min(minPawnDist, distance(ksq, pop_lsb(pawns)));
 
   return shelter - make_score(0, 16 * minPawnDist);
 }
@@ -279,3 +301,5 @@ template Score Entry::do_king_safety<WHITE>(const Position& pos);
 template Score Entry::do_king_safety<BLACK>(const Position& pos);
 
 } // namespace Pawns
+
+} // namespace Stockfish

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
@@ -23,7 +23,7 @@
 #include "position.h"
 #include "types.h"
 
-namespace Pawns {
+namespace Stockfish::Pawns {
 
 /// Pawns::Entry contains various information about a pawn structure. A lookup
 /// to the pawn hash table (performed by calling the probe function) returns a
@@ -65,6 +65,6 @@ typedef HashTable<Entry, 131072> Table;
 
 Entry* probe(const Position& pos);
 
-} // namespace Pawns
+} // namespace Stockfish::Pawns
 
 #endif // #ifndef PAWNS_H_INCLUDED

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
@@ -34,6 +34,8 @@
 
 using std::string;
 
+namespace Stockfish {
+
 namespace Zobrist {
 
   Key psq[PIECE_NB][SQUARE_NB];
@@ -71,12 +73,14 @@ std::ostream& operator<<(std::ostream& os, const Position& pos) {
      << std::setfill(' ') << std::dec << "\nCheckers: ";
 
   for (Bitboard b = pos.checkers(); b; )
-      os << UCI::square(pop_lsb(&b)) << " ";
+      os << UCI::square(pop_lsb(b)) << " ";
 
   if (    int(Tablebases::MaxCardinality) >= popcount(pos.pieces())
       && !pos.can_castle(ANY_CASTLING))
   {
       StateInfo st;
+      ASSERT_ALIGNED(&st, Eval::NNUE::CacheLineSize);
+
       Position p;
       p.set(pos.fen(), pos.is_chess960(), &st, pos.this_thread());
       Tablebases::ProbeState s1, s2;
@@ -195,7 +199,6 @@ 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;
 
   ss >> std::noskipws;
@@ -302,7 +305,7 @@ void Position::set_castling_right(Color c, Square rfrom) {
   Square kto = relative_square(c, cr & KING_SIDE ? SQ_G1 : SQ_C1);
   Square rto = relative_square(c, cr & KING_SIDE ? SQ_F1 : SQ_D1);
 
-  castlingPath[cr] =   (between_bb(rfrom, rto) | between_bb(kfrom, kto) | rto | kto)
+  castlingPath[cr] =   (between_bb(rfrom, rto) | between_bb(kfrom, kto))
                     & ~(kfrom | rfrom);
 }
 
@@ -341,7 +344,7 @@ void Position::set_state(StateInfo* si) const {
 
   for (Bitboard b = pieces(); b; )
   {
-      Square s = pop_lsb(&b);
+      Square s = pop_lsb(b);
       Piece pc = piece_on(s);
       si->key ^= Zobrist::psq[pc][s];
 
@@ -392,7 +395,7 @@ Position& Position::set(const string& code, Color c, StateInfo* si) {
 /// Position::fen() returns a FEN representation of the position. In case of
 /// Chess960 the Shredder-FEN notation is used. This is mainly a debugging function.
 
-const string Position::fen() const {
+string Position::fen() const {
 
   int emptyCnt;
   std::ostringstream ss;
@@ -458,7 +461,7 @@ Bitboard Position::slider_blockers(Bitboard sliders, Square s, Bitboard& pinners
 
   while (snipers)
   {
-    Square sniperSq = pop_lsb(&snipers);
+    Square sniperSq = pop_lsb(snipers);
     Bitboard b = between_bb(s, sniperSq) & occupancy;
 
     if (b && !more_than_one(b))
@@ -502,7 +505,7 @@ bool Position::legal(Move m) const {
   // 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)
+  if (type_of(m) == EN_PASSANT)
   {
       Square ksq = square<KING>(us);
       Square capsq = to - pawn_push(us);
@@ -530,22 +533,20 @@ 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
   // attacked by the opponent.
   if (type_of(piece_on(from)) == KING)
-      return !(attackers_to(to) & pieces(~us));
+      return !(attackers_to(to, pieces() ^ from) & pieces(~us));
 
   // 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));
 }
 
 
@@ -561,8 +562,10 @@ 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)
@@ -607,8 +610,8 @@ bool Position::pseudo_legal(const Move m) const {
           if (more_than_one(checkers()))
               return false;
 
-          // Our move must be a blocking evasion or a capture of the checking piece
-          if (!((between_bb(lsb(checkers()), square<KING>(us)) | checkers()) & to))
+          // Our move must be a blocking interposition or a capture of the checking piece
+          if (!(between_bb(square<KING>(us), lsb(checkers())) & to))
               return false;
       }
       // In case of king moves under check we have to remove king so as to catch
@@ -652,7 +655,7 @@ bool Position::gives_check(Move m) const {
   // 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:
+  case EN_PASSANT:
   {
       Square capsq = make_square(file_of(to), rank_of(from));
       Bitboard b = (pieces() ^ from ^ capsq) | to;
@@ -660,19 +663,15 @@ bool Position::gives_check(Move m) const {
       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;
   }
 }
 
@@ -703,8 +702,8 @@ 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;
+  st->accumulator.computed[WHITE] = false;
+  st->accumulator.computed[BLACK] = false;
   auto& dp = st->dirtyPiece;
   dp.dirty_num = 1;
 
@@ -713,7 +712,7 @@ 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));
@@ -739,7 +738,7 @@ 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);
 
@@ -766,7 +765,7 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
       // 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
@@ -812,7 +811,7 @@ 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)))
       {
@@ -935,7 +934,7 @@ void Position::undo_move(Move m) {
       {
           Square capsq = to;
 
-          if (type_of(m) == ENPASSANT)
+          if (type_of(m) == EN_PASSANT)
           {
               capsq -= pawn_push(us);
 
@@ -989,7 +988,7 @@ void Position::do_castling(Color us, Square from, Square& to, Square& rfrom, Squ
 }
 
 
-/// Position::do(undo)_null_move() is used to do(undo) a "null move": it flips
+/// Position::do_null_move() is used to do a "null move": it flips
 /// the side to move without executing any move on the board.
 
 void Position::do_null_move(StateInfo& newSt) {
@@ -997,17 +996,16 @@ 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.computed[WHITE] = false;
+  st->accumulator.computed[BLACK] = false;
+
   if (st->epSquare != SQ_NONE)
   {
       st->key ^= Zobrist::enpassant[file_of(st->epSquare)];
@@ -1015,7 +1013,7 @@ 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;
@@ -1029,6 +1027,9 @@ void Position::do_null_move(StateInfo& newSt) {
   assert(pos_is_ok());
 }
 
+
+/// Position::undo_null_move() must be used to undo a "null move"
+
 void Position::undo_null_move() {
 
   assert(!checkers());
@@ -1040,7 +1041,7 @@ 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 {
 
@@ -1065,7 +1066,7 @@ 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;
 
@@ -1094,8 +1095,8 @@ bool Position::see_ge(Move m, Value threshold) const {
       if (!(stmAttackers = attackers & pieces(stm)))
           break;
 
-      // Don't allow pinned pieces to attack (except the king) as long as
-      // there are pinners on their original square.
+      // Don't allow pinned pieces to attack as long as there are
+      // pinners on their original square.
       if (pinners(~stm) & occupied)
           stmAttackers &= ~blockers_for_king(stm);
 
@@ -1111,7 +1112,7 @@ bool Position::see_ge(Move m, Value threshold) const {
           if ((swap = PawnValueMg - swap) < res)
               break;
 
-          occupied ^= lsb(bb);
+          occupied ^= least_significant_square_bb(bb);
           attackers |= attacks_bb<BISHOP>(to, occupied) & pieces(BISHOP, QUEEN);
       }
 
@@ -1120,7 +1121,7 @@ bool Position::see_ge(Move m, Value threshold) const {
           if ((swap = KnightValueMg - swap) < res)
               break;
 
-          occupied ^= lsb(bb);
+          occupied ^= least_significant_square_bb(bb);
       }
 
       else if ((bb = stmAttackers & pieces(BISHOP)))
@@ -1128,7 +1129,7 @@ bool Position::see_ge(Move m, Value threshold) const {
           if ((swap = BishopValueMg - swap) < res)
               break;
 
-          occupied ^= lsb(bb);
+          occupied ^= least_significant_square_bb(bb);
           attackers |= attacks_bb<BISHOP>(to, occupied) & pieces(BISHOP, QUEEN);
       }
 
@@ -1137,7 +1138,7 @@ bool Position::see_ge(Move m, Value threshold) const {
           if ((swap = RookValueMg - swap) < res)
               break;
 
-          occupied ^= lsb(bb);
+          occupied ^= least_significant_square_bb(bb);
           attackers |= attacks_bb<ROOK>(to, occupied) & pieces(ROOK, QUEEN);
       }
 
@@ -1146,7 +1147,7 @@ bool Position::see_ge(Move m, Value threshold) const {
           if ((swap = QueenValueMg - swap) < res)
               break;
 
-          occupied ^= lsb(bb);
+          occupied ^= least_significant_square_bb(bb);
           attackers |=  (attacks_bb<BISHOP>(to, occupied) & pieces(BISHOP, QUEEN))
                       | (attacks_bb<ROOK  >(to, occupied) & pieces(ROOK  , QUEEN));
       }
@@ -1220,7 +1221,7 @@ bool Position::has_game_cycle(int ply) const {
           Square s1 = from_sq(move);
           Square s2 = to_sq(move);
 
-          if (!(between_bb(s1, s2) & pieces()))
+          if (!((between_bb(s1, s2) ^ s2) & pieces()))
           {
               if (ply > i)
                   return true;
@@ -1317,21 +1318,17 @@ bool Position::pos_is_ok() const {
               assert(0 && "pos_is_ok: Bitboards");
 
   StateInfo si = *st;
+  ASSERT_ALIGNED(&si, Eval::NNUE::CacheLineSize);
+
   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})
       {
@@ -1346,3 +1343,5 @@ bool Position::pos_is_ok() const {
 
   return true;
 }
+
+} // namespace Stockfish

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
@@ -26,10 +26,12 @@
 
 #include "bitboard.h"
 #include "evaluate.h"
+#include "psqt.h"
 #include "types.h"
 
 #include "nnue/nnue_accumulator.h"
 
+namespace Stockfish {
 
 /// StateInfo struct stores information needed to restore a Position object to
 /// its previous state when we retract a move. Whenever a move is made on the
@@ -49,11 +51,11 @@ struct StateInfo {
   // Not copied when making a move (will be recomputed anyhow)
   Key        key;
   Bitboard   checkersBB;
-  Piece      capturedPiece;
   StateInfo* previous;
   Bitboard   blockersForKing[COLOR_NB];
   Bitboard   pinners[COLOR_NB];
   Bitboard   checkSquares[PIECE_TYPE_NB];
+  Piece      capturedPiece;
   int        repetition;
 
   // Used by NNUE
@@ -86,7 +88,7 @@ public:
   // FEN string input/output
   Position& set(const std::string& fenStr, bool isChess960, StateInfo* si, Thread* th);
   Position& set(const std::string& code, Color c, StateInfo* si);
-  const std::string fen() const;
+  std::string fen() const;
 
   // Position representation
   Bitboard pieces(PieceType pt) const;
@@ -99,7 +101,6 @@ 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;
 
@@ -114,7 +115,6 @@ public:
   Bitboard blockers_for_king(Color c) const;
   Bitboard check_squares(PieceType pt) const;
   Bitboard pinners(Color c) const;
-  bool is_discovery_check_on_king(Color c, Move m) const;
 
   // Attacks to/from a given square
   Bitboard attackers_to(Square s) const;
@@ -127,7 +127,6 @@ public:
   bool capture(Move m) const;
   bool capture_or_promotion(Move m) const;
   bool gives_check(Move m) const;
-  bool advanced_pawn_push(Move m) const;
   Piece moved_piece(Move m) const;
   Piece captured_piece() const;
 
@@ -172,6 +171,9 @@ public:
   // Used by NNUE
   StateInfo* state() const;
 
+  void put_piece(Piece pc, Square s);
+  void remove_piece(Square s);
+
 private:
   // Initialization helpers (used while setting up a position)
   void set_castling_right(Color c, Square rfrom);
@@ -179,8 +181,6 @@ private:
   void set_check_info(StateInfo* si) const;
 
   // Other helpers
-  void put_piece(Piece pc, Square s);
-  void remove_piece(Square s);
   void move_piece(Square from, Square to);
   template<bool Do>
   void do_castling(Color us, Square from, Square& to, Square& rfrom, Square& rto);
@@ -190,23 +190,17 @@ private:
   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];
+  Thread* thisThread;
+  StateInfo* st;
   int gamePly;
   Color sideToMove;
   Score psq;
-  Thread* thisThread;
-  StateInfo* st;
   bool chess960;
 };
 
-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 {
@@ -254,13 +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 {
@@ -311,25 +301,17 @@ 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 {
-  return st->blockersForKing[c] & from_sq(m);
-}
-
 inline bool Position::pawn_passed(Color c, Square s) const {
   return !(pieces(~c, PAWN) & passed_pawn_span(c, s));
 }
 
-inline bool Position::advanced_pawn_push(Move m) const {
-  return   type_of(moved_piece(m)) == PAWN
-        && relative_rank(sideToMove, to_sq(m)) > RANK_5;
-}
-
 inline int Position::pawns_on_same_color_squares(Color c, Square s) const {
   return popcount(pieces(c, PAWN) & ((DarkSquares & s) ? DarkSquares : ~DarkSquares));
 }
 
 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 {
@@ -378,7 +360,7 @@ 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 {
@@ -394,35 +376,25 @@ 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;
+  board[s] = NO_PIECE;
+  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;
@@ -430,8 +402,6 @@ 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];
 }
 
@@ -444,4 +414,6 @@ inline StateInfo* Position::state() const {
   return st;
 }
 
+} // namespace Stockfish
+
 #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 +16,23 @@
   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 {
+namespace Stockfish {
 
-#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 +47,14 @@ 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 +68,13 @@ 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 +91,22 @@ 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.
@@ -107,16 +114,18 @@ 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];
+    }
   }
 }
 
 } // namespace PSQT
+
+} // namespace Stockfish

src/psqt.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
@@ -16,30 +16,23 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-//Common header of input features of NNUE evaluation function
 
-#ifndef NNUE_FEATURES_COMMON_H_INCLUDED
-#define NNUE_FEATURES_COMMON_H_INCLUDED
+#ifndef PSQT_H_INCLUDED
+#define PSQT_H_INCLUDED
 
-#include "../../evaluate.h"
-#include "../nnue_common.h"
 
-namespace Eval::NNUE::Features {
+#include "types.h"
 
-  class IndexList;
 
-  template <typename... FeatureTypes>
-  class FeatureSet;
+namespace Stockfish::PSQT
+{
 
-  // Trigger to perform full calculations instead of difference only
-  enum class TriggerEvent {
-    kFriendKingMoved // calculate full evaluation when own king moves
-  };
+extern Score psq[PIECE_NB][SQUARE_NB];
 
-  enum class Side {
-    kFriend // side to move
-  };
+// Fill psqt array from a set of internally linked parameters
+extern void init();
 
-}  // namespace Eval::NNUE::Features
+} // namespace Stockfish::PSQT
 
-#endif // #ifndef NNUE_FEATURES_COMMON_H_INCLUDED
+
+#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
@@ -25,6 +25,8 @@
 #include "movepick.h"
 #include "types.h"
 
+namespace Stockfish {
+
 class Position;
 
 namespace Search {
@@ -49,6 +51,8 @@ struct Stack {
   int moveCount;
   bool inCheck;
   bool ttPv;
+  bool ttHit;
+  int doubleExtensions;
 };
 
 
@@ -70,7 +74,6 @@ struct RootMove {
   Value previousScore = -VALUE_INFINITE;
   int selDepth = 0;
   int tbRank = 0;
-  int bestMoveCount = 0;
   Value tbScore;
   std::vector<Move> pv;
 };
@@ -106,4 +109,6 @@ void clear();
 
 } // namespace Search
 
+} // namespace Stockfish
+
 #endif // #ifndef SEARCH_H_INCLUDED

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
@@ -50,9 +50,11 @@
 #include <windows.h>
 #endif
 
-using namespace Tablebases;
+using namespace Stockfish::Tablebases;
 
-int Tablebases::MaxCardinality;
+int Stockfish::Tablebases::MaxCardinality;
+
+namespace Stockfish {
 
 namespace {
 
@@ -103,9 +105,6 @@ template<> inline void swap_endian<uint8_t>(uint8_t&) {}
 
 template<typename T, int LE> T number(void* addr)
 {
-    static const union { uint32_t i; char c[4]; } Le = { 0x01020304 };
-    static const bool IsLittleEndian = (Le.c[0] == 4);
-
     T v;
 
     if ((uintptr_t)addr & (alignof(T) - 1)) // Unaligned pointer (very rare)
@@ -190,7 +189,8 @@ public:
         std::stringstream ss(Paths);
         std::string path;
 
-        while (std::getline(ss, path, SepChar)) {
+        while (std::getline(ss, path, SepChar))
+        {
             fname = path + "/" + f;
             std::ifstream::open(fname);
             if (is_open())
@@ -565,7 +565,8 @@ int decompress_pairs(PairsData* d, uint64_t idx) {
     int buf64Size = 64;
     Sym sym;
 
-    while (true) {
+    while (true)
+    {
         int len = 0; // This is the symbol length - d->min_sym_len
 
         // Now get the symbol length. For any symbol s64 of length l right-padded
@@ -603,8 +604,8 @@ int decompress_pairs(PairsData* d, uint64_t idx) {
     // We binary-search for our value recursively expanding into the left and
     // right child symbols until we reach a leaf node where symlen[sym] + 1 == 1
     // that will store the value we need.
-    while (d->symlen[sym]) {
-
+    while (d->symlen[sym])
+    {
         Sym left = d->btree[sym].get<LR::Left>();
 
         // If a symbol contains 36 sub-symbols (d->symlen[sym] + 1 = 36) and
@@ -709,7 +710,7 @@ Ret do_probe_table(const Position& pos, T* entry, WDLScore wdl, ProbeState* resu
 
         leadPawns = b = pos.pieces(color_of(pc), PAWN);
         do
-            squares[size++] = pop_lsb(&b) ^ flipSquares;
+            squares[size++] = pop_lsb(b) ^ flipSquares;
         while (b);
 
         leadPawnsCnt = size;
@@ -729,7 +730,7 @@ Ret do_probe_table(const Position& pos, T* entry, WDLScore wdl, ProbeState* resu
     // directly map them to the correct color and square.
     b = pos.pieces() ^ leadPawns;
     do {
-        Square s = pop_lsb(&b);
+        Square s = pop_lsb(b);
         squares[size] = s ^ flipSquares;
         pieces[size++] = Piece(pos.piece_on(s) ^ flipColor);
     } while (b);
@@ -1000,7 +1001,7 @@ 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;
@@ -1141,7 +1142,7 @@ 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;
@@ -1440,7 +1441,7 @@ 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.
@@ -1535,6 +1536,14 @@ bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves) {
             WDLScore wdl = -probe_wdl(pos, &result);
             dtz = dtz_before_zeroing(wdl);
         }
+        else if (pos.is_draw(1))
+        {
+            // In case a root move leads to a draw by repetition or
+            // 50-move rule, we set dtz to zero. Note: since we are
+            // only 1 ply from the root, this must be a true 3-fold
+            // repetition inside the game history.
+            dtz = 0;
+        }
         else
         {
             // Otherwise, take dtz for the new position and correct by 1 ply
@@ -1585,6 +1594,7 @@ bool Tablebases::root_probe_wdl(Position& pos, Search::RootMoves& rootMoves) {
 
     ProbeState result;
     StateInfo st;
+    WDLScore wdl;
 
     bool rule50 = Options["Syzygy50MoveRule"];
 
@@ -1593,7 +1603,10 @@ bool Tablebases::root_probe_wdl(Position& pos, Search::RootMoves& rootMoves) {
     {
         pos.do_move(m.pv[0], st);
 
-        WDLScore wdl = -probe_wdl(pos, &result);
+        if (pos.is_draw(1))
+            wdl = WDLDraw;
+        else
+            wdl = -probe_wdl(pos, &result);
 
         pos.undo_move(m.pv[0]);
 
@@ -1610,3 +1623,5 @@ bool Tablebases::root_probe_wdl(Position& pos, Search::RootMoves& rootMoves) {
 
     return true;
 }
+
+} // namespace Stockfish

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
@@ -23,7 +23,7 @@
 
 #include "../search.h"
 
-namespace Tablebases {
+namespace Stockfish::Tablebases {
 
 enum WDLScore {
     WDLLoss        = -2, // Loss
@@ -73,6 +73,6 @@ inline std::ostream& operator<<(std::ostream& os, const ProbeState v) {
     return os;
 }
 
-}
+} // namespace Stockfish::Tablebases
 
 #endif

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
@@ -26,6 +26,8 @@
 #include "syzygy/tbprobe.h"
 #include "tt.h"
 
+namespace Stockfish {
+
 ThreadPool Threads; // Global object
 
 
@@ -126,14 +128,16 @@ void Thread::idle_loop() {
 
 void ThreadPool::set(size_t requested) {
 
-  if (size() > 0) { // destroy any existing thread(s)
+  if (size() > 0)   // destroy any existing thread(s)
+  {
       main()->wait_for_search_finished();
 
       while (size() > 0)
           delete back(), pop_back();
   }
 
-  if (requested > 0) { // create new thread(s)
+  if (requested > 0)   // create new thread(s)
+  {
       push_back(new MainThread(0));
 
       while (size() < requested)
@@ -224,16 +228,16 @@ 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;
@@ -258,3 +262,5 @@ void ThreadPool::wait_for_search_finished() const {
         if (th != front())
             th->wait_for_search_finished();
 }
+
+} // namespace Stockfish

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
@@ -32,6 +32,7 @@
 #include "search.h"
 #include "thread_win32_osx.h"
 
+namespace Stockfish {
 
 /// Thread class keeps together all the thread-related stuff. We use
 /// per-thread pawn and material hash tables so that once we get a
@@ -54,6 +55,7 @@ public:
   void idle_loop();
   void start_searching();
   void wait_for_search_finished();
+  size_t id() const { return idx; }
 
   Pawns::Table pawnsTable;
   Material::Table materialTable;
@@ -72,7 +74,7 @@ public:
   LowPlyHistory lowPlyHistory;
   CapturePieceToHistory captureHistory;
   ContinuationHistory continuationHistory[2][2];
-  Score contempt;
+  Score trend;
 };
 
 
@@ -127,4 +129,6 @@ private:
 
 extern ThreadPool Threads;
 
+} // namespace Stockfish
+
 #endif // #ifndef THREAD_H_INCLUDED

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
@@ -31,6 +31,8 @@
 
 #include <pthread.h>
 
+namespace Stockfish {
+
 static const size_t TH_STACK_SIZE = 8 * 1024 * 1024;
 
 template <class T, class P = std::pair<T*, void(T::*)()>>
@@ -57,10 +59,16 @@ public:
   void join() { pthread_join(thread, NULL); }
 };
 
+} // namespace Stockfish
+
 #else // Default case: use STL classes
 
+namespace Stockfish {
+
 typedef std::thread NativeThread;
 
+} // namespace Stockfish
+
 #endif
 
 #endif // #ifndef THREAD_WIN32_OSX_H_INCLUDED

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
@@ -24,6 +24,8 @@
 #include "timeman.h"
 #include "uci.h"
 
+namespace Stockfish {
+
 TimeManagement Time; // Our global time management object
 
 
@@ -75,7 +77,7 @@ 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)
   {
-      optScale = 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));
       maxScale = std::min(7.0, 4.0 + ply / 12.0);
   }
@@ -95,3 +97,5 @@ void TimeManagement::init(Search::LimitsType& limits, Color us, int ply) {
   if (Options["Ponder"])
       optimumTime += optimumTime / 4;
 }
+
+} // namespace Stockfish

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
@@ -23,6 +23,8 @@
 #include "search.h"
 #include "thread.h"
 
+namespace Stockfish {
+
 /// The TimeManagement class computes the optimal time to think depending on
 /// the maximum available time, the game move number and other parameters.
 
@@ -44,4 +46,6 @@ private:
 
 extern TimeManagement Time;
 
+} // namespace Stockfish
+
 #endif // #ifndef TIMEMAN_H_INCLUDED

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
@@ -26,6 +26,8 @@
 #include "tt.h"
 #include "uci.h"
 
+namespace Stockfish {
+
 TranspositionTable TT; // Our global transposition table
 
 /// TTEntry::save() populates the TTEntry with a new node's data, possibly
@@ -62,11 +64,12 @@ 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;
@@ -122,7 +125,7 @@ TTEntry* TranspositionTable::probe(const Key key, bool& found) const {
   for (int i = 0; i < ClusterSize; ++i)
       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].depth8, &tte[i];
       }
@@ -131,11 +134,12 @@ TTEntry* TranspositionTable::probe(const Key key, bool& found) const {
   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;
@@ -150,7 +154,9 @@ 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].depth8 && (table[i].entry[j].genBound8 & 0xF8) == generation8;
+          cnt += table[i].entry[j].depth8 && (table[i].entry[j].genBound8 & GENERATION_MASK) == generation8;
 
   return cnt / ClusterSize;
 }
+
+} // namespace Stockfish

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
@@ -22,6 +22,8 @@
 #include "misc.h"
 #include "types.h"
 
+namespace Stockfish {
+
 /// TTEntry struct is the 10 bytes transposition table entry, defined as below:
 ///
 /// key        16 bit
@@ -72,9 +74,15 @@ 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);
@@ -89,10 +97,11 @@ private:
 
   size_t clusterCount;
   Cluster* table;
-  void* mem;
   uint8_t generation8; // Size must be not bigger than TTEntry::genBound8
 };
 
 extern TranspositionTable TT;
 
+} // namespace Stockfish
+
 #endif // #ifndef TT_H_INCLUDED

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
@@ -26,9 +26,10 @@
 
 using std::string;
 
+namespace Stockfish {
+
 bool Tune::update_on_last;
 const UCI::Option* LastOption = nullptr;
-BoolConditions Conditions;
 static std::map<std::string, int> TuneResults;
 
 string Tune::next(string& names, bool pop) {
@@ -108,23 +109,7 @@ template<> void Tune::Entry<Score>::read_option() {
 template<> void Tune::Entry<Tune::PostUpdate>::init_option() {}
 template<> void Tune::Entry<Tune::PostUpdate>::read_option() { value(); }
 
-
-// Set binary conditions according to a probability that depends
-// on the corresponding parameter value.
-
-void BoolConditions::set() {
-
-  static PRNG rng(now());
-  static bool startup = true; // To workaround fishtest bench
-
-  for (size_t i = 0; i < binary.size(); i++)
-      binary[i] = !startup && (values[i] + int(rng.rand<unsigned>() % variance) > threshold);
-
-  startup = false;
-
-  for (size_t i = 0; i < binary.size(); i++)
-      sync_cout << binary[i] << sync_endl;
-}
+} // namespace Stockfish
 
 
 // Init options with tuning session results instead of default values. Useful to
@@ -138,7 +123,11 @@ void BoolConditions::set() {
 
 #include <cmath>
 
+namespace Stockfish {
+
 void Tune::read_results() {
 
   /* ...insert your values here... */
 }
+
+} // namespace Stockfish

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
@@ -24,6 +24,8 @@
 #include <type_traits>
 #include <vector>
 
+namespace Stockfish {
+
 typedef std::pair<int, int> Range; // Option's min-max values
 typedef Range (RangeFun) (int);
 
@@ -44,27 +46,6 @@ struct SetRange {
 #define SetDefaultRange SetRange(default_range)
 
 
-/// BoolConditions struct is used to tune boolean conditions in the
-/// code by toggling them on/off according to a probability that
-/// depends on the value of a tuned integer parameter: for high
-/// values of the parameter condition is always disabled, for low
-/// values is always enabled, otherwise it is enabled with a given
-/// probability that depnends on the parameter under tuning.
-
-struct BoolConditions {
-  void init(size_t size) { values.resize(size, defaultValue), binary.resize(size, 0); }
-  void set();
-
-  std::vector<int> binary, values;
-  int defaultValue = 465, variance = 40, threshold = 500;
-  SetRange range = SetRange(0, 1000);
-};
-
-extern BoolConditions Conditions;
-
-inline void set_conditions() { Conditions.set(); }
-
-
 /// Tune class implements the 'magic' code that makes the setup of a fishtest
 /// tuning session as easy as it can be. Mainly you have just to remove const
 /// qualifiers from the variables you want to tune and flag them for tuning, so
@@ -130,9 +111,9 @@ 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; }
@@ -157,14 +138,6 @@ class Tune {
     return add(value, (next(names), std::move(names)), args...);
   }
 
-  // Template specialization for BoolConditions
-  template<typename... Args>
-  int add(const SetRange& range, std::string&& names, BoolConditions& cond, Args&&... args) {
-    for (size_t size = cond.values.size(), i = 0; i < size; i++)
-        add(cond.range, next(names, i == size - 1) + "_" + std::to_string(i), cond.values[i]);
-    return add(range, std::move(names), args...);
-  }
-
   std::vector<std::unique_ptr<EntryBase>> list;
 
 public:
@@ -185,9 +158,6 @@ public:
 
 #define UPDATE_ON_LAST() bool UNIQUE(p, __LINE__) = Tune::update_on_last = true
 
-// Some macro to tune toggling of boolean conditions
-#define CONDITION(x) (Conditions.binary[__COUNTER__] || (x))
-#define TUNE_CONDITIONS() int UNIQUE(c, __LINE__) = (Conditions.init(__COUNTER__), 0); \
-                          TUNE(Conditions, set_conditions)
+} // namespace Stockfish
 
 #endif // #ifndef TUNE_H_INCLUDED

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
@@ -77,6 +83,8 @@
 #  define pext(b, m) 0
 #endif
 
+namespace Stockfish {
+
 #ifdef USE_POPCNT
 constexpr bool HasPopCnt = true;
 #else
@@ -107,7 +115,7 @@ 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 +129,7 @@ enum Move : int {
 enum MoveType {
   NORMAL,
   PROMOTION = 1 << 14,
-  ENPASSANT = 2 << 14,
+  EN_PASSANT = 2 << 14,
   CASTLING  = 3 << 14
 };
 
@@ -183,7 +191,6 @@ enum Value : int {
   BishopValueMg = 825,   BishopValueEg = 915,
   RookValueMg   = 1276,  RookValueEg   = 1380,
   QueenValueMg  = 2538,  QueenValueEg  = 2682,
-  Tempo = 28,
 
   MidgameLimit  = 15258, EndgameLimit  = 3915
 };
@@ -196,8 +203,8 @@ 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
 };
 
@@ -476,6 +483,8 @@ constexpr Key make_key(uint64_t seed) {
   return seed * 6364136223846793005ULL + 1442695040888963407ULL;
 }
 
+} // namespace Stockfish
+
 #endif // #ifndef TYPES_H_INCLUDED
 
 #include "tune.h" // Global visibility to tuning setup

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
@@ -34,6 +34,8 @@
 
 using namespace std;
 
+namespace Stockfish {
+
 extern vector<string> setup_bench(const Position&, istream&);
 
 namespace {
@@ -85,7 +87,7 @@ 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 +172,7 @@ 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);
@@ -205,13 +207,13 @@ namespace {
      // Coefficients of a 3rd order polynomial fit based on fishtest data
      // for two parameters needed to transform eval to the argument of a
      // logistic function.
-     double as[] = {-8.24404295, 64.23892342, -95.73056462, 153.86478679};
-     double bs[] = {-3.37154371, 28.44489198, -56.67657741,  72.05858751};
+     double as[] = {-3.68389304,  30.07065921, -60.52878723, 149.53378557};
+     double bs[] = {-2.0181857,   15.85685038, -29.83452023,  47.59078827};
      double a = (((as[0] * m + as[1]) * m + as[2]) * m) + as[3];
      double b = (((bs[0] * m + bs[1]) * m + bs[2]) * m) + bs[3];
 
      // Transform eval to centipawns with limited range
-     double x = std::clamp(double(100 * v) / PawnValueEg, -1000.0, 1000.0);
+     double x = std::clamp(double(100 * v) / PawnValueEg, -2000.0, 2000.0);
 
      // Return win rate in per mille (rounded to nearest)
      return int(0.5 + 1000 / (1 + std::exp((a - x) / b)));
@@ -275,7 +277,15 @@ void UCI::loop(int argc, char* argv[]) {
       else if (token == "d")        sync_cout << pos << sync_endl;
       else if (token == "eval")     trace_eval(pos);
       else if (token == "compiler") sync_cout << compiler_info() << sync_endl;
-      else
+      else if (token == "export_net")
+      {
+          std::optional<std::string> filename;
+          std::string f;
+          if (is >> skipws >> f)
+              filename = f;
+          Eval::NNUE::save_eval(filename);
+      }
+      else if (!token.empty() && token[0] != '#')
           sync_cout << "Unknown command: " << cmd << sync_endl;
 
   } while (token != "quit" && argc == 1); // Command line args are one-shot
@@ -369,3 +379,5 @@ Move UCI::to_move(const Position& pos, string& str) {
 
   return MOVE_NONE;
 }
+
+} // namespace Stockfish

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
@@ -24,6 +24,8 @@
 
 #include "types.h"
 
+namespace Stockfish {
+
 class Position;
 
 namespace UCI {
@@ -78,4 +80,6 @@ Move to_move(const Position& pos, std::string& str);
 
 extern UCI::OptionsMap Options;
 
+} // namespace Stockfish
+
 #endif // #ifndef UCI_H_INCLUDED

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
@@ -31,6 +31,8 @@
 
 using std::string;
 
+namespace Stockfish {
+
 UCI::OptionsMap Options; // Global object
 
 namespace UCI {
@@ -41,8 +43,8 @@ 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 {
@@ -59,8 +61,6 @@ void init(OptionsMap& o) {
   constexpr int MaxHashMB = Is64Bit ? 33554432 : 2048;
 
   o["Debug Log File"]        << Option("", on_logger);
-  o["Contempt"]              << Option(24, -100, 100);
-  o["Analysis Contempt"]     << Option("Both var Off var White var Black var Both", "Both");
   o["Threads"]               << Option(1, 1, 512, on_threads);
   o["Hash"]                  << Option(16, 1, MaxHashMB, on_hash_size);
   o["Clear Hash"]            << Option(on_clear_hash);
@@ -190,3 +190,5 @@ Option& Option::operator=(const string& v) {
 }
 
 } // namespace UCI
+
+} // namespace Stockfish

tests/instrumented.sh

@@ -13,14 +13,14 @@ case $1 in
   --valgrind)
     echo "valgrind testing started"
     prefix=''
-    exeprefix='valgrind --error-exitcode=42'
+    exeprefix='valgrind --error-exitcode=42 --errors-for-leak-kinds=all --leak-check=full'
     postfix='1>/dev/null'
     threads="1"
   ;;
   --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"
   ;;
@@ -39,16 +39,16 @@ case $1 in
     threads="2"
 
 cat << EOF > tsan.supp
-race:TTEntry::move
-race:TTEntry::depth
-race:TTEntry::bound
-race:TTEntry::save
-race:TTEntry::value
-race:TTEntry::eval
-race:TTEntry::is_pv
+race:Stockfish::TTEntry::move
+race:Stockfish::TTEntry::depth
+race:Stockfish::TTEntry::bound
+race:Stockfish::TTEntry::save
+race:Stockfish::TTEntry::value
+race:Stockfish::TTEntry::eval
+race:Stockfish::TTEntry::is_pv
 
-race:TranspositionTable::probe
-race:TranspositionTable::hashfull
+race:Stockfish::TranspositionTable::probe
+race:Stockfish::TranspositionTable::hashfull
 
 EOF
 
@@ -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 20\n"
+ send "go depth 10\n"
  expect "bestmove"
 
  send "quit\n"