Stockfish 16

Official release version of Stockfish 16

Bench: 2593605

---

Stockfish 16

A new major release of Stockfish is now available at

https://stockfishchess.org/download/

*Quality of chess play*

Stockfish continues to demonstrate its ability to discover superior moves
with remarkable speed. In self-play against Stockfish 15, this new
release gains up to 50 Elo[1] and wins up to 12 times more game pairs[2]
than it loses. In major chess engine tournaments, Stockfish reliably tops
the rankings[3] winning the TCEC season 24 Superfinal, Swiss, Fischer
Random, and Double Random Chess tournaments and the CCC 19 Bullet,
20 Blitz, and 20 Rapid competitions. Leela Chess Zero[4] was the
challenger in most finals, putting top-engine chess now firmly in the
hands of teams embracing free and open-source software.

*Progress made*

This updated version of Stockfish introduces several enhancements,
including an upgraded neural net architecture (SFNNv6)[5], improved
implementation, and refined parameterization. The ongoing utilization
of Leela’s data combined with a novel inference approach exploiting
sparsity[6], and network compression[7] ensure a speedy evaluation and
modest binary sizes while allowing for more weights and higher accuracy.
The search has undergone more optimization, leading to improved
performance, particularly in longer analyses[8]. Additionally,
the Fishtest framework has been improved and is now able to run the
tests needed to validate new ideas with 10000s of CPU cores.

*Usability improvements*

Stockfish now comes with documentation, found in the wiki folder when
downloading it or on GitHub[9]. Additionally, Stockfish now includes
a clear and consistent forced tablebase win score, displaying a value
of 200 minus the number of plies required to reach a tablebase win[10].
Furthermore, the UCI_Elo option, to reduce its strength, has been
calibrated[11]. It is worth noting that the evaluation system remains
consistent with Stockfish 15.1[12], maintaining the choice that 100cp
means a 50% chance of winning the game against an equal opponent[13].
Finally, binaries of our latest development version are now provided
continuously as pre-releases on GitHub making it easier for
enthusiasts to download the latest and strongest version of
the program[14], we thank Roman Korba for having provided a similar
service for a long time.

*Thank you*

The success of the Stockfish project relies on the vibrant community
of passionate enthusiasts (we appreciate each and every one of you!)
who generously contribute their knowledge, time, and resources.
Together, this dedicated community works towards the common goal of
developing a powerful, freely accessible, and open-source chess engine.
We invite all chess enthusiasts to join the Fishtest testing framework
and contribute to the project[15].

The Stockfish team

[1] https://tests.stockfishchess.org/tests/view/649409f0dc7002ce609c99cc
[2] https://tests.stockfishchess.org/tests/view/649409d7dc7002ce609c99c6
[3] https://en.wikipedia.org/wiki/Stockfish_(chess)#Competition_results
[4] https://lczero.org/
[5] https://github.com/official-stockfish/Stockfish/commit/c1fff71
[6] https://github.com/official-stockfish/Stockfish/commit/38e6166
[7] https://github.com/official-stockfish/Stockfish/commit/a46087e
[8] https://github.com/official-stockfish/Stockfish/commit/472e726
[9] https://github.com/official-stockfish/Stockfish/wiki/
[10] https://github.com/official-stockfish/Stockfish/commit/def2966
[11] https://github.com/official-stockfish/Stockfish/commit/a08b8d4
[12] https://github.com/official-stockfish/Stockfish/commit/52e84e4
[13] https://github.com/official-stockfish/Stockfish/wiki/Stockfish-FAQ#interpretation-of-the-stockfish-evaluation
[14] https://github.com/official-stockfish/Stockfish/releases?q=prerelease%3Atrue
[15] https://stockfishchess.org/get-involved/

.github/ISSUE_TEMPLATE/BUG-REPORT.yml

@@ -0,0 +1,65 @@
+name: Report issue
+description: Create a report to help us fix issues with the engine
+body:
+- type: textarea
+  attributes:
+    label: Describe the issue
+    description: A clear and concise description of what you're experiencing.
+  validations:
+    required: true
+
+- type: textarea
+  attributes:
+    label: Expected behavior
+    description: A clear and concise description of what you expected to happen.
+  validations:
+    required: true
+
+- type: textarea
+  attributes:
+    label: Steps to reproduce
+    description: |
+      Steps to reproduce the behavior.
+      You can also use this section to paste the command line output.
+    placeholder: |
+      ```
+      position startpos moves g2g4 e7e5 f2f3
+      go mate 1
+      info string NNUE evaluation using nn-6877cd24400e.nnue enabled
+      info depth 1 seldepth 1 multipv 1 score mate 1 nodes 33 nps 11000 tbhits 0 time 3 pv d8h4
+      bestmove d8h4
+      ```
+  validations:
+    required: true
+
+- type: textarea
+  attributes:
+    label: Anything else?
+    description: |
+      Anything that will give us more context about the issue you are encountering.
+      You can also use this section to propose ideas on how to solve the issue. 
+  validations:
+    required: false
+
+- type: dropdown
+  attributes:
+    label: Operating system
+    options:
+      - All
+      - Windows
+      - Linux
+      - MacOS
+      - Android
+      - Other or N/A
+  validations:
+    required: true
+
+- type: input
+  attributes:
+    label: Stockfish version
+    description: |
+      This can be found by running the engine.
+      You can also use the commit ID.
+    placeholder: Stockfish 15 / e6e324e
+  validations:
+    required: true

.github/ISSUE_TEMPLATE/config.yml

@@ -0,0 +1,8 @@
+blank_issues_enabled: false
+contact_links:
+  - name: Discord server
+    url: https://discord.gg/GWDRS3kU6R
+    about: Feel free to ask for support or have a chat with us in our Discord server!
+  - name: Discussions, Q&A, ideas, show us something...
+    url: https://github.com/official-stockfish/Stockfish/discussions/new
+    about: Do you have an idea for Stockfish? Do you want to show something that you made? Please open a discussion about it!

.github/workflows/stockfish.yml

@@ -0,0 +1,51 @@
+name: Stockfish
+on:
+  push:
+    tags:        
+      - '*' 
+    branches:
+      - master
+      - tools
+      - github_ci
+  pull_request:
+    branches:
+      - master
+      - tools
+jobs:
+  Prerelease:
+    if: github.ref == 'refs/heads/master'
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+
+      # returns null if no pre-release exists
+      - name: Get Commit SHA of Latest Pre-release
+        run: |
+          # Install required packages
+          sudo apt-get update
+          sudo apt-get install -y curl jq
+
+          echo "COMMIT_SHA=$(jq -r 'map(select(.prerelease)) | first | .tag_name' <<< $(curl -s https://api.github.com/repos/${{ github.repository_owner }}/Stockfish/releases))" >> $GITHUB_ENV
+
+      # delete old previous pre-release and tag
+      - uses: dev-drprasad/delete-tag-and-release@v0.2.1
+        if: env.COMMIT_SHA != 'null'
+        with:
+          tag_name: ${{ env.COMMIT_SHA }}
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+  Sanitizers:
+    uses: ./.github/workflows/stockfish_sanitizers.yml
+  Tests:
+    uses: ./.github/workflows/stockfish_test.yml
+  Compiles:
+    uses: ./.github/workflows/stockfish_compile_test.yml
+  Binaries:
+    if: github.ref == 'refs/heads/master' || (startsWith(github.ref_name, 'sf_') && github.ref_type == 'tag')
+    uses: ./.github/workflows/stockfish_binaries.yml
+  ARM_Binaries:
+    if: github.ref == 'refs/heads/master' || (startsWith(github.ref_name, 'sf_') && github.ref_type == 'tag')
+    uses: ./.github/workflows/stockfish_arm_binaries.yml

.github/workflows/stockfish_arm_binaries.yml

@@ -0,0 +1,158 @@
+name: Stockfish
+on:
+  workflow_call:
+jobs:
+  Stockfish:
+    name: ${{ matrix.config.name }} ${{ matrix.binaries }}
+    runs-on: ${{ matrix.config.os }}
+    env:
+      COMPILER: ${{ matrix.config.compiler }}
+      COMP: ${{ matrix.config.comp }}
+      EMU: ${{ matrix.config.emu }}
+      EXT: ${{ matrix.config.ext }}
+      OS: ${{ matrix.config.os }}
+      BINARY: ${{ matrix.binaries }}
+    strategy:
+      matrix:
+        config:
+          - name: Android NDK aarch64
+            os: ubuntu-22.04
+            compiler: aarch64-linux-android21-clang++
+            emu: qemu-aarch64
+            comp: ndk
+            shell: bash {0}
+          - name: Android NDK arm
+            os: ubuntu-22.04
+            compiler: armv7a-linux-androideabi21-clang++
+            emu: qemu-arm
+            comp: ndk
+            shell: bash {0}
+        binaries:
+          - armv8
+          - armv7
+          - armv7-neon
+        exclude:
+          - binaries: armv8
+            config: {compiler: armv7a-linux-androideabi21-clang++}
+          - binaries: armv7
+            config: {compiler: aarch64-linux-android21-clang++}
+          - binaries: armv7-neon
+            config: {compiler: aarch64-linux-android21-clang++}
+    defaults:
+      run:
+        working-directory: src
+        shell: ${{ matrix.config.shell }}
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+
+      - name: Download required linux packages
+        if: runner.os == 'Linux'
+        run: |
+          sudo apt update
+          sudo apt install qemu-user
+
+      - name: Install NDK
+        if: runner.os == 'Linux'
+        run: |
+          if [ $COMP == ndk ]; then
+            NDKV="21.4.7075529"
+            ANDROID_ROOT=/usr/local/lib/android
+            ANDROID_SDK_ROOT=$ANDROID_ROOT/sdk
+            SDKMANAGER=$ANDROID_SDK_ROOT/cmdline-tools/latest/bin/sdkmanager
+            echo "y" | $SDKMANAGER "ndk;$NDKV"
+            ANDROID_NDK_ROOT=$ANDROID_SDK_ROOT/ndk/$NDKV
+            ANDROID_NDK_BIN=$ANDROID_NDK_ROOT/toolchains/llvm/prebuilt/linux-x86_64/bin
+            echo "ANDROID_NDK_BIN=$ANDROID_NDK_BIN" >> $GITHUB_ENV
+          fi
+
+      - name: Download the used network from the fishtest framework
+        run: make net
+
+      - name: Check compiler
+        run: |
+          if [ $COMP == ndk ]; then
+            export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
+          fi
+          $COMPILER -v
+
+      - name: Test help target
+        run: make help
+
+      - name: Check git
+        run: git --version
+
+      # Compile profile guided builds
+
+      - name: Compile ${{ matrix.binaries }} build
+        run: |
+          if [ $COMP == ndk ]; then
+            export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
+            export LDFLAGS="-static -Wno-unused-command-line-argument"
+          fi
+          make clean
+          make -j2 profile-build ARCH=$BINARY COMP=$COMP WINE_PATH=$EMU
+          make strip ARCH=$BINARY COMP=$COMP
+          mv ./stockfish$EXT ../stockfish-android-$BINARY$EXT
+
+      - name: Remove non src files
+        run: rm -f *.o .depend *.nnue
+
+      - name: Download wiki
+        run: |
+          git clone https://github.com/official-stockfish/Stockfish.wiki.git ../wiki
+          cd ../wiki
+          rm -rf .git
+
+      - name: Create tar archive.
+        run: |
+          cd ..
+          mkdir stockfish
+          cp -r wiki stockfish/
+          cp -r src stockfish/
+          cp stockfish-android-$BINARY$EXT stockfish/
+          cp "Top CPU Contributors.txt" stockfish/
+          cp Copying.txt stockfish/
+          cp AUTHORS stockfish/
+          cp CITATION.cff stockfish/
+          cp README.md stockfish/
+          tar -cvf stockfish-android-$BINARY.tar stockfish
+
+      - name: Upload binaries
+        uses: actions/upload-artifact@v3
+        with:
+          name: stockfish-android-${{ matrix.binaries }}
+          path: stockfish-android-${{ matrix.binaries }}.tar
+
+      - name: Release
+        if: startsWith(github.ref_name, 'sf_') && github.ref_type == 'tag'
+        uses: softprops/action-gh-release@v1
+        with:
+          files: stockfish-android-${{ matrix.binaries }}.tar
+
+      - name: Get last commit sha
+        id: last_commit
+        run: echo "COMMIT_SHA=$(git rev-parse HEAD | cut -c 1-8)" >> $GITHUB_ENV
+
+      - name: Get commit date
+        id: commit_date
+        run: echo "COMMIT_DATE=$(git show -s --date=format:'%Y%m%d' --format=%cd HEAD)" >> $GITHUB_ENV
+
+      # Make sure that an old ci which still runs on master doesn't recreate a prerelease
+      - name: Check Pullable Commits
+        id: check_commits
+        run: |
+          git fetch
+          CHANGES=$(git rev-list HEAD..origin/master --count)
+          echo "CHANGES=$CHANGES" >> $GITHUB_ENV
+
+      - name: Prerelease
+        if: github.ref_name == 'master' && env.CHANGES == '0'
+        continue-on-error: true
+        uses: softprops/action-gh-release@v1
+        with:
+          name: Stockfish dev-${{ env.COMMIT_DATE }}-${{ env.COMMIT_SHA }}
+          tag_name: stockfish-dev-${{ env.COMMIT_DATE }}-${{ env.COMMIT_SHA }}
+          prerelease: true
+          files: stockfish-android-${{ matrix.binaries }}.tar

.github/workflows/stockfish_binaries.yml

@@ -0,0 +1,168 @@
+name: Stockfish
+on:
+  workflow_call:
+jobs:
+  Stockfish:
+    name: ${{ matrix.config.name }} ${{ matrix.binaries }}
+    runs-on: ${{ matrix.config.os }}
+    env:
+      COMPILER: ${{ matrix.config.compiler }}
+      COMP: ${{ matrix.config.comp }}
+      EXT: ${{ matrix.config.ext }}
+      NAME: ${{ matrix.config.simple_name }}
+      BINARY: ${{ matrix.binaries }}
+    strategy:
+      matrix:
+        config:
+          - name: Ubuntu 20.04 GCC
+            os: ubuntu-20.04
+            simple_name: ubuntu
+            compiler: g++
+            comp: gcc
+            shell: bash {0}
+            archive_ext: tar
+          - name: MacOS 12 Apple Clang
+            os: macos-12
+            simple_name: macos
+            compiler: clang++
+            comp: clang
+            shell: bash {0}
+            archive_ext: tar
+          - name: Windows 2022 Mingw-w64 GCC x86_64
+            os: windows-2022
+            simple_name: windows
+            compiler: g++
+            comp: mingw
+            msys_sys: mingw64
+            msys_env: x86_64-gcc
+            shell: msys2 {0}
+            ext: .exe
+            archive_ext: zip
+        binaries:
+          - x86-64
+          - x86-64-modern
+          - x86-64-avx2
+        exclude:
+          - binaries: x86-64-avx2
+            config: {os: macos-12}
+    defaults:
+      run:
+        working-directory: src
+        shell: ${{ matrix.config.shell }}
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+
+      - name: Download required linux packages
+        if: runner.os == 'Linux'
+        run: sudo apt update
+
+      - name: Setup msys and install required packages
+        if: runner.os == 'Windows'
+        uses: msys2/setup-msys2@v2
+        with:
+          msystem: ${{ matrix.config.msys_sys }}
+          install: mingw-w64-${{ matrix.config.msys_env }} make git zip
+
+      - name: Download the used network from the fishtest framework
+        run: make net
+
+      - name: Check compiler
+        run: $COMPILER -v
+
+      - name: Test help target
+        run: make help
+
+      - name: Check git
+        run: git --version
+
+      # Compile profile guided builds
+
+      - name: Compile ${{ matrix.binaries }} build
+        run: |
+          make -j2 profile-build ARCH=$BINARY COMP=$COMP
+          make strip ARCH=$BINARY COMP=$COMP
+          mv ./stockfish$EXT ../stockfish-$NAME-$BINARY$EXT
+
+      - name: Remove non src files
+        run: git clean -fx
+
+      - name: Download wiki
+        run: |
+          git clone https://github.com/official-stockfish/Stockfish.wiki.git ../wiki
+          rm -rf ../wiki/.git
+
+      - name: Create directory.
+        run: |
+          cd ..
+          mkdir stockfish
+          cp -r wiki stockfish/
+          cp -r src stockfish/
+          cp stockfish-$NAME-$BINARY$EXT stockfish/
+          cp "Top CPU Contributors.txt" stockfish/
+          cp Copying.txt stockfish/
+          cp AUTHORS stockfish/
+          cp CITATION.cff stockfish/
+          cp README.md stockfish/
+
+      - name: Create tar
+        if: runner.os != 'Windows'
+        run: |
+          cd ..
+          tar -cvf stockfish-$NAME-$BINARY.tar stockfish
+
+      - name: Create zip
+        if: runner.os == 'Windows'
+        run: |
+          cd ..
+          zip -r stockfish-$NAME-$BINARY.zip stockfish
+
+      - name: Upload binaries
+        if: runner.os != 'Windows'
+        uses: actions/upload-artifact@v3
+        with:
+          name: stockfish-${{ matrix.config.os }}-${{ matrix.binaries }}
+          path: stockfish-${{ matrix.config.simple_name }}-${{ matrix.binaries }}.tar
+
+      # Artifacts automatically get zipped
+      # to avoid double zipping, we use the unzipped directory
+      - name: Upload binaries
+        if: runner.os == 'Windows'
+        uses: actions/upload-artifact@v3
+        with:
+          name: stockfish-${{ matrix.config.os }}-${{ matrix.binaries }}
+          path: stockfish
+
+      - name: Release
+        if: startsWith(github.ref_name, 'sf_') && github.ref_type == 'tag'
+        uses: softprops/action-gh-release@v1
+        with:
+          files: stockfish-${{ matrix.config.simple_name }}-${{ matrix.binaries }}.${{ matrix.config.archive_ext }}
+
+      - name: Get last commit sha
+        id: last_commit
+        run: echo "COMMIT_SHA=$(git rev-parse HEAD | cut -c 1-8)" >> $GITHUB_ENV
+
+      - name: Get commit date
+        id: commit_date
+        run: echo "COMMIT_DATE=$(git show -s --date=format:'%Y%m%d' --format=%cd HEAD)" >> $GITHUB_ENV
+
+      # Make sure that an old ci which still runs on master doesn't recreate a prerelease
+      - name: Check Pullable Commits
+        id: check_commits
+        run: |
+          git fetch
+          CHANGES=$(git rev-list HEAD..origin/master --count)
+          echo "CHANGES=$CHANGES" >> $GITHUB_ENV
+
+      - name: Prerelease
+        if: github.ref_name == 'master' && env.CHANGES == '0'
+        continue-on-error: true
+        uses: softprops/action-gh-release@v1
+        with:
+          name: Stockfish dev-${{ env.COMMIT_DATE }}-${{ env.COMMIT_SHA }}
+          tag_name: stockfish-dev-${{ env.COMMIT_DATE }}-${{ env.COMMIT_SHA }}
+          prerelease: true
+          files: stockfish-${{ matrix.config.simple_name }}-${{ matrix.binaries }}.${{ matrix.config.archive_ext }}
+

.github/workflows/stockfish_compile_test.yml

@@ -0,0 +1,102 @@
+name: Stockfish
+on:
+  workflow_call:
+jobs:
+  Stockfish:
+    name: ${{ matrix.config.name }}
+    runs-on: ${{ matrix.config.os }}
+    env:
+      COMPILER: ${{ matrix.config.compiler }}
+      COMP: ${{ matrix.config.comp }}
+    strategy:
+      matrix:
+        config:
+          - name: Ubuntu 20.04 GCC
+            os: ubuntu-20.04
+            compiler: g++
+            comp: gcc
+            shell: bash {0}
+          - name: Ubuntu 20.04 Clang
+            os: ubuntu-20.04
+            compiler: clang++
+            comp: clang
+            shell: bash {0}
+          - name: MacOS 12 Apple Clang
+            os: macos-12
+            compiler: clang++
+            comp: clang
+            shell: bash {0}
+          - name: MacOS 12 GCC 11
+            os: macos-12
+            compiler: g++-11
+            comp: gcc
+            shell: bash {0}
+          - name: Windows 2022 Mingw-w64 GCC x86_64
+            os: windows-2022
+            compiler: g++
+            comp: mingw
+            msys_sys: mingw64
+            msys_env: x86_64-gcc
+            shell: msys2 {0}
+          - name: Windows 2022 Mingw-w64 Clang x86_64
+            os: windows-2022
+            compiler: clang++
+            comp: clang
+            msys_sys: clang64
+            msys_env: clang-x86_64-clang
+            shell: msys2 {0}
+
+    defaults:
+      run:
+        working-directory: src
+        shell: ${{ matrix.config.shell }}
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+
+      - name: Setup msys and install required packages
+        if: runner.os == 'Windows'
+        uses: msys2/setup-msys2@v2
+        with:
+          msystem: ${{matrix.config.msys_sys}}
+          install: mingw-w64-${{matrix.config.msys_env}} make git
+
+      - name: Download the used network from the fishtest framework
+        run: make net
+
+      - name: Check compiler
+        run: $COMPILER -v
+
+      - name: Test help target
+        run: make help
+
+      - name: Check git
+        run: git --version
+
+      # 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

.github/workflows/stockfish_sanitizers.yml

@@ -0,0 +1,66 @@
+name: Stockfish
+on:
+  workflow_call:
+jobs:
+  Stockfish:
+    name: ${{ matrix.sanitizers.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
+            shell: bash {0}
+        sanitizers:
+          - name: Run with thread sanitizer
+            make_option: sanitize=thread
+            instrumented_option: sanitizer-thread
+          - name: Run with UB sanitizer
+            make_option: sanitize=undefined
+            instrumented_option: sanitizer-undefined
+          - name: Run under valgrind
+            make_option: ""
+            instrumented_option: valgrind
+          - name: Run under valgrind-thread
+            make_option: ""
+            instrumented_option: valgrind-thread
+    defaults:
+      run:
+        working-directory: src
+        shell: ${{ matrix.config.shell }}
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+
+      - name: Download required linux 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: Check compiler
+        run: $COMPILER -v
+
+      - name: Test help target
+        run: make help
+
+      - name: Check git
+        run: git --version
+
+      # Sanitizers
+
+      - name: ${{ matrix.sanitizers.name }}
+        run: |
+          export CXXFLAGS="-O1 -fno-inline"
+          make clean
+          make -j2 ARCH=x86-64-modern ${{ matrix.sanitizers.make_option }} debug=yes optimize=no build > /dev/null
+          ../tests/instrumented.sh --${{ matrix.sanitizers.instrumented_option }}

.github/workflows/stockfish_test.yml

@@ -0,0 +1,256 @@
+name: Stockfish
+on:
+  workflow_call:
+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_32bit_tests: true
+            run_64bit_tests: true
+            shell: bash {0}
+          - name: Ubuntu 20.04 Clang
+            os: ubuntu-20.04
+            compiler: clang++
+            comp: clang
+            run_32bit_tests: true
+            run_64bit_tests: true
+            shell: bash {0}
+          - name: Android NDK aarch64
+            os: ubuntu-22.04
+            compiler: aarch64-linux-android21-clang++
+            comp: ndk
+            run_armv8_tests: true
+            shell: bash {0}
+          - name: Android NDK arm
+            os: ubuntu-22.04
+            compiler: armv7a-linux-androideabi21-clang++
+            comp: ndk
+            run_armv7_tests: true
+            shell: bash {0}
+          - name: MacOS 12 Apple Clang
+            os: macos-12
+            compiler: clang++
+            comp: clang
+            run_64bit_tests: true
+            shell: bash {0}
+          - name: MacOS 12 GCC 11
+            os: macos-12
+            compiler: g++-11
+            comp: gcc
+            run_64bit_tests: true
+            shell: bash {0}
+          - name: Windows 2022 Mingw-w64 GCC x86_64
+            os: windows-2022
+            compiler: g++
+            comp: mingw
+            run_64bit_tests: true
+            msys_sys: mingw64
+            msys_env: x86_64-gcc
+            shell: msys2 {0}
+          - name: Windows 2022 Mingw-w64 GCC i686
+            os: windows-2022
+            compiler: g++
+            comp: mingw
+            run_32bit_tests: true
+            msys_sys: mingw32
+            msys_env: i686-gcc
+            shell: msys2 {0}
+          - name: Windows 2022 Mingw-w64 Clang x86_64
+            os: windows-2022
+            compiler: clang++
+            comp: clang
+            run_64bit_tests: true
+            msys_sys: clang64
+            msys_env: clang-x86_64-clang
+            shell: msys2 {0}
+    defaults:
+      run:
+        working-directory: src
+        shell: ${{ matrix.config.shell }}
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          fetch-depth: 0
+
+      - name: Download required linux packages
+        if: runner.os == 'Linux'
+        run: |
+          sudo apt update
+          sudo apt install expect valgrind g++-multilib qemu-user
+
+      - name: Install NDK
+        if: runner.os == 'Linux'
+        run: |
+          if [ $COMP == ndk ]; then
+            NDKV="21.4.7075529"
+            ANDROID_ROOT=/usr/local/lib/android
+            ANDROID_SDK_ROOT=$ANDROID_ROOT/sdk
+            SDKMANAGER=$ANDROID_SDK_ROOT/cmdline-tools/latest/bin/sdkmanager
+            echo "y" | $SDKMANAGER "ndk;$NDKV"
+            ANDROID_NDK_ROOT=$ANDROID_SDK_ROOT/ndk/$NDKV
+            ANDROID_NDK_BIN=$ANDROID_NDK_ROOT/toolchains/llvm/prebuilt/linux-x86_64/bin
+            echo "ANDROID_NDK_BIN=$ANDROID_NDK_BIN" >> $GITHUB_ENV
+          fi
+
+      - name: Setup msys and install required packages
+        if: runner.os == 'Windows'
+        uses: msys2/setup-msys2@v2
+        with:
+          msystem: ${{ matrix.config.msys_sys }}
+          install: mingw-w64-${{ matrix.config.msys_env }} make git expect
+
+      - 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 -o "\b[Bb]ench[ :]\+[1-9][0-9]\{5,9\}\b" | head -n 1 | sed "s/[^0-9]//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: |
+          if [ $COMP == ndk ]; then
+            export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
+          fi
+          $COMPILER -v
+
+      - name: Test help target
+        run: make help
+
+      - name: Check git
+        run: git --version
+
+      # x86-32 tests
+
+      - name: Test debug x86-32 build
+        if: ${{ matrix.config.run_32bit_tests }}
+        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
+        if: ${{ matrix.config.run_32bit_tests }}
+        run: |
+          make clean
+          make -j2 ARCH=x86-32 build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-32-sse41-popcnt build
+        if: ${{ matrix.config.run_32bit_tests }}
+        run: |
+          make clean
+          make -j2 ARCH=x86-32-sse41-popcnt build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-32-sse2 build
+        if: ${{ matrix.config.run_32bit_tests }}
+        run: |
+          make clean
+          make -j2 ARCH=x86-32-sse2 build
+          ../tests/signature.sh $benchref
+
+      - name: Test general-32 build
+        if: ${{ matrix.config.run_32bit_tests }}
+        run: |
+          make clean
+          make -j2 ARCH=general-32 build
+          ../tests/signature.sh $benchref
+
+      # x86-64 tests
+
+      - name: Test debug x86-64-modern build
+        if: ${{ matrix.config.run_64bit_tests }}
+        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
+        if: ${{ matrix.config.run_64bit_tests }}
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-modern build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64-ssse3 build
+        if: ${{ matrix.config.run_64bit_tests }}
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-ssse3 build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64-sse3-popcnt build
+        if: ${{ matrix.config.run_64bit_tests }}
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-sse3-popcnt build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64 build
+        if: ${{ matrix.config.run_64bit_tests }}
+        run: |
+          make clean
+          make -j2 ARCH=x86-64 build
+          ../tests/signature.sh $benchref
+
+      - name: Test general-64 build
+        if: matrix.config.run_64bit_tests
+        run: |
+          make clean
+          make -j2 ARCH=general-64 build
+          ../tests/signature.sh $benchref
+
+      # armv8 tests
+
+      - name: Test armv8 build
+        if: ${{ matrix.config.run_armv8_tests }}
+        run: |
+          export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
+          export LDFLAGS="-static -Wno-unused-command-line-argument"
+          make clean
+          make -j2 ARCH=armv8 build
+          ../tests/signature.sh $benchref
+
+      # armv7 tests
+
+      - name: Test armv7 build
+        if: ${{ matrix.config.run_armv7_tests }}
+        run: |
+          export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
+          export LDFLAGS="-static -Wno-unused-command-line-argument"
+          make clean
+          make -j2 ARCH=armv7 build
+          ../tests/signature.sh $benchref
+
+      - name: Test armv7-neon build
+        if: ${{ matrix.config.run_armv7_tests }}
+        run: |
+          export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
+          export LDFLAGS="-static -Wno-unused-command-line-argument"
+          make clean
+          make -j2 ARCH=armv7-neon build
+          ../tests/signature.sh $benchref
+
+      # Other tests
+
+      - name: Check perft and search reproducibility
+        if: ${{ matrix.config.run_64bit_tests }}
+        run: |
+          make clean
+          make -j2 ARCH=x86-64-modern build
+          ../tests/perft.sh
+          ../tests/reprosearch.sh

.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
+

AUTHORS

@@ -0,0 +1,226 @@
+# Founders of the Stockfish project and Fishtest infrastructure
+Tord Romstad (romstad)
+Marco Costalba (mcostalba)
+Joona Kiiski (zamar)
+Gary Linscott (glinscott)
+
+# Authors and inventors of NNUE, training, and NNUE port
+Yu Nasu (ynasu87)
+Motohiro Isozaki (yaneurao)
+Hisayori Noda (nodchip)
+
+# All other authors of Stockfish code (in alphabetical order)
+Aditya (absimaldata)
+Adrian Petrescu (apetresc)
+Ajith Chandy Jose (ajithcj)
+Alain Savard (Rocky640)
+Alayan Feh (Alayan-stk-2)
+Alexander Kure
+Alexander Pagel (Lolligerhans)
+Alfredo Menezes (lonfom169)
+Ali AlZhrani (Cooffe)
+Andreas Matthies (Matthies)
+Andrei Vetrov (proukornew)
+Andrew Grant (AndyGrant)
+Andrey Neporada (nepal)
+Andy Duplain
+Antoine Champion (antoinechampion)
+Aram Tumanian (atumanian)
+Arjun Temurnikar
+Artem Solopiy (EntityFX)
+Auguste Pop
+Balint Pfliegel
+Ben Chaney (Chaneybenjamini)
+Ben Koshy (BKSpurgeon)
+Bill Henry (VoyagerOne)
+Bojun Guo (noobpwnftw, Nooby)
+borg323
+Boštjan Mejak (PedanticHacker)
+braich
+Brian Sheppard (SapphireBrand, briansheppard-toast)
+Bruno de Melo Costa (BM123499)
+Bruno Pellanda (pellanda)
+Bryan Cross (crossbr)
+candirufish
+Chess13234
+Chris Cain (ceebo)
+clefrks
+Dale Weiler (graphitemaster)
+Daniel Axtens (daxtens)
+Daniel Dugovic (ddugovic)
+Dan Schmidt (dfannius)
+Dariusz Orzechowski (dorzechowski)
+David (dav1312)
+David Zar
+Daylen Yang (daylen)
+Deshawn Mohan-Smith (GoldenRare)
+Dieter Dobbelaere (ddobbelaere)
+DiscanX
+Dominik Schlösser (domschl)
+double-beep
+Douglas Matos Gomes (dsmsgms)
+Dubslow
+Eduardo Cáceres (eduherminio)
+Eelco de Groot (KingDefender)
+Elvin Liu (solarlight2)
+erbsenzaehler
+Ernesto Gatti
+Fabian Beuke (madnight)
+Fabian Fichter (ianfab)
+Fanael Linithien (Fanael)
+fanon
+Fauzi Akram Dabat (FauziAkram)
+Felix Wittmann
+gamander
+Gary Heckman (gheckman)
+George Sobala (gsobala)
+gguliash
+Giacomo Lorenzetti (G-Lorenz)
+Gian-Carlo Pascutto (gcp)
+Goh CJ (cj5716)
+Gontran Lemaire (gonlem)
+Goodkov Vasiliy Aleksandrovich (goodkov)
+Gregor Cramer
+GuardianRM
+Guy Vreuls (gvreuls)
+Günther Demetz (pb00067, pb00068)
+Henri Wiechers
+Hiraoka Takuya (HiraokaTakuya)
+homoSapiensSapiens
+Hongzhi Cheng
+Ivan Ivec (IIvec)
+Jacques B. (Timshel)
+Jake Senne (w1wwwwww)
+Jan Ondruš (hxim)
+Jared Kish (Kurtbusch, kurt22i)
+Jarrod Torriero (DU-jdto)
+Jean-Francois Romang (jromang)
+Jean Gauthier (OuaisBla)
+Jekaa
+Jerry Donald Watson (jerrydonaldwatson)
+jjoshua2
+Jonathan Buladas Dumale (SFisGOD)
+Jonathan Calovski (Mysseno)
+Jonathan McDermid (jonathanmcdermid)
+Joost VandeVondele (vondele)
+Joseph Ellis (jhellis3)
+Joseph R. Prostko
+Jörg Oster (joergoster)
+Julian Willemer (NightlyKing)
+jundery
+Justin Blanchard (UncombedCoconut)
+Kelly Wilson
+Ken Takusagawa
+Kian E (KJE-98)
+kinderchocolate
+Kiran Panditrao (Krgp)
+Kojirion
+Krystian Kuzniarek (kuzkry)
+Leonardo Ljubičić (ICCF World Champion)
+Leonid Pechenik (lp--)
+Liam Keegan (lkeegan)
+Linmiao Xu (linrock)
+Linus Arver (listx)
+loco-loco
+Lub van den Berg (ElbertoOne)
+Luca Brivio (lucabrivio)
+Lucas Braesch (lucasart)
+Lyudmil Antonov (lantonov)
+Maciej Żenczykowski (zenczykowski)
+Malcolm Campbell (xoto10)
+Mark Tenzer (31m059)
+marotear
+Matt Ginsberg (mattginsberg)
+Matthew Lai (matthewlai)
+Matthew Sullivan (Matt14916)
+Max A. (Disservin)
+Maxim Masiutin (maximmasiutin)
+Maxim Molchanov (Maxim)
+Michael An (man)
+Michael Byrne (MichaelB7)
+Michael Chaly (Vizvezdenec)
+Michael Stembera (mstembera)
+Michael Whiteley (protonspring)
+Michel Van den Bergh (vdbergh)
+Miguel Lahoz (miguel-l)
+Mikael Bäckman (mbootsector)
+Mike Babigian (Farseer)
+Mira
+Miroslav Fontán (Hexik)
+Moez Jellouli (MJZ1977)
+Mohammed Li (tthsqe12)
+Muzhen J (XInTheDark)
+Nathan Rugg (nmrugg)
+Nguyen Pham (nguyenpham)
+Nicklas Persson (NicklasPersson)
+Nick Pelling (nickpelling)
+Niklas Fiekas (niklasf)
+Nikolay Kostov (NikolayIT)
+Norman Schmidt (FireFather)
+notruck
+Ofek Shochat (OfekShochat, ghostway)
+Ondrej Mosnáček (WOnder93)
+Ondřej Mišina (AndrovT)
+Oskar Werkelin Ahlin
+Pablo Vazquez
+Panthee
+Pascal Romaret
+Pasquale Pigazzini (ppigazzini)
+Patrick Jansen (mibere)
+Peter Schneider (pschneider1968)
+Peter Zsifkovits (CoffeeOne)
+PikaCat
+Praveen Kumar Tummala (praveentml)
+Prokop Randáček (ProkopRandacek)
+Rahul Dsilva (silversolver1)
+Ralph Stößer (Ralph Stoesser)
+Raminder Singh
+renouve
+Reuven Peleg (R-Peleg)
+Richard Lloyd (Richard-Lloyd)
+Rodrigo Exterckötter Tjäder
+Rodrigo Roim (roim)
+Ronald de Man (syzygy1, syzygy)
+Ron Britvich (Britvich)
+rqs
+Rui Coelho (ruicoelhopedro)
+Ryan Schmitt
+Ryan Takker
+Sami Kiminki (skiminki)
+Sebastian Buchwald (UniQP)
+Sergei Antonov (saproj)
+Sergei Ivanov (svivanov72)
+Sergio Vieri (sergiovieri)
+sf-x
+Shahin M. Shahin (peregrine)
+Shane Booth (shane31)
+Shawn Varghese (xXH4CKST3RXx)
+Siad Daboul (Topologist)
+Stefan Geschwentner (locutus2)
+Stefano Cardanobile (Stefano80)
+Stefano Di Martino (StefanoD)
+Steinar Gunderson (sesse)
+Stéphane Nicolet (snicolet)
+Syine Mineta (MinetaS)
+Thanar2
+thaspel
+theo77186
+Tomasz Sobczyk (Sopel97)
+Tom Truscott
+Tom Vijlbrief (tomtor)
+Torsten Franz (torfranz, tfranzer)
+Torsten Hellwig (Torom)
+Tracey Emery (basepr1me)
+tttak
+Unai Corzo (unaiic)
+Uri Blass (uriblass)
+Vince Negri (cuddlestmonkey)
+Viren
+windfishballad
+xefoci7612
+zz4032
+
+# Additionally, we acknowledge the authors and maintainers of fishtest,
+# an amazing and essential framework for Stockfish development!
+#
+# https://github.com/glinscott/fishtest/blob/master/AUTHORS

CITATION.cff

@@ -0,0 +1,23 @@
+# This CITATION.cff file was generated with cffinit.
+# Visit https://bit.ly/cffinit to generate yours today!
+
+cff-version: 1.2.0
+title: Stockfish
+message: >-
+  Please cite this software using the metadata from this
+  file.
+type: software
+authors:
+  - name: The Stockfish developers (see AUTHORS file)
+repository-code: 'https://github.com/official-stockfish/Stockfish'
+url: 'https://stockfishchess.org/'
+repository-artifact: 'https://stockfishchess.org/download/'
+abstract: Stockfish is a free and strong UCI chess engine.
+keywords:
+  - chess
+  - artificial intelligence (AI)
+  - tree search
+  - alpha-beta search
+  - neural networks (NN)
+  - efficiently updatable neural networks (NNUE)
+license: GPL-3.0

README.md

@@ -0,0 +1,164 @@
+<div align="center">
+
+  [![Stockfish][stockfish128-logo]][website-link]
+
+  <h3>Stockfish</h3>
+
+  A free and strong UCI chess engine.
+  <br>
+  <strong>[Explore Stockfish docs »][wiki-link]</strong>
+  <br>
+  <br>
+  [Report bug][issue-link]
+  ·
+  [Open a discussion][discussions-link]
+  ·
+  [Discord][discord-link]
+  ·
+  [Blog][website-blog-link]
+
+  [![Build][build-badge]][build-link]
+  [![License][license-badge]][license-link]
+  <br>
+  [![Release][release-badge]][release-link]
+  [![Commits][commits-badge]][commits-link]
+  <br>
+  [![Website][website-badge]][website-link]
+  [![Fishtest][fishtest-badge]][fishtest-link]
+  [![Discord][discord-badge]][discord-link]
+
+</div>
+
+## Overview
+
+[Stockfish][website-link] is a **free and strong UCI chess engine** derived from
+Glaurung 2.1 that analyzes chess positions and computes the optimal moves.
+
+Stockfish **does not include a graphical user interface** (GUI) that is required
+to display a chessboard and to make it easy to input moves. These GUIs are
+developed independently from Stockfish and are available online. **Read the
+documentation for your GUI** of choice for information about how to use
+Stockfish with it.
+
+See also the Stockfish [documentation][wiki-usage-link] for further usage help.
+
+## Files
+
+This distribution of Stockfish consists of the following files:
+
+  * [README.md][readme-link], the file you are currently reading.
+
+  * [Copying.txt][license-link], a text file containing the GNU General Public
+    License version 3.
+
+  * [AUTHORS][authors-link], a text file with the list of authors for the project.
+
+  * [src][src-link], 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.
+
+## The UCI protocol
+
+The [Universal Chess Interface][uci-link] (UCI) is a standard text-based 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 its options.
+
+Developers can see the default values for the UCI options available in Stockfish
+by typing `./stockfish uci` in a terminal, but most users should typically use a
+chess GUI to interact with Stockfish.
+
+For more information on UCI or debug commands, see our [documentation][wiki-commands-link].
+
+## Compiling Stockfish
+
+Stockfish has support for 32 or 64-bit CPUs, certain hardware instructions,
+big-endian machines such as Power PC, and other platforms.
+
+On Unix-like systems, it should be easy to compile Stockfish directly from the
+source code with the included Makefile in the folder `src`. In general, it is
+recommended to run `make help` to see a list of make targets with corresponding
+descriptions.
+
+```
+cd src
+make -j build ARCH=x86-64-modern
+```
+
+Detailed compilation instructions for all platforms can be found in our
+[documentation][wiki-compile-link].
+
+## Contributing
+
+### Donating hardware
+
+Improving Stockfish requires a massive amount of testing. You can donate your
+hardware resources by installing the [Fishtest Worker][worker-link] and viewing
+the current tests on [Fishtest][fishtest-link].
+
+### Improving the code
+
+In the [chessprogramming wiki][programming-link], many techniques used in
+Stockfish are explained with a lot of background information.
+The [section on Stockfish][programmingsf-link] describes many features
+and techniques used by Stockfish. However, it is generic rather than
+focused on Stockfish's precise implementation.
+
+The engine testing is done on [Fishtest][fishtest-link].
+If you want to help improve Stockfish, please read this [guideline][guideline-link]
+first, where the basics of Stockfish development are explained.
+
+Discussions about Stockfish take place these days mainly in the Stockfish
+[Discord server][discord-link]. This is also the best place to ask questions
+about the codebase and how to improve it.
+
+## Terms of use
+
+Stockfish is free and distributed under the
+[**GNU General Public License version 3**][license-link] (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 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 license and 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 GPL v3.
+
+
+[authors-link]:       https://github.com/official-stockfish/Stockfish/blob/master/AUTHORS
+[build-link]:         https://github.com/official-stockfish/Stockfish/actions/workflows/stockfish.yml
+[commits-link]:       https://github.com/official-stockfish/Stockfish/commits/master
+[discord-link]:       https://discord.gg/GWDRS3kU6R
+[issue-link]:         https://github.com/official-stockfish/Stockfish/issues/new?assignees=&labels=&template=BUG-REPORT.yml
+[discussions-link]:   https://github.com/official-stockfish/Stockfish/discussions/new
+[fishtest-link]:      https://tests.stockfishchess.org/tests
+[guideline-link]:     https://github.com/glinscott/fishtest/wiki/Creating-my-first-test
+[license-link]:       https://github.com/official-stockfish/Stockfish/blob/master/Copying.txt
+[programming-link]:   https://www.chessprogramming.org/Main_Page
+[programmingsf-link]: https://www.chessprogramming.org/Stockfish
+[readme-link]:        https://github.com/official-stockfish/Stockfish/blob/master/README.md
+[release-link]:       https://github.com/official-stockfish/Stockfish/releases/latest
+[src-link]:           https://github.com/official-stockfish/Stockfish/tree/master/src
+[stockfish128-logo]:  https://stockfishchess.org/images/logo/icon_128x128.png
+[uci-link]:           https://backscattering.de/chess/uci/
+[website-link]:       https://stockfishchess.org
+[website-blog-link]:  https://stockfishchess.org/blog/
+[wiki-link]:          https://github.com/official-stockfish/Stockfish/wiki
+[wiki-usage-link]:    https://github.com/official-stockfish/Stockfish/wiki/Download-and-usage
+[wiki-compile-link]:  https://github.com/official-stockfish/Stockfish/wiki/Compiling-from-source
+[wiki-commands-link]: https://github.com/official-stockfish/Stockfish/wiki/Commands
+[worker-link]:        https://github.com/glinscott/fishtest/wiki/Running-the-worker
+
+[build-badge]:        https://img.shields.io/github/actions/workflow/status/official-stockfish/Stockfish/stockfish.yml?branch=master&style=for-the-badge&label=stockfish&logo=github
+[commits-badge]:      https://img.shields.io/github/commits-since/official-stockfish/Stockfish/latest?style=for-the-badge
+[discord-badge]:      https://img.shields.io/discord/435943710472011776?style=for-the-badge&label=discord&logo=Discord
+[fishtest-badge]:     https://img.shields.io/website?style=for-the-badge&down_color=red&down_message=Offline&label=Fishtest&up_color=success&up_message=Online&url=https%3A%2F%2Ftests.stockfishchess.org%2Ftests%2Ffinished
+[license-badge]:      https://img.shields.io/github/license/official-stockfish/Stockfish?style=for-the-badge&label=license&color=success
+[release-badge]:      https://img.shields.io/github/v/release/official-stockfish/Stockfish?style=for-the-badge&label=official%20release
+[website-badge]:      https://img.shields.io/website?style=for-the-badge&down_color=red&down_message=Offline&label=website&up_color=success&up_message=Online&url=https%3A%2F%2Fstockfishchess.org

Readme.txt

@@ -1,81 +0,0 @@
-1. Introduction
----------------
-
-Stockfish is a free UCI chess engine derived from Glaurung 2.1. It is
-not a complete chess program and requires some UCI-compatible GUI
-(e.g. XBoard with PolyGlot, eboard, Arena, Sigma Chess, Shredder, Chess
-Partner or Fritz) in order to be used comfortably. Read the
-documentation for your GUI of choice for information about how to use
-Stockfish with it.
-
-This version of Stockfish supports up to 32 CPUs, but has not been
-tested thoroughly with more than 4.  The program tries to detect the
-number of CPUs on your computer and sets the number of search threads
-accordingly, but please be aware that the detection is not always
-correct. It is therefore recommended to inspect the value of the
-"Threads" UCI parameter, and to make sure it equals the number of CPU
-cores on your computer. If you are using more than eight threads, it is
-recommended to raise the value of the "Min Split Depth" UCI parameter to
-7.
-
-
-2. Files
---------
-
-This distribution of Stockfish consists of the following files:
-
-  * Readme.txt, the file you are currently reading.
-
-  * Copying.txt, a text file containing the GNU General Public License.
-
-  * src/, a subdirectory containing the full source code, including a
-    Makefile that can be used to compile Stockfish on Unix-like systems.
-    For further information about how to compile Stockfish yourself read
-    section 4 below.
-
-  * polyglot.ini, for using Stockfish with Fabien Letouzey's PolyGlot
-    adapter.
-
-
-3. Opening books
-----------------
-
-This version of Stockfish has support for PolyGlot opening books. For
-information about how to create such books, consult the PolyGlot
-documentation. The book file can be selected by setting the "Book File"
-UCI parameter.
-
-
-4. Compiling it yourself
-------------------------
-
-On Unix-like systems, it should be possible to compile Stockfish
-directly from the source code with the included Makefile.
-
-Stockfish has support for 32 or 64-bit CPUs, the hardware POPCNT
-instruction, big-endian machines such as Power PC, and other platforms.
-
-In general it is recommended to run 'make help' to see a list of make
-targets with corresponding descriptions. When not using 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.
-
-
-5. Terms of use
----------------
-
-Stockfish is free, and distributed under the GNU General Public License
-(GPL). Essentially, this means that 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
-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.
-
-For full details, read the copy of the GPL found in the file named
-Copying.txt.

Top CPU Contributors.txt

@@ -0,0 +1,267 @@
+Contributors to Fishtest with >10,000 CPU hours, as of 2023-06-20.
+Thank you!
+
+Username                                CPU Hours     Games played
+------------------------------------------------------------------
+noobpwnftw                               37457426       2850540907
+technologov                              14135647        742892808
+linrock                                   4423514        303254809
+mlang                                     3026000        200065824
+dew                                       1689162        100033738
+okrout                                    1578136        148855886
+pemo                                      1508508         48814305
+grandphish2                               1461406         91540343
+TueRens                                   1194790         70400852
+JojoM                                      947612         61773190
+tvijlbrief                                 796125         51897690
+sebastronomy                               742434         38218524
+mibere                                     703840         46867607
+gvreuls                                    651026         42988582
+oz                                         543438         39314736
+cw                                         517858         34869755
+fastgm                                     503862         30260818
+leszek                                     467278         33514883
+CSU_Dynasty                                464940         31177118
+ctoks                                      434416         28506889
+crunchy                                    427035         27344275
+maximmasiutin                              424795         26577722
+bcross                                     415722         29060963
+olafm                                      395922         32268020
+rpngn                                      348378         24560289
+velislav                                   342567         22138992
+Fisherman                                  327231         21829379
+mgrabiak                                   300612         20608380
+Dantist                                    296386         18031762
+nordlandia                                 246201         16189678
+robal                                      241300         15656382
+marrco                                     234581         17714473
+ncfish1                                    227517         15233777
+glinscott                                  208125         13277240
+drabel                                     204167         13930674
+mhoram                                     202894         12601997
+bking_US                                   198894         11876016
+Thanar                                     179852         12365359
+vdv                                        175544          9904472
+spams                                      157128         10319326
+sqrt2                                      147963          9724586
+DesolatedDodo                              146350          9536172
+Calis007                                   143165          9478764
+vdbergh                                    138650          9064413
+CoffeeOne                                  137100          5024116
+armo9494                                   136191          9460264
+malala                                     136182          8002293
+xoto                                       133759          9159372
+davar                                      129023          8376525
+DMBK                                       122960          8980062
+dsmith                                     122059          7570238
+amicic                                     119661          7938029
+Data                                       113305          8220352
+BrunoBanani                                112960          7436849
+CypressChess                               108331          7759788
+skiminki                                   107583          7218170
+jcAEie                                     105675          8238962
+MaZePallas                                 102823          6633619
+sterni1971                                 100532          5880772
+sunu                                       100167          7040199
+zeryl                                       99331          6221261
+thirdlife                                   99124          2242380
+ElbertoOne                                  99028          7023771
+cuistot                                     98853          6069816
+bigpen0r                                    94809          6529203
+brabos                                      92118          6186135
+Wolfgang                                    91939          6105872
+psk                                         89957          5984901
+sschnee                                     88235          5268000
+racerschmacer                               85805          6122790
+Fifis                                       85722          5709729
+Dubslow                                     84986          6042456
+Vizvezdenec                                 83761          5344740
+0x3C33                                      82614          5271253
+BRAVONE                                     81239          5054681
+nssy                                        76497          5259388
+jromang                                     76106          5236025
+teddybaer                                   75125          5407666
+tolkki963                                   74762          5149662
+megaman7de                                  74351          4940352
+Wencey                                      74181          4711488
+Pking_cda                                   73776          5293873
+yurikvelo                                   73150          5004382
+markkulix                                   72607          5304642
+Bobo1239                                    70579          4794999
+solarlight                                  70517          5028306
+dv8silencer                                 70287          3883992
+manap                                       66273          4121774
+tinker                                      64333          4268790
+qurashee                                    61208          3429862
+Mineta                                      59357          4418202
+Spprtr                                      58723          3911011
+AGI                                         58147          4325994
+robnjr                                      57262          4053117
+Freja                                       56938          3733019
+MaxKlaxxMiner                               56879          3423958
+MarcusTullius                               56746          3762951
+ttruscott                                   56010          3680085
+rkl                                         55132          4164467
+renouve                                     53811          3501516
+javran                                      53785          4627608
+finfish                                     51360          3370515
+eva42                                       51272          3599691
+eastorwest                                  51117          3454811
+rap                                         49985          3219146
+pb00067                                     49733          3298934
+OuaisBla                                    48626          3445134
+ronaldjerum                                 47654          3240695
+biffhero                                    46564          3111352
+VoyagerOne                                  45476          3452465
+jmdana                                      44893          3065205
+maposora                                    44597          4039578
+oryx                                        44570          3454238
+speedycpu                                   43842          3003273
+jbwiebe                                     43305          2805433
+GPUex                                       42378          3133332
+Antihistamine                               41788          2761312
+mhunt                                       41735          2691355
+homyur                                      39893          2850481
+gri                                         39871          2515779
+Garf                                        37741          2999686
+SC                                          37299          2731694
+csnodgrass                                  36207          2688994
+strelock                                    34716          2074055
+szupaw                                      34102          2880346
+EthanOConnor                                33370          2090311
+slakovv                                     32915          2021889
+Gelma                                       31771          1551204
+gopeto                                      31671          2060990
+kdave                                       31157          2198362
+manapbk                                     30987          1810399
+Prcuvu                                      30377          2170122
+anst                                        30301          2190091
+jkiiski                                     30136          1904470
+spcc                                        29925          1901692
+hyperbolic.tom                              29840          2017394
+chuckstablers                               29659          2093438
+Pyafue                                      29650          1902349
+belzedar94                                  28846          1811530
+chriswk                                     26902          1868317
+xwziegtm                                    26897          2124586
+achambord                                   26582          1767323
+Patrick_G                                   26276          1801617
+yorkman                                     26193          1992080
+Ulysses                                     25288          1689730
+SFTUser                                     25182          1675689
+nabildanial                                 24942          1519409
+Sharaf_DG                                   24765          1786697
+Maxim                                       24705          1502062
+rodneyc                                     24376          1416402
+agg177                                      23890          1395014
+Goatminola                                  23763          1956036
+Ente                                        23639          1671638
+Jopo12321                                   23467          1483172
+JanErik                                     23408          1703875
+Isidor                                      23388          1680691
+Norabor                                     23371          1603244
+cisco2015                                   22920          1763301
+jsys14                                      22824          1591906
+Zirie                                       22542          1472937
+team-oh                                     22272          1636708
+Roady                                       22220          1465606
+MazeOfGalious                               21978          1629593
+sg4032                                      21947          1643353
+ianh2105                                    21725          1632562
+xor12                                       21628          1680365
+dex                                         21612          1467203
+nesoneg                                     21494          1463031
+user213718                                  21454          1404128
+sphinx                                      21211          1384728
+AndreasKrug                                 21097          1634811
+jjoshua2                                    21001          1423089
+Zake9298                                    20938          1565848
+horst.prack                                 20878          1465656
+0xB00B1ES                                   20590          1208666
+j3corre                                     20405           941444
+Adrian.Schmidt123                           20316          1281436
+wei                                         19973          1745989
+notchris                                    19958          1800128
+Serpensin                                   19840          1697528
+Gaster319                                   19712          1677310
+fishtester                                  19617          1257388
+rstoesser                                   19569          1293588
+eudhan                                      19274          1283717
+votoanthuan                                 19108          1609992
+vulcan                                      18871          1729392
+Karpovbot                                   18766          1053178
+qoo_charly_cai                              18543          1284937
+jundery                                     18445          1115855
+ville                                       17883          1384026
+chris                                       17698          1487385
+purplefishies                               17595          1092533
+dju                                         17414           981289
+iisiraider                                  17275          1049015
+DragonLord                                  17014          1162790
+redstone59                                  16842          1461780
+Alb11747                                    16787          1213926
+IgorLeMasson                                16064          1147232
+Karby                                       15982           979610
+scuzzi                                      15757           968735
+ako027ako                                   15671          1173203
+Nikolay.IT                                  15154          1068349
+Andrew Grant                                15114           895539
+Naven94                                     15054           834762
+OssumOpossum                                14857          1007129
+ZacHFX                                      14783          1021842
+enedene                                     14476           905279
+bpfliegel                                   14233           882523
+mpx86                                       14019           759568
+jpulman                                     13982           870599
+Skiff84                                     13826           721996
+crocogoat                                   13803          1117422
+Nesa92                                      13786          1114691
+joster                                      13710           946160
+mbeier                                      13650          1044928
+Hjax                                        13535           915487
+Nullvalue                                   13468          1140498
+Dark_wizzie                                 13422          1007152
+Rudolphous                                  13244           883140
+pirt                                        13100          1009897
+Machariel                                   13010           863104
+infinigon                                   12991           943216
+mabichito                                   12903           749391
+thijsk                                      12886           722107
+AdrianSA                                    12860           804972
+Flopzee                                     12698           894821
+korposzczur                                 12606           838168
+fatmurphy                                   12547           853210
+SapphireBrand                               12416           969604
+Oakwen                                      12399           844109
+deflectooor                                 12386           579392
+modolief                                    12386           896470
+Farseer                                     12249           694108
+Jackfish                                    12213           805008
+pgontarz                                    12151           848794
+dbernier                                    12103           860824
+getraideBFF                                 12072          1024966
+stocky                                      11954           699440
+mschmidt                                    11941           803401
+MooTheCow                                   11870           773598
+FormazChar                                  11766           885707
+whelanh                                     11557           245188
+3cho                                        11494          1031076
+infinity                                    11470           727027
+aga                                         11412           695127
+torbjo                                      11395           729145
+Thomas A. Anderson                          11372           732094
+savage84                                    11358           670860
+d64                                         11263           789184
+ali-al-zhrani                               11245           779246
+snicolet                                    11106           869170
+dapper                                      11032           771402
+ols                                         10947           624903
+Karmatron                                   10828           677458
+basepi                                      10637           744851
+Cubox                                       10621           826448
+michaelrpg                                  10509           739239
+OIVAS7572                                   10420           995586
+jojo2357                                    10419           929708
+WoodMan777                                  10380           873720
+Garruk                                      10365           706465
+dzjp                                        10343           732529

polyglot.ini

@@ -1,43 +0,0 @@
-
-[PolyGlot]
-
-EngineDir = .
-EngineCommand = ./stockfish
-
-Book = false
-BookFile = book.bin
-
-Log = false
-LogFile = stockfish.log
-
-Resign = true
-ResignScore = 600
-
-[Engine]
-
-Use Search Log = false
-Search Log Filename = SearchLog.txt
-Book File = book.bin
-Best Book Move = false
-Mobility (Middle Game) = 100
-Mobility (Endgame) = 100
-Passed Pawns (Middle Game) = 100
-Passed Pawns (Endgame) = 100
-Space = 100
-Aggressiveness = 100
-Cowardice = 100
-Min Split Depth = 4
-Max Threads per Split Point = 5
-Threads = 1
-Use Sleeping Threads = true
-Hash = 128
-Ponder = true
-OwnBook = false
-MultiPV = 1
-Skill Level = 20
-Emergency Move Horizon = 40
-Emergency Base Time = 200
-Emergency Move Time = 70
-Minimum Thinking Time = 20
-UCI_Chess960 = false
-UCI_AnalyseMode = false

src/benchmark.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,27 +16,27 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#include "benchmark.h"
+
 #include <fstream>
 #include <iostream>
 #include <istream>
 #include <vector>
 
-#include "misc.h"
 #include "position.h"
-#include "search.h"
-#include "thread.h"
-#include "tt.h"
-#include "ucioption.h"
 
 using namespace std;
 
-static const char* Defaults[] = {
+namespace {
+
+const vector<string> Defaults = {
+  "setoption name UCI_Chess960 value false",
   "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1",
   "r3k2r/p1ppqpb1/bn2pnp1/3PN3/1p2P3/2N2Q1p/PPPBBPPP/R3K2R w KQkq - 0 10",
   "8/2p5/3p4/KP5r/1R3p1k/8/4P1P1/8 w - - 0 11",
   "4rrk1/pp1n3p/3q2pQ/2p1pb2/2PP4/2P3N1/P2B2PP/4RRK1 b - - 7 19",
-  "rq3rk1/ppp2ppp/1bnpb3/3N2B1/3NP3/7P/PPPQ1PP1/2KR3R w - - 7 14",
-  "r1bq1r1k/1pp1n1pp/1p1p4/4p2Q/4Pp2/1BNP4/PPP2PPP/3R1RK1 w - - 2 14",
+  "rq3rk1/ppp2ppp/1bnpb3/3N2B1/3NP3/7P/PPPQ1PP1/2KR3R w - - 7 14 moves d4e6",
+  "r1bq1r1k/1pp1n1pp/1p1p4/4p2Q/4Pp2/1BNP4/PPP2PPP/3R1RK1 w - - 2 14 moves g2g4",
   "r3r1k1/2p2ppp/p1p1bn2/8/1q2P3/2NPQN2/PPP3PP/R4RK1 b - - 2 15",
   "r1bbk1nr/pp3p1p/2n5/1N4p1/2Np1B2/8/PPP2PPP/2KR1B1R w kq - 0 13",
   "r1bq1rk1/ppp1nppp/4n3/3p3Q/3P4/1BP1B3/PP1N2PP/R4RK1 w - - 1 16",
@@ -47,54 +46,95 @@ static const char* Defaults[] = {
   "3r1rk1/p5pp/bpp1pp2/8/q1PP1P2/b3P3/P2NQRPP/1R2B1K1 b - - 6 22",
   "r1q2rk1/2p1bppp/2Pp4/p6b/Q1PNp3/4B3/PP1R1PPP/2K4R w - - 2 18",
   "4k2r/1pb2ppp/1p2p3/1R1p4/3P4/2r1PN2/P4PPP/1R4K1 b - - 3 22",
-  "3q2k1/pb3p1p/4pbp1/2r5/PpN2N2/1P2P2P/5PP1/Q2R2K1 b - - 4 26"
+  "3q2k1/pb3p1p/4pbp1/2r5/PpN2N2/1P2P2P/5PP1/Q2R2K1 b - - 4 26",
+  "6k1/6p1/6Pp/ppp5/3pn2P/1P3K2/1PP2P2/3N4 b - - 0 1",
+  "3b4/5kp1/1p1p1p1p/pP1PpP1P/P1P1P3/3KN3/8/8 w - - 0 1",
+  "2K5/p7/7P/5pR1/8/5k2/r7/8 w - - 0 1 moves g5g6 f3e3 g6g5 e3f3",
+  "8/6pk/1p6/8/PP3p1p/5P2/4KP1q/3Q4 w - - 0 1",
+  "7k/3p2pp/4q3/8/4Q3/5Kp1/P6b/8 w - - 0 1",
+  "8/2p5/8/2kPKp1p/2p4P/2P5/3P4/8 w - - 0 1",
+  "8/1p3pp1/7p/5P1P/2k3P1/8/2K2P2/8 w - - 0 1",
+  "8/pp2r1k1/2p1p3/3pP2p/1P1P1P1P/P5KR/8/8 w - - 0 1",
+  "8/3p4/p1bk3p/Pp6/1Kp1PpPp/2P2P1P/2P5/5B2 b - - 0 1",
+  "5k2/7R/4P2p/5K2/p1r2P1p/8/8/8 b - - 0 1",
+  "6k1/6p1/P6p/r1N5/5p2/7P/1b3PP1/4R1K1 w - - 0 1",
+  "1r3k2/4q3/2Pp3b/3Bp3/2Q2p2/1p1P2P1/1P2KP2/3N4 w - - 0 1",
+  "6k1/4pp1p/3p2p1/P1pPb3/R7/1r2P1PP/3B1P2/6K1 w - - 0 1",
+  "8/3p3B/5p2/5P2/p7/PP5b/k7/6K1 w - - 0 1",
+  "5rk1/q6p/2p3bR/1pPp1rP1/1P1Pp3/P3B1Q1/1K3P2/R7 w - - 93 90",
+  "4rrk1/1p1nq3/p7/2p1P1pp/3P2bp/3Q1Bn1/PPPB4/1K2R1NR w - - 40 21",
+  "r3k2r/3nnpbp/q2pp1p1/p7/Pp1PPPP1/4BNN1/1P5P/R2Q1RK1 w kq - 0 16",
+  "3Qb1k1/1r2ppb1/pN1n2q1/Pp1Pp1Pr/4P2p/4BP2/4B1R1/1R5K b - - 11 40",
+  "4k3/3q1r2/1N2r1b1/3ppN2/2nPP3/1B1R2n1/2R1Q3/3K4 w - - 5 1",
+
+  // 5-man positions
+  "8/8/8/8/5kp1/P7/8/1K1N4 w - - 0 1",     // Kc2 - mate
+  "8/8/8/5N2/8/p7/8/2NK3k w - - 0 1",      // Na2 - mate
+  "8/3k4/8/8/8/4B3/4KB2/2B5 w - - 0 1",    // draw
+
+  // 6-man positions
+  "8/8/1P6/5pr1/8/4R3/7k/2K5 w - - 0 1",   // Re5 - mate
+  "8/2p4P/8/kr6/6R1/8/8/1K6 w - - 0 1",    // Ka2 - mate
+  "8/8/3P3k/8/1p6/8/1P6/1K3n2 b - - 0 1",  // Nd2 - draw
+
+  // 7-man positions
+  "8/R7/2q5/8/6k1/8/1P5p/K6R w - - 0 124", // Draw
+
+  // Mate and stalemate positions
+  "6k1/3b3r/1p1p4/p1n2p2/1PPNpP1q/P3Q1p1/1R1RB1P1/5K2 b - - 0 1",
+  "r2r1n2/pp2bk2/2p1p2p/3q4/3PN1QP/2P3R1/P4PP1/5RK1 w - - 0 1",
+  "8/8/8/8/8/6k1/6p1/6K1 w - -",
+  "7k/7P/6K1/8/3B4/8/8/8 b - -",
+
+  // Chess 960
+  "setoption name UCI_Chess960 value true",
+  "bbqnnrkr/pppppppp/8/8/8/8/PPPPPPPP/BBQNNRKR w HFhf - 0 1 moves g2g3 d7d5 d2d4 c8h3 c1g5 e8d6 g5e7 f7f6",
+  "nqbnrkrb/pppppppp/8/8/8/8/PPPPPPPP/NQBNRKRB w KQkq - 0 1",
+  "setoption name UCI_Chess960 value false"
 };
 
+} // namespace
 
-/// benchmark() runs a simple benchmark by letting Stockfish analyze a set
-/// of positions for a given limit each. There are five parameters; the
-/// transposition table size, the number of search threads that should
-/// be used, the limit value spent for each position (optional, default is
-/// depth 12), an optional file name where to look for positions in fen
-/// format (defaults are the positions defined above) and the type of the
-/// limit value: depth (default), time in secs or number of nodes.
+namespace Stockfish {
 
-void benchmark(const Position& current, istream& is) {
+/// setup_bench() builds a list of UCI commands to be run by bench. There
+/// are five parameters: TT size in MB, number of search threads that
+/// should be used, the limit value spent for each position, a file name
+/// where to look for positions in FEN format, the type of the limit:
+/// depth, perft, nodes and movetime (in millisecs), and evaluation type
+/// mixed (default), classical, NNUE.
+///
+/// bench -> search default positions up to depth 13
+/// bench 64 1 15 -> search default positions up to depth 15 (TT = 64MB)
+/// bench 64 4 5000 current movetime -> search current position with 4 threads for 5 sec
+/// bench 64 1 100000 default nodes -> search default positions for 100K nodes each
+/// bench 16 1 5 default perft -> run a perft 5 on default positions
 
-  string token;
-  Search::LimitsType limits;
-  vector<string> fens;
+vector<string> setup_bench(const Position& current, istream& is) {
+
+  vector<string> fens, list;
+  string go, token;
 
   // Assign default values to missing arguments
-  string ttSize    = (is >> token) ? token : "128";
+  string ttSize    = (is >> token) ? token : "16";
   string threads   = (is >> token) ? token : "1";
-  string limit     = (is >> token) ? token : "12";
+  string limit     = (is >> token) ? token : "13";
   string fenFile   = (is >> token) ? token : "default";
   string limitType = (is >> token) ? token : "depth";
+  string evalType  = (is >> token) ? token : "mixed";
 
-  Options["Hash"]    = ttSize;
-  Options["Threads"] = threads;
-  TT.clear();
-
-  if (limitType == "time")
-      limits.movetime = 1000 * atoi(limit.c_str()); // movetime is in ms
-
-  else if (limitType == "nodes")
-      limits.nodes = atoi(limit.c_str());
-
-  else
-      limits.depth = atoi(limit.c_str());
+  go = limitType == "eval" ? "eval" : "go " + limitType + " " + limit;
 
   if (fenFile == "default")
-      fens.assign(Defaults, Defaults + 16);
+      fens = Defaults;
 
   else if (fenFile == "current")
-      fens.push_back(current.to_fen());
+      fens.push_back(current.fen());
 
   else
   {
       string fen;
-      ifstream file(fenFile.c_str());
+      ifstream file(fenFile);
 
       if (!file.is_open())
       {
@@ -109,34 +149,29 @@ void benchmark(const Position& current, istream& is) {
       file.close();
   }
 
-  int64_t nodes = 0;
-  Search::StateStackPtr st;
-  Time::point elapsed = Time::now();
+  list.emplace_back("setoption name Threads value " + threads);
+  list.emplace_back("setoption name Hash value " + ttSize);
+  list.emplace_back("ucinewgame");
 
-  for (size_t i = 0; i < fens.size(); i++)
-  {
-      Position pos(fens[i], Options["UCI_Chess960"], Threads.main_thread());
+  size_t posCounter = 0;
 
-      cerr << "\nPosition: " << i + 1 << '/' << fens.size() << endl;
-
-      if (limitType == "perft")
-      {
-          size_t cnt = Search::perft(pos, limits.depth * ONE_PLY);
-          cerr << "\nPerft " << limits.depth  << " leaf nodes: " << cnt << endl;
-          nodes += cnt;
-      }
+  for (const string& fen : fens)
+      if (fen.find("setoption") != string::npos)
+          list.emplace_back(fen);
       else
       {
-          Threads.start_searching(pos, limits, vector<Move>(), st);
-          Threads.wait_for_search_finished();
-          nodes += Search::RootPosition.nodes_searched();
+          if (evalType == "classical" || (evalType == "mixed" && posCounter % 2 == 0))
+              list.emplace_back("setoption name Use NNUE value false");
+          else if (evalType == "NNUE" || (evalType == "mixed" && posCounter % 2 != 0))
+              list.emplace_back("setoption name Use NNUE value true");
+          list.emplace_back("position fen " + fen);
+          list.emplace_back(go);
+          ++posCounter;
       }
-  }
 
-  elapsed = Time::now() - elapsed + 1; // Assure positive to avoid a 'divide by zero'
+  list.emplace_back("setoption name Use NNUE value true");
 
-  cerr << "\n==========================="
-       << "\nTotal time (ms) : " << elapsed
-       << "\nNodes searched  : " << nodes
-       << "\nNodes/second    : " << 1000 * nodes / elapsed << endl;
+  return list;
 }
+
+} // namespace Stockfish

src/benchmark.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,19 +16,19 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(NOTATION_H_INCLUDED)
-#define NOTATION_H_INCLUDED
+#ifndef BENCHMARK_H_INCLUDED
+#define BENCHMARK_H_INCLUDED
 
+#include <iosfwd>
 #include <string>
+#include <vector>
 
-#include "types.h"
+namespace Stockfish {
 
 class Position;
 
-std::string score_to_uci(Value v, Value alpha = -VALUE_INFINITE, Value beta = VALUE_INFINITE);
-Move move_from_uci(const Position& pos, std::string& str);
-const std::string move_to_uci(Move m, bool chess960);
-const std::string move_to_san(Position& pos, Move m);
-std::string pretty_pv(Position& pos, int depth, Value score, int64_t msecs, Move pv[]);
+std::vector<std::string> setup_bench(const Position&, std::istream&);
 
-#endif // !defined(NOTATION_H_INCLUDED)
+} // namespace Stockfish
+
+#endif // #ifndef BENCHMARK_H_INCLUDED

src/bitbase.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -18,12 +17,35 @@
 */
 
 #include <cassert>
+#include <vector>
+#include <bitset>
 
 #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.
+  // Positions with the pawn on files E to H will be mirrored before probing.
+  constexpr unsigned MAX_INDEX = 2*24*64*64; // stm * psq * wksq * bksq = 196608
+
+  std::bitset<MAX_INDEX> KPKBitbase;
+
+  // A KPK bitbase index is an integer in [0, IndexMax] range
+  //
+  // Information is mapped in a way that minimizes the number of iterations:
+  //
+  // bit  0- 5: white king square (from SQ_A1 to SQ_H8)
+  // bit  6-11: black king square (from SQ_A1 to SQ_H8)
+  // bit    12: side to move (WHITE or BLACK)
+  // bit 13-14: white pawn file (from FILE_A to FILE_D)
+  // bit 15-17: white pawn RANK_7 - rank (from RANK_7 - RANK_7 to RANK_7 - RANK_2)
+  unsigned index(Color stm, Square bksq, Square wksq, Square psq) {
+    return int(wksq) | (bksq << 6) | (stm << 12) | (file_of(psq) << 13) | ((RANK_7 - rank_of(psq)) << 15);
+  }
+
   enum Result {
     INVALID = 0,
     UNKNOWN = 1,
@@ -31,200 +53,120 @@ namespace {
     WIN     = 4
   };
 
-  inline Result& operator|=(Result& r, Result v) { return r = Result(r | v); }
+  Result& operator|=(Result& r, Result v) { return r = Result(r | v); }
 
   struct KPKPosition {
+    KPKPosition() = default;
+    explicit KPKPosition(unsigned idx);
+    operator Result() const { return result; }
+    Result classify(const std::vector<KPKPosition>& db);
 
-    Result classify_leaf(int idx);
-    Result classify(int idx, Result db[]);
-
-  private:
-    template<Color Us> Result classify(const Result db[]) const;
-
-    template<Color Us> Bitboard k_attacks() const {
-      return Us == WHITE ? StepAttacksBB[W_KING][wksq] : StepAttacksBB[B_KING][bksq];
-    }
-
-    Bitboard p_attacks() const { return StepAttacksBB[W_PAWN][psq]; }
-    void decode_index(int idx);
-
-    Square wksq, bksq, psq;
     Color stm;
+    Square ksq[COLOR_NB], psq;
+    Result result;
   };
 
-  // The possible pawns squares are 24, the first 4 files and ranks from 2 to 7
-  const int IndexMax = 2 * 24 * 64 * 64; // stm * wp_sq * wk_sq * bk_sq = 196608
+} // namespace
 
-  // Each uint32_t stores results of 32 positions, one per bit
-  uint32_t KPKBitbase[IndexMax / 32];
+bool Bitbases::probe(Square wksq, Square wpsq, Square bksq, Color stm) {
 
-  int index(Square wksq, Square bksq, Square psq, Color stm);
+  assert(file_of(wpsq) <= FILE_D);
+
+  return KPKBitbase[index(stm, bksq, wksq, wpsq)];
 }
 
 
-uint32_t Bitbases::probe_kpk(Square wksq, Square wpsq, Square bksq, Color stm) {
+void Bitbases::init() {
 
-  int idx = index(wksq, bksq, wpsq, stm);
-  return KPKBitbase[idx / 32] & (1 << (idx & 31));
-}
+  std::vector<KPKPosition> db(MAX_INDEX);
+  unsigned idx, repeat = 1;
 
+  // Initialize db with known win / draw positions
+  for (idx = 0; idx < MAX_INDEX; ++idx)
+      db[idx] = KPKPosition(idx);
 
-void Bitbases::init_kpk() {
-
-  Result db[IndexMax];
-  KPKPosition pos;
-  int idx, bit, repeat = 1;
-
-  // Initialize table with known win / draw positions
-  for (idx = 0; idx < IndexMax; idx++)
-      db[idx] = pos.classify_leaf(idx);
-
-  // Iterate until all positions are classified (30 cycles needed)
+  // Iterate through the positions until none of the unknown positions can be
+  // changed to either wins or draws (15 cycles needed).
   while (repeat)
-      for (repeat = idx = 0; idx < IndexMax; idx++)
-          if (db[idx] == UNKNOWN && (db[idx] = pos.classify(idx, db)) != UNKNOWN)
-              repeat = 1;
+      for (repeat = idx = 0; idx < MAX_INDEX; ++idx)
+          repeat |= (db[idx] == UNKNOWN && db[idx].classify(db) != UNKNOWN);
 
-  // Map 32 position results into one KPKBitbase[] entry
-  for (idx = 0; idx < IndexMax / 32; idx++)
-      for (bit = 0; bit < 32; bit++)
-          if (db[32 * idx + bit] == WIN)
-              KPKBitbase[idx] |= 1 << bit;
+  // Fill the bitbase with the decisive results
+  for (idx = 0; idx < MAX_INDEX; ++idx)
+      if (db[idx] == WIN)
+          KPKBitbase.set(idx);
 }
 
-
 namespace {
 
-  // A KPK bitbase index is an integer in [0, IndexMax] range
-  //
-  // Information is mapped in this way
-  //
-  // bit     0: side to move (WHITE or BLACK)
-  // bit  1- 6: black king square (from SQ_A1 to SQ_H8)
-  // bit  7-12: white king square (from SQ_A1 to SQ_H8)
-  // bit 13-14: white pawn file (from FILE_A to FILE_D)
-  // bit 15-17: white pawn rank - 1 (from RANK_2 - 1 to RANK_7 - 1)
+  KPKPosition::KPKPosition(unsigned idx) {
 
-  int index(Square w, Square b, Square p, Color c) {
+    ksq[WHITE] = Square((idx >>  0) & 0x3F);
+    ksq[BLACK] = Square((idx >>  6) & 0x3F);
+    stm        = Color ((idx >> 12) & 0x01);
+    psq        = make_square(File((idx >> 13) & 0x3), Rank(RANK_7 - ((idx >> 15) & 0x7)));
 
-    assert(file_of(p) <= FILE_D);
+    // Invalid if two pieces are on the same square or if a king can be captured
+    if (   distance(ksq[WHITE], ksq[BLACK]) <= 1
+        || ksq[WHITE] == psq
+        || ksq[BLACK] == psq
+        || (stm == WHITE && (pawn_attacks_bb(WHITE, psq) & ksq[BLACK])))
+        result = INVALID;
 
-    return c + (b << 1) + (w << 7) + (file_of(p) << 13) + ((rank_of(p) - 1) << 15);
+    // Win if the pawn can be promoted without getting captured
+    else if (   stm == WHITE
+             && rank_of(psq) == RANK_7
+             && ksq[WHITE] != psq + NORTH
+             && (    distance(ksq[BLACK], psq + NORTH) > 1
+                 || (distance(ksq[WHITE], psq + NORTH) == 1)))
+        result = WIN;
+
+    // Draw if it is stalemate or the black king can capture the pawn
+    else if (   stm == BLACK
+             && (  !(attacks_bb<KING>(ksq[BLACK]) & ~(attacks_bb<KING>(ksq[WHITE]) | pawn_attacks_bb(WHITE, psq)))
+                 || (attacks_bb<KING>(ksq[BLACK]) & ~attacks_bb<KING>(ksq[WHITE]) & psq)))
+        result = DRAW;
+
+    // Position will be classified later
+    else
+        result = UNKNOWN;
   }
 
-  void KPKPosition::decode_index(int idx) {
+  Result KPKPosition::classify(const std::vector<KPKPosition>& db) {
 
-    stm  = Color(idx & 1);
-    bksq = Square((idx >> 1) & 63);
-    wksq = Square((idx >> 7) & 63);
-    psq  = File((idx >> 13) & 3) | Rank((idx >> 15) + 1);
-  }
-
-  Result KPKPosition::classify_leaf(int idx) {
-
-    decode_index(idx);
-
-    // Check if two pieces are on the same square or if a king can be captured
-    if (   wksq == psq || wksq == bksq || bksq == psq
-        || (k_attacks<WHITE>() & bksq)
-        || (stm == WHITE && (p_attacks() & bksq)))
-        return INVALID;
-
-    // The position is an immediate win if it is white to move and the white
-    // pawn can be promoted without getting captured.
-    if (   rank_of(psq) == RANK_7
-        && stm == WHITE
-        && wksq != psq + DELTA_N
-        && (   square_distance(bksq, psq + DELTA_N) > 1
-            ||(k_attacks<WHITE>() & (psq + DELTA_N))))
-        return WIN;
-
-    // Check for known draw positions
+    // White to move: If one move leads to a position classified as WIN, the result
+    // of the current position is WIN. If all moves lead to positions classified
+    // as DRAW, the current position is classified as DRAW, otherwise the current
+    // position is classified as UNKNOWN.
     //
-    // Case 1: Stalemate
-    if (   stm == BLACK
-        && !(k_attacks<BLACK>() & ~(k_attacks<WHITE>() | p_attacks())))
-        return DRAW;
-
-    // Case 2: King can capture undefended pawn
-    if (   stm == BLACK
-        && (k_attacks<BLACK>() & psq & ~k_attacks<WHITE>()))
-        return DRAW;
-
-    // Case 3: Black king in front of white pawn
-    if (   bksq == psq + DELTA_N
-        && rank_of(psq) < RANK_7)
-        return DRAW;
-
-    // Case 4: White king in front of pawn and black has opposition
-    if (   stm == WHITE
-        && wksq == psq + DELTA_N
-        && bksq == wksq + DELTA_N + DELTA_N
-        && rank_of(psq) < RANK_5)
-        return DRAW;
-
-    // Case 5: Stalemate with rook pawn
-    if (   bksq == SQ_A8
-        && file_of(psq) == FILE_A)
-        return DRAW;
-
-    // Case 6: White king trapped on the rook file
-    if (   file_of(wksq) == FILE_A
-        && file_of(psq) == FILE_A
-        && rank_of(wksq) > rank_of(psq)
-        && bksq == wksq + 2)
-        return DRAW;
-
-    return UNKNOWN;
-  }
-
-  template<Color Us>
-  Result KPKPosition::classify(const Result db[]) const {
-
-    // White to Move: If one move leads to a position classified as RESULT_WIN,
-    // the result of the current position is RESULT_WIN. If all moves lead to
-    // positions classified as RESULT_DRAW, the current position is classified
-    // RESULT_DRAW otherwise the current position is classified as RESULT_UNKNOWN.
-    //
-    // Black to Move: If one move leads to a position classified as RESULT_DRAW,
-    // the result of the current position is RESULT_DRAW. If all moves lead to
-    // positions classified as RESULT_WIN, the position is classified RESULT_WIN.
-    // Otherwise, the current position is classified as RESULT_UNKNOWN.
+    // Black to move: If one move leads to a position classified as DRAW, the result
+    // of the current position is DRAW. If all moves lead to positions classified
+    // as WIN, the position is classified as WIN, otherwise the current position is
+    // classified as UNKNOWN.
+    const Result Good = (stm == WHITE ? WIN   : DRAW);
+    const Result Bad  = (stm == WHITE ? DRAW  : WIN);
 
     Result r = INVALID;
-    Bitboard b = k_attacks<Us>();
+    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)];
+
+    if (stm == WHITE)
     {
-        r |= Us == WHITE ? db[index(pop_lsb(&b), bksq, psq, BLACK)]
-                         : db[index(wksq, pop_lsb(&b), psq, WHITE)];
+        if (rank_of(psq) < RANK_7)      // Single push
+            r |= db[index(BLACK, ksq[BLACK], ksq[WHITE], psq + NORTH)];
 
-        if (Us == WHITE && (r & WIN))
-            return WIN;
-
-        if (Us == BLACK && (r & DRAW))
-            return DRAW;
+        if (   rank_of(psq) == RANK_2   // Double push
+            && psq + NORTH != ksq[WHITE]
+            && psq + NORTH != ksq[BLACK])
+            r |= db[index(BLACK, ksq[BLACK], ksq[WHITE], psq + NORTH + NORTH)];
     }
 
-    if (Us == WHITE && rank_of(psq) < RANK_7)
-    {
-        Square s = psq + DELTA_N;
-        r |= db[index(wksq, bksq, s, BLACK)]; // Single push
-
-        if (rank_of(s) == RANK_3 && s != wksq && s != bksq)
-            r |= db[index(wksq, bksq, s + DELTA_N, BLACK)]; // Double push
-
-        if (r & WIN)
-            return WIN;
-    }
-
-    return r & UNKNOWN ? UNKNOWN : Us == WHITE ? DRAW : WIN;
+    return result = r & Good  ? Good  : r & UNKNOWN ? UNKNOWN : Bad;
   }
 
-  Result KPKPosition::classify(int idx, Result db[]) {
+} // namespace
 
-    decode_index(idx);
-    return stm == WHITE ? classify<WHITE>(db) : classify<BLACK>(db);
-  }
-
-}
+} // namespace Stockfish

src/bitboard.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -18,282 +17,138 @@
 */
 
 #include <algorithm>
-#include <cstring>
-#include <iostream>
+#include <bitset>
 
 #include "bitboard.h"
-#include "bitcount.h"
 #include "misc.h"
-#include "rkiss.h"
 
-CACHE_LINE_ALIGNMENT
+namespace Stockfish {
 
-Bitboard RMasks[64];
-Bitboard RMagics[64];
-Bitboard* RAttacks[64];
-unsigned RShifts[64];
+uint8_t PopCnt16[1 << 16];
+uint8_t SquareDistance[SQUARE_NB][SQUARE_NB];
 
-Bitboard BMasks[64];
-Bitboard BMagics[64];
-Bitboard* BAttacks[64];
-unsigned BShifts[64];
+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];
 
-Bitboard SquareBB[64];
-Bitboard FileBB[8];
-Bitboard RankBB[8];
-Bitboard AdjacentFilesBB[8];
-Bitboard ThisAndAdjacentFilesBB[8];
-Bitboard InFrontBB[2][8];
-Bitboard StepAttacksBB[16][64];
-Bitboard BetweenBB[64][64];
-Bitboard DistanceRingsBB[64][8];
-Bitboard ForwardBB[2][64];
-Bitboard PassedPawnMask[2][64];
-Bitboard AttackSpanMask[2][64];
-Bitboard PseudoAttacks[6][64];
-
-int SquareDistance[64][64];
+Magic RookMagics[SQUARE_NB];
+Magic BishopMagics[SQUARE_NB];
 
 namespace {
 
-  // De Bruijn sequences. See chessprogramming.wikispaces.com/BitScan
-  const uint64_t DeBruijn_64 = 0x218A392CD3D5DBFULL;
-  const uint32_t DeBruijn_32 = 0x783A9B23;
+  Bitboard RookTable[0x19000];  // To store rook attacks
+  Bitboard BishopTable[0x1480]; // To store bishop attacks
 
-  CACHE_LINE_ALIGNMENT
+  void init_magics(PieceType pt, Bitboard table[], Magic magics[]);
 
-  int MS1BTable[256];
-  Square BSFTable[64];
-  Bitboard RTable[0x19000]; // Storage space for rook attacks
-  Bitboard BTable[0x1480];  // Storage space for bishop attacks
-  uint8_t BitCount8Bit[256];
-
-  typedef unsigned (Fn)(Square, Bitboard);
-
-  void init_magics(Bitboard table[], Bitboard* attacks[], Bitboard magics[],
-                   Bitboard masks[], unsigned shifts[], Square deltas[], Fn index);
-
-  FORCE_INLINE unsigned bsf_index(Bitboard b) {
-
-    if (Is64Bit)
-        return ((b & -b) * DeBruijn_64) >> 58;
-
-    // Use Matt Taylor's folding trick for 32 bit systems
-    b ^= (b - 1);
-    return ((unsigned(b) ^ unsigned(b >> 32)) * DeBruijn_32) >> 26;
-  }
 }
 
-/// lsb()/msb() finds the least/most significant bit in a nonzero bitboard.
-/// pop_lsb() finds and clears the least significant bit in a nonzero bitboard.
+/// 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.
 
-#if !defined(USE_BSFQ)
-
-Square lsb(Bitboard b) { return BSFTable[bsf_index(b)]; }
-
-Square pop_lsb(Bitboard* b) {
-
-  Bitboard bb = *b;
-  *b = bb & (bb - 1);
-  return BSFTable[bsf_index(bb)];
-}
-
-Square msb(Bitboard b) {
-
-  unsigned b32;
-  int result = 0;
-
-  if (b > 0xFFFFFFFF)
-  {
-      b >>= 32;
-      result = 32;
-  }
-
-  b32 = unsigned(b);
-
-  if (b32 > 0xFFFF)
-  {
-      b32 >>= 16;
-      result += 16;
-  }
-
-  if (b32 > 0xFF)
-  {
-      b32 >>= 8;
-      result += 8;
-  }
-
-  return (Square)(result + MS1BTable[b32]);
-}
-
-#endif // !defined(USE_BSFQ)
-
-
-/// Bitboards::print() prints a bitboard in an easily readable format to the
-/// standard output. This is sometimes useful for debugging.
-
-void Bitboards::print(Bitboard b) {
-
-  sync_cout;
-
-  for (Rank rank = RANK_8; rank >= RANK_1; rank--)
-  {
-      std::cout << "+---+---+---+---+---+---+---+---+" << '\n';
-
-      for (File file = FILE_A; file <= FILE_H; file++)
-          std::cout << "| " << (b & (file | rank) ? "X " : "  ");
-
-      std::cout << "|\n";
-  }
-  std::cout << "+---+---+---+---+---+---+---+---+" << sync_endl;
+inline Bitboard safe_destination(Square s, int step) {
+    Square to = Square(s + step);
+    return is_ok(to) && distance(s, to) <= 2 ? square_bb(to) : Bitboard(0);
 }
 
 
-/// Bitboards::init() initializes various bitboard arrays. It is called during
-/// program initialization.
+/// Bitboards::pretty() returns an ASCII representation of a bitboard suitable
+/// to be printed to standard output. Useful for debugging.
+
+std::string Bitboards::pretty(Bitboard b) {
+
+  std::string s = "+---+---+---+---+---+---+---+---+\n";
+
+  for (Rank r = RANK_8; r >= RANK_1; --r)
+  {
+      for (File f = FILE_A; f <= FILE_H; ++f)
+          s += b & make_square(f, r) ? "| X " : "|   ";
+
+      s += "| " + std::to_string(1 + r) + "\n+---+---+---+---+---+---+---+---+\n";
+  }
+  s += "  a   b   c   d   e   f   g   h\n";
+
+  return s;
+}
+
+
+/// Bitboards::init() initializes various bitboard tables. It is called at
+/// startup and relies on global objects to be already zero-initialized.
 
 void Bitboards::init() {
 
-  for (int k = 0, i = 0; i < 8; i++)
-      while (k < (2 << i))
-          MS1BTable[k++] = i;
+  for (unsigned i = 0; i < (1 << 16); ++i)
+      PopCnt16[i] = uint8_t(std::bitset<16>(i).count());
 
-  for (int i = 0; i < 64; i++)
-      BSFTable[bsf_index(1ULL << i)] = Square(i);
+  for (Square s1 = SQ_A1; s1 <= SQ_H8; ++s1)
+      for (Square s2 = SQ_A1; s2 <= SQ_H8; ++s2)
+          SquareDistance[s1][s2] = std::max(distance<File>(s1, s2), distance<Rank>(s1, s2));
 
-  for (Bitboard b = 0; b < 256; b++)
-      BitCount8Bit[b] = (uint8_t)popcount<Max15>(b);
+  init_magics(ROOK, RookTable, RookMagics);
+  init_magics(BISHOP, BishopTable, BishopMagics);
 
-  for (Square s = SQ_A1; s <= SQ_H8; s++)
-      SquareBB[s] = 1ULL << s;
-
-  FileBB[FILE_A] = FileABB;
-  RankBB[RANK_1] = Rank1BB;
-
-  for (int i = 1; i < 8; i++)
+  for (Square s1 = SQ_A1; s1 <= SQ_H8; ++s1)
   {
-      FileBB[i] = FileBB[i - 1] << 1;
-      RankBB[i] = RankBB[i - 1] << 8;
-  }
+      PawnAttacks[WHITE][s1] = pawn_attacks_bb<WHITE>(square_bb(s1));
+      PawnAttacks[BLACK][s1] = pawn_attacks_bb<BLACK>(square_bb(s1));
 
-  for (File f = FILE_A; f <= FILE_H; f++)
-  {
-      AdjacentFilesBB[f] = (f > FILE_A ? FileBB[f - 1] : 0) | (f < FILE_H ? FileBB[f + 1] : 0);
-      ThisAndAdjacentFilesBB[f] = FileBB[f] | AdjacentFilesBB[f];
-  }
+      for (int step : {-9, -8, -7, -1, 1, 7, 8, 9} )
+         PseudoAttacks[KING][s1] |= safe_destination(s1, step);
 
-  for (Rank r = RANK_1; r < RANK_8; r++)
-      InFrontBB[WHITE][r] = ~(InFrontBB[BLACK][r + 1] = InFrontBB[BLACK][r] | RankBB[r]);
+      for (int step : {-17, -15, -10, -6, 6, 10, 15, 17} )
+         PseudoAttacks[KNIGHT][s1] |= safe_destination(s1, step);
 
-  for (Color c = WHITE; c <= BLACK; c++)
-      for (Square s = SQ_A1; s <= SQ_H8; s++)
-      {
-          ForwardBB[c][s]      = InFrontBB[c][rank_of(s)] & FileBB[file_of(s)];
-          PassedPawnMask[c][s] = InFrontBB[c][rank_of(s)] & ThisAndAdjacentFilesBB[file_of(s)];
-          AttackSpanMask[c][s] = InFrontBB[c][rank_of(s)] & AdjacentFilesBB[file_of(s)];
-      }
+      PseudoAttacks[QUEEN][s1]  = PseudoAttacks[BISHOP][s1] = attacks_bb<BISHOP>(s1, 0);
+      PseudoAttacks[QUEEN][s1] |= PseudoAttacks[  ROOK][s1] = attacks_bb<  ROOK>(s1, 0);
 
-  for (Square s1 = SQ_A1; s1 <= SQ_H8; s1++)
-      for (Square s2 = SQ_A1; s2 <= SQ_H8; s2++)
-          SquareDistance[s1][s2] = std::max(file_distance(s1, s2), rank_distance(s1, s2));
-
-  for (Square s1 = SQ_A1; s1 <= SQ_H8; s1++)
-      for (int d = 1; d < 8; d++)
-          for (Square s2 = SQ_A1; s2 <= SQ_H8; s2++)
-              if (SquareDistance[s1][s2] == d)
-                  DistanceRingsBB[s1][d - 1] |= s2;
-
-  int steps[][9] = { {}, { 7, 9 }, { 17, 15, 10, 6, -6, -10, -15, -17 },
-                     {}, {}, {}, { 9, 7, -7, -9, 8, 1, -1, -8 } };
-
-  for (Color c = WHITE; c <= BLACK; c++)
-      for (PieceType pt = PAWN; pt <= KING; pt++)
-          for (Square s = SQ_A1; s <= SQ_H8; s++)
-              for (int k = 0; steps[pt][k]; k++)
-              {
-                  Square to = s + Square(c == WHITE ? steps[pt][k] : -steps[pt][k]);
-
-                  if (is_ok(to) && square_distance(s, to) < 3)
-                      StepAttacksBB[make_piece(c, pt)][s] |= to;
-              }
-
-  Square RDeltas[] = { DELTA_N,  DELTA_E,  DELTA_S,  DELTA_W  };
-  Square BDeltas[] = { DELTA_NE, DELTA_SE, DELTA_SW, DELTA_NW };
-
-  init_magics(RTable, RAttacks, RMagics, RMasks, RShifts, RDeltas, magic_index<ROOK>);
-  init_magics(BTable, BAttacks, BMagics, BMasks, BShifts, BDeltas, magic_index<BISHOP>);
-
-  for (Square s = SQ_A1; s <= SQ_H8; s++)
-  {
-      PseudoAttacks[QUEEN][s]  = PseudoAttacks[BISHOP][s] = attacks_bb<BISHOP>(s, 0);
-      PseudoAttacks[QUEEN][s] |= PseudoAttacks[  ROOK][s] = attacks_bb<  ROOK>(s, 0);
-  }
-
-  for (Square s1 = SQ_A1; s1 <= SQ_H8; s1++)
-      for (Square s2 = SQ_A1; s2 <= SQ_H8; s2++)
-          if (PseudoAttacks[QUEEN][s1] & s2)
+      for (PieceType pt : { BISHOP, ROOK })
+          for (Square s2 = SQ_A1; s2 <= SQ_H8; ++s2)
           {
-              Square delta = (s2 - s1) / square_distance(s1, s2);
-
-              for (Square s = s1 + delta; s != s2; s += delta)
-                  BetweenBB[s1][s2] |= s;
+              if (PseudoAttacks[pt][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(Square deltas[], Square sq, Bitboard occupied) {
+  Bitboard sliding_attack(PieceType pt, Square sq, Bitboard occupied) {
 
-    Bitboard attack = 0;
+    Bitboard attacks = 0;
+    Direction   RookDirections[4] = {NORTH, SOUTH, EAST, WEST};
+    Direction BishopDirections[4] = {NORTH_EAST, SOUTH_EAST, SOUTH_WEST, NORTH_WEST};
 
-    for (int i = 0; i < 4; i++)
-        for (Square s = sq + deltas[i];
-             is_ok(s) && square_distance(s, s - deltas[i]) == 1;
-             s += deltas[i])
-        {
-            attack |= s;
+    for (Direction d : (pt == ROOK ? RookDirections : BishopDirections))
+    {
+        Square s = sq;
+        while (safe_destination(s, d) && !(occupied & s))
+            attacks |= (s += d);
+    }
 
-            if (occupied & s)
-                break;
-        }
-
-    return attack;
-  }
-
-
-  Bitboard pick_random(RKISS& rk, int booster) {
-
-    // Values s1 and s2 are used to rotate the candidate magic of a
-    // quantity known to be the optimal to quickly find the magics.
-    int s1 = booster & 63, s2 = (booster >> 6) & 63;
-
-    Bitboard m = rk.rand<Bitboard>();
-    m = (m >> s1) | (m << (64 - s1));
-    m &= rk.rand<Bitboard>();
-    m = (m >> s2) | (m << (64 - s2));
-    return m & rk.rand<Bitboard>();
+    return attacks;
   }
 
 
   // init_magics() computes all rook and bishop attacks at startup. Magic
   // bitboards are used to look up attacks of sliding pieces. As a reference see
-  // chessprogramming.wikispaces.com/Magic+Bitboards. In particular, here we
-  // use the so called "fancy" approach.
+  // www.chessprogramming.org/Magic_Bitboards. In particular, here we use the so
+  // called "fancy" approach.
 
-  void init_magics(Bitboard table[], Bitboard* attacks[], Bitboard magics[],
-                   Bitboard masks[], unsigned shifts[], Square deltas[], Fn index) {
+  void init_magics(PieceType pt, Bitboard table[], Magic magics[]) {
+
+    // Optimal PRNG seeds to pick the correct magics in the shortest time
+    int seeds[][RANK_NB] = { { 8977, 44560, 54343, 38998,  5731, 95205, 104912, 17020 },
+                             {  728, 10316, 55013, 32803, 12281, 15100,  16645,   255 } };
 
-    int MagicBoosters[][8] = { { 3191, 2184, 1310, 3618, 2091, 1308, 2452, 3996 },
-                               { 1059, 3608,  605, 3234, 3326,   38, 2029, 3043 } };
-    RKISS rk;
     Bitboard occupancy[4096], reference[4096], edges, b;
-    int i, size, booster;
+    int epoch[4096] = {}, cnt = 0, size = 0;
 
-    // attacks[s] is a pointer to the beginning of the attacks table for square 's'
-    attacks[SQ_A1] = table;
-
-    for (Square s = SQ_A1; s <= SQ_H8; s++)
+    for (Square s = SQ_A1; s <= SQ_H8; ++s)
     {
         // Board edges are not considered in the relevant occupancies
         edges = ((Rank1BB | Rank8BB) & ~rank_bb(s)) | ((FileABB | FileHBB) & ~file_bb(s));
@@ -303,49 +158,61 @@ namespace {
         // all the attacks for each possible subset of the mask and so is 2 power
         // the number of 1s of the mask. Hence we deduce the size of the shift to
         // apply to the 64 or 32 bits word to get the index.
-        masks[s]  = sliding_attack(deltas, s, 0) & ~edges;
-        shifts[s] = (Is64Bit ? 64 : 32) - popcount<Max15>(masks[s]);
+        Magic& m = magics[s];
+        m.mask  = sliding_attack(pt, s, 0) & ~edges;
+        m.shift = (Is64Bit ? 64 : 32) - popcount(m.mask);
+
+        // Set the offset for the attacks table of the square. We have individual
+        // table sizes for each square with "Fancy Magic Bitboards".
+        m.attacks = s == SQ_A1 ? table : magics[s - 1].attacks + size;
 
         // Use Carry-Rippler trick to enumerate all subsets of masks[s] and
         // store the corresponding sliding attack bitboard in reference[].
         b = size = 0;
         do {
             occupancy[size] = b;
-            reference[size++] = sliding_attack(deltas, s, b);
-            b = (b - masks[s]) & masks[s];
+            reference[size] = sliding_attack(pt, s, b);
+
+            if (HasPext)
+                m.attacks[pext(b, m.mask)] = reference[size];
+
+            size++;
+            b = (b - m.mask) & m.mask;
         } while (b);
 
-        // Set the offset for the table of the next square. We have individual
-        // table sizes for each square with "Fancy Magic Bitboards".
-        if (s < SQ_H8)
-            attacks[s + 1] = attacks[s] + size;
+        if (HasPext)
+            continue;
 
-        booster = MagicBoosters[Is64Bit][rank_of(s)];
+        PRNG rng(seeds[Is64Bit][rank_of(s)]);
 
         // Find a magic for square 's' picking up an (almost) random number
         // until we find the one that passes the verification test.
-        do {
-            do magics[s] = pick_random(rk, booster);
-            while (BitCount8Bit[(magics[s] * masks[s]) >> 56] < 6);
-
-            memset(attacks[s], 0, size * sizeof(Bitboard));
+        for (int i = 0; i < size; )
+        {
+            for (m.magic = 0; popcount((m.magic * m.mask) >> 56) < 6; )
+                m.magic = rng.sparse_rand<Bitboard>();
 
             // A good magic must map every possible occupancy to an index that
             // looks up the correct sliding attack in the attacks[s] database.
             // Note that we build up the database for square 's' as a side
-            // effect of verifying the magic.
-            for (i = 0; i < size; i++)
+            // effect of verifying the magic. Keep track of the attempt count
+            // and save it in epoch[], little speed-up trick to avoid resetting
+            // m.attacks[] after every failed attempt.
+            for (++cnt, i = 0; i < size; ++i)
             {
-                Bitboard& attack = attacks[s][index(s, occupancy[i])];
+                unsigned idx = m.index(occupancy[i]);
 
-                if (attack && attack != reference[i])
+                if (epoch[idx] < cnt)
+                {
+                    epoch[idx] = cnt;
+                    m.attacks[idx] = reference[i];
+                }
+                else if (m.attacks[idx] != reference[i])
                     break;
-
-                assert(reference[i] != 0);
-
-                attack = reference[i];
             }
-        } while (i != size);
+        }
     }
   }
 }
+
+} // namespace Stockfish

src/bitboard.h

@@ -1,14 +1,12 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -18,263 +16,435 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(BITBOARD_H_INCLUDED)
+#ifndef BITBOARD_H_INCLUDED
 #define BITBOARD_H_INCLUDED
 
+#include <string>
+
 #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();
-void print(Bitboard b);
+std::string pretty(Bitboard b);
 
+} // namespace Stockfish::Bitboards
+
+constexpr Bitboard AllSquares = ~Bitboard(0);
+constexpr Bitboard DarkSquares = 0xAA55AA55AA55AA55ULL;
+
+constexpr Bitboard FileABB = 0x0101010101010101ULL;
+constexpr Bitboard FileBBB = FileABB << 1;
+constexpr Bitboard FileCBB = FileABB << 2;
+constexpr Bitboard FileDBB = FileABB << 3;
+constexpr Bitboard FileEBB = FileABB << 4;
+constexpr Bitboard FileFBB = FileABB << 5;
+constexpr Bitboard FileGBB = FileABB << 6;
+constexpr Bitboard FileHBB = FileABB << 7;
+
+constexpr Bitboard Rank1BB = 0xFF;
+constexpr Bitboard Rank2BB = Rank1BB << (8 * 1);
+constexpr Bitboard Rank3BB = Rank1BB << (8 * 2);
+constexpr Bitboard Rank4BB = Rank1BB << (8 * 3);
+constexpr Bitboard Rank5BB = Rank1BB << (8 * 4);
+constexpr Bitboard Rank6BB = Rank1BB << (8 * 5);
+constexpr Bitboard Rank7BB = Rank1BB << (8 * 6);
+constexpr Bitboard Rank8BB = Rank1BB << (8 * 7);
+
+constexpr Bitboard QueenSide   = FileABB | FileBBB | FileCBB | FileDBB;
+constexpr Bitboard CenterFiles = FileCBB | FileDBB | FileEBB | FileFBB;
+constexpr Bitboard KingSide    = FileEBB | FileFBB | FileGBB | FileHBB;
+constexpr Bitboard Center      = (FileDBB | FileEBB) & (Rank4BB | Rank5BB);
+
+constexpr Bitboard KingFlank[FILE_NB] = {
+  QueenSide ^ FileDBB, QueenSide, QueenSide,
+  CenterFiles, CenterFiles,
+  KingSide, KingSide, KingSide ^ FileEBB
+};
+
+extern uint8_t PopCnt16[1 << 16];
+extern uint8_t SquareDistance[SQUARE_NB][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];
+
+
+/// Magic holds all magic bitboards relevant data for a single square
+struct Magic {
+  Bitboard  mask;
+  Bitboard  magic;
+  Bitboard* attacks;
+  unsigned  shift;
+
+  // Compute the attack's index using the 'magic bitboards' approach
+  unsigned index(Bitboard occupied) const {
+
+    if (HasPext)
+        return unsigned(pext(occupied, mask));
+
+    if (Is64Bit)
+        return unsigned(((occupied & mask) * magic) >> shift);
+
+    unsigned lo = unsigned(occupied) & unsigned(mask);
+    unsigned hi = unsigned(occupied >> 32) & unsigned(mask >> 32);
+    return (lo * unsigned(magic) ^ hi * unsigned(magic >> 32)) >> shift;
+  }
+};
+
+extern Magic RookMagics[SQUARE_NB];
+extern Magic BishopMagics[SQUARE_NB];
+
+inline Bitboard square_bb(Square s) {
+  assert(is_ok(s));
+  return (1ULL << s);
 }
 
-namespace Bitbases {
-
-void init_kpk();
-uint32_t probe_kpk(Square wksq, Square wpsq, Square bksq, Color stm);
-
-}
-
-CACHE_LINE_ALIGNMENT
-
-extern Bitboard RMasks[64];
-extern Bitboard RMagics[64];
-extern Bitboard* RAttacks[64];
-extern unsigned RShifts[64];
-
-extern Bitboard BMasks[64];
-extern Bitboard BMagics[64];
-extern Bitboard* BAttacks[64];
-extern unsigned BShifts[64];
-
-extern Bitboard SquareBB[64];
-extern Bitboard FileBB[8];
-extern Bitboard RankBB[8];
-extern Bitboard AdjacentFilesBB[8];
-extern Bitboard ThisAndAdjacentFilesBB[8];
-extern Bitboard InFrontBB[2][8];
-extern Bitboard StepAttacksBB[16][64];
-extern Bitboard BetweenBB[64][64];
-extern Bitboard DistanceRingsBB[64][8];
-extern Bitboard ForwardBB[2][64];
-extern Bitboard PassedPawnMask[2][64];
-extern Bitboard AttackSpanMask[2][64];
-extern Bitboard PseudoAttacks[6][64];
-
 
 /// Overloads of bitwise operators between a Bitboard and a Square for testing
 /// whether a given bit is set in a bitboard, and for setting and clearing bits.
 
-inline Bitboard operator&(Bitboard b, Square s) {
-  return b & SquareBB[s];
-}
+inline Bitboard  operator&( Bitboard  b, Square s) { return b &  square_bb(s); }
+inline Bitboard  operator|( Bitboard  b, Square s) { return b |  square_bb(s); }
+inline Bitboard  operator^( Bitboard  b, Square s) { return b ^  square_bb(s); }
+inline Bitboard& operator|=(Bitboard& b, Square s) { return b |= square_bb(s); }
+inline Bitboard& operator^=(Bitboard& b, Square s) { return b ^= square_bb(s); }
 
-inline Bitboard& operator|=(Bitboard& b, Square s) {
-  return b |= SquareBB[s];
-}
+inline Bitboard  operator&(Square s, Bitboard b) { return b & s; }
+inline Bitboard  operator|(Square s, Bitboard b) { return b | s; }
+inline Bitboard  operator^(Square s, Bitboard b) { return b ^ s; }
 
-inline Bitboard& operator^=(Bitboard& b, Square s) {
-  return b ^= SquareBB[s];
-}
+inline Bitboard  operator|(Square s1, Square s2) { return square_bb(s1) | s2; }
 
-inline Bitboard operator|(Bitboard b, Square s) {
-  return b | SquareBB[s];
-}
-
-inline Bitboard operator^(Bitboard b, Square s) {
-  return b ^ SquareBB[s];
-}
-
-
-/// more_than_one() returns true if in 'b' there is more than one bit set
-
-inline bool more_than_one(Bitboard b) {
+constexpr bool more_than_one(Bitboard b) {
   return b & (b - 1);
 }
 
 
-/// rank_bb() and file_bb() take a file or a square as input and return
-/// a bitboard representing all squares on the given file or rank.
-
-inline Bitboard rank_bb(Rank r) {
-  return RankBB[r];
-}
-
-inline Bitboard rank_bb(Square s) {
-  return RankBB[rank_of(s)];
-}
-
-inline Bitboard file_bb(File f) {
-  return FileBB[f];
-}
-
-inline Bitboard file_bb(Square s) {
-  return FileBB[file_of(s)];
+constexpr bool opposite_colors(Square s1, Square s2) {
+  return (s1 + rank_of(s1) + s2 + rank_of(s2)) & 1;
 }
 
 
-/// adjacent_files_bb takes a file as input and returns a bitboard representing
-/// all squares on the adjacent files.
+/// rank_bb() and file_bb() return a bitboard representing all the squares on
+/// the given file or rank.
 
-inline Bitboard adjacent_files_bb(File f) {
-  return AdjacentFilesBB[f];
+constexpr Bitboard rank_bb(Rank r) {
+  return Rank1BB << (8 * r);
+}
+
+constexpr Bitboard rank_bb(Square s) {
+  return rank_bb(rank_of(s));
+}
+
+constexpr Bitboard file_bb(File f) {
+  return FileABB << f;
+}
+
+constexpr Bitboard file_bb(Square s) {
+  return file_bb(file_of(s));
 }
 
 
-/// this_and_adjacent_files_bb takes a file as input and returns a bitboard
-/// representing all squares on the given and adjacent files.
+/// shift() moves a bitboard one or two steps as specified by the direction D
 
-inline Bitboard this_and_adjacent_files_bb(File f) {
-  return ThisAndAdjacentFilesBB[f];
+template<Direction D>
+constexpr Bitboard shift(Bitboard b) {
+  return  D == NORTH      ?  b             << 8 : D == SOUTH      ?  b             >> 8
+        : D == NORTH+NORTH?  b             <<16 : D == SOUTH+SOUTH?  b             >>16
+        : D == EAST       ? (b & ~FileHBB) << 1 : D == WEST       ? (b & ~FileABB) >> 1
+        : D == NORTH_EAST ? (b & ~FileHBB) << 9 : D == NORTH_WEST ? (b & ~FileABB) << 7
+        : D == SOUTH_EAST ? (b & ~FileHBB) >> 7 : D == SOUTH_WEST ? (b & ~FileABB) >> 9
+        : 0;
 }
 
 
-/// in_front_bb() takes a color and a rank or square as input, and returns a
-/// bitboard representing all the squares on all ranks in front of the rank
-/// (or square), from the given color's point of view.  For instance,
-/// in_front_bb(WHITE, RANK_5) will give all squares on ranks 6, 7 and 8, while
-/// in_front_bb(BLACK, SQ_D3) will give all squares on ranks 1 and 2.
+/// pawn_attacks_bb() returns the squares attacked by pawns of the given color
+/// from the squares in the given bitboard.
 
-inline Bitboard in_front_bb(Color c, Rank r) {
-  return InFrontBB[c][r];
+template<Color C>
+constexpr Bitboard pawn_attacks_bb(Bitboard b) {
+  return C == WHITE ? shift<NORTH_WEST>(b) | shift<NORTH_EAST>(b)
+                    : shift<SOUTH_WEST>(b) | shift<SOUTH_EAST>(b);
 }
 
-inline Bitboard in_front_bb(Color c, Square s) {
-  return InFrontBB[c][rank_of(s)];
+inline Bitboard pawn_attacks_bb(Color c, Square s) {
+
+  assert(is_ok(s));
+  return PawnAttacks[c][s];
 }
 
 
-/// between_bb returns a bitboard representing all squares between two squares.
-/// For instance, between_bb(SQ_C4, SQ_F7) returns a bitboard with the bits for
-/// square d5 and e6 set.  If s1 and s2 are not on the same line, file or diagonal,
-/// 0 is returned.
+/// pawn_double_attacks_bb() returns the squares doubly attacked by pawns of the
+/// given color from the squares in the given bitboard.
+
+template<Color C>
+constexpr Bitboard pawn_double_attacks_bb(Bitboard b) {
+  return C == WHITE ? shift<NORTH_WEST>(b) & shift<NORTH_EAST>(b)
+                    : shift<SOUTH_WEST>(b) & shift<SOUTH_EAST>(b);
+}
+
+
+/// adjacent_files_bb() returns a bitboard representing all the squares on the
+/// adjacent files of a given square.
+
+constexpr Bitboard adjacent_files_bb(Square s) {
+  return shift<EAST>(file_bb(s)) | shift<WEST>(file_bb(s));
+}
+
+
+/// line_bb() returns a bitboard representing an entire line (from board edge
+/// to board edge) that intersects the two given squares. If the given squares
+/// are not on a same file/rank/diagonal, the function returns 0. For instance,
+/// line_bb(SQ_C4, SQ_F7) will return a bitboard with the A2-G8 diagonal.
+
+inline Bitboard line_bb(Square s1, Square s2) {
+
+  assert(is_ok(s1) && is_ok(s2));
+
+  return LineBB[s1][s2];
+}
+
+
+/// 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) {
+
+  assert(is_ok(s1) && is_ok(s2));
+
   return BetweenBB[s1][s2];
 }
 
 
-/// forward_bb takes a color and a square as input, and returns a bitboard
-/// representing all squares along the line in front of the square, from the
-/// point of view of the given color. Definition of the table is:
-/// ForwardBB[c][s] = in_front_bb(c, s) & file_bb(s)
+/// 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.
 
-inline Bitboard forward_bb(Color c, Square s) {
-  return ForwardBB[c][s];
+constexpr Bitboard forward_ranks_bb(Color c, Square s) {
+  return c == WHITE ? ~Rank1BB << 8 * relative_rank(WHITE, s)
+                    : ~Rank8BB >> 8 * relative_rank(BLACK, s);
 }
 
 
-/// passed_pawn_mask takes a color and a square as input, and returns a
-/// bitboard mask which can be used to test if a pawn of the given color on
-/// the given square is a passed pawn. Definition of the table is:
-/// PassedPawnMask[c][s] = in_front_bb(c, s) & this_and_adjacent_files_bb(s)
+/// forward_file_bb() returns a bitboard representing all the squares along the
+/// line in front of the given one, from the point of view of the given color.
 
-inline Bitboard passed_pawn_mask(Color c, Square s) {
-  return PassedPawnMask[c][s];
+constexpr Bitboard forward_file_bb(Color c, Square s) {
+  return forward_ranks_bb(c, s) & file_bb(s);
 }
 
 
-/// attack_span_mask takes a color and a square as input, and returns a bitboard
-/// representing all squares that can be attacked by a pawn of the given color
-/// when it moves along its file starting from the given square. Definition is:
-/// AttackSpanMask[c][s] = in_front_bb(c, s) & adjacent_files_bb(s);
+/// pawn_attack_span() returns a bitboard representing all the squares that can
+/// be attacked by a pawn of the given color when it moves along its file, starting
+/// from the given square.
 
-inline Bitboard attack_span_mask(Color c, Square s) {
-  return AttackSpanMask[c][s];
+constexpr Bitboard pawn_attack_span(Color c, Square s) {
+  return forward_ranks_bb(c, s) & adjacent_files_bb(s);
 }
 
 
-/// squares_aligned returns true if the squares s1, s2 and s3 are aligned
-/// either on a straight or on a diagonal line.
+/// passed_pawn_span() returns a bitboard which can be used to test if a pawn of
+/// the given color and on the given square is a passed pawn.
 
-inline bool squares_aligned(Square s1, Square s2, Square s3) {
-  return  (BetweenBB[s1][s2] | BetweenBB[s1][s3] | BetweenBB[s2][s3])
-        & (     SquareBB[s1] |      SquareBB[s2] |      SquareBB[s3]);
+constexpr Bitboard passed_pawn_span(Color c, Square s) {
+  return pawn_attack_span(c, s) | forward_file_bb(c, s);
 }
 
 
-/// same_color_squares() returns a bitboard representing all squares with
-/// the same color of the given square.
+/// aligned() returns true if the squares s1, s2 and s3 are aligned either on a
+/// straight or on a diagonal line.
 
-inline Bitboard same_color_squares(Square s) {
-  return Bitboard(0xAA55AA55AA55AA55ULL) & s ?  0xAA55AA55AA55AA55ULL
-                                             : ~0xAA55AA55AA55AA55ULL;
+inline bool aligned(Square s1, Square s2, Square s3) {
+  return line_bb(s1, s2) & s3;
 }
 
 
-/// Functions for computing sliding attack bitboards. Function attacks_bb() takes
-/// a square and a bitboard of occupied squares as input, and returns a bitboard
-/// representing all squares attacked by Pt (bishop or rook) on the given square.
-template<PieceType Pt>
-FORCE_INLINE unsigned magic_index(Square s, Bitboard occ) {
+/// distance() functions return the distance between x and y, defined as the
+/// number of steps for a king in x to reach y.
 
-  Bitboard* const Masks  = Pt == ROOK ? RMasks  : BMasks;
-  Bitboard* const Magics = Pt == ROOK ? RMagics : BMagics;
-  unsigned* const Shifts = Pt == ROOK ? RShifts : BShifts;
+template<typename T1 = Square> inline int distance(Square x, Square y);
+template<> inline int distance<File>(Square x, Square y) { return std::abs(file_of(x) - file_of(y)); }
+template<> inline int distance<Rank>(Square x, Square y) { return std::abs(rank_of(x) - rank_of(y)); }
+template<> inline int distance<Square>(Square x, Square y) { return SquareDistance[x][y]; }
 
-  if (Is64Bit)
-      return unsigned(((occ & Masks[s]) * Magics[s]) >> Shifts[s]);
+inline int edge_distance(File f) { return std::min(f, File(FILE_H - f)); }
+inline int edge_distance(Rank r) { return std::min(r, Rank(RANK_8 - r)); }
 
-  unsigned lo = unsigned(occ) & unsigned(Masks[s]);
-  unsigned hi = unsigned(occ >> 32) & unsigned(Masks[s] >> 32);
-  return (lo * unsigned(Magics[s]) ^ hi * unsigned(Magics[s] >> 32)) >> Shifts[s];
-}
+
+/// attacks_bb(Square) returns the pseudo attacks of the give piece type
+/// assuming an empty board.
 
 template<PieceType Pt>
-inline Bitboard attacks_bb(Square s, Bitboard occ) {
-  return (Pt == ROOK ? RAttacks : BAttacks)[s][magic_index<Pt>(s, occ)];
+inline Bitboard attacks_bb(Square s) {
+
+  assert((Pt != PAWN) && (is_ok(s)));
+
+  return PseudoAttacks[Pt][s];
 }
 
 
-/// lsb()/msb() finds the least/most significant bit in a nonzero bitboard.
-/// pop_lsb() finds and clears the least significant bit in a nonzero bitboard.
+/// attacks_bb(Square, Bitboard) returns the attacks by the given piece
+/// assuming the board is occupied according to the passed Bitboard.
+/// Sliding piece attacks do not continue passed an occupied square.
 
-#if defined(USE_BSFQ)
+template<PieceType Pt>
+inline Bitboard attacks_bb(Square s, Bitboard occupied) {
 
-#  if defined(_MSC_VER) && !defined(__INTEL_COMPILER)
+  assert((Pt != PAWN) && (is_ok(s)));
 
-FORCE_INLINE Square lsb(Bitboard b) {
-  unsigned long index;
-  _BitScanForward64(&index, b);
-  return (Square) index;
+  switch (Pt)
+  {
+  case BISHOP: return BishopMagics[s].attacks[BishopMagics[s].index(occupied)];
+  case ROOK  : return   RookMagics[s].attacks[  RookMagics[s].index(occupied)];
+  case QUEEN : return attacks_bb<BISHOP>(s, occupied) | attacks_bb<ROOK>(s, occupied);
+  default    : return PseudoAttacks[Pt][s];
+  }
 }
 
-FORCE_INLINE Square msb(Bitboard b) {
-  unsigned long index;
-  _BitScanReverse64(&index, b);
-  return (Square) index;
+inline Bitboard attacks_bb(PieceType pt, Square s, Bitboard occupied) {
+
+  assert((pt != PAWN) && (is_ok(s)));
+
+  switch (pt)
+  {
+  case BISHOP: return attacks_bb<BISHOP>(s, occupied);
+  case ROOK  : return attacks_bb<  ROOK>(s, occupied);
+  case QUEEN : return attacks_bb<BISHOP>(s, occupied) | attacks_bb<ROOK>(s, occupied);
+  default    : return PseudoAttacks[pt][s];
+  }
 }
 
-#  else
 
-FORCE_INLINE Square lsb(Bitboard b) { // Assembly code by Heinz van Saanen
-  Bitboard index;
-  __asm__("bsfq %1, %0": "=r"(index): "rm"(b) );
-  return (Square) index;
+/// popcount() counts the number of non-zero bits in a bitboard
+
+inline int popcount(Bitboard b) {
+
+#ifndef USE_POPCNT
+
+  union { Bitboard bb; uint16_t u[4]; } v = { b };
+  return PopCnt16[v.u[0]] + PopCnt16[v.u[1]] + PopCnt16[v.u[2]] + PopCnt16[v.u[3]];
+
+#elif defined(_MSC_VER) || defined(__INTEL_COMPILER)
+
+  return (int)_mm_popcnt_u64(b);
+
+#else // Assumed gcc or compatible compiler
+
+  return __builtin_popcountll(b);
+
+#endif
 }
 
-FORCE_INLINE Square msb(Bitboard b) {
-  Bitboard index;
-  __asm__("bsrq %1, %0": "=r"(index): "rm"(b) );
-  return (Square) index;
+
+/// lsb() and msb() return the least/most significant bit in a non-zero bitboard
+
+#if defined(__GNUC__)  // GCC, Clang, ICC
+
+inline Square lsb(Bitboard b) {
+  assert(b);
+  return Square(__builtin_ctzll(b));
 }
 
-#  endif
-
-FORCE_INLINE Square pop_lsb(Bitboard* b) {
-  const Square s = lsb(*b);
-  *b &= ~(1ULL << s);
-  return s;
+inline Square msb(Bitboard b) {
+  assert(b);
+  return Square(63 ^ __builtin_clzll(b));
 }
 
-#else // if !defined(USE_BSFQ)
+#elif defined(_MSC_VER)  // MSVC
 
-extern Square msb(Bitboard b);
-extern Square lsb(Bitboard b);
-extern Square pop_lsb(Bitboard* b);
+#ifdef _WIN64  // MSVC, WIN64
+
+inline Square lsb(Bitboard b) {
+  assert(b);
+  unsigned long idx;
+  _BitScanForward64(&idx, b);
+  return (Square) idx;
+}
+
+inline Square msb(Bitboard b) {
+  assert(b);
+  unsigned long idx;
+  _BitScanReverse64(&idx, b);
+  return (Square) idx;
+}
+
+#else  // MSVC, WIN32
+
+inline Square lsb(Bitboard b) {
+  assert(b);
+  unsigned long idx;
+
+  if (b & 0xffffffff) {
+      _BitScanForward(&idx, int32_t(b));
+      return Square(idx);
+  } else {
+      _BitScanForward(&idx, int32_t(b >> 32));
+      return Square(idx + 32);
+  }
+}
+
+inline Square msb(Bitboard b) {
+  assert(b);
+  unsigned long idx;
+
+  if (b >> 32) {
+      _BitScanReverse(&idx, int32_t(b >> 32));
+      return Square(idx + 32);
+  } else {
+      _BitScanReverse(&idx, int32_t(b));
+      return Square(idx);
+  }
+}
 
 #endif
 
-#endif // !defined(BITBOARD_H_INCLUDED)
+#else  // Compiler is neither GCC nor MSVC compatible
+
+#error "Compiler not supported."
+
+#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;
+  return s;
+}
+
+
+/// frontmost_sq() returns the most advanced square for the given color,
+/// requires a non-zero bitboard.
+inline Square frontmost_sq(Color c, Bitboard b) {
+  assert(b);
+  return c == WHITE ? msb(b) : lsb(b);
+}
+
+} // namespace Stockfish
+
+#endif // #ifndef BITBOARD_H_INCLUDED

src/bitcount.h

@@ -1,112 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
-
-  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/>.
-*/
-
-#if !defined(BITCOUNT_H_INCLUDED)
-#define BITCOUNT_H_INCLUDED
-
-#include <cassert>
-#include "types.h"
-
-enum BitCountType {
-  CNT_64,
-  CNT_64_MAX15,
-  CNT_32,
-  CNT_32_MAX15,
-  CNT_HW_POPCNT
-};
-
-/// Determine at compile time the best popcount<> specialization according if
-/// platform is 32 or 64 bits, to the maximum number of nonzero bits to count or
-/// use hardware popcnt instruction when available.
-const BitCountType Full  = HasPopCnt ? CNT_HW_POPCNT : Is64Bit ? CNT_64 : CNT_32;
-const BitCountType Max15 = HasPopCnt ? CNT_HW_POPCNT : Is64Bit ? CNT_64_MAX15 : CNT_32_MAX15;
-
-
-/// popcount() counts the number of nonzero bits in a bitboard
-template<BitCountType> inline int popcount(Bitboard);
-
-template<>
-inline int popcount<CNT_64>(Bitboard b) {
-  b -= ((b>>1) & 0x5555555555555555ULL);
-  b = ((b>>2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
-  b = ((b>>4) + b) & 0x0F0F0F0F0F0F0F0FULL;
-  b *= 0x0101010101010101ULL;
-  return int(b >> 56);
-}
-
-template<>
-inline int popcount<CNT_64_MAX15>(Bitboard b) {
-  b -= (b>>1) & 0x5555555555555555ULL;
-  b = ((b>>2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
-  b *= 0x1111111111111111ULL;
-  return int(b >> 60);
-}
-
-template<>
-inline int popcount<CNT_32>(Bitboard b) {
-  unsigned w = unsigned(b >> 32), v = unsigned(b);
-  v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
-  w -= (w >> 1) & 0x55555555;
-  v = ((v >> 2) & 0x33333333) + (v & 0x33333333); // 0-4 in 4 bits
-  w = ((w >> 2) & 0x33333333) + (w & 0x33333333);
-  v = ((v >> 4) + v) & 0x0F0F0F0F; // 0-8 in 8 bits
-  v += (((w >> 4) + w) & 0x0F0F0F0F);  // 0-16 in 8 bits
-  v *= 0x01010101; // mul is fast on amd procs
-  return int(v >> 24);
-}
-
-template<>
-inline int popcount<CNT_32_MAX15>(Bitboard b) {
-  unsigned w = unsigned(b >> 32), v = unsigned(b);
-  v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
-  w -= (w >> 1) & 0x55555555;
-  v = ((v >> 2) & 0x33333333) + (v & 0x33333333); // 0-4 in 4 bits
-  w = ((w >> 2) & 0x33333333) + (w & 0x33333333);
-  v += w; // 0-8 in 4 bits
-  v *= 0x11111111;
-  return int(v >> 28);
-}
-
-template<>
-inline int popcount<CNT_HW_POPCNT>(Bitboard b) {
-
-#if !defined(USE_POPCNT)
-
-  assert(false);
-  return b != 0; // Avoid 'b not used' warning
-
-#elif defined(_MSC_VER) && defined(__INTEL_COMPILER)
-
-  return _mm_popcnt_u64(b);
-
-#elif defined(_MSC_VER)
-
-  return (int)__popcnt64(b);
-
-#else
-
-  unsigned long ret;
-  __asm__("popcnt %1, %0" : "=r" (ret) : "r" (b));
-  return ret;
-
-#endif
-}
-
-#endif // !defined(BITCOUNT_H_INCLUDED)

src/book.cpp

@@ -1,481 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
-
-  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/>.
-*/
-
-/*
-  The code in this file is based on the opening book code in PolyGlot
-  by Fabien Letouzey. PolyGlot is available under the GNU General
-  Public License, and can be downloaded from http://wbec-ridderkerk.nl
-*/
-
-#include <algorithm>
-#include <cassert>
-#include <iostream>
-
-#include "book.h"
-#include "misc.h"
-#include "movegen.h"
-
-using namespace std;
-
-namespace {
-
-  // A Polyglot book is a series of "entries" of 16 bytes. All integers are
-  // stored in big-endian format, with highest byte first (regardless of size).
-  // The entries are ordered according to the key in ascending order.
-  struct BookEntry {
-    uint64_t key;
-    uint16_t move;
-    uint16_t count;
-    uint32_t learn;
-  };
-
-  // Random numbers from PolyGlot, used to compute book hash keys
-  const Key PolyGlotRandoms[781] = {
-    0x9D39247E33776D41ULL, 0x2AF7398005AAA5C7ULL, 0x44DB015024623547ULL,
-    0x9C15F73E62A76AE2ULL, 0x75834465489C0C89ULL, 0x3290AC3A203001BFULL,
-    0x0FBBAD1F61042279ULL, 0xE83A908FF2FB60CAULL, 0x0D7E765D58755C10ULL,
-    0x1A083822CEAFE02DULL, 0x9605D5F0E25EC3B0ULL, 0xD021FF5CD13A2ED5ULL,
-    0x40BDF15D4A672E32ULL, 0x011355146FD56395ULL, 0x5DB4832046F3D9E5ULL,
-    0x239F8B2D7FF719CCULL, 0x05D1A1AE85B49AA1ULL, 0x679F848F6E8FC971ULL,
-    0x7449BBFF801FED0BULL, 0x7D11CDB1C3B7ADF0ULL, 0x82C7709E781EB7CCULL,
-    0xF3218F1C9510786CULL, 0x331478F3AF51BBE6ULL, 0x4BB38DE5E7219443ULL,
-    0xAA649C6EBCFD50FCULL, 0x8DBD98A352AFD40BULL, 0x87D2074B81D79217ULL,
-    0x19F3C751D3E92AE1ULL, 0xB4AB30F062B19ABFULL, 0x7B0500AC42047AC4ULL,
-    0xC9452CA81A09D85DULL, 0x24AA6C514DA27500ULL, 0x4C9F34427501B447ULL,
-    0x14A68FD73C910841ULL, 0xA71B9B83461CBD93ULL, 0x03488B95B0F1850FULL,
-    0x637B2B34FF93C040ULL, 0x09D1BC9A3DD90A94ULL, 0x3575668334A1DD3BULL,
-    0x735E2B97A4C45A23ULL, 0x18727070F1BD400BULL, 0x1FCBACD259BF02E7ULL,
-    0xD310A7C2CE9B6555ULL, 0xBF983FE0FE5D8244ULL, 0x9F74D14F7454A824ULL,
-    0x51EBDC4AB9BA3035ULL, 0x5C82C505DB9AB0FAULL, 0xFCF7FE8A3430B241ULL,
-    0x3253A729B9BA3DDEULL, 0x8C74C368081B3075ULL, 0xB9BC6C87167C33E7ULL,
-    0x7EF48F2B83024E20ULL, 0x11D505D4C351BD7FULL, 0x6568FCA92C76A243ULL,
-    0x4DE0B0F40F32A7B8ULL, 0x96D693460CC37E5DULL, 0x42E240CB63689F2FULL,
-    0x6D2BDCDAE2919661ULL, 0x42880B0236E4D951ULL, 0x5F0F4A5898171BB6ULL,
-    0x39F890F579F92F88ULL, 0x93C5B5F47356388BULL, 0x63DC359D8D231B78ULL,
-    0xEC16CA8AEA98AD76ULL, 0x5355F900C2A82DC7ULL, 0x07FB9F855A997142ULL,
-    0x5093417AA8A7ED5EULL, 0x7BCBC38DA25A7F3CULL, 0x19FC8A768CF4B6D4ULL,
-    0x637A7780DECFC0D9ULL, 0x8249A47AEE0E41F7ULL, 0x79AD695501E7D1E8ULL,
-    0x14ACBAF4777D5776ULL, 0xF145B6BECCDEA195ULL, 0xDABF2AC8201752FCULL,
-    0x24C3C94DF9C8D3F6ULL, 0xBB6E2924F03912EAULL, 0x0CE26C0B95C980D9ULL,
-    0xA49CD132BFBF7CC4ULL, 0xE99D662AF4243939ULL, 0x27E6AD7891165C3FULL,
-    0x8535F040B9744FF1ULL, 0x54B3F4FA5F40D873ULL, 0x72B12C32127FED2BULL,
-    0xEE954D3C7B411F47ULL, 0x9A85AC909A24EAA1ULL, 0x70AC4CD9F04F21F5ULL,
-    0xF9B89D3E99A075C2ULL, 0x87B3E2B2B5C907B1ULL, 0xA366E5B8C54F48B8ULL,
-    0xAE4A9346CC3F7CF2ULL, 0x1920C04D47267BBDULL, 0x87BF02C6B49E2AE9ULL,
-    0x092237AC237F3859ULL, 0xFF07F64EF8ED14D0ULL, 0x8DE8DCA9F03CC54EULL,
-    0x9C1633264DB49C89ULL, 0xB3F22C3D0B0B38EDULL, 0x390E5FB44D01144BULL,
-    0x5BFEA5B4712768E9ULL, 0x1E1032911FA78984ULL, 0x9A74ACB964E78CB3ULL,
-    0x4F80F7A035DAFB04ULL, 0x6304D09A0B3738C4ULL, 0x2171E64683023A08ULL,
-    0x5B9B63EB9CEFF80CULL, 0x506AACF489889342ULL, 0x1881AFC9A3A701D6ULL,
-    0x6503080440750644ULL, 0xDFD395339CDBF4A7ULL, 0xEF927DBCF00C20F2ULL,
-    0x7B32F7D1E03680ECULL, 0xB9FD7620E7316243ULL, 0x05A7E8A57DB91B77ULL,
-    0xB5889C6E15630A75ULL, 0x4A750A09CE9573F7ULL, 0xCF464CEC899A2F8AULL,
-    0xF538639CE705B824ULL, 0x3C79A0FF5580EF7FULL, 0xEDE6C87F8477609DULL,
-    0x799E81F05BC93F31ULL, 0x86536B8CF3428A8CULL, 0x97D7374C60087B73ULL,
-    0xA246637CFF328532ULL, 0x043FCAE60CC0EBA0ULL, 0x920E449535DD359EULL,
-    0x70EB093B15B290CCULL, 0x73A1921916591CBDULL, 0x56436C9FE1A1AA8DULL,
-    0xEFAC4B70633B8F81ULL, 0xBB215798D45DF7AFULL, 0x45F20042F24F1768ULL,
-    0x930F80F4E8EB7462ULL, 0xFF6712FFCFD75EA1ULL, 0xAE623FD67468AA70ULL,
-    0xDD2C5BC84BC8D8FCULL, 0x7EED120D54CF2DD9ULL, 0x22FE545401165F1CULL,
-    0xC91800E98FB99929ULL, 0x808BD68E6AC10365ULL, 0xDEC468145B7605F6ULL,
-    0x1BEDE3A3AEF53302ULL, 0x43539603D6C55602ULL, 0xAA969B5C691CCB7AULL,
-    0xA87832D392EFEE56ULL, 0x65942C7B3C7E11AEULL, 0xDED2D633CAD004F6ULL,
-    0x21F08570F420E565ULL, 0xB415938D7DA94E3CULL, 0x91B859E59ECB6350ULL,
-    0x10CFF333E0ED804AULL, 0x28AED140BE0BB7DDULL, 0xC5CC1D89724FA456ULL,
-    0x5648F680F11A2741ULL, 0x2D255069F0B7DAB3ULL, 0x9BC5A38EF729ABD4ULL,
-    0xEF2F054308F6A2BCULL, 0xAF2042F5CC5C2858ULL, 0x480412BAB7F5BE2AULL,
-    0xAEF3AF4A563DFE43ULL, 0x19AFE59AE451497FULL, 0x52593803DFF1E840ULL,
-    0xF4F076E65F2CE6F0ULL, 0x11379625747D5AF3ULL, 0xBCE5D2248682C115ULL,
-    0x9DA4243DE836994FULL, 0x066F70B33FE09017ULL, 0x4DC4DE189B671A1CULL,
-    0x51039AB7712457C3ULL, 0xC07A3F80C31FB4B4ULL, 0xB46EE9C5E64A6E7CULL,
-    0xB3819A42ABE61C87ULL, 0x21A007933A522A20ULL, 0x2DF16F761598AA4FULL,
-    0x763C4A1371B368FDULL, 0xF793C46702E086A0ULL, 0xD7288E012AEB8D31ULL,
-    0xDE336A2A4BC1C44BULL, 0x0BF692B38D079F23ULL, 0x2C604A7A177326B3ULL,
-    0x4850E73E03EB6064ULL, 0xCFC447F1E53C8E1BULL, 0xB05CA3F564268D99ULL,
-    0x9AE182C8BC9474E8ULL, 0xA4FC4BD4FC5558CAULL, 0xE755178D58FC4E76ULL,
-    0x69B97DB1A4C03DFEULL, 0xF9B5B7C4ACC67C96ULL, 0xFC6A82D64B8655FBULL,
-    0x9C684CB6C4D24417ULL, 0x8EC97D2917456ED0ULL, 0x6703DF9D2924E97EULL,
-    0xC547F57E42A7444EULL, 0x78E37644E7CAD29EULL, 0xFE9A44E9362F05FAULL,
-    0x08BD35CC38336615ULL, 0x9315E5EB3A129ACEULL, 0x94061B871E04DF75ULL,
-    0xDF1D9F9D784BA010ULL, 0x3BBA57B68871B59DULL, 0xD2B7ADEEDED1F73FULL,
-    0xF7A255D83BC373F8ULL, 0xD7F4F2448C0CEB81ULL, 0xD95BE88CD210FFA7ULL,
-    0x336F52F8FF4728E7ULL, 0xA74049DAC312AC71ULL, 0xA2F61BB6E437FDB5ULL,
-    0x4F2A5CB07F6A35B3ULL, 0x87D380BDA5BF7859ULL, 0x16B9F7E06C453A21ULL,
-    0x7BA2484C8A0FD54EULL, 0xF3A678CAD9A2E38CULL, 0x39B0BF7DDE437BA2ULL,
-    0xFCAF55C1BF8A4424ULL, 0x18FCF680573FA594ULL, 0x4C0563B89F495AC3ULL,
-    0x40E087931A00930DULL, 0x8CFFA9412EB642C1ULL, 0x68CA39053261169FULL,
-    0x7A1EE967D27579E2ULL, 0x9D1D60E5076F5B6FULL, 0x3810E399B6F65BA2ULL,
-    0x32095B6D4AB5F9B1ULL, 0x35CAB62109DD038AULL, 0xA90B24499FCFAFB1ULL,
-    0x77A225A07CC2C6BDULL, 0x513E5E634C70E331ULL, 0x4361C0CA3F692F12ULL,
-    0xD941ACA44B20A45BULL, 0x528F7C8602C5807BULL, 0x52AB92BEB9613989ULL,
-    0x9D1DFA2EFC557F73ULL, 0x722FF175F572C348ULL, 0x1D1260A51107FE97ULL,
-    0x7A249A57EC0C9BA2ULL, 0x04208FE9E8F7F2D6ULL, 0x5A110C6058B920A0ULL,
-    0x0CD9A497658A5698ULL, 0x56FD23C8F9715A4CULL, 0x284C847B9D887AAEULL,
-    0x04FEABFBBDB619CBULL, 0x742E1E651C60BA83ULL, 0x9A9632E65904AD3CULL,
-    0x881B82A13B51B9E2ULL, 0x506E6744CD974924ULL, 0xB0183DB56FFC6A79ULL,
-    0x0ED9B915C66ED37EULL, 0x5E11E86D5873D484ULL, 0xF678647E3519AC6EULL,
-    0x1B85D488D0F20CC5ULL, 0xDAB9FE6525D89021ULL, 0x0D151D86ADB73615ULL,
-    0xA865A54EDCC0F019ULL, 0x93C42566AEF98FFBULL, 0x99E7AFEABE000731ULL,
-    0x48CBFF086DDF285AULL, 0x7F9B6AF1EBF78BAFULL, 0x58627E1A149BBA21ULL,
-    0x2CD16E2ABD791E33ULL, 0xD363EFF5F0977996ULL, 0x0CE2A38C344A6EEDULL,
-    0x1A804AADB9CFA741ULL, 0x907F30421D78C5DEULL, 0x501F65EDB3034D07ULL,
-    0x37624AE5A48FA6E9ULL, 0x957BAF61700CFF4EULL, 0x3A6C27934E31188AULL,
-    0xD49503536ABCA345ULL, 0x088E049589C432E0ULL, 0xF943AEE7FEBF21B8ULL,
-    0x6C3B8E3E336139D3ULL, 0x364F6FFA464EE52EULL, 0xD60F6DCEDC314222ULL,
-    0x56963B0DCA418FC0ULL, 0x16F50EDF91E513AFULL, 0xEF1955914B609F93ULL,
-    0x565601C0364E3228ULL, 0xECB53939887E8175ULL, 0xBAC7A9A18531294BULL,
-    0xB344C470397BBA52ULL, 0x65D34954DAF3CEBDULL, 0xB4B81B3FA97511E2ULL,
-    0xB422061193D6F6A7ULL, 0x071582401C38434DULL, 0x7A13F18BBEDC4FF5ULL,
-    0xBC4097B116C524D2ULL, 0x59B97885E2F2EA28ULL, 0x99170A5DC3115544ULL,
-    0x6F423357E7C6A9F9ULL, 0x325928EE6E6F8794ULL, 0xD0E4366228B03343ULL,
-    0x565C31F7DE89EA27ULL, 0x30F5611484119414ULL, 0xD873DB391292ED4FULL,
-    0x7BD94E1D8E17DEBCULL, 0xC7D9F16864A76E94ULL, 0x947AE053EE56E63CULL,
-    0xC8C93882F9475F5FULL, 0x3A9BF55BA91F81CAULL, 0xD9A11FBB3D9808E4ULL,
-    0x0FD22063EDC29FCAULL, 0xB3F256D8ACA0B0B9ULL, 0xB03031A8B4516E84ULL,
-    0x35DD37D5871448AFULL, 0xE9F6082B05542E4EULL, 0xEBFAFA33D7254B59ULL,
-    0x9255ABB50D532280ULL, 0xB9AB4CE57F2D34F3ULL, 0x693501D628297551ULL,
-    0xC62C58F97DD949BFULL, 0xCD454F8F19C5126AULL, 0xBBE83F4ECC2BDECBULL,
-    0xDC842B7E2819E230ULL, 0xBA89142E007503B8ULL, 0xA3BC941D0A5061CBULL,
-    0xE9F6760E32CD8021ULL, 0x09C7E552BC76492FULL, 0x852F54934DA55CC9ULL,
-    0x8107FCCF064FCF56ULL, 0x098954D51FFF6580ULL, 0x23B70EDB1955C4BFULL,
-    0xC330DE426430F69DULL, 0x4715ED43E8A45C0AULL, 0xA8D7E4DAB780A08DULL,
-    0x0572B974F03CE0BBULL, 0xB57D2E985E1419C7ULL, 0xE8D9ECBE2CF3D73FULL,
-    0x2FE4B17170E59750ULL, 0x11317BA87905E790ULL, 0x7FBF21EC8A1F45ECULL,
-    0x1725CABFCB045B00ULL, 0x964E915CD5E2B207ULL, 0x3E2B8BCBF016D66DULL,
-    0xBE7444E39328A0ACULL, 0xF85B2B4FBCDE44B7ULL, 0x49353FEA39BA63B1ULL,
-    0x1DD01AAFCD53486AULL, 0x1FCA8A92FD719F85ULL, 0xFC7C95D827357AFAULL,
-    0x18A6A990C8B35EBDULL, 0xCCCB7005C6B9C28DULL, 0x3BDBB92C43B17F26ULL,
-    0xAA70B5B4F89695A2ULL, 0xE94C39A54A98307FULL, 0xB7A0B174CFF6F36EULL,
-    0xD4DBA84729AF48ADULL, 0x2E18BC1AD9704A68ULL, 0x2DE0966DAF2F8B1CULL,
-    0xB9C11D5B1E43A07EULL, 0x64972D68DEE33360ULL, 0x94628D38D0C20584ULL,
-    0xDBC0D2B6AB90A559ULL, 0xD2733C4335C6A72FULL, 0x7E75D99D94A70F4DULL,
-    0x6CED1983376FA72BULL, 0x97FCAACBF030BC24ULL, 0x7B77497B32503B12ULL,
-    0x8547EDDFB81CCB94ULL, 0x79999CDFF70902CBULL, 0xCFFE1939438E9B24ULL,
-    0x829626E3892D95D7ULL, 0x92FAE24291F2B3F1ULL, 0x63E22C147B9C3403ULL,
-    0xC678B6D860284A1CULL, 0x5873888850659AE7ULL, 0x0981DCD296A8736DULL,
-    0x9F65789A6509A440ULL, 0x9FF38FED72E9052FULL, 0xE479EE5B9930578CULL,
-    0xE7F28ECD2D49EECDULL, 0x56C074A581EA17FEULL, 0x5544F7D774B14AEFULL,
-    0x7B3F0195FC6F290FULL, 0x12153635B2C0CF57ULL, 0x7F5126DBBA5E0CA7ULL,
-    0x7A76956C3EAFB413ULL, 0x3D5774A11D31AB39ULL, 0x8A1B083821F40CB4ULL,
-    0x7B4A38E32537DF62ULL, 0x950113646D1D6E03ULL, 0x4DA8979A0041E8A9ULL,
-    0x3BC36E078F7515D7ULL, 0x5D0A12F27AD310D1ULL, 0x7F9D1A2E1EBE1327ULL,
-    0xDA3A361B1C5157B1ULL, 0xDCDD7D20903D0C25ULL, 0x36833336D068F707ULL,
-    0xCE68341F79893389ULL, 0xAB9090168DD05F34ULL, 0x43954B3252DC25E5ULL,
-    0xB438C2B67F98E5E9ULL, 0x10DCD78E3851A492ULL, 0xDBC27AB5447822BFULL,
-    0x9B3CDB65F82CA382ULL, 0xB67B7896167B4C84ULL, 0xBFCED1B0048EAC50ULL,
-    0xA9119B60369FFEBDULL, 0x1FFF7AC80904BF45ULL, 0xAC12FB171817EEE7ULL,
-    0xAF08DA9177DDA93DULL, 0x1B0CAB936E65C744ULL, 0xB559EB1D04E5E932ULL,
-    0xC37B45B3F8D6F2BAULL, 0xC3A9DC228CAAC9E9ULL, 0xF3B8B6675A6507FFULL,
-    0x9FC477DE4ED681DAULL, 0x67378D8ECCEF96CBULL, 0x6DD856D94D259236ULL,
-    0xA319CE15B0B4DB31ULL, 0x073973751F12DD5EULL, 0x8A8E849EB32781A5ULL,
-    0xE1925C71285279F5ULL, 0x74C04BF1790C0EFEULL, 0x4DDA48153C94938AULL,
-    0x9D266D6A1CC0542CULL, 0x7440FB816508C4FEULL, 0x13328503DF48229FULL,
-    0xD6BF7BAEE43CAC40ULL, 0x4838D65F6EF6748FULL, 0x1E152328F3318DEAULL,
-    0x8F8419A348F296BFULL, 0x72C8834A5957B511ULL, 0xD7A023A73260B45CULL,
-    0x94EBC8ABCFB56DAEULL, 0x9FC10D0F989993E0ULL, 0xDE68A2355B93CAE6ULL,
-    0xA44CFE79AE538BBEULL, 0x9D1D84FCCE371425ULL, 0x51D2B1AB2DDFB636ULL,
-    0x2FD7E4B9E72CD38CULL, 0x65CA5B96B7552210ULL, 0xDD69A0D8AB3B546DULL,
-    0x604D51B25FBF70E2ULL, 0x73AA8A564FB7AC9EULL, 0x1A8C1E992B941148ULL,
-    0xAAC40A2703D9BEA0ULL, 0x764DBEAE7FA4F3A6ULL, 0x1E99B96E70A9BE8BULL,
-    0x2C5E9DEB57EF4743ULL, 0x3A938FEE32D29981ULL, 0x26E6DB8FFDF5ADFEULL,
-    0x469356C504EC9F9DULL, 0xC8763C5B08D1908CULL, 0x3F6C6AF859D80055ULL,
-    0x7F7CC39420A3A545ULL, 0x9BFB227EBDF4C5CEULL, 0x89039D79D6FC5C5CULL,
-    0x8FE88B57305E2AB6ULL, 0xA09E8C8C35AB96DEULL, 0xFA7E393983325753ULL,
-    0xD6B6D0ECC617C699ULL, 0xDFEA21EA9E7557E3ULL, 0xB67C1FA481680AF8ULL,
-    0xCA1E3785A9E724E5ULL, 0x1CFC8BED0D681639ULL, 0xD18D8549D140CAEAULL,
-    0x4ED0FE7E9DC91335ULL, 0xE4DBF0634473F5D2ULL, 0x1761F93A44D5AEFEULL,
-    0x53898E4C3910DA55ULL, 0x734DE8181F6EC39AULL, 0x2680B122BAA28D97ULL,
-    0x298AF231C85BAFABULL, 0x7983EED3740847D5ULL, 0x66C1A2A1A60CD889ULL,
-    0x9E17E49642A3E4C1ULL, 0xEDB454E7BADC0805ULL, 0x50B704CAB602C329ULL,
-    0x4CC317FB9CDDD023ULL, 0x66B4835D9EAFEA22ULL, 0x219B97E26FFC81BDULL,
-    0x261E4E4C0A333A9DULL, 0x1FE2CCA76517DB90ULL, 0xD7504DFA8816EDBBULL,
-    0xB9571FA04DC089C8ULL, 0x1DDC0325259B27DEULL, 0xCF3F4688801EB9AAULL,
-    0xF4F5D05C10CAB243ULL, 0x38B6525C21A42B0EULL, 0x36F60E2BA4FA6800ULL,
-    0xEB3593803173E0CEULL, 0x9C4CD6257C5A3603ULL, 0xAF0C317D32ADAA8AULL,
-    0x258E5A80C7204C4BULL, 0x8B889D624D44885DULL, 0xF4D14597E660F855ULL,
-    0xD4347F66EC8941C3ULL, 0xE699ED85B0DFB40DULL, 0x2472F6207C2D0484ULL,
-    0xC2A1E7B5B459AEB5ULL, 0xAB4F6451CC1D45ECULL, 0x63767572AE3D6174ULL,
-    0xA59E0BD101731A28ULL, 0x116D0016CB948F09ULL, 0x2CF9C8CA052F6E9FULL,
-    0x0B090A7560A968E3ULL, 0xABEEDDB2DDE06FF1ULL, 0x58EFC10B06A2068DULL,
-    0xC6E57A78FBD986E0ULL, 0x2EAB8CA63CE802D7ULL, 0x14A195640116F336ULL,
-    0x7C0828DD624EC390ULL, 0xD74BBE77E6116AC7ULL, 0x804456AF10F5FB53ULL,
-    0xEBE9EA2ADF4321C7ULL, 0x03219A39EE587A30ULL, 0x49787FEF17AF9924ULL,
-    0xA1E9300CD8520548ULL, 0x5B45E522E4B1B4EFULL, 0xB49C3B3995091A36ULL,
-    0xD4490AD526F14431ULL, 0x12A8F216AF9418C2ULL, 0x001F837CC7350524ULL,
-    0x1877B51E57A764D5ULL, 0xA2853B80F17F58EEULL, 0x993E1DE72D36D310ULL,
-    0xB3598080CE64A656ULL, 0x252F59CF0D9F04BBULL, 0xD23C8E176D113600ULL,
-    0x1BDA0492E7E4586EULL, 0x21E0BD5026C619BFULL, 0x3B097ADAF088F94EULL,
-    0x8D14DEDB30BE846EULL, 0xF95CFFA23AF5F6F4ULL, 0x3871700761B3F743ULL,
-    0xCA672B91E9E4FA16ULL, 0x64C8E531BFF53B55ULL, 0x241260ED4AD1E87DULL,
-    0x106C09B972D2E822ULL, 0x7FBA195410E5CA30ULL, 0x7884D9BC6CB569D8ULL,
-    0x0647DFEDCD894A29ULL, 0x63573FF03E224774ULL, 0x4FC8E9560F91B123ULL,
-    0x1DB956E450275779ULL, 0xB8D91274B9E9D4FBULL, 0xA2EBEE47E2FBFCE1ULL,
-    0xD9F1F30CCD97FB09ULL, 0xEFED53D75FD64E6BULL, 0x2E6D02C36017F67FULL,
-    0xA9AA4D20DB084E9BULL, 0xB64BE8D8B25396C1ULL, 0x70CB6AF7C2D5BCF0ULL,
-    0x98F076A4F7A2322EULL, 0xBF84470805E69B5FULL, 0x94C3251F06F90CF3ULL,
-    0x3E003E616A6591E9ULL, 0xB925A6CD0421AFF3ULL, 0x61BDD1307C66E300ULL,
-    0xBF8D5108E27E0D48ULL, 0x240AB57A8B888B20ULL, 0xFC87614BAF287E07ULL,
-    0xEF02CDD06FFDB432ULL, 0xA1082C0466DF6C0AULL, 0x8215E577001332C8ULL,
-    0xD39BB9C3A48DB6CFULL, 0x2738259634305C14ULL, 0x61CF4F94C97DF93DULL,
-    0x1B6BACA2AE4E125BULL, 0x758F450C88572E0BULL, 0x959F587D507A8359ULL,
-    0xB063E962E045F54DULL, 0x60E8ED72C0DFF5D1ULL, 0x7B64978555326F9FULL,
-    0xFD080D236DA814BAULL, 0x8C90FD9B083F4558ULL, 0x106F72FE81E2C590ULL,
-    0x7976033A39F7D952ULL, 0xA4EC0132764CA04BULL, 0x733EA705FAE4FA77ULL,
-    0xB4D8F77BC3E56167ULL, 0x9E21F4F903B33FD9ULL, 0x9D765E419FB69F6DULL,
-    0xD30C088BA61EA5EFULL, 0x5D94337FBFAF7F5BULL, 0x1A4E4822EB4D7A59ULL,
-    0x6FFE73E81B637FB3ULL, 0xDDF957BC36D8B9CAULL, 0x64D0E29EEA8838B3ULL,
-    0x08DD9BDFD96B9F63ULL, 0x087E79E5A57D1D13ULL, 0xE328E230E3E2B3FBULL,
-    0x1C2559E30F0946BEULL, 0x720BF5F26F4D2EAAULL, 0xB0774D261CC609DBULL,
-    0x443F64EC5A371195ULL, 0x4112CF68649A260EULL, 0xD813F2FAB7F5C5CAULL,
-    0x660D3257380841EEULL, 0x59AC2C7873F910A3ULL, 0xE846963877671A17ULL,
-    0x93B633ABFA3469F8ULL, 0xC0C0F5A60EF4CDCFULL, 0xCAF21ECD4377B28CULL,
-    0x57277707199B8175ULL, 0x506C11B9D90E8B1DULL, 0xD83CC2687A19255FULL,
-    0x4A29C6465A314CD1ULL, 0xED2DF21216235097ULL, 0xB5635C95FF7296E2ULL,
-    0x22AF003AB672E811ULL, 0x52E762596BF68235ULL, 0x9AEBA33AC6ECC6B0ULL,
-    0x944F6DE09134DFB6ULL, 0x6C47BEC883A7DE39ULL, 0x6AD047C430A12104ULL,
-    0xA5B1CFDBA0AB4067ULL, 0x7C45D833AFF07862ULL, 0x5092EF950A16DA0BULL,
-    0x9338E69C052B8E7BULL, 0x455A4B4CFE30E3F5ULL, 0x6B02E63195AD0CF8ULL,
-    0x6B17B224BAD6BF27ULL, 0xD1E0CCD25BB9C169ULL, 0xDE0C89A556B9AE70ULL,
-    0x50065E535A213CF6ULL, 0x9C1169FA2777B874ULL, 0x78EDEFD694AF1EEDULL,
-    0x6DC93D9526A50E68ULL, 0xEE97F453F06791EDULL, 0x32AB0EDB696703D3ULL,
-    0x3A6853C7E70757A7ULL, 0x31865CED6120F37DULL, 0x67FEF95D92607890ULL,
-    0x1F2B1D1F15F6DC9CULL, 0xB69E38A8965C6B65ULL, 0xAA9119FF184CCCF4ULL,
-    0xF43C732873F24C13ULL, 0xFB4A3D794A9A80D2ULL, 0x3550C2321FD6109CULL,
-    0x371F77E76BB8417EULL, 0x6BFA9AAE5EC05779ULL, 0xCD04F3FF001A4778ULL,
-    0xE3273522064480CAULL, 0x9F91508BFFCFC14AULL, 0x049A7F41061A9E60ULL,
-    0xFCB6BE43A9F2FE9BULL, 0x08DE8A1C7797DA9BULL, 0x8F9887E6078735A1ULL,
-    0xB5B4071DBFC73A66ULL, 0x230E343DFBA08D33ULL, 0x43ED7F5A0FAE657DULL,
-    0x3A88A0FBBCB05C63ULL, 0x21874B8B4D2DBC4FULL, 0x1BDEA12E35F6A8C9ULL,
-    0x53C065C6C8E63528ULL, 0xE34A1D250E7A8D6BULL, 0xD6B04D3B7651DD7EULL,
-    0x5E90277E7CB39E2DULL, 0x2C046F22062DC67DULL, 0xB10BB459132D0A26ULL,
-    0x3FA9DDFB67E2F199ULL, 0x0E09B88E1914F7AFULL, 0x10E8B35AF3EEAB37ULL,
-    0x9EEDECA8E272B933ULL, 0xD4C718BC4AE8AE5FULL, 0x81536D601170FC20ULL,
-    0x91B534F885818A06ULL, 0xEC8177F83F900978ULL, 0x190E714FADA5156EULL,
-    0xB592BF39B0364963ULL, 0x89C350C893AE7DC1ULL, 0xAC042E70F8B383F2ULL,
-    0xB49B52E587A1EE60ULL, 0xFB152FE3FF26DA89ULL, 0x3E666E6F69AE2C15ULL,
-    0x3B544EBE544C19F9ULL, 0xE805A1E290CF2456ULL, 0x24B33C9D7ED25117ULL,
-    0xE74733427B72F0C1ULL, 0x0A804D18B7097475ULL, 0x57E3306D881EDB4FULL,
-    0x4AE7D6A36EB5DBCBULL, 0x2D8D5432157064C8ULL, 0xD1E649DE1E7F268BULL,
-    0x8A328A1CEDFE552CULL, 0x07A3AEC79624C7DAULL, 0x84547DDC3E203C94ULL,
-    0x990A98FD5071D263ULL, 0x1A4FF12616EEFC89ULL, 0xF6F7FD1431714200ULL,
-    0x30C05B1BA332F41CULL, 0x8D2636B81555A786ULL, 0x46C9FEB55D120902ULL,
-    0xCCEC0A73B49C9921ULL, 0x4E9D2827355FC492ULL, 0x19EBB029435DCB0FULL,
-    0x4659D2B743848A2CULL, 0x963EF2C96B33BE31ULL, 0x74F85198B05A2E7DULL,
-    0x5A0F544DD2B1FB18ULL, 0x03727073C2E134B1ULL, 0xC7F6AA2DE59AEA61ULL,
-    0x352787BAA0D7C22FULL, 0x9853EAB63B5E0B35ULL, 0xABBDCDD7ED5C0860ULL,
-    0xCF05DAF5AC8D77B0ULL, 0x49CAD48CEBF4A71EULL, 0x7A4C10EC2158C4A6ULL,
-    0xD9E92AA246BF719EULL, 0x13AE978D09FE5557ULL, 0x730499AF921549FFULL,
-    0x4E4B705B92903BA4ULL, 0xFF577222C14F0A3AULL, 0x55B6344CF97AAFAEULL,
-    0xB862225B055B6960ULL, 0xCAC09AFBDDD2CDB4ULL, 0xDAF8E9829FE96B5FULL,
-    0xB5FDFC5D3132C498ULL, 0x310CB380DB6F7503ULL, 0xE87FBB46217A360EULL,
-    0x2102AE466EBB1148ULL, 0xF8549E1A3AA5E00DULL, 0x07A69AFDCC42261AULL,
-    0xC4C118BFE78FEAAEULL, 0xF9F4892ED96BD438ULL, 0x1AF3DBE25D8F45DAULL,
-    0xF5B4B0B0D2DEEEB4ULL, 0x962ACEEFA82E1C84ULL, 0x046E3ECAAF453CE9ULL,
-    0xF05D129681949A4CULL, 0x964781CE734B3C84ULL, 0x9C2ED44081CE5FBDULL,
-    0x522E23F3925E319EULL, 0x177E00F9FC32F791ULL, 0x2BC60A63A6F3B3F2ULL,
-    0x222BBFAE61725606ULL, 0x486289DDCC3D6780ULL, 0x7DC7785B8EFDFC80ULL,
-    0x8AF38731C02BA980ULL, 0x1FAB64EA29A2DDF7ULL, 0xE4D9429322CD065AULL,
-    0x9DA058C67844F20CULL, 0x24C0E332B70019B0ULL, 0x233003B5A6CFE6ADULL,
-    0xD586BD01C5C217F6ULL, 0x5E5637885F29BC2BULL, 0x7EBA726D8C94094BULL,
-    0x0A56A5F0BFE39272ULL, 0xD79476A84EE20D06ULL, 0x9E4C1269BAA4BF37ULL,
-    0x17EFEE45B0DEE640ULL, 0x1D95B0A5FCF90BC6ULL, 0x93CBE0B699C2585DULL,
-    0x65FA4F227A2B6D79ULL, 0xD5F9E858292504D5ULL, 0xC2B5A03F71471A6FULL,
-    0x59300222B4561E00ULL, 0xCE2F8642CA0712DCULL, 0x7CA9723FBB2E8988ULL,
-    0x2785338347F2BA08ULL, 0xC61BB3A141E50E8CULL, 0x150F361DAB9DEC26ULL,
-    0x9F6A419D382595F4ULL, 0x64A53DC924FE7AC9ULL, 0x142DE49FFF7A7C3DULL,
-    0x0C335248857FA9E7ULL, 0x0A9C32D5EAE45305ULL, 0xE6C42178C4BBB92EULL,
-    0x71F1CE2490D20B07ULL, 0xF1BCC3D275AFE51AULL, 0xE728E8C83C334074ULL,
-    0x96FBF83A12884624ULL, 0x81A1549FD6573DA5ULL, 0x5FA7867CAF35E149ULL,
-    0x56986E2EF3ED091BULL, 0x917F1DD5F8886C61ULL, 0xD20D8C88C8FFE65FULL,
-    0x31D71DCE64B2C310ULL, 0xF165B587DF898190ULL, 0xA57E6339DD2CF3A0ULL,
-    0x1EF6E6DBB1961EC9ULL, 0x70CC73D90BC26E24ULL, 0xE21A6B35DF0C3AD7ULL,
-    0x003A93D8B2806962ULL, 0x1C99DED33CB890A1ULL, 0xCF3145DE0ADD4289ULL,
-    0xD0E4427A5514FB72ULL, 0x77C621CC9FB3A483ULL, 0x67A34DAC4356550BULL,
-    0xF8D626AAAF278509ULL
-  };
-
-  // Offsets to the PolyGlotRandoms[] array of zobrist keys
-  const Key* ZobPiece     = PolyGlotRandoms;
-  const Key* ZobCastle    = ZobPiece     + 12 * 64; // Pieces * squares
-  const Key* ZobEnPassant = ZobCastle    + 4;       // Castle flags
-  const Key* ZobTurn      = ZobEnPassant + 8;       // Number of files
-
-  // book_key() returns the PolyGlot hash key of the given position
-  uint64_t book_key(const Position& pos) {
-
-    uint64_t key = 0;
-    Bitboard b = pos.pieces();
-
-    while (b)
-    {
-        // In PolyGlotRandoms[] pieces are stored in the following sequence:
-        // BP = 0, WP = 1, BN = 2, WN = 3, ... BK = 10, WK = 11
-        Square s = pop_lsb(&b);
-        Piece p = pos.piece_on(s);
-        int pieceOfs = 2 * (type_of(p) - 1) + (color_of(p) == WHITE);
-        key ^= ZobPiece[64 * pieceOfs + s];
-    }
-
-    b = pos.can_castle(ALL_CASTLES);
-
-    while (b)
-        key ^= ZobCastle[pop_lsb(&b)];
-
-    if (pos.ep_square() != SQ_NONE)
-        key ^= ZobEnPassant[file_of(pos.ep_square())];
-
-    if (pos.side_to_move() == WHITE)
-        key ^= ZobTurn[0];
-
-    return key;
-  }
-
-} // namespace
-
-PolyglotBook::PolyglotBook() {
-
-  for (int i = Time::now() % 10000; i > 0; i--)
-      RKiss.rand<unsigned>(); // Make random number generation less deterministic
-}
-
-PolyglotBook::~PolyglotBook() { if (is_open()) close(); }
-
-
-/// operator>>() reads sizeof(T) chars from the file's binary byte stream and
-/// converts them in a number of type T. A Polyglot book stores numbers in
-/// big-endian format.
-
-template<typename T> PolyglotBook& PolyglotBook::operator>>(T& n) {
-
-  n = 0;
-  for (size_t i = 0; i < sizeof(T); i++)
-      n = T((n << 8) + ifstream::get());
-
-  return *this;
-}
-
-template<> PolyglotBook& PolyglotBook::operator>>(BookEntry& e) {
-  return *this >> e.key >> e.move >> e.count >> e.learn;
-}
-
-
-/// open() tries to open a book file with the given name after closing any
-/// exsisting one.
-
-bool PolyglotBook::open(const char* fName) {
-
-  if (is_open()) // Cannot close an already closed file
-      close();
-
-  ifstream::open(fName, ifstream::in | ifstream::binary);
-
-  fileName = is_open() ? fName : "";
-  ifstream::clear(); // Reset any error flag to allow retry ifstream::open()
-  return !fileName.empty();
-}
-
-
-/// probe() tries to find a book move for the given position. If no move is
-/// found returns MOVE_NONE. If pickBest is true returns always the highest
-/// rated move, otherwise randomly chooses one, based on the move score.
-
-Move PolyglotBook::probe(const Position& pos, const string& fName, bool pickBest) {
-
-  if (fileName != fName && !open(fName.c_str()))
-      return MOVE_NONE;
-
-  BookEntry e;
-  uint16_t best = 0;
-  unsigned sum = 0;
-  Move move = MOVE_NONE;
-  uint64_t key = book_key(pos);
-
-  seekg(find_first(key) * sizeof(BookEntry), ios_base::beg);
-
-  while (*this >> e, e.key == key && good())
-  {
-      best = max(best, e.count);
-      sum += e.count;
-
-      // Choose book move according to its score. If a move has a very
-      // high score it has higher probability to be choosen than a move
-      // with lower score. Note that first entry is always chosen.
-      if (   (sum && RKiss.rand<unsigned>() % sum < e.count)
-          || (pickBest && e.count == best))
-          move = Move(e.move);
-  }
-
-  if (!move)
-      return MOVE_NONE;
-
-  // A PolyGlot book move is encoded as follows:
-  //
-  // bit  0- 5: destination square (from 0 to 63)
-  // bit  6-11: origin square (from 0 to 63)
-  // bit 12-14: promotion piece (from KNIGHT == 1 to QUEEN == 4)
-  //
-  // Castling moves follow "king captures rook" representation. So in case book
-  // move is a promotion we have to convert to our representation, in all the
-  // other cases we can directly compare with a Move after having masked out
-  // the special Move's flags (bit 14-15) that are not supported by PolyGlot.
-  int pt = (move >> 12) & 7;
-  if (pt)
-      move = make<PROMOTION>(from_sq(move), to_sq(move), PieceType(pt + 1));
-
-  // Add 'special move' flags and verify it is legal
-  for (MoveList<LEGAL> ml(pos); !ml.end(); ++ml)
-      if (move == (ml.move() & 0x3FFF))
-          return ml.move();
-
-  return MOVE_NONE;
-}
-
-
-/// find_first() takes a book key as input, and does a binary search through
-/// the book file for the given key. Returns the index of the leftmost book
-/// entry with the same key as the input.
-
-size_t PolyglotBook::find_first(uint64_t key) {
-
-  seekg(0, ios::end); // Move pointer to end, so tellg() gets file's size
-
-  size_t low = 0, mid, high = (size_t)tellg() / sizeof(BookEntry) - 1;
-  BookEntry e;
-
-  assert(low <= high);
-
-  while (low < high && good())
-  {
-      mid = (low + high) / 2;
-
-      assert(mid >= low && mid < high);
-
-      seekg(mid * sizeof(BookEntry), ios_base::beg);
-      *this >> e;
-
-      if (key <= e.key)
-          high = mid;
-      else
-          low = mid + 1;
-  }
-
-  assert(low == high);
-
-  return low;
-}

src/endgame.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,104 +16,111 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(ENDGAME_H_INCLUDED)
+#ifndef ENDGAME_H_INCLUDED
 #define ENDGAME_H_INCLUDED
 
-#include <map>
+#include <memory>
 #include <string>
+#include <type_traits>
+#include <unordered_map>
+#include <utility>
 
 #include "position.h"
 #include "types.h"
 
+namespace Stockfish {
 
-/// EndgameType lists all supported endgames
+/// EndgameCode lists all supported endgame functions by corresponding codes
 
-enum EndgameType {
-
-  // Evaluation functions
+enum EndgameCode {
 
+  EVALUATION_FUNCTIONS,
+  KNNK,  // KNN vs K
+  KNNKP, // KNN vs KP
   KXK,   // Generic "mate lone king" eval
   KBNK,  // KBN vs K
   KPK,   // KP vs K
   KRKP,  // KR vs KP
   KRKB,  // KR vs KB
   KRKN,  // KR vs KN
+  KQKP,  // KQ vs KP
   KQKR,  // KQ vs KR
-  KBBKN, // KBB vs KN
-  KNNK,  // KNN vs K
-  KmmKm, // K and two minors vs K and one or two minors
 
-
-  // Scaling functions
-  SCALE_FUNS,
-
-  KBPsK,   // KB+pawns vs K
-  KQKRPs,  // KQ vs KR+pawns
+  SCALING_FUNCTIONS,
+  KBPsK,   // KB and pawns vs K
+  KQKRPs,  // KQ vs KR and pawns
   KRPKR,   // KRP vs KR
+  KRPKB,   // KRP vs KB
   KRPPKRP, // KRPP vs KRP
-  KPsK,    // King and pawns vs king
+  KPsK,    // K and pawns vs K
   KBPKB,   // KBP vs KB
   KBPPKB,  // KBPP vs KB
   KBPKN,   // KBP vs KN
-  KNPK,    // KNP vs K
   KPKP     // KP vs KP
 };
 
 
-/// Endgame functions can be of two types according if return a Value or a
-/// ScaleFactor. Type eg_fun<int>::type equals to either ScaleFactor or Value
-/// depending if the template parameter is 0 or 1.
+/// Endgame functions can be of two types depending on whether they return a
+/// Value or a ScaleFactor.
 
-template<int> struct eg_fun { typedef Value type; };
-template<> struct eg_fun<1> { typedef ScaleFactor type; };
+template<EndgameCode E> using
+eg_type = typename std::conditional<(E < SCALING_FUNCTIONS), Value, ScaleFactor>::type;
 
 
-/// Base and derived templates for endgame evaluation and scaling functions
+/// Base and derived functors for endgame evaluation and scaling functions
 
 template<typename T>
 struct EndgameBase {
 
-  virtual ~EndgameBase() {}
-  virtual Color color() const = 0;
+  explicit EndgameBase(Color c) : strongSide(c), weakSide(~c) {}
+  virtual ~EndgameBase() = default;
   virtual T operator()(const Position&) const = 0;
+
+  const Color strongSide, weakSide;
 };
 
 
-template<EndgameType E, typename T = typename eg_fun<(E > SCALE_FUNS)>::type>
+template<EndgameCode E, typename T = eg_type<E>>
 struct Endgame : public EndgameBase<T> {
 
-  explicit Endgame(Color c) : strongerSide(c), weakerSide(~c) {}
-  Color color() const { return strongerSide; }
-  T operator()(const Position&) const;
-
-private:
-  Color strongerSide, weakerSide;
+  explicit Endgame(Color c) : EndgameBase<T>(c) {}
+  T operator()(const Position&) const override;
 };
 
 
-/// Endgames class stores in two std::map the pointers to endgame evaluation
-/// and scaling base objects. Then we use polymorphism to invoke the actual
-/// endgame function calling its operator() that is virtual.
+/// The Endgames namespace handles the pointers to endgame evaluation and scaling
+/// base objects in two std::map. We use polymorphism to invoke the actual
+/// endgame function by calling its virtual operator().
 
-class Endgames {
+namespace Endgames {
 
-  typedef std::map<Key, EndgameBase<eg_fun<0>::type>*> M1;
-  typedef std::map<Key, EndgameBase<eg_fun<1>::type>*> M2;
+  template<typename T> using Ptr = std::unique_ptr<EndgameBase<T>>;
+  template<typename T> using Map = std::unordered_map<Key, Ptr<T>>;
 
-  M1 m1;
-  M2 m2;
+  extern std::pair<Map<Value>, Map<ScaleFactor>> maps;
 
-  M1& map(M1::mapped_type) { return m1; }
-  M2& map(M2::mapped_type) { return m2; }
+  void init();
 
-  template<EndgameType E> void add(const std::string& code);
+  template<typename T>
+  Map<T>& map() {
+    return std::get<std::is_same<T, ScaleFactor>::value>(maps);
+  }
 
-public:
-  Endgames();
-  ~Endgames();
+  template<EndgameCode E, typename T = eg_type<E>>
+  void add(const std::string& code) {
 
-  template<typename T> T probe(Key key, T& eg)
-  { return eg = map(eg).count(key) ? map(eg)[key] : NULL; }
-};
+    StateInfo st;
+    map<T>()[Position().set(code, WHITE, &st).material_key()] = Ptr<T>(new Endgame<E>(WHITE));
+    map<T>()[Position().set(code, BLACK, &st).material_key()] = Ptr<T>(new Endgame<E>(BLACK));
+  }
 
-#endif // !defined(ENDGAME_H_INCLUDED)
+  template<typename T>
+  const EndgameBase<T>* probe(Key key) {
+    auto it = map<T>().find(key);
+    return it != map<T>().end() ? it->second.get() : nullptr;
+  }
+}
+
+} // namespace Stockfish
+
+#endif // #ifndef ENDGAME_H_INCLUDED

src/evaluate.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,21 +16,40 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(EVALUATE_H_INCLUDED)
+#ifndef EVALUATE_H_INCLUDED
 #define EVALUATE_H_INCLUDED
 
+#include <string>
+#include <optional>
+
 #include "types.h"
 
+namespace Stockfish {
+
 class Position;
 
 namespace Eval {
 
-extern Color RootColor;
+  std::string trace(Position& pos);
+  Value evaluate(const Position& pos);
 
-extern void init();
-extern Value evaluate(const Position& pos, Value& margin);
-extern std::string trace(const Position& pos);
+  extern bool useNNUE;
+  extern std::string currentEvalFileName;
 
-}
+  // 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-5af11540bbfe.nnue"
 
-#endif // !defined(EVALUATE_H_INCLUDED)
+  namespace NNUE {
+
+    void init();
+    void verify();
+
+  } // namespace NNUE
+
+} // namespace Eval
+
+} // namespace Stockfish
+
+#endif // #ifndef EVALUATE_H_INCLUDED

src/history.h

@@ -1,72 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
-
-  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/>.
-*/
-
-#if !defined(HISTORY_H_INCLUDED)
-#define HISTORY_H_INCLUDED
-
-#include <algorithm>
-#include <cstring>
-
-#include "types.h"
-
-/// The History class stores statistics about how often different moves
-/// have been successful or unsuccessful during the current search. These
-/// statistics are used for reduction and move ordering decisions. History
-/// entries are stored according only to moving piece and destination square,
-/// in particular two moves with different origin but same destination and
-/// same piece will be considered identical.
-
-class History {
-
-public:
-  void clear();
-  Value value(Piece p, Square to) const;
-  void add(Piece p, Square to, Value bonus);
-  Value gain(Piece p, Square to) const;
-  void update_gain(Piece p, Square to, Value g);
-
-  static const Value MaxValue = Value(2000);
-
-private:
-  Value history[16][64];  // [piece][to_square]
-  Value maxGains[16][64]; // [piece][to_square]
-};
-
-inline void History::clear() {
-  memset(history,  0, 16 * 64 * sizeof(Value));
-  memset(maxGains, 0, 16 * 64 * sizeof(Value));
-}
-
-inline Value History::value(Piece p, Square to) const {
-  return history[p][to];
-}
-
-inline void History::add(Piece p, Square to, Value bonus) {
-  if (abs(history[p][to] + bonus) < MaxValue) history[p][to] += bonus;
-}
-
-inline Value History::gain(Piece p, Square to) const {
-  return maxGains[p][to];
-}
-
-inline void History::update_gain(Piece p, Square to, Value g) {
-  maxGains[p][to] = std::max(g, maxGains[p][to] - 1);
-}
-
-#endif // !defined(HISTORY_H_INCLUDED)

src/incbin/UNLICENCE

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

src/incbin/incbin.h

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

src/main.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -18,35 +17,37 @@
 */
 
 #include <iostream>
-#include <string>
 
 #include "bitboard.h"
-#include "evaluate.h"
+#include "endgame.h"
 #include "position.h"
+#include "psqt.h"
 #include "search.h"
+#include "syzygy/tbprobe.h"
 #include "thread.h"
 #include "tt.h"
-#include "ucioption.h"
+#include "uci.h"
+
+using namespace Stockfish;
 
 int main(int argc, char* argv[]) {
 
   std::cout << engine_info() << std::endl;
 
+  CommandLine::init(argc, argv);
   UCI::init(Options);
+  Tune::init();
+  PSQT::init();
   Bitboards::init();
-  Zobrist::init();
-  Bitbases::init_kpk();
-  Search::init();
-  Eval::init();
-  Threads.init();
-  TT.set_size(Options["Hash"]);
+  Position::init();
+  Bitbases::init();
+  Endgames::init();
+  Threads.set(size_t(Options["Threads"]));
+  Search::clear(); // After threads are up
+  Eval::NNUE::init();
 
-  std::string args;
+  UCI::loop(argc, argv);
 
-  for (int i = 1; i < argc; i++)
-      args += std::string(argv[i]) + " ";
-
-  UCI::loop(args);
-
-  Threads.exit();
+  Threads.set(0);
+  return 0;
 }

src/material.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,171 +16,182 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#include <algorithm>
 #include <cassert>
-#include <cstring>
+#include <cstring>   // For std::memset
 
 #include "material.h"
+#include "thread.h"
 
 using namespace std;
 
+namespace Stockfish {
+
 namespace {
+  #define S(mg, eg) make_score(mg, eg)
 
-  // Values modified by Joona Kiiski
-  const Value MidgameLimit = Value(15581);
-  const Value EndgameLimit = Value(3998);
+  // Polynomial material imbalance parameters
 
-  // Scale factors used when one side has no more pawns
-  const int NoPawnsSF[4] = { 6, 12, 32 };
+  // 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
+  };
 
-  // Polynomial material balance parameters
-  const Value RedundantQueenPenalty = Value(320);
-  const Value RedundantRookPenalty  = Value(554);
+  // 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
+  };
 
-  const int LinearCoefficients[6] = { 1617, -162, -1172, -190, 105, 26 };
+  #undef S
 
-  const int QuadraticCoefficientsSameColor[][8] = {
-  { 7, 7, 7, 7, 7, 7 }, { 39, 2, 7, 7, 7, 7 }, { 35, 271, -4, 7, 7, 7 },
-  { 7, 25, 4, 7, 7, 7 }, { -27, -2, 46, 100, 56, 7 }, { 58, 29, 83, 148, -3, -25 } };
-
-  const int QuadraticCoefficientsOppositeColor[][8] = {
-  { 41, 41, 41, 41, 41, 41 }, { 37, 41, 41, 41, 41, 41 }, { 10, 62, 41, 41, 41, 41 },
-  { 57, 64, 39, 41, 41, 41 }, { 50, 40, 23, -22, 41, 41 }, { 106, 101, 3, 151, 171, 41 } };
-
-  // Endgame evaluation and scaling functions accessed direcly and not through
-  // the function maps because correspond to more then one material hash key.
-  Endgame<KmmKm> EvaluateKmmKm[] = { Endgame<KmmKm>(WHITE), Endgame<KmmKm>(BLACK) };
-  Endgame<KXK>   EvaluateKXK[]   = { Endgame<KXK>(WHITE),   Endgame<KXK>(BLACK) };
+  // 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) };
 
   Endgame<KBPsK>  ScaleKBPsK[]  = { Endgame<KBPsK>(WHITE),  Endgame<KBPsK>(BLACK) };
   Endgame<KQKRPs> ScaleKQKRPs[] = { Endgame<KQKRPs>(WHITE), Endgame<KQKRPs>(BLACK) };
   Endgame<KPsK>   ScaleKPsK[]   = { Endgame<KPsK>(WHITE),   Endgame<KPsK>(BLACK) };
   Endgame<KPKP>   ScaleKPKP[]   = { Endgame<KPKP>(WHITE),   Endgame<KPKP>(BLACK) };
 
-  // Helper templates used to detect a given material distribution
-  template<Color Us> bool is_KXK(const Position& pos) {
-    const Color Them = (Us == WHITE ? BLACK : WHITE);
-    return   pos.non_pawn_material(Them) == VALUE_ZERO
-          && pos.piece_count(Them, PAWN) == 0
-          && pos.non_pawn_material(Us)   >= RookValueMg;
+  // Helper used to detect a given material distribution
+  bool is_KXK(const Position& pos, Color us) {
+    return  !more_than_one(pos.pieces(~us))
+          && pos.non_pawn_material(us) >= RookValueMg;
   }
 
-  template<Color Us> bool is_KBPsKs(const Position& pos) {
-    return   pos.non_pawn_material(Us)   == BishopValueMg
-          && pos.piece_count(Us, BISHOP) == 1
-          && pos.piece_count(Us, PAWN)   >= 1;
+  bool is_KBPsK(const Position& pos, Color us) {
+    return   pos.non_pawn_material(us) == BishopValueMg
+          && pos.count<PAWN>(us) >= 1;
   }
 
-  template<Color Us> bool is_KQKRPs(const Position& pos) {
-    const Color Them = (Us == WHITE ? BLACK : WHITE);
-    return   pos.piece_count(Us, PAWN)    == 0
-          && pos.non_pawn_material(Us)    == QueenValueMg
-          && pos.piece_count(Us, QUEEN)   == 1
-          && pos.piece_count(Them, ROOK)  == 1
-          && pos.piece_count(Them, PAWN)  >= 1;
+  bool is_KQKRPs(const Position& pos, Color us) {
+    return  !pos.count<PAWN>(us)
+          && pos.non_pawn_material(us) == QueenValueMg
+          && pos.count<ROOK>(~us) == 1
+          && pos.count<PAWN>(~us) >= 1;
+  }
+
+
+  /// imbalance() calculates the imbalance by comparing the piece count of each
+  /// piece type for both colors.
+
+  template<Color Us>
+  Score imbalance(const int pieceCount[][PIECE_TYPE_NB]) {
+
+    constexpr Color Them = ~Us;
+
+    Score bonus = SCORE_ZERO;
+
+    // Second-degree polynomial material imbalance, by Tord Romstad
+    for (int pt1 = NO_PIECE_TYPE; pt1 <= QUEEN; ++pt1)
+    {
+        if (!pieceCount[Us][pt1])
+            continue;
+
+        int v = QuadraticOurs[pt1][pt1] * pieceCount[Us][pt1];
+
+        for (int pt2 = NO_PIECE_TYPE; pt2 < pt1; ++pt2)
+            v +=  QuadraticOurs[pt1][pt2] * pieceCount[Us][pt2]
+                + QuadraticTheirs[pt1][pt2] * pieceCount[Them][pt2];
+
+        bonus += pieceCount[Us][pt1] * v;
+    }
+
+    return bonus;
   }
 
 } // namespace
 
+namespace Material {
 
-/// MaterialTable::probe() takes a position object as input, looks up a MaterialEntry
-/// object, and returns a pointer to it. If the material configuration is not
-/// already present in the table, it is computed and stored there, so we don't
-/// have to recompute everything when the same material configuration occurs again.
 
-MaterialEntry* MaterialTable::probe(const Position& pos) {
+/// Material::probe() looks up the current position's material configuration in
+/// the material hash table. It returns a pointer to the Entry if the position
+/// is found. Otherwise a new Entry is computed and stored there, so we don't
+/// have to recompute all when the same material configuration occurs again.
+
+Entry* probe(const Position& pos) {
 
   Key key = pos.material_key();
-  MaterialEntry* e = entries[key];
+  Entry* e = pos.this_thread()->materialTable[key];
 
-  // If e->key matches the position's material hash key, it means that we
-  // have analysed this material configuration before, and we can simply
-  // return the information we found the last time instead of recomputing it.
   if (e->key == key)
       return e;
 
-  memset(e, 0, sizeof(MaterialEntry));
+  std::memset(e, 0, sizeof(Entry));
   e->key = key;
   e->factor[WHITE] = e->factor[BLACK] = (uint8_t)SCALE_FACTOR_NORMAL;
-  e->gamePhase = MaterialTable::game_phase(pos);
-
-  // Let's look if we have a specialized evaluation function for this
-  // particular material configuration. First we look for a fixed
-  // configuration one, then a generic one if previous search failed.
-  if (endgames.probe(key, e->evaluationFunction))
-      return e;
-
-  if (is_KXK<WHITE>(pos))
-  {
-      e->evaluationFunction = &EvaluateKXK[WHITE];
-      return e;
-  }
-
-  if (is_KXK<BLACK>(pos))
-  {
-      e->evaluationFunction = &EvaluateKXK[BLACK];
-      return e;
-  }
-
-  if (!pos.pieces(PAWN) && !pos.pieces(ROOK) && !pos.pieces(QUEEN))
-  {
-      // Minor piece endgame with at least one minor piece per side and
-      // no pawns. Note that the case KmmK is already handled by KXK.
-      assert((pos.pieces(WHITE, KNIGHT) | pos.pieces(WHITE, BISHOP)));
-      assert((pos.pieces(BLACK, KNIGHT) | pos.pieces(BLACK, BISHOP)));
-
-      if (   pos.piece_count(WHITE, BISHOP) + pos.piece_count(WHITE, KNIGHT) <= 2
-          && pos.piece_count(BLACK, BISHOP) + pos.piece_count(BLACK, KNIGHT) <= 2)
-      {
-          e->evaluationFunction = &EvaluateKmmKm[pos.side_to_move()];
-          return e;
-      }
-  }
-
-  // OK, we didn't find any special evaluation function for the current
-  // material configuration. Is there a suitable scaling function?
-  //
-  // We face problems when there are several conflicting applicable
-  // scaling functions and we need to decide which one to use.
-  EndgameBase<ScaleFactor>* sf;
-
-  if (endgames.probe(key, sf))
-  {
-      e->scalingFunction[sf->color()] = sf;
-      return e;
-  }
-
-  // Generic scaling functions that refer to more then one material
-  // distribution. Should be probed after the specialized ones.
-  // Note that these ones don't return after setting the function.
-  if (is_KBPsKs<WHITE>(pos))
-      e->scalingFunction[WHITE] = &ScaleKBPsK[WHITE];
-
-  if (is_KBPsKs<BLACK>(pos))
-      e->scalingFunction[BLACK] = &ScaleKBPsK[BLACK];
-
-  if (is_KQKRPs<WHITE>(pos))
-      e->scalingFunction[WHITE] = &ScaleKQKRPs[WHITE];
-
-  else if (is_KQKRPs<BLACK>(pos))
-      e->scalingFunction[BLACK] = &ScaleKQKRPs[BLACK];
 
   Value npm_w = pos.non_pawn_material(WHITE);
   Value npm_b = pos.non_pawn_material(BLACK);
+  Value npm   = std::clamp(npm_w + npm_b, EndgameLimit, MidgameLimit);
 
-  if (npm_w + npm_b == VALUE_ZERO)
-  {
-      if (pos.piece_count(BLACK, PAWN) == 0)
+  // Map total non-pawn material into [PHASE_ENDGAME, PHASE_MIDGAME]
+  e->gamePhase = Phase(((npm - EndgameLimit) * PHASE_MIDGAME) / (MidgameLimit - EndgameLimit));
+
+  // Let's look if we have a specialized evaluation function for this particular
+  // material configuration. Firstly we look for a fixed configuration one, then
+  // for a generic one if the previous search failed.
+  if ((e->evaluationFunction = Endgames::probe<Value>(key)) != nullptr)
+      return e;
+
+  for (Color c : { WHITE, BLACK })
+      if (is_KXK(pos, c))
       {
-          assert(pos.piece_count(WHITE, PAWN) >= 2);
+          e->evaluationFunction = &EvaluateKXK[c];
+          return e;
+      }
+
+  // OK, we didn't find any special evaluation function for the current material
+  // configuration. Is there a suitable specialized scaling function?
+  const auto* sf = Endgames::probe<ScaleFactor>(key);
+
+  if (sf)
+  {
+      e->scalingFunction[sf->strongSide] = sf; // Only strong color assigned
+      return e;
+  }
+
+  // We didn't find any specialized scaling function, so fall back on generic
+  // ones that refer to more than one material distribution. Note that in this
+  // case we don't return after setting the function.
+  for (Color c : { WHITE, BLACK })
+  {
+    if (is_KBPsK(pos, c))
+        e->scalingFunction[c] = &ScaleKBPsK[c];
+
+    else if (is_KQKRPs(pos, c))
+        e->scalingFunction[c] = &ScaleKQKRPs[c];
+  }
+
+  if (npm_w + npm_b == VALUE_ZERO && pos.pieces(PAWN)) // Only pawns on the board
+  {
+      if (!pos.count<PAWN>(BLACK))
+      {
+          assert(pos.count<PAWN>(WHITE) >= 2);
+
           e->scalingFunction[WHITE] = &ScaleKPsK[WHITE];
       }
-      else if (pos.piece_count(WHITE, PAWN) == 0)
+      else if (!pos.count<PAWN>(WHITE))
       {
-          assert(pos.piece_count(BLACK, PAWN) >= 2);
+          assert(pos.count<PAWN>(BLACK) >= 2);
+
           e->scalingFunction[BLACK] = &ScaleKPsK[BLACK];
       }
-      else if (pos.piece_count(WHITE, PAWN) == 1 && pos.piece_count(BLACK, PAWN) == 1)
+      else if (pos.count<PAWN>(WHITE) == 1 && pos.count<PAWN>(BLACK) == 1)
       {
           // This is a special case because we set scaling functions
           // for both colors instead of only one.
@@ -190,87 +200,30 @@ MaterialEntry* MaterialTable::probe(const Position& pos) {
       }
   }
 
-  // No pawns makes it difficult to win, even with a material advantage
-  if (pos.piece_count(WHITE, PAWN) == 0 && npm_w - npm_b <= BishopValueMg)
-  {
-      e->factor[WHITE] = (uint8_t)
-      (npm_w == npm_b || npm_w < RookValueMg ? 0 : NoPawnsSF[std::min(pos.piece_count(WHITE, BISHOP), 2)]);
-  }
+  // Zero or just one pawn makes it difficult to win, even with a small material
+  // advantage. This catches some trivial draws like KK, KBK and KNK and gives a
+  // drawish scale factor for cases such as KRKBP and KmmKm (except for KBBKN).
+  if (!pos.count<PAWN>(WHITE) && npm_w - npm_b <= BishopValueMg)
+      e->factor[WHITE] = uint8_t(npm_w <  RookValueMg   ? SCALE_FACTOR_DRAW :
+                                 npm_b <= BishopValueMg ? 4 : 14);
 
-  if (pos.piece_count(BLACK, PAWN) == 0 && npm_b - npm_w <= BishopValueMg)
-  {
-      e->factor[BLACK] = (uint8_t)
-      (npm_w == npm_b || npm_b < RookValueMg ? 0 : NoPawnsSF[std::min(pos.piece_count(BLACK, BISHOP), 2)]);
-  }
-
-  // Compute the space weight
-  if (npm_w + npm_b >= 2 * QueenValueMg + 4 * RookValueMg + 2 * KnightValueMg)
-  {
-      int minorPieceCount =  pos.piece_count(WHITE, KNIGHT) + pos.piece_count(WHITE, BISHOP)
-                           + pos.piece_count(BLACK, KNIGHT) + pos.piece_count(BLACK, BISHOP);
-
-      e->spaceWeight = minorPieceCount * minorPieceCount;
-  }
+  if (!pos.count<PAWN>(BLACK) && npm_b - npm_w <= BishopValueMg)
+      e->factor[BLACK] = uint8_t(npm_b <  RookValueMg   ? SCALE_FACTOR_DRAW :
+                                 npm_w <= BishopValueMg ? 4 : 14);
 
   // Evaluate the material imbalance. We use PIECE_TYPE_NONE as a place holder
-  // for the bishop pair "extended piece", this allow us to be more flexible
+  // for the bishop pair "extended piece", which allows us to be more flexible
   // in defining bishop pair bonuses.
-  const int pieceCount[2][8] = {
-  { pos.piece_count(WHITE, BISHOP) > 1, pos.piece_count(WHITE, PAWN), pos.piece_count(WHITE, KNIGHT),
-    pos.piece_count(WHITE, BISHOP)    , pos.piece_count(WHITE, ROOK), pos.piece_count(WHITE, QUEEN) },
-  { pos.piece_count(BLACK, BISHOP) > 1, pos.piece_count(BLACK, PAWN), pos.piece_count(BLACK, KNIGHT),
-    pos.piece_count(BLACK, BISHOP)    , pos.piece_count(BLACK, ROOK), pos.piece_count(BLACK, QUEEN) } };
+  const int pieceCount[COLOR_NB][PIECE_TYPE_NB] = {
+  { pos.count<BISHOP>(WHITE) > 1, pos.count<PAWN>(WHITE), pos.count<KNIGHT>(WHITE),
+    pos.count<BISHOP>(WHITE)    , pos.count<ROOK>(WHITE), pos.count<QUEEN >(WHITE) },
+  { 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
 
-/// MaterialTable::imbalance() calculates imbalance comparing piece count of each
-/// piece type for both colors.
-
-template<Color Us>
-int MaterialTable::imbalance(const int pieceCount[][8]) {
-
-  const Color Them = (Us == WHITE ? BLACK : WHITE);
-
-  int pt1, pt2, pc, v;
-  int value = 0;
-
-  // Redundancy of major pieces, formula based on Kaufman's paper
-  // "The Evaluation of Material Imbalances in Chess"
-  if (pieceCount[Us][ROOK] > 0)
-      value -=  RedundantRookPenalty * (pieceCount[Us][ROOK] - 1)
-              + RedundantQueenPenalty * pieceCount[Us][QUEEN];
-
-  // Second-degree polynomial material imbalance by Tord Romstad
-  for (pt1 = NO_PIECE_TYPE; pt1 <= QUEEN; pt1++)
-  {
-      pc = pieceCount[Us][pt1];
-      if (!pc)
-          continue;
-
-      v = LinearCoefficients[pt1];
-
-      for (pt2 = NO_PIECE_TYPE; pt2 <= pt1; pt2++)
-          v +=  QuadraticCoefficientsSameColor[pt1][pt2] * pieceCount[Us][pt2]
-              + QuadraticCoefficientsOppositeColor[pt1][pt2] * pieceCount[Them][pt2];
-
-      value += pc * v;
-  }
-  return value;
-}
-
-
-/// MaterialTable::game_phase() calculates the phase given the current
-/// position. Because the phase is strictly a function of the material, it
-/// is stored in MaterialEntry.
-
-Phase MaterialTable::game_phase(const Position& pos) {
-
-  Value npm = pos.non_pawn_material(WHITE) + pos.non_pawn_material(BLACK);
-
-  return  npm >= MidgameLimit ? PHASE_MIDGAME
-        : npm <= EndgameLimit ? PHASE_ENDGAME
-        : Phase(((npm - EndgameLimit) * 128) / (MidgameLimit - EndgameLimit));
-}
+} // namespace Stockfish

src/material.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,7 +16,7 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(MATERIAL_H_INCLUDED)
+#ifndef MATERIAL_H_INCLUDED
 #define MATERIAL_H_INCLUDED
 
 #include "endgame.h"
@@ -25,96 +24,48 @@
 #include "position.h"
 #include "types.h"
 
-const int MaterialTableSize = 8192;
+namespace Stockfish::Material {
 
-/// Game phase
-enum Phase {
-  PHASE_ENDGAME = 0,
-  PHASE_MIDGAME = 128
-};
-
-
-/// MaterialEntry is a class which contains various information about a
-/// material configuration. It contains a material balance evaluation,
-/// a function pointer to a special endgame evaluation function (which in
-/// most cases is NULL, meaning that the standard evaluation function will
-/// be used), and "scale factors" for black and white.
+/// Material::Entry contains various information about a material configuration.
+/// It contains a material imbalance evaluation, a function pointer to a special
+/// endgame evaluation function (which in most cases is nullptr, meaning that the
+/// standard evaluation function will be used), and scale factors.
 ///
-/// The scale factors are used to scale the evaluation score up or down.
-/// For instance, in KRB vs KR endgames, the score is scaled down by a factor
-/// of 4, which will result in scores of absolute value less than one pawn.
+/// The scale factors are used to scale the evaluation score up or down. For
+/// instance, in KRB vs KR endgames, the score is scaled down by a factor of 4,
+/// which will result in scores of absolute value less than one pawn.
 
-class MaterialEntry {
+struct Entry {
 
-  friend struct MaterialTable;
+  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); }
 
-public:
-  Score material_value() const;
-  ScaleFactor scale_factor(const Position& pos, Color c) const;
-  int space_weight() const;
-  Phase game_phase() const;
-  bool specialized_eval_exists() const;
-  Value evaluate(const Position& pos) const;
+  // scale_factor() takes a position and a color as input and returns a scale factor
+  // for the given color. We have to provide the position in addition to the color
+  // because the scale factor may also be a function which should be applied to
+  // the position. For instance, in KBP vs K endgames, the scaling function looks
+  // for rook pawns and wrong-colored bishops.
+  ScaleFactor scale_factor(const Position& pos, Color c) const {
+    ScaleFactor sf = scalingFunction[c] ? (*scalingFunction[c])(pos)
+                                        :  SCALE_FACTOR_NONE;
+    return sf != SCALE_FACTOR_NONE ? sf : ScaleFactor(factor[c]);
+  }
 
-private:
   Key key;
-  int16_t value;
-  uint8_t factor[2];
-  EndgameBase<Value>* evaluationFunction;
-  EndgameBase<ScaleFactor>* scalingFunction[2];
-  int spaceWeight;
-  Phase gamePhase;
+  const EndgameBase<Value>* evaluationFunction;
+  const EndgameBase<ScaleFactor>* scalingFunction[COLOR_NB]; // Could be one for each
+                                                             // side (e.g. KPKP, KBPsK)
+  Score score;
+  int16_t gamePhase;
+  uint8_t factor[COLOR_NB];
 };
 
+using Table = HashTable<Entry, 8192>;
 
-/// The MaterialTable class represents a material hash table. The most important
-/// method is probe(), which returns a pointer to a MaterialEntry object.
+Entry* probe(const Position& pos);
 
-struct MaterialTable {
+} // namespace Stockfish::Material
 
-  MaterialEntry* probe(const Position& pos);
-  static Phase game_phase(const Position& pos);
-  template<Color Us> static int imbalance(const int pieceCount[][8]);
-
-  HashTable<MaterialEntry, MaterialTableSize> entries;
-  Endgames endgames;
-};
-
-
-/// MaterialEntry::scale_factor takes a position and a color as input, and
-/// returns a scale factor for the given color. We have to provide the
-/// position in addition to the color, because the scale factor need not
-/// to be a constant: It can also be a function which should be applied to
-/// the position. For instance, in KBP vs K endgames, a scaling function
-/// which checks for draws with rook pawns and wrong-colored bishops.
-
-inline ScaleFactor MaterialEntry::scale_factor(const Position& pos, Color c) const {
-
-  if (!scalingFunction[c])
-      return ScaleFactor(factor[c]);
-
-  ScaleFactor sf = (*scalingFunction[c])(pos);
-  return sf == SCALE_FACTOR_NONE ? ScaleFactor(factor[c]) : sf;
-}
-
-inline Value MaterialEntry::evaluate(const Position& pos) const {
-  return (*evaluationFunction)(pos);
-}
-
-inline Score MaterialEntry::material_value() const {
-  return make_score(value, value);
-}
-
-inline int MaterialEntry::space_weight() const {
-  return spaceWeight;
-}
-
-inline Phase MaterialEntry::game_phase() const {
-  return gamePhase;
-}
-
-inline bool MaterialEntry::specialized_eval_exists() const {
-  return evaluationFunction != NULL;
-}
-
-#endif // !defined(MATERIAL_H_INCLUDED)
+#endif // #ifndef MATERIAL_H_INCLUDED

src/misc.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,153 +16,393 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#ifdef _WIN32
+#if _WIN32_WINNT < 0x0601
+#undef  _WIN32_WINNT
+#define _WIN32_WINNT 0x0601 // Force to include needed API prototypes
+#endif
+
+#ifndef NOMINMAX
+#define NOMINMAX
+#endif
+
+#include <windows.h>
+// The needed Windows API for processor groups could be missed from old Windows
+// versions, so instead of calling them directly (forcing the linker to resolve
+// the calls at compile time), try to load them at runtime. To do this we need
+// first to define the corresponding function pointers.
+extern "C" {
+using fun1_t = bool(*)(LOGICAL_PROCESSOR_RELATIONSHIP,
+                       PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX, PDWORD);
+using fun2_t = bool(*)(USHORT, PGROUP_AFFINITY);
+using fun3_t = bool(*)(HANDLE, CONST GROUP_AFFINITY*, PGROUP_AFFINITY);
+using fun4_t = bool(*)(USHORT, PGROUP_AFFINITY, USHORT, PUSHORT);
+using fun5_t = WORD(*)();
+using fun6_t = bool(*)(HANDLE, DWORD, PHANDLE);
+using fun7_t = bool(*)(LPCSTR, LPCSTR, PLUID);
+using fun8_t = bool(*)(HANDLE, BOOL, PTOKEN_PRIVILEGES, DWORD, PTOKEN_PRIVILEGES, PDWORD);
+}
+#endif
+
+#include <cmath>
+#include <cstdlib>
+#include <fstream>
 #include <iomanip>
 #include <iostream>
 #include <sstream>
+#include <string_view>
+#include <vector>
+
+#if defined(__linux__) && !defined(__ANDROID__)
+#include <stdlib.h>
+#include <sys/mman.h>
+#endif
+
+#if defined(__APPLE__) || defined(__ANDROID__) || defined(__OpenBSD__) || (defined(__GLIBCXX__) && !defined(_GLIBCXX_HAVE_ALIGNED_ALLOC) && !defined(_WIN32)) || defined(__e2k__)
+#define POSIXALIGNEDALLOC
+#include <stdlib.h>
+#endif
 
 #include "misc.h"
 #include "thread.h"
 
-#if defined(__hpux)
-#    include <sys/pstat.h>
-#endif
-
 using namespace std;
 
-/// Version number. If Version is left empty, then Tag plus current
-/// date (in the format YYMMDD) is used as a version number.
+namespace Stockfish {
 
-static const string Version = "";
-static const string Tag = "";
-
-
-/// engine_info() returns the full name of the current Stockfish version.
-/// This will be either "Stockfish YYMMDD" (where YYMMDD is the date when
-/// the program was compiled) or "Stockfish <version number>", depending
-/// on whether Version is empty.
-
-const string engine_info(bool to_uci) {
-
-  const string months("Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec");
-  const string cpu64(Is64Bit ? " 64bit" : "");
-  const string popcnt(HasPopCnt ? " SSE4.2" : "");
-
-  string month, day, year;
-  stringstream s, date(__DATE__); // From compiler, format is "Sep 21 2008"
-
-  s << "Stockfish " << Version;
-
-  if (Version.empty())
-  {
-      date >> month >> day >> year;
-
-      s << Tag << setfill('0') << " " << year.substr(2)
-        << setw(2) << (1 + months.find(month) / 4) << setw(2) << day;
-  }
-
-  s << cpu64 << popcnt << (to_uci ? "\nid author ": " by ")
-    << "Tord Romstad, Marco Costalba and Joona Kiiski";
-
-  return s.str();
-}
-
-
-/// Convert system time to milliseconds. That's all we need.
-
-Time::point Time::now() {
-  sys_time_t t; system_time(&t); return time_to_msec(t);
-}
-
-
-/// Debug functions used mainly to collect run-time statistics
-
-static uint64_t hits[2], means[2];
-
-void dbg_hit_on(bool b) { hits[0]++; if (b) hits[1]++; }
-void dbg_hit_on_c(bool c, bool b) { if (c) dbg_hit_on(b); }
-void dbg_mean_of(int v) { means[0]++; means[1] += v; }
-
-void dbg_print() {
-
-  if (hits[0])
-      cerr << "Total " << hits[0] << " Hits " << hits[1]
-           << " hit rate (%) " << 100 * hits[1] / hits[0] << endl;
-
-  if (means[0])
-      cerr << "Total " << means[0] << " Mean "
-           << (float)means[1] / means[0] << endl;
-}
+namespace {
 
+/// Version number or dev.
+constexpr string_view version = "16";
 
 /// 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
-/// can toggle the logging of std::cout and std:cin at runtime while preserving
-/// usual i/o functionality and without changing a single line of code!
+/// can toggle the logging of std::cout and std:cin at runtime whilst preserving
+/// usual I/O functionality, all without changing a single line of code!
 /// Idea from http://groups.google.com/group/comp.lang.c++/msg/1d941c0f26ea0d81
 
-struct Tie: public streambuf { // MSVC requires splitted streambuf for cin and cout
+struct Tie: public streambuf { // MSVC requires split streambuf for cin and cout
 
-  Tie(streambuf* b, ofstream* f) : buf(b), file(f) {}
+  Tie(streambuf* b, streambuf* l) : buf(b), logBuf(l) {}
 
-  int sync() { return file->rdbuf()->pubsync(), buf->pubsync(); }
-  int overflow(int c) { return log(buf->sputc((char)c), "<< "); }
-  int underflow() { return buf->sgetc(); }
-  int uflow() { return log(buf->sbumpc(), ">> "); }
+  int sync() override { return logBuf->pubsync(), buf->pubsync(); }
+  int overflow(int c) override { return log(buf->sputc((char)c), "<< "); }
+  int underflow() override { return buf->sgetc(); }
+  int uflow() override { return log(buf->sbumpc(), ">> "); }
 
-  streambuf* buf;
-  ofstream* file;
+  streambuf *buf, *logBuf;
 
   int log(int c, const char* prefix) {
 
-    static int last = '\n';
+    static int last = '\n'; // Single log file
 
     if (last == '\n')
-        file->rdbuf()->sputn(prefix, 3);
+        logBuf->sputn(prefix, 3);
 
-    return last = file->rdbuf()->sputc((char)c);
+    return last = logBuf->sputc((char)c);
   }
 };
 
 class Logger {
 
-  Logger() : in(cin.rdbuf(), &file), out(cout.rdbuf(), &file) {}
- ~Logger() { start(false); }
+  Logger() : in(cin.rdbuf(), file.rdbuf()), out(cout.rdbuf(), file.rdbuf()) {}
+ ~Logger() { start(""); }
 
   ofstream file;
   Tie in, out;
 
 public:
-  static void start(bool b) {
+  static void start(const std::string& fname) {
 
     static Logger l;
 
-    if (b && !l.file.is_open())
-    {
-        l.file.open("io_log.txt", ifstream::out | ifstream::app);
-        cin.rdbuf(&l.in);
-        cout.rdbuf(&l.out);
-    }
-    else if (!b && l.file.is_open())
+    if (l.file.is_open())
     {
         cout.rdbuf(l.out.buf);
         cin.rdbuf(l.in.buf);
         l.file.close();
     }
+
+    if (!fname.empty())
+    {
+        l.file.open(fname, ifstream::out);
+
+        if (!l.file.is_open())
+        {
+            cerr << "Unable to open debug log file " << fname << endl;
+            exit(EXIT_FAILURE);
+        }
+
+        cin.rdbuf(&l.in);
+        cout.rdbuf(&l.out);
+    }
   }
 };
 
+} // namespace
 
-/// Used to serialize access to std::cout to avoid multiple threads to write at
+
+/// engine_info() returns the full name of the current Stockfish version.
+/// For local dev compiles we try to append the commit sha and commit date
+/// from git if that fails only the local compilation date is set and "nogit" is specified:
+/// Stockfish dev-YYYYMMDD-SHA
+/// or
+/// Stockfish dev-YYYYMMDD-nogit
+///
+/// For releases (non dev builds) we only include the version number:
+/// Stockfish version
+
+string engine_info(bool to_uci) {
+  stringstream ss;
+  ss << "Stockfish " << version << setfill('0');
+
+  if constexpr (version == "dev")
+  {
+      ss << "-";
+      #ifdef GIT_DATE
+      ss << stringify(GIT_DATE);
+      #else
+      constexpr string_view months("Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec");
+      string month, day, year;
+      stringstream date(__DATE__); // From compiler, format is "Sep 21 2008"
+
+      date >> month >> day >> year;
+      ss << year << setw(2) << setfill('0') << (1 + months.find(month) / 4) << setw(2) << setfill('0') << day;
+      #endif
+
+      ss << "-";
+
+      #ifdef GIT_SHA
+      ss << stringify(GIT_SHA);
+      #else
+      ss << "nogit";
+      #endif
+  }
+
+  ss << (to_uci  ? "\nid author ": " by ")
+     << "the Stockfish developers (see AUTHORS file)";
+
+  return ss.str();
+}
+
+
+/// compiler_info() returns a string trying to describe the compiler we use
+
+std::string compiler_info() {
+
+  #define make_version_string(major, minor, patch) stringify(major) "." stringify(minor) "." stringify(patch)
+
+/// Predefined macros hell:
+///
+/// __GNUC__           Compiler is gcc, Clang or Intel on Linux
+/// __INTEL_COMPILER   Compiler is Intel
+/// _MSC_VER           Compiler is MSVC or Intel on Windows
+/// _WIN32             Building on Windows (any)
+/// _WIN64             Building on Windows 64 bit
+
+  std::string compiler = "\nCompiled by ";
+
+  #ifdef __clang__
+     compiler += "clang++ ";
+     compiler += make_version_string(__clang_major__, __clang_minor__, __clang_patchlevel__);
+  #elif __INTEL_COMPILER
+     compiler += "Intel compiler ";
+     compiler += "(version ";
+     compiler += stringify(__INTEL_COMPILER) " update " stringify(__INTEL_COMPILER_UPDATE);
+     compiler += ")";
+  #elif _MSC_VER
+     compiler += "MSVC ";
+     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__);
+  #else
+     compiler += "Unknown compiler ";
+     compiler += "(unknown version)";
+  #endif
+
+  #if defined(__APPLE__)
+     compiler += " on Apple";
+  #elif defined(__CYGWIN__)
+     compiler += " on Cygwin";
+  #elif defined(__MINGW64__)
+     compiler += " on MinGW64";
+  #elif defined(__MINGW32__)
+     compiler += " on MinGW32";
+  #elif defined(__ANDROID__)
+     compiler += " on Android";
+  #elif defined(__linux__)
+     compiler += " on Linux";
+  #elif defined(_WIN64)
+     compiler += " on Microsoft Windows 64-bit";
+  #elif defined(_WIN32)
+     compiler += " on Microsoft Windows 32-bit";
+  #else
+     compiler += " on unknown system";
+  #endif
+
+  compiler += "\nCompilation settings include: ";
+  compiler += (Is64Bit ? " 64bit" : " 32bit");
+  #if defined(USE_VNNI)
+    compiler += " VNNI";
+  #endif
+  #if defined(USE_AVX512)
+    compiler += " AVX512";
+  #endif
+  compiler += (HasPext ? " BMI2" : "");
+  #if defined(USE_AVX2)
+    compiler += " AVX2";
+  #endif
+  #if defined(USE_SSE41)
+    compiler += " SSE41";
+  #endif
+  #if defined(USE_SSSE3)
+    compiler += " SSSE3";
+  #endif
+  #if defined(USE_SSE2)
+    compiler += " SSE2";
+  #endif
+  compiler += (HasPopCnt ? " POPCNT" : "");
+  #if defined(USE_MMX)
+    compiler += " MMX";
+  #endif
+  #if defined(USE_NEON)
+    compiler += " NEON";
+  #endif
+
+  #if !defined(NDEBUG)
+    compiler += " DEBUG";
+  #endif
+
+  compiler += "\n__VERSION__ macro expands to: ";
+  #ifdef __VERSION__
+     compiler += __VERSION__;
+  #else
+     compiler += "(undefined macro)";
+  #endif
+  compiler += "\n";
+
+  return compiler;
+}
+
+
+/// Debug functions used mainly to collect run-time statistics
+constexpr int MaxDebugSlots = 32;
+
+namespace {
+
+template<size_t N>
+struct DebugInfo {
+    std::atomic<int64_t> data[N] = { 0 };
+
+    constexpr inline std::atomic<int64_t>& operator[](int index) { return data[index]; }
+};
+
+DebugInfo<2> hit[MaxDebugSlots];
+DebugInfo<2> mean[MaxDebugSlots];
+DebugInfo<3> stdev[MaxDebugSlots];
+DebugInfo<6> correl[MaxDebugSlots];
+
+}  // namespace
+
+void dbg_hit_on(bool cond, int slot) {
+
+    ++hit[slot][0];
+    if (cond)
+        ++hit[slot][1];
+}
+
+void dbg_mean_of(int64_t value, int slot) {
+
+    ++mean[slot][0];
+    mean[slot][1] += value;
+}
+
+void dbg_stdev_of(int64_t value, int slot) {
+
+    ++stdev[slot][0];
+    stdev[slot][1] += value;
+    stdev[slot][2] += value * value;
+}
+
+void dbg_correl_of(int64_t value1, int64_t value2, int slot) {
+
+    ++correl[slot][0];
+    correl[slot][1] += value1;
+    correl[slot][2] += value1 * value1;
+    correl[slot][3] += value2;
+    correl[slot][4] += value2 * value2;
+    correl[slot][5] += value1 * value2;
+}
+
+void dbg_print() {
+
+    int64_t n;
+    auto E   = [&n](int64_t x) { return double(x) / n; };
+    auto sqr = [](double x) { return x * x; };
+
+    for (int i = 0; i < MaxDebugSlots; ++i)
+        if ((n = hit[i][0]))
+            std::cerr << "Hit #" << i
+                      << ": Total " << n << " Hits " << hit[i][1]
+                      << " Hit Rate (%) " << 100.0 * E(hit[i][1])
+                      << std::endl;
+
+    for (int i = 0; i < MaxDebugSlots; ++i)
+        if ((n = mean[i][0]))
+        {
+            std::cerr << "Mean #" << i
+                      << ": Total " << n << " Mean " << E(mean[i][1])
+                      << std::endl;
+        }
+
+    for (int i = 0; i < MaxDebugSlots; ++i)
+        if ((n = stdev[i][0]))
+        {
+            double r = sqrtl(E(stdev[i][2]) - sqr(E(stdev[i][1])));
+            std::cerr << "Stdev #" << i
+                      << ": Total " << n << " Stdev " << r
+                      << std::endl;
+        }
+
+    for (int i = 0; i < MaxDebugSlots; ++i)
+        if ((n = correl[i][0]))
+        {
+            double r = (E(correl[i][5]) - E(correl[i][1]) * E(correl[i][3]))
+                       / (  sqrtl(E(correl[i][2]) - sqr(E(correl[i][1])))
+                          * sqrtl(E(correl[i][4]) - sqr(E(correl[i][3]))));
+            std::cerr << "Correl. #" << i
+                      << ": Total " << n << " Coefficient " << r
+                      << std::endl;
+        }
+}
+
+
+/// Used to serialize access to std::cout to avoid multiple threads writing at
 /// the same time.
 
 std::ostream& operator<<(std::ostream& os, SyncCout sc) {
 
-  static Mutex m;
+  static std::mutex m;
 
-  if (sc == io_lock)
+  if (sc == IO_LOCK)
       m.lock();
 
-  if (sc == io_unlock)
+  if (sc == IO_UNLOCK)
       m.unlock();
 
   return os;
@@ -171,74 +410,378 @@ std::ostream& operator<<(std::ostream& os, SyncCout sc) {
 
 
 /// Trampoline helper to avoid moving Logger to misc.h
-void start_logger(bool b) { Logger::start(b); }
+void start_logger(const std::string& fname) { Logger::start(fname); }
 
 
-/// cpu_count() tries to detect the number of CPU cores
+/// prefetch() preloads the given address in L1/L2 cache. This is a non-blocking
+/// function that doesn't stall the CPU waiting for data to be loaded from memory,
+/// which can be quite slow.
+#ifdef NO_PREFETCH
 
-int cpu_count() {
-
-#if defined(_WIN32) || defined(_WIN64)
-  SYSTEM_INFO s;
-  GetSystemInfo(&s);
-  return s.dwNumberOfProcessors;
-#else
-
-#  if defined(_SC_NPROCESSORS_ONLN)
-  return sysconf(_SC_NPROCESSORS_ONLN);
-#  elif defined(__hpux)
-  struct pst_dynamic psd;
-  if (pstat_getdynamic(&psd, sizeof(psd), (size_t)1, 0) == -1)
-      return 1;
-  return psd.psd_proc_cnt;
-#  else
-  return 1;
-#  endif
-
-#endif
-}
-
-
-/// timed_wait() waits for msec milliseconds. It is mainly an helper to wrap
-/// conversion from milliseconds to struct timespec, as used by pthreads.
-
-void timed_wait(WaitCondition& sleepCond, Lock& sleepLock, int msec) {
-
-#if defined(_WIN32) || defined(_WIN64)
-  int tm = msec;
-#else
-  timespec ts, *tm = &ts;
-  uint64_t ms = Time::now() + msec;
-
-  ts.tv_sec = ms / 1000;
-  ts.tv_nsec = (ms % 1000) * 1000000LL;
-#endif
-
-  cond_timedwait(sleepCond, sleepLock, tm);
-}
-
-
-/// prefetch() preloads the given address in L1/L2 cache. This is a non
-/// blocking function and do not stalls the CPU waiting for data to be
-/// loaded from memory, that can be quite slow.
-#if defined(NO_PREFETCH)
-
-void prefetch(char*) {}
+void prefetch(void*) {}
 
 #else
 
-#   include <xmmintrin.h>
+void prefetch(void* addr) {
 
-void prefetch(char* addr) {
-
-#  if defined(__INTEL_COMPILER) || defined(__ICL)
-   // This hack prevents prefetches to be optimized away by
-   // Intel compiler. Both MSVC and gcc seems not affected.
+#  if defined(__INTEL_COMPILER)
+   // This hack prevents prefetches from being optimized away by
+   // Intel compiler. Both MSVC and gcc seem not be affected by this.
    __asm__ ("");
 #  endif
 
-  _mm_prefetch(addr, _MM_HINT_T2);
-  _mm_prefetch(addr+64, _MM_HINT_T2); // 64 bytes ahead
+#  if defined(__INTEL_COMPILER) || defined(_MSC_VER)
+  _mm_prefetch((char*)addr, _MM_HINT_T0);
+#  else
+  __builtin_prefetch(addr);
+#  endif
 }
 
 #endif
+
+
+/// std_aligned_alloc() is our wrapper for systems where the c++17 implementation
+/// does not guarantee the availability of aligned_alloc(). Memory allocated with
+/// std_aligned_alloc() must be freed with std_aligned_free().
+
+void* std_aligned_alloc(size_t alignment, size_t size) {
+
+#if defined(POSIXALIGNEDALLOC)
+  void *mem;
+  return posix_memalign(&mem, alignment, size) ? nullptr : mem;
+#elif defined(_WIN32) && !defined(_M_ARM) && !defined(_M_ARM64)
+  return _mm_malloc(size, alignment);
+#elif defined(_WIN32)
+  return _aligned_malloc(size, alignment);
+#else
+  return std::aligned_alloc(alignment, size);
+#endif
+}
+
+void std_aligned_free(void* ptr) {
+
+#if defined(POSIXALIGNEDALLOC)
+  free(ptr);
+#elif defined(_WIN32) && !defined(_M_ARM) && !defined(_M_ARM64)
+  _mm_free(ptr);
+#elif defined(_WIN32)
+  _aligned_free(ptr);
+#else
+  free(ptr);
+#endif
+}
+
+/// aligned_large_pages_alloc() will return suitably aligned memory, if possible using large pages.
+
+#if defined(_WIN32)
+
+static void* aligned_large_pages_alloc_windows([[maybe_unused]] size_t allocSize) {
+
+  #if !defined(_WIN64)
+    return nullptr;
+  #else
+
+  HANDLE hProcessToken { };
+  LUID luid { };
+  void* mem = nullptr;
+
+  const size_t largePageSize = GetLargePageMinimum();
+  if (!largePageSize)
+      return nullptr;
+
+  // Dynamically link OpenProcessToken, LookupPrivilegeValue and AdjustTokenPrivileges
+
+  HMODULE hAdvapi32 = GetModuleHandle(TEXT("advapi32.dll"));
+
+  if (!hAdvapi32)
+      hAdvapi32 = LoadLibrary(TEXT("advapi32.dll"));
+
+  auto fun6 = (fun6_t)(void(*)())GetProcAddress(hAdvapi32, "OpenProcessToken");
+  if (!fun6)
+      return nullptr;
+  auto fun7 = (fun7_t)(void(*)())GetProcAddress(hAdvapi32, "LookupPrivilegeValueA");
+  if (!fun7)
+      return nullptr;
+  auto fun8 = (fun8_t)(void(*)())GetProcAddress(hAdvapi32, "AdjustTokenPrivileges");
+  if (!fun8)
+      return nullptr;
+
+  // We need SeLockMemoryPrivilege, so try to enable it for the process
+  if (!fun6( // OpenProcessToken()
+      GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES | TOKEN_QUERY, &hProcessToken))
+          return nullptr;
+
+  if (fun7( // LookupPrivilegeValue(nullptr, SE_LOCK_MEMORY_NAME, &luid)
+      nullptr, "SeLockMemoryPrivilege", &luid))
+  {
+      TOKEN_PRIVILEGES tp { };
+      TOKEN_PRIVILEGES prevTp { };
+      DWORD prevTpLen = 0;
+
+      tp.PrivilegeCount = 1;
+      tp.Privileges[0].Luid = luid;
+      tp.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED;
+
+      // Try to enable SeLockMemoryPrivilege. Note that even if AdjustTokenPrivileges() succeeds,
+      // we still need to query GetLastError() to ensure that the privileges were actually obtained.
+      if (fun8( // AdjustTokenPrivileges()
+              hProcessToken, FALSE, &tp, sizeof(TOKEN_PRIVILEGES), &prevTp, &prevTpLen) &&
+          GetLastError() == ERROR_SUCCESS)
+      {
+          // Round up size to full pages and allocate
+          allocSize = (allocSize + largePageSize - 1) & ~size_t(largePageSize - 1);
+          mem = VirtualAlloc(
+              nullptr, allocSize, MEM_RESERVE | MEM_COMMIT | MEM_LARGE_PAGES, PAGE_READWRITE);
+
+          // Privilege no longer needed, restore previous state
+          fun8( // AdjustTokenPrivileges ()
+              hProcessToken, FALSE, &prevTp, 0, nullptr, nullptr);
+      }
+  }
+
+  CloseHandle(hProcessToken);
+
+  return mem;
+
+  #endif
+}
+
+void* aligned_large_pages_alloc(size_t allocSize) {
+
+  // Try to allocate large pages
+  void* mem = aligned_large_pages_alloc_windows(allocSize);
+
+  // Fall back to regular, page aligned, allocation if necessary
+  if (!mem)
+      mem = VirtualAlloc(nullptr, allocSize, MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE);
+
+  return mem;
+}
+
+#else
+
+void* aligned_large_pages_alloc(size_t allocSize) {
+
+#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_large_pages_free() will free the previously allocated ttmem
+
+#if defined(_WIN32)
+
+void aligned_large_pages_free(void* mem) {
+
+  if (mem && !VirtualFree(mem, 0, MEM_RELEASE))
+  {
+      DWORD err = GetLastError();
+      std::cerr << "Failed to free large page memory. Error code: 0x"
+                << std::hex << err
+                << std::dec << std::endl;
+      exit(EXIT_FAILURE);
+  }
+}
+
+#else
+
+void aligned_large_pages_free(void *mem) {
+  std_aligned_free(mem);
+}
+
+#endif
+
+
+namespace WinProcGroup {
+
+#ifndef _WIN32
+
+void bindThisThread(size_t) {}
+
+#else
+
+/// best_node() retrieves logical processor information using Windows specific
+/// API and returns the best node id for the thread with index idx. Original
+/// code from Texel by Peter Österlund.
+
+static int best_node(size_t idx) {
+
+  int threads = 0;
+  int nodes = 0;
+  int cores = 0;
+  DWORD returnLength = 0;
+  DWORD byteOffset = 0;
+
+  // Early exit if the needed API is not available at runtime
+  HMODULE k32 = GetModuleHandle(TEXT("Kernel32.dll"));
+  auto fun1 = (fun1_t)(void(*)())GetProcAddress(k32, "GetLogicalProcessorInformationEx");
+  if (!fun1)
+      return -1;
+
+  // First call to GetLogicalProcessorInformationEx() to get returnLength.
+  // We expect the call to fail due to null buffer.
+  if (fun1(RelationAll, nullptr, &returnLength))
+      return -1;
+
+  // Once we know returnLength, allocate the buffer
+  SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *buffer, *ptr;
+  ptr = buffer = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)malloc(returnLength);
+
+  // Second call to GetLogicalProcessorInformationEx(), now we expect to succeed
+  if (!fun1(RelationAll, buffer, &returnLength))
+  {
+      free(buffer);
+      return -1;
+  }
+
+  while (byteOffset < returnLength)
+  {
+      if (ptr->Relationship == RelationNumaNode)
+          nodes++;
+
+      else if (ptr->Relationship == RelationProcessorCore)
+      {
+          cores++;
+          threads += (ptr->Processor.Flags == LTP_PC_SMT) ? 2 : 1;
+      }
+
+      assert(ptr->Size);
+      byteOffset += ptr->Size;
+      ptr = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)(((char*)ptr) + ptr->Size);
+  }
+
+  free(buffer);
+
+  std::vector<int> groups;
+
+  // Run as many threads as possible on the same node until core limit is
+  // reached, then move on filling the next node.
+  for (int n = 0; n < nodes; n++)
+      for (int i = 0; i < cores / nodes; i++)
+          groups.push_back(n);
+
+  // In case a core has more than one logical processor (we assume 2) and we
+  // have still threads to allocate, then spread them evenly across available
+  // nodes.
+  for (int t = 0; t < threads - cores; t++)
+      groups.push_back(t % nodes);
+
+  // If we still have more threads than the total number of logical processors
+  // then return -1 and let the OS to decide what to do.
+  return idx < groups.size() ? groups[idx] : -1;
+}
+
+
+/// bindThisThread() set the group affinity of the current thread
+
+void bindThisThread(size_t idx) {
+
+  // Use only local variables to be thread-safe
+  int node = best_node(idx);
+
+  if (node == -1)
+      return;
+
+  // Early exit if the needed API are not available at runtime
+  HMODULE k32 = GetModuleHandle(TEXT("Kernel32.dll"));
+  auto fun2 = (fun2_t)(void(*)())GetProcAddress(k32, "GetNumaNodeProcessorMaskEx");
+  auto fun3 = (fun3_t)(void(*)())GetProcAddress(k32, "SetThreadGroupAffinity");
+  auto fun4 = (fun4_t)(void(*)())GetProcAddress(k32, "GetNumaNodeProcessorMask2");
+  auto fun5 = (fun5_t)(void(*)())GetProcAddress(k32, "GetMaximumProcessorGroupCount");
+
+  if (!fun2 || !fun3)
+      return;
+
+  if (!fun4 || !fun5)
+  {
+      GROUP_AFFINITY affinity;
+      if (fun2(node, &affinity))                                                 // GetNumaNodeProcessorMaskEx
+          fun3(GetCurrentThread(), &affinity, nullptr);                          // SetThreadGroupAffinity
+  }
+  else
+  {
+      // If a numa node has more than one processor group, we assume they are
+      // sized equal and we spread threads evenly across the groups.
+      USHORT elements, returnedElements;
+      elements = fun5();                                                         // GetMaximumProcessorGroupCount
+      GROUP_AFFINITY *affinity = (GROUP_AFFINITY*)malloc(elements * sizeof(GROUP_AFFINITY));
+      if (fun4(node, affinity, elements, &returnedElements))                     // GetNumaNodeProcessorMask2
+          fun3(GetCurrentThread(), &affinity[idx % returnedElements], nullptr);  // SetThreadGroupAffinity
+      free(affinity);
+  }
+}
+
+#endif
+
+} // namespace WinProcGroup
+
+#ifdef _WIN32
+#include <direct.h>
+#define GETCWD _getcwd
+#else
+#include <unistd.h>
+#define GETCWD getcwd
+#endif
+
+namespace CommandLine {
+
+string argv0;            // path+name of the executable binary, as given by argv[0]
+string binaryDirectory;  // path of the executable directory
+string workingDirectory; // path of the working directory
+
+void init([[maybe_unused]] int argc, char* argv[]) {
+    string pathSeparator;
+
+    // extract the path+name of the executable binary
+    argv0 = argv[0];
+
+#ifdef _WIN32
+    pathSeparator = "\\";
+  #ifdef _MSC_VER
+    // Under windows argv[0] may not have the extension. Also _get_pgmptr() had
+    // issues in some windows 10 versions, so check returned values carefully.
+    char* pgmptr = nullptr;
+    if (!_get_pgmptr(&pgmptr) && pgmptr != nullptr && *pgmptr)
+        argv0 = pgmptr;
+  #endif
+#else
+    pathSeparator = "/";
+#endif
+
+    // extract the working directory
+    workingDirectory = "";
+    char buff[40000];
+    char* cwd = GETCWD(buff, 40000);
+    if (cwd)
+        workingDirectory = cwd;
+
+    // extract the binary directory path from argv0
+    binaryDirectory = argv0;
+    size_t pos = binaryDirectory.find_last_of("\\/");
+    if (pos == std::string::npos)
+        binaryDirectory = "." + pathSeparator;
+    else
+        binaryDirectory.resize(pos + 1);
+
+    // pattern replacement: "./" at the start of path is replaced by the working directory
+    if (binaryDirectory.find("." + pathSeparator) == 0)
+        binaryDirectory.replace(0, 1, workingDirectory);
+}
+
+
+} // namespace CommandLine
+
+} // namespace Stockfish

src/misc.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,53 +16,161 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(MISC_H_INCLUDED)
+#ifndef MISC_H_INCLUDED
 #define MISC_H_INCLUDED
 
-#include <fstream>
+#include <cassert>
+#include <chrono>
+#include <ostream>
 #include <string>
 #include <vector>
+#include <cstdint>
 
 #include "types.h"
 
-extern const std::string engine_info(bool to_uci = false);
-extern int cpu_count();
-extern void timed_wait(WaitCondition&, Lock&, int);
-extern void prefetch(char* addr);
-extern void start_logger(bool b);
+#define stringify2(x) #x
+#define stringify(x) stringify2(x)
 
-extern void dbg_hit_on(bool b);
-extern void dbg_hit_on_c(bool c, bool b);
-extern void dbg_mean_of(int v);
-extern void dbg_print();
+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_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
 
-struct Log : public std::ofstream {
-  Log(const std::string& f = "log.txt") : std::ofstream(f.c_str(), std::ios::out | std::ios::app) {}
- ~Log() { if (is_open()) close(); }
-};
+void dbg_hit_on(bool cond, int slot = 0);
+void dbg_mean_of(int64_t value, int slot = 0);
+void dbg_stdev_of(int64_t value, int slot = 0);
+void dbg_correl_of(int64_t value1, int64_t value2, int slot = 0);
+void dbg_print();
 
-
-namespace Time {
-  typedef int64_t point;
-  point now();
+using TimePoint = std::chrono::milliseconds::rep; // A value in milliseconds
+static_assert(sizeof(TimePoint) == sizeof(int64_t), "TimePoint should be 64 bits");
+inline TimePoint now() {
+  return std::chrono::duration_cast<std::chrono::milliseconds>
+        (std::chrono::steady_clock::now().time_since_epoch()).count();
 }
 
-
 template<class Entry, int Size>
 struct HashTable {
-  HashTable() : e(Size, Entry()) {}
-  Entry* operator[](Key k) { return &e[(uint32_t)k & (Size - 1)]; }
+  Entry* operator[](Key key) { return &table[(uint32_t)key & (Size - 1)]; }
 
 private:
-  std::vector<Entry> e;
+  std::vector<Entry> table = std::vector<Entry>(Size); // Allocate on the heap
 };
 
 
-enum SyncCout { io_lock, io_unlock };
+enum SyncCout { IO_LOCK, IO_UNLOCK };
 std::ostream& operator<<(std::ostream&, SyncCout);
 
-#define sync_cout std::cout << io_lock
-#define sync_endl std::endl << io_unlock
+#define sync_cout std::cout << IO_LOCK
+#define sync_endl std::endl << IO_UNLOCK
 
-#endif // !defined(MISC_H_INCLUDED)
+
+// 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, std::size_t MaxSize>
+class ValueList {
+
+public:
+  std::size_t size() const { return size_; }
+  void push_back(const T& value) { values_[size_++] = value; }
+  const T* begin() const { return values_; }
+  const T* end() const { return values_ + 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).
+/// It has the following characteristics:
+///
+///  -  Outputs 64-bit numbers
+///  -  Passes Dieharder and SmallCrush test batteries
+///  -  Does not require warm-up, no zeroland to escape
+///  -  Internal state is a single 64-bit integer
+///  -  Period is 2^64 - 1
+///  -  Speed: 1.60 ns/call (Core i7 @3.40GHz)
+///
+/// For further analysis see
+///   <http://vigna.di.unimi.it/ftp/papers/xorshift.pdf>
+
+class PRNG {
+
+  uint64_t s;
+
+  uint64_t rand64() {
+
+    s ^= s >> 12, s ^= s << 25, s ^= s >> 27;
+    return s * 2685821657736338717LL;
+  }
+
+public:
+  PRNG(uint64_t seed) : s(seed) { assert(seed); }
+
+  template<typename T> T rand() { return T(rand64()); }
+
+  /// Special generator used to fast init magic numbers.
+  /// Output values only have 1/8th of their bits set on average.
+  template<typename T> T sparse_rand()
+  { return T(rand64() & rand64() & rand64()); }
+};
+
+inline uint64_t mul_hi64(uint64_t a, uint64_t b) {
+#if defined(__GNUC__) && defined(IS_64BIT)
+    __extension__ using uint128 = unsigned __int128;
+    return ((uint128)a * (uint128)b) >> 64;
+#else
+    uint64_t aL = (uint32_t)a, aH = a >> 32;
+    uint64_t bL = (uint32_t)b, bH = b >> 32;
+    uint64_t c1 = (aL * bL) >> 32;
+    uint64_t c2 = aH * bL + c1;
+    uint64_t c3 = aL * bH + (uint32_t)c2;
+    return aH * bH + (c2 >> 32) + (c3 >> 32);
+#endif
+}
+
+/// Under Windows it is not possible for a process to run on more than one
+/// logical processor group. This usually means to be limited to use max 64
+/// cores. To overcome this, some special platform specific API should be
+/// called to set group affinity for each thread. Original code from Texel by
+/// Peter Österlund.
+
+namespace WinProcGroup {
+  void bindThisThread(size_t idx);
+}
+
+namespace CommandLine {
+  void init(int argc, char* argv[]);
+
+  extern std::string binaryDirectory;  // path of the executable directory
+  extern std::string workingDirectory; // path of the working directory
+}
+
+} // namespace Stockfish
+
+#endif // #ifndef MISC_H_INCLUDED

src/movegen.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -22,424 +21,264 @@
 #include "movegen.h"
 #include "position.h"
 
-/// Simple macro to wrap a very common while loop, no facny, no flexibility,
-/// hardcoded names 'mlist' and 'from'.
-#define SERIALIZE(b) while (b) (*mlist++).move = make_move(from, pop_lsb(&b))
+namespace Stockfish {
 
-/// Version used for pawns, where the 'from' square is given as a delta from the 'to' square
-#define SERIALIZE_PAWNS(b, d) while (b) { Square to = pop_lsb(&b); \
-                                         (*mlist++).move = make_move(to - (d), to); }
 namespace {
 
-  template<CastlingSide Side, bool Checks, bool Chess960>
-  MoveStack* generate_castle(const Position& pos, MoveStack* mlist, Color us) {
+  template<GenType Type, Direction D, bool Enemy>
+  ExtMove* make_promotions(ExtMove* moveList, [[maybe_unused]] Square to) {
 
-    if (pos.castle_impeded(us, Side) || !pos.can_castle(make_castle_right(us, Side)))
-        return mlist;
-
-    // After castling, the rook and king final positions are the same in Chess960
-    // as they would be in standard chess.
-    Square kfrom = pos.king_square(us);
-    Square rfrom = pos.castle_rook_square(us, Side);
-    Square kto = relative_square(us, Side == KING_SIDE ? SQ_G1 : SQ_C1);
-    Bitboard enemies = pos.pieces(~us);
-
-    assert(!pos.in_check());
-
-    const int K = Chess960 ? kto > kfrom ? -1 : 1
-                           : Side == KING_SIDE ? -1 : 1;
-
-    for (Square s = kto; s != kfrom; s += (Square)K)
-        if (pos.attackers_to(s) & enemies)
-            return mlist;
-
-    // Because we generate only legal castling moves we need to verify that
-    // when moving the castling rook we do not discover some hidden checker.
-    // For instance an enemy queen in SQ_A1 when castling rook is in SQ_B1.
-    if (Chess960 && (pos.attackers_to(kto, pos.pieces() ^ rfrom) & enemies))
-        return mlist;
-
-    (*mlist++).move = make<CASTLE>(kfrom, rfrom);
-
-    if (Checks && !pos.move_gives_check((mlist - 1)->move, CheckInfo(pos)))
-        mlist--;
-
-    return mlist;
-  }
-
-
-  template<Square Delta>
-  inline Bitboard move_pawns(Bitboard p) {
-
-    return  Delta == DELTA_N  ?  p << 8
-          : Delta == DELTA_S  ?  p >> 8
-          : Delta == DELTA_NE ? (p & ~FileHBB) << 9
-          : Delta == DELTA_SE ? (p & ~FileHBB) >> 7
-          : Delta == DELTA_NW ? (p & ~FileABB) << 7
-          : Delta == DELTA_SW ? (p & ~FileABB) >> 9 : 0;
-  }
-
-
-  template<GenType Type, Square Delta>
-  inline MoveStack* generate_promotions(MoveStack* mlist, Bitboard pawnsOn7,
-                                        Bitboard target, const CheckInfo* ci) {
-
-    Bitboard b = move_pawns<Delta>(pawnsOn7) & target;
-
-    while (b)
+    if constexpr (Type == CAPTURES || Type == EVASIONS || Type == NON_EVASIONS)
     {
-        Square to = pop_lsb(&b);
-
-        if (Type == CAPTURES || Type == EVASIONS || Type == NON_EVASIONS)
-            (*mlist++).move = make<PROMOTION>(to - Delta, to, QUEEN);
-
-        if (Type == QUIETS || Type == EVASIONS || Type == NON_EVASIONS)
+        *moveList++ = make<PROMOTION>(to - D, to, QUEEN);
+        if constexpr (Enemy && Type == CAPTURES)
         {
-            (*mlist++).move = make<PROMOTION>(to - Delta, to, ROOK);
-            (*mlist++).move = make<PROMOTION>(to - Delta, to, BISHOP);
-            (*mlist++).move = make<PROMOTION>(to - Delta, to, KNIGHT);
+            *moveList++ = make<PROMOTION>(to - D, to, ROOK);
+            *moveList++ = make<PROMOTION>(to - D, to, BISHOP);
+            *moveList++ = make<PROMOTION>(to - D, to, KNIGHT);
         }
-
-        // Knight-promotion is the only one that can give a direct check not
-        // already included in the queen-promotion.
-        if (Type == QUIET_CHECKS && (StepAttacksBB[W_KNIGHT][to] & ci->ksq))
-            (*mlist++).move = make<PROMOTION>(to - Delta, to, KNIGHT);
-        else
-            (void)ci; // Silence a warning under MSVC
     }
 
-    return mlist;
+    if constexpr ((Type == QUIETS && !Enemy) || Type == EVASIONS || Type == NON_EVASIONS)
+    {
+        *moveList++ = make<PROMOTION>(to - D, to, ROOK);
+        *moveList++ = make<PROMOTION>(to - D, to, BISHOP);
+        *moveList++ = make<PROMOTION>(to - D, to, KNIGHT);
+    }
+
+    return moveList;
   }
 
 
   template<Color Us, GenType Type>
-  MoveStack* generate_pawn_moves(const Position& pos, MoveStack* mlist,
-                                 Bitboard target, const CheckInfo* ci) {
+  ExtMove* generate_pawn_moves(const Position& pos, ExtMove* moveList, Bitboard target) {
 
-    // Compute our parametrized parameters at compile time, named according to
-    // the point of view of white side.
-    const Color    Them     = (Us == WHITE ? BLACK    : WHITE);
-    const Bitboard TRank8BB = (Us == WHITE ? Rank8BB  : Rank1BB);
-    const Bitboard TRank7BB = (Us == WHITE ? Rank7BB  : Rank2BB);
-    const Bitboard TRank3BB = (Us == WHITE ? Rank3BB  : Rank6BB);
-    const Square   UP       = (Us == WHITE ? DELTA_N  : DELTA_S);
-    const Square   RIGHT    = (Us == WHITE ? DELTA_NE : DELTA_SW);
-    const Square   LEFT     = (Us == WHITE ? DELTA_NW : DELTA_SE);
+    constexpr Color     Them     = ~Us;
+    constexpr Bitboard  TRank7BB = (Us == WHITE ? Rank7BB    : Rank2BB);
+    constexpr Bitboard  TRank3BB = (Us == WHITE ? Rank3BB    : Rank6BB);
+    constexpr Direction Up       = pawn_push(Us);
+    constexpr Direction UpRight  = (Us == WHITE ? NORTH_EAST : SOUTH_WEST);
+    constexpr Direction UpLeft   = (Us == WHITE ? NORTH_WEST : SOUTH_EAST);
 
-    Bitboard b1, b2, dc1, dc2, emptySquares;
+    const Bitboard emptySquares = ~pos.pieces();
+    const Bitboard enemies      =  Type == EVASIONS ? pos.checkers()
+                                                    : 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)
+    if constexpr (Type != CAPTURES)
     {
-        emptySquares = (Type == QUIETS || Type == QUIET_CHECKS ? target : ~pos.pieces());
+        Bitboard b1 = shift<Up>(pawnsNotOn7)   & emptySquares;
+        Bitboard b2 = shift<Up>(b1 & TRank3BB) & emptySquares;
 
-        b1 = move_pawns<UP>(pawnsNotOn7)   & emptySquares;
-        b2 = move_pawns<UP>(b1 & TRank3BB) & emptySquares;
-
-        if (Type == EVASIONS) // Consider only blocking squares
+        if constexpr (Type == EVASIONS) // Consider only blocking squares
         {
             b1 &= target;
             b2 &= target;
         }
 
-        if (Type == QUIET_CHECKS)
+        if constexpr (Type == QUIET_CHECKS)
         {
-            b1 &= pos.attacks_from<PAWN>(ci->ksq, Them);
-            b2 &= pos.attacks_from<PAWN>(ci->ksq, Them);
-
-            // 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 among captures.
-            if (pawnsNotOn7 & ci->dcCandidates)
-            {
-                dc1 = move_pawns<UP>(pawnsNotOn7 & ci->dcCandidates) & emptySquares & ~file_bb(ci->ksq);
-                dc2 = move_pawns<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);
         }
 
-        SERIALIZE_PAWNS(b1, UP);
-        SERIALIZE_PAWNS(b2, UP + UP);
+        while (b1)
+        {
+            Square to = pop_lsb(b1);
+            *moveList++ = make_move(to - Up, to);
+        }
+
+        while (b2)
+        {
+            Square to = pop_lsb(b2);
+            *moveList++ = make_move(to - Up - Up, to);
+        }
     }
 
     // Promotions and underpromotions
-    if (pawnsOn7 && (Type != EVASIONS || (target & TRank8BB)))
+    if (pawnsOn7)
     {
-        if (Type == CAPTURES)
-            emptySquares = ~pos.pieces();
+        Bitboard b1 = shift<UpRight>(pawnsOn7) & enemies;
+        Bitboard b2 = shift<UpLeft >(pawnsOn7) & enemies;
+        Bitboard b3 = shift<Up     >(pawnsOn7) & emptySquares;
 
-        if (Type == EVASIONS)
-            emptySquares &= target;
+        if constexpr (Type == EVASIONS)
+            b3 &= target;
 
-        mlist = generate_promotions<Type, RIGHT>(mlist, pawnsOn7, enemies, ci);
-        mlist = generate_promotions<Type, LEFT>(mlist, pawnsOn7, enemies, ci);
-        mlist = generate_promotions<Type, UP>(mlist, pawnsOn7, emptySquares, ci);
+        while (b1)
+            moveList = make_promotions<Type, UpRight, true>(moveList, pop_lsb(b1));
+
+        while (b2)
+            moveList = make_promotions<Type, UpLeft, true>(moveList, pop_lsb(b2));
+
+        while (b3)
+            moveList = make_promotions<Type, Up,    false>(moveList, pop_lsb(b3));
     }
 
-    // Standard and en-passant captures
-    if (Type == CAPTURES || Type == EVASIONS || Type == NON_EVASIONS)
+    // Standard and en passant captures
+    if constexpr (Type == CAPTURES || Type == EVASIONS || Type == NON_EVASIONS)
     {
-        b1 = move_pawns<RIGHT>(pawnsNotOn7) & enemies;
-        b2 = move_pawns<LEFT >(pawnsNotOn7) & enemies;
+        Bitboard b1 = shift<UpRight>(pawnsNotOn7) & enemies;
+        Bitboard b2 = shift<UpLeft >(pawnsNotOn7) & enemies;
 
-        SERIALIZE_PAWNS(b1, RIGHT);
-        SERIALIZE_PAWNS(b2, LEFT);
+        while (b1)
+        {
+            Square to = pop_lsb(b1);
+            *moveList++ = make_move(to - UpRight, to);
+        }
+
+        while (b2)
+        {
+            Square to = pop_lsb(b2);
+            *moveList++ = make_move(to - UpLeft, to);
+        }
 
         if (pos.ep_square() != SQ_NONE)
         {
             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)))
-                return mlist;
+            // An en passant capture cannot resolve a discovered check
+            if (Type == EVASIONS && (target & (pos.ep_square() + Up)))
+                return moveList;
 
-            b1 = pawnsNotOn7 & pos.attacks_from<PAWN>(pos.ep_square(), Them);
+            b1 = pawnsNotOn7 & pawn_attacks_bb(Them, pos.ep_square());
 
             assert(b1);
 
             while (b1)
-                (*mlist++).move = make<ENPASSANT>(pop_lsb(&b1), pos.ep_square());
+                *moveList++ = make<EN_PASSANT>(pop_lsb(b1), pos.ep_square());
         }
     }
 
-    return mlist;
+    return moveList;
   }
 
 
-  template<PieceType Pt, bool Checks> FORCE_INLINE
-  MoveStack* generate_moves(const Position& pos, MoveStack* mlist, Color us,
-                            Bitboard target, const CheckInfo* ci) {
+  template<Color Us, PieceType Pt, bool Checks>
+  ExtMove* generate_moves(const Position& pos, ExtMove* moveList, Bitboard target) {
 
-    assert(Pt != KING && Pt != PAWN);
+    static_assert(Pt != KING && Pt != PAWN, "Unsupported piece type in generate_moves()");
 
-    const Square* pl = pos.piece_list(us, Pt);
+    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)
-                && !(PseudoAttacks[Pt][from] & target & ci->checkSq[Pt]))
-                continue;
+        Square from = pop_lsb(bb);
+        Bitboard b = attacks_bb<Pt>(from, pos.pieces()) & target;
 
-            if (ci->dcCandidates && (ci->dcCandidates & from))
-                continue;
-        }
+        // 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);
 
-        Bitboard b = pos.attacks_from<Pt>(from) & target;
-
-        if (Checks)
-            b &= ci->checkSq[Pt];
-
-        SERIALIZE(b);
+        while (b)
+            *moveList++ = make_move(from, pop_lsb(b));
     }
 
-    return mlist;
+    return moveList;
   }
 
 
-  FORCE_INLINE MoveStack* generate_king_moves(const Position& pos, MoveStack* mlist,
-                                              Color us, Bitboard target) {
-    Square from = pos.king_square(us);
-    Bitboard b = pos.attacks_from<KING>(from) & target;
-    SERIALIZE(b);
-    return mlist;
-  }
+  template<Color Us, GenType Type>
+  ExtMove* generate_all(const Position& pos, ExtMove* moveList) {
 
+    static_assert(Type != LEGAL, "Unsupported type in generate_all()");
 
-  template<GenType Type> FORCE_INLINE
-  MoveStack* generate_all_moves(const Position& pos, MoveStack* mlist, Color us,
-                                Bitboard target, const CheckInfo* ci = NULL) {
+    constexpr bool Checks = Type == QUIET_CHECKS; // Reduce template instantiations
+    const Square ksq = pos.square<KING>(Us);
+    Bitboard target;
 
-    mlist = (us == WHITE ? generate_pawn_moves<WHITE, Type>(pos, mlist, target, ci)
-                         : generate_pawn_moves<BLACK, Type>(pos, mlist, target, ci));
-
-    mlist = generate_moves<KNIGHT, Type == QUIET_CHECKS>(pos, mlist, us, target, ci);
-    mlist = generate_moves<BISHOP, Type == QUIET_CHECKS>(pos, mlist, us, target, ci);
-    mlist = generate_moves<ROOK,   Type == QUIET_CHECKS>(pos, mlist, us, target, ci);
-    mlist = generate_moves<QUEEN,  Type == QUIET_CHECKS>(pos, mlist, us, target, ci);
-
-    if (Type != QUIET_CHECKS && Type != EVASIONS)
-        mlist = generate_king_moves(pos, mlist, us, target);
-
-    if (Type != CAPTURES && Type != EVASIONS && pos.can_castle(us))
+    // Skip generating non-king moves when in double check
+    if (Type != EVASIONS || !more_than_one(pos.checkers()))
     {
-        if (pos.is_chess960())
-        {
-            mlist = generate_castle<KING_SIDE,  Type == QUIET_CHECKS, true>(pos, mlist, us);
-            mlist = generate_castle<QUEEN_SIDE, Type == QUIET_CHECKS, true>(pos, mlist, us);
-        }
-        else
-        {
-            mlist = generate_castle<KING_SIDE,  Type == QUIET_CHECKS, false>(pos, mlist, us);
-            mlist = generate_castle<QUEEN_SIDE, Type == QUIET_CHECKS, false>(pos, mlist, us);
-        }
+        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);
     }
 
-    return mlist;
-  }
+    if (!Checks || pos.blockers_for_king(~Us) & ksq)
+    {
+        Bitboard b = attacks_bb<KING>(ksq) & (Type == EVASIONS ? ~pos.pieces(Us) : target);
+        if (Checks)
+            b &= ~attacks_bb<QUEEN>(pos.square<KING>(~Us));
 
+        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));
+    }
+
+    return moveList;
+  }
 
 } // namespace
 
 
-/// generate<CAPTURES> generates all pseudo-legal captures and queen
-/// promotions. Returns a pointer to the end of the move list.
+/// <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
 ///
-/// generate<QUIETS> generates all pseudo-legal non-captures and
-/// underpromotions. Returns a pointer to the end of the move list.
-///
-/// generate<NON_EVASIONS> generates all pseudo-legal captures and
-/// non-captures. Returns a pointer to the end of the move list.
+/// Returns a pointer to the end of the move list.
 
 template<GenType Type>
-MoveStack* generate(const Position& pos, MoveStack* mlist) {
+ExtMove* generate(const Position& pos, ExtMove* moveList) {
 
-  assert(Type == CAPTURES || Type == QUIETS || Type == NON_EVASIONS);
-  assert(!pos.in_check());
+  static_assert(Type != LEGAL, "Unsupported type in generate()");
+  assert((Type == EVASIONS) == (bool)pos.checkers());
 
   Color us = pos.side_to_move();
-  Bitboard target;
 
-  if (Type == CAPTURES)
-      target = pos.pieces(~us);
-
-  else if (Type == QUIETS)
-      target = ~pos.pieces();
-
-  else if (Type == NON_EVASIONS)
-      target = ~pos.pieces(us);
-
-  return generate_all_moves<Type>(pos, mlist, us, target);
+  return us == WHITE ? generate_all<WHITE, Type>(pos, moveList)
+                     : generate_all<BLACK, Type>(pos, moveList);
 }
 
 // Explicit template instantiations
-template MoveStack* generate<CAPTURES>(const Position&, MoveStack*);
-template MoveStack* generate<QUIETS>(const Position&, MoveStack*);
-template MoveStack* generate<NON_EVASIONS>(const Position&, MoveStack*);
-
-
-/// generate<QUIET_CHECKS> generates all pseudo-legal non-captures and knight
-/// underpromotions that give check. Returns a pointer to the end of the move list.
-template<>
-MoveStack* generate<QUIET_CHECKS>(const Position& pos, MoveStack* mlist) {
-
-  assert(!pos.in_check());
-
-  Color us = pos.side_to_move();
-  CheckInfo ci(pos);
-  Bitboard dc = ci.dcCandidates;
-
-  while (dc)
-  {
-     Square from = pop_lsb(&dc);
-     PieceType pt = type_of(pos.piece_on(from));
-
-     if (pt == PAWN)
-         continue; // Will be generated togheter with direct checks
-
-     Bitboard b = pos.attacks_from(Piece(pt), from) & ~pos.pieces();
-
-     if (pt == KING)
-         b &= ~PseudoAttacks[QUEEN][ci.ksq];
-
-     SERIALIZE(b);
-  }
-
-  return generate_all_moves<QUIET_CHECKS>(pos, mlist, us, ~pos.pieces(), &ci);
-}
-
-
-/// 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<>
-MoveStack* generate<EVASIONS>(const Position& pos, MoveStack* mlist) {
-
-  assert(pos.in_check());
-
-  Square from, checksq;
-  int checkersCnt = 0;
-  Color us = pos.side_to_move();
-  Square ksq = pos.king_square(us);
-  Bitboard sliderAttacks = 0;
-  Bitboard b = pos.checkers();
-
-  assert(pos.checkers());
-
-  // Find squares attacked by slider checkers, we will remove them from the king
-  // evasions so to skip known illegal moves avoiding useless legality check later.
-  do
-  {
-      checkersCnt++;
-      checksq = pop_lsb(&b);
-
-      assert(color_of(pos.piece_on(checksq)) == ~us);
-
-      switch (type_of(pos.piece_on(checksq)))
-      {
-      case BISHOP: sliderAttacks |= PseudoAttacks[BISHOP][checksq]; break;
-      case ROOK:   sliderAttacks |= PseudoAttacks[ROOK][checksq];   break;
-      case QUEEN:
-          // If queen and king are far or not on a diagonal line we can safely
-          // remove all the squares attacked in the other direction becuase are
-          // not reachable by the king anyway.
-          if (between_bb(ksq, checksq) || !(PseudoAttacks[BISHOP][checksq] & ksq))
-              sliderAttacks |= PseudoAttacks[QUEEN][checksq];
-
-          // Otherwise we need to use real rook attacks to check if king is safe
-          // to move in the other direction. For example: king in B2, queen in A1
-          // a knight in B1, and we can safely move to C1.
-          else
-              sliderAttacks |= PseudoAttacks[BISHOP][checksq] | pos.attacks_from<ROOK>(checksq);
-
-      default:
-          break;
-      }
-  } while (b);
-
-  // Generate evasions for king, capture and non capture moves
-  b = pos.attacks_from<KING>(ksq) & ~pos.pieces(us) & ~sliderAttacks;
-  from = ksq;
-  SERIALIZE(b);
-
-  if (checkersCnt > 1)
-      return mlist; // Double check, only a king move can save the day
-
-  // Generate blocking evasions or captures of the checking piece
-  Bitboard target = between_bb(checksq, ksq) | pos.checkers();
-
-  return generate_all_moves<EVASIONS>(pos, mlist, us, target);
-}
+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<LEGAL> generates all the legal moves in the given position
 
 template<>
-MoveStack* generate<LEGAL>(const Position& pos, MoveStack* mlist) {
+ExtMove* generate<LEGAL>(const Position& pos, ExtMove* moveList) {
 
-  MoveStack *end, *cur = mlist;
-  Bitboard pinned = pos.pinned_pieces();
-  Square ksq = pos.king_square(pos.side_to_move());
+  Color us = pos.side_to_move();
+  Bitboard pinned = pos.blockers_for_king(us) & pos.pieces(us);
+  Square ksq = pos.square<KING>(us);
+  ExtMove* cur = moveList;
 
-  end = pos.in_check() ? generate<EVASIONS>(pos, mlist)
-                       : generate<NON_EVASIONS>(pos, mlist);
-  while (cur != end)
-      if (   (pinned || from_sq(cur->move) == ksq || type_of(cur->move) == ENPASSANT)
-          && !pos.pl_move_is_legal(cur->move, pinned))
-          cur->move = (--end)->move;
+  moveList = pos.checkers() ? generate<EVASIONS    >(pos, moveList)
+                            : generate<NON_EVASIONS>(pos, moveList);
+  while (cur != moveList)
+      if (  ((pinned & from_sq(*cur)) || from_sq(*cur) == ksq || type_of(*cur) == EN_PASSANT)
+          && !pos.legal(*cur))
+          *cur = (--moveList)->move;
       else
-          cur++;
+          ++cur;
 
-  return end;
+  return moveList;
 }
+
+} // namespace Stockfish

src/movegen.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,11 +16,17 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(MOVEGEN_H_INCLUDED)
+#ifndef MOVEGEN_H_INCLUDED
 #define MOVEGEN_H_INCLUDED
 
+#include <algorithm>
+
 #include "types.h"
 
+namespace Stockfish {
+
+class Position;
+
 enum GenType {
   CAPTURES,
   QUIETS,
@@ -31,25 +36,42 @@ enum GenType {
   LEGAL
 };
 
-class Position;
+struct ExtMove {
+  Move move;
+  int value;
+
+  operator Move() const { return move; }
+  void operator=(Move m) { move = m; }
+
+  // Inhibit unwanted implicit conversions to Move
+  // with an ambiguity that yields to a compile error.
+  operator float() const = delete;
+};
+
+inline bool operator<(const ExtMove& f, const ExtMove& s) {
+  return f.value < s.value;
+}
 
 template<GenType>
-MoveStack* generate(const Position& pos, MoveStack* mlist);
+ExtMove* generate(const Position& pos, ExtMove* moveList);
 
-/// The MoveList struct is a simple wrapper around generate(), sometimes comes
-/// handy to use this class instead of the low level generate() function.
+/// The MoveList struct is a simple wrapper around generate(). It sometimes comes
+/// in handy to use this class instead of the low level generate() function.
 template<GenType T>
 struct MoveList {
 
-  explicit MoveList(const Position& pos) : cur(mlist), last(generate<T>(pos, mlist)) {}
-  void operator++() { cur++; }
-  bool end() const { return cur == last; }
-  Move move() const { return cur->move; }
-  size_t size() const { return last - mlist; }
+  explicit MoveList(const Position& pos) : last(generate<T>(pos, moveList)) {}
+  const ExtMove* begin() const { return moveList; }
+  const ExtMove* end() const { return last; }
+  size_t size() const { return last - moveList; }
+  bool contains(Move move) const {
+    return std::find(begin(), end(), move) != end();
+  }
 
 private:
-  MoveStack mlist[MAX_MOVES];
-  MoveStack *cur, *last;
+  ExtMove moveList[MAX_MOVES], *last;
 };
 
-#endif // !defined(MOVEGEN_H_INCLUDED)
+} // namespace Stockfish
+
+#endif // #ifndef MOVEGEN_H_INCLUDED

src/movepick.cpp

@@ -1,14 +1,12 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -18,347 +16,280 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#include <algorithm>
 #include <cassert>
 
-#include "movegen.h"
+#include "bitboard.h"
 #include "movepick.h"
-#include "thread.h"
+
+namespace Stockfish {
 
 namespace {
 
-  enum Sequencer {
-    MAIN_SEARCH, CAPTURES_S1, KILLERS_S1, QUIETS_1_S1, QUIETS_2_S1, BAD_CAPTURES_S1,
-    EVASION,     EVASIONS_S2,
-    QSEARCH_0,   CAPTURES_S3, QUIET_CHECKS_S3,
-    QSEARCH_1,   CAPTURES_S4,
-    PROBCUT,     CAPTURES_S5,
-    RECAPTURE,   CAPTURES_S6,
-    STOP
+  enum Stages {
+    MAIN_TT, CAPTURE_INIT, GOOD_CAPTURE, REFUTATION, QUIET_INIT, QUIET, BAD_CAPTURE,
+    EVASION_TT, EVASION_INIT, EVASION,
+    PROBCUT_TT, PROBCUT_INIT, PROBCUT,
+    QSEARCH_TT, QCAPTURE_INIT, QCAPTURE, QCHECK_INIT, QCHECK
   };
 
-  // Unary predicate used by std::partition to split positive scores from remaining
-  // ones so to sort separately the two sets, and with the second sort delayed.
-  inline bool has_positive_score(const MoveStack& ms) { return ms.score > 0; }
+  // partial_insertion_sort() sorts moves in descending order up to and including
+  // a given limit. The order of moves smaller than the limit is left unspecified.
+  void partial_insertion_sort(ExtMove* begin, ExtMove* end, int limit) {
 
-  // Picks and moves to the front the best move in the range [begin, end),
-  // it is faster than sorting all the moves in advance when moves are few, as
-  // normally are the possible captures.
-  inline MoveStack* pick_best(MoveStack* begin, MoveStack* end)
-  {
-      std::swap(*begin, *std::max_element(begin, end));
-      return begin;
+    for (ExtMove *sortedEnd = begin, *p = begin + 1; p < end; ++p)
+        if (p->value >= limit)
+        {
+            ExtMove tmp = *p, *q;
+            *p = *++sortedEnd;
+            for (q = sortedEnd; q != begin && *(q - 1) < tmp; --q)
+                *q = *(q - 1);
+            *q = tmp;
+        }
   }
-}
+
+} // namespace
 
 
 /// Constructors of the MovePicker class. As arguments we pass information
-/// to help it to return the presumably good moves first, to decide which
+/// to help it to return the (presumably) good moves first, to decide which
 /// moves to return (in the quiescence search, for instance, we only want to
-/// search captures, promotions and some checks) and about how important good
-/// move ordering is at the current node.
+/// search captures, promotions, and some checks) and how important good move
+/// ordering is at the current node.
 
-MovePicker::MovePicker(const Position& p, Move ttm, Depth d, const History& h,
-                       Search::Stack* s, Value beta) : pos(p), H(h), depth(d) {
+/// MovePicker constructor for the main search
+MovePicker::MovePicker(const Position& p, Move ttm, Depth d, const ButterflyHistory* mh,
+                                                             const CapturePieceToHistory* cph,
+                                                             const PieceToHistory** ch,
+                                                             Move cm,
+                                                             const Move* killers)
+           : pos(p), mainHistory(mh), captureHistory(cph), continuationHistory(ch),
+             ttMove(ttm), refutations{{killers[0], 0}, {killers[1], 0}, {cm, 0}}, depth(d)
+{
+  assert(d > 0);
 
-  assert(d > DEPTH_ZERO);
-
-  captureThreshold = 0;
-  cur = end = moves;
-  endBadCaptures = moves + MAX_MOVES - 1;
-  ss = s;
-
-  if (p.in_check())
-      phase = EVASION;
-
-  else
-  {
-      phase = MAIN_SEARCH;
-
-      killers[0].move = ss->killers[0];
-      killers[1].move = ss->killers[1];
-
-      // Consider sligtly negative captures as good if at low depth and far from beta
-      if (ss && ss->eval < beta - PawnValueMg && d < 3 * ONE_PLY)
-          captureThreshold = -PawnValueMg;
-
-      // Consider negative captures as good if still enough to reach beta
-      else if (ss && ss->eval > beta)
-          captureThreshold = beta - ss->eval;
-  }
-
-  ttMove = (ttm && pos.is_pseudo_legal(ttm) ? ttm : MOVE_NONE);
-  end += (ttMove != MOVE_NONE);
+  stage = (pos.checkers() ? EVASION_TT : MAIN_TT) +
+          !(ttm && pos.pseudo_legal(ttm));
 }
 
-MovePicker::MovePicker(const Position& p, Move ttm, Depth d, const History& h,
-                       Square sq) : pos(p), H(h), cur(moves), end(moves) {
+/// MovePicker constructor for quiescence search
+MovePicker::MovePicker(const Position& p, Move ttm, Depth d, const ButterflyHistory* mh,
+                                                             const CapturePieceToHistory* cph,
+                                                             const PieceToHistory** ch,
+                                                             Square rs)
+           : pos(p), mainHistory(mh), captureHistory(cph), continuationHistory(ch), ttMove(ttm), recaptureSquare(rs), depth(d)
+{
+  assert(d <= 0);
 
-  assert(d <= DEPTH_ZERO);
-
-  if (p.in_check())
-      phase = EVASION;
-
-  else if (d > DEPTH_QS_NO_CHECKS)
-      phase = QSEARCH_0;
-
-  else if (d > DEPTH_QS_RECAPTURES)
-  {
-      phase = QSEARCH_1;
-
-      // Skip TT move if is not a capture or a promotion, this avoids qsearch
-      // tree explosion due to a possible perpetual check or similar rare cases
-      // when TT table is full.
-      if (ttm && !pos.is_capture_or_promotion(ttm))
-          ttm = MOVE_NONE;
-  }
-  else
-  {
-      phase = RECAPTURE;
-      recaptureSquare = sq;
-      ttm = MOVE_NONE;
-  }
-
-  ttMove = (ttm && pos.is_pseudo_legal(ttm) ? ttm : MOVE_NONE);
-  end += (ttMove != MOVE_NONE);
+  stage = (pos.checkers() ? EVASION_TT : QSEARCH_TT) +
+          !(   ttm
+            && pos.pseudo_legal(ttm));
 }
 
-MovePicker::MovePicker(const Position& p, Move ttm, const History& h, PieceType pt)
-                       : pos(p), H(h), cur(moves), end(moves) {
+/// MovePicker constructor for ProbCut: we generate captures with SEE greater
+/// than or equal to the given threshold.
+MovePicker::MovePicker(const Position& p, Move ttm, Value th, const CapturePieceToHistory* cph)
+           : pos(p), captureHistory(cph), ttMove(ttm), threshold(th)
+{
+  assert(!pos.checkers());
 
-  assert(!pos.in_check());
-
-  phase = PROBCUT;
-
-  // In ProbCut we generate only captures better than parent's captured piece
-  captureThreshold = PieceValue[Mg][pt];
-  ttMove = (ttm && pos.is_pseudo_legal(ttm) ? ttm : MOVE_NONE);
-
-  if (ttMove && (!pos.is_capture(ttMove) ||  pos.see(ttMove) <= captureThreshold))
-      ttMove = MOVE_NONE;
-
-  end += (ttMove != MOVE_NONE);
+  stage = PROBCUT_TT + !(ttm && pos.capture_stage(ttm)
+                             && pos.pseudo_legal(ttm)
+                             && pos.see_ge(ttm, threshold));
 }
 
+/// MovePicker::score() assigns a numerical value to each move in a list, used
+/// for sorting. Captures are ordered by Most Valuable Victim (MVV), preferring
+/// captures with a good history. Quiets moves are ordered using the history tables.
+template<GenType Type>
+void MovePicker::score() {
 
-/// MovePicker::score_captures(), MovePicker::score_noncaptures() and
-/// MovePicker::score_evasions() assign a numerical move ordering score
-/// to each move in a move list.  The moves with highest scores will be
-/// picked first by next_move().
+  static_assert(Type == CAPTURES || Type == QUIETS || Type == EVASIONS, "Wrong type");
 
-void MovePicker::score_captures() {
-  // Winning and equal captures in the main search are ordered by MVV/LVA.
-  // Suprisingly, this appears to perform slightly better than SEE based
-  // move ordering. The reason is probably that in a position with a winning
-  // capture, capturing a more valuable (but sufficiently defended) piece
-  // first usually doesn't hurt. The opponent will have to recapture, and
-  // the hanging piece will still be hanging (except in the unusual cases
-  // where it is possible to recapture with the hanging piece). Exchanging
-  // big pieces before capturing a hanging piece probably helps to reduce
-  // the subtree size.
-  // In main search we want to push captures with negative SEE values to
-  // badCaptures[] array, but instead of doing it now we delay till when
-  // the move has been picked up in pick_move_from_list(), this way we save
-  // some SEE calls in case we get a cutoff (idea from Pablo Vazquez).
-  Move m;
-
-  for (MoveStack* it = moves; it != end; ++it)
+  [[maybe_unused]] Bitboard threatenedByPawn, threatenedByMinor, threatenedByRook, threatenedPieces;
+  if constexpr (Type == QUIETS)
   {
-      m = it->move;
-      it->score =  PieceValue[Mg][pos.piece_on(to_sq(m))]
-                 - type_of(pos.piece_moved(m));
+      Color us = pos.side_to_move();
 
-      if (type_of(m) == PROMOTION)
-          it->score += PieceValue[Mg][promotion_type(m)];
+      threatenedByPawn  = pos.attacks_by<PAWN>(~us);
+      threatenedByMinor = pos.attacks_by<KNIGHT>(~us) | pos.attacks_by<BISHOP>(~us) | threatenedByPawn;
+      threatenedByRook  = pos.attacks_by<ROOK>(~us) | threatenedByMinor;
+
+      // Pieces threatened by pieces of lesser material value
+      threatenedPieces = (pos.pieces(us, QUEEN) & threatenedByRook)
+                       | (pos.pieces(us, ROOK)  & threatenedByMinor)
+                       | (pos.pieces(us, KNIGHT, BISHOP) & threatenedByPawn);
   }
+
+  for (auto& m : *this)
+      if constexpr (Type == CAPTURES)
+          m.value =  (7 * int(PieceValue[MG][pos.piece_on(to_sq(m))])
+                   +     (*captureHistory)[pos.moved_piece(m)][to_sq(m)][type_of(pos.piece_on(to_sq(m)))]) / 16;
+
+      else if constexpr (Type == QUIETS)
+          m.value =  2 * (*mainHistory)[pos.side_to_move()][from_to(m)]
+                   + 2 * (*continuationHistory[0])[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)]
+                   +     (threatenedPieces & from_sq(m) ?
+                           (type_of(pos.moved_piece(m)) == QUEEN && !(to_sq(m) & threatenedByRook)  ? 50000
+                          : type_of(pos.moved_piece(m)) == ROOK  && !(to_sq(m) & threatenedByMinor) ? 25000
+                          :                                         !(to_sq(m) & threatenedByPawn)  ? 15000
+                          :                                                                           0)
+                          :                                                                           0)
+                   +     bool(pos.check_squares(type_of(pos.moved_piece(m))) & to_sq(m)) * 16384;
+      else // Type == EVASIONS
+      {
+          if (pos.capture_stage(m))
+              m.value =  PieceValue[MG][pos.piece_on(to_sq(m))]
+                       - Value(type_of(pos.moved_piece(m)))
+                       + (1 << 28);
+          else
+              m.value =  (*mainHistory)[pos.side_to_move()][from_to(m)]
+                       + (*continuationHistory[0])[pos.moved_piece(m)][to_sq(m)];
+      }
 }
 
-void MovePicker::score_noncaptures() {
+/// MovePicker::select() returns the next move satisfying a predicate function.
+/// It never returns the TT move.
+template<MovePicker::PickType T, typename Pred>
+Move MovePicker::select(Pred filter) {
 
-  Move m;
-
-  for (MoveStack* it = moves; it != end; ++it)
+  while (cur < endMoves)
   {
-      m = it->move;
-      it->score = H.value(pos.piece_moved(m), to_sq(m));
+      if constexpr (T == Best)
+          std::swap(*cur, *std::max_element(cur, endMoves));
+
+      if (*cur != ttMove && filter())
+          return *cur++;
+
+      cur++;
   }
+  return MOVE_NONE;
 }
 
-void MovePicker::score_evasions() {
-  // Try good captures ordered by MVV/LVA, then non-captures if destination square
-  // is not under attack, ordered by history value, then bad-captures and quiet
-  // moves with a negative SEE. This last group is ordered by the SEE score.
-  Move m;
-  int seeScore;
+/// 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
+/// moves left, picking the move with the highest score from a list of generated moves.
+Move MovePicker::next_move(bool skipQuiets) {
 
-  if (end < moves + 2)
-      return;
+top:
+  switch (stage) {
 
-  for (MoveStack* it = moves; it != end; ++it)
-  {
-      m = it->move;
-      if ((seeScore = pos.see_sign(m)) < 0)
-          it->score = seeScore - History::MaxValue; // Be sure we are at the bottom
-      else if (pos.is_capture(m))
-          it->score =  PieceValue[Mg][pos.piece_on(to_sq(m))]
-                     - type_of(pos.piece_moved(m)) + History::MaxValue;
-      else
-          it->score = H.value(pos.piece_moved(m), to_sq(m));
-  }
-}
+  case MAIN_TT:
+  case EVASION_TT:
+  case QSEARCH_TT:
+  case PROBCUT_TT:
+      ++stage;
+      return ttMove;
 
+  case CAPTURE_INIT:
+  case PROBCUT_INIT:
+  case QCAPTURE_INIT:
+      cur = endBadCaptures = moves;
+      endMoves = generate<CAPTURES>(pos, cur);
 
-/// MovePicker::generate_next() generates, scores and sorts the next bunch of moves,
-/// when there are no more moves to try for the current phase.
+      score<CAPTURES>();
+      partial_insertion_sort(cur, endMoves, std::numeric_limits<int>::min());
+      ++stage;
+      goto top;
 
-void MovePicker::generate_next() {
+  case GOOD_CAPTURE:
+      if (select<Next>([&](){
+                       return pos.see_ge(*cur, Value(-cur->value)) ?
+                              // Move losing capture to endBadCaptures to be tried later
+                              true : (*endBadCaptures++ = *cur, false); }))
+          return *(cur - 1);
 
-  cur = moves;
+      // Prepare the pointers to loop over the refutations array
+      cur = std::begin(refutations);
+      endMoves = std::end(refutations);
 
-  switch (++phase) {
+      // If the countermove is the same as a killer, skip it
+      if (   refutations[0].move == refutations[2].move
+          || refutations[1].move == refutations[2].move)
+          --endMoves;
 
-  case CAPTURES_S1: case CAPTURES_S3: case CAPTURES_S4: case CAPTURES_S5: case CAPTURES_S6:
-      end = generate<CAPTURES>(pos, moves);
-      score_captures();
-      return;
+      ++stage;
+      [[fallthrough]];
 
-  case KILLERS_S1:
-      cur = killers;
-      end = cur + 2;
-      return;
+  case REFUTATION:
+      if (select<Next>([&](){ return    *cur != MOVE_NONE
+                                    && !pos.capture_stage(*cur)
+                                    &&  pos.pseudo_legal(*cur); }))
+          return *(cur - 1);
+      ++stage;
+      [[fallthrough]];
 
-  case QUIETS_1_S1:
-      endQuiets = end = generate<QUIETS>(pos, moves);
-      score_noncaptures();
-      end = std::partition(cur, end, has_positive_score);
-      sort<MoveStack>(cur, end);
-      return;
+  case QUIET_INIT:
+      if (!skipQuiets)
+      {
+          cur = endBadCaptures;
+          endMoves = generate<QUIETS>(pos, cur);
 
-  case QUIETS_2_S1:
-      cur = end;
-      end = endQuiets;
-      if (depth >= 3 * ONE_PLY)
-          sort<MoveStack>(cur, end);
-      return;
+          score<QUIETS>();
+          partial_insertion_sort(cur, endMoves, -3000 * depth);
+      }
 
-  case BAD_CAPTURES_S1:
-      // Just pick them in reverse order to get MVV/LVA ordering
-      cur = moves + MAX_MOVES - 1;
-      end = endBadCaptures;
-      return;
+      ++stage;
+      [[fallthrough]];
 
-  case EVASIONS_S2:
-      end = generate<EVASIONS>(pos, moves);
-      score_evasions();
-      return;
+  case QUIET:
+      if (   !skipQuiets
+          && select<Next>([&](){return   *cur != refutations[0].move
+                                      && *cur != refutations[1].move
+                                      && *cur != refutations[2].move;}))
+          return *(cur - 1);
 
-  case QUIET_CHECKS_S3:
-      end = generate<QUIET_CHECKS>(pos, moves);
-      return;
+      // Prepare the pointers to loop over the bad captures
+      cur = moves;
+      endMoves = endBadCaptures;
 
-  case EVASION: case QSEARCH_0: case QSEARCH_1: case PROBCUT: case RECAPTURE:
-      phase = STOP;
-  case STOP:
-      end = cur + 1; // Avoid another next_phase() call
-      return;
+      ++stage;
+      [[fallthrough]];
 
-  default:
-      assert(false);
-  }
-}
+  case BAD_CAPTURE:
+      return select<Next>([](){ return true; });
 
+  case EVASION_INIT:
+      cur = moves;
+      endMoves = generate<EVASIONS>(pos, cur);
 
-/// 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 moves left. It picks the move with the biggest score from a list
-/// of generated moves taking care not to return the tt move if has already been
-/// searched previously.
-template<>
-Move MovePicker::next_move<false>() {
+      score<EVASIONS>();
+      ++stage;
+      [[fallthrough]];
 
-  Move move;
+  case EVASION:
+      return select<Best>([](){ return true; });
 
-  while (true)
-  {
-      while (cur == end)
-          generate_next();
+  case PROBCUT:
+      return select<Next>([&](){ return pos.see_ge(*cur, threshold); });
 
-      switch (phase) {
+  case QCAPTURE:
+      if (select<Next>([&](){ return   depth > DEPTH_QS_RECAPTURES
+                                    || to_sq(*cur) == recaptureSquare; }))
+          return *(cur - 1);
 
-      case MAIN_SEARCH: case EVASION: case QSEARCH_0: case QSEARCH_1: case PROBCUT:
-          cur++;
-          return ttMove;
-
-      case CAPTURES_S1:
-          move = pick_best(cur++, end)->move;
-          if (move != ttMove)
-          {
-              assert(captureThreshold <= 0); // Otherwise we cannot use see_sign()
-
-              if (pos.see_sign(move) >= captureThreshold)
-                  return move;
-
-              // Losing capture, move it to the tail of the array
-              (endBadCaptures--)->move = move;
-          }
-          break;
-
-      case KILLERS_S1:
-          move = (cur++)->move;
-          if (    move != MOVE_NONE
-              &&  pos.is_pseudo_legal(move)
-              &&  move != ttMove
-              && !pos.is_capture(move))
-              return move;
-          break;
-
-      case QUIETS_1_S1: case QUIETS_2_S1:
-          move = (cur++)->move;
-          if (   move != ttMove
-              && move != killers[0].move
-              && move != killers[1].move)
-              return move;
-          break;
-
-      case BAD_CAPTURES_S1:
-          return (cur--)->move;
-
-      case EVASIONS_S2: case CAPTURES_S3: case CAPTURES_S4:
-          move = pick_best(cur++, end)->move;
-          if (move != ttMove)
-              return move;
-          break;
-
-      case CAPTURES_S5:
-           move = pick_best(cur++, end)->move;
-           if (move != ttMove && pos.see(move) > captureThreshold)
-               return move;
-           break;
-
-      case CAPTURES_S6:
-          move = pick_best(cur++, end)->move;
-          if (to_sq(move) == recaptureSquare)
-              return move;
-          break;
-
-      case QUIET_CHECKS_S3:
-          move = (cur++)->move;
-          if (move != ttMove)
-              return move;
-          break;
-
-      case STOP:
+      // If we did not find any move and we do not try checks, we have finished
+      if (depth != DEPTH_QS_CHECKS)
           return MOVE_NONE;
 
-      default:
-          assert(false);
-      }
+      ++stage;
+      [[fallthrough]];
+
+  case QCHECK_INIT:
+      cur = moves;
+      endMoves = generate<QUIET_CHECKS>(pos, cur);
+
+      ++stage;
+      [[fallthrough]];
+
+  case QCHECK:
+      return select<Next>([](){ return true; });
   }
+
+  assert(false);
+  return MOVE_NONE; // Silence warning
 }
 
-
-/// Version of next_move() to use at split point nodes where the move is grabbed
-/// from the split point's shared MovePicker object. This function is not thread
-/// safe so should be lock protected by the caller.
-template<>
-Move MovePicker::next_move<true>() { return ss->sp->mp->next_move<false>(); }
+} // namespace Stockfish

src/movepick.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,48 +16,140 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined MOVEPICK_H_INCLUDED
+#ifndef MOVEPICK_H_INCLUDED
 #define MOVEPICK_H_INCLUDED
 
-#include "history.h"
+#include <array>
+#include <limits>
+#include <type_traits>
+
+#include "movegen.h"
 #include "position.h"
-#include "search.h"
 #include "types.h"
 
+namespace Stockfish {
 
-/// 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.
+/// 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
+/// tables at caller sites as simple multi-dim arrays.
+template<typename T, int D>
+class StatsEntry {
 
-class MovePicker {
-
-  MovePicker& operator=(const MovePicker&); // Silence a warning under MSVC
+  T entry;
 
 public:
-  MovePicker(const Position&, Move, Depth, const History&, Search::Stack*, Value);
-  MovePicker(const Position&, Move, Depth, const History&, Square);
-  MovePicker(const Position&, Move, const History&, PieceType);
-  template<bool SpNode> Move next_move();
+  void operator=(const T& v) { entry = v; }
+  T* operator&() { return &entry; }
+  T* operator->() { return &entry; }
+  operator const T&() const { return entry; }
 
-private:
-  void score_captures();
-  void score_noncaptures();
-  void score_evasions();
-  void generate_next();
+  void operator<<(int bonus) {
+    assert(abs(bonus) <= D); // Ensure range is [-D, D]
+    static_assert(D <= std::numeric_limits<T>::max(), "D overflows T");
 
-  const Position& pos;
-  const History& H;
-  Search::Stack* ss;
-  Depth depth;
-  Move ttMove;
-  MoveStack killers[2];
-  Square recaptureSquare;
-  int captureThreshold, phase;
-  MoveStack *cur, *end, *endQuiets, *endBadCaptures;
-  MoveStack moves[MAX_MOVES];
+    entry += bonus - entry * abs(bonus) / D;
+
+    assert(abs(entry) <= D);
+  }
 };
 
-#endif // !defined(MOVEPICK_H_INCLUDED)
+/// Stats is a generic N-dimensional array used to store various statistics.
+/// The first template parameter T is the base type of the array, the second
+/// template parameter D limits the range of updates in [-D, D] when we update
+/// values with the << operator, while the last parameters (Size and Sizes)
+/// encode the dimensions of the array.
+template <typename T, int D, int Size, int... Sizes>
+struct Stats : public std::array<Stats<T, D, Sizes...>, Size>
+{
+  using stats = Stats<T, D, Size, Sizes...>;
+
+  void fill(const T& v) {
+
+    // For standard-layout 'this' points to first struct member
+    assert(std::is_standard_layout<stats>::value);
+
+    using entry = StatsEntry<T, D>;
+    entry* p = reinterpret_cast<entry*>(this);
+    std::fill(p, p + sizeof(*this) / sizeof(entry), v);
+  }
+};
+
+template <typename T, int D, int Size>
+struct Stats<T, D, Size> : public std::array<StatsEntry<T, D>, Size> {};
+
+/// In stats table, D=0 means that the template parameter is not used
+enum StatsParams { NOT_USED = 0 };
+enum StatsType { NoCaptures, Captures };
+
+/// ButterflyHistory records how often quiet moves have been successful or
+/// 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
+/// (~11 elo)
+using ButterflyHistory = Stats<int16_t, 7183, COLOR_NB, int(SQUARE_NB) * int(SQUARE_NB)>;
+
+/// CounterMoveHistory stores counter moves indexed by [piece][to] of the previous
+/// move, see www.chessprogramming.org/Countermove_Heuristic
+using CounterMoveHistory = Stats<Move, NOT_USED, PIECE_NB, SQUARE_NB>;
+
+/// CapturePieceToHistory is addressed by a move's [piece][to][captured piece type]
+using CapturePieceToHistory = Stats<int16_t, 10692, PIECE_NB, SQUARE_NB, PIECE_TYPE_NB>;
+
+/// PieceToHistory is like ButterflyHistory but is addressed by a move's [piece][to]
+using PieceToHistory = Stats<int16_t, 29952, PIECE_NB, SQUARE_NB>;
+
+/// ContinuationHistory is the combined history of a given pair of moves, usually
+/// the current one given a previous one. The nested history table is based on
+/// PieceToHistory instead of ButterflyBoards.
+/// (~63 elo)
+using ContinuationHistory = Stats<PieceToHistory, NOT_USED, PIECE_NB, SQUARE_NB>;
+
+
+/// 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.
+class MovePicker {
+
+  enum PickType { Next, Best };
+
+public:
+  MovePicker(const MovePicker&) = delete;
+  MovePicker& operator=(const MovePicker&) = delete;
+  MovePicker(const Position&, Move, Depth, const ButterflyHistory*,
+                                           const CapturePieceToHistory*,
+                                           const PieceToHistory**,
+                                           Move,
+                                           const Move*);
+  MovePicker(const Position&, Move, Depth, const ButterflyHistory*,
+                                           const CapturePieceToHistory*,
+                                           const PieceToHistory**,
+                                           Square);
+  MovePicker(const Position&, Move, Value, const CapturePieceToHistory*);
+  Move next_move(bool skipQuiets = false);
+
+private:
+  template<PickType T, typename Pred> Move select(Pred);
+  template<GenType> void score();
+  ExtMove* begin() { return cur; }
+  ExtMove* end() { return endMoves; }
+
+  const Position& pos;
+  const ButterflyHistory* mainHistory;
+  const CapturePieceToHistory* captureHistory;
+  const PieceToHistory** continuationHistory;
+  Move ttMove;
+  ExtMove refutations[3], *cur, *endMoves, *endBadCaptures;
+  int stage;
+  Square recaptureSquare;
+  Value threshold;
+  Depth depth;
+  ExtMove moves[MAX_MOVES];
+};
+
+} // namespace Stockfish
+
+#endif // #ifndef MOVEPICK_H_INCLUDED

src/nnue/evaluate_nnue.cpp

@@ -0,0 +1,405 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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/>.
+*/
+
+// Code for calculating NNUE evaluation function
+
+#include <fstream>
+#include <iomanip>
+#include <iostream>
+#include <set>
+#include <sstream>
+#include <string_view>
+
+#include "../evaluate.h"
+#include "../position.h"
+#include "../uci.h"
+#include "../types.h"
+
+#include "evaluate_nnue.h"
+
+namespace Stockfish::Eval::NNUE {
+
+  // Input feature converter
+  LargePagePtr<FeatureTransformer> featureTransformer;
+
+  // Evaluation function
+  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) {
+
+    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 read_parameters(std::istream& stream, T& reference) {
+
+    std::uint32_t header;
+    header = read_little_endian<std::uint32_t>(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
+  static void initialize() {
+
+    Detail::initialize(featureTransformer);
+    for (std::size_t i = 0; i < LayerStacks; ++i)
+      Detail::initialize(network[i]);
+  }
+
+  // Read network header
+  static 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);
+    *hashValue  = read_little_endian<std::uint32_t>(stream);
+    size        = read_little_endian<std::uint32_t>(stream);
+    if (!stream || version != Version) return false;
+    desc->resize(size);
+    stream.read(&(*desc)[0], size);
+    return !stream.fail();
+  }
+
+  // Write network header
+  static 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, (std::uint32_t)desc.size());
+    stream.write(&desc[0], desc.size());
+    return !stream.fail();
+  }
+
+  // Read network parameters
+  static bool read_parameters(std::istream& stream) {
+
+    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();
+  }
+
+  // Write network parameters
+  static bool write_parameters(std::ostream& 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;
+  }
+
+  void hint_common_parent_position(const Position& pos) {
+    if (Eval::useNNUE)
+        featureTransformer->hint_common_access(pos);
+  }
+
+  // Evaluation function. Perform differential calculation.
+  Value evaluate(const Position& pos, bool adjusted, int* complexity) {
+
+    // 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;
+    constexpr int delta = 24;
+
+#if defined(ALIGNAS_ON_STACK_VARIABLES_BROKEN)
+    TransformedFeatureType transformedFeaturesUnaligned[
+      FeatureTransformer::BufferSize + alignment / sizeof(TransformedFeatureType)];
+
+    auto* transformedFeatures = align_ptr_up<alignment>(&transformedFeaturesUnaligned[0]);
+#else
+    alignas(alignment)
+      TransformedFeatureType transformedFeatures[FeatureTransformer::BufferSize];
+#endif
+
+    ASSERT_ALIGNED(transformedFeatures, alignment);
+
+    const int bucket = (pos.count<ALL_PIECES>() - 1) / 4;
+    const auto psqt = featureTransformer->transform(pos, transformedFeatures, bucket);
+    const auto positional = network[bucket]->propagate(transformedFeatures);
+
+    if (complexity)
+        *complexity = abs(psqt - positional) / OutputScale;
+
+    // Give more value to positional evaluation when adjusted flag is set
+    if (adjusted)
+        return static_cast<Value>(((1024 - delta) * psqt + (1024 + delta) * positional) / (1024 * OutputScale));
+    else
+        return static_cast<Value>((psqt + positional) / OutputScale);
+  }
+
+  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)];
+
+    auto* transformedFeatures = align_ptr_up<alignment>(&transformedFeaturesUnaligned[0]);
+#else
+    alignas(alignment)
+      TransformedFeatureType transformedFeatures[FeatureTransformer::BufferSize];
+#endif
+
+    ASSERT_ALIGNED(transformedFeatures, alignment);
+
+    NnueEvalTrace t{};
+    t.correctBucket = (pos.count<ALL_PIECES>() - 1) / 4;
+    for (IndexType bucket = 0; bucket < LayerStacks; ++bucket) {
+      const auto materialist = featureTransformer->transform(pos, transformedFeatures, bucket);
+      const auto positional = network[bucket]->propagate(transformedFeatures);
+
+      t.psqt[bucket] = static_cast<Value>( materialist / OutputScale );
+      t.positional[bucket] = static_cast<Value>( positional / OutputScale );
+    }
+
+    return t;
+  }
+
+  constexpr std::string_view PieceToChar(" PNBRQK  pnbrqk");
+
+
+  // format_cp_compact() converts a Value into (centi)pawns and writes it in a buffer.
+  // The buffer must have capacity for at least 5 chars.
+  static void format_cp_compact(Value v, char* buffer) {
+
+    buffer[0] = (v < 0 ? '-' : v > 0 ? '+' : ' ');
+
+    int cp = std::abs(100 * v / UCI::NormalizeToPawnValue);
+    if (cp >= 10000)
+    {
+        buffer[1] = '0' + cp / 10000; cp %= 10000;
+        buffer[2] = '0' + cp / 1000; cp %= 1000;
+        buffer[3] = '0' + 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;
+    }
+  }
+
+
+  // format_cp_aligned_dot() converts a Value into (centi)pawns, always keeping two decimals.
+  static void format_cp_aligned_dot(Value v, std::stringstream &stream) {
+    const double cp = 1.0 * std::abs(int(v)) / UCI::NormalizeToPawnValue;
+
+    stream << (v < 0 ? '-' : v > 0 ? '+' : ' ')
+           << std::setiosflags(std::ios::fixed)
+           << std::setw(6)
+           << std::setprecision(2)
+           << cp;
+  }
+
+
+  // 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)
+    {
+      ss <<  "|  " << bucket    << "        ";
+      ss << " |  "; format_cp_aligned_dot(t.psqt[bucket], ss); ss << "  "
+         << " |  "; format_cp_aligned_dot(t.positional[bucket], ss); ss << "  "
+         << " |  "; format_cp_aligned_dot(t.psqt[bucket] + t.positional[bucket], ss); ss << "  "
+         << " |";
+      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 (currentEvalFileName != 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

@@ -0,0 +1,68 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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/>.
+*/
+
+// header used in NNUE evaluation function
+
+#ifndef NNUE_EVALUATE_NNUE_H_INCLUDED
+#define NNUE_EVALUATE_NNUE_H_INCLUDED
+
+#include "nnue_feature_transformer.h"
+
+#include <memory>
+
+namespace Stockfish::Eval::NNUE {
+
+  // Hash value of evaluation function structure
+  constexpr std::uint32_t HashValue =
+      FeatureTransformer::get_hash_value() ^ Network::get_hash_value();
+
+
+  // Deleter for automating release of memory area
+  template <typename T>
+  struct AlignedDeleter {
+    void operator()(T* ptr) const {
+      ptr->~T();
+      std_aligned_free(ptr);
+    }
+  };
+
+  template <typename T>
+  struct LargePageDeleter {
+    void operator()(T* ptr) const {
+      ptr->~T();
+      aligned_large_pages_free(ptr);
+    }
+  };
+
+  template <typename T>
+  using AlignedPtr = std::unique_ptr<T, AlignedDeleter<T>>;
+
+  template <typename T>
+  using LargePagePtr = std::unique_ptr<T, LargePageDeleter<T>>;
+
+  std::string trace(Position& pos);
+  Value evaluate(const Position& pos, bool adjusted = false, int* complexity = nullptr);
+  void hint_common_parent_position(const Position& pos);
+
+  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 Stockfish::Eval::NNUE
+
+#endif // #ifndef NNUE_EVALUATE_NNUE_H_INCLUDED

src/nnue/features/half_ka_v2_hm.cpp

@@ -0,0 +1,84 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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_hm of NNUE evaluation function
+
+#include "half_ka_v2_hm.h"
+
+#include "../../position.h"
+
+namespace Stockfish::Eval::NNUE::Features {
+
+  // Index of a feature for a given king position and another piece on some square
+  template<Color Perspective>
+  inline IndexType HalfKAv2_hm::make_index(Square s, Piece pc, Square ksq) {
+    return IndexType((int(s) ^ OrientTBL[Perspective][ksq]) + PieceSquareIndex[Perspective][pc] + KingBuckets[Perspective][ksq]);
+  }
+
+  // Get a list of indices for active features
+  template<Color Perspective>
+  void HalfKAv2_hm::append_active_indices(
+    const Position& pos,
+    IndexList& active
+  ) {
+    Square ksq = 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));
+    }
+  }
+
+  // Explicit template instantiations
+  template void HalfKAv2_hm::append_active_indices<WHITE>(const Position& pos, IndexList& active);
+  template void HalfKAv2_hm::append_active_indices<BLACK>(const Position& pos, IndexList& active);
+
+  // append_changed_indices() : get a list of indices for recently changed features
+  template<Color Perspective>
+  void HalfKAv2_hm::append_changed_indices(
+    Square ksq,
+    const DirtyPiece& dp,
+    IndexList& removed,
+    IndexList& added
+  ) {
+    for (int i = 0; i < dp.dirty_num; ++i) {
+      if (dp.from[i] != SQ_NONE)
+        removed.push_back(make_index<Perspective>(dp.from[i], dp.piece[i], ksq));
+      if (dp.to[i] != SQ_NONE)
+        added.push_back(make_index<Perspective>(dp.to[i], dp.piece[i], ksq));
+    }
+  }
+
+  // Explicit template instantiations
+  template void HalfKAv2_hm::append_changed_indices<WHITE>(Square ksq, const DirtyPiece& dp, IndexList& removed, IndexList& added);
+  template void HalfKAv2_hm::append_changed_indices<BLACK>(Square ksq, const DirtyPiece& dp, IndexList& removed, IndexList& added);
+
+  int HalfKAv2_hm::update_cost(const StateInfo* st) {
+    return st->dirtyPiece.dirty_num;
+  }
+
+  int HalfKAv2_hm::refresh_cost(const Position& pos) {
+    return pos.count<ALL_PIECES>();
+  }
+
+  bool HalfKAv2_hm::requires_refresh(const StateInfo* st, Color perspective) {
+    return st->dirtyPiece.piece[0] == make_piece(perspective, KING);
+  }
+
+}  // namespace Stockfish::Eval::NNUE::Features

src/nnue/features/half_ka_v2_hm.h

@@ -0,0 +1,152 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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_HM_H_INCLUDED
+#define NNUE_FEATURES_HALF_KA_V2_HM_H_INCLUDED
+
+#include "../nnue_common.h"
+
+#include "../../evaluate.h"
+#include "../../misc.h"
+
+namespace Stockfish {
+  struct StateInfo;
+}
+
+namespace Stockfish::Eval::NNUE::Features {
+
+  // Feature HalfKAv2_hm: Combination of the position of own king
+  // and the position of pieces. Position mirrored such that king always on e..h files.
+  class HalfKAv2_hm {
+
+    // 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 }
+    };
+
+    // Index of a feature for a given king position and another piece on some square
+    template<Color Perspective>
+    static IndexType make_index(Square s, Piece pc, Square ksq);
+
+   public:
+    // Feature name
+    static constexpr const char* Name = "HalfKAv2_hm(Friend)";
+
+    // Hash value embedded in the evaluation file
+    static constexpr std::uint32_t HashValue = 0x7f234cb8u;
+
+    // Number of feature dimensions
+    static constexpr IndexType Dimensions =
+        static_cast<IndexType>(SQUARE_NB) * static_cast<IndexType>(PS_NB) / 2;
+
+#define B(v) (v * PS_NB)
+    static constexpr int KingBuckets[COLOR_NB][SQUARE_NB] = {
+      { B(28), B(29), B(30), B(31), B(31), B(30), B(29), B(28),
+        B(24), B(25), B(26), B(27), B(27), B(26), B(25), B(24),
+        B(20), B(21), B(22), B(23), B(23), B(22), B(21), B(20),
+        B(16), B(17), B(18), B(19), B(19), B(18), B(17), B(16),
+        B(12), B(13), B(14), B(15), B(15), B(14), B(13), B(12),
+        B( 8), B( 9), B(10), B(11), B(11), B(10), B( 9), B( 8),
+        B( 4), B( 5), B( 6), B( 7), B( 7), B( 6), B( 5), B( 4),
+        B( 0), B( 1), B( 2), B( 3), B( 3), B( 2), B( 1), B( 0) },
+      { B( 0), B( 1), B( 2), B( 3), B( 3), B( 2), B( 1), B( 0),
+        B( 4), B( 5), B( 6), B( 7), B( 7), B( 6), B( 5), B( 4),
+        B( 8), B( 9), B(10), B(11), B(11), B(10), B( 9), B( 8),
+        B(12), B(13), B(14), B(15), B(15), B(14), B(13), B(12),
+        B(16), B(17), B(18), B(19), B(19), B(18), B(17), B(16),
+        B(20), B(21), B(22), B(23), B(23), B(22), B(21), B(20),
+        B(24), B(25), B(26), B(27), B(27), B(26), B(25), B(24),
+        B(28), B(29), B(30), B(31), B(31), B(30), B(29), B(28) }
+    };
+#undef B
+
+    // Orient a square according to perspective (rotates by 180 for black)
+    static constexpr int OrientTBL[COLOR_NB][SQUARE_NB] = {
+      { SQ_H1, SQ_H1, SQ_H1, SQ_H1, SQ_A1, SQ_A1, SQ_A1, SQ_A1,
+        SQ_H1, SQ_H1, SQ_H1, SQ_H1, SQ_A1, SQ_A1, SQ_A1, SQ_A1,
+        SQ_H1, SQ_H1, SQ_H1, SQ_H1, SQ_A1, SQ_A1, SQ_A1, SQ_A1,
+        SQ_H1, SQ_H1, SQ_H1, SQ_H1, SQ_A1, SQ_A1, SQ_A1, SQ_A1,
+        SQ_H1, SQ_H1, SQ_H1, SQ_H1, SQ_A1, SQ_A1, SQ_A1, SQ_A1,
+        SQ_H1, SQ_H1, SQ_H1, SQ_H1, SQ_A1, SQ_A1, SQ_A1, SQ_A1,
+        SQ_H1, SQ_H1, SQ_H1, SQ_H1, SQ_A1, SQ_A1, SQ_A1, SQ_A1,
+        SQ_H1, SQ_H1, SQ_H1, SQ_H1, SQ_A1, SQ_A1, SQ_A1, SQ_A1 },
+      { SQ_H8, SQ_H8, SQ_H8, SQ_H8, SQ_A8, SQ_A8, SQ_A8, SQ_A8,
+        SQ_H8, SQ_H8, SQ_H8, SQ_H8, SQ_A8, SQ_A8, SQ_A8, SQ_A8,
+        SQ_H8, SQ_H8, SQ_H8, SQ_H8, SQ_A8, SQ_A8, SQ_A8, SQ_A8,
+        SQ_H8, SQ_H8, SQ_H8, SQ_H8, SQ_A8, SQ_A8, SQ_A8, SQ_A8,
+        SQ_H8, SQ_H8, SQ_H8, SQ_H8, SQ_A8, SQ_A8, SQ_A8, SQ_A8,
+        SQ_H8, SQ_H8, SQ_H8, SQ_H8, SQ_A8, SQ_A8, SQ_A8, SQ_A8,
+        SQ_H8, SQ_H8, SQ_H8, SQ_H8, SQ_A8, SQ_A8, SQ_A8, SQ_A8,
+        SQ_H8, SQ_H8, SQ_H8, SQ_H8, SQ_A8, SQ_A8, SQ_A8, SQ_A8 }
+    };
+
+    // Maximum number of simultaneously active features.
+    static constexpr IndexType MaxActiveDimensions = 32;
+    using IndexList = ValueList<IndexType, MaxActiveDimensions>;
+
+    // Get a list of indices for active features
+    template<Color Perspective>
+    static void append_active_indices(
+      const Position& pos,
+      IndexList& active);
+
+    // Get a list of indices for recently changed features
+    template<Color Perspective>
+    static void append_changed_indices(
+      Square ksq,
+      const DirtyPiece& dp,
+      IndexList& removed,
+      IndexList& added
+    );
+
+    // Returns the cost of updating one perspective, the most costly one.
+    // Assumes no refresh needed.
+    static int update_cost(const 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(const StateInfo* st, Color perspective);
+  };
+
+}  // namespace Stockfish::Eval::NNUE::Features
+
+#endif // #ifndef NNUE_FEATURES_HALF_KA_V2_HM_H_INCLUDED

src/nnue/layers/affine_transform.h

@@ -0,0 +1,563 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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 layer AffineTransform of NNUE evaluation function
+
+#ifndef NNUE_LAYERS_AFFINE_TRANSFORM_H_INCLUDED
+#define NNUE_LAYERS_AFFINE_TRANSFORM_H_INCLUDED
+
+#include <iostream>
+#include <algorithm>
+#include <type_traits>
+#include "../nnue_common.h"
+#include "simd.h"
+
+/*
+  This file contains the definition for a fully connected layer (aka affine transform).
+  Two approaches are employed, depending on the sizes of the transform.
+
+  Approach 1 (a specialization for large inputs):
+    - used when the PaddedInputDimensions >= 128
+    - uses AVX512 if possible
+    - processes inputs in batches of 2*InputSimdWidth
+      - so in batches of 128 for AVX512
+    - the weight blocks of size InputSimdWidth are transposed such that
+      access is sequential
+    - N columns of the weight matrix are processed a time, where N
+      depends on the architecture (the amount of registers)
+    - accumulate + hadd is used
+
+  Approach 2 (a specialization for small inputs):
+    - used when the PaddedInputDimensions < 128
+    - expected use-case is for when PaddedInputDimensions == 32 and InputDimensions <= 32.
+      - that's why AVX512 is hard to implement
+    - expected use-case is small layers
+      - not optimized as well as the approach 1
+    - inputs are processed in chunks of 4, weights are respectively transposed
+    - accumulation happens directly to int32s
+*/
+
+namespace Stockfish::Eval::NNUE::Layers {
+
+// Fallback implementation for older/other architectures.
+// Identical for both approaches. Requires the input to be padded to at least 16 values.
+#if !defined(USE_SSSE3)
+  template <IndexType InputDimensions, IndexType PaddedInputDimensions, IndexType OutputDimensions>
+  static void affine_transform_non_ssse3(std::int32_t* output, const std::int8_t* weights, const std::int32_t* biases, const std::uint8_t* input)
+  {
+# if defined(USE_SSE2)
+    // At least a multiple of 16, with SSE2.
+    constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
+    const __m128i Zeros = _mm_setzero_si128();
+    const auto inputVector = reinterpret_cast<const __m128i*>(input);
+
+# elif defined(USE_MMX)
+    constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 8) / 8;
+    const __m64 Zeros = _mm_setzero_si64();
+    const auto inputVector = reinterpret_cast<const __m64*>(input);
+
+# elif defined(USE_NEON_DOTPROD)
+    constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
+    const auto inputVector = reinterpret_cast<const int8x16_t*>(input);
+
+# elif defined(USE_NEON)
+    constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
+    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(&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(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 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 = 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(sumLo, sumHi);
+      sum = _mm_add_pi32(sum, _mm_unpackhi_pi32(sum, sum));
+      output[i] = _mm_cvtsi64_si32(sum);
+
+# elif defined(USE_NEON_DOTPROD)
+      int32x4_t sum = {biases[i]};
+      const auto row = reinterpret_cast<const int8x16_t*>(&weights[offset]);
+      for (IndexType j = 0; j < NumChunks; ++j) {
+        sum = vdotq_s32(sum, inputVector[j], row[j]);
+      }
+      output[i] = vaddvq_s32(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 < 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
+      std::int32_t sum = biases[i];
+      for (IndexType j = 0; j < InputDimensions; ++j) {
+        sum += weights[offset + j] * input[j];
+      }
+      output[i] = sum;
+# endif
+    }
+
+# if defined(USE_MMX)
+    _mm_empty();
+# endif
+  }
+#endif
+
+  template <IndexType InDims, IndexType OutDims, typename Enabled = void>
+  class AffineTransform;
+
+#if defined (USE_AVX512)
+  constexpr IndexType LargeInputSize = 2 * 64;
+#else
+  constexpr IndexType LargeInputSize = std::numeric_limits<IndexType>::max();
+#endif
+
+  // A specialization for large inputs
+  template <IndexType InDims, IndexType OutDims>
+  class AffineTransform<InDims, OutDims, std::enable_if_t<(ceil_to_multiple<IndexType>(InDims, MaxSimdWidth) >= LargeInputSize)>> {
+   public:
+    // Input/output type
+    using InputType = std::uint8_t;
+    using OutputType = std::int32_t;
+
+    // Number of input/output dimensions
+    static constexpr IndexType InputDimensions = InDims;
+    static constexpr IndexType OutputDimensions = OutDims;
+
+    static constexpr IndexType PaddedInputDimensions =
+      ceil_to_multiple<IndexType>(InputDimensions, MaxSimdWidth);
+    static constexpr IndexType PaddedOutputDimensions =
+      ceil_to_multiple<IndexType>(OutputDimensions, MaxSimdWidth);
+
+    using OutputBuffer = OutputType[PaddedOutputDimensions];
+
+    static_assert(PaddedInputDimensions >= LargeInputSize, "Something went wrong. This specialization (for large inputs) should not have been chosen.");
+
+#if defined (USE_AVX512)
+    static constexpr IndexType InputSimdWidth = 64;
+    static constexpr IndexType MaxNumOutputRegs = 16;
+#elif defined (USE_AVX2)
+    static constexpr IndexType InputSimdWidth = 32;
+    static constexpr IndexType MaxNumOutputRegs = 8;
+#elif defined (USE_SSSE3)
+    static constexpr IndexType InputSimdWidth = 16;
+    static constexpr IndexType MaxNumOutputRegs = 8;
+#elif defined (USE_NEON_DOTPROD)
+    static constexpr IndexType InputSimdWidth = 16;
+    static constexpr IndexType MaxNumOutputRegs = 8;
+#elif defined (USE_NEON)
+    static constexpr IndexType InputSimdWidth = 8;
+    static constexpr IndexType MaxNumOutputRegs = 8;
+#else
+    // The fallback implementation will not have permuted weights.
+    // We define these to avoid a lot of ifdefs later.
+    static constexpr IndexType InputSimdWidth = 1;
+    static constexpr IndexType MaxNumOutputRegs = 1;
+#endif
+
+    // A big block is a region in the weight matrix of the size [PaddedInputDimensions, NumOutputRegs].
+    // A small block is a region of size [InputSimdWidth, 1]
+
+    static constexpr IndexType NumOutputRegs = std::min(MaxNumOutputRegs, OutputDimensions);
+    static constexpr IndexType SmallBlockSize = InputSimdWidth;
+    static constexpr IndexType BigBlockSize = NumOutputRegs * PaddedInputDimensions;
+    static constexpr IndexType NumSmallBlocksInBigBlock = BigBlockSize / SmallBlockSize;
+    static constexpr IndexType NumSmallBlocksPerOutput = PaddedInputDimensions / SmallBlockSize;
+    static constexpr IndexType NumBigBlocks = OutputDimensions / NumOutputRegs;
+
+    static_assert(OutputDimensions % NumOutputRegs == 0);
+
+    // Hash value embedded in the evaluation file
+    static constexpr std::uint32_t get_hash_value(std::uint32_t prevHash) {
+      std::uint32_t hashValue = 0xCC03DAE4u;
+      hashValue += OutputDimensions;
+      hashValue ^= prevHash >> 1;
+      hashValue ^= prevHash << 31;
+      return hashValue;
+    }
+
+    /*
+      Transposes the small blocks within a block.
+      Effectively means that weights can be traversed sequentially during inference.
+    */
+    static IndexType get_weight_index(IndexType i)
+    {
+      const IndexType smallBlock = (i / SmallBlockSize) % NumSmallBlocksInBigBlock;
+      const IndexType smallBlockCol = smallBlock / NumSmallBlocksPerOutput;
+      const IndexType smallBlockRow = smallBlock % NumSmallBlocksPerOutput;
+      const IndexType bigBlock   = i / BigBlockSize;
+      const IndexType rest       = i % SmallBlockSize;
+
+      const IndexType idx =
+          bigBlock * BigBlockSize
+        + smallBlockRow * SmallBlockSize * NumOutputRegs
+        + smallBlockCol * SmallBlockSize
+        + rest;
+
+      return idx;
+    }
+
+    // Read network parameters
+    bool read_parameters(std::istream& stream) {
+      read_little_endian<BiasType>(stream, biases, OutputDimensions);
+
+      for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+        weights[get_weight_index(i)] = read_little_endian<WeightType>(stream);
+
+      return !stream.fail();
+    }
+
+    // Write network parameters
+    bool write_parameters(std::ostream& stream) const {
+      write_little_endian<BiasType>(stream, biases, OutputDimensions);
+
+      for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+        write_little_endian<WeightType>(stream, weights[get_weight_index(i)]);
+
+      return !stream.fail();
+    }
+
+    // Forward propagation
+    const OutputType* propagate(
+        const InputType* input, OutputType* output) const {
+
+#if defined (USE_AVX512)
+      using acc_vec_t = __m512i;
+      using bias_vec_t = __m128i;
+      using weight_vec_t = __m512i;
+      using in_vec_t = __m512i;
+      #define vec_zero _mm512_setzero_si512()
+      #define vec_add_dpbusd_32x2 Simd::m512_add_dpbusd_epi32x2
+      #define vec_hadd Simd::m512_hadd
+      #define vec_haddx4 Simd::m512_haddx4
+#elif defined (USE_AVX2)
+      using acc_vec_t = __m256i;
+      using bias_vec_t = __m128i;
+      using weight_vec_t = __m256i;
+      using in_vec_t = __m256i;
+      #define vec_zero _mm256_setzero_si256()
+      #define vec_add_dpbusd_32x2 Simd::m256_add_dpbusd_epi32x2
+      #define vec_hadd Simd::m256_hadd
+      #define vec_haddx4 Simd::m256_haddx4
+#elif defined (USE_SSSE3)
+      using acc_vec_t = __m128i;
+      using bias_vec_t = __m128i;
+      using weight_vec_t = __m128i;
+      using in_vec_t = __m128i;
+      #define vec_zero _mm_setzero_si128()
+      #define vec_add_dpbusd_32x2 Simd::m128_add_dpbusd_epi32x2
+      #define vec_hadd Simd::m128_hadd
+      #define vec_haddx4 Simd::m128_haddx4
+#elif defined (USE_NEON_DOTPROD)
+      using acc_vec_t = int32x4_t;
+      using bias_vec_t = int32x4_t;
+      using weight_vec_t = int8x16_t;
+      using in_vec_t = int8x16_t;
+      #define vec_zero {0}
+      #define vec_add_dpbusd_32x2 Simd::dotprod_m128_add_dpbusd_epi32x2
+      #define vec_hadd Simd::neon_m128_hadd
+      #define vec_haddx4 Simd::neon_m128_haddx4
+#elif defined (USE_NEON)
+      using acc_vec_t = int32x4_t;
+      using bias_vec_t = int32x4_t;
+      using weight_vec_t = int8x8_t;
+      using in_vec_t = int8x8_t;
+      #define vec_zero {0}
+      #define vec_add_dpbusd_32x2 Simd::neon_m128_add_dpbusd_epi32x2
+      #define vec_hadd Simd::neon_m128_hadd
+      #define vec_haddx4 Simd::neon_m128_haddx4
+#endif
+
+#if defined (USE_SSSE3) || defined (USE_NEON)
+      const in_vec_t* invec = reinterpret_cast<const in_vec_t*>(input);
+
+      // Perform accumulation to registers for each big block
+      for (IndexType bigBlock = 0; bigBlock < NumBigBlocks; ++bigBlock)
+      {
+        acc_vec_t acc[NumOutputRegs] = { vec_zero };
+
+        // Each big block has NumOutputRegs small blocks in each "row", one per register.
+        // We process two small blocks at a time to save on one addition without VNNI.
+        for (IndexType smallBlock = 0; smallBlock < NumSmallBlocksPerOutput; smallBlock += 2)
+        {
+          const weight_vec_t* weightvec =
+            reinterpret_cast<const weight_vec_t*>(
+                weights
+              + bigBlock * BigBlockSize
+              + smallBlock * SmallBlockSize * NumOutputRegs);
+
+          const in_vec_t in0 = invec[smallBlock + 0];
+          const in_vec_t in1 = invec[smallBlock + 1];
+
+          for (IndexType k = 0; k < NumOutputRegs; ++k)
+            vec_add_dpbusd_32x2(acc[k], in0, weightvec[k], in1, weightvec[k + NumOutputRegs]);
+        }
+
+        // Horizontally add all accumulators.
+        if constexpr (NumOutputRegs % 4 == 0)
+        {
+          bias_vec_t* outputvec = reinterpret_cast<bias_vec_t*>(output);
+          const bias_vec_t* biasvec = reinterpret_cast<const bias_vec_t*>(biases);
+
+          for (IndexType k = 0; k < NumOutputRegs; k += 4)
+          {
+            const IndexType idx = (bigBlock * NumOutputRegs + k) / 4;
+            outputvec[idx] = vec_haddx4(acc[k+0], acc[k+1], acc[k+2], acc[k+3], biasvec[idx]);
+          }
+        }
+        else
+        {
+          for (IndexType k = 0; k < NumOutputRegs; ++k)
+          {
+            const IndexType idx = (bigBlock * NumOutputRegs + k);
+            output[idx] = vec_hadd(acc[k], biases[idx]);
+          }
+        }
+      }
+
+# undef vec_zero
+# undef vec_add_dpbusd_32x2
+# undef vec_hadd
+# undef vec_haddx4
+#else
+      // Use old implementation for the other architectures.
+      affine_transform_non_ssse3<
+        InputDimensions,
+        PaddedInputDimensions,
+        OutputDimensions>(output, weights, biases, input);
+
+#endif
+
+      return output;
+    }
+
+   private:
+    using BiasType = OutputType;
+    using WeightType = std::int8_t;
+
+    alignas(CacheLineSize) BiasType biases[OutputDimensions];
+    alignas(CacheLineSize) WeightType weights[OutputDimensions * PaddedInputDimensions];
+  };
+
+  // A specialization for small inputs
+  template <IndexType InDims, IndexType OutDims>
+  class AffineTransform<InDims, OutDims, std::enable_if_t<(ceil_to_multiple<IndexType>(InDims, MaxSimdWidth) < LargeInputSize)>> {
+   public:
+    // Input/output type
+    // Input/output type
+    using InputType = std::uint8_t;
+    using OutputType = std::int32_t;
+
+    // Number of input/output dimensions
+    static constexpr IndexType InputDimensions = InDims;
+    static constexpr IndexType OutputDimensions = OutDims;
+
+    static constexpr IndexType PaddedInputDimensions =
+      ceil_to_multiple<IndexType>(InputDimensions, MaxSimdWidth);
+    static constexpr IndexType PaddedOutputDimensions =
+      ceil_to_multiple<IndexType>(OutputDimensions, MaxSimdWidth);
+
+    using OutputBuffer = OutputType[PaddedOutputDimensions];
+
+    static_assert(PaddedInputDimensions < LargeInputSize, "Something went wrong. This specialization (for small inputs) should not have been chosen.");
+
+    // Hash value embedded in the evaluation file
+    static constexpr std::uint32_t get_hash_value(std::uint32_t prevHash) {
+      std::uint32_t hashValue = 0xCC03DAE4u;
+      hashValue += OutputDimensions;
+      hashValue ^= prevHash >> 1;
+      hashValue ^= prevHash << 31;
+      return hashValue;
+    }
+
+    static IndexType get_weight_index_scrambled(IndexType i)
+    {
+      return
+        (i / 4) % (PaddedInputDimensions / 4) * OutputDimensions * 4 +
+        i / PaddedInputDimensions * 4 +
+        i % 4;
+    }
+
+    static IndexType get_weight_index(IndexType i)
+    {
+#if defined (USE_SSSE3)
+      return get_weight_index_scrambled(i);
+#else
+      return i;
+#endif
+    }
+
+    // Read network parameters
+    bool read_parameters(std::istream& stream) {
+      read_little_endian<BiasType>(stream, biases, OutputDimensions);
+      for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+        weights[get_weight_index(i)] = read_little_endian<WeightType>(stream);
+
+      return !stream.fail();
+    }
+
+    // Write network parameters
+    bool write_parameters(std::ostream& stream) const {
+      write_little_endian<BiasType>(stream, biases, OutputDimensions);
+
+      for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+        write_little_endian<WeightType>(stream, weights[get_weight_index(i)]);
+
+      return !stream.fail();
+    }
+    // Forward propagation
+    const OutputType* propagate(
+        const InputType* input, OutputType* output) const {
+
+#if defined (USE_AVX512)
+      using vec_t = __m512i;
+      #define vec_setzero _mm512_setzero_si512
+      #define vec_set_32 _mm512_set1_epi32
+      #define vec_add_dpbusd_32 Simd::m512_add_dpbusd_epi32
+      #define vec_add_dpbusd_32x2 Simd::m512_add_dpbusd_epi32x2
+      #define vec_hadd Simd::m512_hadd
+#elif defined (USE_AVX2)
+      using vec_t = __m256i;
+      #define vec_setzero _mm256_setzero_si256
+      #define vec_set_32 _mm256_set1_epi32
+      #define vec_add_dpbusd_32 Simd::m256_add_dpbusd_epi32
+      #define vec_add_dpbusd_32x2 Simd::m256_add_dpbusd_epi32x2
+      #define vec_hadd Simd::m256_hadd
+#elif defined (USE_SSSE3)
+      using vec_t = __m128i;
+      #define vec_setzero _mm_setzero_si128
+      #define vec_set_32 _mm_set1_epi32
+      #define vec_add_dpbusd_32 Simd::m128_add_dpbusd_epi32
+      #define vec_add_dpbusd_32x2 Simd::m128_add_dpbusd_epi32x2
+      #define vec_hadd Simd::m128_hadd
+#endif
+
+#if defined (USE_SSSE3)
+      const auto inputVector = reinterpret_cast<const vec_t*>(input);
+
+      static constexpr IndexType OutputSimdWidth = sizeof(vec_t) / sizeof(OutputType);
+
+      static_assert(OutputDimensions % OutputSimdWidth == 0 || OutputDimensions == 1);
+
+      if constexpr (OutputDimensions % OutputSimdWidth == 0)
+      {
+        constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 8) / 4;
+        constexpr IndexType NumRegs = OutputDimensions / OutputSimdWidth;
+
+        const auto input32 = reinterpret_cast<const std::int32_t*>(input);
+        const vec_t* biasvec = reinterpret_cast<const vec_t*>(biases);
+        vec_t acc[NumRegs];
+        for (IndexType k = 0; k < NumRegs; ++k)
+          acc[k] = biasvec[k];
+
+        for (IndexType i = 0; i < NumChunks; i += 2)
+        {
+          const vec_t in0 = vec_set_32(input32[i + 0]);
+          const vec_t in1 = vec_set_32(input32[i + 1]);
+          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]);
+          for (IndexType k = 0; k < NumRegs; ++k)
+            vec_add_dpbusd_32x2(acc[k], in0, col0[k], in1, col1[k]);
+        }
+
+        vec_t* outptr = reinterpret_cast<vec_t*>(output);
+        for (IndexType k = 0; k < NumRegs; ++k)
+          outptr[k] = acc[k];
+      }
+      else if constexpr (OutputDimensions == 1)
+      {
+        constexpr IndexType NumChunks = PaddedInputDimensions / SimdWidth;
+        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]);
+      }
+
+# undef vec_setzero
+# undef vec_set_32
+# undef vec_add_dpbusd_32
+# undef vec_add_dpbusd_32x2
+# undef vec_hadd
+#else
+      // Use old implementation for the other architectures.
+      affine_transform_non_ssse3<
+        InputDimensions,
+        PaddedInputDimensions,
+        OutputDimensions>(output, weights, biases, input);
+#endif
+
+      return output;
+    }
+
+   private:
+    using BiasType = OutputType;
+    using WeightType = std::int8_t;
+
+    alignas(CacheLineSize) BiasType biases[OutputDimensions];
+    alignas(CacheLineSize) WeightType weights[OutputDimensions * PaddedInputDimensions];
+  };
+
+}  // namespace Stockfish::Eval::NNUE::Layers
+
+#endif // #ifndef NNUE_LAYERS_AFFINE_TRANSFORM_H_INCLUDED

src/nnue/layers/affine_transform_sparse_input.h

@@ -0,0 +1,286 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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 layer AffineTransformSparseInput of NNUE evaluation function
+
+#ifndef NNUE_LAYERS_AFFINE_TRANSFORM_SPARSE_INPUT_H_INCLUDED
+#define NNUE_LAYERS_AFFINE_TRANSFORM_SPARSE_INPUT_H_INCLUDED
+
+#include <iostream>
+#include <algorithm>
+#include <array>
+#include <type_traits>
+#include "../nnue_common.h"
+#include "affine_transform.h"
+#include "simd.h"
+
+/*
+  This file contains the definition for a fully connected layer (aka affine transform) with block sparse input.
+*/
+
+namespace Stockfish::Eval::NNUE::Layers {
+#if defined(__GNUC__)  // GCC, Clang, ICC
+
+  static inline IndexType lsb_(std::uint32_t b) {
+    assert(b);
+    return IndexType(__builtin_ctzl(b));
+  }
+
+#elif defined(_MSC_VER)  // MSVC
+
+  static inline IndexType lsb_(std::uint32_t b) {
+    assert(b);
+    unsigned long idx;
+    _BitScanForward(&idx, b);
+    return (IndexType) idx;
+  }
+
+#else  // Compiler is neither GCC nor MSVC compatible
+
+#error "Compiler not supported."
+
+#endif
+
+
+#if defined(USE_SSSE3)
+  alignas(CacheLineSize) static inline const std::array<std::array<std::uint16_t, 8>, 256> lookup_indices = [](){
+    std::array<std::array<std::uint16_t, 8>, 256> v{};
+    for (int i = 0; i < 256; ++i)
+    {
+      int j = i;
+      int k = 0;
+      while(j)
+      {
+        const IndexType lsbIndex = lsb_(std::uint32_t(j));
+        j &= j - 1;
+        v[i][k] = lsbIndex;
+        ++k;
+      }
+    }
+    return v;
+  }();
+  alignas(CacheLineSize) static inline const std::array<unsigned, 256> lookup_count = [](){
+    std::array<unsigned, 256> v;
+    for (int i = 0; i < 256; ++i)
+    {
+      int j = i;
+      int k = 0;
+      while(j)
+      {
+        j &= j - 1;
+        ++k;
+      }
+      v[i] = k;
+    }
+    return v;
+  }();
+
+  // Find indices of nonzero numbers in an int32_t array
+  template<const IndexType InputDimensions>
+  void find_nnz(const std::int32_t* input, std::uint16_t* out, IndexType& count_out) {
+#if defined (USE_AVX512)
+    using vec_t = __m512i;
+    #define vec_nnz(a) _mm512_cmpgt_epi32_mask(a, _mm512_setzero_si512())
+#elif defined (USE_AVX2)
+    using vec_t = __m256i;
+    #define vec_nnz(a) _mm256_movemask_ps(_mm256_castsi256_ps(_mm256_cmpgt_epi32(a, _mm256_setzero_si256())))
+#elif defined (USE_SSSE3)
+    using vec_t = __m128i;
+    #define vec_nnz(a) _mm_movemask_ps(_mm_castsi128_ps(_mm_cmpgt_epi32(a, _mm_setzero_si128())))
+#endif
+    constexpr IndexType InputSimdWidth = sizeof(vec_t) / sizeof(std::int32_t);
+    // Inputs are processed InputSimdWidth at a time and outputs are processed 8 at a time so we process in chunks of max(InputSimdWidth, 8)
+    constexpr IndexType ChunkSize = std::max<IndexType>(InputSimdWidth, 8);
+    constexpr IndexType NumChunks = InputDimensions / ChunkSize;
+    constexpr IndexType InputsPerChunk = ChunkSize / InputSimdWidth;
+    constexpr IndexType OutputsPerChunk = ChunkSize / 8;
+
+    const auto inputVector = reinterpret_cast<const vec_t*>(input);
+    IndexType count = 0;
+    __m128i base = _mm_set1_epi16(0);
+    __m128i increment = _mm_set1_epi16(8);
+    for (IndexType i = 0; i < NumChunks; ++i)
+    {
+      // bitmask of nonzero values in this chunk
+      unsigned nnz = 0;
+      for (IndexType j = 0; j < InputsPerChunk; ++j)
+      {
+        const vec_t inputChunk = inputVector[i * InputsPerChunk + j];
+        nnz |= (unsigned)vec_nnz(inputChunk) << (j * InputSimdWidth);
+      }
+      for (IndexType j = 0; j < OutputsPerChunk; ++j)
+      {
+        const auto lookup = (nnz >> (j * 8)) & 0xFF;
+        const auto offsets = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&lookup_indices[lookup]));
+        _mm_storeu_si128(reinterpret_cast<__m128i*>(out + count), _mm_add_epi16(base, offsets));
+        count += lookup_count[lookup];
+        base = _mm_add_epi16(base, increment);
+      }
+    }
+    count_out = count;
+  }
+# undef vec_nnz
+#endif
+
+  // Sparse input implementation
+  template <IndexType InDims, IndexType OutDims>
+  class AffineTransformSparseInput {
+   public:
+    // Input/output type
+    // Input/output type
+    using InputType = std::uint8_t;
+    using OutputType = std::int32_t;
+
+    // Number of input/output dimensions
+    static constexpr IndexType InputDimensions = InDims;
+    static constexpr IndexType OutputDimensions = OutDims;
+
+    static_assert(OutputDimensions % 16 == 0, "Only implemented for OutputDimensions divisible by 16.");
+
+    static constexpr IndexType PaddedInputDimensions =
+      ceil_to_multiple<IndexType>(InputDimensions, MaxSimdWidth);
+    static constexpr IndexType PaddedOutputDimensions =
+      ceil_to_multiple<IndexType>(OutputDimensions, MaxSimdWidth);
+
+#if defined (USE_SSSE3)
+    static constexpr IndexType ChunkSize = 4;
+#else
+    static constexpr IndexType ChunkSize = 1;
+#endif
+
+    using OutputBuffer = OutputType[PaddedOutputDimensions];
+
+    // Hash value embedded in the evaluation file
+    static constexpr std::uint32_t get_hash_value(std::uint32_t prevHash) {
+      std::uint32_t hashValue = 0xCC03DAE4u;
+      hashValue += OutputDimensions;
+      hashValue ^= prevHash >> 1;
+      hashValue ^= prevHash << 31;
+      return hashValue;
+    }
+
+    static IndexType get_weight_index_scrambled(IndexType i)
+    {
+      return
+        (i / ChunkSize) % (PaddedInputDimensions / ChunkSize) * OutputDimensions * ChunkSize +
+        i / PaddedInputDimensions * ChunkSize +
+        i % ChunkSize;
+    }
+
+    static IndexType get_weight_index(IndexType i)
+    {
+#if defined (USE_SSSE3)
+      return get_weight_index_scrambled(i);
+#else
+      return i;
+#endif
+    }
+
+    // Read network parameters
+    bool read_parameters(std::istream& stream) {
+      read_little_endian<BiasType>(stream, biases, OutputDimensions);
+      for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+        weights[get_weight_index(i)] = read_little_endian<WeightType>(stream);
+
+      return !stream.fail();
+    }
+
+    // Write network parameters
+    bool write_parameters(std::ostream& stream) const {
+      write_little_endian<BiasType>(stream, biases, OutputDimensions);
+
+      for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+        write_little_endian<WeightType>(stream, weights[get_weight_index(i)]);
+
+      return !stream.fail();
+    }
+    // Forward propagation
+    const OutputType* propagate(
+        const InputType* input, OutputType* output) const {
+
+#if defined (USE_SSSE3)
+#if defined (USE_AVX512)
+      using vec_t = __m512i;
+      #define vec_setzero _mm512_setzero_si512
+      #define vec_set_32 _mm512_set1_epi32
+      #define vec_add_dpbusd_32 Simd::m512_add_dpbusd_epi32
+#elif defined (USE_AVX2)
+      using vec_t = __m256i;
+      #define vec_setzero _mm256_setzero_si256
+      #define vec_set_32 _mm256_set1_epi32
+      #define vec_add_dpbusd_32 Simd::m256_add_dpbusd_epi32
+#elif defined (USE_SSSE3)
+      using vec_t = __m128i;
+      #define vec_setzero _mm_setzero_si128
+      #define vec_set_32 _mm_set1_epi32
+      #define vec_add_dpbusd_32 Simd::m128_add_dpbusd_epi32
+#endif
+      static constexpr IndexType OutputSimdWidth = sizeof(vec_t) / sizeof(OutputType);
+
+      constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 8) / ChunkSize;
+      constexpr IndexType NumRegs = OutputDimensions / OutputSimdWidth;
+      std::uint16_t nnz[NumChunks];
+      IndexType count;
+
+      const auto input32 = reinterpret_cast<const std::int32_t*>(input);
+
+      // Find indices of nonzero 32bit blocks
+      find_nnz<NumChunks>(input32, nnz, count);
+
+      const vec_t* biasvec = reinterpret_cast<const vec_t*>(biases);
+      vec_t acc[NumRegs];
+      for (IndexType k = 0; k < NumRegs; ++k)
+        acc[k] = biasvec[k];
+
+      for (IndexType j = 0; j < count; ++j)
+      {
+        const auto i = nnz[j];
+        const vec_t in = vec_set_32(input32[i]);
+        const auto col = reinterpret_cast<const vec_t*>(&weights[i * OutputDimensions * ChunkSize]);
+        for (IndexType k = 0; k < NumRegs; ++k)
+          vec_add_dpbusd_32(acc[k], in, col[k]);
+      }
+
+      vec_t* outptr = reinterpret_cast<vec_t*>(output);
+      for (IndexType k = 0; k < NumRegs; ++k)
+        outptr[k] = acc[k];
+# undef vec_setzero
+# undef vec_set_32
+# undef vec_add_dpbusd_32
+#else
+      // Use dense implementation for the other architectures.
+      affine_transform_non_ssse3<
+        InputDimensions,
+        PaddedInputDimensions,
+        OutputDimensions>(output, weights, biases, input);
+#endif
+
+      return output;
+    }
+
+   private:
+    using BiasType = OutputType;
+    using WeightType = std::int8_t;
+
+    alignas(CacheLineSize) BiasType biases[OutputDimensions];
+    alignas(CacheLineSize) WeightType weights[OutputDimensions * PaddedInputDimensions];
+  };
+
+}  // namespace Stockfish::Eval::NNUE::Layers
+
+#endif // #ifndef NNUE_LAYERS_AFFINE_TRANSFORM_SPARSE_INPUT_H_INCLUDED

src/nnue/layers/clipped_relu.h

@@ -0,0 +1,180 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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 layer ClippedReLU of NNUE evaluation function
+
+#ifndef NNUE_LAYERS_CLIPPED_RELU_H_INCLUDED
+#define NNUE_LAYERS_CLIPPED_RELU_H_INCLUDED
+
+#include "../nnue_common.h"
+
+namespace Stockfish::Eval::NNUE::Layers {
+
+  // Clipped ReLU
+  template <IndexType InDims>
+  class ClippedReLU {
+   public:
+    // Input/output type
+    using InputType = std::int32_t;
+    using OutputType = std::uint8_t;
+
+    // Number of input/output dimensions
+    static constexpr IndexType InputDimensions = InDims;
+    static constexpr IndexType OutputDimensions = InputDimensions;
+    static constexpr IndexType PaddedOutputDimensions =
+        ceil_to_multiple<IndexType>(OutputDimensions, 32);
+
+    using OutputBuffer = OutputType[PaddedOutputDimensions];
+
+    // Hash value embedded in the evaluation file
+    static constexpr std::uint32_t get_hash_value(std::uint32_t prevHash) {
+      std::uint32_t hashValue = 0x538D24C7u;
+      hashValue += prevHash;
+      return hashValue;
+    }
+
+    // Read network parameters
+    bool read_parameters(std::istream&) {
+      return true;
+    }
+
+    // Write network parameters
+    bool write_parameters(std::ostream&) const {
+      return true;
+    }
+
+    // Forward propagation
+    const OutputType* propagate(
+        const InputType* input, OutputType* output) const {
+
+  #if defined(USE_AVX2)
+      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 Start =
+        InputDimensions % SimdWidth == 0
+        ? InputDimensions / SimdWidth * SimdWidth
+        : InputDimensions / (SimdWidth / 2) * (SimdWidth / 2);
+
+  #elif defined(USE_SSE2)
+      constexpr IndexType NumChunks = InputDimensions / SimdWidth;
+
+  #ifdef USE_SSE41
+      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 < 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],
+
+  #ifdef USE_SSE41
+          _mm_max_epi8(packedbytes, Zero)
+  #else
+          _mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s)
+  #endif
+
+        );
+      }
+      constexpr IndexType Start = NumChunks * SimdWidth;
+
+  #elif defined(USE_MMX)
+      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 < NumChunks; ++i) {
+        const __m64 words0 = _mm_srai_pi16(
+            _mm_packs_pi32(in[i * 4 + 0], in[i * 4 + 1]),
+            WeightScaleBits);
+        const __m64 words1 = _mm_srai_pi16(
+            _mm_packs_pi32(in[i * 4 + 2], in[i * 4 + 3]),
+            WeightScaleBits);
+        const __m64 packedbytes = _mm_packs_pi16(words0, words1);
+        out[i] = _mm_subs_pi8(_mm_adds_pi8(packedbytes, k0x80s), k0x80s);
+      }
+      _mm_empty();
+      constexpr IndexType Start = NumChunks * SimdWidth;
+
+  #elif defined(USE_NEON)
+      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 < NumChunks; ++i) {
+        int16x8_t shifted;
+        const auto pack = reinterpret_cast<int16x4_t*>(&shifted);
+        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 Start = NumChunks * (SimdWidth / 2);
+  #else
+      constexpr IndexType Start = 0;
+  #endif
+
+      for (IndexType i = Start; i < InputDimensions; ++i) {
+        output[i] = static_cast<OutputType>(
+            std::max(0, std::min(127, input[i] >> WeightScaleBits)));
+      }
+
+      return output;
+    }
+  };
+
+}  // namespace Stockfish::Eval::NNUE::Layers
+
+#endif // NNUE_LAYERS_CLIPPED_RELU_H_INCLUDED

src/nnue/layers/simd.h

@@ -0,0 +1,403 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+
+  Stockfish is free software: you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation, either version 3 of the License, or
+  (at your option) any later version.
+
+  Stockfish is distributed in the hope that it will be useful,
+  but WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+  GNU General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef STOCKFISH_SIMD_H_INCLUDED
+#define STOCKFISH_SIMD_H_INCLUDED
+
+#if defined(USE_AVX2)
+# include <immintrin.h>
+
+#elif defined(USE_SSE41)
+# include <smmintrin.h>
+
+#elif defined(USE_SSSE3)
+# include <tmmintrin.h>
+
+#elif defined(USE_SSE2)
+# include <emmintrin.h>
+
+#elif defined(USE_MMX)
+# include <mmintrin.h>
+
+#elif defined(USE_NEON)
+# include <arm_neon.h>
+#endif
+
+// The inline asm is only safe for GCC, where it is necessary to get good codegen.
+// See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101693
+// Clang does fine without it.
+// Play around here: https://godbolt.org/z/7EWqrYq51
+#if (defined(__GNUC__) && !defined(__clang__) && !defined(__INTEL_COMPILER))
+#define USE_INLINE_ASM
+#endif
+
+// Use either the AVX512 or AVX-VNNI version of the VNNI instructions.
+#if defined(USE_AVXVNNI)
+#define VNNI_PREFIX "%{vex%} "
+#else
+#define VNNI_PREFIX ""
+#endif
+
+namespace Stockfish::Simd {
+
+#if defined (USE_AVX512)
+
+    [[maybe_unused]] static int m512_hadd(__m512i sum, int bias) {
+      return _mm512_reduce_add_epi32(sum) + bias;
+    }
+
+    /*
+      Parameters:
+        sum0 = [zmm0.i128[0], zmm0.i128[1], zmm0.i128[2], zmm0.i128[3]]
+        sum1 = [zmm1.i128[0], zmm1.i128[1], zmm1.i128[2], zmm1.i128[3]]
+        sum2 = [zmm2.i128[0], zmm2.i128[1], zmm2.i128[2], zmm2.i128[3]]
+        sum3 = [zmm3.i128[0], zmm3.i128[1], zmm3.i128[2], zmm3.i128[3]]
+
+      Returns:
+        ret = [
+          reduce_add_epi32(zmm0.i128[0]), reduce_add_epi32(zmm1.i128[0]), reduce_add_epi32(zmm2.i128[0]), reduce_add_epi32(zmm3.i128[0]),
+          reduce_add_epi32(zmm0.i128[1]), reduce_add_epi32(zmm1.i128[1]), reduce_add_epi32(zmm2.i128[1]), reduce_add_epi32(zmm3.i128[1]),
+          reduce_add_epi32(zmm0.i128[2]), reduce_add_epi32(zmm1.i128[2]), reduce_add_epi32(zmm2.i128[2]), reduce_add_epi32(zmm3.i128[2]),
+          reduce_add_epi32(zmm0.i128[3]), reduce_add_epi32(zmm1.i128[3]), reduce_add_epi32(zmm2.i128[3]), reduce_add_epi32(zmm3.i128[3])
+        ]
+    */
+    [[maybe_unused]] static __m512i m512_hadd128x16_interleave(
+        __m512i sum0, __m512i sum1, __m512i sum2, __m512i sum3) {
+
+      __m512i sum01a = _mm512_unpacklo_epi32(sum0, sum1);
+      __m512i sum01b = _mm512_unpackhi_epi32(sum0, sum1);
+
+      __m512i sum23a = _mm512_unpacklo_epi32(sum2, sum3);
+      __m512i sum23b = _mm512_unpackhi_epi32(sum2, sum3);
+
+      __m512i sum01 = _mm512_add_epi32(sum01a, sum01b);
+      __m512i sum23 = _mm512_add_epi32(sum23a, sum23b);
+
+      __m512i sum0123a = _mm512_unpacklo_epi64(sum01, sum23);
+      __m512i sum0123b = _mm512_unpackhi_epi64(sum01, sum23);
+
+      return _mm512_add_epi32(sum0123a, sum0123b);
+    }
+
+    [[maybe_unused]] static __m128i m512_haddx4(
+        __m512i sum0, __m512i sum1, __m512i sum2, __m512i sum3,
+        __m128i bias) {
+
+      __m512i sum = m512_hadd128x16_interleave(sum0, sum1, sum2, sum3);
+
+      __m256i sum256lo = _mm512_castsi512_si256(sum);
+      __m256i sum256hi = _mm512_extracti64x4_epi64(sum, 1);
+
+      sum256lo = _mm256_add_epi32(sum256lo, sum256hi);
+
+      __m128i sum128lo = _mm256_castsi256_si128(sum256lo);
+      __m128i sum128hi = _mm256_extracti128_si256(sum256lo, 1);
+
+      return _mm_add_epi32(_mm_add_epi32(sum128lo, sum128hi), bias);
+    }
+
+    [[maybe_unused]] static void m512_add_dpbusd_epi32(
+        __m512i& acc,
+        __m512i a,
+        __m512i b) {
+
+# if defined (USE_VNNI)
+#   if defined (USE_INLINE_ASM)
+      asm(
+        "vpdpbusd %[b], %[a], %[acc]\n\t"
+        : [acc]"+v"(acc)
+        : [a]"v"(a), [b]"vm"(b)
+      );
+#   else
+      acc = _mm512_dpbusd_epi32(acc, a, b);
+#   endif
+# else
+#   if defined (USE_INLINE_ASM)
+      __m512i tmp = _mm512_maddubs_epi16(a, b);
+      asm(
+          "vpmaddwd    %[tmp], %[ones], %[tmp]\n\t"
+          "vpaddd      %[acc], %[tmp], %[acc]\n\t"
+          : [acc]"+v"(acc), [tmp]"+&v"(tmp)
+          : [ones]"v"(_mm512_set1_epi16(1))
+      );
+#   else
+      __m512i product0 = _mm512_maddubs_epi16(a, b);
+      product0 = _mm512_madd_epi16(product0, _mm512_set1_epi16(1));
+      acc = _mm512_add_epi32(acc, product0);
+#   endif
+# endif
+    }
+
+    [[maybe_unused]] static void m512_add_dpbusd_epi32x2(
+        __m512i& acc,
+        __m512i a0, __m512i b0,
+        __m512i a1, __m512i b1) {
+
+# if defined (USE_VNNI)
+#   if defined (USE_INLINE_ASM)
+      asm(
+        "vpdpbusd %[b0], %[a0], %[acc]\n\t"
+        "vpdpbusd %[b1], %[a1], %[acc]\n\t"
+        : [acc]"+&v"(acc)
+        : [a0]"v"(a0), [b0]"vm"(b0), [a1]"v"(a1), [b1]"vm"(b1)
+      );
+#   else
+      acc = _mm512_dpbusd_epi32(acc, a0, b0);
+      acc = _mm512_dpbusd_epi32(acc, a1, b1);
+#   endif
+# else
+#   if defined (USE_INLINE_ASM)
+      __m512i tmp0 = _mm512_maddubs_epi16(a0, b0);
+      __m512i tmp1 = _mm512_maddubs_epi16(a1, b1);
+      asm(
+          "vpmaddwd    %[tmp0], %[ones], %[tmp0]\n\t"
+          "vpmaddwd    %[tmp1], %[ones], %[tmp1]\n\t"
+          "vpaddd      %[tmp0], %[tmp1], %[tmp0]\n\t"
+          "vpaddd      %[acc], %[tmp0], %[acc]\n\t"
+          : [acc]"+v"(acc), [tmp0]"+&v"(tmp0), [tmp1]"+&v"(tmp1)
+          : [ones]"v"(_mm512_set1_epi16(1))
+      );
+#   else
+      __m512i product0 = _mm512_maddubs_epi16(a0, b0);
+      __m512i product1 = _mm512_maddubs_epi16(a1, b1);
+      product0 = _mm512_madd_epi16(product0, _mm512_set1_epi16(1));
+      product1 = _mm512_madd_epi16(product1, _mm512_set1_epi16(1));
+      acc = _mm512_add_epi32(acc, _mm512_add_epi32(product0, product1));
+#   endif
+# endif
+    }
+
+#endif
+
+#if defined (USE_AVX2)
+
+    [[maybe_unused]] static int m256_hadd(__m256i sum, int bias) {
+      __m128i sum128 = _mm_add_epi32(_mm256_castsi256_si128(sum), _mm256_extracti128_si256(sum, 1));
+      sum128 = _mm_add_epi32(sum128, _mm_shuffle_epi32(sum128, _MM_PERM_BADC));
+      sum128 = _mm_add_epi32(sum128, _mm_shuffle_epi32(sum128, _MM_PERM_CDAB));
+      return _mm_cvtsi128_si32(sum128) + bias;
+    }
+
+    [[maybe_unused]] static __m128i m256_haddx4(
+        __m256i sum0, __m256i sum1, __m256i sum2, __m256i sum3,
+        __m128i bias) {
+
+      sum0 = _mm256_hadd_epi32(sum0, sum1);
+      sum2 = _mm256_hadd_epi32(sum2, sum3);
+
+      sum0 = _mm256_hadd_epi32(sum0, sum2);
+
+      __m128i sum128lo = _mm256_castsi256_si128(sum0);
+      __m128i sum128hi = _mm256_extracti128_si256(sum0, 1);
+
+      return _mm_add_epi32(_mm_add_epi32(sum128lo, sum128hi), bias);
+    }
+
+    [[maybe_unused]] static void m256_add_dpbusd_epi32(
+        __m256i& acc,
+        __m256i a,
+        __m256i b) {
+
+# if defined (USE_VNNI)
+#   if defined (USE_INLINE_ASM)
+      asm(
+        VNNI_PREFIX "vpdpbusd %[b], %[a], %[acc]\n\t"
+        : [acc]"+v"(acc)
+        : [a]"v"(a), [b]"vm"(b)
+      );
+#   else
+      acc = _mm256_dpbusd_epi32(acc, a, b);
+#   endif
+# else
+#   if defined (USE_INLINE_ASM)
+      __m256i tmp = _mm256_maddubs_epi16(a, b);
+      asm(
+          "vpmaddwd    %[tmp], %[ones], %[tmp]\n\t"
+          "vpaddd      %[acc], %[tmp], %[acc]\n\t"
+          : [acc]"+v"(acc), [tmp]"+&v"(tmp)
+          : [ones]"v"(_mm256_set1_epi16(1))
+      );
+#   else
+      __m256i product0 = _mm256_maddubs_epi16(a, b);
+      product0 = _mm256_madd_epi16(product0, _mm256_set1_epi16(1));
+      acc = _mm256_add_epi32(acc, product0);
+#   endif
+# endif
+    }
+
+    [[maybe_unused]] static void m256_add_dpbusd_epi32x2(
+        __m256i& acc,
+        __m256i a0, __m256i b0,
+        __m256i a1, __m256i b1) {
+
+# if defined (USE_VNNI)
+#   if defined (USE_INLINE_ASM)
+      asm(
+        VNNI_PREFIX "vpdpbusd %[b0], %[a0], %[acc]\n\t"
+        VNNI_PREFIX "vpdpbusd %[b1], %[a1], %[acc]\n\t"
+        : [acc]"+&v"(acc)
+        : [a0]"v"(a0), [b0]"vm"(b0), [a1]"v"(a1), [b1]"vm"(b1)
+      );
+#   else
+      acc = _mm256_dpbusd_epi32(acc, a0, b0);
+      acc = _mm256_dpbusd_epi32(acc, a1, b1);
+#   endif
+# else
+#   if defined (USE_INLINE_ASM)
+      __m256i tmp0 = _mm256_maddubs_epi16(a0, b0);
+      __m256i tmp1 = _mm256_maddubs_epi16(a1, b1);
+      asm(
+          "vpmaddwd    %[tmp0], %[ones], %[tmp0]\n\t"
+          "vpmaddwd    %[tmp1], %[ones], %[tmp1]\n\t"
+          "vpaddd      %[tmp0], %[tmp1], %[tmp0]\n\t"
+          "vpaddd      %[acc], %[tmp0], %[acc]\n\t"
+          : [acc]"+v"(acc), [tmp0]"+&v"(tmp0), [tmp1]"+&v"(tmp1)
+          : [ones]"v"(_mm256_set1_epi16(1))
+      );
+#   else
+      __m256i product0 = _mm256_maddubs_epi16(a0, b0);
+      __m256i product1 = _mm256_maddubs_epi16(a1, b1);
+      product0 = _mm256_madd_epi16(product0, _mm256_set1_epi16(1));
+      product1 = _mm256_madd_epi16(product1, _mm256_set1_epi16(1));
+      acc = _mm256_add_epi32(acc, _mm256_add_epi32(product0, product1));
+#   endif
+# endif
+    }
+
+#endif
+
+#if defined (USE_SSSE3)
+
+    [[maybe_unused]] static int m128_hadd(__m128i sum, int bias) {
+      sum = _mm_add_epi32(sum, _mm_shuffle_epi32(sum, 0x4E)); //_MM_PERM_BADC
+      sum = _mm_add_epi32(sum, _mm_shuffle_epi32(sum, 0xB1)); //_MM_PERM_CDAB
+      return _mm_cvtsi128_si32(sum) + bias;
+    }
+
+    [[maybe_unused]] static __m128i m128_haddx4(
+        __m128i sum0, __m128i sum1, __m128i sum2, __m128i sum3,
+        __m128i bias) {
+
+      sum0 = _mm_hadd_epi32(sum0, sum1);
+      sum2 = _mm_hadd_epi32(sum2, sum3);
+      sum0 = _mm_hadd_epi32(sum0, sum2);
+      return _mm_add_epi32(sum0, bias);
+    }
+
+    [[maybe_unused]] static void m128_add_dpbusd_epi32(
+        __m128i& acc,
+        __m128i a,
+        __m128i b) {
+
+#   if defined (USE_INLINE_ASM)
+      __m128i tmp = _mm_maddubs_epi16(a, b);
+      asm(
+          "pmaddwd    %[ones], %[tmp]\n\t"
+          "paddd      %[tmp], %[acc]\n\t"
+          : [acc]"+v"(acc), [tmp]"+&v"(tmp)
+          : [ones]"v"(_mm_set1_epi16(1))
+      );
+#   else
+      __m128i product0 = _mm_maddubs_epi16(a, b);
+      product0 = _mm_madd_epi16(product0, _mm_set1_epi16(1));
+      acc = _mm_add_epi32(acc, product0);
+#   endif
+    }
+
+    [[maybe_unused]] static void m128_add_dpbusd_epi32x2(
+        __m128i& acc,
+        __m128i a0, __m128i b0,
+        __m128i a1, __m128i b1) {
+
+#   if defined (USE_INLINE_ASM)
+      __m128i tmp0 = _mm_maddubs_epi16(a0, b0);
+      __m128i tmp1 = _mm_maddubs_epi16(a1, b1);
+      asm(
+          "pmaddwd    %[ones], %[tmp0]\n\t"
+          "pmaddwd    %[ones], %[tmp1]\n\t"
+          "paddd      %[tmp1], %[tmp0]\n\t"
+          "paddd      %[tmp0], %[acc]\n\t"
+          : [acc]"+v"(acc), [tmp0]"+&v"(tmp0), [tmp1]"+&v"(tmp1)
+          : [ones]"v"(_mm_set1_epi16(1))
+      );
+#   else
+      __m128i product0 = _mm_maddubs_epi16(a0, b0);
+      __m128i product1 = _mm_maddubs_epi16(a1, b1);
+      product0 = _mm_madd_epi16(product0, _mm_set1_epi16(1));
+      product1 = _mm_madd_epi16(product1, _mm_set1_epi16(1));
+      acc = _mm_add_epi32(acc, _mm_add_epi32(product0, product1));
+#   endif
+    }
+
+#endif
+
+#if defined (USE_NEON_DOTPROD)
+
+    [[maybe_unused]] static void dotprod_m128_add_dpbusd_epi32x2(
+        int32x4_t& acc,
+        int8x16_t a0, int8x16_t b0,
+        int8x16_t a1, int8x16_t b1) {
+
+        acc = vdotq_s32(acc, a0, b0);
+        acc = vdotq_s32(acc, a1, b1);
+    }
+
+#endif
+
+#if defined (USE_NEON)
+
+    [[maybe_unused]] static int neon_m128_reduce_add_epi32(int32x4_t s) {
+#   if USE_NEON >= 8
+      return vaddvq_s32(s);
+#   else
+      return s[0] + s[1] + s[2] + s[3];
+#   endif
+    }
+
+    [[maybe_unused]] static int neon_m128_hadd(int32x4_t sum, int bias) {
+      return neon_m128_reduce_add_epi32(sum) + bias;
+    }
+
+    [[maybe_unused]] static int32x4_t neon_m128_haddx4(
+        int32x4_t sum0, int32x4_t sum1, int32x4_t sum2, int32x4_t sum3,
+        int32x4_t bias) {
+
+      int32x4_t hsums {
+        neon_m128_reduce_add_epi32(sum0),
+        neon_m128_reduce_add_epi32(sum1),
+        neon_m128_reduce_add_epi32(sum2),
+        neon_m128_reduce_add_epi32(sum3)
+      };
+      return vaddq_s32(hsums, bias);
+    }
+
+    [[maybe_unused]] static void neon_m128_add_dpbusd_epi32x2(
+        int32x4_t& acc,
+        int8x8_t a0, int8x8_t b0,
+        int8x8_t a1, int8x8_t b1) {
+
+      int16x8_t product = vmull_s8(a0, b0);
+      product = vmlal_s8(product, a1, b1);
+      acc = vpadalq_s16(acc, product);
+    }
+
+#endif
+
+}
+
+#endif // STOCKFISH_SIMD_H_INCLUDED

src/nnue/layers/sqr_clipped_relu.h

@@ -0,0 +1,120 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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 layer ClippedReLU of NNUE evaluation function
+
+#ifndef NNUE_LAYERS_SQR_CLIPPED_RELU_H_INCLUDED
+#define NNUE_LAYERS_SQR_CLIPPED_RELU_H_INCLUDED
+
+#include "../nnue_common.h"
+
+namespace Stockfish::Eval::NNUE::Layers {
+
+  // Clipped ReLU
+  template <IndexType InDims>
+  class SqrClippedReLU {
+   public:
+    // Input/output type
+    using InputType = std::int32_t;
+    using OutputType = std::uint8_t;
+
+    // Number of input/output dimensions
+    static constexpr IndexType InputDimensions = InDims;
+    static constexpr IndexType OutputDimensions = InputDimensions;
+    static constexpr IndexType PaddedOutputDimensions =
+        ceil_to_multiple<IndexType>(OutputDimensions, 32);
+
+    using OutputBuffer = OutputType[PaddedOutputDimensions];
+
+    // Hash value embedded in the evaluation file
+    static constexpr std::uint32_t get_hash_value(std::uint32_t prevHash) {
+      std::uint32_t hashValue = 0x538D24C7u;
+      hashValue += prevHash;
+      return hashValue;
+    }
+
+    // Read network parameters
+    bool read_parameters(std::istream&) {
+      return true;
+    }
+
+    // Write network parameters
+    bool write_parameters(std::ostream&) const {
+      return true;
+    }
+
+    // Forward propagation
+    const OutputType* propagate(
+        const InputType* input, OutputType* output) const {
+
+  #if defined(USE_SSE2)
+      constexpr IndexType NumChunks = InputDimensions / 16;
+
+  #ifdef USE_SSE41
+      const __m128i Zero = _mm_setzero_si128();
+  #else
+      const __m128i k0x80s = _mm_set1_epi8(-128);
+  #endif
+
+      static_assert(WeightScaleBits == 6);
+      const auto in = reinterpret_cast<const __m128i*>(input);
+      const auto out = reinterpret_cast<__m128i*>(output);
+      for (IndexType i = 0; i < NumChunks; ++i) {
+        __m128i words0 = _mm_packs_epi32(
+            _mm_load_si128(&in[i * 4 + 0]),
+            _mm_load_si128(&in[i * 4 + 1]));
+        __m128i words1 = _mm_packs_epi32(
+            _mm_load_si128(&in[i * 4 + 2]),
+            _mm_load_si128(&in[i * 4 + 3]));
+
+        // Not sure if
+        words0 = _mm_srli_epi16(_mm_mulhi_epi16(words0, words0), 3);
+        words1 = _mm_srli_epi16(_mm_mulhi_epi16(words1, words1), 3);
+
+        const __m128i packedbytes = _mm_packs_epi16(words0, words1);
+
+        _mm_store_si128(&out[i],
+
+  #ifdef USE_SSE41
+          _mm_max_epi8(packedbytes, Zero)
+  #else
+          _mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s)
+  #endif
+
+        );
+      }
+      constexpr IndexType Start = NumChunks * 16;
+
+  #else
+      constexpr IndexType Start = 0;
+  #endif
+
+      for (IndexType i = Start; i < InputDimensions; ++i) {
+        output[i] = static_cast<OutputType>(
+            // really should be /127 but we need to make it fast
+            // needs to be accounted for in the trainer
+            std::max(0ll, std::min(127ll, (((long long)input[i] * input[i]) >> (2 * WeightScaleBits)) / 128)));
+      }
+
+      return output;
+    }
+  };
+
+}  // namespace Stockfish::Eval::NNUE::Layers
+
+#endif // NNUE_LAYERS_SQR_CLIPPED_RELU_H_INCLUDED

src/nnue/nnue_accumulator.h

@@ -0,0 +1,37 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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/>.
+*/
+
+// Class for difference calculation of NNUE evaluation function
+
+#ifndef NNUE_ACCUMULATOR_H_INCLUDED
+#define NNUE_ACCUMULATOR_H_INCLUDED
+
+#include "nnue_architecture.h"
+
+namespace Stockfish::Eval::NNUE {
+
+  // Class that holds the result of affine transformation of input features
+  struct alignas(CacheLineSize) Accumulator {
+    std::int16_t accumulation[2][TransformedFeatureDimensions];
+    std::int32_t psqtAccumulation[2][PSQTBuckets];
+    bool computed[2];
+  };
+
+}  // namespace Stockfish::Eval::NNUE
+
+#endif // NNUE_ACCUMULATOR_H_INCLUDED

src/nnue/nnue_architecture.h

@@ -0,0 +1,137 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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/>.
+*/
+
+// Input features and network structure used in NNUE evaluation function
+
+#ifndef NNUE_ARCHITECTURE_H_INCLUDED
+#define NNUE_ARCHITECTURE_H_INCLUDED
+
+#include <memory>
+
+#include "nnue_common.h"
+
+#include "features/half_ka_v2_hm.h"
+
+#include "layers/affine_transform_sparse_input.h"
+#include "layers/affine_transform.h"
+#include "layers/clipped_relu.h"
+#include "layers/sqr_clipped_relu.h"
+
+#include "../misc.h"
+
+namespace Stockfish::Eval::NNUE {
+
+// Input features used in evaluation function
+using FeatureSet = Features::HalfKAv2_hm;
+
+// Number of input feature dimensions after conversion
+constexpr IndexType TransformedFeatureDimensions = 1536;
+constexpr IndexType PSQTBuckets = 8;
+constexpr IndexType LayerStacks = 8;
+
+struct Network
+{
+  static constexpr int FC_0_OUTPUTS = 15;
+  static constexpr int FC_1_OUTPUTS = 32;
+
+  Layers::AffineTransformSparseInput<TransformedFeatureDimensions, FC_0_OUTPUTS + 1> fc_0;
+  Layers::SqrClippedReLU<FC_0_OUTPUTS + 1> ac_sqr_0;
+  Layers::ClippedReLU<FC_0_OUTPUTS + 1> ac_0;
+  Layers::AffineTransform<FC_0_OUTPUTS * 2, FC_1_OUTPUTS> fc_1;
+  Layers::ClippedReLU<FC_1_OUTPUTS> ac_1;
+  Layers::AffineTransform<FC_1_OUTPUTS, 1> fc_2;
+
+  // Hash value embedded in the evaluation file
+  static constexpr std::uint32_t get_hash_value() {
+    // input slice hash
+    std::uint32_t hashValue = 0xEC42E90Du;
+    hashValue ^= TransformedFeatureDimensions * 2;
+
+    hashValue = decltype(fc_0)::get_hash_value(hashValue);
+    hashValue = decltype(ac_0)::get_hash_value(hashValue);
+    hashValue = decltype(fc_1)::get_hash_value(hashValue);
+    hashValue = decltype(ac_1)::get_hash_value(hashValue);
+    hashValue = decltype(fc_2)::get_hash_value(hashValue);
+
+    return hashValue;
+  }
+
+  // Read network parameters
+  bool read_parameters(std::istream& stream) {
+    return   fc_0.read_parameters(stream)
+          && ac_0.read_parameters(stream)
+          && fc_1.read_parameters(stream)
+          && ac_1.read_parameters(stream)
+          && fc_2.read_parameters(stream);
+  }
+
+  // Write network parameters
+  bool write_parameters(std::ostream& stream) const {
+    return   fc_0.write_parameters(stream)
+          && ac_0.write_parameters(stream)
+          && fc_1.write_parameters(stream)
+          && ac_1.write_parameters(stream)
+          && fc_2.write_parameters(stream);
+  }
+
+  std::int32_t propagate(const TransformedFeatureType* transformedFeatures)
+  {
+    struct alignas(CacheLineSize) Buffer
+    {
+      alignas(CacheLineSize) decltype(fc_0)::OutputBuffer fc_0_out;
+      alignas(CacheLineSize) decltype(ac_sqr_0)::OutputType ac_sqr_0_out[ceil_to_multiple<IndexType>(FC_0_OUTPUTS * 2, 32)];
+      alignas(CacheLineSize) decltype(ac_0)::OutputBuffer ac_0_out;
+      alignas(CacheLineSize) decltype(fc_1)::OutputBuffer fc_1_out;
+      alignas(CacheLineSize) decltype(ac_1)::OutputBuffer ac_1_out;
+      alignas(CacheLineSize) decltype(fc_2)::OutputBuffer fc_2_out;
+
+      Buffer()
+      {
+          std::memset(this, 0, sizeof(*this));
+      }
+    };
+
+#if defined(__clang__) && (__APPLE__)
+    // workaround for a bug reported with xcode 12
+    static thread_local auto tlsBuffer = std::make_unique<Buffer>();
+    // Access TLS only once, cache result.
+    Buffer& buffer = *tlsBuffer;
+#else
+    alignas(CacheLineSize) static thread_local Buffer buffer;
+#endif
+
+    fc_0.propagate(transformedFeatures, buffer.fc_0_out);
+    ac_sqr_0.propagate(buffer.fc_0_out, buffer.ac_sqr_0_out);
+    ac_0.propagate(buffer.fc_0_out, buffer.ac_0_out);
+    std::memcpy(buffer.ac_sqr_0_out + FC_0_OUTPUTS, buffer.ac_0_out, FC_0_OUTPUTS * sizeof(decltype(ac_0)::OutputType));
+    fc_1.propagate(buffer.ac_sqr_0_out, buffer.fc_1_out);
+    ac_1.propagate(buffer.fc_1_out, buffer.ac_1_out);
+    fc_2.propagate(buffer.ac_1_out, buffer.fc_2_out);
+
+    // buffer.fc_0_out[FC_0_OUTPUTS] is such that 1.0 is equal to 127*(1<<WeightScaleBits) in quantized form
+    // but we want 1.0 to be equal to 600*OutputScale
+    std::int32_t fwdOut = int(buffer.fc_0_out[FC_0_OUTPUTS]) * (600*OutputScale) / (127*(1<<WeightScaleBits));
+    std::int32_t outputValue = buffer.fc_2_out[0] + fwdOut;
+
+    return outputValue;
+  }
+};
+
+}  // namespace Stockfish::Eval::NNUE
+
+#endif // #ifndef NNUE_ARCHITECTURE_H_INCLUDED

src/nnue/nnue_common.h

@@ -0,0 +1,241 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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/>.
+*/
+
+// Constants used in NNUE evaluation function
+
+#ifndef NNUE_COMMON_H_INCLUDED
+#define NNUE_COMMON_H_INCLUDED
+
+#include <cstring>
+#include <iostream>
+
+#include "../misc.h"  // for IsLittleEndian
+
+#if defined(USE_AVX2)
+#include <immintrin.h>
+
+#elif defined(USE_SSE41)
+#include <smmintrin.h>
+
+#elif defined(USE_SSSE3)
+#include <tmmintrin.h>
+
+#elif defined(USE_SSE2)
+#include <emmintrin.h>
+
+#elif defined(USE_MMX)
+#include <mmintrin.h>
+
+#elif defined(USE_NEON)
+#include <arm_neon.h>
+#endif
+
+namespace Stockfish::Eval::NNUE {
+
+  // Version of the evaluation file
+  constexpr std::uint32_t Version = 0x7AF32F20u;
+
+  // Constant used in evaluation value calculation
+  constexpr int OutputScale = 16;
+  constexpr int WeightScaleBits = 6;
+
+  // Size of cache line (in bytes)
+  constexpr std::size_t CacheLineSize = 64;
+
+  constexpr const char Leb128MagicString[] = "COMPRESSED_LEB128";
+  constexpr const std::size_t Leb128MagicStringSize = sizeof(Leb128MagicString) - 1;
+
+  // SIMD width (in bytes)
+  #if defined(USE_AVX2)
+  constexpr std::size_t SimdWidth = 32;
+
+  #elif defined(USE_SSE2)
+  constexpr std::size_t SimdWidth = 16;
+
+  #elif defined(USE_MMX)
+  constexpr std::size_t SimdWidth = 8;
+
+  #elif defined(USE_NEON)
+  constexpr std::size_t SimdWidth = 16;
+  #endif
+
+  constexpr std::size_t MaxSimdWidth = 32;
+
+  // Type of input feature after conversion
+  using TransformedFeatureType = std::uint8_t;
+  using IndexType = std::uint32_t;
+
+  // Round n up to be a multiple of base
+  template <typename IntType>
+  constexpr IntType ceil_to_multiple(IntType n, IntType base) {
+      return (n + base - 1) / base * base;
+  }
+
+  // read_little_endian() is our utility to read an integer (signed or unsigned, any size)
+  // from a stream in little-endian order. We swap the byte order after the read if
+  // necessary to return a result with the byte ordering of the compiling machine.
+  template <typename IntType>
+  inline IntType read_little_endian(std::istream& stream) {
+      IntType result;
+
+      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));
+      }
+
+      return result;
+  }
+
+  // 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, independently 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] = (std::uint8_t)v;
+                v >>= 8;
+            }
+          }
+          u[i] = (std::uint8_t)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]);
+  }
+
+  template <typename IntType>
+  inline void read_leb_128(std::istream& stream, IntType* out, std::size_t count) {
+      static_assert(std::is_signed_v<IntType>, "Not implemented for unsigned types");
+      char leb128MagicString[Leb128MagicStringSize];
+      stream.read(leb128MagicString, Leb128MagicStringSize);
+      assert(strncmp(Leb128MagicString, leb128MagicString, Leb128MagicStringSize) == 0);
+      const std::uint32_t BUF_SIZE = 4096;
+      std::uint8_t buf[BUF_SIZE];
+      auto bytes_left = read_little_endian<std::uint32_t>(stream);
+      std::uint32_t buf_pos = BUF_SIZE;
+      for (std::size_t i = 0; i < count; ++i) {
+          IntType result = 0;
+          size_t shift = 0;
+          do {
+              if (buf_pos == BUF_SIZE) {
+                  stream.read(reinterpret_cast<char*>(buf), std::min(bytes_left, BUF_SIZE));
+                  buf_pos = 0;
+              }
+              std::uint8_t byte = buf[buf_pos++];
+              --bytes_left;
+              result |= (byte & 0x7f) << shift;
+              shift += 7;
+              if ((byte & 0x80) == 0) {
+                  out[i] = sizeof(IntType) * 8 <= shift || (byte & 0x40) == 0 ? result : result | ~((1 << shift) - 1);
+                  break;
+              }
+          } while (shift < sizeof(IntType) * 8);
+      }
+      assert(bytes_left == 0);
+  }
+
+  template <typename IntType>
+  inline void write_leb_128(std::ostream& stream, const IntType* values, std::size_t count) {
+      static_assert(std::is_signed_v<IntType>, "Not implemented for unsigned types");
+      stream.write(Leb128MagicString, Leb128MagicStringSize);
+      std::uint32_t byte_count = 0;
+      for (std::size_t i = 0; i < count; ++i) {
+          IntType value = values[i];
+          std::uint8_t byte;
+          do {
+              byte = value & 0x7f;
+              value >>= 7;
+              ++byte_count;
+          } while ((byte & 0x40) == 0 ? value != 0 : value != -1);
+      }
+      write_little_endian(stream, byte_count);
+      const std::uint32_t BUF_SIZE = 4096;
+      std::uint8_t buf[BUF_SIZE];
+      std::uint32_t buf_pos = 0;
+      auto flush = [&]() {
+          if (buf_pos > 0) {
+              stream.write(reinterpret_cast<char*>(buf), buf_pos);
+              buf_pos = 0;
+          }
+      };
+      auto write = [&](std::uint8_t byte) {
+          buf[buf_pos++] = byte;
+          if (buf_pos == BUF_SIZE) flush();
+      };
+      for (std::size_t i = 0; i < count; ++i) {
+          IntType value = values[i];
+          while (true) {
+              std::uint8_t byte = value & 0x7f;
+              value >>= 7;
+              if ((byte & 0x40) == 0 ? value == 0 : value == -1) {
+                  write(byte);
+                  break;
+              }
+              write(byte | 0x80);
+          }
+      }
+      flush();
+  }
+
+}  // namespace Stockfish::Eval::NNUE
+
+#endif // #ifndef NNUE_COMMON_H_INCLUDED

src/nnue/nnue_feature_transformer.h

@@ -0,0 +1,674 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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 that converts the input features of the NNUE evaluation function
+
+#ifndef NNUE_FEATURE_TRANSFORMER_H_INCLUDED
+#define NNUE_FEATURE_TRANSFORMER_H_INCLUDED
+
+#include "nnue_common.h"
+#include "nnue_architecture.h"
+
+#include <cstring> // std::memset()
+#include <utility> // std::pair
+
+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
+  using vec_t = __m512i;
+  using psqt_vec_t = __m256i;
+  #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_mul_16(a,b) _mm512_mullo_epi16(a,b)
+  #define vec_zero() _mm512_setzero_epi32()
+  #define vec_set_16(a) _mm512_set1_epi16(a)
+  #define vec_max_16(a,b) _mm512_max_epi16(a,b)
+  #define vec_min_16(a,b) _mm512_min_epi16(a,b)
+  inline vec_t vec_msb_pack_16(vec_t a, vec_t b){
+    vec_t compacted = _mm512_packs_epi16(_mm512_srli_epi16(a,7),_mm512_srli_epi16(b,7));
+    return _mm512_permutexvar_epi64(_mm512_setr_epi64(0, 2, 4, 6, 1, 3, 5, 7), compacted);
+  }
+  #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
+  #define MaxChunkSize 64
+
+  #elif USE_AVX2
+  using vec_t = __m256i;
+  using psqt_vec_t = __m256i;
+  #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_mul_16(a,b) _mm256_mullo_epi16(a,b)
+  #define vec_zero() _mm256_setzero_si256()
+  #define vec_set_16(a) _mm256_set1_epi16(a)
+  #define vec_max_16(a,b) _mm256_max_epi16(a,b)
+  #define vec_min_16(a,b) _mm256_min_epi16(a,b)
+  inline vec_t vec_msb_pack_16(vec_t a, vec_t b){
+    vec_t compacted = _mm256_packs_epi16(_mm256_srli_epi16(a,7), _mm256_srli_epi16(b,7));
+    return _mm256_permute4x64_epi64(compacted, 0b11011000);
+  }
+  #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
+  #define MaxChunkSize 32
+
+  #elif USE_SSE2
+  using vec_t = __m128i;
+  using psqt_vec_t = __m128i;
+  #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_mul_16(a,b) _mm_mullo_epi16(a,b)
+  #define vec_zero() _mm_setzero_si128()
+  #define vec_set_16(a) _mm_set1_epi16(a)
+  #define vec_max_16(a,b) _mm_max_epi16(a,b)
+  #define vec_min_16(a,b) _mm_min_epi16(a,b)
+  #define vec_msb_pack_16(a,b) _mm_packs_epi16(_mm_srli_epi16(a,7),_mm_srli_epi16(b,7))
+  #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)
+  #define MaxChunkSize 16
+
+  #elif USE_MMX
+  using vec_t = __m64;
+  using psqt_vec_t = __m64;
+  #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_mul_16(a,b) _mm_mullo_pi16(a,b)
+  #define vec_zero() _mm_setzero_si64()
+  #define vec_set_16(a) _mm_set1_pi16(a)
+  inline vec_t vec_max_16(vec_t a,vec_t b){
+    vec_t comparison = _mm_cmpgt_pi16(a,b);
+    return _mm_or_si64(_mm_and_si64(comparison, a), _mm_andnot_si64(comparison, b));
+  }
+  inline vec_t vec_min_16(vec_t a,vec_t b){
+    vec_t comparison = _mm_cmpgt_pi16(a,b);
+    return _mm_or_si64(_mm_and_si64(comparison, b), _mm_andnot_si64(comparison, a));
+  }
+  #define vec_msb_pack_16(a,b) _mm_packs_pi16(_mm_srli_pi16(a,7),_mm_srli_pi16(b,7))
+  #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 vec_cleanup() _mm_empty()
+  #define NumRegistersSIMD 8
+  #define MaxChunkSize 8
+
+  #elif USE_NEON
+  using vec_t = int16x8_t;
+  using psqt_vec_t = int32x4_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_mul_16(a,b) vmulq_s16(a,b)
+  #define vec_zero() vec_t{0}
+  #define vec_set_16(a) vdupq_n_s16(a)
+  #define vec_max_16(a,b) vmaxq_s16(a,b)
+  #define vec_min_16(a,b) vminq_s16(a,b)
+  inline vec_t vec_msb_pack_16(vec_t a, vec_t b){
+    const int8x8_t shifta = vshrn_n_s16(a, 7);
+    const int8x8_t shiftb = vshrn_n_s16(b, 7);
+    const int8x16_t compacted = vcombine_s8(shifta,shiftb);
+    return *reinterpret_cast<const vec_t*> (&compacted);
+  }
+  #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
+  #define MaxChunkSize 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.
+      #if defined(__GNUC__)
+      #pragma GCC diagnostic push
+      #pragma GCC diagnostic ignored "-Wignored-attributes"
+      #endif
+
+      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>();
+      #if defined(__GNUC__)
+      #pragma GCC diagnostic pop
+      #endif
+  #endif
+
+
+
+  // Input feature converter
+  class FeatureTransformer {
+
+   private:
+    // Number of output dimensions for one side
+    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 InputDimensions = FeatureSet::Dimensions;
+    static constexpr IndexType OutputDimensions = HalfDimensions;
+
+    // Size of forward propagation buffer
+    static constexpr std::size_t BufferSize =
+        OutputDimensions * sizeof(OutputType);
+
+    // Hash value embedded in the evaluation file
+    static constexpr std::uint32_t get_hash_value() {
+      return FeatureSet::HashValue ^ (OutputDimensions * 2);
+    }
+
+    // Read network parameters
+    bool read_parameters(std::istream& stream) {
+
+      read_leb_128<BiasType      >(stream, biases     , HalfDimensions                  );
+      read_leb_128<WeightType    >(stream, weights    , HalfDimensions * InputDimensions);
+      read_leb_128<PSQTWeightType>(stream, psqtWeights, PSQTBuckets    * InputDimensions);
+
+      return !stream.fail();
+    }
+
+    // Write network parameters
+    bool write_parameters(std::ostream& stream) const {
+
+      write_leb_128<BiasType      >(stream, biases     , HalfDimensions                  );
+      write_leb_128<WeightType    >(stream, weights    , HalfDimensions * InputDimensions);
+      write_leb_128<PSQTWeightType>(stream, psqtWeights, PSQTBuckets    * InputDimensions);
+
+      return !stream.fail();
+    }
+
+    // Convert input features
+    std::int32_t transform(const Position& pos, OutputType* output, int bucket) const {
+      update_accumulator<WHITE>(pos);
+      update_accumulator<BLACK>(pos);
+
+      const Color perspectives[2] = {pos.side_to_move(), ~pos.side_to_move()};
+      const auto& accumulation = pos.state()->accumulator.accumulation;
+      const auto& psqtAccumulation = pos.state()->accumulator.psqtAccumulation;
+
+      const auto psqt = (
+            psqtAccumulation[perspectives[0]][bucket]
+          - psqtAccumulation[perspectives[1]][bucket]
+        ) / 2;
+
+
+      for (IndexType p = 0; p < 2; ++p)
+      {
+          const IndexType offset = (HalfDimensions / 2) * p;
+
+#if defined(VECTOR)
+
+          constexpr IndexType OutputChunkSize = MaxChunkSize;
+          static_assert((HalfDimensions / 2) % OutputChunkSize == 0);
+          constexpr IndexType NumOutputChunks = HalfDimensions / 2 / OutputChunkSize;
+
+          vec_t Zero = vec_zero();
+          vec_t One = vec_set_16(127);
+
+          const vec_t* in0 = reinterpret_cast<const vec_t*>(&(accumulation[perspectives[p]][0]));
+          const vec_t* in1 = reinterpret_cast<const vec_t*>(&(accumulation[perspectives[p]][HalfDimensions / 2]));
+                vec_t* out = reinterpret_cast<      vec_t*>(output + offset);
+
+          for (IndexType j = 0; j < NumOutputChunks; j += 1)
+          {
+              const vec_t sum0a = vec_max_16(vec_min_16(in0[j * 2 + 0], One), Zero);
+              const vec_t sum0b = vec_max_16(vec_min_16(in0[j * 2 + 1], One), Zero);
+              const vec_t sum1a = vec_max_16(vec_min_16(in1[j * 2 + 0], One), Zero);
+              const vec_t sum1b = vec_max_16(vec_min_16(in1[j * 2 + 1], One), Zero);
+
+              const vec_t pa = vec_mul_16(sum0a, sum1a);
+              const vec_t pb = vec_mul_16(sum0b, sum1b);
+
+              out[j] = vec_msb_pack_16(pa, pb);
+          }
+
+#else
+
+          for (IndexType j = 0; j < HalfDimensions / 2; ++j) {
+              BiasType sum0 = accumulation[static_cast<int>(perspectives[p])][j + 0];
+              BiasType sum1 = accumulation[static_cast<int>(perspectives[p])][j + HalfDimensions / 2];
+              sum0 = std::max<int>(0, std::min<int>(127, sum0));
+              sum1 = std::max<int>(0, std::min<int>(127, sum1));
+              output[offset + j] = static_cast<OutputType>(sum0 * sum1 / 128);
+          }
+
+#endif
+      }
+
+#if defined(vec_cleanup)
+      vec_cleanup();
+#endif
+
+      return psqt;
+    } // end of function transform()
+
+    void hint_common_access(const Position& pos) const {
+      hint_common_access_for_perspective<WHITE>(pos);
+      hint_common_access_for_perspective<BLACK>(pos);
+    }
+
+   private:
+    template<Color Perspective>
+    [[nodiscard]] std::pair<StateInfo*, StateInfo*> try_find_computed_accumulator(const Position& pos) const {
+      // 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;
+      }
+      return { st, next };
+    }
+
+    // NOTE: The parameter states_to_update is an array of position states, ending with nullptr.
+    //       All states must be sequential, that is states_to_update[i] must either be reachable
+    //       by repeatedly applying ->previous from states_to_update[i+1] or states_to_update[i] == nullptr.
+    //       computed_st must be reachable by repeatedly applying ->previous on states_to_update[0], if not nullptr.
+    template<Color Perspective, size_t N>
+    void update_accumulator_incremental(const Position& pos, StateInfo* computed_st, StateInfo* states_to_update[N]) const {
+      static_assert(N > 0);
+      assert(states_to_update[N-1] == nullptr);
+
+  #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 (states_to_update[0] == nullptr)
+        return;
+
+      // Update incrementally going back through states_to_update.
+
+      // Gather all features to be updated.
+      const Square ksq = pos.square<KING>(Perspective);
+
+      // 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.
+      FeatureSet::IndexList removed[N-1], added[N-1];
+
+      {
+        int i = N-2; // last potential state to update. Skip last element because it must be nullptr.
+        while (states_to_update[i] == nullptr)
+          --i;
+
+        StateInfo *st2 = states_to_update[i];
+
+        for (; i >= 0; --i)
+        {
+          states_to_update[i]->accumulator.computed[Perspective] = true;
+
+          StateInfo* end_state = i == 0 ? computed_st : states_to_update[i - 1];
+
+          for (; st2 != end_state; st2 = st2->previous)
+            FeatureSet::append_changed_indices<Perspective>(
+              ksq, st2->dirtyPiece, removed[i], added[i]);
+        }
+      }
+
+      StateInfo* st = computed_st;
+
+      // Now update the accumulators listed in states_to_update[], where the last element is a sentinel.
+#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[i])
+        {
+          const IndexType offset = HalfDimensions * index;
+
+          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[i])
+        {
+          const IndexType offset = HalfDimensions * index;
+
+          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];
+        }
+      }
+#endif
+
+  #if defined(USE_MMX)
+      _mm_empty();
+  #endif
+    }
+
+    template<Color Perspective>
+    void update_accumulator_refresh(const Position& pos) const {
+  #ifdef VECTOR
+      // Gcc-10.2 unnecessarily spills AVX2 registers if this array
+      // is defined in the VECTOR code below, once in each branch
+      vec_t acc[NumRegs];
+      psqt_vec_t psqt[NumPsqtRegs];
+  #endif
+
+      // Refresh the accumulator
+      // Could be extracted to a separate function because it's done in 2 places,
+      // but it's unclear if compilers would correctly handle register allocation.
+      auto& accumulator = pos.state()->accumulator;
+      accumulator.computed[Perspective] = true;
+      FeatureSet::IndexList active;
+      FeatureSet::append_active_indices<Perspective>(pos, 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
+    }
+
+    template<Color Perspective>
+    void hint_common_access_for_perspective(const Position& pos) const {
+
+      // Works like update_accumulator, but performs less work.
+      // Updates ONLY the accumulator for pos.
+
+      // Look for a usable accumulator of an earlier position. We keep track
+      // of the estimated gain in terms of features to be added/subtracted.
+      // Fast early exit.
+      if (pos.state()->accumulator.computed[Perspective])
+        return;
+
+      auto [oldest_st, _] = try_find_computed_accumulator<Perspective>(pos);
+
+      if (oldest_st->accumulator.computed[Perspective])
+      {
+        // Only update current position accumulator to minimize work.
+        StateInfo* states_to_update[2] = { pos.state(), nullptr };
+        update_accumulator_incremental<Perspective, 2>(pos, oldest_st, states_to_update);
+      }
+      else
+      {
+        update_accumulator_refresh<Perspective>(pos);
+      }
+    }
+
+    template<Color Perspective>
+    void update_accumulator(const Position& pos) const {
+
+      auto [oldest_st, next] = try_find_computed_accumulator<Perspective>(pos);
+
+      if (oldest_st->accumulator.computed[Perspective])
+      {
+        if (next == nullptr)
+          return;
+
+        // Now update the accumulators listed in states_to_update[], where the last element is a sentinel.
+        // Currently we update 2 accumulators.
+        //     1. for the current position
+        //     2. the next accumulator after the computed one
+        // The heuristic may change in the future.
+        StateInfo *states_to_update[3] =
+          { next, next == pos.state() ? nullptr : pos.state(), nullptr };
+
+        update_accumulator_incremental<Perspective, 3>(pos, oldest_st, states_to_update);
+      }
+      else
+      {
+        update_accumulator_refresh<Perspective>(pos);
+      }
+    }
+
+    alignas(CacheLineSize) BiasType biases[HalfDimensions];
+    alignas(CacheLineSize) WeightType weights[HalfDimensions * InputDimensions];
+    alignas(CacheLineSize) PSQTWeightType psqtWeights[InputDimensions * PSQTBuckets];
+  };
+
+}  // namespace Stockfish::Eval::NNUE
+
+#endif // #ifndef NNUE_FEATURE_TRANSFORMER_H_INCLUDED

src/notation.cpp

@@ -1,271 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
-
-  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/>.
-*/
-
-#include <cassert>
-#include <iomanip>
-#include <sstream>
-#include <string>
-
-#include "movegen.h"
-#include "notation.h"
-#include "position.h"
-
-using namespace std;
-
-static const char* PieceToChar = " PNBRQK  pnbrqk";
-
-
-/// score_to_uci() converts a value to a string suitable for use with the UCI
-/// protocol specifications:
-///
-/// cp <x>     The score from the engine's point of view in centipawns.
-/// mate <y>   Mate in y moves, not plies. If the engine is getting mated
-///            use negative values for y.
-
-string score_to_uci(Value v, Value alpha, Value beta) {
-
-  stringstream s;
-
-  if (abs(v) < VALUE_MATE_IN_MAX_PLY)
-      s << "cp " << v * 100 / int(PawnValueMg);
-  else
-      s << "mate " << (v > 0 ? VALUE_MATE - v + 1 : -VALUE_MATE - v) / 2;
-
-  s << (v >= beta ? " lowerbound" : v <= alpha ? " upperbound" : "");
-
-  return s.str();
-}
-
-
-/// move_to_uci() converts a move to a string in coordinate notation
-/// (g1f3, a7a8q, etc.). The only special case is castling moves, where we print
-/// in the e1g1 notation in normal chess mode, and in e1h1 notation in chess960
-/// mode. Internally castle moves are always coded as "king captures rook".
-
-const string move_to_uci(Move m, bool chess960) {
-
-  Square from = from_sq(m);
-  Square to = to_sq(m);
-
-  if (m == MOVE_NONE)
-      return "(none)";
-
-  if (m == MOVE_NULL)
-      return "0000";
-
-  if (type_of(m) == CASTLE && !chess960)
-      to = (to > from ? FILE_G : FILE_C) | rank_of(from);
-
-  string move = square_to_string(from) + square_to_string(to);
-
-  if (type_of(m) == PROMOTION)
-      move += PieceToChar[make_piece(BLACK, promotion_type(m))]; // Lower case
-
-  return move;
-}
-
-
-/// move_from_uci() takes a position and a string representing a move in
-/// simple coordinate notation and returns an equivalent legal Move if any.
-
-Move move_from_uci(const Position& pos, string& str) {
-
-  if (str.length() == 5) // Junior could send promotion piece in uppercase
-      str[4] = char(tolower(str[4]));
-
-  for (MoveList<LEGAL> ml(pos); !ml.end(); ++ml)
-      if (str == move_to_uci(ml.move(), pos.is_chess960()))
-          return ml.move();
-
-  return MOVE_NONE;
-}
-
-
-/// move_to_san() takes a position and a legal Move as input and returns its
-/// short algebraic notation representation.
-
-const string move_to_san(Position& pos, Move m) {
-
-  if (m == MOVE_NONE)
-      return "(none)";
-
-  if (m == MOVE_NULL)
-      return "(null)";
-
-  assert(pos.move_is_legal(m));
-
-  Bitboard attackers;
-  bool ambiguousMove, ambiguousFile, ambiguousRank;
-  string san;
-  Color us = pos.side_to_move();
-  Square from = from_sq(m);
-  Square to = to_sq(m);
-  Piece pc = pos.piece_on(from);
-  PieceType pt = type_of(pc);
-
-  if (type_of(m) == CASTLE)
-      san = to > from ? "O-O" : "O-O-O";
-  else
-  {
-      if (pt != PAWN)
-      {
-          san = PieceToChar[pt]; // Upper case
-
-          // Disambiguation if we have more then one piece with destination 'to'
-          // note that for pawns is not needed because starting file is explicit.
-          ambiguousMove = ambiguousFile = ambiguousRank = false;
-
-          attackers = (pos.attacks_from(pc, to) & pos.pieces(us, pt)) ^ from;
-
-          while (attackers)
-          {
-              Square sq = pop_lsb(&attackers);
-
-              // Pinned pieces are not included in the possible sub-set
-              if (!pos.pl_move_is_legal(make_move(sq, to), pos.pinned_pieces()))
-                  continue;
-
-              ambiguousFile |= file_of(sq) == file_of(from);
-              ambiguousRank |= rank_of(sq) == rank_of(from);
-              ambiguousMove = true;
-          }
-
-          if (ambiguousMove)
-          {
-              if (!ambiguousFile)
-                  san += file_to_char(file_of(from));
-
-              else if (!ambiguousRank)
-                  san += rank_to_char(rank_of(from));
-
-              else
-                  san += square_to_string(from);
-          }
-      }
-      else if (pos.is_capture(m))
-          san = file_to_char(file_of(from));
-
-      if (pos.is_capture(m))
-          san += 'x';
-
-      san += square_to_string(to);
-
-      if (type_of(m) == PROMOTION)
-          san += string("=") + PieceToChar[promotion_type(m)];
-  }
-
-  if (pos.move_gives_check(m, CheckInfo(pos)))
-  {
-      StateInfo st;
-      pos.do_move(m, st);
-      san += MoveList<LEGAL>(pos).size() ? "+" : "#";
-      pos.undo_move(m);
-  }
-
-  return san;
-}
-
-
-/// pretty_pv() formats human-readable search information, typically to be
-/// appended to the search log file. It uses the two helpers below to pretty
-/// format time and score respectively.
-
-static string time_to_string(int64_t msecs) {
-
-  const int MSecMinute = 1000 * 60;
-  const int MSecHour   = 1000 * 60 * 60;
-
-  int64_t hours   =   msecs / MSecHour;
-  int64_t minutes =  (msecs % MSecHour) / MSecMinute;
-  int64_t seconds = ((msecs % MSecHour) % MSecMinute) / 1000;
-
-  stringstream s;
-
-  if (hours)
-      s << hours << ':';
-
-  s << setfill('0') << setw(2) << minutes << ':' << setw(2) << seconds;
-
-  return s.str();
-}
-
-static string score_to_string(Value v) {
-
-  stringstream s;
-
-  if (v >= VALUE_MATE_IN_MAX_PLY)
-      s << "#" << (VALUE_MATE - v + 1) / 2;
-
-  else if (v <= VALUE_MATED_IN_MAX_PLY)
-      s << "-#" << (VALUE_MATE + v) / 2;
-
-  else
-      s << setprecision(2) << fixed << showpos << float(v) / PawnValueMg;
-
-  return s.str();
-}
-
-string pretty_pv(Position& pos, int depth, Value value, int64_t msecs, Move pv[]) {
-
-  const int64_t K = 1000;
-  const int64_t M = 1000000;
-
-  StateInfo state[MAX_PLY_PLUS_2], *st = state;
-  Move* m = pv;
-  string san, padding;
-  size_t length;
-  stringstream s;
-
-  s << setw(2) << depth
-    << setw(8) << score_to_string(value)
-    << setw(8) << time_to_string(msecs);
-
-  if (pos.nodes_searched() < M)
-      s << setw(8) << pos.nodes_searched() / 1 << "  ";
-
-  else if (pos.nodes_searched() < K * M)
-      s << setw(7) << pos.nodes_searched() / K << "K  ";
-
-  else
-      s << setw(7) << pos.nodes_searched() / M << "M  ";
-
-  padding = string(s.str().length(), ' ');
-  length = padding.length();
-
-  while (*m != MOVE_NONE)
-  {
-      san = move_to_san(pos, *m);
-
-      if (length + san.length() > 80)
-      {
-          s << "\n" + padding;
-          length = padding.length();
-      }
-
-      s << san << ' ';
-      length += san.length() + 1;
-
-      pos.do_move(*m++, *st++);
-  }
-
-  while (m != pv)
-      pos.undo_move(*--m);
-
-  return s.str();
-}

src/pawns.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,271 +16,290 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#include <algorithm>
 #include <cassert>
 
 #include "bitboard.h"
-#include "bitcount.h"
 #include "pawns.h"
 #include "position.h"
+#include "thread.h"
+
+namespace Stockfish {
 
 namespace {
 
   #define V Value
   #define S(mg, eg) make_score(mg, eg)
 
-  // Doubled pawn penalty by opposed flag and file
-  const Score DoubledPawnPenalty[2][8] = {
-  { S(13, 43), S(20, 48), S(23, 48), S(23, 48),
-    S(23, 48), S(23, 48), S(20, 48), S(13, 43) },
-  { S(13, 43), S(20, 48), S(23, 48), S(23, 48),
-    S(23, 48), S(23, 48), S(20, 48), S(13, 43) }};
+  // Pawn penalties
+  constexpr Score Backward      = S( 6, 19);
+  constexpr Score Doubled       = S(11, 51);
+  constexpr Score DoubledEarly  = S(17,  7);
+  constexpr Score Isolated      = S( 1, 20);
+  constexpr Score WeakLever     = S( 2, 57);
+  constexpr Score WeakUnopposed = S(15, 18);
 
-  // Isolated pawn penalty by opposed flag and file
-  const Score IsolatedPawnPenalty[2][8] = {
-  { S(37, 45), S(54, 52), S(60, 52), S(60, 52),
-    S(60, 52), S(60, 52), S(54, 52), S(37, 45) },
-  { S(25, 30), S(36, 35), S(40, 35), S(40, 35),
-    S(40, 35), S(40, 35), S(36, 35), S(25, 30) }};
+  // Bonus for blocked pawns at 5th or 6th rank
+  constexpr Score BlockedPawn[2] = { S(-19, -8), S(-7, 3) };
 
-  // Backward pawn penalty by opposed flag and file
-  const Score BackwardPawnPenalty[2][8] = {
-  { S(30, 42), S(43, 46), S(49, 46), S(49, 46),
-    S(49, 46), S(49, 46), S(43, 46), S(30, 42) },
-  { S(20, 28), S(29, 31), S(33, 31), S(33, 31),
-    S(33, 31), S(33, 31), S(29, 31), S(20, 28) }};
-
-  // Pawn chain membership bonus by file
-  const Score ChainBonus[8] = {
-    S(11,-1), S(13,-1), S(13,-1), S(14,-1),
-    S(14,-1), S(13,-1), S(13,-1), S(11,-1)
+  constexpr Score BlockedStorm[RANK_NB] = {
+    S(0, 0), S(0, 0), S(64, 75), S(-3, 14), S(-12, 19), S(-7, 4), S(-10, 5)
   };
 
-  // Candidate passed pawn bonus by rank
-  const Score CandidateBonus[8] = {
-    S( 0, 0), S( 6, 13), S(6,13), S(14,29),
-    S(34,68), S(83,166), S(0, 0), S( 0, 0)
+  // Connected pawn bonus
+  constexpr int Connected[RANK_NB] = { 0, 3, 7, 7, 15, 54, 86 };
+
+  // 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(-2), V(85), V(95), V(53), V(39), V(23), V(25) },
+    { V(-55), V(64), V(32), V(-55), V(-30), V(-11), V(-61) },
+    { V(-11), V(75), V(19), V(-6), V(26), V(9), V(-47) },
+    { V(-41), V(-11), V(-27), V(-58), V(-42), V(-66), V(-163) }
   };
 
-  const Score PawnStructureWeight = S(233, 201);
+  // Danger of enemy pawns moving toward our king by [distance from edge][rank].
+  // RANK_1 = 0 is used for files where the enemy has no pawn, or their pawn
+  // 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(94), V(-280), V(-170), V(90), V(59), V(47), V(53) },
+    { V(43), V(-17), V(128), V(39), V(26), V(-17), V(15) },
+    { V(-9), V(62), V(170), V(34), V(-5), V(-20), V(-11) },
+    { V(-27), V(-19), V(106), V(10), V(2), V(-13), V(-24) }
+  };
 
-  // Weakness of our pawn shelter in front of the king indexed by [king pawn][rank]
-  const Value ShelterWeakness[2][8] =
-  { { V(141), V(0), V(38), V(102), V(128), V(141), V(141) },
-    { V( 61), V(0), V(16), V( 44), V( 56), V( 61), V( 61) } };
 
-  // Danger of enemy pawns moving toward our king indexed by [pawn blocked][rank]
-  const Value StormDanger[2][8] =
-  { { V(26), V(0), V(128), V(51), V(26) },
-    { V(13), V(0), V( 64), V(25), V(13) } };
-
-  // Max bonus for king safety. Corresponds to start position with all the pawns
-  // in front of the king and no enemy pawn on the horizont.
-  const Value MaxSafetyBonus = V(263);
+  // 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(-18,11), S(-6,-3)  },
+                                     {  S(  0, 0), S( 5,-4) }};
 
   #undef S
   #undef V
-}
 
 
-/// PawnTable::probe() takes a position object as input, computes a PawnEntry
-/// object, and returns a pointer to it. The result is also stored in a hash
-/// table, so we don't have to recompute everything when the same pawn structure
-/// occurs again.
+  /// evaluate() calculates a score for the static pawn structure of the given position.
+  /// We cannot use the location of pieces or king in this function, as the evaluation
+  /// of the pawn structure will be stored in a small cache for speed reasons, and will
+  /// be re-used even when the pieces have moved.
 
-PawnEntry* PawnTable::probe(const Position& pos) {
+  template<Color Us>
+  Score evaluate(const Position& pos, Pawns::Entry* e) {
+
+    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;
+    Bitboard b = pos.pieces(Us, PAWN);
+
+    Bitboard ourPawns   = pos.pieces(  Us, PAWN);
+    Bitboard theirPawns = pos.pieces(Them, PAWN);
+
+    Bitboard doubleAttackThem = pawn_double_attacks_bb<Them>(theirPawns);
+
+    e->passedPawns[Us] = 0;
+    e->kingSquares[Us] = SQ_NONE;
+    e->pawnAttacks[Us] = e->pawnAttacksSpan[Us] = pawn_attacks_bb<Us>(ourPawns);
+    e->blockedCount += popcount(shift<Up>(ourPawns) & (theirPawns | doubleAttackThem));
+
+    // Loop through all pawns of the current color and score each pawn
+    while (b)
+    {
+        s = pop_lsb(b);
+
+        assert(pos.piece_on(s) == make_piece(Us, PAWN));
+
+        Rank r = relative_rank(Us, s);
+
+        // Flag the pawn
+        opposed    = theirPawns & forward_file_bb(Us, s);
+        blocked    = theirPawns & (s + Up);
+        stoppers   = theirPawns & passed_pawn_span(Us, s);
+        lever      = theirPawns & pawn_attacks_bb(Us, s);
+        leverPush  = theirPawns & pawn_attacks_bb(Us, s + Up);
+        doubled    = ourPawns   & (s - Up);
+        neighbours = ourPawns   & adjacent_files_bb(s);
+        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))
+                  && (leverPush | blocked);
+
+        // Compute additional span if pawn is not backward nor blocked
+        if (!backward && !blocked)
+            e->pawnAttacksSpan[Us] |= pawn_attack_span(Us, s);
+
+        // A pawn is passed if one of the three following conditions is true:
+        // (a) there is no stoppers except some levers
+        // (b) the only stoppers are the leverPush, but we outnumber them
+        // (c) there is only one front stopper which can be levered.
+        //     (Refined in Evaluation::passed)
+        passed =   !(stoppers ^ lever)
+                || (   !(stoppers ^ leverPush)
+                    && popcount(phalanx) >= popcount(leverPush))
+                || (   stoppers == blocked && r >= RANK_5
+                    && (shift<Up>(support) & ~(theirPawns | doubleAttackThem)));
+
+        passed &= !(forward_file_bb(Us, s) & ourPawns);
+
+        // Passed pawns will be properly scored later in evaluation when we have
+        // full attack info.
+        if (passed)
+            e->passedPawns[Us] |= s;
+
+        // Score this pawn
+        if (support | phalanx)
+        {
+            int v =  Connected[r] * (2 + bool(phalanx) - bool(opposed))
+                   + 22 * popcount(support);
+
+            score += make_score(v, v * (r - 2) / 4);
+        }
+
+        else if (!neighbours)
+        {
+            if (     opposed
+                &&  (ourPawns & forward_file_bb(Them, s))
+                && !(theirPawns & adjacent_files_bb(s)))
+                score -= Doubled;
+            else
+                score -=  Isolated
+                        + WeakUnopposed * !opposed;
+        }
+
+        else if (backward)
+            score -=  Backward
+                    + WeakUnopposed * !opposed * bool(~(FileABB | FileHBB) & s);
+
+        if (!support)
+            score -=  Doubled * doubled
+                    + WeakLever * more_than_one(lever);
+
+        if (blocked && r >= RANK_5)
+            score += BlockedPawn[r - RANK_5];
+    }
+
+    return score;
+  }
+
+} // namespace
+
+namespace Pawns {
+
+
+/// Pawns::probe() looks up the current position's pawns configuration in
+/// the pawns hash table. It returns a pointer to the Entry if the position
+/// is found. Otherwise a new Entry is computed and stored there, so we don't
+/// have to recompute all when the same pawns configuration occurs again.
+
+Entry* probe(const Position& pos) {
 
   Key key = pos.pawn_key();
-  PawnEntry* e = entries[key];
+  Entry* e = pos.this_thread()->pawnsTable[key];
 
-  // If e->key matches the position's pawn hash key, it means that we
-  // have analysed this pawn structure before, and we can simply return
-  // the information we found the last time instead of recomputing it.
   if (e->key == key)
       return e;
 
   e->key = key;
-  e->passedPawns[WHITE] = e->passedPawns[BLACK] = 0;
-  e->kingSquares[WHITE] = e->kingSquares[BLACK] = SQ_NONE;
-  e->halfOpenFiles[WHITE] = e->halfOpenFiles[BLACK] = 0xFF;
-
-  Bitboard wPawns = pos.pieces(WHITE, PAWN);
-  Bitboard bPawns = pos.pieces(BLACK, PAWN);
-  e->pawnAttacks[WHITE] = ((wPawns & ~FileHBB) << 9) | ((wPawns & ~FileABB) << 7);
-  e->pawnAttacks[BLACK] = ((bPawns & ~FileHBB) >> 7) | ((bPawns & ~FileABB) >> 9);
-
-  e->value =  evaluate_pawns<WHITE>(pos, wPawns, bPawns, e)
-            - evaluate_pawns<BLACK>(pos, bPawns, wPawns, e);
-
-  e->value = apply_weight(e->value, PawnStructureWeight);
+  e->blockedCount = 0;
+  e->scores[WHITE] = evaluate<WHITE>(pos, e);
+  e->scores[BLACK] = evaluate<BLACK>(pos, e);
 
   return e;
 }
 
 
-/// PawnTable::evaluate_pawns() evaluates each pawn of the given color
+/// Entry::evaluate_shelter() calculates the shelter bonus and the storm
+/// penalty for a king, looking at the king file and the two closest files.
 
 template<Color Us>
-Score PawnTable::evaluate_pawns(const Position& pos, Bitboard ourPawns,
-                                Bitboard theirPawns, PawnEntry* e) {
+Score Entry::evaluate_shelter(const Position& pos, Square ksq) const {
 
-  const Color Them = (Us == WHITE ? BLACK : WHITE);
+  constexpr Color Them = ~Us;
 
-  Bitboard b;
-  Square s;
-  File f;
-  Rank r;
-  bool passed, isolated, doubled, opposed, chain, backward, candidate;
-  Score value = SCORE_ZERO;
-  const Square* pl = pos.piece_list(Us, PAWN);
-
-  // Loop through all pawns of the current color and score each pawn
-  while ((s = *pl++) != SQ_NONE)
-  {
-      assert(pos.piece_on(s) == make_piece(Us, PAWN));
-
-      f = file_of(s);
-      r = rank_of(s);
-
-      // This file cannot be half open
-      e->halfOpenFiles[Us] &= ~(1 << f);
-
-      // Our rank plus previous one. Used for chain detection
-      b = rank_bb(r) | rank_bb(Us == WHITE ? r - Rank(1) : r + Rank(1));
-
-      // Flag the pawn as passed, isolated, doubled or member of a pawn
-      // chain (but not the backward one).
-      chain    =   ourPawns   & adjacent_files_bb(f) & b;
-      isolated = !(ourPawns   & adjacent_files_bb(f));
-      doubled  =   ourPawns   & forward_bb(Us, s);
-      opposed  =   theirPawns & forward_bb(Us, s);
-      passed   = !(theirPawns & passed_pawn_mask(Us, s));
-
-      // Test for backward pawn
-      backward = false;
-
-      // If the pawn is passed, isolated, or member of a pawn chain it cannot
-      // be backward. If there are friendly pawns behind on adjacent files
-      // or if can capture an enemy pawn it cannot be backward either.
-      if (   !(passed | isolated | chain)
-          && !(ourPawns & attack_span_mask(Them, s))
-          && !(pos.attacks_from<PAWN>(s, Us) & theirPawns))
-      {
-          // We now know that there are no friendly pawns beside or behind this
-          // pawn on adjacent files. We now check whether the pawn is
-          // backward by looking in the forward direction on the adjacent
-          // files, and seeing whether we meet a friendly or an enemy pawn first.
-          b = pos.attacks_from<PAWN>(s, Us);
-
-          // Note that we are sure to find something because pawn is not passed
-          // nor isolated, so loop is potentially infinite, but it isn't.
-          while (!(b & (ourPawns | theirPawns)))
-              Us == WHITE ? b <<= 8 : b >>= 8;
-
-          // The friendly pawn needs to be at least two ranks closer than the
-          // enemy pawn in order to help the potentially backward pawn advance.
-          backward = (b | (Us == WHITE ? b << 8 : b >> 8)) & theirPawns;
-      }
-
-      assert(opposed | passed | (attack_span_mask(Us, s) & theirPawns));
-
-      // A not passed pawn is a candidate to become passed if it is free to
-      // advance and if the number of friendly pawns beside or behind this
-      // pawn on adjacent files is higher or equal than the number of
-      // enemy pawns in the forward direction on the adjacent files.
-      candidate =   !(opposed | passed | backward | isolated)
-                 && (b = attack_span_mask(Them, s + pawn_push(Us)) & ourPawns) != 0
-                 &&  popcount<Max15>(b) >= popcount<Max15>(attack_span_mask(Us, s) & theirPawns);
-
-      // Passed pawns will be properly scored in evaluation because we need
-      // full attack info to evaluate passed pawns. Only the frontmost passed
-      // pawn on each file is considered a true passed pawn.
-      if (passed && !doubled)
-          e->passedPawns[Us] |= s;
-
-      // Score this pawn
-      if (isolated)
-          value -= IsolatedPawnPenalty[opposed][f];
-
-      if (doubled)
-          value -= DoubledPawnPenalty[opposed][f];
-
-      if (backward)
-          value -= BackwardPawnPenalty[opposed][f];
-
-      if (chain)
-          value += ChainBonus[f];
-
-      if (candidate)
-          value += CandidateBonus[relative_rank(Us, s)];
-  }
-
-  return value;
-}
-
-
-/// PawnEntry::shelter_storm() calculates shelter and storm penalties for the file
-/// the king is on, as well as the two adjacent files.
-
-template<Color Us>
-Value PawnEntry::shelter_storm(const Position& pos, Square ksq) {
-
-  const Color Them = (Us == WHITE ? BLACK : WHITE);
-
-  Value safety = MaxSafetyBonus;
-  Bitboard b = pos.pieces(PAWN) & (in_front_bb(Us, ksq) | rank_bb(ksq));
-  Bitboard ourPawns = b & pos.pieces(Us) & ~rank_bb(ksq);
+  Bitboard b = pos.pieces(PAWN) & ~forward_ranks_bb(Them, ksq);
+  Bitboard ourPawns = b & pos.pieces(Us) & ~pawnAttacks[Them];
   Bitboard theirPawns = b & pos.pieces(Them);
-  Rank rkUs, rkThem;
-  File kf = file_of(ksq);
 
-  kf = (kf == FILE_A) ? kf++ : (kf == FILE_H) ? kf-- : kf;
+  Score bonus = make_score(5, 5);
 
-  for (int f = kf - 1; f <= kf + 1; f++)
+  File center = std::clamp(file_of(ksq), FILE_B, FILE_G);
+  for (File f = File(center - 1); f <= File(center + 1); ++f)
   {
-      // Shelter penalty is higher for the pawn in front of the king
-      b = ourPawns & FileBB[f];
-      rkUs = b ? rank_of(Us == WHITE ? lsb(b) : ~msb(b)) : RANK_1;
-      safety -= ShelterWeakness[f != kf][rkUs];
+      b = ourPawns & file_bb(f);
+      int ourRank = b ? relative_rank(Us, frontmost_sq(Them, b)) : 0;
 
-      // Storm danger is smaller if enemy pawn is blocked
-      b  = theirPawns & FileBB[f];
-      rkThem = b ? rank_of(Us == WHITE ? lsb(b) : ~msb(b)) : RANK_1;
-      safety -= StormDanger[rkThem == rkUs + 1][rkThem];
+      b = theirPawns & file_bb(f);
+      int theirRank = b ? relative_rank(Us, frontmost_sq(Them, b)) : 0;
+
+      int d = edge_distance(f);
+      bonus += make_score(ShelterStrength[d][ourRank], 0);
+
+      if (ourRank && (ourRank == theirRank - 1))
+          bonus -= BlockedStorm[theirRank];
+      else
+          bonus -= make_score(UnblockedStorm[d][theirRank], 0);
   }
 
-  return safety;
+  // King On File
+  bonus -= KingOnFile[pos.is_on_semiopen_file(Us, ksq)][pos.is_on_semiopen_file(Them, ksq)];
+
+  return bonus;
 }
 
 
-/// PawnEntry::update_safety() calculates and caches a bonus for king safety. It is
-/// called only when king square changes, about 20% of total king_safety() calls.
+/// Entry::do_king_safety() calculates a bonus for king safety. It is called only
+/// when king square changes, which is about 20% of total king_safety() calls.
 
 template<Color Us>
-Score PawnEntry::update_safety(const Position& pos, Square ksq) {
+Score Entry::do_king_safety(const Position& pos) {
 
+  Square ksq = pos.square<KING>(Us);
   kingSquares[Us] = ksq;
-  castleRights[Us] = pos.can_castle(Us);
-  minKPdistance[Us] = 0;
+  castlingRights[Us] = pos.castling_rights(Us);
+  auto compare = [](Score a, Score b) { return mg_value(a) < mg_value(b); };
 
+  Score shelter = evaluate_shelter<Us>(pos, ksq);
+
+  // If we can castle use the bonus after castling if it is bigger
+
+  if (pos.can_castle(Us & KING_SIDE))
+      shelter = std::max(shelter, evaluate_shelter<Us>(pos, relative_square(Us, SQ_G1)), compare);
+
+  if (pos.can_castle(Us & QUEEN_SIDE))
+      shelter = std::max(shelter, evaluate_shelter<Us>(pos, relative_square(Us, SQ_C1)), compare);
+
+  // In endgame we like to bring our king near our closest pawn
   Bitboard pawns = pos.pieces(Us, PAWN);
-  if (pawns)
-      while (!(DistanceRingsBB[ksq][minKPdistance[Us]++] & pawns)) {}
+  int minPawnDist = 6;
 
-  if (relative_rank(Us, ksq) > RANK_4)
-      return kingSafety[Us] = make_score(0, -16 * minKPdistance[Us]);
+  if (pawns & attacks_bb<KING>(ksq))
+      minPawnDist = 1;
+  else while (pawns)
+      minPawnDist = std::min(minPawnDist, distance(ksq, pop_lsb(pawns)));
 
-  Value bonus = shelter_storm<Us>(pos, ksq);
-
-  // If we can castle use the bonus after the castle if is bigger
-  if (pos.can_castle(make_castle_right(Us, KING_SIDE)))
-      bonus = std::max(bonus, shelter_storm<Us>(pos, relative_square(Us, SQ_G1)));
-
-  if (pos.can_castle(make_castle_right(Us, QUEEN_SIDE)))
-      bonus = std::max(bonus, shelter_storm<Us>(pos, relative_square(Us, SQ_C1)));
-
-  return kingSafety[Us] = make_score(bonus, -16 * minKPdistance[Us]);
+  return shelter - make_score(0, 16 * minPawnDist);
 }
 
 // Explicit template instantiation
-template Score PawnEntry::update_safety<WHITE>(const Position& pos, Square ksq);
-template Score PawnEntry::update_safety<BLACK>(const Position& pos, Square ksq);
+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

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,99 +16,55 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(PAWNS_H_INCLUDED)
+#ifndef PAWNS_H_INCLUDED
 #define PAWNS_H_INCLUDED
 
 #include "misc.h"
 #include "position.h"
 #include "types.h"
 
-const int PawnTableSize = 16384;
+namespace Stockfish::Pawns {
 
-/// PawnEntry is a class which contains various information about a pawn
-/// structure. Currently, it only includes a middle game and an end game
-/// pawn structure evaluation, and a bitboard of passed pawns. We may want
-/// to add further information in the future. A lookup to the pawn hash
-/// table (performed by calling the probe method in a PawnTable object)
-/// returns a pointer to a PawnEntry object.
+/// Pawns::Entry contains various information about a pawn structure. A lookup
+/// to the pawn hash table (performed by calling the probe function) returns a
+/// pointer to an Entry object.
 
-class PawnEntry {
+struct Entry {
 
-  friend struct PawnTable;
-
-public:
-  Score pawns_value() const;
-  Bitboard pawn_attacks(Color c) const;
-  Bitboard passed_pawns(Color c) const;
-  int file_is_half_open(Color c, File f) const;
-  int has_open_file_to_left(Color c, File f) const;
-  int has_open_file_to_right(Color c, File f) const;
+  Score pawn_score(Color c) const { return scores[c]; }
+  Bitboard pawn_attacks(Color c) const { return pawnAttacks[c]; }
+  Bitboard passed_pawns(Color c) const { return passedPawns[c]; }
+  Bitboard pawn_attacks_span(Color c) const { return pawnAttacksSpan[c]; }
+  int passed_count() const { return popcount(passedPawns[WHITE] | passedPawns[BLACK]); }
+  int blocked_count() const { return blockedCount; }
 
   template<Color Us>
-  Score king_safety(const Position& pos, Square ksq);
-
-private:
-  template<Color Us>
-  Score update_safety(const Position& pos, Square ksq);
+  Score king_safety(const Position& pos) {
+    return  kingSquares[Us] == pos.square<KING>(Us) && castlingRights[Us] == pos.castling_rights(Us)
+          ? kingSafety[Us] : (kingSafety[Us] = do_king_safety<Us>(pos));
+  }
 
   template<Color Us>
-  Value shelter_storm(const Position& pos, Square ksq);
+  Score do_king_safety(const Position& pos);
+
+  template<Color Us>
+  Score evaluate_shelter(const Position& pos, Square ksq) const;
 
   Key key;
-  Bitboard passedPawns[2];
-  Bitboard pawnAttacks[2];
-  Square kingSquares[2];
-  int minKPdistance[2];
-  int castleRights[2];
-  Score value;
-  int halfOpenFiles[2];
-  Score kingSafety[2];
+  Score scores[COLOR_NB];
+  Bitboard passedPawns[COLOR_NB];
+  Bitboard pawnAttacks[COLOR_NB];
+  Bitboard pawnAttacksSpan[COLOR_NB];
+  Square kingSquares[COLOR_NB];
+  Score kingSafety[COLOR_NB];
+  int castlingRights[COLOR_NB];
+  int blockedCount;
 };
 
+using Table = HashTable<Entry, 131072>;
 
-/// The PawnTable class represents a pawn hash table. The most important
-/// method is probe, which returns a pointer to a PawnEntry object.
+Entry* probe(const Position& pos);
 
-struct PawnTable {
+} // namespace Stockfish::Pawns
 
-  PawnEntry* probe(const Position& pos);
-
-  template<Color Us>
-  static Score evaluate_pawns(const Position& pos, Bitboard ourPawns,
-                              Bitboard theirPawns, PawnEntry* e);
-
-  HashTable<PawnEntry, PawnTableSize> entries;
-};
-
-
-inline Score PawnEntry::pawns_value() const {
-  return value;
-}
-
-inline Bitboard PawnEntry::pawn_attacks(Color c) const {
-  return pawnAttacks[c];
-}
-
-inline Bitboard PawnEntry::passed_pawns(Color c) const {
-  return passedPawns[c];
-}
-
-inline int PawnEntry::file_is_half_open(Color c, File f) const {
-  return halfOpenFiles[c] & (1 << int(f));
-}
-
-inline int PawnEntry::has_open_file_to_left(Color c, File f) const {
-  return halfOpenFiles[c] & ((1 << int(f)) - 1);
-}
-
-inline int PawnEntry::has_open_file_to_right(Color c, File f) const {
-  return halfOpenFiles[c] & ~((1 << int(f+1)) - 1);
-}
-
-template<Color Us>
-inline Score PawnEntry::king_safety(const Position& pos, Square ksq) {
-  return kingSquares[Us] == ksq && castleRights[Us] == pos.can_castle(Us)
-       ? kingSafety[Us] : update_safety<Us>(pos, ksq);
-}
-
-#endif // !defined(PAWNS_H_INCLUDED)
+#endif // #ifndef PAWNS_H_INCLUDED

src/platform.h

@@ -1,109 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
-
-  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/>.
-*/
-
-#if !defined(PLATFORM_H_INCLUDED)
-#define PLATFORM_H_INCLUDED
-
-#if defined(_MSC_VER)
-
-// Disable some silly and noisy warning from MSVC compiler
-#pragma warning(disable: 4127) // Conditional expression is constant
-#pragma warning(disable: 4146) // Unary minus operator applied to unsigned type
-#pragma warning(disable: 4800) // Forcing value to bool 'true' or 'false'
-#pragma warning(disable: 4996) // Function _ftime() may be unsafe
-
-// MSVC does not support <inttypes.h>
-typedef   signed __int8    int8_t;
-typedef unsigned __int8   uint8_t;
-typedef   signed __int16  int16_t;
-typedef unsigned __int16 uint16_t;
-typedef   signed __int32  int32_t;
-typedef unsigned __int32 uint32_t;
-typedef   signed __int64  int64_t;
-typedef unsigned __int64 uint64_t;
-
-#else
-#  include <inttypes.h>
-#endif
-
-#if !defined(_WIN32) && !defined(_WIN64) // Linux - Unix
-
-#  include <sys/time.h>
-typedef timeval sys_time_t;
-
-inline void system_time(sys_time_t* t) { gettimeofday(t, NULL); }
-inline int64_t time_to_msec(const sys_time_t& t) { return t.tv_sec * 1000LL + t.tv_usec / 1000; }
-
-#  include <pthread.h>
-typedef pthread_mutex_t Lock;
-typedef pthread_cond_t WaitCondition;
-typedef pthread_t NativeHandle;
-typedef void*(*pt_start_fn)(void*);
-
-#  define lock_init(x) pthread_mutex_init(&(x), NULL)
-#  define lock_grab(x) pthread_mutex_lock(&(x))
-#  define lock_release(x) pthread_mutex_unlock(&(x))
-#  define lock_destroy(x) pthread_mutex_destroy(&(x))
-#  define cond_destroy(x) pthread_cond_destroy(&(x))
-#  define cond_init(x) pthread_cond_init(&(x), NULL)
-#  define cond_signal(x) pthread_cond_signal(&(x))
-#  define cond_wait(x,y) pthread_cond_wait(&(x),&(y))
-#  define cond_timedwait(x,y,z) pthread_cond_timedwait(&(x),&(y),z)
-#  define thread_create(x,f,t) !pthread_create(&(x),NULL,(pt_start_fn)f,t)
-#  define thread_join(x) pthread_join(x, NULL)
-
-#else // Windows and MinGW
-
-#  include <sys/timeb.h>
-typedef _timeb sys_time_t;
-
-inline void system_time(sys_time_t* t) { _ftime(t); }
-inline int64_t time_to_msec(const sys_time_t& t) { return t.time * 1000LL + t.millitm; }
-
-#if !defined(NOMINMAX)
-#  define NOMINMAX // disable macros min() and max()
-#endif
-
-#define WIN32_LEAN_AND_MEAN
-#include <windows.h>
-#undef WIN32_LEAN_AND_MEAN
-#undef NOMINMAX
-
-// We use critical sections on Windows to support Windows XP and older versions,
-// unfortunatly cond_wait() is racy between lock_release() and WaitForSingleObject()
-// but apart from this they have the same speed performance of SRW locks.
-typedef CRITICAL_SECTION Lock;
-typedef HANDLE WaitCondition;
-typedef HANDLE NativeHandle;
-
-#  define lock_init(x) InitializeCriticalSection(&(x))
-#  define lock_grab(x) EnterCriticalSection(&(x))
-#  define lock_release(x) LeaveCriticalSection(&(x))
-#  define lock_destroy(x) DeleteCriticalSection(&(x))
-#  define cond_init(x) { x = CreateEvent(0, FALSE, FALSE, 0); }
-#  define cond_destroy(x) CloseHandle(x)
-#  define cond_signal(x) SetEvent(x)
-#  define cond_wait(x,y) { lock_release(y); WaitForSingleObject(x, INFINITE); lock_grab(y); }
-#  define cond_timedwait(x,y,z) { lock_release(y); WaitForSingleObject(x,z); lock_grab(y); }
-#  define thread_create(x,f,t) (x = CreateThread(NULL,0,(LPTHREAD_START_ROUTINE)f,t,0,NULL), x != NULL)
-#  define thread_join(x) { WaitForSingleObject(x, INFINITE); CloseHandle(x); }
-
-#endif
-
-#endif // !defined(PLATFORM_H_INCLUDED)

src/position.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,345 +16,326 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(POSITION_H_INCLUDED)
+#ifndef POSITION_H_INCLUDED
 #define POSITION_H_INCLUDED
 
 #include <cassert>
+#include <deque>
+#include <memory> // For std::unique_ptr
+#include <string>
 
 #include "bitboard.h"
+#include "evaluate.h"
+#include "psqt.h"
 #include "types.h"
 
+#include "nnue/nnue_accumulator.h"
 
-/// The checkInfo struct is initialized at c'tor time and keeps info used
-/// to detect if a move gives check.
-class Position;
-class Thread;
+namespace Stockfish {
 
-struct CheckInfo {
-
-  explicit CheckInfo(const Position&);
-
-  Bitboard dcCandidates;
-  Bitboard pinned;
-  Bitboard checkSq[8];
-  Square ksq;
-};
-
-
-/// The StateInfo struct stores information we need to restore a Position
-/// object to its previous state when we retract a move. Whenever a move
-/// is made on the board (by calling Position::do_move), an StateInfo object
-/// must be passed as a parameter.
+/// 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
+/// board (by calling Position::do_move), a StateInfo object must be passed.
 
 struct StateInfo {
-  Key pawnKey, materialKey;
-  Value npMaterial[2];
-  int castleRights, rule50, pliesFromNull;
-  Score psqScore;
+
+  // Copied when making a move
+  Key    pawnKey;
+  Key    materialKey;
+  Value  nonPawnMaterial[COLOR_NB];
+  int    castlingRights;
+  int    rule50;
+  int    pliesFromNull;
   Square epSquare;
 
-  Key key;
-  Bitboard checkersBB;
-  PieceType capturedType;
+  // Not copied when making a move (will be recomputed anyhow)
+  Key        key;
+  Bitboard   checkersBB;
   StateInfo* previous;
-};
+  Bitboard   blockersForKing[COLOR_NB];
+  Bitboard   pinners[COLOR_NB];
+  Bitboard   checkSquares[PIECE_TYPE_NB];
+  Piece      capturedPiece;
+  int        repetition;
 
-struct ReducedStateInfo {
-  Key pawnKey, materialKey;
-  Value npMaterial[2];
-  int castleRights, rule50, pliesFromNull;
-  Score psqScore;
-  Square epSquare;
+  // Used by NNUE
+  Eval::NNUE::Accumulator accumulator;
+  DirtyPiece dirtyPiece;
 };
 
 
-/// The position data structure. A position consists of the following data:
-///
-///    * For each piece type, a bitboard representing the squares occupied
-///      by pieces of that type.
-///    * For each color, a bitboard representing the squares occupied by
-///      pieces of that color.
-///    * A bitboard of all occupied squares.
-///    * A bitboard of all checking pieces.
-///    * A 64-entry array of pieces, indexed by the squares of the board.
-///    * The current side to move.
-///    * Information about the castling rights for both sides.
-///    * The initial files of the kings and both pairs of rooks. This is
-///      used to implement the Chess960 castling rules.
-///    * The en passant square (which is SQ_NONE if no en passant capture is
-///      possible).
-///    * The squares of the kings for both sides.
-///    * Hash keys for the position itself, the current pawn structure, and
-///      the current material situation.
-///    * Hash keys for all previous positions in the game for detecting
-///      repetition draws.
-///    * A counter for detecting 50 move rule draws.
+/// A list to keep track of the position states along the setup moves (from the
+/// start position to the position just before the search starts). Needed by
+/// 'draw by repetition' detection. Use a std::deque because pointers to
+/// elements are not invalidated upon list resizing.
+using StateListPtr = std::unique_ptr<std::deque<StateInfo>>;
+
+
+/// Position class stores information regarding the board representation as
+/// pieces, side to move, hash keys, castling info, etc. Important methods are
+/// do_move() and undo_move(), used by the search to update node info when
+/// traversing the search tree.
+class Thread;
 
 class Position {
 public:
-  Position() {}
-  Position(const Position& p, Thread* t) { *this = p; thisThread = t; }
-  Position(const std::string& f, bool c960, Thread* t) { from_fen(f, c960, t); }
-  Position& operator=(const Position&);
+  static void init();
 
-  // Text input/output
-  void from_fen(const std::string& fen, bool isChess960, Thread* th);
-  const std::string to_fen() const;
-  void print(Move m = MOVE_NONE) const;
+  Position() = default;
+  Position(const Position&) = delete;
+  Position& operator=(const Position&) = delete;
+
+  // 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);
+  std::string fen() const;
 
   // Position representation
-  Bitboard pieces() const;
   Bitboard pieces(PieceType pt) const;
-  Bitboard pieces(PieceType pt1, PieceType pt2) const;
+  template<typename ...PieceTypes> Bitboard pieces(PieceType pt, PieceTypes... pts) const;
   Bitboard pieces(Color c) const;
-  Bitboard pieces(Color c, PieceType pt) const;
-  Bitboard pieces(Color c, PieceType pt1, PieceType pt2) const;
+  template<typename ...PieceTypes> Bitboard pieces(Color c, PieceTypes... pts) const;
   Piece piece_on(Square s) const;
-  Square king_square(Color c) const;
   Square ep_square() const;
-  bool is_empty(Square s) const;
-  const Square* piece_list(Color c, PieceType pt) const;
-  int piece_count(Color c, PieceType pt) const;
+  bool empty(Square s) const;
+  template<PieceType Pt> int count(Color c) const;
+  template<PieceType Pt> int count() const;
+  template<PieceType Pt> Square square(Color c) const;
+  bool is_on_semiopen_file(Color c, Square s) const;
 
   // Castling
-  int can_castle(CastleRight f) const;
-  int can_castle(Color c) const;
-  bool castle_impeded(Color c, CastlingSide s) const;
-  Square castle_rook_square(Color c, CastlingSide s) const;
+  CastlingRights castling_rights(Color c) const;
+  bool can_castle(CastlingRights cr) const;
+  bool castling_impeded(CastlingRights cr) const;
+  Square castling_rook_square(CastlingRights cr) const;
 
   // Checking
-  bool in_check() const;
   Bitboard checkers() const;
-  Bitboard discovered_check_candidates() const;
-  Bitboard pinned_pieces() const;
+  Bitboard blockers_for_king(Color c) const;
+  Bitboard check_squares(PieceType pt) const;
+  Bitboard pinners(Color c) const;
 
   // Attacks to/from a given square
   Bitboard attackers_to(Square s) const;
-  Bitboard attackers_to(Square s, Bitboard occ) const;
-  Bitboard attacks_from(Piece p, Square s) const;
-  static Bitboard attacks_from(Piece p, Square s, Bitboard occ);
-  template<PieceType> Bitboard attacks_from(Square s) const;
-  template<PieceType> Bitboard attacks_from(Square s, Color c) const;
+  Bitboard attackers_to(Square s, Bitboard occupied) const;
+  Bitboard slider_blockers(Bitboard sliders, Square s, Bitboard& pinners) const;
+  template<PieceType Pt> Bitboard attacks_by(Color c) const;
 
   // Properties of moves
-  bool move_gives_check(Move m, const CheckInfo& ci) const;
-  bool move_attacks_square(Move m, Square s) const;
-  bool move_is_legal(const Move m) const;
-  bool pl_move_is_legal(Move m, Bitboard pinned) const;
-  bool is_pseudo_legal(const Move m) const;
-  bool is_capture(Move m) const;
-  bool is_capture_or_promotion(Move m) const;
-  bool is_passed_pawn_push(Move m) const;
-  Piece piece_moved(Move m) const;
-  PieceType captured_piece_type() const;
+  bool legal(Move m) const;
+  bool pseudo_legal(const Move m) const;
+  bool capture(Move m) const;
+  bool capture_stage(Move m) const;
+  bool gives_check(Move m) const;
+  Piece moved_piece(Move m) const;
+  Piece captured_piece() const;
 
   // Piece specific
-  bool pawn_is_passed(Color c, Square s) const;
-  bool pawn_on_7th(Color c) const;
+  bool pawn_passed(Color c, Square s) const;
   bool opposite_bishops() const;
-  bool bishop_pair(Color c) const;
+  int  pawns_on_same_color_squares(Color c, Square s) const;
 
   // Doing and undoing moves
-  void do_move(Move m, StateInfo& st);
-  void do_move(Move m, StateInfo& st, const CheckInfo& ci, bool moveIsCheck);
+  void do_move(Move m, StateInfo& newSt);
+  void do_move(Move m, StateInfo& newSt, bool givesCheck);
   void undo_move(Move m);
-  template<bool Do> void do_null_move(StateInfo& st);
+  void do_null_move(StateInfo& newSt);
+  void undo_null_move();
 
-  // Static exchange evaluation
-  int see(Move m) const;
-  int see_sign(Move m) const;
+  // Static Exchange Evaluation
+  bool see_ge(Move m, Value threshold = VALUE_ZERO) const;
+  bool see_ge(Move m, Bitboard& occupied, Value threshold = VALUE_ZERO) const;
 
   // Accessing hash keys
   Key key() const;
-  Key exclusion_key() const;
-  Key pawn_key() const;
+  Key key_after(Move m) const;
   Key material_key() const;
-
-  // Incremental piece-square evaluation
-  Score psq_score() const;
-  Score psq_delta(Piece p, Square from, Square to) const;
-  Value non_pawn_material(Color c) const;
+  Key pawn_key() const;
 
   // Other properties of the position
   Color side_to_move() const;
-  int startpos_ply_counter() const;
+  int game_ply() const;
   bool is_chess960() const;
   Thread* this_thread() const;
-  int64_t nodes_searched() const;
-  void set_nodes_searched(int64_t n);
-  template<bool SkipRepetition> bool is_draw() const;
+  bool is_draw(int ply) const;
+  bool has_game_cycle(int ply) const;
+  bool has_repeated() const;
+  int rule50_count() const;
+  Score psq_score() const;
+  Value psq_eg_stm() const;
+  Value non_pawn_material(Color c) const;
+  Value non_pawn_material() const;
 
   // Position consistency check, for debugging
-  bool pos_is_ok(int* failedStep = NULL) const;
+  bool pos_is_ok() const;
   void flip();
 
+  // 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 clear();
-  void put_piece(Piece p, Square s);
-  void set_castle_right(Color c, Square rfrom);
+  void set_castling_right(Color c, Square rfrom);
+  void set_state() const;
+  void set_check_info() const;
 
-  // Helper template functions
-  template<bool Do> void do_castle_move(Move m);
-  template<bool FindPinned> Bitboard hidden_checkers() const;
+  // Other helpers
+  void move_piece(Square from, Square to);
+  template<bool Do>
+  void do_castling(Color us, Square from, Square& to, Square& rfrom, Square& rto);
+  template<bool AfterMove>
+  Key adjust_key50(Key k) const;
 
-  // Computing hash keys from scratch (for initialization and debugging)
-  Key compute_key() const;
-  Key compute_pawn_key() const;
-  Key compute_material_key() const;
-
-  // Computing incremental evaluation scores and material counts
-  Score compute_psq_score() const;
-  Value compute_non_pawn_material(Color c) const;
-
-  // Board and pieces
-  Piece board[64];             // [square]
-  Bitboard byTypeBB[8];        // [pieceType]
-  Bitboard byColorBB[2];       // [color]
-  int pieceCount[2][8];        // [color][pieceType]
-  Square pieceList[2][8][16];  // [color][pieceType][index]
-  int index[64];               // [square]
-
-  // Other info
-  int castleRightsMask[64];      // [square]
-  Square castleRookSquare[2][2]; // [color][side]
-  Bitboard castlePath[2][2];     // [color][side]
-  StateInfo startState;
-  int64_t nodes;
-  int startPosPly;
-  Color sideToMove;
+  // Data members
+  Piece board[SQUARE_NB];
+  Bitboard byTypeBB[PIECE_TYPE_NB];
+  Bitboard byColorBB[COLOR_NB];
+  int pieceCount[PIECE_NB];
+  int castlingRightsMask[SQUARE_NB];
+  Square castlingRookSquare[CASTLING_RIGHT_NB];
+  Bitboard castlingPath[CASTLING_RIGHT_NB];
   Thread* thisThread;
   StateInfo* st;
-  int chess960;
+  int gamePly;
+  Color sideToMove;
+  Score psq;
+  bool chess960;
 };
 
-inline int64_t Position::nodes_searched() const {
-  return nodes;
-}
-
-inline void Position::set_nodes_searched(int64_t n) {
-  nodes = n;
-}
-
-inline Piece Position::piece_on(Square s) const {
-  return board[s];
-}
-
-inline Piece Position::piece_moved(Move m) const {
-  return board[from_sq(m)];
-}
-
-inline bool Position::is_empty(Square s) const {
-  return board[s] == NO_PIECE;
-}
+std::ostream& operator<<(std::ostream& os, const Position& pos);
 
 inline Color Position::side_to_move() const {
   return sideToMove;
 }
 
-inline Bitboard Position::pieces() const {
-  return byTypeBB[ALL_PIECES];
+inline Piece Position::piece_on(Square s) const {
+  assert(is_ok(s));
+  return board[s];
 }
 
-inline Bitboard Position::pieces(PieceType pt) const {
+inline bool Position::empty(Square s) const {
+  return piece_on(s) == NO_PIECE;
+}
+
+inline Piece Position::moved_piece(Move m) const {
+  return piece_on(from_sq(m));
+}
+
+inline Bitboard Position::pieces(PieceType pt = ALL_PIECES) const {
   return byTypeBB[pt];
 }
 
-inline Bitboard Position::pieces(PieceType pt1, PieceType pt2) const {
-  return byTypeBB[pt1] | byTypeBB[pt2];
+template<typename ...PieceTypes>
+inline Bitboard Position::pieces(PieceType pt, PieceTypes... pts) const {
+  return pieces(pt) | pieces(pts...);
 }
 
 inline Bitboard Position::pieces(Color c) const {
   return byColorBB[c];
 }
 
-inline Bitboard Position::pieces(Color c, PieceType pt) const {
-  return byColorBB[c] & byTypeBB[pt];
+template<typename ...PieceTypes>
+inline Bitboard Position::pieces(Color c, PieceTypes... pts) const {
+  return pieces(c) & pieces(pts...);
 }
 
-inline Bitboard Position::pieces(Color c, PieceType pt1, PieceType pt2) const {
-  return byColorBB[c] & (byTypeBB[pt1] | byTypeBB[pt2]);
+template<PieceType Pt> inline int Position::count(Color c) const {
+  return pieceCount[make_piece(c, Pt)];
 }
 
-inline int Position::piece_count(Color c, PieceType pt) const {
-  return pieceCount[c][pt];
+template<PieceType Pt> inline int Position::count() const {
+  return count<Pt>(WHITE) + count<Pt>(BLACK);
 }
 
-inline const Square* Position::piece_list(Color c, PieceType pt) const {
-  return pieceList[c][pt];
+template<PieceType Pt> inline Square Position::square(Color c) const {
+  assert(count<Pt>(c) == 1);
+  return lsb(pieces(c, Pt));
 }
 
 inline Square Position::ep_square() const {
   return st->epSquare;
 }
 
-inline Square Position::king_square(Color c) const {
-  return pieceList[c][KING][0];
+inline bool Position::is_on_semiopen_file(Color c, Square s) const {
+  return !(pieces(c, PAWN) & file_bb(s));
 }
 
-inline int Position::can_castle(CastleRight f) const {
-  return st->castleRights & f;
+inline bool Position::can_castle(CastlingRights cr) const {
+  return st->castlingRights & cr;
 }
 
-inline int Position::can_castle(Color c) const {
-  return st->castleRights & ((WHITE_OO | WHITE_OOO) << (2 * c));
+inline CastlingRights Position::castling_rights(Color c) const {
+  return c & CastlingRights(st->castlingRights);
 }
 
-inline bool Position::castle_impeded(Color c, CastlingSide s) const {
-  return byTypeBB[ALL_PIECES] & castlePath[c][s];
+inline bool Position::castling_impeded(CastlingRights cr) const {
+  assert(cr == WHITE_OO || cr == WHITE_OOO || cr == BLACK_OO || cr == BLACK_OOO);
+
+  return pieces() & castlingPath[cr];
 }
 
-inline Square Position::castle_rook_square(Color c, CastlingSide s) const {
-  return castleRookSquare[c][s];
-}
+inline Square Position::castling_rook_square(CastlingRights cr) const {
+  assert(cr == WHITE_OO || cr == WHITE_OOO || cr == BLACK_OO || cr == BLACK_OOO);
 
-template<PieceType Pt>
-inline Bitboard Position::attacks_from(Square s) const {
-
-  return  Pt == BISHOP || Pt == ROOK ? attacks_bb<Pt>(s, pieces())
-        : Pt == QUEEN  ? attacks_from<ROOK>(s) | attacks_from<BISHOP>(s)
-        : StepAttacksBB[Pt][s];
-}
-
-template<>
-inline Bitboard Position::attacks_from<PAWN>(Square s, Color c) const {
-  return StepAttacksBB[make_piece(c, PAWN)][s];
-}
-
-inline Bitboard Position::attacks_from(Piece p, Square s) const {
-  return attacks_from(p, s, byTypeBB[ALL_PIECES]);
+  return castlingRookSquare[cr];
 }
 
 inline Bitboard Position::attackers_to(Square s) const {
-  return attackers_to(s, byTypeBB[ALL_PIECES]);
+  return attackers_to(s, pieces());
+}
+
+template<PieceType Pt>
+inline Bitboard Position::attacks_by(Color c) const {
+
+  if constexpr (Pt == PAWN)
+      return c == WHITE ? pawn_attacks_bb<WHITE>(pieces(WHITE, PAWN))
+                        : pawn_attacks_bb<BLACK>(pieces(BLACK, PAWN));
+  else
+  {
+      Bitboard threats = 0;
+      Bitboard attackers = pieces(c, Pt);
+      while (attackers)
+          threats |= attacks_bb<Pt>(pop_lsb(attackers), pieces());
+      return threats;
+  }
 }
 
 inline Bitboard Position::checkers() const {
   return st->checkersBB;
 }
 
-inline bool Position::in_check() const {
-  return st->checkersBB != 0;
+inline Bitboard Position::blockers_for_king(Color c) const {
+  return st->blockersForKing[c];
 }
 
-inline Bitboard Position::discovered_check_candidates() const {
-  return hidden_checkers<false>();
+inline Bitboard Position::pinners(Color c) const {
+  return st->pinners[c];
 }
 
-inline Bitboard Position::pinned_pieces() const {
-  return hidden_checkers<true>();
+inline Bitboard Position::check_squares(PieceType pt) const {
+  return st->checkSquares[pt];
 }
 
-inline bool Position::pawn_is_passed(Color c, Square s) const {
-  return !(pieces(~c, PAWN) & passed_pawn_mask(c, s));
+inline bool Position::pawn_passed(Color c, Square s) const {
+  return !(pieces(~c, PAWN) & passed_pawn_span(c, s));
+}
+
+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 adjust_key50<false>(st->key);
 }
 
-inline Key Position::exclusion_key() const {
-  return st->key ^ Zobrist::exclusion;
+template<bool AfterMove>
+inline Key Position::adjust_key50(Key k) const
+{
+  return st->rule50 < 14 - AfterMove
+      ? k : k ^ make_key((st->rule50 - (14 - AfterMove)) / 8);
 }
 
 inline Key Position::pawn_key() const {
@@ -366,68 +346,105 @@ inline Key Position::material_key() const {
   return st->materialKey;
 }
 
-inline Score Position::psq_delta(Piece p, Square from, Square to) const {
-  return pieceSquareTable[p][to] - pieceSquareTable[p][from];
+inline Score Position::psq_score() const {
+  return psq;
 }
 
-inline Score Position::psq_score() const {
-  return st->psqScore;
+inline Value Position::psq_eg_stm() const {
+  return (sideToMove == WHITE ? 1 : -1) * eg_value(psq);
 }
 
 inline Value Position::non_pawn_material(Color c) const {
-  return st->npMaterial[c];
+  return st->nonPawnMaterial[c];
 }
 
-inline bool Position::is_passed_pawn_push(Move m) const {
-
-  return   type_of(piece_moved(m)) == PAWN
-        && pawn_is_passed(sideToMove, to_sq(m));
+inline Value Position::non_pawn_material() const {
+  return non_pawn_material(WHITE) + non_pawn_material(BLACK);
 }
 
-inline int Position::startpos_ply_counter() const {
-  return startPosPly + st->pliesFromNull; // HACK
+inline int Position::game_ply() const {
+  return gamePly;
+}
+
+inline int Position::rule50_count() const {
+  return st->rule50;
 }
 
 inline bool Position::opposite_bishops() const {
-
-  return   pieceCount[WHITE][BISHOP] == 1
-        && pieceCount[BLACK][BISHOP] == 1
-        && opposite_colors(pieceList[WHITE][BISHOP][0], pieceList[BLACK][BISHOP][0]);
-}
-
-inline bool Position::bishop_pair(Color c) const {
-
-  return   pieceCount[c][BISHOP] >= 2
-        && opposite_colors(pieceList[c][BISHOP][0], pieceList[c][BISHOP][1]);
-}
-
-inline bool Position::pawn_on_7th(Color c) const {
-  return pieces(c, PAWN) & rank_bb(relative_rank(c, RANK_7));
+  return   count<BISHOP>(WHITE) == 1
+        && count<BISHOP>(BLACK) == 1
+        && opposite_colors(square<BISHOP>(WHITE), square<BISHOP>(BLACK));
 }
 
 inline bool Position::is_chess960() const {
   return chess960;
 }
 
-inline bool Position::is_capture_or_promotion(Move m) const {
-
+inline bool Position::capture(Move m) const {
   assert(is_ok(m));
-  return type_of(m) ? type_of(m) != CASTLE : !is_empty(to_sq(m));
+  return     (!empty(to_sq(m)) && type_of(m) != CASTLING)
+          ||  type_of(m) == EN_PASSANT;
 }
 
-inline bool Position::is_capture(Move m) const {
-
-  // Note that castle is coded as "king captures the rook"
+// returns true if a move is generated from the capture stage
+// having also queen promotions covered, i.e. consistency with the capture stage move generation
+// is needed to avoid the generation of duplicate moves.
+inline bool Position::capture_stage(Move m) const {
   assert(is_ok(m));
-  return (!is_empty(to_sq(m)) && type_of(m) != CASTLE) || type_of(m) == ENPASSANT;
+  return  capture(m) || promotion_type(m) == QUEEN;
 }
 
-inline PieceType Position::captured_piece_type() const {
-  return st->capturedType;
+inline Piece Position::captured_piece() const {
+  return st->capturedPiece;
 }
 
 inline Thread* Position::this_thread() const {
   return thisThread;
 }
 
-#endif // !defined(POSITION_H_INCLUDED)
+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;
+  pieceCount[pc]++;
+  pieceCount[make_piece(color_of(pc), ALL_PIECES)]++;
+  psq += PSQT::psq[pc][s];
+}
+
+inline void Position::remove_piece(Square s) {
+
+  Piece pc = board[s];
+  byTypeBB[ALL_PIECES] ^= s;
+  byTypeBB[type_of(pc)] ^= s;
+  byColorBB[color_of(pc)] ^= s;
+  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) {
+
+  Piece pc = board[from];
+  Bitboard fromTo = from | to;
+  byTypeBB[ALL_PIECES] ^= fromTo;
+  byTypeBB[type_of(pc)] ^= fromTo;
+  byColorBB[color_of(pc)] ^= fromTo;
+  board[from] = NO_PIECE;
+  board[to] = pc;
+  psq += PSQT::psq[pc][to] - PSQT::psq[pc][from];
+}
+
+inline void Position::do_move(Move m, StateInfo& newSt) {
+  do_move(m, newSt, gives_check(m));
+}
+
+inline StateInfo* Position::state() const {
+
+  return st;
+}
+
+} // namespace Stockfish
+
+#endif // #ifndef POSITION_H_INCLUDED

src/psqt.cpp

@@ -0,0 +1,131 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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/>.
+*/
+
+
+#include "psqt.h"
+
+#include <algorithm>
+
+#include "bitboard.h"
+#include "types.h"
+
+namespace Stockfish {
+
+namespace
+{
+
+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] = {
+  { },
+  { },
+  { // Knight
+   { S(-175, -96), S(-92,-65), S(-74,-49), S(-73,-21) },
+   { S( -77, -67), S(-41,-54), S(-27,-18), S(-15,  8) },
+   { S( -61, -40), S(-17,-27), S(  6, -8), S( 12, 29) },
+   { S( -35, -35), S(  8, -2), S( 40, 13), S( 49, 28) },
+   { S( -34, -45), S( 13,-16), S( 44,  9), S( 51, 39) },
+   { S(  -9, -51), S( 22,-44), S( 58,-16), S( 53, 17) },
+   { S( -67, -69), S(-27,-50), S(  4,-51), S( 37, 12) },
+   { S(-201,-100), S(-83,-88), S(-56,-56), S(-26,-17) }
+  },
+  { // Bishop
+   { 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) },
+   { S(-21,-12), S(-13, -9), S( -8, -1), S( 6, -2) },
+   { S(-25,  6), S(-11, -8), S( -1, -2), S( 3, -6) },
+   { S(-13, -6), S( -5,  1), S( -4, -9), S(-6,  7) },
+   { S(-27, -5), S(-15,  8), S( -4,  7), S( 3, -6) },
+   { S(-22,  6), S( -2,  1), S(  6, -7), S(12, 10) },
+   { S( -2,  4), S( 12,  5), S( 16, 20), S(18, -5) },
+   { S(-17, 18), S(-19,  0), S( -1, 19), S( 9, 13) }
+  },
+  { // Queen
+   { S( 3,-69), S(-5,-57), S(-5,-47), S( 4,-26) },
+   { 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,-11), S( 8,  1) },
+   { S(-5,-50), S( 6,-27), S(10,-24), S( 8, -8) },
+   { 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) },
+   { S(278, 53), S(303,100), S(234,133), S(179,135) },
+   { S(195, 88), S(258,130), S(169,169), S(120,175) },
+   { S(164,103), S(190,156), S(138,172), S( 98,172) },
+   { S(154, 96), S(179,166), S(105,199), S( 70,199) },
+   { S(123, 92), S(145,172), S( 81,184), S( 31,191) },
+   { S( 88, 47), S(120,121), S( 65,116), S( 33,131) },
+   { S( 59, 11), S( 89, 59), S( 45, 73), S( -1, 78) }
+  }
+};
+
+constexpr Score PBonus[RANK_NB][FILE_NB] =
+  { // Pawn (asymmetric distribution)
+   { },
+   { 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) }
+  };
+
+} // 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.
+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]);
+
+    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

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,29 +16,23 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(BOOK_H_INCLUDED)
-#define BOOK_H_INCLUDED
 
-#include <fstream>
-#include <string>
+#ifndef PSQT_H_INCLUDED
+#define PSQT_H_INCLUDED
 
-#include "position.h"
-#include "rkiss.h"
 
-class PolyglotBook : private std::ifstream {
-public:
-  PolyglotBook();
- ~PolyglotBook();
-  Move probe(const Position& pos, const std::string& fName, bool pickBest);
+#include "types.h"
 
-private:
-  template<typename T> PolyglotBook& operator>>(T& n);
 
-  bool open(const char* fName);
-  size_t find_first(uint64_t key);
+namespace Stockfish::PSQT
+{
 
-  RKISS RKiss;
-  std::string fileName;
-};
+extern Score psq[PIECE_NB][SQUARE_NB];
 
-#endif // !defined(BOOK_H_INCLUDED)
+// Fill psqt array from a set of internally linked parameters
+void init();
+
+} // namespace Stockfish::PSQT
+
+
+#endif // PSQT_H_INCLUDED

src/psqtab.h

@@ -1,98 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
-
-  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/>.
-*/
-
-#if !defined(PSQTAB_H_INCLUDED)
-#define PSQTAB_H_INCLUDED
-
-#include "types.h"
-
-#define S(mg, eg) make_score(mg, eg)
-
-
-/// PSQT[PieceType][Square] contains Piece-Square scores. For each piece type on
-/// a given square a (midgame, endgame) score pair is assigned. PSQT is defined
-/// for white side, for black side the tables are symmetric.
-
-static const Score PSQT[][64] = {
-  { },
-  { // Pawn
-   S(  0, 0), S( 0, 0), S( 0, 0), S( 0, 0), S(0,  0), S( 0, 0), S( 0, 0), S(  0, 0),
-   S(-28,-8), S(-6,-8), S( 4,-8), S(14,-8), S(14,-8), S( 4,-8), S(-6,-8), S(-28,-8),
-   S(-28,-8), S(-6,-8), S( 9,-8), S(36,-8), S(36,-8), S( 9,-8), S(-6,-8), S(-28,-8),
-   S(-28,-8), S(-6,-8), S(17,-8), S(58,-8), S(58,-8), S(17,-8), S(-6,-8), S(-28,-8),
-   S(-28,-8), S(-6,-8), S(17,-8), S(36,-8), S(36,-8), S(17,-8), S(-6,-8), S(-28,-8),
-   S(-28,-8), S(-6,-8), S( 9,-8), S(14,-8), S(14,-8), S( 9,-8), S(-6,-8), S(-28,-8),
-   S(-28,-8), S(-6,-8), S( 4,-8), S(14,-8), S(14,-8), S( 4,-8), S(-6,-8), S(-28,-8),
-   S(  0, 0), S( 0, 0), S( 0, 0), S( 0, 0), S(0,  0), S( 0, 0), S( 0, 0), S(  0, 0)
-  },
-  { // Knight
-   S(-135,-104), S(-107,-79), S(-80,-55), S(-67,-42), S(-67,-42), S(-80,-55), S(-107,-79), S(-135,-104),
-   S( -93, -79), S( -67,-55), S(-39,-30), S(-25,-17), S(-25,-17), S(-39,-30), S( -67,-55), S( -93, -79),
-   S( -53, -55), S( -25,-30), S(  1, -6), S( 13,  5), S( 13,  5), S(  1, -6), S( -25,-30), S( -53, -55),
-   S( -25, -42), S(   1,-17), S( 27,  5), S( 41, 18), S( 41, 18), S( 27,  5), S(   1,-17), S( -25, -42),
-   S( -11, -42), S(  13,-17), S( 41,  5), S( 55, 18), S( 55, 18), S( 41,  5), S(  13,-17), S( -11, -42),
-   S( -11, -55), S(  13,-30), S( 41, -6), S( 55,  5), S( 55,  5), S( 41, -6), S(  13,-30), S( -11, -55),
-   S( -53, -79), S( -25,-55), S(  1,-30), S( 13,-17), S( 13,-17), S(  1,-30), S( -25,-55), S( -53, -79),
-   S(-193,-104), S( -67,-79), S(-39,-55), S(-25,-42), S(-25,-42), S(-39,-55), S( -67,-79), S(-193,-104)
-  },
-  { // Bishop
-   S(-40,-59), S(-40,-42), S(-35,-35), S(-30,-26), S(-30,-26), S(-35,-35), S(-40,-42), S(-40,-59),
-   S(-17,-42), S(  0,-26), S( -4,-18), S(  0,-11), S(  0,-11), S( -4,-18), S(  0,-26), S(-17,-42),
-   S(-13,-35), S( -4,-18), S(  8,-11), S(  4, -4), S(  4, -4), S(  8,-11), S( -4,-18), S(-13,-35),
-   S( -8,-26), S(  0,-11), S(  4, -4), S( 17,  4), S( 17,  4), S(  4, -4), S(  0,-11), S( -8,-26),
-   S( -8,-26), S(  0,-11), S(  4, -4), S( 17,  4), S( 17,  4), S(  4, -4), S(  0,-11), S( -8,-26),
-   S(-13,-35), S( -4,-18), S(  8,-11), S(  4, -4), S(  4, -4), S(  8,-11), S( -4,-18), S(-13,-35),
-   S(-17,-42), S(  0,-26), S( -4,-18), S(  0,-11), S(  0,-11), S( -4,-18), S(  0,-26), S(-17,-42),
-   S(-17,-59), S(-17,-42), S(-13,-35), S( -8,-26), S( -8,-26), S(-13,-35), S(-17,-42), S(-17,-59)
-  },
-  { // Rook
-   S(-12, 3), S(-7, 3), S(-2, 3), S(2, 3), S(2, 3), S(-2, 3), S(-7, 3), S(-12, 3),
-   S(-12, 3), S(-7, 3), S(-2, 3), S(2, 3), S(2, 3), S(-2, 3), S(-7, 3), S(-12, 3),
-   S(-12, 3), S(-7, 3), S(-2, 3), S(2, 3), S(2, 3), S(-2, 3), S(-7, 3), S(-12, 3),
-   S(-12, 3), S(-7, 3), S(-2, 3), S(2, 3), S(2, 3), S(-2, 3), S(-7, 3), S(-12, 3),
-   S(-12, 3), S(-7, 3), S(-2, 3), S(2, 3), S(2, 3), S(-2, 3), S(-7, 3), S(-12, 3),
-   S(-12, 3), S(-7, 3), S(-2, 3), S(2, 3), S(2, 3), S(-2, 3), S(-7, 3), S(-12, 3),
-   S(-12, 3), S(-7, 3), S(-2, 3), S(2, 3), S(2, 3), S(-2, 3), S(-7, 3), S(-12, 3),
-   S(-12, 3), S(-7, 3), S(-2, 3), S(2, 3), S(2, 3), S(-2, 3), S(-7, 3), S(-12, 3)
-  },
-  { // Queen
-   S(8,-80), S(8,-54), S(8,-42), S(8,-30), S(8,-30), S(8,-42), S(8,-54), S(8,-80),
-   S(8,-54), S(8,-30), S(8,-18), S(8, -6), S(8, -6), S(8,-18), S(8,-30), S(8,-54),
-   S(8,-42), S(8,-18), S(8, -6), S(8,  6), S(8,  6), S(8, -6), S(8,-18), S(8,-42),
-   S(8,-30), S(8, -6), S(8,  6), S(8, 18), S(8, 18), S(8,  6), S(8, -6), S(8,-30),
-   S(8,-30), S(8, -6), S(8,  6), S(8, 18), S(8, 18), S(8,  6), S(8, -6), S(8,-30),
-   S(8,-42), S(8,-18), S(8, -6), S(8,  6), S(8,  6), S(8, -6), S(8,-18), S(8,-42),
-   S(8,-54), S(8,-30), S(8,-18), S(8, -6), S(8, -6), S(8,-18), S(8,-30), S(8,-54),
-   S(8,-80), S(8,-54), S(8,-42), S(8,-30), S(8,-30), S(8,-42), S(8,-54), S(8,-80)
-  },
-  { // King
-   S(287, 18), S(311, 77), S(262,105), S(214,135), S(214,135), S(262,105), S(311, 77), S(287, 18),
-   S(262, 77), S(287,135), S(238,165), S(190,193), S(190,193), S(238,165), S(287,135), S(262, 77),
-   S(214,105), S(238,165), S(190,193), S(142,222), S(142,222), S(190,193), S(238,165), S(214,105),
-   S(190,135), S(214,193), S(167,222), S(119,251), S(119,251), S(167,222), S(214,193), S(190,135),
-   S(167,135), S(190,193), S(142,222), S( 94,251), S( 94,251), S(142,222), S(190,193), S(167,135),
-   S(142,105), S(167,165), S(119,193), S( 69,222), S( 69,222), S(119,193), S(167,165), S(142,105),
-   S(119, 77), S(142,135), S( 94,165), S( 46,193), S( 46,193), S( 94,165), S(142,135), S(119, 77),
-   S(94,  18), S(119, 77), S( 69,105), S( 21,135), S( 21,135), S( 69,105), S(119, 77), S( 94, 18)
-  }
-};
-
-#undef S
-
-#endif // !defined(PSQTAB_H_INCLUDED)

src/rkiss.h

@@ -1,78 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
-
-  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/>.
-
-  This file is based on original code by Heinz van Saanen and is
-  available under 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.
-*/
-
-#if !defined(RKISS_H_INCLUDED)
-#define RKISS_H_INCLUDED
-
-#include "types.h"
-
-/// RKISS is our pseudo random number generator (PRNG) used to compute hash keys.
-/// George Marsaglia invented the RNG-Kiss-family in the early 90's. This is a
-/// specific version that Heinz van Saanen derived from some public domain code
-/// by Bob Jenkins. Following the feature list, as tested by Heinz.
-///
-/// - Quite platform independent
-/// - Passes ALL dieharder tests! Here *nix sys-rand() e.g. fails miserably:-)
-/// - ~12 times faster than my *nix sys-rand()
-/// - ~4 times faster than SSE2-version of Mersenne twister
-/// - Average cycle length: ~2^126
-/// - 64 bit seed
-/// - Return doubles with a full 53 bit mantissa
-/// - Thread safe
-
-class RKISS {
-
-  // Keep variables always together
-  struct S { uint64_t a, b, c, d; } s;
-
-  uint64_t rotate(uint64_t x, uint64_t k) const {
-    return (x << k) | (x >> (64 - k));
-  }
-
-  // Return 64 bit unsigned integer in between [0, 2^64 - 1]
-  uint64_t rand64() {
-
-    const uint64_t
-      e = s.a - rotate(s.b,  7);
-    s.a = s.b ^ rotate(s.c, 13);
-    s.b = s.c + rotate(s.d, 37);
-    s.c = s.d + e;
-    return s.d = e + s.a;
-  }
-
-  // Init seed and scramble a few rounds
-  void raninit() {
-
-    s.a = 0xf1ea5eed;
-    s.b = s.c = s.d = 0xd4e12c77;
-    for (int i = 0; i < 73; i++)
-        rand64();
-  }
-
-public:
-  RKISS() { raninit(); }
-  template<typename T> T rand() { return T(rand64()); }
-};
-
-#endif // !defined(RKISS_H_INCLUDED)

src/search.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,95 +16,101 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(SEARCH_H_INCLUDED)
+#ifndef SEARCH_H_INCLUDED
 #define SEARCH_H_INCLUDED
 
-#include <cstring>
-#include <memory>
-#include <stack>
 #include <vector>
 
 #include "misc.h"
-#include "position.h"
+#include "movepick.h"
 #include "types.h"
 
-struct SplitPoint;
+namespace Stockfish {
+
+class Position;
 
 namespace Search {
 
-/// The Stack struct keeps track of the information we need to remember from
-/// nodes shallower and deeper in the tree during the search. Each search thread
-/// has its own array of Stack objects, indexed by the current ply.
+
+/// Stack struct keeps track of the information we need to remember from nodes
+/// shallower and deeper in the tree during the search. Each search thread has
+/// its own array of Stack objects, indexed by the current ply.
 
 struct Stack {
-  SplitPoint* sp;
+  Move* pv;
+  PieceToHistory* continuationHistory;
   int ply;
   Move currentMove;
   Move excludedMove;
   Move killers[2];
-  Depth reduction;
-  Value eval;
-  Value evalMargin;
-  int skipNullMove;
+  Value staticEval;
+  int statScore;
+  int moveCount;
+  bool inCheck;
+  bool ttPv;
+  bool ttHit;
+  int doubleExtensions;
+  int cutoffCnt;
 };
 
 
-/// RootMove struct is used for moves at the root of the tree. For each root
-/// move we store a score, a node count, and a PV (really a refutation in the
-/// case of moves which fail low). Score is normally set at -VALUE_INFINITE for
-/// all non-pv moves.
+/// RootMove struct is used for moves at the root of the tree. For each root move
+/// we store a score and a PV (really a refutation in the case of moves which
+/// fail low). Score is normally set at -VALUE_INFINITE for all non-pv moves.
+
 struct RootMove {
 
-  RootMove(){} // Needed by sort()
-  RootMove(Move m) : score(-VALUE_INFINITE), prevScore(-VALUE_INFINITE) {
-    pv.push_back(m); pv.push_back(MOVE_NONE);
+  explicit RootMove(Move m) : pv(1, m) {}
+  bool extract_ponder_from_tt(Position& pos);
+  bool operator==(const Move& m) const { return pv[0] == m; }
+  bool operator<(const RootMove& m) const { // Sort in descending order
+    return m.score != score ? m.score < score
+                            : m.previousScore < previousScore;
   }
 
-  bool operator<(const RootMove& m) const { return score < m.score; }
-  bool operator==(const Move& m) const { return pv[0] == m; }
-
-  void extract_pv_from_tt(Position& pos);
-  void insert_pv_in_tt(Position& pos);
-
-  Value score;
-  Value prevScore;
+  Value score = -VALUE_INFINITE;
+  Value previousScore = -VALUE_INFINITE;
+  Value averageScore = -VALUE_INFINITE;
+  Value uciScore = -VALUE_INFINITE;
+  bool scoreLowerbound = false;
+  bool scoreUpperbound = false;
+  int selDepth = 0;
+  int tbRank = 0;
+  Value tbScore;
   std::vector<Move> pv;
 };
 
+using RootMoves = std::vector<RootMove>;
 
-/// The LimitsType struct stores information sent by GUI about available time
-/// to search the current move, maximum depth/time, if we are in analysis mode
-/// or if we have to ponder while is our opponent's side to move.
+
+/// LimitsType struct stores information sent by GUI about available time to
+/// search the current move, maximum depth/time, or if we are in analysis mode.
 
 struct LimitsType {
 
-  LimitsType() { memset(this, 0, sizeof(LimitsType)); }
-  bool use_time_management() const { return !(movetime | depth | nodes | infinite); }
+  LimitsType() { // Init explicitly due to broken value-initialization of non POD in MSVC
+    time[WHITE] = time[BLACK] = inc[WHITE] = inc[BLACK] = npmsec = movetime = TimePoint(0);
+    movestogo = depth = mate = perft = infinite = 0;
+    nodes = 0;
+  }
 
-  int time[2], inc[2], movestogo, depth, nodes, movetime, infinite, ponder;
+  bool use_time_management() const {
+    return time[WHITE] || time[BLACK];
+  }
+
+  std::vector<Move> searchmoves;
+  TimePoint time[COLOR_NB], inc[COLOR_NB], npmsec, movetime, startTime;
+  int movestogo, depth, mate, perft, infinite;
+  int64_t nodes;
 };
 
-
-/// The SignalsType struct stores volatile flags updated during the search
-/// typically in an async fashion, for instance to stop the search by the GUI.
-
-struct SignalsType {
-  bool stopOnPonderhit, firstRootMove, stop, failedLowAtRoot;
-};
-
-typedef std::auto_ptr<std::stack<StateInfo> > StateStackPtr;
-
-extern volatile SignalsType Signals;
 extern LimitsType Limits;
-extern std::vector<RootMove> RootMoves;
-extern Position RootPosition;
-extern Time::point SearchTime;
-extern StateStackPtr SetupStates;
 
-extern void init();
-extern size_t perft(Position& pos, Depth depth);
-extern void think();
+void init();
+void clear();
 
 } // namespace Search
 
-#endif // !defined(SEARCH_H_INCLUDED)
+} // namespace Stockfish
+
+#endif // #ifndef SEARCH_H_INCLUDED

src/syzygy/tbprobe.h

@@ -0,0 +1,76 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+
+  Stockfish is free software: you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation, either version 3 of the License, or
+  (at your option) any later version.
+
+  Stockfish is distributed in the hope that it will be useful,
+  but WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+  GNU General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef TBPROBE_H
+#define TBPROBE_H
+
+#include <ostream>
+
+#include "../search.h"
+
+namespace Stockfish::Tablebases {
+
+enum WDLScore {
+    WDLLoss        = -2, // Loss
+    WDLBlessedLoss = -1, // Loss, but draw under 50-move rule
+    WDLDraw        =  0, // Draw
+    WDLCursedWin   =  1, // Win, but draw under 50-move rule
+    WDLWin         =  2, // Win
+};
+
+// Possible states after a probing operation
+enum ProbeState {
+    FAIL              =  0, // Probe failed (missing file table)
+    OK                =  1, // Probe successful
+    CHANGE_STM        = -1, // DTZ should check the other side
+    ZEROING_BEST_MOVE =  2  // Best move zeroes DTZ (capture or pawn move)
+};
+
+extern int MaxCardinality;
+
+void init(const std::string& paths);
+WDLScore probe_wdl(Position& pos, ProbeState* result);
+int probe_dtz(Position& pos, ProbeState* result);
+bool root_probe(Position& pos, Search::RootMoves& rootMoves);
+bool root_probe_wdl(Position& pos, Search::RootMoves& rootMoves);
+void rank_root_moves(Position& pos, Search::RootMoves& rootMoves);
+
+inline std::ostream& operator<<(std::ostream& os, const WDLScore v) {
+
+    os << (v == WDLLoss        ? "Loss" :
+           v == WDLBlessedLoss ? "Blessed loss" :
+           v == WDLDraw        ? "Draw" :
+           v == WDLCursedWin   ? "Cursed win" :
+           v == WDLWin         ? "Win" : "None");
+
+    return os;
+}
+
+inline std::ostream& operator<<(std::ostream& os, const ProbeState v) {
+
+    os << (v == FAIL              ? "Failed" :
+           v == OK                ? "Success" :
+           v == CHANGE_STM        ? "Probed opponent side" :
+           v == ZEROING_BEST_MOVE ? "Best move zeroes DTZ" : "None");
+
+    return os;
+}
+
+} // namespace Stockfish::Tablebases
+
+#endif

src/thread.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -18,419 +17,252 @@
 */
 
 #include <cassert>
-#include <iostream>
 
+#include <algorithm> // For std::count
 #include "movegen.h"
 #include "search.h"
 #include "thread.h"
-#include "ucioption.h"
+#include "uci.h"
+#include "syzygy/tbprobe.h"
+#include "tt.h"
 
-using namespace Search;
+namespace Stockfish {
 
 ThreadPool Threads; // Global object
 
-namespace { extern "C" {
 
- // start_routine() is the C function which is called when a new thread
- // is launched. It is a wrapper to member function pointed by start_fn.
+/// Thread constructor launches the thread and waits until it goes to sleep
+/// in idle_loop(). Note that 'searching' and 'exit' should be already set.
 
- long start_routine(Thread* th) { (th->*(th->start_fn))(); return 0; }
+Thread::Thread(size_t n) : idx(n), stdThread(&Thread::idle_loop, this) {
 
-} }
-
-
-// Thread c'tor starts a newly-created thread of execution that will call
-// the idle loop function pointed by start_fn going immediately to sleep.
-
-Thread::Thread(Fn fn) {
-
-  is_searching = do_exit = false;
-  maxPly = splitPointsCnt = 0;
-  curSplitPoint = NULL;
-  start_fn = fn;
-  idx = Threads.size();
-
-  do_sleep = (fn != &Thread::main_loop); // Avoid a race with start_searching()
-
-  if (!thread_create(handle, start_routine, this))
-  {
-      std::cerr << "Failed to create thread number " << idx << std::endl;
-      ::exit(EXIT_FAILURE);
-  }
+  wait_for_search_finished();
 }
 
 
-// Thread d'tor waits for thread termination before to return.
+/// Thread destructor wakes up the thread in idle_loop() and waits
+/// for its termination. Thread should be already waiting.
 
 Thread::~Thread() {
 
-  assert(do_sleep);
+  assert(!searching);
 
-  do_exit = true; // Search must be already finished
-  wake_up();
-  thread_join(handle); // Wait for thread termination
+  exit = true;
+  start_searching();
+  stdThread.join();
 }
 
 
-// Thread::timer_loop() is where the timer thread waits maxPly milliseconds and
-// then calls check_time(). If maxPly is 0 thread sleeps until is woken up.
-extern void check_time();
+/// Thread::clear() reset histories, usually before a new game
 
-void Thread::timer_loop() {
+void Thread::clear() {
 
-  while (!do_exit)
-  {
-      mutex.lock();
-      sleepCondition.wait_for(mutex, maxPly ? maxPly : INT_MAX);
-      mutex.unlock();
-      check_time();
-  }
+  counterMoves.fill(MOVE_NONE);
+  mainHistory.fill(0);
+  captureHistory.fill(0);
+
+  for (bool inCheck : { false, true })
+      for (StatsType c : { NoCaptures, Captures })
+          for (auto& to : continuationHistory[inCheck][c])
+              for (auto& h : to)
+                  h->fill(-71);
 }
 
 
-// Thread::main_loop() is where the main thread is parked waiting to be started
-// when there is a new search. Main thread will launch all the slave threads.
+/// Thread::start_searching() wakes up the thread that will start the search
 
-void Thread::main_loop() {
+void Thread::start_searching() {
+  mutex.lock();
+  searching = true;
+  mutex.unlock(); // Unlock before notifying saves a few CPU-cycles
+  cv.notify_one(); // Wake up the thread in idle_loop()
+}
+
+
+/// Thread::wait_for_search_finished() blocks on the condition variable
+/// until the thread has finished searching.
+
+void Thread::wait_for_search_finished() {
+
+  std::unique_lock<std::mutex> lk(mutex);
+  cv.wait(lk, [&]{ return !searching; });
+}
+
+
+/// Thread::idle_loop() is where the thread is parked, blocked on the
+/// condition variable, when it has no work to do.
+
+void Thread::idle_loop() {
+
+  // If OS already scheduled us on a different group than 0 then don't overwrite
+  // the choice, eventually we are one of many one-threaded processes running on
+  // some Windows NUMA hardware, for instance in fishtest. To make it simple,
+  // just check if running threads are below a threshold, in this case all this
+  // NUMA machinery is not needed.
+  if (Options["Threads"] > 8)
+      WinProcGroup::bindThisThread(idx);
 
   while (true)
   {
-      mutex.lock();
+      std::unique_lock<std::mutex> lk(mutex);
+      searching = false;
+      cv.notify_one(); // Wake up anyone waiting for search finished
+      cv.wait(lk, [&]{ return searching; });
 
-      do_sleep = true; // Always return to sleep after a search
-      is_searching = false;
-
-      while (do_sleep && !do_exit)
-      {
-          Threads.sleepCondition.notify_one(); // Wake up UI thread if needed
-          sleepCondition.wait(mutex);
-      }
-
-      mutex.unlock();
-
-      if (do_exit)
+      if (exit)
           return;
 
-      is_searching = true;
+      lk.unlock();
 
-      Search::think();
-
-      assert(is_searching);
+      search();
   }
 }
 
+/// ThreadPool::set() creates/destroys threads to match the requested number.
+/// Created and launched threads will immediately go to sleep in idle_loop.
+/// Upon resizing, threads are recreated to allow for binding if necessary.
 
-// Thread::wake_up() wakes up the thread, normally at the beginning of the search
-// or, if "sleeping threads" is used at split time.
+void ThreadPool::set(size_t requested) {
 
-void Thread::wake_up() {
-
-  mutex.lock();
-  sleepCondition.notify_one();
-  mutex.unlock();
-}
-
-
-// Thread::wait_for_stop_or_ponderhit() is called when the maximum depth is
-// reached while the program is pondering. The point is to work around a wrinkle
-// in the UCI protocol: When pondering, the engine is not allowed to give a
-// "bestmove" before the GUI sends it a "stop" or "ponderhit" command. We simply
-// wait here until one of these commands (that raise StopRequest) is sent and
-// then return, after which the bestmove and pondermove will be printed.
-
-void Thread::wait_for_stop_or_ponderhit() {
-
-  Signals.stopOnPonderhit = true;
-
-  mutex.lock();
-  while (!Signals.stop) sleepCondition.wait(mutex);;
-  mutex.unlock();
-}
-
-
-// Thread::cutoff_occurred() checks whether a beta cutoff has occurred in the
-// current active split point, or in some ancestor of the split point.
-
-bool Thread::cutoff_occurred() const {
-
-  for (SplitPoint* sp = curSplitPoint; sp; sp = sp->parent)
-      if (sp->cutoff)
-          return true;
-
-  return false;
-}
-
-
-// Thread::is_available_to() checks whether the thread is available to help the
-// thread 'master' at a split point. An obvious requirement is that thread must
-// be idle. With more than two threads, this is not sufficient: If the thread is
-// the master of some active split point, it is only available as a slave to the
-// slaves which are busy searching the split point at the top of slaves split
-// point stack (the "helpful master concept" in YBWC terminology).
-
-bool Thread::is_available_to(Thread* master) const {
-
-  if (is_searching)
-      return false;
-
-  // Make a local copy to be sure doesn't become zero under our feet while
-  // testing next condition and so leading to an out of bound access.
-  int spCnt = splitPointsCnt;
-
-  // No active split points means that the thread is available as a slave for any
-  // other thread otherwise apply the "helpful master" concept if possible.
-  return !spCnt || (splitPoints[spCnt - 1].slavesMask & (1ULL << master->idx));
-}
-
-
-// init() is called at startup. Initializes lock and condition variable and
-// launches requested threads sending them immediately to sleep. We cannot use
-// a c'tor becuase Threads is a static object and we need a fully initialized
-// engine at this point due to allocation of endgames in Thread c'tor.
-
-void ThreadPool::init() {
-
-  timer = new Thread(&Thread::timer_loop);
-  threads.push_back(new Thread(&Thread::main_loop));
-  read_uci_options();
-}
-
-
-// exit() cleanly terminates the threads before the program exits.
-
-void ThreadPool::exit() {
-
-  for (size_t i = 0; i < threads.size(); i++)
-      delete threads[i];
-
-  delete timer;
-}
-
-
-// read_uci_options() updates internal threads parameters from the corresponding
-// UCI options and creates/destroys threads to match the requested number. Thread
-// objects are dynamically allocated to avoid creating in advance all possible
-// threads, with included pawns and material tables, if only few are used.
-
-void ThreadPool::read_uci_options() {
-
-  maxThreadsPerSplitPoint = Options["Max Threads per Split Point"];
-  minimumSplitDepth       = Options["Min Split Depth"] * ONE_PLY;
-  useSleepingThreads      = Options["Use Sleeping Threads"];
-  size_t requested        = Options["Threads"];
-
-  assert(requested > 0);
-
-  while (threads.size() < requested)
-      threads.push_back(new Thread(&Thread::idle_loop));
-
-  while (threads.size() > requested)
+  if (threads.size() > 0)   // destroy any existing thread(s)
   {
-      delete threads.back();
-      threads.pop_back();
+      main()->wait_for_search_finished();
+
+      while (threads.size() > 0)
+          delete threads.back(), threads.pop_back();
   }
-}
 
-
-// wake_up() is called before a new search to start the threads that are waiting
-// on the sleep condition and to reset maxPly. When useSleepingThreads is set
-// threads will be woken up at split time.
-
-void ThreadPool::wake_up() const {
-
-  for (size_t i = 0; i < threads.size(); i++)
+  if (requested > 0)   // create new thread(s)
   {
-      threads[i]->maxPly = 0;
-      threads[i]->do_sleep = false;
+      threads.push_back(new MainThread(0));
 
-      if (!useSleepingThreads)
-          threads[i]->wake_up();
+      while (threads.size() < requested)
+          threads.push_back(new Thread(threads.size()));
+      clear();
+
+      // Reallocate the hash with the new threadpool size
+      TT.resize(size_t(Options["Hash"]));
+
+      // Init thread number dependent search params.
+      Search::init();
   }
 }
 
 
-// sleep() is called after the search finishes to ask all the threads but the
-// main one to go waiting on a sleep condition.
+/// ThreadPool::clear() sets threadPool data to initial values
 
-void ThreadPool::sleep() const {
+void ThreadPool::clear() {
 
-  // Main thread will go to sleep by itself to avoid a race with start_searching()
-  for (size_t i = 1; i < threads.size(); i++)
-      threads[i]->do_sleep = true;
+  for (Thread* th : threads)
+      th->clear();
+
+  main()->callsCnt = 0;
+  main()->bestPreviousScore = VALUE_INFINITE;
+  main()->bestPreviousAverageScore = VALUE_INFINITE;
+  main()->previousTimeReduction = 1.0;
 }
 
 
-// available_slave_exists() tries to find an idle thread which is available as
-// a slave for the thread 'master'.
+/// ThreadPool::start_thinking() wakes up main thread waiting in idle_loop() and
+/// returns immediately. Main thread will wake up other threads and start the search.
 
-bool ThreadPool::available_slave_exists(Thread* master) const {
+void ThreadPool::start_thinking(Position& pos, StateListPtr& states,
+                                const Search::LimitsType& limits, bool ponderMode) {
 
-  for (size_t i = 0; i < threads.size(); i++)
-      if (threads[i]->is_available_to(master))
-          return true;
+  main()->wait_for_search_finished();
 
-  return false;
-}
+  main()->stopOnPonderhit = stop = false;
+  increaseDepth = true;
+  main()->ponder = ponderMode;
+  Search::Limits = limits;
+  Search::RootMoves rootMoves;
 
+  for (const auto& m : MoveList<LEGAL>(pos))
+      if (   limits.searchmoves.empty()
+          || std::count(limits.searchmoves.begin(), limits.searchmoves.end(), m))
+          rootMoves.emplace_back(m);
 
-// split() does the actual work of distributing the work at a node between
-// several available threads. If it does not succeed in splitting the node
-// (because no idle threads are available, or because we have no unused split
-// point objects), the function immediately returns. If splitting is possible, a
-// SplitPoint object is initialized with all the data that must be copied to the
-// helper threads and then helper threads are told that they have been assigned
-// work. This will cause them to instantly leave their idle loops and call
-// search(). When all threads have returned from search() then split() returns.
+  if (!rootMoves.empty())
+      Tablebases::rank_root_moves(pos, rootMoves);
 
-template <bool Fake>
-Value ThreadPool::split(Position& pos, Stack* ss, Value alpha, Value beta,
-                        Value bestValue, Move* bestMove, Depth depth,
-                        Move threatMove, int moveCount, MovePicker* mp, int nodeType) {
+  // After ownership transfer 'states' becomes empty, so if we stop the search
+  // and call 'go' again without setting a new position states.get() == nullptr.
+  assert(states.get() || setupStates.get());
 
-  assert(pos.pos_is_ok());
-  assert(bestValue > -VALUE_INFINITE);
-  assert(bestValue <= alpha);
-  assert(alpha < beta);
-  assert(beta <= VALUE_INFINITE);
-  assert(depth > DEPTH_ZERO);
+  if (states.get())
+      setupStates = std::move(states); // Ownership transfer, states is now empty
 
-  Thread* master = pos.this_thread();
-
-  if (master->splitPointsCnt >= MAX_SPLITPOINTS_PER_THREAD)
-      return bestValue;
-
-  // Pick the next available split point from the split point stack
-  SplitPoint& sp = master->splitPoints[master->splitPointsCnt];
-
-  sp.parent = master->curSplitPoint;
-  sp.master = master;
-  sp.cutoff = false;
-  sp.slavesMask = 1ULL << master->idx;
-  sp.depth = depth;
-  sp.bestMove = *bestMove;
-  sp.threatMove = threatMove;
-  sp.alpha = alpha;
-  sp.beta = beta;
-  sp.nodeType = nodeType;
-  sp.bestValue = bestValue;
-  sp.mp = mp;
-  sp.moveCount = moveCount;
-  sp.pos = &pos;
-  sp.nodes = 0;
-  sp.ss = ss;
-
-  assert(master->is_searching);
-
-  master->curSplitPoint = &sp;
-  int slavesCnt = 0;
-
-  // Try to allocate available threads and ask them to start searching setting
-  // is_searching flag. This must be done under lock protection to avoid concurrent
-  // allocation of the same slave by another master.
-  sp.mutex.lock();
-  mutex.lock();
-
-  for (size_t i = 0; i < threads.size() && !Fake; ++i)
-      if (threads[i]->is_available_to(master))
-      {
-          sp.slavesMask |= 1ULL << i;
-          threads[i]->curSplitPoint = &sp;
-          threads[i]->is_searching = true; // Slave leaves idle_loop()
-
-          if (useSleepingThreads)
-              threads[i]->wake_up();
-
-          if (++slavesCnt + 1 >= maxThreadsPerSplitPoint) // Master is always included
-              break;
-      }
-
-  master->splitPointsCnt++;
-
-  mutex.unlock();
-  sp.mutex.unlock();
-
-  // Everything is set up. The master thread enters the idle loop, from which
-  // it will instantly launch a search, because its is_searching flag is set.
-  // The thread will return from the idle loop when all slaves have finished
-  // their work at this split point.
-  if (slavesCnt || Fake)
+  // We use Position::set() to set root position across threads. But there are
+  // some StateInfo fields (previous, pliesFromNull, capturedPiece) that cannot
+  // be deduced from a fen string, so set() clears them and they are set from
+  // setupStates->back() later. The rootState is per thread, earlier states are shared
+  // since they are read-only.
+  for (Thread* th : threads)
   {
-      master->idle_loop();
-
-      // In helpful master concept a master can help only a sub-tree of its split
-      // point, and because here is all finished is not possible master is booked.
-      assert(!master->is_searching);
+      th->nodes = th->tbHits = th->nmpMinPly = th->bestMoveChanges = 0;
+      th->rootDepth = th->completedDepth = 0;
+      th->rootMoves = rootMoves;
+      th->rootPos.set(pos.fen(), pos.is_chess960(), &th->rootState, th);
+      th->rootState = setupStates->back();
   }
 
-  // We have returned from the idle loop, which means that all threads are
-  // finished. Note that setting is_searching and decreasing splitPointsCnt is
-  // done under lock protection to avoid a race with Thread::is_available_to().
-  sp.mutex.lock(); // To protect sp.nodes
-  mutex.lock();
-
-  master->is_searching = true;
-  master->splitPointsCnt--;
-  master->curSplitPoint = sp.parent;
-  pos.set_nodes_searched(pos.nodes_searched() + sp.nodes);
-  *bestMove = sp.bestMove;
-
-  mutex.unlock();
-  sp.mutex.unlock();
-
-  return sp.bestValue;
+  main()->start_searching();
 }
 
-// Explicit template instantiations
-template Value ThreadPool::split<false>(Position&, Stack*, Value, Value, Value, Move*, Depth, Move, int, MovePicker*, int);
-template Value ThreadPool::split<true>(Position&, Stack*, Value, Value, Value, Move*, Depth, Move, int, MovePicker*, int);
+Thread* ThreadPool::get_best_thread() const {
 
+    Thread* bestThread = threads.front();
+    std::map<Move, int64_t> votes;
+    Value minScore = VALUE_NONE;
 
-// set_timer() is used to set the timer to trigger after msec milliseconds.
-// If msec is 0 then timer is stopped.
+    // Find minimum score of all threads
+    for (Thread* th: threads)
+        minScore = std::min(minScore, th->rootMoves[0].score);
 
-void ThreadPool::set_timer(int msec) {
+    // Vote according to score and depth, and select the best thread
+    auto thread_value = [minScore](Thread* th) {
+            return (th->rootMoves[0].score - minScore + 14) * int(th->completedDepth);
+        };
 
-  timer->mutex.lock();
-  timer->maxPly = msec;
-  timer->sleepCondition.notify_one(); // Wake up and restart the timer
-  timer->mutex.unlock();
+    for (Thread* th : threads)
+        votes[th->rootMoves[0].pv[0]] += thread_value(th);
+
+    for (Thread* th : threads)
+        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]]
+                         || (   votes[th->rootMoves[0].pv[0]] == votes[bestThread->rootMoves[0].pv[0]]
+                             &&   thread_value(th) * int(th->rootMoves[0].pv.size() > 2)
+                                > thread_value(bestThread) * int(bestThread->rootMoves[0].pv.size() > 2)))))
+            bestThread = th;
+
+    return bestThread;
 }
 
 
-// wait_for_search_finished() waits for main thread to go to sleep, this means
-// search is finished. Then returns.
+/// Start non-main threads
 
-void ThreadPool::wait_for_search_finished() {
+void ThreadPool::start_searching() {
 
-  Thread* t = main_thread();
-  t->mutex.lock();
-  t->sleepCondition.notify_one(); // In case is waiting for stop or ponderhit
-  while (!t->do_sleep) sleepCondition.wait(t->mutex);
-  t->mutex.unlock();
+    for (Thread* th : threads)
+        if (th != threads.front())
+            th->start_searching();
 }
 
 
-// start_searching() wakes up the main thread sleeping in  main_loop() so to start
-// a new search, then returns immediately.
+/// Wait for non-main threads
 
-void ThreadPool::start_searching(const Position& pos, const LimitsType& limits,
-                                 const std::vector<Move>& searchMoves, StateStackPtr& states) {
-  wait_for_search_finished();
+void ThreadPool::wait_for_search_finished() const {
 
-  SearchTime = Time::now(); // As early as possible
-
-  Signals.stopOnPonderhit = Signals.firstRootMove = false;
-  Signals.stop = Signals.failedLowAtRoot = false;
-
-  RootPosition = pos;
-  Limits = limits;
-  SetupStates = states; // Ownership transfer here
-  RootMoves.clear();
-
-  for (MoveList<LEGAL> ml(pos); !ml.end(); ++ml)
-      if (searchMoves.empty() || count(searchMoves.begin(), searchMoves.end(), ml.move()))
-          RootMoves.push_back(RootMove(ml.move()));
-
-  main_thread()->do_sleep = false;
-  main_thread()->wake_up();
+    for (Thread* th : threads)
+        if (th != threads.front())
+            th->wait_for_search_finished();
 }
+
+} // namespace Stockfish

src/thread.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,9 +16,13 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(THREAD_H_INCLUDED)
+#ifndef THREAD_H_INCLUDED
 #define THREAD_H_INCLUDED
 
+#include <atomic>
+#include <condition_variable>
+#include <mutex>
+#include <thread>
 #include <vector>
 
 #include "material.h"
@@ -27,142 +30,112 @@
 #include "pawns.h"
 #include "position.h"
 #include "search.h"
+#include "thread_win32_osx.h"
 
-const int MAX_THREADS = 32;
-const int MAX_SPLITPOINTS_PER_THREAD = 8;
+namespace Stockfish {
 
-struct Mutex {
-  Mutex() { lock_init(l); }
- ~Mutex() { lock_destroy(l); }
-
-  void lock() { lock_grab(l); }
-  void unlock() { lock_release(l); }
-
-private:
-  friend struct ConditionVariable;
-
-  Lock l;
-};
-
-struct ConditionVariable {
-  ConditionVariable() { cond_init(c); }
- ~ConditionVariable() { cond_destroy(c); }
-
-  void wait(Mutex& m) { cond_wait(c, m.l); }
-  void wait_for(Mutex& m, int ms) { timed_wait(c, m.l, ms); }
-  void notify_one() { cond_signal(c); }
-
-private:
-  WaitCondition c;
-};
-
-class Thread;
-
-struct SplitPoint {
-
-  // Const data after split point has been setup
-  const Position* pos;
-  const Search::Stack* ss;
-  Depth depth;
-  Value beta;
-  int nodeType;
-  Thread* master;
-  Move threatMove;
-
-  // Const pointers to shared data
-  MovePicker* mp;
-  SplitPoint* parent;
-
-  // Shared data
-  Mutex mutex;
-  volatile uint64_t slavesMask;
-  volatile int64_t nodes;
-  volatile Value alpha;
-  volatile Value bestValue;
-  volatile Move bestMove;
-  volatile int moveCount;
-  volatile bool cutoff;
-};
-
-
-/// Thread struct keeps together all the thread related stuff like locks, state
-/// and especially split points. We also use per-thread pawn and material hash
-/// tables so that once we get a pointer to an entry its life time is unlimited
-/// and we don't have to care about someone changing the entry under our feet.
+/// Thread class keeps together all the thread-related stuff. We use
+/// per-thread pawn and material hash tables so that once we get a
+/// pointer to an entry its life time is unlimited and we don't have
+/// to care about someone changing the entry under our feet.
 
 class Thread {
 
-  typedef void (Thread::* Fn) (); // Pointer to member function
+  std::mutex mutex;
+  std::condition_variable cv;
+  size_t idx;
+  bool exit = false, searching = true; // Set before starting std::thread
+  NativeThread stdThread;
 
 public:
-  Thread(Fn fn);
- ~Thread();
-
-  void wake_up();
-  bool cutoff_occurred() const;
-  bool is_available_to(Thread* master) const;
+  explicit Thread(size_t);
+  virtual ~Thread();
+  virtual void search();
+  void clear();
   void idle_loop();
-  void main_loop();
-  void timer_loop();
-  void wait_for_stop_or_ponderhit();
+  void start_searching();
+  void wait_for_search_finished();
+  size_t id() const { return idx; }
 
-  SplitPoint splitPoints[MAX_SPLITPOINTS_PER_THREAD];
-  MaterialTable materialTable;
-  PawnTable pawnTable;
-  size_t idx;
-  int maxPly;
-  Mutex mutex;
-  ConditionVariable sleepCondition;
-  NativeHandle handle;
-  Fn start_fn;
-  SplitPoint* volatile curSplitPoint;
-  volatile int splitPointsCnt;
-  volatile bool is_searching;
-  volatile bool do_sleep;
-  volatile bool do_exit;
+  Pawns::Table pawnsTable;
+  Material::Table materialTable;
+  size_t pvIdx, pvLast;
+  std::atomic<uint64_t> nodes, tbHits, bestMoveChanges;
+  int selDepth, nmpMinPly;
+  Value bestValue, optimism[COLOR_NB];
+
+  Position rootPos;
+  StateInfo rootState;
+  Search::RootMoves rootMoves;
+  Depth rootDepth, completedDepth;
+  Value rootDelta;
+  CounterMoveHistory counterMoves;
+  ButterflyHistory mainHistory;
+  CapturePieceToHistory captureHistory;
+  ContinuationHistory continuationHistory[2][2];
 };
 
 
-/// ThreadPool class handles all the threads related stuff like init, starting,
-/// parking and, the most important, launching a slave thread at a split point.
-/// All the access to shared thread data is done through this class.
+/// MainThread is a derived class specific for main thread
 
-class ThreadPool {
+struct MainThread : public Thread {
 
-public:
-  void init(); // No c'tor and d'tor, threads rely on globals that should
-  void exit(); // be initialized and valid during the whole thread lifetime.
+  using Thread::Thread;
 
-  Thread& operator[](size_t id) { return *threads[id]; }
-  bool use_sleeping_threads() const { return useSleepingThreads; }
-  int min_split_depth() const { return minimumSplitDepth; }
-  size_t size() const { return threads.size(); }
-  Thread* main_thread() { return threads[0]; }
+  void search() override;
+  void check_time();
 
-  void wake_up() const;
-  void sleep() const;
-  void read_uci_options();
-  bool available_slave_exists(Thread* master) const;
-  void set_timer(int msec);
-  void wait_for_search_finished();
-  void start_searching(const Position&, const Search::LimitsType&,
-                       const std::vector<Move>&, Search::StateStackPtr&);
+  double previousTimeReduction;
+  Value bestPreviousScore;
+  Value bestPreviousAverageScore;
+  Value iterValue[4];
+  int callsCnt;
+  bool stopOnPonderhit;
+  std::atomic_bool ponder;
+};
+
+
+/// ThreadPool struct handles all the threads-related stuff like init, starting,
+/// parking and, most importantly, launching a thread. All the access to threads
+/// is done through this class.
+
+struct ThreadPool {
+
+  void start_thinking(Position&, StateListPtr&, const Search::LimitsType&, bool = false);
+  void clear();
+  void set(size_t);
+
+  MainThread* main()        const { return static_cast<MainThread*>(threads.front()); }
+  uint64_t nodes_searched() const { return accumulate(&Thread::nodes); }
+  uint64_t tb_hits()        const { return accumulate(&Thread::tbHits); }
+  Thread* get_best_thread() const;
+  void start_searching();
+  void wait_for_search_finished() const;
+
+  std::atomic_bool stop, increaseDepth;
+
+  auto cbegin() const noexcept { return threads.cbegin(); }
+  auto begin() noexcept { return threads.begin(); }
+  auto end() noexcept { return threads.end(); }
+  auto cend() const noexcept { return threads.cend(); }
+  auto size() const noexcept { return threads.size(); }
+  auto empty() const noexcept { return threads.empty(); }
 
-  template <bool Fake>
-  Value split(Position& pos, Search::Stack* ss, Value alpha, Value beta, Value bestValue, Move* bestMove,
-              Depth depth, Move threatMove, int moveCount, MovePicker* mp, int nodeType);
 private:
-  friend class Thread;
-
+  StateListPtr setupStates;
   std::vector<Thread*> threads;
-  Thread* timer;
-  Mutex mutex;
-  ConditionVariable sleepCondition;
-  Depth minimumSplitDepth;
-  int maxThreadsPerSplitPoint;
-  bool useSleepingThreads;
+
+  uint64_t accumulate(std::atomic<uint64_t> Thread::* member) const {
+
+    uint64_t sum = 0;
+    for (Thread* th : threads)
+        sum += (th->*member).load(std::memory_order_relaxed);
+    return sum;
+  }
 };
 
 extern ThreadPool Threads;
 
-#endif // !defined(THREAD_H_INCLUDED)
+} // namespace Stockfish
+
+#endif // #ifndef THREAD_H_INCLUDED

src/thread_win32_osx.h

@@ -0,0 +1,74 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+
+  Stockfish is free software: you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation, either version 3 of the License, or
+  (at your option) any later version.
+
+  Stockfish is distributed in the hope that it will be useful,
+  but WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+  GNU General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef THREAD_WIN32_OSX_H_INCLUDED
+#define THREAD_WIN32_OSX_H_INCLUDED
+
+#include <thread>
+
+/// On OSX threads other than the main thread are created with a reduced stack
+/// size of 512KB by default, this is too low for deep searches, which require
+/// somewhat more than 1MB stack, so adjust it to TH_STACK_SIZE.
+/// The implementation calls pthread_create() with the stack size parameter
+/// equal to the linux 8MB default, on platforms that support it.
+
+#if defined(__APPLE__) || defined(__MINGW32__) || defined(__MINGW64__) || defined(USE_PTHREADS)
+
+#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::*)()>>
+void* start_routine(void* ptr)
+{
+   P* p = reinterpret_cast<P*>(ptr);
+   (p->first->*(p->second))(); // Call member function pointer
+   delete p;
+   return nullptr;
+}
+
+class NativeThread {
+
+   pthread_t thread;
+
+public:
+  template<class T, class P = std::pair<T*, void(T::*)()>>
+  explicit NativeThread(void(T::*fun)(), T* obj) {
+    pthread_attr_t attr_storage, *attr = &attr_storage;
+    pthread_attr_init(attr);
+    pthread_attr_setstacksize(attr, TH_STACK_SIZE);
+    pthread_create(&thread, attr, start_routine<T>, new P(obj, fun));
+  }
+  void join() { pthread_join(thread, nullptr); }
+};
+
+} // namespace Stockfish
+
+#else // Default case: use STL classes
+
+namespace Stockfish {
+
+using NativeThread = std::thread;
+
+} // namespace Stockfish
+
+#endif
+
+#endif // #ifndef THREAD_WIN32_OSX_H_INCLUDED

src/timeman.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,146 +16,94 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#include <cmath>
 #include <algorithm>
+#include <cfloat>
+#include <cmath>
 
 #include "search.h"
 #include "timeman.h"
-#include "ucioption.h"
+#include "uci.h"
 
-namespace {
+namespace Stockfish {
 
-  /// Constants
-
-  const int MoveHorizon  = 50;    // Plan time management at most this many moves ahead
-  const float MaxRatio   = 3.0f;  // When in trouble, we can step over reserved time with this ratio
-  const float StealRatio = 0.33f; // However we must not steal time from remaining moves over this ratio
+TimeManagement Time; // Our global time management object
 
 
-  // MoveImportance[] is based on naive statistical analysis of "how many games are still undecided
-  // after n half-moves". Game is considered "undecided" as long as neither side has >275cp advantage.
-  // Data was extracted from CCRL game database with some simple filtering criteria.
-  const int MoveImportance[512] = {
-    7780, 7780, 7780, 7780, 7780, 7780, 7780, 7780, 7780, 7780, 7780, 7780, 7780, 7780, 7780, 7780,
-    7780, 7780, 7780, 7780, 7778, 7778, 7776, 7776, 7776, 7773, 7770, 7768, 7766, 7763, 7757, 7751,
-    7743, 7735, 7724, 7713, 7696, 7689, 7670, 7656, 7627, 7605, 7571, 7549, 7522, 7493, 7462, 7425,
-    7385, 7350, 7308, 7272, 7230, 7180, 7139, 7094, 7055, 7010, 6959, 6902, 6841, 6778, 6705, 6651,
-    6569, 6508, 6435, 6378, 6323, 6253, 6152, 6085, 5995, 5931, 5859, 5794, 5717, 5646, 5544, 5462,
-    5364, 5282, 5172, 5078, 4988, 4901, 4831, 4764, 4688, 4609, 4536, 4443, 4365, 4293, 4225, 4155,
-    4085, 4005, 3927, 3844, 3765, 3693, 3634, 3560, 3479, 3404, 3331, 3268, 3207, 3146, 3077, 3011,
-    2947, 2894, 2828, 2776, 2727, 2676, 2626, 2589, 2538, 2490, 2442, 2394, 2345, 2302, 2243, 2192,
-    2156, 2115, 2078, 2043, 2004, 1967, 1922, 1893, 1845, 1809, 1772, 1736, 1702, 1674, 1640, 1605,
-    1566, 1536, 1509, 1479, 1452, 1423, 1388, 1362, 1332, 1304, 1289, 1266, 1250, 1228, 1206, 1180,
-    1160, 1134, 1118, 1100, 1080, 1068, 1051, 1034, 1012, 1001, 980, 960, 945, 934, 916, 900, 888,
-    878, 865, 852, 828, 807, 787, 770, 753, 744, 731, 722, 706, 700, 683, 676, 671, 664, 652, 641,
-    634, 627, 613, 604, 591, 582, 568, 560, 552, 540, 534, 529, 519, 509, 495, 484, 474, 467, 460,
-    450, 438, 427, 419, 410, 406, 399, 394, 387, 382, 377, 372, 366, 359, 353, 348, 343, 337, 333,
-    328, 321, 315, 309, 303, 298, 293, 287, 284, 281, 277, 273, 265, 261, 255, 251, 247, 241, 240,
-    235, 229, 218, 217, 213, 212, 208, 206, 197, 193, 191, 189, 185, 184, 180, 177, 172, 170, 170,
-    170, 166, 163, 159, 158, 156, 155, 151, 146, 141, 138, 136, 132, 130, 128, 125, 123, 122, 118,
-    118, 118, 117, 115, 114, 108, 107, 105, 105, 105, 102, 97, 97, 95, 94, 93, 91, 88, 86, 83, 80,
-    80, 79, 79, 79, 78, 76, 75, 72, 72, 71, 70, 68, 65, 63, 61, 61, 59, 59, 59, 58, 56, 55, 54, 54,
-    52, 49, 48, 48, 48, 48, 45, 45, 45, 44, 43, 41, 41, 41, 41, 40, 40, 38, 37, 36, 34, 34, 34, 33,
-    31, 29, 29, 29, 28, 28, 28, 28, 28, 28, 28, 27, 27, 27, 27, 27, 24, 24, 23, 23, 22, 21, 20, 20,
-    19, 19, 19, 19, 19, 18, 18, 18, 18, 17, 17, 17, 17, 17, 16, 16, 15, 15, 14, 14, 14, 12, 12, 11,
-    9, 9, 9, 9, 9, 9, 9, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
-    8, 8, 8, 8, 7, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
-    4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, 2, 2, 2, 2,
-    2, 1, 1, 1, 1, 1, 1, 1 };
+/// TimeManagement::init() is called at the beginning of the search and calculates
+/// the bounds of time allowed for the current game ply. We currently support:
+//      1) x basetime (+ z increment)
+//      2) x moves in y seconds (+ z increment)
 
-  int move_importance(int ply) { return MoveImportance[std::min(ply, 511)]; }
+void TimeManagement::init(Search::LimitsType& limits, Color us, int ply) {
 
+  // if we have no time, no need to initialize TM, except for the start time,
+  // which is used by movetime.
+  startTime = limits.startTime;
+  if (limits.time[us] == 0)
+      return;
 
-  /// Function Prototypes
+  TimePoint moveOverhead    = TimePoint(Options["Move Overhead"]);
+  TimePoint slowMover       = TimePoint(Options["Slow Mover"]);
+  TimePoint npmsec          = TimePoint(Options["nodestime"]);
 
-  enum TimeType { OptimumTime, MaxTime };
+  // optScale is a percentage of available time to use for the current move.
+  // maxScale is a multiplier applied to optimumTime.
+  double optScale, maxScale;
 
-  template<TimeType>
-  int remaining(int myTime, int movesToGo, int fullMoveNumber, int slowMover);
-}
-
-
-void TimeManager::pv_instability(int curChanges, int prevChanges) {
-
-  unstablePVExtraTime =  curChanges  * (optimumSearchTime / 2)
-                       + prevChanges * (optimumSearchTime / 3);
-}
-
-
-void TimeManager::init(const Search::LimitsType& limits, int currentPly, Color us)
-{
-  /* We support four different kind of time controls:
-
-      increment == 0 && movesToGo == 0 means: x basetime  [sudden death!]
-      increment == 0 && movesToGo != 0 means: x moves in y minutes
-      increment >  0 && movesToGo == 0 means: x basetime + z increment
-      increment >  0 && movesToGo != 0 means: x moves in y minutes + z increment
-
-    Time management is adjusted by following UCI parameters:
-
-      emergencyMoveHorizon: Be prepared to always play at least this many moves
-      emergencyBaseTime   : Always attempt to keep at least this much time (in ms) at clock
-      emergencyMoveTime   : Plus attempt to keep at least this much time for each remaining emergency move
-      minThinkingTime     : No matter what, use at least this much thinking before doing the move
-  */
-
-  int hypMTG, hypMyTime, t1, t2;
-
-  // Read uci parameters
-  int emergencyMoveHorizon = Options["Emergency Move Horizon"];
-  int emergencyBaseTime    = Options["Emergency Base Time"];
-  int emergencyMoveTime    = Options["Emergency Move Time"];
-  int minThinkingTime      = Options["Minimum Thinking Time"];
-  int slowMover            = Options["Slow Mover"];
-
-  // Initialize to maximum values but unstablePVExtraTime that is reset
-  unstablePVExtraTime = 0;
-  optimumSearchTime = maximumSearchTime = limits.time[us];
-
-  // We calculate optimum time usage for different hypothetic "moves to go"-values and choose the
-  // minimum of calculated search time values. Usually the greatest hypMTG gives the minimum values.
-  for (hypMTG = 1; hypMTG <= (limits.movestogo ? std::min(limits.movestogo, MoveHorizon) : MoveHorizon); hypMTG++)
+  // If we have to play in 'nodes as time' mode, then convert from time
+  // to nodes, and use resulting values in time management formulas.
+  // WARNING: to avoid time losses, the given npmsec (nodes per millisecond)
+  // must be much lower than the real engine speed.
+  if (npmsec)
   {
-      // Calculate thinking time for hypothetic "moves to go"-value
-      hypMyTime =  limits.time[us]
-                 + limits.inc[us] * (hypMTG - 1)
-                 - emergencyBaseTime
-                 - emergencyMoveTime * std::min(hypMTG, emergencyMoveHorizon);
+      if (!availableNodes) // Only once at game start
+          availableNodes = npmsec * limits.time[us]; // Time is in msec
 
-      hypMyTime = std::max(hypMyTime, 0);
-
-      t1 = minThinkingTime + remaining<OptimumTime>(hypMyTime, hypMTG, currentPly, slowMover);
-      t2 = minThinkingTime + remaining<MaxTime>(hypMyTime, hypMTG, currentPly, slowMover);
-
-      optimumSearchTime = std::min(optimumSearchTime, t1);
-      maximumSearchTime = std::min(maximumSearchTime, t2);
+      // Convert from milliseconds to nodes
+      limits.time[us] = TimePoint(availableNodes);
+      limits.inc[us] *= npmsec;
+      limits.npmsec = npmsec;
   }
 
+  // Maximum move horizon of 50 moves
+  int mtg = limits.movestogo ? std::min(limits.movestogo, 50) : 50;
+
+  // Make sure timeLeft is > 0 since we may use it as a divisor
+  TimePoint timeLeft =  std::max(TimePoint(1),
+      limits.time[us] + limits.inc[us] * (mtg - 1) - moveOverhead * (2 + mtg));
+
+  // Use extra time with larger increments
+  double optExtra = std::clamp(1.0 + 12.0 * limits.inc[us] / limits.time[us], 1.0, 1.12);
+
+  // A user may scale time usage by setting UCI option "Slow Mover"
+  // Default is 100 and changing this value will probably lose elo.
+  timeLeft = slowMover * timeLeft / 100;
+
+  // x basetime (+ z increment)
+  // If there is a healthy increment, timeLeft can exceed actual available
+  // game time for the current move, so also cap to 20% of available game time.
+  if (limits.movestogo == 0)
+  {
+      optScale = std::min(0.0120 + std::pow(ply + 3.0, 0.45) * 0.0039,
+                           0.2 * limits.time[us] / double(timeLeft))
+                 * optExtra;
+      maxScale = std::min(7.0, 4.0 + ply / 12.0);
+  }
+
+  // x moves in y seconds (+ z increment)
+  else
+  {
+      optScale = std::min((0.88 + ply / 116.4) / mtg,
+                            0.88 * limits.time[us] / double(timeLeft));
+      maxScale = std::min(6.3, 1.5 + 0.11 * mtg);
+  }
+
+  // Never use more than 80% of the available time for this move
+  optimumTime = TimePoint(optScale * timeLeft);
+  maximumTime = TimePoint(std::min(0.8 * limits.time[us] - moveOverhead, maxScale * optimumTime));
+
   if (Options["Ponder"])
-      optimumSearchTime += optimumSearchTime / 4;
-
-  // Make sure that maxSearchTime is not over absoluteMaxSearchTime
-  optimumSearchTime = std::min(optimumSearchTime, maximumSearchTime);
+      optimumTime += optimumTime / 4;
 }
 
-
-namespace {
-
-  template<TimeType T>
-  int remaining(int myTime, int movesToGo, int currentPly, int slowMover)
-  {
-    const float TMaxRatio   = (T == OptimumTime ? 1 : MaxRatio);
-    const float TStealRatio = (T == OptimumTime ? 0 : StealRatio);
-
-    int thisMoveImportance = move_importance(currentPly) * slowMover / 100;
-    int otherMovesImportance = 0;
-
-    for (int i = 1; i < movesToGo; i++)
-        otherMovesImportance += move_importance(currentPly + 2 * i);
-
-    float ratio1 = (TMaxRatio * thisMoveImportance) / float(TMaxRatio * thisMoveImportance + otherMovesImportance);
-    float ratio2 = (thisMoveImportance + TStealRatio * otherMovesImportance) / float(thisMoveImportance + otherMovesImportance);
-
-    return int(floor(myTime * std::min(ratio1, ratio2)));
-  }
-}
+} // namespace Stockfish

src/timeman.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,23 +16,36 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(TIMEMAN_H_INCLUDED)
+#ifndef TIMEMAN_H_INCLUDED
 #define TIMEMAN_H_INCLUDED
 
-/// The TimeManager class computes the optimal time to think depending on the
-/// maximum available time, the move game number and other parameters.
+#include "misc.h"
+#include "search.h"
+#include "thread.h"
 
-class TimeManager {
+namespace Stockfish {
+
+/// The TimeManagement class computes the optimal time to think depending on
+/// the maximum available time, the game move number and other parameters.
+
+class TimeManagement {
 public:
-  void init(const Search::LimitsType& limits, int currentPly, Color us);
-  void pv_instability(int curChanges, int prevChanges);
-  int available_time() const { return optimumSearchTime + unstablePVExtraTime; }
-  int maximum_time() const { return maximumSearchTime; }
+  void init(Search::LimitsType& limits, Color us, int ply);
+  TimePoint optimum() const { return optimumTime; }
+  TimePoint maximum() const { return maximumTime; }
+  TimePoint elapsed() const { return Search::Limits.npmsec ?
+                                     TimePoint(Threads.nodes_searched()) : now() - startTime; }
+
+  int64_t availableNodes; // When in 'nodes as time' mode
 
 private:
-  int optimumSearchTime;
-  int maximumSearchTime;
-  int unstablePVExtraTime;
+  TimePoint startTime;
+  TimePoint optimumTime;
+  TimePoint maximumTime;
 };
 
-#endif // !defined(TIMEMAN_H_INCLUDED)
+extern TimeManagement Time;
+
+} // namespace Stockfish
+
+#endif // #ifndef TIMEMAN_H_INCLUDED

src/tt.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,125 +16,147 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#include <cstring>
+#include <cstring>   // For std::memset
 #include <iostream>
+#include <thread>
 
 #include "bitboard.h"
+#include "misc.h"
+#include "thread.h"
 #include "tt.h"
+#include "uci.h"
+
+namespace Stockfish {
 
 TranspositionTable TT; // Our global transposition table
 
-TranspositionTable::TranspositionTable() {
+/// TTEntry::save() populates the TTEntry with a new node's data, possibly
+/// overwriting an old position. Update is not atomic and can be racy.
 
-  size = generation = 0;
-  entries = NULL;
-}
+void TTEntry::save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev) {
 
-TranspositionTable::~TranspositionTable() {
+  // Preserve any existing move for the same position
+  if (m || (uint16_t)k != key16)
+      move16 = (uint16_t)m;
 
-  delete [] entries;
+  // Overwrite less valuable entries (cheapest checks first)
+  if (   b == BOUND_EXACT
+      || (uint16_t)k != key16
+      || d - DEPTH_OFFSET + 2 * pv > depth8 - 4)
+  {
+      assert(d > DEPTH_OFFSET);
+      assert(d < 256 + DEPTH_OFFSET);
+
+      key16     = (uint16_t)k;
+      depth8    = (uint8_t)(d - DEPTH_OFFSET);
+      genBound8 = (uint8_t)(TT.generation8 | uint8_t(pv) << 2 | b);
+      value16   = (int16_t)v;
+      eval16    = (int16_t)ev;
+  }
 }
 
 
-/// TranspositionTable::set_size() sets the size of the transposition table,
-/// measured in megabytes. Transposition table consists of a power of 2 number of
-/// TTCluster and each cluster consists of ClusterSize number of TTEntries. Each
-/// non-empty entry contains information of exactly one position.
+/// TranspositionTable::resize() sets the size of the transposition table,
+/// measured in megabytes. Transposition table consists of a power of 2 number
+/// of clusters and each cluster consists of ClusterSize number of TTEntry.
 
-void TranspositionTable::set_size(size_t mbSize) {
+void TranspositionTable::resize(size_t mbSize) {
 
-  size_t newSize = 1ULL << msb((mbSize << 20) / sizeof(TTCluster));
+  Threads.main()->wait_for_search_finished();
 
-  if (newSize == size)
-      return;
+  aligned_large_pages_free(table);
 
-  size = newSize;
-  delete [] entries;
-  entries = new (std::nothrow) TTCluster[size];
+  clusterCount = mbSize * 1024 * 1024 / sizeof(Cluster);
 
-  if (!entries)
+  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;
       exit(EXIT_FAILURE);
   }
 
-  clear(); // Operator new is not guaranteed to initialize memory to zero
+  clear();
 }
 
 
-/// TranspositionTable::clear() overwrites the entire transposition table
-/// with zeroes. It is called whenever the table is resized, or when the
-/// user asks the program to clear the table (from the UCI interface).
+/// TranspositionTable::clear() initializes the entire transposition table to zero,
+//  in a multi-threaded way.
 
 void TranspositionTable::clear() {
 
-  memset(entries, 0, size * sizeof(TTCluster));
+  std::vector<std::thread> threads;
+
+  for (size_t idx = 0; idx < size_t(Options["Threads"]); ++idx)
+  {
+      threads.emplace_back([this, idx]() {
+
+          // Thread binding gives faster search on systems with a first-touch policy
+          if (Options["Threads"] > 8)
+              WinProcGroup::bindThisThread(idx);
+
+          // Each thread will zero its part of the hash table
+          const size_t stride = size_t(clusterCount / Options["Threads"]),
+                       start  = size_t(stride * idx),
+                       len    = idx != size_t(Options["Threads"]) - 1 ?
+                                stride : clusterCount - start;
+
+          std::memset(&table[start], 0, len * sizeof(Cluster));
+      });
+  }
+
+  for (std::thread& th : threads)
+      th.join();
 }
 
 
-/// TranspositionTable::store() writes a new entry containing position key and
-/// valuable information of current position. The lowest order bits of position
-/// key are used to decide on which cluster the position will be placed.
-/// When a new entry is written and there are no empty entries available in cluster,
-/// it replaces the least valuable of entries. A TTEntry t1 is considered to be
-/// more valuable than a TTEntry t2 if t1 is from the current search and t2 is from
-/// a previous search, or if the depth of t1 is bigger than the depth of t2.
+/// TranspositionTable::probe() looks up the current position in the transposition
+/// table. It returns true and a pointer to the TTEntry if the position is found.
+/// Otherwise, it returns false and a pointer to an empty or least valuable TTEntry
+/// to be replaced later. The replace value of an entry is calculated as its depth
+/// minus 8 times its relative age. TTEntry t1 is considered more valuable than
+/// TTEntry t2 if its replace value is greater than that of t2.
 
-void TranspositionTable::store(const Key posKey, Value v, Bound t, Depth d, Move m, Value statV, Value kingD) {
+TTEntry* TranspositionTable::probe(const Key key, bool& found) const {
 
-  int c1, c2, c3;
-  TTEntry *tte, *replace;
-  uint32_t posKey32 = posKey >> 32; // Use the high 32 bits as key inside the cluster
+  TTEntry* const tte = first_entry(key);
+  const uint16_t key16 = (uint16_t)key;  // Use the low 16 bits as key inside the cluster
 
-  tte = replace = first_entry(posKey);
-
-  for (int i = 0; i < ClusterSize; i++, tte++)
-  {
-      if (!tte->key() || tte->key() == posKey32) // Empty or overwrite old
+  for (int i = 0; i < ClusterSize; ++i)
+      if (tte[i].key16 == key16 || !tte[i].depth8)
       {
-          // Preserve any existing ttMove
-          if (m == MOVE_NONE)
-              m = tte->move();
+          tte[i].genBound8 = uint8_t(generation8 | (tte[i].genBound8 & (GENERATION_DELTA - 1))); // Refresh
 
-          tte->save(posKey32, v, t, d, m, generation, statV, kingD);
-          return;
+          return found = (bool)tte[i].depth8, &tte[i];
       }
 
-      // Implement replace strategy
-      c1 = (replace->generation() == generation ?  2 : 0);
-      c2 = (tte->generation() == generation || tte->type() == BOUND_EXACT ? -2 : 0);
-      c3 = (tte->depth() < replace->depth() ?  1 : 0);
+  // Find an entry to be replaced according to the replacement strategy
+  TTEntry* replace = tte;
+  for (int i = 1; i < ClusterSize; ++i)
+      // Due to our packed storage format for generation and its cyclic
+      // nature we add 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];
 
-      if (c1 + c2 + c3 > 0)
-          replace = tte;
-  }
-  replace->save(posKey32, v, t, d, m, generation, statV, kingD);
+  return found = false, replace;
 }
 
 
-/// TranspositionTable::probe() looks up the current position in the
-/// transposition table. Returns a pointer to the TTEntry or NULL if
-/// position is not found.
+/// TranspositionTable::hashfull() returns an approximation of the hashtable
+/// occupation during a search. The hash is x permill full, as per UCI protocol.
 
-TTEntry* TranspositionTable::probe(const Key posKey) const {
+int TranspositionTable::hashfull() const {
 
-  uint32_t posKey32 = posKey >> 32;
-  TTEntry* tte = first_entry(posKey);
+  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 & GENERATION_MASK) == generation8;
 
-  for (int i = 0; i < ClusterSize; i++, tte++)
-      if (tte->key() == posKey32)
-          return tte;
-
-  return NULL;
+  return cnt / ClusterSize;
 }
 
-
-/// TranspositionTable::new_search() is called at the beginning of every new
-/// search. It increments the "generation" variable, which is used to
-/// distinguish transposition table entries from previous searches from
-/// entries from the current search.
-
-void TranspositionTable::new_search() {
-  generation++;
-}
+} // namespace Stockfish

src/tt.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,120 +16,92 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(TT_H_INCLUDED)
+#ifndef TT_H_INCLUDED
 #define TT_H_INCLUDED
 
 #include "misc.h"
 #include "types.h"
 
-/// The TTEntry is the class of transposition table entries
-///
-/// A TTEntry needs 128 bits to be stored
-///
-/// bit  0-31: key
-/// bit 32-63: data
-/// bit 64-79: value
-/// bit 80-95: depth
-/// bit 96-111: static value
-/// bit 112-127: margin of static value
-///
-/// the 32 bits of the data field are so defined
-///
-/// bit  0-15: move
-/// bit 16-20: not used
-/// bit 21-22: value type
-/// bit 23-31: generation
+namespace Stockfish {
 
-class TTEntry {
+/// TTEntry struct is the 10 bytes transposition table entry, defined as below:
+///
+/// key        16 bit
+/// depth       8 bit
+/// generation  5 bit
+/// pv node     1 bit
+/// bound type  2 bit
+/// move       16 bit
+/// value      16 bit
+/// eval value 16 bit
 
-public:
-  void save(uint32_t k, Value v, Bound b, Depth d, Move m, int g, Value statV, Value statM) {
+struct TTEntry {
 
-    key32        = (uint32_t)k;
-    move16       = (uint16_t)m;
-    bound        = (uint8_t)b;
-    generation8  = (uint8_t)g;
-    value16      = (int16_t)v;
-    depth16      = (int16_t)d;
-    staticValue  = (int16_t)statV;
-    staticMargin = (int16_t)statM;
-  }
-  void set_generation(int g) { generation8 = (uint8_t)g; }
-
-  uint32_t key() const              { return key32; }
-  Depth depth() const               { return (Depth)depth16; }
-  Move move() const                 { return (Move)move16; }
-  Value value() const               { return (Value)value16; }
-  Bound type() const                { return (Bound)bound; }
-  int generation() const            { return (int)generation8; }
-  Value static_value() const        { return (Value)staticValue; }
-  Value static_value_margin() const { return (Value)staticMargin; }
+  Move  move()  const { return (Move )move16; }
+  Value value() const { return (Value)value16; }
+  Value eval()  const { return (Value)eval16; }
+  Depth depth() const { return (Depth)depth8 + DEPTH_OFFSET; }
+  bool is_pv()  const { return (bool)(genBound8 & 0x4); }
+  Bound bound() const { return (Bound)(genBound8 & 0x3); }
+  void save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev);
 
 private:
-  uint32_t key32;
+  friend class TranspositionTable;
+
+  uint16_t key16;
+  uint8_t  depth8;
+  uint8_t  genBound8;
   uint16_t move16;
-  uint8_t bound, generation8;
-  int16_t value16, depth16, staticValue, staticMargin;
+  int16_t  value16;
+  int16_t  eval16;
 };
 
 
-/// This is the number of TTEntry slots for each cluster
-const int ClusterSize = 4;
-
-
-/// TTCluster consists of ClusterSize number of TTEntries. Size of TTCluster
-/// must not be bigger than a cache line size. In case it is less, it should
-/// be padded to guarantee always aligned accesses.
-
-struct TTCluster {
-  TTEntry data[ClusterSize];
-};
-
-
-/// The transposition table class. This is basically just a huge array containing
-/// TTCluster objects, and a few methods for writing and reading entries.
+/// A TranspositionTable is an array of Cluster, of size clusterCount. Each
+/// cluster consists of ClusterSize number of TTEntry. Each non-empty TTEntry
+/// contains information on exactly one position. The size of a Cluster should
+/// divide the size of a cache line for best performance, as the cacheline is
+/// prefetched when possible.
 
 class TranspositionTable {
 
-  TranspositionTable(const TranspositionTable&);
-  TranspositionTable& operator=(const TranspositionTable&);
+  static constexpr int ClusterSize = 3;
+
+  struct Cluster {
+    TTEntry entry[ClusterSize];
+    char padding[2]; // Pad to 32 bytes
+  };
+
+  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();
-  ~TranspositionTable();
-  void set_size(size_t mbSize);
+ ~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);
   void clear();
-  void store(const Key posKey, Value v, Bound type, Depth d, Move m, Value statV, Value kingD);
-  TTEntry* probe(const Key posKey) const;
-  void new_search();
-  TTEntry* first_entry(const Key posKey) const;
-  void refresh(const TTEntry* tte) const;
+
+  TTEntry* first_entry(const Key key) const {
+    return &table[mul_hi64(key, clusterCount)].entry[0];
+  }
 
 private:
-  size_t size;
-  TTCluster* entries;
-  uint8_t generation; // Size must be not bigger then TTEntry::generation8
+  friend struct TTEntry;
+
+  size_t clusterCount;
+  Cluster* table;
+  uint8_t generation8; // Size must be not bigger than TTEntry::genBound8
 };
 
 extern TranspositionTable TT;
 
+} // namespace Stockfish
 
-/// TranspositionTable::first_entry() returns a pointer to the first entry of
-/// a cluster given a position. The lowest order bits of the key are used to
-/// get the index of the cluster.
-
-inline TTEntry* TranspositionTable::first_entry(const Key posKey) const {
-
-  return entries[((uint32_t)posKey) & (size - 1)].data;
-}
-
-
-/// TranspositionTable::refresh() updates the 'generation' value of the TTEntry
-/// to avoid aging. Normally called after a TT hit.
-
-inline void TranspositionTable::refresh(const TTEntry* tte) const {
-
-  const_cast<TTEntry*>(tte)->set_generation(generation);
-}
-
-#endif // !defined(TT_H_INCLUDED)
+#endif // #ifndef TT_H_INCLUDED

src/tune.cpp

@@ -0,0 +1,133 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 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/>.
+*/
+
+#include <algorithm>
+#include <iostream>
+#include <sstream>
+
+#include "types.h"
+#include "misc.h"
+#include "uci.h"
+
+using std::string;
+
+namespace Stockfish {
+
+bool Tune::update_on_last;
+const UCI::Option* LastOption = nullptr;
+static std::map<std::string, int> TuneResults;
+
+string Tune::next(string& names, bool pop) {
+
+  string name;
+
+  do {
+      string token = names.substr(0, names.find(','));
+
+      if (pop)
+          names.erase(0, token.size() + 1);
+
+      std::stringstream ws(token);
+      name += (ws >> token, token); // Remove trailing whitespace
+
+  } while (  std::count(name.begin(), name.end(), '(')
+           - std::count(name.begin(), name.end(), ')'));
+
+  return name;
+}
+
+static void on_tune(const UCI::Option& o) {
+
+  if (!Tune::update_on_last || LastOption == &o)
+      Tune::read_options();
+}
+
+static void make_option(const string& n, int v, const SetRange& r) {
+
+  // Do not generate option when there is nothing to tune (ie. min = max)
+  if (r(v).first == r(v).second)
+      return;
+
+  if (TuneResults.count(n))
+      v = TuneResults[n];
+
+  Options[n] << UCI::Option(v, r(v).first, r(v).second, on_tune);
+  LastOption = &Options[n];
+
+  // Print formatted parameters, ready to be copy-pasted in Fishtest
+  std::cout << n << ","
+            << v << ","
+            << r(v).first << "," << r(v).second << ","
+            << (r(v).second - r(v).first) / 20.0 << ","
+            << "0.0020"
+            << std::endl;
+}
+
+template<> void Tune::Entry<int>::init_option() { make_option(name, value, range); }
+
+template<> void Tune::Entry<int>::read_option() {
+  if (Options.count(name))
+      value = int(Options[name]);
+}
+
+template<> void Tune::Entry<Value>::init_option() { make_option(name, value, range); }
+
+template<> void Tune::Entry<Value>::read_option() {
+  if (Options.count(name))
+      value = Value(int(Options[name]));
+}
+
+template<> void Tune::Entry<Score>::init_option() {
+  make_option("m" + name, mg_value(value), range);
+  make_option("e" + name, eg_value(value), range);
+}
+
+template<> void Tune::Entry<Score>::read_option() {
+  if (Options.count("m" + name))
+      value = make_score(int(Options["m" + name]), eg_value(value));
+
+  if (Options.count("e" + name))
+      value = make_score(mg_value(value), int(Options["e" + name]));
+}
+
+// Instead of a variable here we have a PostUpdate function: just call it
+template<> void Tune::Entry<Tune::PostUpdate>::init_option() {}
+template<> void Tune::Entry<Tune::PostUpdate>::read_option() { value(); }
+
+} // namespace Stockfish
+
+
+// Init options with tuning session results instead of default values. Useful to
+// get correct bench signature after a tuning session or to test tuned values.
+// Just copy fishtest tuning results in a result.txt file and extract the
+// values with:
+//
+// cat results.txt | sed 's/^param: \([^,]*\), best: \([^,]*\).*/  TuneResults["\1"] = int(round(\2));/'
+//
+// Then paste the output below, as the function body
+
+#include <cmath>
+
+namespace Stockfish {
+
+void Tune::read_results() {
+
+  /* ...insert your values here... */
+}
+
+} // namespace Stockfish

src/tune.h

@@ -0,0 +1,163 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+
+  Stockfish is free software: you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation, either version 3 of the License, or
+  (at your option) any later version.
+
+  Stockfish is distributed in the hope that it will be useful,
+  but WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+  GNU General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef TUNE_H_INCLUDED
+#define TUNE_H_INCLUDED
+
+#include <memory>
+#include <string>
+#include <type_traits>
+#include <vector>
+
+namespace Stockfish {
+
+using Range = std::pair<int, int>; // Option's min-max values
+using RangeFun = Range (int);
+
+// Default Range function, to calculate Option's min-max values
+inline Range default_range(int v) {
+  return v > 0 ? Range(0, 2 * v) : Range(2 * v, 0);
+}
+
+struct SetRange {
+  explicit SetRange(RangeFun f) : fun(f) {}
+  SetRange(int min, int max) : fun(nullptr), range(min, max) {}
+  Range operator()(int v) const { return fun ? fun(v) : range; }
+
+  RangeFun* fun;
+  Range range;
+};
+
+#define SetDefaultRange SetRange(default_range)
+
+
+/// 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
+/// if you have:
+///
+///   const Score myScore = S(10, 15);
+///   const Value myValue[][2] = { { V(100), V(20) }, { V(7), V(78) } };
+///
+/// If you have a my_post_update() function to run after values have been updated,
+/// and a my_range() function to set custom Option's min-max values, then you just
+/// remove the 'const' qualifiers and write somewhere below in the file:
+///
+///   TUNE(SetRange(my_range), myScore, myValue, my_post_update);
+///
+/// You can also set the range directly, and restore the default at the end
+///
+///   TUNE(SetRange(-100, 100), myScore, SetDefaultRange);
+///
+/// In case update function is slow and you have many parameters, you can add:
+///
+///   UPDATE_ON_LAST();
+///
+/// And the values update, including post update function call, will be done only
+/// once, after the engine receives the last UCI option, that is the one defined
+/// and created as the last one, so the GUI should send the options in the same
+/// order in which have been defined.
+
+class Tune {
+
+  using PostUpdate = void (); // Post-update function
+
+  Tune() { read_results(); }
+  Tune(const Tune&) = delete;
+  void operator=(const Tune&) = delete;
+  void read_results();
+
+  static Tune& instance() { static Tune t; return t; } // Singleton
+
+  // Use polymorphism to accommodate Entry of different types in the same vector
+  struct EntryBase {
+    virtual ~EntryBase() = default;
+    virtual void init_option() = 0;
+    virtual void read_option() = 0;
+  };
+
+  template<typename T>
+  struct Entry : public EntryBase {
+
+    static_assert(!std::is_const<T>::value, "Parameter cannot be const!");
+
+    static_assert(   std::is_same<T,   int>::value
+                  || std::is_same<T, Value>::value
+                  || std::is_same<T, Score>::value
+                  || std::is_same<T, PostUpdate>::value, "Parameter type not supported!");
+
+    Entry(const std::string& n, T& v, const SetRange& r) : name(n), value(v), range(r) {}
+    void operator=(const Entry&) = delete; // Because 'value' is a reference
+    void init_option() override;
+    void read_option() override;
+
+    std::string name;
+    T& value;
+    SetRange range;
+  };
+
+  // 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; }
+
+  template<typename T, typename... Args>
+  int add(const SetRange& range, std::string&& names, T& value, Args&&... args) {
+    list.push_back(std::unique_ptr<EntryBase>(new Entry<T>(next(names), value, range)));
+    return add(range, std::move(names), args...);
+  }
+
+  // Template specialization for arrays: recursively handle multi-dimensional arrays
+  template<typename T, size_t N, typename... Args>
+  int add(const SetRange& range, std::string&& names, T (&value)[N], Args&&... args) {
+    for (size_t i = 0; i < N; i++)
+        add(range, next(names, i == N - 1) + "[" + std::to_string(i) + "]", value[i]);
+    return add(range, std::move(names), args...);
+  }
+
+  // Template specialization for SetRange
+  template<typename... Args>
+  int add(const SetRange&, std::string&& names, SetRange& value, Args&&... args) {
+    return add(value, (next(names), std::move(names)), args...);
+  }
+
+  std::vector<std::unique_ptr<EntryBase>> list;
+
+public:
+  template<typename... Args>
+  static int add(const std::string& names, Args&&... args) {
+    return instance().add(SetDefaultRange, names.substr(1, names.size() - 2), args...); // Remove trailing parenthesis
+  }
+  static void init() { for (auto& e : instance().list) e->init_option(); read_options(); } // Deferred, due to UCI::Options access
+  static void read_options() { for (auto& e : instance().list) e->read_option(); }
+  static bool update_on_last;
+};
+
+// Some macro magic :-) we define a dummy int variable that compiler initializes calling Tune::add()
+#define STRINGIFY(x) #x
+#define UNIQUE2(x, y) x ## y
+#define UNIQUE(x, y) UNIQUE2(x, y) // Two indirection levels to expand __LINE__
+#define TUNE(...) int UNIQUE(p, __LINE__) = Tune::add(STRINGIFY((__VA_ARGS__)), __VA_ARGS__)
+
+#define UPDATE_ON_LAST() bool UNIQUE(p, __LINE__) = Tune::update_on_last = true
+
+} // namespace Stockfish
+
+#endif // #ifndef TUNE_H_INCLUDED

src/types.h

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,194 +16,221 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#if !defined(TYPES_H_INCLUDED)
+#ifndef TYPES_H_INCLUDED
 #define TYPES_H_INCLUDED
 
-/// For Linux and OSX configuration is done automatically using Makefile. To get
-/// started type 'make help'.
+/// When compiling with provided Makefile (e.g. for Linux and OSX), configuration
+/// is done automatically. To get started type 'make help'.
 ///
-/// For Windows, part of the configuration is detected automatically, but some
-/// switches need to be set manually:
+/// When Makefile is not used (e.g. with Microsoft Visual Studio) some switches
+/// need to be set manually:
 ///
-/// -DNDEBUG      | Disable debugging mode. Use always.
+/// -DNDEBUG      | Disable debugging mode. Always use this for release.
 ///
-/// -DNO_PREFETCH | Disable use of prefetch asm-instruction. A must if you want
-///               | the executable to run on some very old machines.
+/// -DNO_PREFETCH | Disable use of prefetch asm-instruction. You may need this to
+///               | run on some very old machines.
 ///
 /// -DUSE_POPCNT  | Add runtime support for use of popcnt asm-instruction. Works
-///               | only in 64-bit mode. For compiling requires hardware with
-///               | popcnt support.
+///               | only in 64-bit mode and requires hardware with popcnt support.
+///
+/// -DUSE_PEXT    | Add runtime support for use of pext asm-instruction. Works
+///               | only in 64-bit mode and requires hardware with pext support.
 
+#include <cassert>
 #include <cctype>
-#include <climits>
+#include <cstdint>
 #include <cstdlib>
-
-#include "platform.h"
-
-#if defined(_WIN64)
-#  include <intrin.h> // MSVC popcnt and bsfq instrinsics
-#  define IS_64BIT
-#  define USE_BSFQ
-#endif
-
-#if defined(USE_POPCNT) && defined(_MSC_VER) && defined(__INTEL_COMPILER)
-#  include <nmmintrin.h> // Intel header for _mm_popcnt_u64() intrinsic
-#endif
-
-#if defined(_MSC_VER) || defined(__INTEL_COMPILER)
-#  define CACHE_LINE_ALIGNMENT __declspec(align(64))
-#else
-#  define CACHE_LINE_ALIGNMENT  __attribute__ ((aligned(64)))
-#endif
+#include <algorithm>
 
 #if defined(_MSC_VER)
-#  define FORCE_INLINE  __forceinline
-#elif defined(__GNUC__)
-#  define FORCE_INLINE  inline __attribute__((always_inline))
-#else
-#  define FORCE_INLINE  inline
+// Disable some silly and noisy warning from MSVC compiler
+#pragma warning(disable: 4127) // Conditional expression is constant
+#pragma warning(disable: 4146) // Unary minus operator applied to unsigned type
+#pragma warning(disable: 4800) // Forcing value to bool 'true' or 'false'
 #endif
 
-#if defined(USE_POPCNT)
-const bool HasPopCnt = true;
-#else
-const bool HasPopCnt = false;
+/// Predefined macros hell:
+///
+/// __GNUC__           Compiler is gcc, Clang or Intel on Linux
+/// __INTEL_COMPILER   Compiler is Intel
+/// _MSC_VER           Compiler is MSVC or Intel on Windows
+/// _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
 
-#if defined(IS_64BIT)
-const bool Is64Bit = true;
-#else
-const bool Is64Bit = false;
+#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
 #endif
 
-typedef uint64_t Key;
-typedef uint64_t Bitboard;
+#if defined(USE_POPCNT) && (defined(__INTEL_COMPILER) || defined(_MSC_VER))
+#  include <nmmintrin.h> // Intel and Microsoft header for _mm_popcnt_u64()
+#endif
 
-const int MAX_MOVES      = 192;
-const int MAX_PLY        = 100;
-const int MAX_PLY_PLUS_2 = MAX_PLY + 2;
+#if !defined(NO_PREFETCH) && (defined(__INTEL_COMPILER) || defined(_MSC_VER))
+#  include <xmmintrin.h> // Intel and Microsoft header for _mm_prefetch()
+#endif
 
-const Bitboard FileABB = 0x0101010101010101ULL;
-const Bitboard FileBBB = FileABB << 1;
-const Bitboard FileCBB = FileABB << 2;
-const Bitboard FileDBB = FileABB << 3;
-const Bitboard FileEBB = FileABB << 4;
-const Bitboard FileFBB = FileABB << 5;
-const Bitboard FileGBB = FileABB << 6;
-const Bitboard FileHBB = FileABB << 7;
+#if defined(USE_PEXT)
+#  include <immintrin.h> // Header for _pext_u64() intrinsic
+#  define pext(b, m) _pext_u64(b, m)
+#else
+#  define pext(b, m) 0
+#endif
 
-const Bitboard Rank1BB = 0xFF;
-const Bitboard Rank2BB = Rank1BB << (8 * 1);
-const Bitboard Rank3BB = Rank1BB << (8 * 2);
-const Bitboard Rank4BB = Rank1BB << (8 * 3);
-const Bitboard Rank5BB = Rank1BB << (8 * 4);
-const Bitboard Rank6BB = Rank1BB << (8 * 5);
-const Bitboard Rank7BB = Rank1BB << (8 * 6);
-const Bitboard Rank8BB = Rank1BB << (8 * 7);
+namespace Stockfish {
 
+#ifdef USE_POPCNT
+constexpr bool HasPopCnt = true;
+#else
+constexpr bool HasPopCnt = false;
+#endif
+
+#ifdef USE_PEXT
+constexpr bool HasPext = true;
+#else
+constexpr bool HasPext = false;
+#endif
+
+#ifdef IS_64BIT
+constexpr bool Is64Bit = true;
+#else
+constexpr bool Is64Bit = false;
+#endif
+
+using Key = uint64_t;
+using Bitboard = uint64_t;
+
+constexpr int MAX_MOVES = 256;
+constexpr int MAX_PLY   = 246;
 
 /// A move needs 16 bits to be stored
 ///
 /// bit  0- 5: destination square (from 0 to 63)
 /// 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), castle (3)
+/// 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
 ///
 /// 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
 /// while MOVE_NONE and MOVE_NULL have the same origin and destination square.
 
-enum Move {
-  MOVE_NONE = 0,
+enum Move : int {
+  MOVE_NONE,
   MOVE_NULL = 65
 };
 
 enum MoveType {
-  NORMAL    = 0,
+  NORMAL,
   PROMOTION = 1 << 14,
-  ENPASSANT = 2 << 14,
-  CASTLE    = 3 << 14
-};
-
-enum CastleRight {  // Defined as in PolyGlot book hash key
-  CASTLES_NONE = 0,
-  WHITE_OO     = 1,
-  WHITE_OOO    = 2,
-  BLACK_OO     = 4,
-  BLACK_OOO    = 8,
-  ALL_CASTLES  = 15
-};
-
-enum CastlingSide {
-  KING_SIDE,
-  QUEEN_SIDE
-};
-
-enum ScaleFactor {
-  SCALE_FACTOR_DRAW   = 0,
-  SCALE_FACTOR_NORMAL = 64,
-  SCALE_FACTOR_MAX    = 128,
-  SCALE_FACTOR_NONE   = 255
-};
-
-enum Bound {
-  BOUND_NONE  = 0,
-  BOUND_UPPER = 1,
-  BOUND_LOWER = 2,
-  BOUND_EXACT = BOUND_UPPER | BOUND_LOWER
-};
-
-enum Value {
-  VALUE_ZERO      = 0,
-  VALUE_DRAW      = 0,
-  VALUE_KNOWN_WIN = 15000,
-  VALUE_MATE      = 30000,
-  VALUE_INFINITE  = 30001,
-  VALUE_NONE      = 30002,
-
-  VALUE_MATE_IN_MAX_PLY  =  VALUE_MATE - MAX_PLY,
-  VALUE_MATED_IN_MAX_PLY = -VALUE_MATE + MAX_PLY,
-
-  VALUE_ENSURE_INTEGER_SIZE_P = INT_MAX,
-  VALUE_ENSURE_INTEGER_SIZE_N = INT_MIN,
-
-  Mg = 0, Eg = 1,
-
-  PawnValueMg   = 198,   PawnValueEg   = 258,
-  KnightValueMg = 817,   KnightValueEg = 846,
-  BishopValueMg = 836,   BishopValueEg = 857,
-  RookValueMg   = 1270,  RookValueEg   = 1278,
-  QueenValueMg  = 2521,  QueenValueEg  = 2558
-};
-
-enum PieceType {
-  NO_PIECE_TYPE = 0, ALL_PIECES = 0,
-  PAWN = 1, KNIGHT = 2, BISHOP = 3, ROOK = 4, QUEEN = 5, KING = 6
-};
-
-enum Piece {
-  NO_PIECE = 16, // color_of(NO_PIECE) == NO_COLOR
-  W_PAWN = 1, W_KNIGHT =  2, W_BISHOP =  3, W_ROOK =  4, W_QUEEN =  5, W_KING =  6,
-  B_PAWN = 9, B_KNIGHT = 10, B_BISHOP = 11, B_ROOK = 12, B_QUEEN = 13, B_KING = 14
+  EN_PASSANT = 2 << 14,
+  CASTLING  = 3 << 14
 };
 
 enum Color {
-  WHITE, BLACK, NO_COLOR
+  WHITE, BLACK, COLOR_NB = 2
 };
 
-enum Depth {
+enum CastlingRights {
+  NO_CASTLING,
+  WHITE_OO,
+  WHITE_OOO = WHITE_OO << 1,
+  BLACK_OO  = WHITE_OO << 2,
+  BLACK_OOO = WHITE_OO << 3,
 
-  ONE_PLY = 2,
+  KING_SIDE      = WHITE_OO  | BLACK_OO,
+  QUEEN_SIDE     = WHITE_OOO | BLACK_OOO,
+  WHITE_CASTLING = WHITE_OO  | WHITE_OOO,
+  BLACK_CASTLING = BLACK_OO  | BLACK_OOO,
+  ANY_CASTLING   = WHITE_CASTLING | BLACK_CASTLING,
 
-  DEPTH_ZERO          =  0 * ONE_PLY,
-  DEPTH_QS_CHECKS     = -1 * ONE_PLY,
-  DEPTH_QS_NO_CHECKS  = -2 * ONE_PLY,
-  DEPTH_QS_RECAPTURES = -5 * ONE_PLY,
-
-  DEPTH_NONE = -127 * ONE_PLY
+  CASTLING_RIGHT_NB = 16
 };
 
-enum Square {
+enum Phase {
+  PHASE_ENDGAME,
+  PHASE_MIDGAME = 128,
+  MG = 0, EG = 1, PHASE_NB = 2
+};
+
+enum ScaleFactor {
+  SCALE_FACTOR_DRAW    = 0,
+  SCALE_FACTOR_NORMAL  = 64,
+  SCALE_FACTOR_MAX     = 128,
+  SCALE_FACTOR_NONE    = 255
+};
+
+enum Bound {
+  BOUND_NONE,
+  BOUND_UPPER,
+  BOUND_LOWER,
+  BOUND_EXACT = BOUND_UPPER | BOUND_LOWER
+};
+
+enum Value : int {
+  VALUE_ZERO      = 0,
+  VALUE_DRAW      = 0,
+  VALUE_KNOWN_WIN = 10000,
+  VALUE_MATE      = 32000,
+  VALUE_INFINITE  = 32001,
+  VALUE_NONE      = 32002,
+
+  VALUE_TB_WIN_IN_MAX_PLY  =  VALUE_MATE - 2 * MAX_PLY,
+  VALUE_TB_LOSS_IN_MAX_PLY = -VALUE_TB_WIN_IN_MAX_PLY,
+  VALUE_MATE_IN_MAX_PLY  =  VALUE_MATE - MAX_PLY,
+  VALUE_MATED_IN_MAX_PLY = -VALUE_MATE_IN_MAX_PLY,
+
+  // In the code, we make the assumption that these values
+  // are such that non_pawn_material() can be used to uniquely
+  // identify the material on the board.
+  PawnValueMg   = 126,   PawnValueEg   = 208,
+  KnightValueMg = 781,   KnightValueEg = 854,
+  BishopValueMg = 825,   BishopValueEg = 915,
+  RookValueMg   = 1276,  RookValueEg   = 1380,
+  QueenValueMg  = 2538,  QueenValueEg  = 2682,
+
+  MidgameLimit  = 15258, EndgameLimit  = 3915
+};
+
+enum PieceType {
+  NO_PIECE_TYPE, PAWN, KNIGHT, BISHOP, ROOK, QUEEN, KING,
+  ALL_PIECES = 0,
+  PIECE_TYPE_NB = 8
+};
+
+enum Piece {
+  NO_PIECE,
+  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
+};
+
+constexpr Value PieceValue[PHASE_NB][PIECE_NB] = {
+  { VALUE_ZERO, PawnValueMg, KnightValueMg, BishopValueMg, RookValueMg, QueenValueMg, VALUE_ZERO, VALUE_ZERO,
+    VALUE_ZERO, PawnValueMg, KnightValueMg, BishopValueMg, RookValueMg, QueenValueMg, VALUE_ZERO, VALUE_ZERO },
+  { VALUE_ZERO, PawnValueEg, KnightValueEg, BishopValueEg, RookValueEg, QueenValueEg, VALUE_ZERO, VALUE_ZERO,
+    VALUE_ZERO, PawnValueEg, KnightValueEg, BishopValueEg, RookValueEg, QueenValueEg, VALUE_ZERO, VALUE_ZERO }
+};
+
+using Depth = int;
+
+enum : int {
+  DEPTH_QS_CHECKS     =  0,
+  DEPTH_QS_NO_CHECKS  = -1,
+  DEPTH_QS_RECAPTURES = -5,
+
+  DEPTH_NONE   = -6,
+
+  DEPTH_OFFSET = -7 // value used only for TT entry occupancy check
+};
+
+enum Square : int {
   SQ_A1, SQ_B1, SQ_C1, SQ_D1, SQ_E1, SQ_F1, SQ_G1, SQ_H1,
   SQ_A2, SQ_B2, SQ_C2, SQ_D2, SQ_E2, SQ_F2, SQ_G2, SQ_H2,
   SQ_A3, SQ_B3, SQ_C3, SQ_D3, SQ_E3, SQ_F3, SQ_G3, SQ_H3,
@@ -215,279 +241,251 @@ enum Square {
   SQ_A8, SQ_B8, SQ_C8, SQ_D8, SQ_E8, SQ_F8, SQ_G8, SQ_H8,
   SQ_NONE,
 
-  DELTA_N =  8,
-  DELTA_E =  1,
-  DELTA_S = -8,
-  DELTA_W = -1,
-
-  DELTA_NN = DELTA_N + DELTA_N,
-  DELTA_NE = DELTA_N + DELTA_E,
-  DELTA_SE = DELTA_S + DELTA_E,
-  DELTA_SS = DELTA_S + DELTA_S,
-  DELTA_SW = DELTA_S + DELTA_W,
-  DELTA_NW = DELTA_N + DELTA_W
+  SQUARE_ZERO = 0,
+  SQUARE_NB   = 64
 };
 
-enum File {
-  FILE_A, FILE_B, FILE_C, FILE_D, FILE_E, FILE_F, FILE_G, FILE_H
+enum Direction : int {
+  NORTH =  8,
+  EAST  =  1,
+  SOUTH = -NORTH,
+  WEST  = -EAST,
+
+  NORTH_EAST = NORTH + EAST,
+  SOUTH_EAST = SOUTH + EAST,
+  SOUTH_WEST = SOUTH + WEST,
+  NORTH_WEST = NORTH + WEST
 };
 
-enum Rank {
-  RANK_1, RANK_2, RANK_3, RANK_4, RANK_5, RANK_6, RANK_7, RANK_8
+enum File : int {
+  FILE_A, FILE_B, FILE_C, FILE_D, FILE_E, FILE_F, FILE_G, FILE_H, FILE_NB
 };
 
-
-/// Score enum keeps a midgame and an endgame value in a single integer (enum),
-/// first LSB 16 bits are used to store endgame value, while upper bits are used
-/// for midgame value. Compiler is free to choose the enum type as long as can
-/// keep its data, so ensure Score to be an integer type.
-enum Score {
-  SCORE_ZERO = 0,
-  SCORE_ENSURE_INTEGER_SIZE_P = INT_MAX,
-  SCORE_ENSURE_INTEGER_SIZE_N = INT_MIN
+enum Rank : int {
+  RANK_1, RANK_2, RANK_3, RANK_4, RANK_5, RANK_6, RANK_7, RANK_8, RANK_NB
 };
 
-inline Score make_score(int mg, int eg) { return Score((mg << 16) + eg); }
+// Keep track of what a move changes on the board (used by NNUE)
+struct DirtyPiece {
 
-/// Extracting the signed lower and upper 16 bits it not so trivial because
-/// according to the standard a simple cast to short is implementation defined
-/// and so is a right shift of a signed integer.
-inline Value mg_value(Score s) { return Value(((s + 32768) & ~0xffff) / 0x10000); }
+  // Number of changed pieces
+  int dirty_num;
 
-/// On Intel 64 bit we have a small speed regression with the standard conforming
-/// version, so use a faster code in this case that, although not 100% standard
-/// compliant it seems to work for Intel and MSVC.
-#if defined(IS_64BIT) && (!defined(__GNUC__) || defined(__INTEL_COMPILER))
+  // Max 3 pieces can change in one move. A promotion with capture moves
+  // both the pawn and the captured piece to SQ_NONE and the piece promoted
+  // to from SQ_NONE to the capture square.
+  Piece piece[3];
 
-inline Value eg_value(Score s) { return Value(int16_t(s & 0xffff)); }
+  // From and to squares, which may be SQ_NONE
+  Square from[3];
+  Square to[3];
+};
 
-#else
+/// Score enum stores a middlegame and an endgame value in a single integer (enum).
+/// The least significant 16 bits are used to store the middlegame value and the
+/// upper 16 bits are used to store the endgame value. We have to take care to
+/// avoid left-shifting a signed int to avoid undefined behavior.
+enum Score : int { SCORE_ZERO };
 
-inline Value eg_value(Score s) {
-  return Value((int)(unsigned(s) & 0x7fffu) - (int)(unsigned(s) & 0x8000u));
+constexpr Score make_score(int mg, int eg) {
+  return Score((int)((unsigned int)eg << 16) + mg);
 }
 
-#endif
+/// Extracting the signed lower and upper 16 bits is not so trivial because
+/// according to the standard a simple cast to short is implementation defined
+/// and so is a right shift of a signed integer.
+inline Value eg_value(Score s) {
+  union { uint16_t u; int16_t s; } eg = { uint16_t(unsigned(s + 0x8000) >> 16) };
+  return Value(eg.s);
+}
 
-#define ENABLE_SAFE_OPERATORS_ON(T)                                         \
-inline T operator+(const T d1, const T d2) { return T(int(d1) + int(d2)); } \
-inline T operator-(const T d1, const T d2) { return T(int(d1) - int(d2)); } \
-inline T operator*(int i, const T d) { return T(i * int(d)); }              \
-inline T operator*(const T d, int i) { return T(int(d) * i); }              \
-inline T operator-(const T d) { return T(-int(d)); }                        \
-inline T& operator+=(T& d1, const T d2) { d1 = d1 + d2; return d1; }        \
-inline T& operator-=(T& d1, const T d2) { d1 = d1 - d2; return d1; }        \
-inline T& operator*=(T& d, int i) { d = T(int(d) * i); return d; }
+inline Value mg_value(Score s) {
+  union { uint16_t u; int16_t s; } mg = { uint16_t(unsigned(s)) };
+  return Value(mg.s);
+}
 
-#define ENABLE_OPERATORS_ON(T) ENABLE_SAFE_OPERATORS_ON(T)                  \
-inline T operator++(T& d, int) { d = T(int(d) + 1); return d; }             \
-inline T operator--(T& d, int) { d = T(int(d) - 1); return d; }             \
-inline T operator/(const T d, int i) { return T(int(d) / i); }              \
-inline T& operator/=(T& d, int i) { d = T(int(d) / i); return d; }
+#define ENABLE_BASE_OPERATORS_ON(T)                                \
+constexpr T operator+(T d1, int d2) { return T(int(d1) + d2); }    \
+constexpr T operator-(T d1, int d2) { return T(int(d1) - d2); }    \
+constexpr T operator-(T d) { return T(-int(d)); }                  \
+inline T& operator+=(T& d1, int d2) { return d1 = d1 + d2; }       \
+inline T& operator-=(T& d1, int d2) { return d1 = d1 - d2; }
 
-ENABLE_OPERATORS_ON(Value)
-ENABLE_OPERATORS_ON(PieceType)
-ENABLE_OPERATORS_ON(Piece)
-ENABLE_OPERATORS_ON(Color)
-ENABLE_OPERATORS_ON(Depth)
-ENABLE_OPERATORS_ON(Square)
-ENABLE_OPERATORS_ON(File)
-ENABLE_OPERATORS_ON(Rank)
+#define ENABLE_INCR_OPERATORS_ON(T)                                \
+inline T& operator++(T& d) { return d = T(int(d) + 1); }           \
+inline T& operator--(T& d) { return d = T(int(d) - 1); }
 
-/// Added operators for adding integers to a Value
-inline Value operator+(Value v, int i) { return Value(int(v) + i); }
-inline Value operator-(Value v, int i) { return Value(int(v) - i); }
+#define ENABLE_FULL_OPERATORS_ON(T)                                \
+ENABLE_BASE_OPERATORS_ON(T)                                        \
+constexpr T operator*(int i, T d) { return T(i * int(d)); }        \
+constexpr T operator*(T d, int i) { return T(int(d) * i); }        \
+constexpr T operator/(T d, int i) { return T(int(d) / i); }        \
+constexpr int operator/(T d1, T d2) { return int(d1) / int(d2); }  \
+inline T& operator*=(T& d, int i) { return d = T(int(d) * i); }    \
+inline T& operator/=(T& d, int i) { return d = T(int(d) / i); }
 
-ENABLE_SAFE_OPERATORS_ON(Score)
+ENABLE_FULL_OPERATORS_ON(Value)
+ENABLE_FULL_OPERATORS_ON(Direction)
+
+ENABLE_INCR_OPERATORS_ON(Piece)
+ENABLE_INCR_OPERATORS_ON(PieceType)
+ENABLE_INCR_OPERATORS_ON(Square)
+ENABLE_INCR_OPERATORS_ON(File)
+ENABLE_INCR_OPERATORS_ON(Rank)
+
+ENABLE_BASE_OPERATORS_ON(Score)
+
+#undef ENABLE_FULL_OPERATORS_ON
+#undef ENABLE_INCR_OPERATORS_ON
+#undef ENABLE_BASE_OPERATORS_ON
+
+/// Additional operators to add a Direction to a Square
+constexpr Square operator+(Square s, Direction d) { return Square(int(s) + int(d)); }
+constexpr Square operator-(Square s, Direction d) { return Square(int(s) - int(d)); }
+inline Square& operator+=(Square& s, Direction d) { return s = s + d; }
+inline Square& operator-=(Square& s, Direction d) { return s = s - d; }
 
 /// Only declared but not defined. We don't want to multiply two scores due to
 /// a very high risk of overflow. So user should explicitly convert to integer.
-inline Score operator*(Score s1, Score s2);
+Score operator*(Score, Score) = delete;
 
 /// Division of a Score must be handled separately for each term
 inline Score operator/(Score s, int i) {
   return make_score(mg_value(s) / i, eg_value(s) / i);
 }
 
-/// Weight score v by score w trying to prevent overflow
-inline Score apply_weight(Score v, Score w) {
-  return make_score((int(mg_value(v)) * mg_value(w)) / 0x100,
-                    (int(eg_value(v)) * eg_value(w)) / 0x100);
+/// Multiplication of a Score by an integer. We check for overflow in debug mode.
+inline Score operator*(Score s, int i) {
+
+  Score result = Score(int(s) * i);
+
+  assert(eg_value(result) == (i * eg_value(s)));
+  assert(mg_value(result) == (i * mg_value(s)));
+  assert((i == 0) || (result / i) == s);
+
+  return result;
 }
 
-#undef ENABLE_OPERATORS_ON
-#undef ENABLE_SAFE_OPERATORS_ON
-
-namespace Zobrist {
-
-  extern Key psq[2][8][64]; // [color][pieceType][square / piece count]
-  extern Key enpassant[8];  // [file]
-  extern Key castle[16];    // [castleRight]
-  extern Key side;
-  extern Key exclusion;
-
-  void init();
+/// Multiplication of a Score by a boolean
+inline Score operator*(Score s, bool b) {
+  return b ? s : SCORE_ZERO;
 }
 
-CACHE_LINE_ALIGNMENT
-
-extern Score pieceSquareTable[16][64]; // [piece][square]
-extern Value PieceValue[2][18];        // [Mg / Eg][piece / pieceType]
-extern int SquareDistance[64][64];     // [square][square]
-
-struct MoveStack {
-  Move move;
-  int score;
-};
-
-inline bool operator<(const MoveStack& f, const MoveStack& s) {
-  return f.score < s.score;
+constexpr Color operator~(Color c) {
+  return Color(c ^ BLACK); // Toggle color
 }
 
-inline Color operator~(Color c) {
-  return Color(c ^ 1);
+constexpr Square flip_rank(Square s) { // Swap A1 <-> A8
+  return Square(s ^ SQ_A8);
 }
 
-inline Square operator~(Square s) {
-  return Square(s ^ 56); // Vertical flip SQ_A1 -> SQ_A8
+constexpr Square flip_file(Square s) { // Swap A1 <-> H1
+  return Square(s ^ SQ_H1);
 }
 
-inline Square operator|(File f, Rank r) {
-  return Square((r << 3) | f);
+constexpr Piece operator~(Piece pc) {
+  return Piece(pc ^ 8); // Swap color of piece B_KNIGHT <-> W_KNIGHT
 }
 
-inline Value mate_in(int ply) {
+constexpr CastlingRights operator&(Color c, CastlingRights cr) {
+  return CastlingRights((c == WHITE ? WHITE_CASTLING : BLACK_CASTLING) & cr);
+}
+
+constexpr Value mate_in(int ply) {
   return VALUE_MATE - ply;
 }
 
-inline Value mated_in(int ply) {
+constexpr Value mated_in(int ply) {
   return -VALUE_MATE + ply;
 }
 
-inline Piece make_piece(Color c, PieceType pt) {
-  return Piece((c << 3) | pt);
+constexpr Square make_square(File f, Rank r) {
+  return Square((r << 3) + f);
 }
 
-inline CastleRight make_castle_right(Color c, CastlingSide s) {
-  return CastleRight(WHITE_OO << ((s == QUEEN_SIDE) + 2 * c));
+constexpr Piece make_piece(Color c, PieceType pt) {
+  return Piece((c << 3) + pt);
 }
 
-inline PieceType type_of(Piece p)  {
-  return PieceType(p & 7);
+constexpr PieceType type_of(Piece pc) {
+  return PieceType(pc & 7);
 }
 
-inline Color color_of(Piece p) {
-  return Color(p >> 3);
+inline Color color_of(Piece pc) {
+  assert(pc != NO_PIECE);
+  return Color(pc >> 3);
 }
 
-inline bool is_ok(Square s) {
+constexpr bool is_ok(Move m) {
+  return m != MOVE_NONE && m != MOVE_NULL;
+}
+
+constexpr bool is_ok(Square s) {
   return s >= SQ_A1 && s <= SQ_H8;
 }
 
-inline File file_of(Square s) {
+constexpr File file_of(Square s) {
   return File(s & 7);
 }
 
-inline Rank rank_of(Square s) {
+constexpr Rank rank_of(Square s) {
   return Rank(s >> 3);
 }
 
-inline Square mirror(Square s) {
-  return Square(s ^ 7); // Horizontal flip SQ_A1 -> SQ_H1
-}
-
-inline Square relative_square(Color c, Square s) {
+constexpr Square relative_square(Color c, Square s) {
   return Square(s ^ (c * 56));
 }
 
-inline Rank relative_rank(Color c, Rank r) {
+constexpr Rank relative_rank(Color c, Rank r) {
   return Rank(r ^ (c * 7));
 }
 
-inline Rank relative_rank(Color c, Square s) {
+constexpr Rank relative_rank(Color c, Square s) {
   return relative_rank(c, rank_of(s));
 }
 
-inline bool opposite_colors(Square s1, Square s2) {
-  int s = int(s1) ^ int(s2);
-  return ((s >> 3) ^ s) & 1;
+constexpr Direction pawn_push(Color c) {
+  return c == WHITE ? NORTH : SOUTH;
 }
 
-inline int file_distance(Square s1, Square s2) {
-  return abs(file_of(s1) - file_of(s2));
-}
-
-inline int rank_distance(Square s1, Square s2) {
-  return abs(rank_of(s1) - rank_of(s2));
-}
-
-inline int square_distance(Square s1, Square s2) {
-  return SquareDistance[s1][s2];
-}
-
-inline char file_to_char(File f) {
-  return char(f - FILE_A + int('a'));
-}
-
-inline char rank_to_char(Rank r) {
-  return char(r - RANK_1 + int('1'));
-}
-
-inline Square pawn_push(Color c) {
-  return c == WHITE ? DELTA_N : DELTA_S;
-}
-
-inline Square from_sq(Move m) {
+constexpr Square from_sq(Move m) {
+  assert(is_ok(m));
   return Square((m >> 6) & 0x3F);
 }
 
-inline Square to_sq(Move m) {
+constexpr Square to_sq(Move m) {
+  assert(is_ok(m));
   return Square(m & 0x3F);
 }
 
-inline MoveType type_of(Move m) {
+constexpr int from_to(Move m) {
+  return m & 0xFFF;
+}
+
+constexpr MoveType type_of(Move m) {
   return MoveType(m & (3 << 14));
 }
 
-inline PieceType promotion_type(Move m) {
-  return PieceType(((m >> 12) & 3) + 2);
+constexpr PieceType promotion_type(Move m) {
+  return PieceType(((m >> 12) & 3) + KNIGHT);
 }
 
-inline Move make_move(Square from, Square to) {
-  return Move(to | (from << 6));
+constexpr Move make_move(Square from, Square to) {
+  return Move((from << 6) + to);
 }
 
 template<MoveType T>
-inline Move make(Square from, Square to, PieceType pt = KNIGHT) {
-  return Move(to | (from << 6) | T | ((pt - KNIGHT) << 12)) ;
+constexpr Move make(Square from, Square to, PieceType pt = KNIGHT) {
+  return Move(T + ((pt - KNIGHT) << 12) + (from << 6) + to);
 }
 
-inline bool is_ok(Move m) {
-  return from_sq(m) != to_sq(m); // Catches also MOVE_NULL and MOVE_NONE
+/// Based on a congruential pseudo random number generator
+constexpr Key make_key(uint64_t seed) {
+  return seed * 6364136223846793005ULL + 1442695040888963407ULL;
 }
 
-#include <string>
+} // namespace Stockfish
 
-inline const std::string square_to_string(Square s) {
-  char ch[] = { file_to_char(file_of(s)), rank_to_char(rank_of(s)), 0 };
-  return ch;
-}
+#endif // #ifndef TYPES_H_INCLUDED
 
-/// Our insertion sort implementation, works with pointers and iterators and is
-/// guaranteed to be stable, as is needed.
-template<typename T, typename K>
-void sort(K first, K last)
-{
-  T tmp;
-  K p, q;
-
-  for (p = first + 1; p < last; p++)
-  {
-      tmp = *p;
-      for (q = p; q != first && *(q-1) < tmp; --q)
-          *q = *(q-1);
-      *q = tmp;
-  }
-}
-
-#endif // !defined(TYPES_H_INCLUDED)
+#include "tune.h" // Global visibility to tuning setup

src/uci.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -17,146 +16,40 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#include <cassert>
+#include <cmath>
 #include <iostream>
 #include <sstream>
 #include <string>
 
+#include "benchmark.h"
 #include "evaluate.h"
-#include "notation.h"
+#include "movegen.h"
 #include "position.h"
 #include "search.h"
 #include "thread.h"
+#include "timeman.h"
 #include "tt.h"
-#include "ucioption.h"
+#include "uci.h"
+#include "syzygy/tbprobe.h"
+#include "nnue/evaluate_nnue.h"
 
 using namespace std;
 
-extern void benchmark(const Position& pos, istream& is);
+namespace Stockfish {
 
 namespace {
 
-  // FEN string of the initial position, normal chess
+  // FEN string for the initial position in standard chess
   const char* StartFEN = "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1";
 
-  // Keep track of position keys along the setup moves (from start position to the
-  // position just before to start searching). Needed by repetition draw detection.
-  Search::StateStackPtr SetupStates;
 
-  void set_option(istringstream& up);
-  void set_position(Position& pos, istringstream& up);
-  void go(Position& pos, istringstream& up);
-}
+  // position() is called when the engine receives the "position" UCI command.
+  // It sets up the position that is described in the given FEN string ("fen") or
+  // the initial position ("startpos") and then makes the moves given in the following
+  // move list ("moves").
 
-
-/// Wait for a command from the user, parse this text string as an UCI command,
-/// and call the appropriate functions. Also intercepts EOF from stdin to ensure
-/// that we exit gracefully if the GUI dies unexpectedly. In addition to the UCI
-/// commands, the function also supports a few debug commands.
-
-void UCI::loop(const string& args) {
-
-  Position pos(StartFEN, false, Threads.main_thread()); // The root position
-  string cmd, token;
-
-  while (token != "quit")
-  {
-      if (!args.empty())
-          cmd = args;
-
-      else if (!getline(cin, cmd)) // Block here waiting for input
-          cmd = "quit";
-
-      istringstream is(cmd);
-
-      is >> skipws >> token;
-
-      if (token == "quit" || token == "stop")
-      {
-          Search::Signals.stop = true;
-          Threads.wait_for_search_finished(); // Cannot quit while threads are running
-      }
-
-      else if (token == "ponderhit")
-      {
-          // The opponent has played the expected move. GUI sends "ponderhit" if
-          // we were told to ponder on the same move the opponent has played. We
-          // should continue searching but switching from pondering to normal search.
-          Search::Limits.ponder = false;
-
-          if (Search::Signals.stopOnPonderhit)
-          {
-              Search::Signals.stop = true;
-              Threads.main_thread()->wake_up(); // Could be sleeping
-          }
-      }
-
-      else if (token == "go")
-          go(pos, is);
-
-      else if (token == "ucinewgame")
-          TT.clear();
-
-      else if (token == "isready")
-          sync_cout << "readyok" << sync_endl;
-
-      else if (token == "position")
-          set_position(pos, is);
-
-      else if (token == "setoption")
-          set_option(is);
-
-      else if (token == "d")
-          pos.print();
-
-      else if (token == "flip")
-          pos.flip();
-
-      else if (token == "eval")
-          sync_cout << Eval::trace(pos) << sync_endl;
-
-      else if (token == "bench")
-          benchmark(pos, is);
-
-      else if (token == "key")
-          sync_cout << "key: " << hex     << pos.key()
-                    << "\nmaterial key: " << pos.material_key()
-                    << "\npawn key: "     << pos.pawn_key() << sync_endl;
-
-      else if (token == "uci")
-          sync_cout << "id name " << engine_info(true)
-                    << "\n"       << Options
-                    << "\nuciok"  << sync_endl;
-
-      else if (token == "perft" && (is >> token)) // Read depth
-      {
-          stringstream ss;
-
-          ss << Options["Hash"]    << " "
-             << Options["Threads"] << " " << token << " current perft";
-
-          benchmark(pos, ss);
-      }
-
-      else
-          sync_cout << "Unknown command: " << cmd << sync_endl;
-
-      if (!args.empty()) // Command line arguments have one-shot behaviour
-      {
-          Threads.wait_for_search_finished();
-          break;
-      }
-  }
-}
-
-
-namespace {
-
-  // set_position() is called when engine receives the "position" UCI command.
-  // The function sets up the position described in the given fen string ("fen")
-  // or the starting position ("startpos") and then makes the moves given in the
-  // following move list ("moves").
-
-  void set_position(Position& pos, istringstream& is) {
+  void position(Position& pos, istringstream& is, StateListPtr& states) {
 
     Move m;
     string token, fen;
@@ -166,7 +59,7 @@ namespace {
     if (token == "startpos")
     {
         fen = StartFEN;
-        is >> token; // Consume "moves" token if any
+        is >> token; // Consume the "moves" token, if any
     }
     else if (token == "fen")
         while (is >> token && token != "moves")
@@ -174,34 +67,48 @@ namespace {
     else
         return;
 
-    pos.from_fen(fen, Options["UCI_Chess960"], Threads.main_thread());
-    SetupStates = Search::StateStackPtr(new std::stack<StateInfo>());
+    states = StateListPtr(new std::deque<StateInfo>(1)); // Drop the old state and create a new one
+    pos.set(fen, Options["UCI_Chess960"], &states->back(), Threads.main());
 
-    // Parse move list (if any)
-    while (is >> token && (m = move_from_uci(pos, token)) != MOVE_NONE)
+    // Parse the move list, if any
+    while (is >> token && (m = UCI::to_move(pos, token)) != MOVE_NONE)
     {
-        SetupStates->push(StateInfo());
-        pos.do_move(m, SetupStates->top());
+        states->emplace_back();
+        pos.do_move(m, states->back());
     }
   }
 
+  // trace_eval() prints the evaluation of the current position, consistent with
+  // the UCI options set so far.
 
-  // set_option() is called when engine receives the "setoption" UCI command. The
-  // function updates the UCI option ("name") to the given value ("value").
+  void trace_eval(Position& pos) {
 
-  void set_option(istringstream& is) {
+    StateListPtr states(new std::deque<StateInfo>(1));
+    Position p;
+    p.set(pos.fen(), Options["UCI_Chess960"], &states->back(), Threads.main());
+
+    Eval::NNUE::verify();
+
+    sync_cout << "\n" << Eval::trace(p) << sync_endl;
+  }
+
+
+  // setoption() is called when the engine receives the "setoption" UCI command.
+  // The function updates the UCI option ("name") to the given value ("value").
+
+  void setoption(istringstream& is) {
 
     string token, name, value;
 
-    is >> token; // Consume "name" token
+    is >> token; // Consume the "name" token
 
-    // Read option name (can contain spaces)
+    // Read the option name (can contain spaces)
     while (is >> token && token != "value")
-        name += string(" ", !name.empty()) + token;
+        name += (name.empty() ? "" : " ") + token;
 
-    // Read option value (can contain spaces)
+    // Read the option value (can contain spaces)
     while (is >> token)
-        value += string(" ", !value.empty()) + token;
+        value += (value.empty() ? "" : " ") + token;
 
     if (Options.count(name))
         Options[name] = value;
@@ -210,43 +117,282 @@ namespace {
   }
 
 
-  // go() is called when engine receives the "go" UCI command. The function sets
-  // the thinking time and other parameters from the input string, and then starts
-  // the search.
+  // go() is called when the engine receives the "go" UCI command. The function
+  // sets the thinking time and other parameters from the input string, then starts
+  // with a search.
 
-  void go(Position& pos, istringstream& is) {
+  void go(Position& pos, istringstream& is, StateListPtr& states) {
 
     Search::LimitsType limits;
-    vector<Move> searchMoves;
     string token;
+    bool ponderMode = false;
+
+    limits.startTime = now(); // The search starts as early as possible
 
     while (is >> token)
-    {
-        if (token == "wtime")
-            is >> limits.time[WHITE];
-        else if (token == "btime")
-            is >> limits.time[BLACK];
-        else if (token == "winc")
-            is >> limits.inc[WHITE];
-        else if (token == "binc")
-            is >> limits.inc[BLACK];
-        else if (token == "movestogo")
-            is >> limits.movestogo;
-        else if (token == "depth")
-            is >> limits.depth;
-        else if (token == "nodes")
-            is >> limits.nodes;
-        else if (token == "movetime")
-            is >> limits.movetime;
-        else if (token == "infinite")
-            limits.infinite = true;
-        else if (token == "ponder")
-            limits.ponder = true;
-        else if (token == "searchmoves")
+        if (token == "searchmoves") // Needs to be the last command on the line
             while (is >> token)
-                searchMoves.push_back(move_from_uci(pos, token));
+                limits.searchmoves.push_back(UCI::to_move(pos, token));
+
+        else if (token == "wtime")     is >> limits.time[WHITE];
+        else if (token == "btime")     is >> limits.time[BLACK];
+        else if (token == "winc")      is >> limits.inc[WHITE];
+        else if (token == "binc")      is >> limits.inc[BLACK];
+        else if (token == "movestogo") is >> limits.movestogo;
+        else if (token == "depth")     is >> limits.depth;
+        else if (token == "nodes")     is >> limits.nodes;
+        else if (token == "movetime")  is >> limits.movetime;
+        else if (token == "mate")      is >> limits.mate;
+        else if (token == "perft")     is >> limits.perft;
+        else if (token == "infinite")  limits.infinite = 1;
+        else if (token == "ponder")    ponderMode = true;
+
+    Threads.start_thinking(pos, states, limits, ponderMode);
+  }
+
+
+  // bench() is called when the engine receives the "bench" command.
+  // Firstly, a list of UCI commands is set up according to the bench
+  // parameters, then it is run one by one, printing a summary at the end.
+
+  void bench(Position& pos, istream& args, StateListPtr& states) {
+
+    string token;
+    uint64_t num, nodes = 0, cnt = 1;
+
+    vector<string> list = setup_bench(pos, args);
+    num = count_if(list.begin(), list.end(), [](const string& s) { return s.find("go ") == 0 || s.find("eval") == 0; });
+
+    TimePoint elapsed = now();
+
+    for (const auto& cmd : list)
+    {
+        istringstream is(cmd);
+        is >> skipws >> token;
+
+        if (token == "go" || token == "eval")
+        {
+            cerr << "\nPosition: " << cnt++ << '/' << num << " (" << pos.fen() << ")" << endl;
+            if (token == "go")
+            {
+               go(pos, is, states);
+               Threads.main()->wait_for_search_finished();
+               nodes += Threads.nodes_searched();
+            }
+            else
+               trace_eval(pos);
+        }
+        else if (token == "setoption")  setoption(is);
+        else if (token == "position")   position(pos, is, states);
+        else if (token == "ucinewgame") { Search::clear(); elapsed = now(); } // Search::clear() may take a while
     }
 
-    Threads.start_searching(pos, limits, searchMoves, SetupStates);
+    elapsed = now() - elapsed + 1; // Ensure positivity to avoid a 'divide by zero'
+
+    dbg_print();
+
+    cerr << "\n==========================="
+         << "\nTotal time (ms) : " << elapsed
+         << "\nNodes searched  : " << nodes
+         << "\nNodes/second    : " << 1000 * nodes / elapsed << endl;
   }
+
+  // The win rate model returns the probability of winning (in per mille units) given an
+  // eval and a game ply. It fits the LTC fishtest statistics rather accurately.
+  int win_rate_model(Value v, int ply) {
+
+     // The model only captures up to 240 plies, so limit the input and then rescale
+     double m = std::min(240, ply) / 64.0;
+
+     // The coefficients of a third-order polynomial fit is based on the fishtest data
+     // for two parameters that need to transform eval to the argument of a logistic
+     // function.
+     constexpr double as[] = {   0.38036525,   -2.82015070,   23.17882135,  307.36768407};
+     constexpr double bs[] = {  -2.29434733,   13.27689788,  -14.26828904,   63.45318330 };
+
+     // Enforce that NormalizeToPawnValue corresponds to a 50% win rate at ply 64
+     static_assert(UCI::NormalizeToPawnValue == int(as[0] + as[1] + as[2] + as[3]));
+
+     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 the eval to centipawns with limited range
+     double x = std::clamp(double(v), -4000.0, 4000.0);
+
+     // Return the win rate in per mille units rounded to the nearest value
+     return int(0.5 + 1000 / (1 + std::exp((a - x) / b)));
+  }
+
+} // namespace
+
+
+/// UCI::loop() waits for a command from the stdin, parses it and then calls the appropriate
+/// function. It also intercepts an end-of-file (EOF) indication from the stdin to ensure a
+/// graceful exit if the GUI dies unexpectedly. When called with some command-line arguments,
+/// like running 'bench', the function returns immediately after the command is executed.
+/// In addition to the UCI ones, some additional debug commands are also supported.
+
+void UCI::loop(int argc, char* argv[]) {
+
+  Position pos;
+  string token, cmd;
+  StateListPtr states(new std::deque<StateInfo>(1));
+
+  pos.set(StartFEN, false, &states->back(), Threads.main());
+
+  for (int i = 1; i < argc; ++i)
+      cmd += std::string(argv[i]) + " ";
+
+  do {
+      if (argc == 1 && !getline(cin, cmd)) // Wait for an input or an end-of-file (EOF) indication
+          cmd = "quit";
+
+      istringstream is(cmd);
+
+      token.clear(); // Avoid a stale if getline() returns nothing or a blank line
+      is >> skipws >> token;
+
+      if (    token == "quit"
+          ||  token == "stop")
+          Threads.stop = true;
+
+      // The GUI sends 'ponderhit' to tell that the user has played the expected move.
+      // So, 'ponderhit' is sent if pondering was done on the same move that the user
+      // has played. The search should continue, but should also switch from pondering
+      // to the normal search.
+      else if (token == "ponderhit")
+          Threads.main()->ponder = false; // Switch to the normal search
+
+      else if (token == "uci")
+          sync_cout << "id name " << engine_info(true)
+                    << "\n"       << Options
+                    << "\nuciok"  << sync_endl;
+
+      else if (token == "setoption")  setoption(is);
+      else if (token == "go")         go(pos, is, states);
+      else if (token == "position")   position(pos, is, states);
+      else if (token == "ucinewgame") Search::clear();
+      else if (token == "isready")    sync_cout << "readyok" << sync_endl;
+
+      // Add custom non-UCI commands, mainly for debugging purposes.
+      // These commands must not be used during a search!
+      else if (token == "flip")     pos.flip();
+      else if (token == "bench")    bench(pos, is, states);
+      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 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 == "--help" || token == "help" || token == "--license" || token == "license")
+          sync_cout << "\nStockfish is a powerful chess engine for playing and analyzing."
+                       "\nIt is released as free software licensed under the GNU GPLv3 License."
+                       "\nStockfish is normally used with a graphical user interface (GUI) and implements"
+                       "\nthe Universal Chess Interface (UCI) protocol to communicate with a GUI, an API, etc."
+                       "\nFor any further information, visit https://github.com/official-stockfish/Stockfish#readme"
+                       "\nor read the corresponding README.md and Copying.txt files distributed along with this program.\n" << sync_endl;
+      else if (!token.empty() && token[0] != '#')
+          sync_cout << "Unknown command: '" << cmd << "'. Type help for more information." << sync_endl;
+
+  } while (token != "quit" && argc == 1); // The command-line arguments are one-shot
 }
+
+
+/// UCI::value() converts a Value to a string by adhering to the UCI protocol specification:
+///
+/// cp <x>    The score from the engine's point of view in centipawns.
+/// mate <y>  Mate in 'y' moves (not plies). If the engine is getting mated,
+///           uses negative values for 'y'.
+
+string UCI::value(Value v) {
+
+  assert(-VALUE_INFINITE < v && v < VALUE_INFINITE);
+
+  stringstream ss;
+
+  if (abs(v) < VALUE_TB_WIN_IN_MAX_PLY)
+      ss << "cp " << v * 100 / NormalizeToPawnValue;
+  else if (abs(v) < VALUE_MATE_IN_MAX_PLY)
+  {
+      const int ply = VALUE_MATE_IN_MAX_PLY - 1 - std::abs(v);  // recompute ss->ply
+      ss << "cp " << (v > 0 ? 20000 - ply : -20000 + ply);
+  }
+  else
+      ss << "mate " << (v > 0 ? VALUE_MATE - v + 1 : -VALUE_MATE - v) / 2;
+
+  return ss.str();
+}
+
+
+/// UCI::wdl() reports the win-draw-loss (WDL) statistics given an evaluation
+/// and a game ply based on the data gathered for fishtest LTC games.
+
+string UCI::wdl(Value v, int ply) {
+
+  stringstream ss;
+
+  int wdl_w = win_rate_model( v, ply);
+  int wdl_l = win_rate_model(-v, ply);
+  int wdl_d = 1000 - wdl_w - wdl_l;
+  ss << " wdl " << wdl_w << " " << wdl_d << " " << wdl_l;
+
+  return ss.str();
+}
+
+
+/// UCI::square() converts a Square to a string in algebraic notation (g1, a7, etc.)
+
+std::string UCI::square(Square s) {
+  return std::string{ char('a' + file_of(s)), char('1' + rank_of(s)) };
+}
+
+
+/// UCI::move() converts a Move to a string in coordinate notation (g1f3, a7a8q).
+/// The only special case is castling where the e1g1 notation is printed in
+/// standard chess mode and in e1h1 notation it is printed in Chess960 mode.
+/// Internally, all castling moves are always encoded as 'king captures rook'.
+
+string UCI::move(Move m, bool chess960) {
+
+  if (m == MOVE_NONE)
+      return "(none)";
+
+  if (m == MOVE_NULL)
+      return "0000";
+
+  Square from = from_sq(m);
+  Square to = to_sq(m);
+
+  if (type_of(m) == CASTLING && !chess960)
+      to = make_square(to > from ? FILE_G : FILE_C, rank_of(from));
+
+  string move = UCI::square(from) + UCI::square(to);
+
+  if (type_of(m) == PROMOTION)
+      move += " pnbrqk"[promotion_type(m)];
+
+  return move;
+}
+
+
+/// UCI::to_move() converts a string representing a move in coordinate notation
+/// (g1f3, a7a8q) to the corresponding legal Move, if any.
+
+Move UCI::to_move(const Position& pos, string& str) {
+
+  if (str.length() == 5)
+      str[4] = char(tolower(str[4])); // The promotion piece character must be lowercased
+
+  for (const auto& m : MoveList<LEGAL>(pos))
+      if (str == UCI::move(m, pos.is_chess960()))
+          return m;
+
+  return MOVE_NONE;
+}
+
+} // namespace Stockfish

src/uci.h

@@ -0,0 +1,92 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+
+  Stockfish is free software: you can redistribute it and/or modify
+  it under the terms of the GNU General Public License as published by
+  the Free Software Foundation, either version 3 of the License, or
+  (at your option) any later version.
+
+  Stockfish is distributed in the hope that it will be useful,
+  but WITHOUT ANY WARRANTY; without even the implied warranty of
+  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+  GNU General Public License for more details.
+
+  You should have received a copy of the GNU General Public License
+  along with this program.  If not, see <http://www.gnu.org/licenses/>.
+*/
+
+#ifndef UCI_H_INCLUDED
+#define UCI_H_INCLUDED
+
+#include <map>
+#include <string>
+
+#include "types.h"
+
+namespace Stockfish {
+
+class Position;
+
+namespace UCI {
+
+// Normalizes the internal value as reported by evaluate or search
+// to the UCI centipawn result used in output. This value is derived from
+// the win_rate_model() such that Stockfish outputs an advantage of
+// "100 centipawns" for a position if the engine has a 50% probability to win
+// from this position in selfplay at fishtest LTC time control.
+const int NormalizeToPawnValue = 328;
+
+class Option;
+
+/// Define a custom comparator, because the UCI options should be case-insensitive
+struct CaseInsensitiveLess {
+  bool operator() (const std::string&, const std::string&) const;
+};
+
+/// The options container is defined as a std::map
+using OptionsMap = std::map<std::string, Option, CaseInsensitiveLess>;
+
+/// The Option class implements each option as specified by the UCI protocol
+class Option {
+
+  using OnChange = void (*)(const Option&);
+
+public:
+  Option(OnChange = nullptr);
+  Option(bool v, OnChange = nullptr);
+  Option(const char* v, OnChange = nullptr);
+  Option(double v, int minv, int maxv, OnChange = nullptr);
+  Option(const char* v, const char* cur, OnChange = nullptr);
+
+  Option& operator=(const std::string&);
+  void operator<<(const Option&);
+  operator int() const;
+  operator std::string() const;
+  bool operator==(const char*) const;
+
+private:
+  friend std::ostream& operator<<(std::ostream&, const OptionsMap&);
+
+  std::string defaultValue, currentValue, type;
+  int min, max;
+  size_t idx;
+  OnChange on_change;
+};
+
+void init(OptionsMap&);
+void loop(int argc, char* argv[]);
+std::string value(Value v);
+std::string square(Square s);
+std::string move(Move m, bool chess960);
+std::string pv(const Position& pos, Depth depth);
+std::string wdl(Value v, int ply);
+Move to_move(const Position& pos, std::string& str);
+
+} // namespace UCI
+
+extern UCI::OptionsMap Options;
+
+} // namespace Stockfish
+
+#endif // #ifndef UCI_H_INCLUDED

src/ucioption.cpp

@@ -1,7 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
+  Copyright (C) 2004-2023 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
@@ -19,75 +18,69 @@
 
 #include <algorithm>
 #include <cassert>
-#include <cstdlib>
+#include <ostream>
 #include <sstream>
 
 #include "evaluate.h"
 #include "misc.h"
+#include "search.h"
 #include "thread.h"
 #include "tt.h"
-#include "ucioption.h"
+#include "uci.h"
+#include "syzygy/tbprobe.h"
 
 using std::string;
 
+namespace Stockfish {
+
 UCI::OptionsMap Options; // Global object
 
 namespace UCI {
 
 /// 'On change' actions, triggered by an option's value change
-void on_logger(const Option& o) { start_logger(o); }
-void on_eval(const Option&) { Eval::init(); }
-void on_threads(const Option&) { Threads.read_uci_options(); }
-void on_hash_size(const Option& o) { TT.set_size(o); }
-void on_clear_hash(const Option&) { TT.clear(); }
-
+static void on_clear_hash(const Option&) { Search::clear(); }
+static void on_hash_size(const Option& o) { TT.resize(size_t(o)); }
+static void on_logger(const Option& o) { start_logger(o); }
+static void on_threads(const Option& o) { Threads.set(size_t(o)); }
+static void on_tb_path(const Option& o) { Tablebases::init(o); }
+static void on_use_NNUE(const Option&) { Eval::NNUE::init(); }
+static void on_eval_file(const Option&) { Eval::NNUE::init(); }
 
 /// Our case insensitive less() function as required by UCI protocol
-bool ci_less(char c1, char c2) { return tolower(c1) < tolower(c2); }
-
 bool CaseInsensitiveLess::operator() (const string& s1, const string& s2) const {
-  return std::lexicographical_compare(s1.begin(), s1.end(), s2.begin(), s2.end(), ci_less);
+
+  return std::lexicographical_compare(s1.begin(), s1.end(), s2.begin(), s2.end(),
+         [](char c1, char c2) { return tolower(c1) < tolower(c2); });
 }
 
 
-/// init() initializes the UCI options to their hard coded default values
-/// and initializes the default value of "Threads" and "Min Split Depth"
-/// parameters according to the number of CPU cores detected.
+/// UCI::init() initializes the UCI options to their hard-coded default values
 
 void init(OptionsMap& o) {
 
-  int cpus = std::min(cpu_count(), MAX_THREADS);
-  int msd = cpus < 8 ? 4 : 7;
+  constexpr int MaxHashMB = Is64Bit ? 33554432 : 2048;
 
-  o["Use Debug Log"]               = Option(false, on_logger);
-  o["Use Search Log"]              = Option(false);
-  o["Search Log Filename"]         = Option("SearchLog.txt");
-  o["Book File"]                   = Option("book.bin");
-  o["Best Book Move"]              = Option(false);
-  o["Mobility (Middle Game)"]      = Option(100, 0, 200, on_eval);
-  o["Mobility (Endgame)"]          = Option(100, 0, 200, on_eval);
-  o["Passed Pawns (Middle Game)"]  = Option(100, 0, 200, on_eval);
-  o["Passed Pawns (Endgame)"]      = Option(100, 0, 200, on_eval);
-  o["Space"]                       = Option(100, 0, 200, on_eval);
-  o["Aggressiveness"]              = Option(100, 0, 200, on_eval);
-  o["Cowardice"]                   = Option(100, 0, 200, on_eval);
-  o["Min Split Depth"]             = Option(msd, 4, 7, on_threads);
-  o["Max Threads per Split Point"] = Option(5, 4, 8, on_threads);
-  o["Threads"]                     = Option(cpus, 1, MAX_THREADS, on_threads);
-  o["Use Sleeping Threads"]        = Option(true, on_threads);
-  o["Hash"]                        = Option(32, 4, 8192, on_hash_size);
-  o["Clear Hash"]                  = Option(on_clear_hash);
-  o["Ponder"]                      = Option(true);
-  o["OwnBook"]                     = Option(false);
-  o["MultiPV"]                     = Option(1, 1, 500);
-  o["Skill Level"]                 = Option(20, 0, 20);
-  o["Emergency Move Horizon"]      = Option(40, 0, 50);
-  o["Emergency Base Time"]         = Option(200, 0, 30000);
-  o["Emergency Move Time"]         = Option(70, 0, 5000);
-  o["Minimum Thinking Time"]       = Option(20, 0, 5000);
-  o["Slow Mover"]                  = Option(100, 10, 1000);
-  o["UCI_Chess960"]                = Option(false);
-  o["UCI_AnalyseMode"]             = Option(false, on_eval);
+  o["Debug Log File"]        << Option("", on_logger);
+  o["Threads"]               << Option(1, 1, 1024, on_threads);
+  o["Hash"]                  << Option(16, 1, MaxHashMB, on_hash_size);
+  o["Clear Hash"]            << Option(on_clear_hash);
+  o["Ponder"]                << Option(false);
+  o["MultiPV"]               << Option(1, 1, 500);
+  o["Skill Level"]           << Option(20, 0, 20);
+  o["Move Overhead"]         << Option(10, 0, 5000);
+  o["Slow Mover"]            << Option(100, 10, 1000);
+  o["nodestime"]             << Option(0, 0, 10000);
+  o["UCI_Chess960"]          << Option(false);
+  o["UCI_AnalyseMode"]       << Option(false);
+  o["UCI_LimitStrength"]     << Option(false);
+  o["UCI_Elo"]               << Option(1320, 1320, 3190);
+  o["UCI_ShowWDL"]           << Option(false);
+  o["SyzygyPath"]            << Option("<empty>", on_tb_path);
+  o["SyzygyProbeDepth"]      << Option(1, 1, 100);
+  o["Syzygy50MoveRule"]      << Option(true);
+  o["SyzygyProbeLimit"]      << Option(7, 0, 7);
+  o["Use NNUE"]              << Option(true, on_use_NNUE);
+  o["EvalFile"]              << Option(EvalFileDefaultName, on_eval_file);
 }
 
 
@@ -96,43 +89,48 @@ void init(OptionsMap& o) {
 
 std::ostream& operator<<(std::ostream& os, const OptionsMap& om) {
 
-  for (size_t idx = 0; idx < om.size(); idx++)
-      for (OptionsMap::const_iterator it = om.begin(); it != om.end(); ++it)
-          if (it->second.idx == idx)
+  for (size_t idx = 0; idx < om.size(); ++idx)
+      for (const auto& it : om)
+          if (it.second.idx == idx)
           {
-              const Option& o = it->second;
-              os << "\noption name " << it->first << " type " << o.type;
+              const Option& o = it.second;
+              os << "\noption name " << it.first << " type " << o.type;
 
-              if (o.type != "button")
+              if (o.type == "string" || o.type == "check" || o.type == "combo")
                   os << " default " << o.defaultValue;
 
               if (o.type == "spin")
-                  os << " min " << o.min << " max " << o.max;
+                  os << " default " << int(stof(o.defaultValue))
+                     << " min "     << o.min
+                     << " max "     << o.max;
 
               break;
           }
+
   return os;
 }
 
 
-/// Option c'tors and conversion operators
+/// Option class constructors and conversion operators
 
-Option::Option(const char* v, Fn* f) : type("string"), min(0), max(0), idx(Options.size()), on_change(f)
+Option::Option(const char* v, OnChange f) : type("string"), min(0), max(0), on_change(f)
 { defaultValue = currentValue = v; }
 
-Option::Option(bool v, Fn* f) : type("check"), min(0), max(0), idx(Options.size()), on_change(f)
+Option::Option(bool v, OnChange f) : type("check"), min(0), max(0), on_change(f)
 { defaultValue = currentValue = (v ? "true" : "false"); }
 
-Option::Option(Fn* f) : type("button"), min(0), max(0), idx(Options.size()), on_change(f)
+Option::Option(OnChange f) : type("button"), min(0), max(0), on_change(f)
 {}
 
-Option::Option(int v, int minv, int maxv, Fn* f) : type("spin"), min(minv), max(maxv), idx(Options.size()), on_change(f)
-{ std::ostringstream ss; ss << v; defaultValue = currentValue = ss.str(); }
+Option::Option(double v, int minv, int maxv, OnChange f) : type("spin"), min(minv), max(maxv), on_change(f)
+{ defaultValue = currentValue = std::to_string(v); }
 
+Option::Option(const char* v, const char* cur, OnChange f) : type("combo"), min(0), max(0), on_change(f)
+{ defaultValue = v; currentValue = cur; }
 
 Option::operator int() const {
   assert(type == "check" || type == "spin");
-  return (type == "spin" ? atoi(currentValue.c_str()) : currentValue == "true");
+  return (type == "spin" ? std::stoi(currentValue) : currentValue == "true");
 }
 
 Option::operator std::string() const {
@@ -140,27 +138,57 @@ Option::operator std::string() const {
   return currentValue;
 }
 
+bool Option::operator==(const char* s) const {
+  assert(type == "combo");
+  return   !CaseInsensitiveLess()(currentValue, s)
+        && !CaseInsensitiveLess()(s, currentValue);
+}
+
+
+/// operator<<() inits options and assigns idx in the correct printing order
+
+void Option::operator<<(const Option& o) {
+
+  static size_t insert_order = 0;
+
+  *this = o;
+  idx = insert_order++;
+}
+
 
 /// operator=() updates currentValue and triggers on_change() action. It's up to
-/// the GUI to check for option's limits, but we could receive the new value from
-/// the user by console window, so let's check the bounds anyway.
+/// the GUI to check for option's limits, but we could receive the new value
+/// from the user by console window, so let's check the bounds anyway.
 
 Option& Option::operator=(const string& v) {
 
   assert(!type.empty());
 
-  if (   (type != "button" && v.empty())
+  if (   (type != "button" && type != "string" && v.empty())
       || (type == "check" && v != "true" && v != "false")
-      || (type == "spin" && (atoi(v.c_str()) < min || atoi(v.c_str()) > max)))
+      || (type == "spin" && (stof(v) < min || stof(v) > max)))
       return *this;
 
+  if (type == "combo")
+  {
+      OptionsMap comboMap; // To have case insensitive compare
+      string token;
+      std::istringstream ss(defaultValue);
+      while (ss >> token)
+          comboMap[token] << Option();
+      if (!comboMap.count(v) || v == "var")
+          return *this;
+  }
+
   if (type != "button")
       currentValue = v;
 
   if (on_change)
-      (*on_change)(*this);
+      on_change(*this);
 
   return *this;
 }
 
 } // namespace UCI
+
+} // namespace Stockfish

src/ucioption.h

@@ -1,69 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
-  Copyright (C) 2008-2012 Marco Costalba, Joona Kiiski, Tord Romstad
-
-  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/>.
-*/
-
-#if !defined(UCIOPTION_H_INCLUDED)
-#define UCIOPTION_H_INCLUDED
-
-#include <map>
-#include <string>
-
-namespace UCI {
-
-class Option;
-
-/// Custom comparator because UCI options should be case insensitive
-struct CaseInsensitiveLess {
-  bool operator() (const std::string&, const std::string&) const;
-};
-
-/// Our options container is actually a std::map
-typedef std::map<std::string, Option, CaseInsensitiveLess> OptionsMap;
-
-/// Option class implements an option as defined by UCI protocol
-class Option {
-
-  typedef void (Fn)(const Option&);
-
-public:
-  Option(Fn* = NULL);
-  Option(bool v, Fn* = NULL);
-  Option(const char* v, Fn* = NULL);
-  Option(int v, int min, int max, Fn* = NULL);
-
-  Option& operator=(const std::string& v);
-  operator int() const;
-  operator std::string() const;
-
-private:
-  friend std::ostream& operator<<(std::ostream&, const OptionsMap&);
-
-  std::string defaultValue, currentValue, type;
-  int min, max;
-  size_t idx;
-  Fn* on_change;
-};
-
-void init(OptionsMap&);
-void loop(const std::string&);
-
-} // namespace UCI
-
-extern UCI::OptionsMap Options;
-
-#endif // !defined(UCIOPTION_H_INCLUDED)

tests/instrumented.sh

@@ -0,0 +1,151 @@
+#!/bin/bash
+# check for errors under valgrind or sanitizers.
+
+error()
+{
+  echo "instrumented testing failed on line $1"
+  exit 1
+}
+trap 'error ${LINENO}' ERR
+
+# define suitable post and prefixes for testing options
+case $1 in
+  --valgrind)
+    echo "valgrind testing started"
+    prefix=''
+    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 --fair-sched=try --error-exitcode=42'
+    postfix='1>/dev/null'
+    threads="2"
+  ;;
+  --sanitizer-undefined)
+    echo "sanitizer-undefined testing started"
+    prefix='!'
+    exeprefix=''
+    postfix='2>&1 | grep -A50 "runtime error:"'
+    threads="1"
+  ;;
+  --sanitizer-thread)
+    echo "sanitizer-thread testing started"
+    prefix='!'
+    exeprefix=''
+    postfix='2>&1 | grep -A50 "WARNING: ThreadSanitizer:"'
+    threads="2"
+
+cat << EOF > tsan.supp
+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:Stockfish::TranspositionTable::probe
+race:Stockfish::TranspositionTable::hashfull
+
+EOF
+
+    export TSAN_OPTIONS="suppressions=./tsan.supp"
+
+  ;;
+  *)
+    echo "unknown testing started"
+    prefix=''
+    exeprefix=''
+    postfix=''
+    threads="1"
+  ;;
+esac
+
+# simple command line testing
+for args in "eval" \
+            "go nodes 1000" \
+            "go depth 10" \
+            "go movetime 1000" \
+            "go wtime 8000 btime 8000 winc 500 binc 500" \
+            "bench 128 $threads 8 default depth" \
+            "export_net verify.nnue"
+do
+
+   echo "$prefix $exeprefix ./stockfish $args $postfix"
+   eval "$prefix $exeprefix ./stockfish $args $postfix"
+
+done
+
+# verify the generated net equals the base net
+network=`./stockfish uci | grep 'option name EvalFile type string default' | awk '{print $NF}'`
+echo "Comparing $network to the written verify.nnue"
+diff $network verify.nnue
+
+# more general testing, following an uci protocol exchange
+cat << EOF > game.exp
+ set timeout 240
+ spawn $exeprefix ./stockfish
+
+ send "uci\n"
+ expect "uciok"
+
+ send "setoption name Threads value $threads\n"
+
+ send "ucinewgame\n"
+ send "position startpos\n"
+ send "go nodes 1000\n"
+ expect "bestmove"
+
+ send "position startpos moves e2e4 e7e6\n"
+ send "go nodes 1000\n"
+ expect "bestmove"
+
+ send "position fen 5rk1/1K4p1/8/8/3B4/8/8/8 b - - 0 1\n"
+ send "go depth 10\n"
+ expect "bestmove"
+
+ send "quit\n"
+ expect eof
+
+ # return error code of the spawned program, useful for valgrind
+ lassign [wait] pid spawnid os_error_flag value
+ exit \$value
+EOF
+
+#download TB as needed
+if [ ! -d ../tests/syzygy ]; then
+   curl -sL https://api.github.com/repos/niklasf/python-chess/tarball/9b9aa13f9f36d08aadfabff872882f4ab1494e95 | tar -xzf -
+   mv niklasf-python-chess-9b9aa13 ../tests/syzygy
+fi
+
+cat << EOF > syzygy.exp
+ set timeout 240
+ spawn $exeprefix ./stockfish
+ send "uci\n"
+ send "setoption name SyzygyPath value ../tests/syzygy/\n"
+ expect "info string Found 35 tablebases" {} timeout {exit 1}
+ send "bench 128 1 8 default depth\n"
+ send "quit\n"
+ expect eof
+
+ # return error code of the spawned program, useful for valgrind
+ lassign [wait] pid spawnid os_error_flag value
+ exit \$value
+EOF
+
+for exp in game.exp syzygy.exp
+do
+
+  echo "$prefix expect $exp $postfix"
+  eval "$prefix expect $exp $postfix"
+
+  rm $exp
+
+done
+
+rm -f tsan.supp
+
+echo "instrumented testing OK"

tests/perft.sh

@@ -0,0 +1,32 @@
+#!/bin/bash
+# verify perft numbers (positions from www.chessprogramming.org/Perft_Results)
+
+error()
+{
+  echo "perft testing failed on line $1"
+  exit 1
+}
+trap 'error ${LINENO}' ERR
+
+echo "perft testing started"
+
+cat << EOF > perft.exp
+   set timeout 10
+   lassign \$argv pos depth result
+   spawn ./stockfish
+   send "position \$pos\\ngo perft \$depth\\n"
+   expect "Nodes searched? \$result" {} timeout {exit 1}
+   send "quit\\n"
+   expect eof
+EOF
+
+expect perft.exp startpos 5 4865609 > /dev/null
+expect perft.exp "fen r3k2r/p1ppqpb1/bn2pnp1/3PN3/1p2P3/2N2Q1p/PPPBBPPP/R3K2R w KQkq -" 5 193690690 > /dev/null
+expect perft.exp "fen 8/2p5/3p4/KP5r/1R3p1k/8/4P1P1/8 w - -" 6 11030083 > /dev/null
+expect perft.exp "fen r3k2r/Pppp1ppp/1b3nbN/nP6/BBP1P3/q4N2/Pp1P2PP/R2Q1RK1 w kq - 0 1" 5 15833292 > /dev/null
+expect perft.exp "fen rnbq1k1r/pp1Pbppp/2p5/8/2B5/8/PPP1NnPP/RNBQK2R w KQ - 1 8" 5 89941194 > /dev/null
+expect perft.exp "fen r4rk1/1pp1qppp/p1np1n2/2b1p1B1/2B1P1b1/P1NP1N2/1PP1QPPP/R4RK1 w - - 0 10" 5 164075551 > /dev/null
+
+rm perft.exp
+
+echo "perft testing OK"

tests/reprosearch.sh

@@ -0,0 +1,61 @@
+#!/bin/bash
+# verify reproducible search
+
+error()
+{
+  echo "reprosearch testing failed on line $1"
+  exit 1
+}
+trap 'error ${LINENO}' ERR
+
+echo "reprosearch testing started"
+
+# repeat two short games, separated by ucinewgame.
+# with go nodes $nodes they should result in exactly
+# the same node count for each iteration.
+cat << EOF > repeat.exp
+ set timeout 10
+ spawn ./stockfish
+ lassign \$argv nodes
+
+ send "uci\n"
+ expect "uciok"
+
+ send "ucinewgame\n"
+ send "position startpos\n"
+ send "go nodes \$nodes\n"
+ expect "bestmove"
+
+ send "position startpos moves e2e4 e7e6\n"
+ send "go nodes \$nodes\n"
+ expect "bestmove"
+
+ send "ucinewgame\n"
+ send "position startpos\n"
+ send "go nodes \$nodes\n"
+ expect "bestmove"
+
+ send "position startpos moves e2e4 e7e6\n"
+ send "go nodes \$nodes\n"
+ expect "bestmove"
+
+ send "quit\n"
+ expect eof
+EOF
+
+# to increase the likelihood of finding a non-reproducible case,
+# the allowed number of nodes are varied systematically
+for i in `seq 1 20`
+do
+
+  nodes=$((100*3**i/2**i))
+  echo "reprosearch testing with $nodes nodes"
+
+  # each line should appear exactly an even number of times
+  expect repeat.exp $nodes 2>&1 | grep -o "nodes [0-9]*" | sort | uniq -c | awk '{if ($1%2!=0) exit(1)}'
+
+done
+
+rm repeat.exp
+
+echo "reprosearch testing OK"

tests/signature.sh

@@ -0,0 +1,31 @@
+#!/bin/bash
+# obtain and optionally verify Bench / signature
+# if no reference is given, the output is deliberately limited to just the signature
+
+error()
+{
+  echo "running bench for signature failed on line $1"
+  exit 1
+}
+trap 'error ${LINENO}' ERR
+
+# obtain
+
+signature=`./stockfish bench 2>&1 | grep "Nodes searched  : " | awk '{print $4}'`
+
+if [ $# -gt 0 ]; then
+   # compare to given reference
+   if [ "$1" != "$signature" ]; then
+      if [ -z "$signature" ]; then
+         echo "No signature obtained from bench. Code crashed or assert triggered ?"
+      else
+         echo "signature mismatch: reference $1 obtained: $signature ."
+      fi
+      exit 1
+   else
+      echo "signature OK: $signature"
+   fi
+else
+   # just report signature
+   echo $signature
+fi