Update evaluate.h

Passed STC:
LLR: 2.95 (-2.94,2.94) <0.00,2.00>
Total: 29600 W: 7979 L: 7662 D: 13959
Ptnml(0-2): 138, 3446, 7324, 3745, 147
https://tests.stockfishchess.org/tests/view/67ac7dff52879dfd14d7e7da

Passed LTC:
LLR: 2.95 (-2.94,2.94) <0.50,2.50>
Total: 49662 W: 12850 L: 12502 D: 24310
Ptnml(0-2): 41, 5354, 13689, 5710, 37
https://tests.stockfishchess.org/tests/view/67acc1b252879dfd14d7e81d

closes https://github.com/official-stockfish/Stockfish/pull/5879

Bench: 2581469

.clang-format

@@ -0,0 +1,44 @@
+AccessModifierOffset: -1
+AlignAfterOpenBracket: Align
+AlignConsecutiveAssignments: Consecutive
+AlignConsecutiveDeclarations: Consecutive
+AlignEscapedNewlines: DontAlign
+AlignOperands: AlignAfterOperator
+AlignTrailingComments: true
+AllowAllParametersOfDeclarationOnNextLine: true
+AllowShortCaseLabelsOnASingleLine: false
+AllowShortEnumsOnASingleLine: false
+AllowShortIfStatementsOnASingleLine: false
+AlwaysBreakTemplateDeclarations: Yes
+BasedOnStyle: WebKit
+BitFieldColonSpacing: After
+BinPackParameters: false
+BreakBeforeBinaryOperators: NonAssignment
+BreakBeforeBraces: Custom
+BraceWrapping:
+  AfterFunction: false 
+  AfterClass: false
+  AfterControlStatement: true
+  BeforeElse: true
+BreakBeforeTernaryOperators: true
+BreakConstructorInitializers: AfterColon
+BreakStringLiterals: false
+ColumnLimit: 100
+ContinuationIndentWidth: 2
+Cpp11BracedListStyle: true
+IndentGotoLabels: false
+IndentPPDirectives: BeforeHash
+IndentWidth: 4
+MaxEmptyLinesToKeep: 2
+NamespaceIndentation: None
+PackConstructorInitializers: Never
+ReflowComments: false
+SortIncludes: false
+SortUsingDeclarations: false
+SpaceAfterCStyleCast: true
+SpaceAfterTemplateKeyword: false
+SpaceBeforeCaseColon: true
+SpaceBeforeCpp11BracedList: false
+SpaceBeforeInheritanceColon: false
+SpaceInEmptyBlock: false
+SpacesBeforeTrailingComments: 2

.git-blame-ignore-revs

@@ -0,0 +1,7 @@
+# .git-blame-ignore-revs
+# Ignore commit which added clang-format
+2d0237db3f0e596fb06e3ffbadba84dcc4e018f6
+
+# Post commit formatting fixes
+0fca5605fa2e5e7240fde5e1aae50952b2612231
+08ed4c90db31959521b7ef3186c026edd1e90307

.github/ISSUE_TEMPLATE/config.yml

@@ -2,7 +2,7 @@ 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!
+    about: Feel free to ask for support or have a chat with us on 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/ci/arm_matrix.json

@@ -0,0 +1,51 @@
+{
+  "config": [
+    {
+      "name": "Android NDK aarch64",
+      "os": "ubuntu-22.04",
+      "simple_name": "android",
+      "compiler": "aarch64-linux-android21-clang++",
+      "emu": "qemu-aarch64",
+      "comp": "ndk",
+      "shell": "bash",
+      "archive_ext": "tar"
+    },
+    {
+      "name": "Android NDK arm",
+      "os": "ubuntu-22.04",
+      "simple_name": "android",
+      "compiler": "armv7a-linux-androideabi21-clang++",
+      "emu": "qemu-arm",
+      "comp": "ndk",
+      "shell": "bash",
+      "archive_ext": "tar"
+    }
+  ],
+  "binaries": ["armv8-dotprod", "armv8", "armv7", "armv7-neon"],
+  "exclude": [
+    {
+      "binaries": "armv8-dotprod",
+      "config": {
+        "compiler": "armv7a-linux-androideabi21-clang++"
+      }
+    },
+    {
+      "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++"
+      }
+    }
+  ]
+}

.github/ci/libcxx17.imp

@@ -0,0 +1,22 @@
+[
+    # Mappings for libcxx's internal headers
+    { include: [ "<__fwd/fstream.h>", private, "<iosfwd>", public ] },
+    { include: [ "<__fwd/ios.h>", private, "<iosfwd>", public ] },
+    { include: [ "<__fwd/istream.h>", private, "<iosfwd>", public ] },
+    { include: [ "<__fwd/ostream.h>", private, "<iosfwd>", public ] },
+    { include: [ "<__fwd/sstream.h>", private, "<iosfwd>", public ] },
+    { include: [ "<__fwd/streambuf.h>", private, "<iosfwd>", public ] },
+    { include: [ "<__fwd/string_view.h>", private, "<string_view>", public ] },
+    { include: [ "<__system_error/errc.h>", private, "<system_error>", public ] },
+
+    # Mappings for includes between public headers
+    { include: [ "<ios>", public, "<iostream>", public ] },
+    { include: [ "<streambuf>", public, "<iostream>", public ] },
+    { include: [ "<istream>", public, "<iostream>", public ] },
+    { include: [ "<ostream>", public, "<iostream>", public ] },
+    { include: [ "<iosfwd>", public, "<iostream>", public ] },
+
+    # Missing mappings in include-what-you-use's libcxx.imp
+    { include: ["@<__condition_variable/.*>", private, "<condition_variable>", public ] },
+    { include: ["@<__mutex/.*>", private, "<mutex>", public ] },
+]

.github/ci/matrix.json

@@ -0,0 +1,160 @@
+{
+  "config": [
+    {
+      "name": "Ubuntu 22.04 GCC",
+      "os": "ubuntu-22.04",
+      "simple_name": "ubuntu",
+      "compiler": "g++",
+      "comp": "gcc",
+      "shell": "bash",
+      "archive_ext": "tar",
+      "sde": "/home/runner/work/Stockfish/Stockfish/.output/sde-temp-files/sde-external-9.27.0-2023-09-13-lin/sde -future --"
+    },
+    {
+      "name": "MacOS 13 Apple Clang",
+      "os": "macos-13",
+      "simple_name": "macos",
+      "compiler": "clang++",
+      "comp": "clang",
+      "shell": "bash",
+      "archive_ext": "tar"
+    },
+    {
+      "name": "MacOS 14 Apple Clang M1",
+      "os": "macos-14",
+      "simple_name": "macos-m1",
+      "compiler": "clang++",
+      "comp": "clang",
+      "shell": "bash",
+      "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",
+      "sde": "/d/a/Stockfish/Stockfish/.output/sde-temp-files/sde-external-9.27.0-2023-09-13-win/sde.exe -future --",
+      "archive_ext": "zip"
+    }
+  ],
+  "binaries": [
+    "x86-64",
+    "x86-64-sse41-popcnt",
+    "x86-64-avx2",
+    "x86-64-bmi2",
+    "x86-64-avxvnni",
+    "x86-64-avx512",
+    "x86-64-vnni256",
+    "x86-64-vnni512",
+    "apple-silicon"
+  ],
+  "exclude": [
+    {
+      "binaries": "x86-64",
+      "config": {
+        "os": "macos-14"
+      }
+    },
+    {
+      "binaries": "x86-64-sse41-popcnt",
+      "config": {
+        "os": "macos-14"
+      }
+    },
+    {
+      "binaries": "x86-64-avx2",
+      "config": {
+        "os": "macos-14"
+      }
+    },
+    {
+      "binaries": "x86-64-bmi2",
+      "config": {
+        "os": "macos-14"
+      }
+    },
+    {
+      "binaries": "x86-64-avxvnni",
+      "config": {
+        "os": "macos-14"
+      }
+    },
+    {
+      "binaries": "x86-64-avxvnni",
+      "config": {
+        "os": "macos-14"
+      }
+    },
+    {
+      "binaries": "x86-64-avx512",
+      "config": {
+        "os": "macos-14"
+      }
+    },
+    {
+      "binaries": "x86-64-vnni256",
+      "config": {
+        "os": "macos-14"
+      }
+    },
+    {
+      "binaries": "x86-64-vnni512",
+      "config": {
+        "os": "macos-14"
+      }
+    },
+    {
+      "binaries": "x86-64-avxvnni",
+      "config": {
+        "ubuntu-22.04": null
+      }
+    },
+    {
+      "binaries": "x86-64-avxvnni",
+      "config": {
+        "os": "macos-13"
+      }
+    },
+    {
+      "binaries": "x86-64-avx512",
+      "config": {
+        "os": "macos-13"
+      }
+    },
+    {
+      "binaries": "x86-64-vnni256",
+      "config": {
+        "os": "macos-13"
+      }
+    },
+    {
+      "binaries": "x86-64-vnni512",
+      "config": {
+        "os": "macos-13"
+      }
+    },
+    {
+      "binaries": "apple-silicon",
+      "config": {
+        "os": "windows-2022"
+      }
+    },
+    {
+      "binaries": "apple-silicon",
+      "config": {
+        "os": "macos-13"
+      }
+    },
+    {
+      "binaries": "apple-silicon",
+      "config": {
+        "os": "ubuntu-22.04"
+      }
+    }
+  ]
+}

.github/workflows/arm_compilation.yml

@@ -0,0 +1,98 @@
+name: Compilation
+on:
+  workflow_call:
+    inputs:
+      matrix:
+        type: string
+        required: true
+jobs:
+  Compilation:
+    name: ${{ matrix.config.name }} ${{ matrix.binaries }}
+    runs-on: ${{ matrix.config.os }}
+    env:
+      COMPCXX: ${{ matrix.config.compiler }}
+      COMP: ${{ matrix.config.comp }}
+      EMU: ${{ matrix.config.emu }}
+      EXT: ${{ matrix.config.ext }}
+      BINARY: ${{ matrix.binaries }}
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJson(inputs.matrix) }}
+    defaults:
+      run:
+        working-directory: src
+        shell: ${{ matrix.config.shell }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+
+      - 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: Extract the bench number from the commit history
+        run: |
+          for hash in $(git rev-list -100 HEAD); do
+            benchref=$(git show -s $hash | tac | grep -m 1 -o -x '[[:space:]]*\b[Bb]ench[ :]\+[1-9][0-9]\{5,7\}\b[[:space:]]*' | sed 's/[^0-9]//g') && break || true
+          done
+          [[ -n "$benchref" ]] && echo "benchref=$benchref" >> $GITHUB_ENV && echo "From commit: $hash" && echo "Reference bench: $benchref" || echo "No bench found"
+
+      - 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
+          $COMPCXX -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 }}:$PATH
+            export LDFLAGS="-static -Wno-unused-command-line-argument"
+          fi
+          make clean
+          make -j4 profile-build ARCH=$BINARY COMP=$COMP WINE_PATH=$EMU
+          make strip ARCH=$BINARY COMP=$COMP
+          WINE_PATH=$EMU ../tests/signature.sh $benchref
+          mv ./stockfish$EXT ../stockfish-android-$BINARY$EXT
+
+      - name: Remove non src files
+        run: git clean -fx
+
+      - name: Upload artifact for (pre)-release
+        uses: actions/upload-artifact@v4
+        with:
+          name: ${{ matrix.config.simple_name }} ${{ matrix.binaries }}
+          path: |
+            .
+            !.git
+            !.output

.github/workflows/clang-format.yml

@@ -0,0 +1,57 @@
+# This workflow will run clang-format and comment on the PR.
+# Because of security reasons, it is crucial that this workflow
+# executes no shell script nor runs make.
+# Read this before editing: https://securitylab.github.com/research/github-actions-preventing-pwn-requests/
+
+name: Clang-Format
+on:
+  pull_request_target:
+    branches:
+      - "master"
+    paths:
+      - "**.cpp"
+      - "**.h"
+
+permissions:
+  pull-requests: write
+
+jobs:
+  Clang-Format:
+    name: Clang-Format
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+
+      - name: Run clang-format style check
+        uses: jidicula/clang-format-action@f62da5e3d3a2d88ff364771d9d938773a618ab5e # @v4.11.0
+        id: clang-format
+        continue-on-error: true
+        with:
+          clang-format-version: "18"
+          exclude-regex: "incbin"
+
+      - name: Comment on PR
+        if: steps.clang-format.outcome == 'failure'
+        uses: thollander/actions-comment-pull-request@fabd468d3a1a0b97feee5f6b9e499eab0dd903f6 # @v2.5.0
+        with:
+          message: |
+            clang-format 18 needs to be run on this PR.
+            If you do not have clang-format installed, the maintainer will run it when merging.
+            For the exact version please see https://packages.ubuntu.com/noble/clang-format-18.
+
+            _(execution **${{ github.run_id }}** / attempt **${{ github.run_attempt }}**)_
+          comment_tag: execution
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Comment on PR
+        if: steps.clang-format.outcome != 'failure'
+        uses: thollander/actions-comment-pull-request@fabd468d3a1a0b97feee5f6b9e499eab0dd903f6 # @v2.5.0
+        with:
+          message: |
+            _(execution **${{ github.run_id }}** / attempt **${{ github.run_attempt }}**)_
+          create_if_not_exists: false
+          comment_tag: execution
+          mode: delete
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/codeql.yml

@@ -0,0 +1,55 @@
+name: "CodeQL"
+
+on:
+  push:
+    branches: ["master"]
+  pull_request:
+    # The branches below must be a subset of the branches above
+    branches: ["master"]
+  schedule:
+    - cron: "17 18 * * 1"
+
+jobs:
+  analyze:
+    name: Analyze
+    runs-on: ubuntu-latest
+    permissions:
+      actions: read
+      contents: read
+      security-events: write
+
+    strategy:
+      fail-fast: false
+      matrix:
+        language: ["cpp"]
+        # CodeQL supports [ 'cpp', 'csharp', 'go', 'java', 'javascript', 'python', 'ruby' ]
+        # Use only 'java' to analyze code written in Java, Kotlin, or both
+        # Use only 'javascript' to analyze code written in JavaScript, TypeScript or both
+        # Learn more about CodeQL language support at https://aka.ms/codeql-docs/language-support
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+
+      # Initializes the CodeQL tools for scanning.
+      - name: Initialize CodeQL
+        uses: github/codeql-action/init@v3
+        with:
+          languages: ${{ matrix.language }}
+          # If you wish to specify custom queries, you can do so here or in a config file.
+          # By default, queries listed here will override any specified in a config file.
+          # Prefix the list here with "+" to use these queries and those in the config file.
+
+          # For more details on CodeQL's query packs, refer to: https://docs.github.com/en/code-security/code-scanning/automatically-scanning-your-code-for-vulnerabilities-and-errors/configuring-code-scanning#using-queries-in-ql-packs
+          # queries: security-extended,security-and-quality
+
+      - name: Build
+        working-directory: src
+        run: make -j build ARCH=x86-64-modern
+
+      - name: Perform CodeQL Analysis
+        uses: github/codeql-action/analyze@v3
+        with:
+          category: "/language:${{matrix.language}}"

.github/workflows/compilation.yml

@@ -0,0 +1,94 @@
+name: Compilation
+on:
+  workflow_call:
+    inputs:
+      matrix:
+        type: string
+        required: true
+jobs:
+  Compilation:
+    name: ${{ matrix.config.name }} ${{ matrix.binaries }}
+    runs-on: ${{ matrix.config.os }}
+    env:
+      COMPCXX: ${{ matrix.config.compiler }}
+      COMP: ${{ matrix.config.comp }}
+      EXT: ${{ matrix.config.ext }}
+      NAME: ${{ matrix.config.simple_name }}
+      BINARY: ${{ matrix.binaries }}
+      SDE: ${{ matrix.config.sde }}
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJson(inputs.matrix) }}
+    defaults:
+      run:
+        working-directory: src
+        shell: ${{ matrix.config.shell }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+
+      - name: Install fixed GCC on Linux
+        if: runner.os == 'Linux'
+        uses: egor-tensin/setup-gcc@eaa888eb19115a521fa72b65cd94fe1f25bbcaac # @v1.3
+        with:
+          version: 11
+
+      - 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 SDE package
+        if: runner.os == 'Linux' || runner.os == 'Windows'
+        uses: petarpetrovt/setup-sde@91a1a03434384e064706634125a15f7446d2aafb # @v2.3
+        with:
+          environmentVariableName: SDE_DIR
+          sdeVersion: 9.27.0
+
+      - name: Download the used network from the fishtest framework
+        run: make net
+
+      - name: Check compiler
+        run: $COMPCXX -v
+
+      - name: Test help target
+        run: make help
+
+      - name: Check git
+        run: git --version
+
+      - name: Check compiler
+        run: $COMPCXX -v
+
+      - name: Show g++ cpu info
+        if: runner.os != 'macOS'
+        run: g++ -Q -march=native --help=target
+
+      - name: Show clang++ cpu info
+        if: runner.os == 'macOS'
+        run: clang++ -E - -march=native -###
+
+      # x86-64 with newer extensions tests
+
+      - name: Compile ${{ matrix.config.binaries }} build
+        run: |
+          make clean
+          make -j4 profile-build ARCH=$BINARY COMP=$COMP WINE_PATH="$SDE"
+          make strip ARCH=$BINARY COMP=$COMP
+          WINE_PATH="$SDE" ../tests/signature.sh $benchref
+          mv ./stockfish$EXT ../stockfish-$NAME-$BINARY$EXT
+
+      - name: Remove non src files
+        run: git clean -fx
+
+      - name: Upload artifact for (pre)-release
+        uses: actions/upload-artifact@v4
+        with:
+          name: ${{ matrix.config.simple_name }} ${{ matrix.binaries }}
+          path: |
+             .
+             !.git
+             !.output

.github/workflows/games.yml

@@ -0,0 +1,43 @@
+# This workflow will play games with a debug enabled SF using the PR
+
+name: Games
+on:
+  workflow_call:
+jobs:
+  Matetrack:
+    name: Games
+    runs-on: ubuntu-22.04
+    steps:
+      - name: Checkout SF repo 
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+          path: Stockfish
+          persist-credentials: false
+
+      - name: build debug enabled version of SF
+        working-directory: Stockfish/src
+        run: make -j build debug=yes
+
+      - name: Checkout fastchess repo
+        uses: actions/checkout@v4
+        with:
+          repository: Disservin/fastchess
+          path: fastchess
+          ref: 894616028492ae6114835195f14a899f6fa237d3
+          persist-credentials: false
+
+      - name: fastchess build
+        working-directory: fastchess
+        run: make -j
+
+      - name: Run games
+        working-directory: fastchess
+        run: |
+          ./fastchess -rounds 4 -games 2 -repeat -concurrency 4 -openings file=app/tests/data/openings.epd format=epd order=random -srand $RANDOM\
+               -engine name=sf1 cmd=/home/runner/work/Stockfish/Stockfish/Stockfish/src/stockfish\
+               -engine name=sf2 cmd=/home/runner/work/Stockfish/Stockfish/Stockfish/src/stockfish\
+               -ratinginterval 1 -report penta=true -each proto=uci tc=4+0.04 -log file=fast.log | tee fast.out
+          cat fast.log
+          ! grep "Assertion" fast.log > /dev/null
+          ! grep "disconnect" fast.out > /dev/null

.github/workflows/iwyu.yml

@@ -0,0 +1,49 @@
+name: IWYU
+on:
+  workflow_call:
+jobs:
+  Analyzers:
+    name: Check includes
+    runs-on: ubuntu-22.04
+    defaults:
+      run:
+        working-directory: Stockfish/src
+        shell: bash
+    steps:
+      - name: Checkout Stockfish
+        uses: actions/checkout@v4
+        with:
+          path: Stockfish
+          persist-credentials: false
+
+      - name: Checkout include-what-you-use
+        uses: actions/checkout@v4
+        with:
+          repository: include-what-you-use/include-what-you-use
+          ref: f25caa280dc3277c4086ec345ad279a2463fea0f
+          path: include-what-you-use
+          persist-credentials: false
+
+      - name: Download required linux packages
+        run: |
+          sudo add-apt-repository 'deb http://apt.llvm.org/jammy/ llvm-toolchain-jammy-17 main'
+          wget -O - https://apt.llvm.org/llvm-snapshot.gpg.key | sudo apt-key add -
+          sudo apt update
+          sudo apt install -y libclang-17-dev clang-17 libc++-17-dev
+
+      - name: Set up include-what-you-use
+        run: |
+          mkdir build && cd build
+          cmake -G "Unix Makefiles" -DCMAKE_PREFIX_PATH="/usr/lib/llvm-17" ..
+          sudo make install
+        working-directory: include-what-you-use
+
+      - name: Check include-what-you-use
+        run: include-what-you-use --version
+
+      - name: Check includes
+        run: >
+          make analyze
+          COMP=clang
+          CXX=include-what-you-use
+          CXXFLAGS="-stdlib=libc++ -Xiwyu --comment_style=long -Xiwyu --mapping='${{ github.workspace }}/Stockfish/.github/ci/libcxx17.imp' -Xiwyu --error"

.github/workflows/matetrack.yml

@@ -0,0 +1,71 @@
+# This workflow will run matetrack on the PR
+
+name: Matetrack
+on:
+  workflow_call:
+jobs:
+  Matetrack:
+    name: Matetrack
+    runs-on: ubuntu-22.04
+    steps:
+      - name: Checkout SF repo 
+        uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha }}
+          path: Stockfish
+          persist-credentials: false
+
+      - name: build SF
+        working-directory: Stockfish/src
+        run: make -j profile-build
+
+      - name: Checkout matetrack repo
+        uses: actions/checkout@v4
+        with:
+          repository: vondele/matetrack
+          path: matetrack
+          ref: 4f8a80860ed8f3607f05a9195df8b40203bdc360
+          persist-credentials: false
+
+      - name: matetrack install deps
+        working-directory: matetrack
+        run: pip install -r requirements.txt
+
+      - name: cache syzygy
+        id: cache-syzygy
+        uses: actions/cache@v4
+        with:
+           path: |
+              matetrack/3-4-5-wdl/
+              matetrack/3-4-5-dtz/
+           key: key-syzygy
+
+      - name: download syzygy 3-4-5 if needed
+        working-directory: matetrack
+        if: steps.cache-syzygy.outputs.cache-hit != 'true'
+        run: |
+          wget --no-verbose -r -nH --cut-dirs=2 --no-parent --reject="index.html*" -e robots=off https://tablebase.lichess.ovh/tables/standard/3-4-5-wdl/
+          wget --no-verbose -r -nH --cut-dirs=2 --no-parent --reject="index.html*" -e robots=off https://tablebase.lichess.ovh/tables/standard/3-4-5-dtz/
+
+      - name: Run matetrack
+        working-directory: matetrack
+        run: |
+          python matecheck.py --syzygyPath 3-4-5-wdl/:3-4-5-dtz/ --engine /home/runner/work/Stockfish/Stockfish/Stockfish/src/stockfish --epdFile mates2000.epd --nodes 100000 | tee matecheckout.out
+          ! grep "issues were detected" matecheckout.out > /dev/null
+
+      - name: Run matetrack with --syzygy50MoveRule false
+        working-directory: matetrack
+        run: |
+          grep 5men cursed.epd > cursed5.epd
+          python matecheck.py --syzygyPath 3-4-5-wdl/:3-4-5-dtz/ --engine /home/runner/work/Stockfish/Stockfish/Stockfish/src/stockfish --epdFile cursed5.epd --nodes 100000 --syzygy50MoveRule false | tee matecheckcursed.out
+          ! grep "issues were detected" matecheckcursed.out > /dev/null
+
+      - name: Verify mate and TB win count for matecheckcursed.out
+        working-directory: matetrack
+        run: |
+          mates=$(grep "Found mates:" matecheckcursed.out | awk '{print $3}')
+          tbwins=$(grep "Found TB wins:" matecheckcursed.out | awk '{print $4}')
+          if [ $(($mates + $tbwins)) -ne 32 ]; then
+            echo "Sum of mates and TB wins is not 32 in matecheckcursed.out" >&2
+            exit 1
+          fi

.github/workflows/sanitizers.yml

@@ -0,0 +1,87 @@
+name: Sanitizers
+on:
+  workflow_call:
+jobs:
+  Test-under-sanitizers:
+    name: ${{ matrix.sanitizers.name }}
+    runs-on: ${{ matrix.config.os }}
+    env:
+      COMPCXX: ${{ matrix.config.compiler }}
+      COMP: ${{ matrix.config.comp }}
+      CXXFLAGS: "-Werror"
+    strategy:
+      fail-fast: false
+      matrix:
+        config:
+          - name: Ubuntu 22.04 GCC
+            os: ubuntu-22.04
+            compiler: g++
+            comp: gcc
+            shell: bash
+        sanitizers:
+          - name: Run with thread sanitizer
+            make_option: sanitize=thread
+            cxx_extra_flags: ""
+            instrumented_option: sanitizer-thread
+          - name: Run with UB sanitizer
+            make_option: sanitize=undefined
+            cxx_extra_flags: ""
+            instrumented_option: sanitizer-undefined
+          - name: Run under valgrind
+            make_option: ""
+            cxx_extra_flags: ""
+            instrumented_option: valgrind
+          - name: Run under valgrind-thread
+            make_option: ""
+            cxx_extra_flags: ""
+            instrumented_option: valgrind-thread
+          - name: Run non-instrumented
+            make_option: ""
+            cxx_extra_flags: ""
+            instrumented_option: none
+          - name: Run with glibcxx assertions
+            make_option: ""
+            cxx_extra_flags: -D_GLIBCXX_ASSERTIONS
+            instrumented_option: non
+    defaults:
+      run:
+        working-directory: src
+        shell: ${{ matrix.config.shell }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+
+      - 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: $COMPCXX -v
+
+      - name: Test help target
+        run: make help
+
+      - name: Check git
+        run: git --version
+
+      # Since Linux Kernel 6.5 we are getting false positives from the ci,
+      # lower the ALSR entropy to disable ALSR, which works as a temporary workaround.
+      # https://github.com/google/sanitizers/issues/1716
+      # https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2056762
+
+      - name: Lower ALSR entropy
+        run: sudo sysctl -w vm.mmap_rnd_bits=28
+
+      # Sanitizers
+
+      - name: ${{ matrix.sanitizers.name }}
+        run: |
+          export CXXFLAGS="-O1 -fno-inline ${{ matrix.sanitizers.cxx_extra_flags }}"
+          make clean
+          make -j4 ARCH=x86-64-sse41-popcnt ${{ matrix.sanitizers.make_option }} debug=yes optimize=no build > /dev/null
+          python3 ../tests/instrumented.py --${{ matrix.sanitizers.instrumented_option }} ./stockfish

.github/workflows/stockfish.yml

@@ -1,8 +1,8 @@
 name: Stockfish
 on:
   push:
-    tags:        
+    tags:
-      - '*' 
+      - "*"
     branches:
       - master
       - tools
@@ -13,12 +13,14 @@ on:
       - tools
 jobs:
   Prerelease:
-    if: github.ref == 'refs/heads/master'
+    if: github.repository == 'official-stockfish/Stockfish' && (github.ref == 'refs/heads/master' || (startsWith(github.ref_name, 'sf_') && github.ref_type == 'tag'))
     runs-on: ubuntu-latest
+    permissions:
+      contents: write # For deleting/creating a prerelease
     steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v4
         with:
-          fetch-depth: 0
+          persist-credentials: false
 
       # returns null if no pre-release exists
       - name: Get Commit SHA of Latest Pre-release
@@ -27,25 +29,94 @@ jobs:
           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))" >> $GITHUB_ENV
+          echo "COMMIT_SHA_TAG=$(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
+      - run: gh release delete ${{ env.COMMIT_SHA_TAG }} --cleanup-tag
-        if: env.COMMIT_SHA != 'null'
+        if: env.COMMIT_SHA_TAG != 'null'
-        with:
-          tag_name: ${{ env.COMMIT_SHA }}
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
 
+      # Make sure that an old ci that 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" >> $GITHUB_ENV
+
+      - name: Get last commit SHA
+        id: last_commit
+        run: echo "COMMIT_SHA=$(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)" >> $GITHUB_ENV
+
+      # Create a new pre-release, the other upload_binaries.yml will upload the binaries
+      # to this pre-release.
+      - name: Create Prerelease
+        if: github.ref_name == 'master' && env.CHANGES == '0'
+        uses: softprops/action-gh-release@4634c16e79c963813287e889244c50009e7f0981
+        with:
+          name: Stockfish dev-${{ env.COMMIT_DATE }}-${{ env.COMMIT_SHA }}
+          tag_name: stockfish-dev-${{ env.COMMIT_DATE }}-${{ env.COMMIT_SHA }}
+          prerelease: true
+
+  Matrix:
+    runs-on: ubuntu-latest
+    outputs:
+      matrix: ${{ steps.set-matrix.outputs.matrix }}
+      arm_matrix: ${{ steps.set-arm-matrix.outputs.arm_matrix }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+      - id: set-matrix
+        run: |
+          TASKS=$(echo $(cat .github/ci/matrix.json) )
+          echo "MATRIX=$TASKS" >> $GITHUB_OUTPUT
+      - id: set-arm-matrix
+        run: |
+          TASKS_ARM=$(echo $(cat .github/ci/arm_matrix.json) )
+          echo "ARM_MATRIX=$TASKS_ARM" >> $GITHUB_OUTPUT
+  Compilation:
+    needs: [Matrix]
+    uses: ./.github/workflows/compilation.yml
+    with:
+      matrix: ${{ needs.Matrix.outputs.matrix }}
+  ARMCompilation:
+    needs: [Matrix]
+    uses: ./.github/workflows/arm_compilation.yml
+    with:
+      matrix: ${{ needs.Matrix.outputs.arm_matrix }}
+  IWYU:
+    uses: ./.github/workflows/iwyu.yml
   Sanitizers:
-    uses: ./.github/workflows/stockfish_sanitizers.yml
+    uses: ./.github/workflows/sanitizers.yml
   Tests:
-    uses: ./.github/workflows/stockfish_test.yml
+    uses: ./.github/workflows/tests.yml
-  Compiles:
+  Matetrack:
-    uses: ./.github/workflows/stockfish_compile_test.yml
+    uses: ./.github/workflows/matetrack.yml
+  Games:
+    uses: ./.github/workflows/games.yml
   Binaries:
-    if: github.ref == 'refs/heads/master' || (startsWith(github.ref_name, 'sf_') && github.ref_type == 'tag')
+    if: github.repository == 'official-stockfish/Stockfish'
-    uses: ./.github/workflows/stockfish_binaries.yml
+    needs: [Matrix, Prerelease, Compilation]
+    uses: ./.github/workflows/upload_binaries.yml
+    with:
+      matrix: ${{ needs.Matrix.outputs.matrix }}
+    permissions:
+      contents: write # For deleting/creating a (pre)release
+    secrets:
+      token: ${{ secrets.GITHUB_TOKEN }}
   ARM_Binaries:
-    if: github.ref == 'refs/heads/master' || (startsWith(github.ref_name, 'sf_') && github.ref_type == 'tag')
+    if: github.repository == 'official-stockfish/Stockfish'
-    uses: ./.github/workflows/stockfish_arm_binaries.yml
+    needs: [Matrix, Prerelease, ARMCompilation]
+    uses: ./.github/workflows/upload_binaries.yml
+    with:
+      matrix: ${{ needs.Matrix.outputs.arm_matrix }}
+    permissions:
+      contents: write # For deleting/creating a (pre)release
+    secrets:
+      token: ${{ secrets.GITHUB_TOKEN }}

.github/workflows/stockfish_arm_binaries.yml

@@ -1,158 +0,0 @@
-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

@@ -1,168 +0,0 @@
-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

@@ -1,102 +0,0 @@
-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

@@ -1,66 +0,0 @@
-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

@@ -1,256 +0,0 @@
-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

.github/workflows/tests.yml

@@ -0,0 +1,366 @@
+name: Tests
+on:
+  workflow_call:
+jobs:
+  Test-Targets:
+    name: ${{ matrix.config.name }}
+    runs-on: ${{ matrix.config.os }}
+    env:
+      COMPCXX: ${{ matrix.config.compiler }}
+      COMP: ${{ matrix.config.comp }}
+      CXXFLAGS: "-Werror"
+    strategy:
+      fail-fast: false
+      matrix:
+        config:
+          - name: Ubuntu 22.04 GCC
+            os: ubuntu-22.04
+            compiler: g++
+            comp: gcc
+            run_32bit_tests: true
+            run_64bit_tests: true
+            shell: bash
+          - name: Ubuntu 22.04 Clang
+            os: ubuntu-22.04
+            compiler: clang++
+            comp: clang
+            run_32bit_tests: true
+            run_64bit_tests: true
+            shell: bash
+          - name: Android NDK aarch64
+            os: ubuntu-22.04
+            compiler: aarch64-linux-android21-clang++
+            comp: ndk
+            run_armv8_tests: true
+            shell: bash
+          - name: Android NDK arm
+            os: ubuntu-22.04
+            compiler: armv7a-linux-androideabi21-clang++
+            comp: ndk
+            run_armv7_tests: true
+            shell: bash
+          - name: Linux GCC riscv64
+            os: ubuntu-22.04
+            compiler: g++
+            comp: gcc
+            run_riscv64_tests: true
+            base_image: "riscv64/alpine:edge"
+            platform: linux/riscv64
+            shell: bash
+          - name: Linux GCC ppc64
+            os: ubuntu-22.04
+            compiler: g++
+            comp: gcc
+            run_ppc64_tests: true
+            base_image: "ppc64le/alpine:latest"
+            platform: linux/ppc64le
+            shell: bash
+          - name: MacOS 13 Apple Clang
+            os: macos-13
+            compiler: clang++
+            comp: clang
+            run_64bit_tests: true
+            shell: bash
+          - name: MacOS 14 Apple Clang M1
+            os: macos-14
+            compiler: clang++
+            comp: clang
+            run_64bit_tests: false
+            run_m1_tests: true
+            shell: bash
+          - name: MacOS 13 GCC 11
+            os: macos-13
+            compiler: g++-11
+            comp: gcc
+            run_64bit_tests: true
+            shell: bash
+          - 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@v4
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+
+      - name: Download required linux packages
+        if: runner.os == 'Linux'
+        run: |
+          sudo apt update
+          sudo apt install expect valgrind g++-multilib qemu-user-static
+
+      - 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: Set up QEMU
+        if: matrix.config.base_image
+        uses: docker/setup-qemu-action@v3
+
+      - name: Set up Docker Buildx
+        if: matrix.config.base_image
+        uses: docker/setup-buildx-action@v3
+
+      - name: Build Docker container
+        if: matrix.config.base_image
+        run: |
+          docker buildx build --platform ${{ matrix.config.platform }} --load -t sf_builder - << EOF
+          FROM ${{ matrix.config.base_image }}
+          WORKDIR /app
+          RUN apk update && apk add make g++
+          CMD ["sh", "src/script.sh"]
+          EOF
+
+      - name: Download required macOS packages
+        if: runner.os == 'macOS'
+        run: brew install coreutils gcc@11
+
+      - 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: |
+          for hash in $(git rev-list -100 HEAD); do
+            benchref=$(git show -s $hash | tac | grep -m 1 -o -x '[[:space:]]*\b[Bb]ench[ :]\+[1-9][0-9]\{5,7\}\b[[:space:]]*' | sed 's/[^0-9]//g') && break || true
+          done
+          [[ -n "$benchref" ]] && echo "benchref=$benchref" >> $GITHUB_ENV && echo "From commit: $hash" && echo "Reference bench: $benchref" || echo "No bench found"
+
+      - name: Check compiler
+        run: |
+          if [ -z "${{ matrix.config.base_image }}" ]; then
+            if [ $COMP == ndk ]; then
+              export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
+            fi
+            $COMPCXX -v
+          else
+            echo "$COMPCXX -v" > script.sh
+            docker run --rm --platform ${{ matrix.config.platform }} -v ${{ github.workspace }}:/app sf_builder
+          fi
+
+      - 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 -j4 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 -j4 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 -j4 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 -j4 ARCH=x86-32-sse2 build
+          ../tests/signature.sh $benchref
+
+      - name: Test general-32 build
+        if: matrix.config.run_32bit_tests
+        run: |
+          make clean
+          make -j4 ARCH=general-32 build
+          ../tests/signature.sh $benchref
+
+      # x86-64 tests
+
+      - name: Test debug x86-64-avx2 build
+        if: matrix.config.run_64bit_tests
+        run: |
+          export CXXFLAGS="-Werror -D_GLIBCXX_DEBUG"
+          make clean
+          make -j4 ARCH=x86-64-avx2 optimize=no debug=yes build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64-bmi2 build
+        if: matrix.config.run_64bit_tests
+        run: |
+          make clean
+          make -j4 ARCH=x86-64-bmi2 build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64-avx2 build
+        if: matrix.config.run_64bit_tests
+        run: |
+          make clean
+          make -j4 ARCH=x86-64-avx2 build
+          ../tests/signature.sh $benchref
+
+      # Test a deprecated arch
+      - name: Test x86-64-modern build
+        if: matrix.config.run_64bit_tests
+        run: |
+          make clean
+          make -j4 ARCH=x86-64-modern build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64-sse41-popcnt build
+        if: matrix.config.run_64bit_tests
+        run: |
+          make clean
+          make -j4 ARCH=x86-64-sse41-popcnt build
+          ../tests/signature.sh $benchref
+
+      - name: Test x86-64-ssse3 build
+        if: matrix.config.run_64bit_tests
+        run: |
+          make clean
+          make -j4 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 -j4 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 -j4 ARCH=x86-64 build
+          ../tests/signature.sh $benchref
+
+      - name: Test general-64 build
+        if: matrix.config.run_64bit_tests
+        run: |
+          make clean
+          make -j4 ARCH=general-64 build
+          ../tests/signature.sh $benchref
+
+      - name: Test apple-silicon build
+        if: matrix.config.run_m1_tests
+        run: |
+          make clean
+          make -j4 ARCH=apple-silicon 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 -j4 ARCH=armv8 build
+          ../tests/signature.sh $benchref
+
+      - name: Test armv8-dotprod build
+        if: matrix.config.run_armv8_tests
+        run: |
+          export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
+          export LDFLAGS="-static -Wno-unused-command-line-argument"
+          make clean
+          make -j4 ARCH=armv8-dotprod 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 -j4 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 -j4 ARCH=armv7-neon build
+          ../tests/signature.sh $benchref
+
+      # riscv64 tests
+
+      - name: Test riscv64 build
+        if: matrix.config.run_riscv64_tests
+        run: |
+          echo "cd src && export LDFLAGS='-static' && make clean && make -j4 ARCH=riscv64 build" > script.sh
+          docker run --rm --platform ${{ matrix.config.platform }} -v ${{ github.workspace }}:/app sf_builder
+          ../tests/signature.sh $benchref
+
+      # ppc64 tests
+
+      - name: Test ppc64 build
+        if: matrix.config.run_ppc64_tests
+        run: |
+          echo "cd src && export LDFLAGS='-static' && make clean && make -j4 ARCH=ppc-64 build" > script.sh
+          docker run --rm --platform ${{ matrix.config.platform }} -v ${{ github.workspace }}:/app sf_builder
+          ../tests/signature.sh $benchref
+
+      # Other tests
+
+      - name: Check perft and search reproducibility
+        if: matrix.config.run_64bit_tests
+        run: |
+          make clean
+          make -j4 ARCH=x86-64-avx2 build
+          ../tests/perft.sh
+          ../tests/reprosearch.sh

.github/workflows/upload_binaries.yml

@@ -0,0 +1,114 @@
+name: Upload Binaries
+on:
+  workflow_call:
+    inputs:
+      matrix:
+        type: string
+        required: true
+    secrets:
+      token:
+        required: true
+
+jobs:
+  Artifacts:
+    name: ${{ matrix.config.name }} ${{ matrix.binaries }}
+    runs-on: ${{ matrix.config.os }}
+    env:
+      COMPCXX: ${{ matrix.config.compiler }}
+      COMP: ${{ matrix.config.comp }}
+      EXT: ${{ matrix.config.ext }}
+      NAME: ${{ matrix.config.simple_name }}
+      BINARY: ${{ matrix.binaries }}
+      SDE: ${{ matrix.config.sde }}
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJson(inputs.matrix) }}
+    defaults:
+      run:
+        shell: ${{ matrix.config.shell }}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
+
+      - name: Download artifact from compilation
+        uses: actions/download-artifact@v4
+        with:
+          name: ${{ matrix.config.simple_name }} ${{ matrix.binaries }}
+          path: ${{ matrix.config.simple_name }} ${{ matrix.binaries }}
+
+      - 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: Create Package
+        run: |
+          mkdir stockfish
+
+      - name: Download wiki
+        run: |
+          git clone https://github.com/official-stockfish/Stockfish.wiki.git wiki
+          rm -rf wiki/.git
+          mv wiki stockfish/
+
+      - name: Copy files
+        run: |
+          mv "${{ matrix.config.simple_name }} ${{ matrix.binaries }}" stockfish-workflow
+          cd stockfish-workflow
+          cp -r src ../stockfish/
+          cp -r scripts ../stockfish/
+          cp stockfish-$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/
+          cp CONTRIBUTING.md ../stockfish/
+
+      - name: Create tar
+        if: runner.os != 'Windows'
+        run: |
+          chmod +x ./stockfish/stockfish-$NAME-$BINARY$EXT
+          tar -cvf stockfish-$NAME-$BINARY.tar stockfish
+
+      - name: Create zip
+        if: runner.os == 'Windows'
+        run: |
+          zip -r stockfish-$NAME-$BINARY.zip stockfish
+
+      - name: Release
+        if: startsWith(github.ref_name, 'sf_') && github.ref_type == 'tag'
+        uses: softprops/action-gh-release@4634c16e79c963813287e889244c50009e7f0981
+        with:
+          files: stockfish-${{ matrix.config.simple_name }}-${{ matrix.binaries }}.${{ matrix.config.archive_ext }}
+          token: ${{ secrets.token }}
+
+      - name: Get last commit sha
+        id: last_commit
+        run: echo "COMMIT_SHA=$(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)" >> $GITHUB_ENV
+
+      # Make sure that an old ci that still runs on master doesn't recreate a prerelease
+      - name: Check Pullable Commits
+        id: check_commits
+        run: |
+          git fetch
+          CHANGES=$(git rev-list HEAD..origin/master --count)
+          echo "CHANGES=$CHANGES" >> $GITHUB_ENV
+
+      - name: Prerelease
+        if: github.ref_name == 'master' && env.CHANGES == '0'
+        continue-on-error: true
+        uses: softprops/action-gh-release@4634c16e79c963813287e889244c50009e7f0981
+        with:
+          name: Stockfish dev-${{ 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 }}
+          token: ${{ secrets.token }}

.gitignore

@@ -10,3 +10,8 @@ src/-lstdc++.res
 # Neural network for the NNUE evaluation
 **/*.nnue
 
+# Files generated by the instrumented tests
+tsan.supp
+__pycache__/
+tests/syzygy
+tests/bench_tmp.epd

AUTHORS

@@ -12,6 +12,7 @@ Hisayori Noda (nodchip)
 # All other authors of Stockfish code (in alphabetical order)
 Aditya (absimaldata)
 Adrian Petrescu (apetresc)
+Ahmed Kerimov (wcdbmv)
 Ajith Chandy Jose (ajithcj)
 Alain Savard (Rocky640)
 Alayan Feh (Alayan-stk-2)
@@ -19,6 +20,7 @@ Alexander Kure
 Alexander Pagel (Lolligerhans)
 Alfredo Menezes (lonfom169)
 Ali AlZhrani (Cooffe)
+Andreas Jan van der Meulen (Andyson007)
 Andreas Matthies (Matthies)
 Andrei Vetrov (proukornew)
 Andrew Grant (AndyGrant)
@@ -29,6 +31,7 @@ Aram Tumanian (atumanian)
 Arjun Temurnikar
 Artem Solopiy (EntityFX)
 Auguste Pop
+Balazs Szilagyi
 Balint Pfliegel
 Ben Chaney (Chaneybenjamini)
 Ben Koshy (BKSpurgeon)
@@ -42,12 +45,18 @@ Bruno de Melo Costa (BM123499)
 Bruno Pellanda (pellanda)
 Bryan Cross (crossbr)
 candirufish
+Carlos Esparza Sánchez (ces42)
 Chess13234
+Chris Bao (sscg13)
 Chris Cain (ceebo)
+Ciekce
 clefrks
+Clemens L. (rn5f107s2)
+Cody Ho (aesrentai)
 Dale Weiler (graphitemaster)
 Daniel Axtens (daxtens)
 Daniel Dugovic (ddugovic)
+Daniel Monroe (Ergodice)
 Dan Schmidt (dfannius)
 Dariusz Orzechowski (dorzechowski)
 David (dav1312)
@@ -62,16 +71,20 @@ Douglas Matos Gomes (dsmsgms)
 Dubslow
 Eduardo Cáceres (eduherminio)
 Eelco de Groot (KingDefender)
+Ehsan Rashid (erashid)
 Elvin Liu (solarlight2)
 erbsenzaehler
 Ernesto Gatti
+evqsx
 Fabian Beuke (madnight)
 Fabian Fichter (ianfab)
 Fanael Linithien (Fanael)
 fanon
-Fauzi Akram Dabat (FauziAkram)
+Fauzi Akram Dabat (fauzi2)
 Felix Wittmann
 gamander
+Gabriele Lombardo (gabe)
+Gahtan Nahdi
 Gary Heckman (gheckman)
 George Sobala (gsobala)
 gguliash
@@ -94,6 +107,7 @@ Jake Senne (w1wwwwww)
 Jan Ondruš (hxim)
 Jared Kish (Kurtbusch, kurt22i)
 Jarrod Torriero (DU-jdto)
+Jasper Shovelton (Beanie496)
 Jean-Francois Romang (jromang)
 Jean Gauthier (OuaisBla)
 Jekaa
@@ -111,13 +125,16 @@ jundery
 Justin Blanchard (UncombedCoconut)
 Kelly Wilson
 Ken Takusagawa
+Kenneth Lee (kennethlee33)
 Kian E (KJE-98)
 kinderchocolate
 Kiran Panditrao (Krgp)
 Kojirion
+Krisztián Peőcz
 Krystian Kuzniarek (kuzkry)
 Leonardo Ljubičić (ICCF World Champion)
 Leonid Pechenik (lp--)
+Li Ying (yl25946)
 Liam Keegan (lkeegan)
 Linmiao Xu (linrock)
 Linus Arver (listx)
@@ -130,6 +147,7 @@ Maciej Żenczykowski (zenczykowski)
 Malcolm Campbell (xoto10)
 Mark Tenzer (31m059)
 marotear
+Mathias Parnaudeau (mparnaudeau)
 Matt Ginsberg (mattginsberg)
 Matthew Lai (matthewlai)
 Matthew Sullivan (Matt14916)
@@ -158,10 +176,12 @@ Niklas Fiekas (niklasf)
 Nikolay Kostov (NikolayIT)
 Norman Schmidt (FireFather)
 notruck
+Nour Berakdar (Nonlinear)
 Ofek Shochat (OfekShochat, ghostway)
 Ondrej Mosnáček (WOnder93)
 Ondřej Mišina (AndrovT)
 Oskar Werkelin Ahlin
+Ömer Faruk Tutkun (OmerFarukTutkun)
 Pablo Vazquez
 Panthee
 Pascal Romaret
@@ -178,6 +198,7 @@ Raminder Singh
 renouve
 Reuven Peleg (R-Peleg)
 Richard Lloyd (Richard-Lloyd)
+Robert Nürnberg (robertnurnberg)
 Rodrigo Exterckötter Tjäder
 Rodrigo Roim (roim)
 Ronald de Man (syzygy1, syzygy)
@@ -195,16 +216,22 @@ sf-x
 Shahin M. Shahin (peregrine)
 Shane Booth (shane31)
 Shawn Varghese (xXH4CKST3RXx)
+Shawn Xu (xu-shawn)
 Siad Daboul (Topologist)
 Stefan Geschwentner (locutus2)
 Stefano Cardanobile (Stefano80)
 Stefano Di Martino (StefanoD)
 Steinar Gunderson (sesse)
 Stéphane Nicolet (snicolet)
+Stephen Touset (stouset)
 Syine Mineta (MinetaS)
+Taras Vuk (TarasVuk)
 Thanar2
 thaspel
 theo77186
+TierynnB
+Ting-Hsuan Huang (fffelix-huang)
+Tobias Steinmann
 Tomasz Sobczyk (Sopel97)
 Tom Truscott
 Tom Vijlbrief (tomtor)
@@ -216,11 +243,13 @@ Unai Corzo (unaiic)
 Uri Blass (uriblass)
 Vince Negri (cuddlestmonkey)
 Viren
+Wencey Wang
 windfishballad
 xefoci7612
+Xiang Wang (KatyushaScarlet)
 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
+# https://github.com/official-stockfish/fishtest/blob/master/AUTHORS

CONTRIBUTING.md

@@ -0,0 +1,96 @@
+# Contributing to Stockfish
+
+Welcome to the Stockfish project! We are excited that you are interested in
+contributing. This document outlines the guidelines and steps to follow when
+making contributions to Stockfish.
+
+## Table of Contents
+
+- [Building Stockfish](#building-stockfish)
+- [Making Contributions](#making-contributions)
+  - [Reporting Issues](#reporting-issues)
+  - [Submitting Pull Requests](#submitting-pull-requests)
+- [Code Style](#code-style)
+- [Community and Communication](#community-and-communication)
+- [License](#license)
+
+## Building Stockfish
+
+In case you do not have a C++ compiler installed, you can follow the
+instructions from our wiki.
+
+- [Ubuntu][ubuntu-compiling-link]
+- [Windows][windows-compiling-link]
+- [macOS][macos-compiling-link]
+
+## Making Contributions
+
+### Reporting Issues
+
+If you find a bug, please open an issue on the
+[issue tracker][issue-tracker-link]. Be sure to include relevant information
+like your operating system, build environment, and a detailed description of the
+problem.
+
+_Please note that Stockfish's development is not focused on adding new features.
+Thus any issue regarding missing features will potentially be closed without
+further discussion._
+
+### Submitting Pull Requests
+
+- Functional changes need to be tested on fishtest. See
+  [Creating my First Test][creating-my-first-test] for more details.
+  The accompanying pull request should include a link to the test results and
+  the new bench.
+
+- Non-functional changes (e.g. refactoring, code style, documentation) do not
+  need to be tested on fishtest, unless they might impact performance.
+
+- Provide a clear and concise description of the changes in the pull request
+  description.
+
+_First time contributors should add their name to [AUTHORS](./AUTHORS)._
+
+_Stockfish's development is not focused on adding new features. Thus any pull
+request introducing new features will potentially be closed without further
+discussion._
+
+## Code Style
+
+Changes to Stockfish C++ code should respect our coding style defined by
+[.clang-format](.clang-format). You can format your changes by running
+`make format`. This requires clang-format version 18 to be installed on your system.
+
+## Navigate
+
+For experienced Git users who frequently use git blame, it is recommended to
+configure the blame.ignoreRevsFile setting.
+This setting is useful for excluding noisy formatting commits.
+
+```bash
+git config blame.ignoreRevsFile .git-blame-ignore-revs
+```
+
+## Community and Communication
+
+- Join the [Stockfish discord][discord-link] to discuss ideas, issues, and
+  development.
+- Participate in the [Stockfish GitHub discussions][discussions-link] for
+  broader conversations.
+
+## License
+
+By contributing to Stockfish, you agree that your contributions will be licensed
+under the GNU General Public License v3.0. See [Copying.txt][copying-link] for
+more details.
+
+Thank you for contributing to Stockfish and helping us make it even better!
+
+[copying-link]:           https://github.com/official-stockfish/Stockfish/blob/master/Copying.txt
+[discord-link]:           https://discord.gg/GWDRS3kU6R
+[discussions-link]:       https://github.com/official-stockfish/Stockfish/discussions/new
+[creating-my-first-test]: https://github.com/official-stockfish/fishtest/wiki/Creating-my-first-test#create-your-test
+[issue-tracker-link]:     https://github.com/official-stockfish/Stockfish/issues
+[ubuntu-compiling-link]:  https://github.com/official-stockfish/Stockfish/wiki/Developers#user-content-installing-a-compiler-1
+[windows-compiling-link]: https://github.com/official-stockfish/Stockfish/wiki/Developers#user-content-installing-a-compiler
+[macos-compiling-link]:   https://github.com/official-stockfish/Stockfish/wiki/Developers#user-content-installing-a-compiler-2

README.md

@@ -59,39 +59,10 @@ This distribution of Stockfish consists of the following files:
   * 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
 
+__See [Contributing Guide](CONTRIBUTING.md).__
+
 ### Donating hardware
 
 Improving Stockfish requires a massive amount of testing. You can donate your
@@ -114,6 +85,25 @@ 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.
 
+## 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. An example suitable for most Intel and AMD chips:
+
+```
+cd src
+make -j profile-build
+```
+
+Detailed compilation instructions for all platforms can be found in our
+[documentation][wiki-compile-link]. Our wiki also has information about
+the [UCI commands][wiki-uci-link] supported by Stockfish.
+
 ## Terms of use
 
 Stockfish is free and distributed under the
@@ -130,6 +120,11 @@ 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.
 
+## Acknowledgements
+
+Stockfish uses neural networks trained on [data provided by the Leela Chess Zero
+project][lc0-data-link], which is made available under the [Open Database License][odbl-link] (ODbL).
+
 
 [authors-link]:       https://github.com/official-stockfish/Stockfish/blob/master/AUTHORS
 [build-link]:         https://github.com/official-stockfish/Stockfish/actions/workflows/stockfish.yml
@@ -138,7 +133,7 @@ also be made available under GPL v3.
 [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
+[guideline-link]:     https://github.com/official-stockfish/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
@@ -150,10 +145,12 @@ also be made available under GPL v3.
 [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
+[wiki-uci-link]:      https://github.com/official-stockfish/Stockfish/wiki/UCI-&-Commands
-[worker-link]:        https://github.com/glinscott/fishtest/wiki/Running-the-worker
+[wiki-usage-link]:    https://github.com/official-stockfish/Stockfish/wiki/Download-and-usage
+[worker-link]:        https://github.com/official-stockfish/fishtest/wiki/Running-the-worker
+[lc0-data-link]:      https://storage.lczero.org/files/training_data
+[odbl-link]:          https://opendatacommons.org/licenses/odbl/odbl-10.txt
 
 [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

Top CPU Contributors.txt

@@ -1,139 +1,155 @@
-Contributors to Fishtest with >10,000 CPU hours, as of 2023-06-20.
+Contributors to Fishtest with >10,000 CPU hours, as of 2024-08-31.
 Thank you!
 
 Username                                CPU Hours     Games played
 ------------------------------------------------------------------
-noobpwnftw                               37457426       2850540907
+noobpwnftw                               40428649       3164740143
-technologov                              14135647        742892808
+technologov                              23581394       1076895482
-linrock                                   4423514        303254809
+vdv                                      19425375        718302718
+linrock                                  10034115        643194527
 mlang                                     3026000        200065824
+okrout                                    2572676        237511408
+pemo                                      1836785         62226157
 dew                                       1689162        100033738
-okrout                                    1578136        148855886
+TueRens                                   1648780         77891164
-pemo                                      1508508         48814305
+sebastronomy                              1468328         60859092
-grandphish2                               1461406         91540343
+grandphish2                               1466110         91776075
-TueRens                                   1194790         70400852
+JojoM                                     1130625         73666098
-JojoM                                      947612         61773190
+olafm                                     1067009         74807270
 tvijlbrief                                 796125         51897690
-sebastronomy                               742434         38218524
+oz                                         781847         53910686
+rpngn                                      768460         49812975
+gvreuls                                    751085         52177668
 mibere                                     703840         46867607
-gvreuls                                    651026         42988582
+leszek                                     566598         42024615
-oz                                         543438         39314736
+cw                                         519601         34988161
-cw                                         517858         34869755
 fastgm                                     503862         30260818
-leszek                                     467278         33514883
+CSU_Dynasty                                468784         31385034
-CSU_Dynasty                                464940         31177118
+maximmasiutin                              439192         27893522
-ctoks                                      434416         28506889
+ctoks                                      435148         28541909
-crunchy                                    427035         27344275
+crunchy                                    427414         27371625
-maximmasiutin                              424795         26577722
+bcross                                     415724         29061187
-bcross                                     415722         29060963
+robal                                      371112         24642270
-olafm                                      395922         32268020
+mgrabiak                                   367963         26464704
-rpngn                                      348378         24560289
+velislav                                   342588         22140902
-velislav                                   342567         22138992
+ncfish1                                    329039         20624527
 Fisherman                                  327231         21829379
-mgrabiak                                   300612         20608380
 Dantist                                    296386         18031762
-nordlandia                                 246201         16189678
+tolkki963                                  262050         22049676
-robal                                      241300         15656382
+Sylvain27                                  255595          8864404
+nordlandia                                 249322         16420192
+Fifis                                      237657         13065577
 marrco                                     234581         17714473
-ncfish1                                    227517         15233777
+Calis007                                   217537         14450582
 glinscott                                  208125         13277240
 drabel                                     204167         13930674
 mhoram                                     202894         12601997
 bking_US                                   198894         11876016
 Thanar                                     179852         12365359
-vdv                                        175544          9904472
+javran                                     169679         13481966
+armo9494                                   162863         10937118
 spams                                      157128         10319326
+DesolatedDodo                              156683         10211206
+Wencey                                     152308          8375444
 sqrt2                                      147963          9724586
-DesolatedDodo                              146350          9536172
+vdbergh                                    140311          9225125
-Calis007                                   143165          9478764
+jcAEie                                     140086         10603658
-vdbergh                                    138650          9064413
 CoffeeOne                                  137100          5024116
-armo9494                                   136191          9460264
 malala                                     136182          8002293
 xoto                                       133759          9159372
+Dubslow                                    129614          8519312
 davar                                      129023          8376525
 DMBK                                       122960          8980062
 dsmith                                     122059          7570238
+CypressChess                               120784          8672620
+sschnee                                    120526          7547722
+maposora                                   119734         10749710
 amicic                                     119661          7938029
+Wolfgang                                   115713          8159062
 Data                                       113305          8220352
 BrunoBanani                                112960          7436849
-CypressChess                               108331          7759788
+markkulix                                  112897          9133168
+cuistot                                    109802          7121030
 skiminki                                   107583          7218170
-jcAEie                                     105675          8238962
+sterni1971                                 104431          5938282
 MaZePallas                                 102823          6633619
-sterni1971                                 100532          5880772
 sunu                                       100167          7040199
 zeryl                                       99331          6221261
-thirdlife                                   99124          2242380
+thirdlife                                   99156          2245320
 ElbertoOne                                  99028          7023771
-cuistot                                     98853          6069816
+megaman7de                                  98456          6675076
-bigpen0r                                    94809          6529203
+Goatminola                                  96765          8257832
+bigpen0r                                    94825          6529241
 brabos                                      92118          6186135
-Wolfgang                                    91939          6105872
+Maxim                                       90818          3283364
 psk                                         89957          5984901
-sschnee                                     88235          5268000
 racerschmacer                               85805          6122790
-Fifis                                       85722          5709729
-Dubslow                                     84986          6042456
 Vizvezdenec                                 83761          5344740
 0x3C33                                      82614          5271253
+szupaw                                      82495          7151686
 BRAVONE                                     81239          5054681
 nssy                                        76497          5259388
+cody                                        76126          4492126
 jromang                                     76106          5236025
+MarcusTullius                               76103          5061991
+woutboat                                    76072          6022922
+Spprtr                                      75977          5252287
 teddybaer                                   75125          5407666
-tolkki963                                   74762          5149662
-megaman7de                                  74351          4940352
-Wencey                                      74181          4711488
 Pking_cda                                   73776          5293873
-yurikvelo                                   73150          5004382
+yurikvelo                                   73611          5046822
-markkulix                                   72607          5304642
+Mineta                                      71130          4711422
 Bobo1239                                    70579          4794999
 solarlight                                  70517          5028306
 dv8silencer                                 70287          3883992
 manap                                       66273          4121774
 tinker                                      64333          4268790
 qurashee                                    61208          3429862
-Mineta                                      59357          4418202
+AGI                                         58195          4329580
-Spprtr                                      58723          3911011
-AGI                                         58147          4325994
 robnjr                                      57262          4053117
 Freja                                       56938          3733019
 MaxKlaxxMiner                               56879          3423958
-MarcusTullius                               56746          3762951
 ttruscott                                   56010          3680085
 rkl                                         55132          4164467
+jmdana                                      54697          4012593
+notchris                                    53936          4184018
 renouve                                     53811          3501516
-javran                                      53785          4627608
 finfish                                     51360          3370515
 eva42                                       51272          3599691
 eastorwest                                  51117          3454811
 rap                                         49985          3219146
 pb00067                                     49733          3298934
+GPUex                                       48686          3684998
 OuaisBla                                    48626          3445134
 ronaldjerum                                 47654          3240695
 biffhero                                    46564          3111352
+oryx                                        45639          3546530
 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
+jibarbosa                                   41640          4145702
 homyur                                      39893          2850481
 gri                                         39871          2515779
+DeepnessFulled                              39020          3323102
 Garf                                        37741          2999686
 SC                                          37299          2731694
+Gaster319                                   37118          3279678
+naclosagc                                   36562          1279618
 csnodgrass                                  36207          2688994
 strelock                                    34716          2074055
-szupaw                                      34102          2880346
+gopeto                                      33717          2245606
 EthanOConnor                                33370          2090311
 slakovv                                     32915          2021889
+jojo2357                                    32890          2826662
+shawnxu                                     32019          2802552
 Gelma                                       31771          1551204
-gopeto                                      31671          2060990
+vidar808                                    31560          1351810
 kdave                                       31157          2198362
 manapbk                                     30987          1810399
+ZacHFX                                      30966          2272416
+TataneSan                                   30713          1513402
+votoanthuan                                 30691          2460856
 Prcuvu                                      30377          2170122
 anst                                        30301          2190091
 jkiiski                                     30136          1904470
@@ -142,126 +158,144 @@ hyperbolic.tom                              29840          2017394
 chuckstablers                               29659          2093438
 Pyafue                                      29650          1902349
 belzedar94                                  28846          1811530
+mecevdimitar                                27610          1721382
 chriswk                                     26902          1868317
 xwziegtm                                    26897          2124586
 achambord                                   26582          1767323
+somethingintheshadows                       26496          2186404
 Patrick_G                                   26276          1801617
 yorkman                                     26193          1992080
-Ulysses                                     25288          1689730
+srowen                                      25743          1490684
+Ulysses                                     25413          1702830
+Jopo12321                                   25227          1652482
 SFTUser                                     25182          1675689
-nabildanial                                 24942          1519409
+nabildanial                                 25068          1531665
 Sharaf_DG                                   24765          1786697
-Maxim                                       24705          1502062
 rodneyc                                     24376          1416402
+jsys14                                      24297          1721230
 agg177                                      23890          1395014
-Goatminola                                  23763          1956036
+AndreasKrug                                 23754          1890115
-Ente                                        23639          1671638
+Ente                                        23752          1678188
-Jopo12321                                   23467          1483172
 JanErik                                     23408          1703875
 Isidor                                      23388          1680691
 Norabor                                     23371          1603244
+WoodMan777                                  23253          2023048
+Nullvalue                                   23155          2022752
 cisco2015                                   22920          1763301
-jsys14                                      22824          1591906
 Zirie                                       22542          1472937
 team-oh                                     22272          1636708
 Roady                                       22220          1465606
 MazeOfGalious                               21978          1629593
-sg4032                                      21947          1643353
+sg4032                                      21950          1643373
+tsim67                                      21747          1330880
 ianh2105                                    21725          1632562
+Skiff84                                     21711          1014212
 xor12                                       21628          1680365
 dex                                         21612          1467203
 nesoneg                                     21494          1463031
 user213718                                  21454          1404128
+Serpensin                                   21452          1790510
 sphinx                                      21211          1384728
-AndreasKrug                                 21097          1634811
+qoo_charly_cai                              21136          1514927
+IslandLambda                                21062          1220838
 jjoshua2                                    21001          1423089
 Zake9298                                    20938          1565848
 horst.prack                                 20878          1465656
+fishtester                                  20729          1348888
 0xB00B1ES                                   20590          1208666
+ols                                         20477          1195945
+Dinde                                       20459          1292774
 j3corre                                     20405           941444
 Adrian.Schmidt123                           20316          1281436
 wei                                         19973          1745989
-notchris                                    19958          1800128
+teenychess                                  19819          1762006
-Serpensin                                   19840          1697528
-Gaster319                                   19712          1677310
-fishtester                                  19617          1257388
 rstoesser                                   19569          1293588
 eudhan                                      19274          1283717
-votoanthuan                                 19108          1609992
 vulcan                                      18871          1729392
+wizardassassin                              18795          1376884
 Karpovbot                                   18766          1053178
-qoo_charly_cai                              18543          1284937
 jundery                                     18445          1115855
+mkstockfishtester                           18350          1690676
 ville                                       17883          1384026
 chris                                       17698          1487385
 purplefishies                               17595          1092533
 dju                                         17414           981289
 iisiraider                                  17275          1049015
 DragonLord                                  17014          1162790
+Karby                                       17008          1013160
+pirt                                        16965          1271519
 redstone59                                  16842          1461780
-Alb11747                                    16787          1213926
+Alb11747                                    16787          1213990
+Naven94                                     16414           951718
+scuzzi                                      16115           994341
 IgorLeMasson                                16064          1147232
-Karby                                       15982           979610
-scuzzi                                      15757           968735
 ako027ako                                   15671          1173203
+infinigon                                   15285           965966
 Nikolay.IT                                  15154          1068349
 Andrew Grant                                15114           895539
-Naven94                                     15054           834762
 OssumOpossum                                14857          1007129
-ZacHFX                                      14783          1021842
+LunaticBFF57                                14525          1190310
 enedene                                     14476           905279
+Hjax                                        14394          1005013
 bpfliegel                                   14233           882523
+YELNAMRON                                   14230          1128094
 mpx86                                       14019           759568
 jpulman                                     13982           870599
-Skiff84                                     13826           721996
+getraideBFF                                 13871          1172846
+Nesa92                                      13806          1116101
 crocogoat                                   13803          1117422
-Nesa92                                      13786          1114691
 joster                                      13710           946160
 mbeier                                      13650          1044928
-Hjax                                        13535           915487
+Pablohn26                                   13552          1088532
-Nullvalue                                   13468          1140498
+wxt9861                                     13550          1312306
 Dark_wizzie                                 13422          1007152
 Rudolphous                                  13244           883140
-pirt                                        13100          1009897
 Machariel                                   13010           863104
-infinigon                                   12991           943216
+nalanzeyu                                   12996           232590
 mabichito                                   12903           749391
+Jackfish                                    12895           868928
 thijsk                                      12886           722107
 AdrianSA                                    12860           804972
 Flopzee                                     12698           894821
+whelanh                                     12682           266404
+mschmidt                                    12644           863193
 korposzczur                                 12606           838168
 fatmurphy                                   12547           853210
+Oakwen                                      12532           855759
+icewulf                                     12447           854878
 SapphireBrand                               12416           969604
-Oakwen                                      12399           844109
 deflectooor                                 12386           579392
 modolief                                    12386           896470
 Farseer                                     12249           694108
-Jackfish                                    12213           805008
+Hongildong                                  12201           648712
 pgontarz                                    12151           848794
 dbernier                                    12103           860824
-getraideBFF                                 12072          1024966
+szczur90                                    12035           942376
+FormazChar                                  12019           910409
+rensonthemove                               11999           971993
 stocky                                      11954           699440
-mschmidt                                    11941           803401
+MooTheCow                                   11923           779432
-MooTheCow                                   11870           773598
+3cho                                        11842          1036786
-FormazChar                                  11766           885707
+ckaz                                        11792           732276
-whelanh                                     11557           245188
-3cho                                        11494          1031076
 infinity                                    11470           727027
 aga                                         11412           695127
 torbjo                                      11395           729145
 Thomas A. Anderson                          11372           732094
 savage84                                    11358           670860
+Def9Infinity                                11345           696552
 d64                                         11263           789184
 ali-al-zhrani                               11245           779246
+ImperiumAeternum                            11155           952000
 snicolet                                    11106           869170
 dapper                                      11032           771402
-ols                                         10947           624903
+Ethnikoi                                    10993           945906
-Karmatron                                   10828           677458
+Snuuka                                      10938           435504
+Karmatron                                   10871           678306
 basepi                                      10637           744851
 Cubox                                       10621           826448
+gerbil                                      10519           971688
 michaelrpg                                  10509           739239
 OIVAS7572                                   10420           995586
-jojo2357                                    10419           929708
-WoodMan777                                  10380           873720
 Garruk                                      10365           706465
 dzjp                                        10343           732529
+RickGroszkiewicz                            10263           990798

scripts/.gitattributes

@@ -0,0 +1 @@
+*.sh text eol=lf

scripts/get_native_properties.sh

@@ -0,0 +1,153 @@
+#!/bin/sh
+
+#
+# Returns properties of the native system.
+# best architecture as supported by the CPU
+# filename of the best binary uploaded as an artifact during CI
+#
+
+# Check if all the given flags are present in the CPU flags list
+check_flags() {
+  for flag; do
+    printf '%s\n' "$flags" | grep -q -w "$flag" || return 1
+  done
+}
+
+# Set the CPU flags list
+# remove underscores and points from flags, e.g. gcc uses avx512vnni, while some cpuinfo can have avx512_vnni, some systems use sse4_1 others sse4.1
+get_flags() {
+  flags=$(awk '/^flags[ \t]*:|^Features[ \t]*:/{gsub(/^flags[ \t]*:[ \t]*|^Features[ \t]*:[ \t]*|[_.]/, ""); line=$0} END{print line}' /proc/cpuinfo)
+}
+
+# Check for gcc march "znver1" or "znver2" https://en.wikichip.org/wiki/amd/cpuid
+check_znver_1_2() {
+  vendor_id=$(awk '/^vendor_id/{print $3; exit}' /proc/cpuinfo)
+  cpu_family=$(awk '/^cpu family/{print $4; exit}' /proc/cpuinfo)
+  [ "$vendor_id" = "AuthenticAMD" ] && [ "$cpu_family" = "23" ] && znver_1_2=true
+}
+
+# Set the file CPU loongarch64 architecture
+set_arch_loongarch64() {
+  if check_flags 'lasx'; then
+    true_arch='loongarch64-lasx'
+  elif check_flags 'lsx'; then
+    true_arch='lonngarch64-lsx'
+  else
+    true_arch='loongarch64'
+  fi
+}
+
+# Set the file CPU x86_64 architecture
+set_arch_x86_64() {
+  if check_flags 'avx512vnni' 'avx512dq' 'avx512f' 'avx512bw' 'avx512vl'; then
+    true_arch='x86-64-vnni256'
+  elif check_flags 'avx512f' 'avx512bw'; then
+    true_arch='x86-64-avx512'
+  elif [ -z "${znver_1_2+1}" ] && check_flags 'bmi2'; then
+    true_arch='x86-64-bmi2'
+  elif check_flags 'avx2'; then
+    true_arch='x86-64-avx2'
+  elif check_flags 'sse41' && check_flags 'popcnt'; then
+    true_arch='x86-64-sse41-popcnt'
+  else
+    true_arch='x86-64'
+  fi
+}
+
+set_arch_ppc_64() {
+  if $(grep -q -w "altivec" /proc/cpuinfo); then
+    power=$(grep -oP -m 1 'cpu\t+: POWER\K\d+' /proc/cpuinfo)
+    if [ "0$power" -gt 7 ]; then
+      # VSX started with POWER8
+      true_arch='ppc-64-vsx'
+    else
+      true_arch='ppc-64-altivec'
+    fi
+  else
+    true_arch='ppc-64'
+  fi
+}
+
+# Check the system type
+uname_s=$(uname -s)
+uname_m=$(uname -m)
+case $uname_s in
+  'Darwin') # Mac OSX system
+    case $uname_m in
+      'arm64')
+        true_arch='apple-silicon'
+        file_arch='m1-apple-silicon'
+        ;;
+      'x86_64')
+        flags=$(sysctl -n machdep.cpu.features machdep.cpu.leaf7_features | tr '\n' ' ' | tr '[:upper:]' '[:lower:]' | tr -d '_.')
+        set_arch_x86_64
+        if [ "$true_arch" = 'x86-64-vnni256' ] || [ "$true_arch" = 'x86-64-avx512' ]; then
+           file_arch='x86-64-bmi2'
+        fi
+        ;;
+    esac
+    file_os='macos'
+    file_ext='tar'
+    ;;
+  'Linux') # Linux system
+    get_flags
+    case $uname_m in
+      'x86_64')
+        file_os='ubuntu'
+        check_znver_1_2
+        set_arch_x86_64
+        ;;
+      'i686')
+        file_os='ubuntu'
+        true_arch='x86-32'
+        ;;
+      'ppc64'*)
+        file_os='ubuntu'
+        set_arch_ppc_64
+        ;;
+      'aarch64')
+        file_os='android'
+        true_arch='armv8'
+        if check_flags 'asimddp'; then
+          true_arch="$true_arch-dotprod"
+        fi
+        ;;
+      'armv7'*)
+        file_os='android'
+        true_arch='armv7'
+        if check_flags 'neon'; then
+          true_arch="$true_arch-neon"
+        fi
+        ;;
+      'loongarch64'*)
+        file_os='linux'
+        set_arch_loongarch64
+        ;;
+      *) # Unsupported machine type, exit with error
+        printf 'Unsupported machine type: %s\n' "$uname_m"
+        exit 1
+        ;;
+    esac
+    file_ext='tar'
+    ;;
+  'CYGWIN'*|'MINGW'*|'MSYS'*) # Windows system with POSIX compatibility layer
+    get_flags
+    check_znver_1_2
+    set_arch_x86_64
+    file_os='windows'
+    file_ext='zip'
+    ;;
+  *)
+    # Unknown system type, exit with error
+    printf 'Unsupported system type: %s\n' "$uname_s"
+    exit 1
+    ;;
+esac
+
+if [ -z "$file_arch" ]; then
+  file_arch=$true_arch
+fi
+
+file_name="stockfish-$file_os-$file_arch.$file_ext"
+
+printf '%s %s\n' "$true_arch" "$file_name"

scripts/net.sh

@@ -0,0 +1,75 @@
+#!/bin/sh
+
+wget_or_curl=$( (command -v wget > /dev/null 2>&1 && echo "wget -qO-") || \
+                (command -v curl > /dev/null 2>&1 && echo "curl -skL"))
+
+if [ -z "$wget_or_curl" ]; then
+  >&2 printf "%s\n" "Neither wget or curl is installed." \
+	         "Install one of these tools to download NNUE files automatically."
+  exit 1
+fi
+
+sha256sum=$( (command -v shasum > /dev/null 2>&1 && echo "shasum -a 256") || \
+             (command -v sha256sum > /dev/null 2>&1 && echo "sha256sum"))
+
+if [ -z "$sha256sum" ]; then
+  >&2 echo "sha256sum not found, NNUE files will be assumed valid."
+fi
+
+get_nnue_filename() {
+  grep "$1" evaluate.h | grep "#define" | sed "s/.*\(nn-[a-z0-9]\{12\}.nnue\).*/\1/"
+}
+
+validate_network() {
+  # If no sha256sum command is available, assume the file is always valid.
+  if [ -n "$sha256sum" ] && [ -f "$1" ]; then
+    if [ "$1" != "nn-$($sha256sum "$1" | cut -c 1-12).nnue" ]; then
+      rm -f "$1"
+      return 1
+    fi
+  fi
+}
+
+fetch_network() {
+  _filename="$(get_nnue_filename "$1")"
+
+  if [ -z "$_filename" ]; then
+    >&2 echo "NNUE file name not found for: $1"
+    return 1
+  fi
+
+  if [ -f "$_filename" ]; then
+    if validate_network "$_filename"; then
+      echo "Existing $_filename validated, skipping download"
+      return
+    else
+      echo "Removing invalid NNUE file: $_filename"
+    fi
+  fi
+
+  for url in \
+    "https://tests.stockfishchess.org/api/nn/$_filename" \
+    "https://github.com/official-stockfish/networks/raw/master/$_filename"; do
+    echo "Downloading from $url ..."
+    if $wget_or_curl "$url" > "$_filename"; then
+      if validate_network "$_filename"; then
+        echo "Successfully validated $_filename"
+      else
+        echo "Downloaded $_filename is invalid"
+        continue
+      fi
+    else
+      echo "Failed to download from $url"
+    fi
+    if [ -f "$_filename" ]; then
+      return
+    fi
+  done
+
+  # Download was not successful in the loop, return false.
+  >&2 echo "Failed to download $_filename"
+  return 1
+}
+
+fetch_network EvalFileDefaultNameBig && \
+fetch_network EvalFileDefaultNameSmall

src/Makefile

@@ -1,5 +1,5 @@
 # Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-# Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+# Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 #
 # Stockfish is free software: you can redistribute it and/or modify
 # it under the terms of the GNU General Public License as published by
@@ -20,9 +20,9 @@
 ### ==========================================================================
 
 ### Establish the operating system name
-KERNEL = $(shell uname -s)
+KERNEL := $(shell uname -s)
 ifeq ($(KERNEL),Linux)
-	OS = $(shell uname -o)
+	OS := $(shell uname -o)
 endif
 
 ### Target Windows OS
@@ -33,7 +33,7 @@ ifeq ($(OS),Windows_NT)
 else ifeq ($(COMP),mingw)
 	target_windows = yes
 	ifeq ($(WINE_PATH),)
-		WINE_PATH = $(shell which wine)
+		WINE_PATH := $(shell which wine)
 	endif
 endif
 
@@ -49,17 +49,21 @@ PREFIX = /usr/local
 BINDIR = $(PREFIX)/bin
 
 ### Built-in benchmark for pgo-builds
-ifeq ($(SDE_PATH),)
+PGOBENCH = $(WINE_PATH) ./$(EXE) bench
-	PGOBENCH = $(WINE_PATH) ./$(EXE) bench
-else
-	PGOBENCH = $(SDE_PATH) -- $(WINE_PATH) ./$(EXE) bench
-endif
 
 ### Source and object files
-SRCS = benchmark.cpp bitbase.cpp bitboard.cpp endgame.cpp evaluate.cpp main.cpp \
+SRCS = benchmark.cpp bitboard.cpp evaluate.cpp main.cpp \
-	material.cpp misc.cpp movegen.cpp movepick.cpp pawns.cpp position.cpp psqt.cpp \
+	misc.cpp movegen.cpp movepick.cpp position.cpp \
 	search.cpp thread.cpp timeman.cpp tt.cpp uci.cpp ucioption.cpp tune.cpp syzygy/tbprobe.cpp \
-	nnue/evaluate_nnue.cpp nnue/features/half_ka_v2_hm.cpp
+	nnue/nnue_misc.cpp nnue/features/half_ka_v2_hm.cpp nnue/network.cpp engine.cpp score.cpp memory.cpp
+
+HEADERS = benchmark.h bitboard.h evaluate.h misc.h movegen.h movepick.h history.h \
+		nnue/nnue_misc.h nnue/features/half_ka_v2_hm.h nnue/layers/affine_transform.h \
+		nnue/layers/affine_transform_sparse_input.h nnue/layers/clipped_relu.h nnue/layers/simd.h \
+		nnue/layers/sqr_clipped_relu.h nnue/nnue_accumulator.h nnue/nnue_architecture.h \
+		nnue/nnue_common.h nnue/nnue_feature_transformer.h position.h \
+		search.h syzygy/tbprobe.h thread.h thread_win32_osx.h timeman.h \
+		tt.h tune.h types.h uci.h ucioption.h perft.h nnue/network.h engine.h score.h numa.h memory.h
 
 OBJS = $(notdir $(SRCS:.cpp=.o))
 
@@ -94,8 +98,12 @@ VPATH = syzygy:nnue:nnue/features
 # avx512 = yes/no     --- -mavx512bw         --- Use Intel Advanced Vector Extensions 512
 # vnni256 = yes/no    --- -mavx256vnni       --- Use Intel Vector Neural Network Instructions 512 with 256bit operands
 # vnni512 = yes/no    --- -mavx512vnni       --- Use Intel Vector Neural Network Instructions 512
+# altivec = yes/no    --- -maltivec          --- Use PowerPC Altivec SIMD extension
+# vsx = yes/no        --- -mvsx              --- Use POWER VSX SIMD extension
 # neon = yes/no       --- -DUSE_NEON         --- Use ARM SIMD architecture
 # dotprod = yes/no    --- -DUSE_NEON_DOTPROD --- Use ARM advanced SIMD Int8 dot product instructions
+# lsx = yes/no        --- -mlsx              --- Use Loongson SIMD eXtension
+# lasx = yes/no       --- -mlasx             --- use Loongson Advanced SIMD eXtension
 #
 # Note that Makefile is space sensitive, so when adding new architectures
 # or modifying existing flags, you have to make sure there are no extra spaces
@@ -108,16 +116,21 @@ VPATH = syzygy:nnue:nnue/features
 ### 2.1. General and architecture defaults
 
 ifeq ($(ARCH),)
-   ARCH = x86-64-modern
+   ARCH = native
-   help_skip_sanity = yes
 endif
+
+ifeq ($(ARCH), native)
+   override ARCH := $(shell $(SHELL) ../scripts/get_native_properties.sh | cut -d " " -f 1)
+endif
+
 # explicitly check for the list of supported architectures (as listed with make help),
 # the user can override with `make ARCH=x86-32-vnni256 SUPPORTED_ARCH=true`
 ifeq ($(ARCH), $(filter $(ARCH), \
                  x86-64-vnni512 x86-64-vnni256 x86-64-avx512 x86-64-avxvnni x86-64-bmi2 \
                  x86-64-avx2 x86-64-sse41-popcnt x86-64-modern x86-64-ssse3 x86-64-sse3-popcnt \
-                 x86-64 x86-32-sse41-popcnt x86-32-sse2 x86-32 ppc-64 ppc-32 e2k \
+                 x86-64 x86-32-sse41-popcnt x86-32-sse2 x86-32 ppc-64 ppc-64-altivec ppc-64-vsx ppc-32 e2k \
-                 armv7 armv7-neon armv8 armv8-dotprod apple-silicon general-64 general-32 riscv64))
+                 armv7 armv7-neon armv8 armv8-dotprod apple-silicon general-64 general-32 riscv64 \
+                 loongarch64 loongarch64-lsx loongarch64-lasx))
    SUPPORTED_ARCH=true
 else
    SUPPORTED_ARCH=false
@@ -140,11 +153,21 @@ avxvnni = no
 avx512 = no
 vnni256 = no
 vnni512 = no
+altivec = no
+vsx = no
 neon = no
 dotprod = no
 arm_version = 0
+lsx = no
+lasx = no
 STRIP = strip
 
+ifneq ($(shell which clang-format-18 2> /dev/null),)
+	CLANG-FORMAT = clang-format-18
+else
+	CLANG-FORMAT = clang-format
+endif
+
 ### 2.2 Architecture specific
 
 ifeq ($(findstring x86,$(ARCH)),x86)
@@ -193,6 +216,8 @@ ifeq ($(findstring -sse41,$(ARCH)),-sse41)
 endif
 
 ifeq ($(findstring -modern,$(ARCH)),-modern)
+        $(warning *** ARCH=$(ARCH) is deprecated, defaulting to ARCH=x86-64-sse41-popcnt. Execute `make help` for a list of available architectures. ***)
+        $(shell sleep 5)
 	popcnt = yes
 	sse = yes
 	sse2 = yes
@@ -339,6 +364,20 @@ ifeq ($(ARCH),ppc-64)
 	prefetch = yes
 endif
 
+ifeq ($(ARCH),ppc-64-altivec)
+	arch = ppc64
+	popcnt = yes
+	prefetch = yes
+	altivec = yes
+endif
+
+ifeq ($(ARCH),ppc-64-vsx)
+	arch = ppc64
+	popcnt = yes
+	prefetch = yes
+	vsx = yes
+endif
+
 ifeq ($(findstring e2k,$(ARCH)),e2k)
 	arch = e2k
 	mmx = yes
@@ -353,6 +392,21 @@ endif
 ifeq ($(ARCH),riscv64)
 	arch = riscv64
 endif
+
+ifeq ($(findstring loongarch64,$(ARCH)),loongarch64)
+	arch = loongarch64
+	prefetch = yes
+
+ifeq ($(findstring -lasx,$(ARCH)),-lasx)
+	lsx = yes
+	lasx = yes
+endif
+
+ifeq ($(findstring -lsx,$(ARCH)),-lsx)
+	lsx = yes
+endif
+
+endif
 endif
 
 
@@ -388,6 +442,8 @@ ifeq ($(COMP),gcc)
 		ifeq ($(ARCH),riscv64)
 			CXXFLAGS += -latomic
 		endif
+	else ifeq ($(arch),loongarch64)
+		CXXFLAGS += -latomic
 	else
 		CXXFLAGS += -m$(bits)
 		LDFLAGS += -m$(bits)
@@ -458,6 +514,8 @@ ifeq ($(COMP),clang)
 		ifeq ($(ARCH),riscv64)
 			CXXFLAGS += -latomic
 		endif
+	else ifeq ($(arch),loongarch64)
+		CXXFLAGS += -latomic
 	else
 		CXXFLAGS += -m$(bits)
 		LDFLAGS += -m$(bits)
@@ -465,8 +523,8 @@ ifeq ($(COMP),clang)
 endif
 
 ifeq ($(KERNEL),Darwin)
-	CXXFLAGS += -mmacosx-version-min=10.14
+	CXXFLAGS += -mmacosx-version-min=10.15
-	LDFLAGS += -mmacosx-version-min=10.14
+	LDFLAGS += -mmacosx-version-min=10.15
 	ifneq ($(arch),any)
 		CXXFLAGS += -arch $(arch)
 		LDFLAGS += -arch $(arch)
@@ -497,6 +555,14 @@ ifeq ($(COMP),ndk)
 			STRIP=llvm-strip
 		endif
 	endif
+	ifeq ($(arch),x86_64)
+		CXX=x86_64-linux-android21-clang++
+		ifneq ($(shell which x86_64-linux-android-strip 2>/dev/null),)
+			STRIP=x86_64-linux-android-strip
+		else
+			STRIP=llvm-strip
+		endif
+	endif
 	LDFLAGS += -static-libstdc++ -pie -lm -latomic
 endif
 
@@ -514,11 +580,6 @@ else
 	endif
 endif
 
-### Travis CI script uses COMPILER to overwrite CXX
-ifdef COMPILER
-	COMPCXX=$(COMPILER)
-endif
-
 ### Allow overwriting CXX from command line
 ifdef COMPCXX
 	CXX=$(COMPCXX)
@@ -526,8 +587,8 @@ endif
 
 ### Sometimes gcc is really clang
 ifeq ($(COMP),gcc)
-	gccversion = $(shell $(CXX) --version 2>/dev/null)
+	gccversion := $(shell $(CXX) --version 2>/dev/null)
-	gccisclang = $(findstring clang,$(gccversion))
+	gccisclang := $(findstring clang,$(gccversion))
 	ifneq ($(gccisclang),)
 		profile_make = clang-profile-make
 		profile_use = clang-profile-use
@@ -564,7 +625,7 @@ endif
 ### 3.3 Optimization
 ifeq ($(optimize),yes)
 
-	CXXFLAGS += -O3
+	CXXFLAGS += -O3 -funroll-loops
 
 	ifeq ($(comp),gcc)
 		ifeq ($(OS), Android)
@@ -585,7 +646,7 @@ ifeq ($(optimize),yes)
 	endif
 
 	ifeq ($(comp),clang)
-		clangmajorversion = $(shell $(CXX) -dumpversion 2>/dev/null | cut -f1 -d.)
+		clangmajorversion := $(shell $(CXX) -dumpversion 2>/dev/null | cut -f1 -d.)
 		ifeq ($(shell expr $(clangmajorversion) \< 16),1)
 			CXXFLAGS += -fexperimental-new-pass-manager
 		endif
@@ -607,7 +668,7 @@ else
 endif
 
 ifeq ($(popcnt),yes)
-	ifeq ($(arch),$(filter $(arch),ppc64 armv7 armv8 arm64))
+	ifeq ($(arch),$(filter $(arch),ppc64 ppc64-altivec ppc64-vsx armv7 armv8 arm64))
 		CXXFLAGS += -DUSE_POPCNT
 	else
 		CXXFLAGS += -msse3 -mpopcnt -DUSE_POPCNT
@@ -672,12 +733,25 @@ ifeq ($(sse2),yes)
 endif
 
 ifeq ($(mmx),yes)
-	CXXFLAGS += -DUSE_MMX
 	ifeq ($(comp),$(filter $(comp),gcc clang mingw icx))
 		CXXFLAGS += -mmmx
 	endif
 endif
 
+ifeq ($(altivec),yes)
+	CXXFLAGS += -maltivec
+	ifeq ($(COMP),gcc)
+		CXXFLAGS += -mabi=altivec
+	endif
+endif
+
+ifeq ($(vsx),yes)
+	CXXFLAGS += -mvsx
+	ifeq ($(COMP),gcc)
+		CXXFLAGS += -DNO_WARN_X86_INTRINSICS -DUSE_SSE2
+	endif
+endif
+
 ifeq ($(neon),yes)
 	CXXFLAGS += -DUSE_NEON=$(arm_version)
 	ifeq ($(KERNEL),Linux)
@@ -693,6 +767,18 @@ ifeq ($(dotprod),yes)
 	CXXFLAGS += -march=armv8.2-a+dotprod -DUSE_NEON_DOTPROD
 endif
 
+ifeq ($(lasx),yes)
+	ifeq ($(comp),$(filter $(comp),gcc clang mingw icx))
+		CXXFLAGS += -mlasx
+	endif
+endif
+
+ifeq ($(lsx),yes)
+	ifeq ($(comp),$(filter $(comp),gcc clang mingw icx))
+		CXXFLAGS += -mlsx
+	endif
+endif
+
 ### 3.7 pext
 ifeq ($(pext),yes)
 	CXXFLAGS += -DUSE_PEXT
@@ -701,19 +787,24 @@ ifeq ($(pext),yes)
 	endif
 endif
 
-### 3.7.1 Try to include git commit sha for versioning
+### 3.8.1 Try to include git commit sha for versioning
-GIT_SHA = $(shell git rev-parse HEAD 2>/dev/null | cut -c 1-8)
+GIT_SHA := $(shell git rev-parse HEAD 2>/dev/null | cut -c 1-8)
 ifneq ($(GIT_SHA), )
 	CXXFLAGS += -DGIT_SHA=$(GIT_SHA)
 endif
 
-### 3.7.2 Try to include git commit date for versioning
+### 3.8.2 Try to include git commit date for versioning
-GIT_DATE = $(shell git show -s --date=format:'%Y%m%d' --format=%cd HEAD 2>/dev/null)
+GIT_DATE := $(shell git show -s --date=format:'%Y%m%d' --format=%cd HEAD 2>/dev/null)
 ifneq ($(GIT_DATE), )
 	CXXFLAGS += -DGIT_DATE=$(GIT_DATE)
 endif
 
-### 3.8 Link Time Optimization
+### 3.8.3 Try to include architecture
+ifneq ($(ARCH), )
+	CXXFLAGS += -DARCH=$(ARCH)
+endif
+
+### 3.9 Link Time Optimization
 ### This is a mix of compile and link time options because the lto link phase
 ### needs access to the optimization flags.
 ifeq ($(optimize),yes)
@@ -748,7 +839,7 @@ ifeq ($(debug), no)
 endif
 endif
 
-### 3.9 Android 5 can only run position independent executables. Note that this
+### 3.10 Android 5 can only run position independent executables. Note that this
 ### breaks Android 4.0 and earlier.
 ifeq ($(OS), Android)
 	CXXFLAGS += -fPIE
@@ -759,86 +850,91 @@ endif
 ### Section 4. Public Targets
 ### ==========================================================================
 
-
 help:
-	@echo ""
+	@echo "" && \
-	@echo "To compile stockfish, type: "
+	echo "To compile stockfish, type: " && \
-	@echo ""
+	echo "" && \
-	@echo "make target ARCH=arch [COMP=compiler] [COMPCXX=cxx]"
+	echo "make -j target [ARCH=arch] [COMP=compiler] [COMPCXX=cxx]" && \
-	@echo ""
+	echo "" && \
-	@echo "Supported targets:"
+	echo "Supported targets:" && \
-	@echo ""
+	echo "" && \
-	@echo "help                    > Display architecture details"
+	echo "help                    > Display architecture details" && \
-	@echo "profile-build           > standard build with profile-guided optimization"
+	echo "profile-build           > standard build with profile-guided optimization" && \
-	@echo "build                   > skip profile-guided optimization"
+	echo "build                   > skip profile-guided optimization" && \
-	@echo "net                     > Download the default nnue net"
+	echo "net                     > Download the default nnue nets" && \
-	@echo "strip                   > Strip executable"
+	echo "strip                   > Strip executable" && \
-	@echo "install                 > Install executable"
+	echo "install                 > Install executable" && \
-	@echo "clean                   > Clean up"
+	echo "clean                   > Clean up" && \
-	@echo ""
+	echo "" && \
-	@echo "Supported archs:"
+	echo "Supported archs:" && \
-	@echo ""
+	echo "" && \
-	@echo "x86-64-vnni512          > x86 64-bit with vnni 512bit support"
+	echo "native                  > select the best architecture for the host processor (default)" && \
-	@echo "x86-64-vnni256          > x86 64-bit with vnni 512bit support, limit operands to 256bit wide"
+	echo "x86-64-vnni512          > x86 64-bit with vnni 512bit support" && \
-	@echo "x86-64-avx512           > x86 64-bit with avx512 support"
+	echo "x86-64-vnni256          > x86 64-bit with vnni 512bit support, limit operands to 256bit wide" && \
-	@echo "x86-64-avxvnni          > x86 64-bit with vnni 256bit support"
+	echo "x86-64-avx512           > x86 64-bit with avx512 support" && \
-	@echo "x86-64-bmi2             > x86 64-bit with bmi2 support"
+	echo "x86-64-avxvnni          > x86 64-bit with vnni 256bit support" && \
-	@echo "x86-64-avx2             > x86 64-bit with avx2 support"
+	echo "x86-64-bmi2             > x86 64-bit with bmi2 support" && \
-	@echo "x86-64-sse41-popcnt     > x86 64-bit with sse41 and popcnt support"
+	echo "x86-64-avx2             > x86 64-bit with avx2 support" && \
-	@echo "x86-64-modern           > common modern CPU, currently x86-64-sse41-popcnt"
+	echo "x86-64-sse41-popcnt     > x86 64-bit with sse41 and popcnt support" && \
-	@echo "x86-64-ssse3            > x86 64-bit with ssse3 support"
+	echo "x86-64-modern           > deprecated, currently x86-64-sse41-popcnt" && \
-	@echo "x86-64-sse3-popcnt      > x86 64-bit with sse3 and popcnt support"
+	echo "x86-64-ssse3            > x86 64-bit with ssse3 support" && \
-	@echo "x86-64                  > x86 64-bit generic (with sse2 support)"
+	echo "x86-64-sse3-popcnt      > x86 64-bit with sse3 compile and popcnt support" && \
-	@echo "x86-32-sse41-popcnt     > x86 32-bit with sse41 and popcnt support"
+	echo "x86-64                  > x86 64-bit generic (with sse2 support)" && \
-	@echo "x86-32-sse2             > x86 32-bit with sse2 support"
+	echo "x86-32-sse41-popcnt     > x86 32-bit with sse41 and popcnt support" && \
-	@echo "x86-32                  > x86 32-bit generic (with mmx and sse support)"
+	echo "x86-32-sse2             > x86 32-bit with sse2 support" && \
-	@echo "ppc-64                  > PPC 64-bit"
+	echo "x86-32                  > x86 32-bit generic (with mmx compile support)" && \
-	@echo "ppc-32                  > PPC 32-bit"
+	echo "ppc-64                  > PPC 64-bit" && \
-	@echo "armv7                   > ARMv7 32-bit"
+	echo "ppc-64-altivec          > PPC 64-bit with altivec support" && \
-	@echo "armv7-neon              > ARMv7 32-bit with popcnt and neon"
+	echo "ppc-64-vsx              > PPC 64-bit with vsx support" && \
-	@echo "armv8                   > ARMv8 64-bit with popcnt and neon"
+	echo "ppc-32                  > PPC 32-bit" && \
-	@echo "armv8-dotprod           > ARMv8 64-bit with popcnt, neon and dot product support"
+	echo "armv7                   > ARMv7 32-bit" && \
-	@echo "e2k                     > Elbrus 2000"
+	echo "armv7-neon              > ARMv7 32-bit with popcnt and neon" && \
-	@echo "apple-silicon           > Apple silicon ARM64"
+	echo "armv8                   > ARMv8 64-bit with popcnt and neon" && \
-	@echo "general-64              > unspecified 64-bit"
+	echo "armv8-dotprod           > ARMv8 64-bit with popcnt, neon and dot product support" && \
-	@echo "general-32              > unspecified 32-bit"
+	echo "e2k                     > Elbrus 2000" && \
-	@echo "riscv64                 > RISC-V 64-bit"
+	echo "apple-silicon           > Apple silicon ARM64" && \
-	@echo ""
+	echo "general-64              > unspecified 64-bit" && \
-	@echo "Supported compilers:"
+	echo "general-32              > unspecified 32-bit" && \
-	@echo ""
+	echo "riscv64                 > RISC-V 64-bit" && \
-	@echo "gcc                     > Gnu compiler (default)"
+	echo "loongarch64             > LoongArch 64-bit" && \
-	@echo "mingw                   > Gnu compiler with MinGW under Windows"
+	echo "loongarch64-lsx         > LoongArch 64-bit with SIMD eXtension" && \
-	@echo "clang                   > LLVM Clang compiler"
+	echo "loongarch64-lasx        > LoongArch 64-bit with Advanced SIMD eXtension" && \
-	@echo "icx                     > Intel oneAPI DPC++/C++ Compiler"
+	echo "" && \
-	@echo "ndk                     > Google NDK to cross-compile for Android"
+	echo "Supported compilers:" && \
-	@echo ""
+	echo "" && \
-	@echo "Simple examples. If you don't know what to do, you likely want to run one of: "
+	echo "gcc                     > GNU compiler (default)" && \
-	@echo ""
+	echo "mingw                   > GNU compiler with MinGW under Windows" && \
-	@echo "make -j profile-build ARCH=x86-64-avx2    # typically a fast compile for common systems "
+	echo "clang                   > LLVM Clang compiler" && \
-	@echo "make -j profile-build ARCH=x86-64-modern  # A more portable compile for 64-bit systems "
+	echo "icx                     > Intel oneAPI DPC++/C++ Compiler" && \
-	@echo "make -j profile-build ARCH=x86-64         # A portable compile for 64-bit systems "
+	echo "ndk                     > Google NDK to cross-compile for Android" && \
-	@echo ""
+	echo "" && \
-	@echo "Advanced examples, for experienced users: "
+	echo "Simple examples. If you don't know what to do, you likely want to run one of: " && \
-	@echo ""
+	echo "" && \
-	@echo "make -j profile-build ARCH=x86-64-bmi2"
+	echo "make -j profile-build ARCH=x86-64-avx2    # typically a fast compile for common systems " && \
-	@echo "make -j profile-build ARCH=x86-64-bmi2 COMP=gcc COMPCXX=g++-9.0"
+	echo "make -j profile-build ARCH=x86-64-sse41-popcnt  # A more portable compile for 64-bit systems " && \
-	@echo "make -j build ARCH=x86-64-ssse3 COMP=clang"
+	echo "make -j profile-build ARCH=x86-64         # A portable compile for 64-bit systems " && \
-	@echo ""
+	echo "" && \
-	@echo "-------------------------------"
+	echo "Advanced examples, for experienced users: " && \
-ifeq ($(SUPPORTED_ARCH)$(help_skip_sanity), true)
+	echo "" && \
-	@echo "The selected architecture $(ARCH) will enable the following configuration: "
+	echo "make -j profile-build ARCH=x86-64-avxvnni" && \
-	@$(MAKE) ARCH=$(ARCH) COMP=$(COMP) config-sanity
+	echo "make -j profile-build ARCH=x86-64-avxvnni COMP=gcc COMPCXX=g++-12.0" && \
-else
+	echo "make -j build ARCH=x86-64-ssse3 COMP=clang" && \
+	echo ""
+ifneq ($(SUPPORTED_ARCH), true)
 	@echo "Specify a supported architecture with the ARCH option for more details"
 	@echo ""
 endif
 
 
-.PHONY: help build profile-build strip install clean net objclean profileclean \
+.PHONY: help analyze build profile-build strip install clean net \
-	config-sanity \
+	objclean profileclean config-sanity \
 	icx-profile-use icx-profile-make \
 	gcc-profile-use gcc-profile-make \
-	clang-profile-use clang-profile-make FORCE
+	clang-profile-use clang-profile-make FORCE \
+	format analyze
+
+analyze: net config-sanity objclean
+	$(MAKE) -k ARCH=$(ARCH) COMP=$(COMP) $(OBJS)
 
 build: net config-sanity
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) all
@@ -849,7 +945,8 @@ profile-build: net config-sanity objclean profileclean
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) $(profile_make)
 	@echo ""
 	@echo "Step 2/4. Running benchmark for pgo-build ..."
-	$(PGOBENCH) 2>&1 | tail -n 4
+	$(PGOBENCH) > PGOBENCH.out 2>&1
+	tail -n 4 PGOBENCH.out
 	@echo ""
 	@echo "Step 3/4. Building optimized executable ..."
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) objclean
@@ -870,42 +967,6 @@ install:
 clean: objclean profileclean
 	@rm -f .depend *~ core
 
-# evaluation network (nnue)
-net:
-	$(eval nnuenet := $(shell grep EvalFileDefaultName evaluate.h | grep define | sed 's/.*\(nn-[a-z0-9]\{12\}.nnue\).*/\1/'))
-	@echo "Default net: $(nnuenet)"
-	$(eval nnuedownloadurl1 := https://tests.stockfishchess.org/api/nn/$(nnuenet))
-	$(eval nnuedownloadurl2 := https://github.com/official-stockfish/networks/raw/master/$(nnuenet))
-	$(eval curl_or_wget := $(shell if hash curl 2>/dev/null; then echo "curl -skL"; elif hash wget 2>/dev/null; then echo "wget -qO-"; fi))
-	@if [ "x$(curl_or_wget)" = "x" ]; then \
-	    echo "Neither curl nor wget is installed. Install one of these tools unless the net has been downloaded manually"; \
-        fi
-	$(eval shasum_command := $(shell if hash shasum 2>/dev/null; then echo "shasum -a 256 "; elif hash sha256sum 2>/dev/null; then echo "sha256sum "; fi))
-	@if [ "x$(shasum_command)" = "x" ]; then \
-            echo "shasum / sha256sum not found, skipping net validation"; \
-        fi
-	@for nnuedownloadurl in "$(nnuedownloadurl1)" "$(nnuedownloadurl2)"; do \
-	   if test -f "$(nnuenet)"; then \
-	      echo "$(nnuenet) available."; \
-	   else \
-	      if [ "x$(curl_or_wget)" != "x" ]; then \
-                 echo "Downloading $${nnuedownloadurl}"; $(curl_or_wget) $${nnuedownloadurl} > $(nnuenet);\
-              else \
-                 echo "No net found and download not possible"; exit 1;\
-	      fi; \
-	   fi; \
-	   if [ "x$(shasum_command)" != "x" ]; then \
-	      if [ "$(nnuenet)" != "nn-"`$(shasum_command) $(nnuenet) | cut -c1-12`".nnue" ]; then \
-	         echo "Removing failed download"; rm -f $(nnuenet); \
-	      else \
-	         echo "Network validated"; break; \
-	      fi; \
-	   fi; \
-	done
-	@if ! test -f "$(nnuenet)"; then \
-	    echo "Failed to download $(nnuenet)."; \
-	fi
-
 # clean binaries and objects
 objclean:
 	@rm -f stockfish stockfish.exe *.o ./syzygy/*.o ./nnue/*.o ./nnue/features/*.o
@@ -913,13 +974,21 @@ objclean:
 # clean auxiliary profiling files
 profileclean:
 	@rm -rf profdir
-	@rm -f bench.txt *.gcda *.gcno ./syzygy/*.gcda ./nnue/*.gcda ./nnue/features/*.gcda *.s
+	@rm -f bench.txt *.gcda *.gcno ./syzygy/*.gcda ./nnue/*.gcda ./nnue/features/*.gcda *.s PGOBENCH.out
 	@rm -f stockfish.profdata *.profraw
 	@rm -f stockfish.*args*
 	@rm -f stockfish.*lt*
 	@rm -f stockfish.res
 	@rm -f ./-lstdc++.res
 
+# evaluation network (nnue)
+net:
+	@$(SHELL) ../scripts/net.sh
+
+format:
+	$(CLANG-FORMAT) -i $(SRCS) $(HEADERS) -style=file
+
+# default target
 default:
 	help
 
@@ -931,61 +1000,71 @@ all: $(EXE) .depend
 
 config-sanity: net
 	@echo ""
-	@echo "Config:"
+	@echo "Config:" && \
-	@echo "debug: '$(debug)'"
+	echo "debug: '$(debug)'" && \
-	@echo "sanitize: '$(sanitize)'"
+	echo "sanitize: '$(sanitize)'" && \
-	@echo "optimize: '$(optimize)'"
+	echo "optimize: '$(optimize)'" && \
-	@echo "arch: '$(arch)'"
+	echo "arch: '$(arch)'" && \
-	@echo "bits: '$(bits)'"
+	echo "bits: '$(bits)'" && \
-	@echo "kernel: '$(KERNEL)'"
+	echo "kernel: '$(KERNEL)'" && \
-	@echo "os: '$(OS)'"
+	echo "os: '$(OS)'" && \
-	@echo "prefetch: '$(prefetch)'"
+	echo "prefetch: '$(prefetch)'" && \
-	@echo "popcnt: '$(popcnt)'"
+	echo "popcnt: '$(popcnt)'" && \
-	@echo "pext: '$(pext)'"
+	echo "pext: '$(pext)'" && \
-	@echo "sse: '$(sse)'"
+	echo "sse: '$(sse)'" && \
-	@echo "mmx: '$(mmx)'"
+	echo "mmx: '$(mmx)'" && \
-	@echo "sse2: '$(sse2)'"
+	echo "sse2: '$(sse2)'" && \
-	@echo "ssse3: '$(ssse3)'"
+	echo "ssse3: '$(ssse3)'" && \
-	@echo "sse41: '$(sse41)'"
+	echo "sse41: '$(sse41)'" && \
-	@echo "avx2: '$(avx2)'"
+	echo "avx2: '$(avx2)'" && \
-	@echo "avxvnni: '$(avxvnni)'"
+	echo "avxvnni: '$(avxvnni)'" && \
-	@echo "avx512: '$(avx512)'"
+	echo "avx512: '$(avx512)'" && \
-	@echo "vnni256: '$(vnni256)'"
+	echo "vnni256: '$(vnni256)'" && \
-	@echo "vnni512: '$(vnni512)'"
+	echo "vnni512: '$(vnni512)'" && \
-	@echo "neon: '$(neon)'"
+	echo "altivec: '$(altivec)'" && \
-	@echo "dotprod: '$(dotprod)'"
+	echo "vsx: '$(vsx)'" && \
-	@echo "arm_version: '$(arm_version)'"
+	echo "neon: '$(neon)'" && \
-	@echo "target_windows: '$(target_windows)'"
+	echo "dotprod: '$(dotprod)'" && \
-	@echo ""
+	echo "arm_version: '$(arm_version)'" && \
-	@echo "Flags:"
+	echo "lsx: '$(lsx)'" && \
-	@echo "CXX: $(CXX)"
+	echo "lasx: '$(lasx)'" && \
-	@echo "CXXFLAGS: $(CXXFLAGS)"
+	echo "target_windows: '$(target_windows)'" && \
-	@echo "LDFLAGS: $(LDFLAGS)"
+	echo "" && \
-	@echo ""
+	echo "Flags:" && \
-	@echo "Testing config sanity. If this fails, try 'make help' ..."
+	echo "CXX: $(CXX)" && \
-	@echo ""
+	echo "CXXFLAGS: $(CXXFLAGS)" && \
-	@test "$(debug)" = "yes" || test "$(debug)" = "no"
+	echo "LDFLAGS: $(LDFLAGS)" && \
-	@test "$(optimize)" = "yes" || test "$(optimize)" = "no"
+	echo "" && \
-	@test "$(SUPPORTED_ARCH)" = "true"
+	echo "Testing config sanity. If this fails, try 'make help' ..." && \
-	@test "$(arch)" = "any" || test "$(arch)" = "x86_64" || test "$(arch)" = "i386" || \
+	echo "" && \
+	(test "$(debug)" = "yes" || test "$(debug)" = "no") && \
+	(test "$(optimize)" = "yes" || test "$(optimize)" = "no") && \
+	(test "$(SUPPORTED_ARCH)" = "true") && \
+	(test "$(arch)" = "any" || test "$(arch)" = "x86_64" || test "$(arch)" = "i386" || \
 	 test "$(arch)" = "ppc64" || test "$(arch)" = "ppc" || test "$(arch)" = "e2k" || \
-	 test "$(arch)" = "armv7" || test "$(arch)" = "armv8" || test "$(arch)" = "arm64" || test "$(arch)" = "riscv64"
+	 test "$(arch)" = "armv7" || test "$(arch)" = "armv8" || test "$(arch)" = "arm64" || \
-	@test "$(bits)" = "32" || test "$(bits)" = "64"
+	 test "$(arch)" = "riscv64" || test "$(arch)" = "loongarch64") && \
-	@test "$(prefetch)" = "yes" || test "$(prefetch)" = "no"
+	(test "$(bits)" = "32" || test "$(bits)" = "64") && \
-	@test "$(popcnt)" = "yes" || test "$(popcnt)" = "no"
+	(test "$(prefetch)" = "yes" || test "$(prefetch)" = "no") && \
-	@test "$(pext)" = "yes" || test "$(pext)" = "no"
+	(test "$(popcnt)" = "yes" || test "$(popcnt)" = "no") && \
-	@test "$(sse)" = "yes" || test "$(sse)" = "no"
+	(test "$(pext)" = "yes" || test "$(pext)" = "no") && \
-	@test "$(mmx)" = "yes" || test "$(mmx)" = "no"
+	(test "$(sse)" = "yes" || test "$(sse)" = "no") && \
-	@test "$(sse2)" = "yes" || test "$(sse2)" = "no"
+	(test "$(mmx)" = "yes" || test "$(mmx)" = "no") && \
-	@test "$(ssse3)" = "yes" || test "$(ssse3)" = "no"
+	(test "$(sse2)" = "yes" || test "$(sse2)" = "no") && \
-	@test "$(sse41)" = "yes" || test "$(sse41)" = "no"
+	(test "$(ssse3)" = "yes" || test "$(ssse3)" = "no") && \
-	@test "$(avx2)" = "yes" || test "$(avx2)" = "no"
+	(test "$(sse41)" = "yes" || test "$(sse41)" = "no") && \
-	@test "$(avx512)" = "yes" || test "$(avx512)" = "no"
+	(test "$(avx2)" = "yes" || test "$(avx2)" = "no") && \
-	@test "$(vnni256)" = "yes" || test "$(vnni256)" = "no"
+	(test "$(avx512)" = "yes" || test "$(avx512)" = "no") && \
-	@test "$(vnni512)" = "yes" || test "$(vnni512)" = "no"
+	(test "$(vnni256)" = "yes" || test "$(vnni256)" = "no") && \
-	@test "$(neon)" = "yes" || test "$(neon)" = "no"
+	(test "$(vnni512)" = "yes" || test "$(vnni512)" = "no") && \
-	@test "$(comp)" = "gcc" || test "$(comp)" = "icx" || test "$(comp)" = "mingw" || test "$(comp)" = "clang" \
+	(test "$(altivec)" = "yes" || test "$(altivec)" = "no") && \
-	|| test "$(comp)" = "armv7a-linux-androideabi16-clang"  || test "$(comp)" = "aarch64-linux-android21-clang"
+	(test "$(vsx)" = "yes" || test "$(vsx)" = "no") && \
+	(test "$(neon)" = "yes" || test "$(neon)" = "no") && \
+	(test "$(lsx)" = "yes" || test "$(lsx)" = "no") && \
+	(test "$(lasx)" = "yes" || test "$(lasx)" = "no") && \
+	(test "$(comp)" = "gcc" || test "$(comp)" = "icx" || test "$(comp)" = "mingw" || \
+	 test "$(comp)" = "clang" || test "$(comp)" = "armv7a-linux-androideabi16-clang" || \
+	 test "$(comp)" = "aarch64-linux-android21-clang")
 
 $(EXE): $(OBJS)
 	+$(CXX) -o $@ $(OBJS) $(LDFLAGS)
@@ -996,14 +1075,14 @@ FORCE:
 
 clang-profile-make:
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) \
-	EXTRACXXFLAGS='-fprofile-instr-generate ' \
+	EXTRACXXFLAGS='-fprofile-generate ' \
-	EXTRALDFLAGS=' -fprofile-instr-generate' \
+	EXTRALDFLAGS=' -fprofile-generate' \
 	all
 
 clang-profile-use:
 	$(XCRUN) llvm-profdata merge -output=stockfish.profdata *.profraw
 	$(MAKE) ARCH=$(ARCH) COMP=$(COMP) \
-	EXTRACXXFLAGS='-fprofile-instr-use=stockfish.profdata' \
+	EXTRACXXFLAGS='-fprofile-use=stockfish.profdata' \
 	EXTRALDFLAGS='-fprofile-use ' \
 	all
 

src/benchmark.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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 +17,17 @@
 */
 
 #include "benchmark.h"
+#include "numa.h"
 
+#include <cstdlib>
 #include <fstream>
 #include <iostream>
-#include <istream>
 #include <vector>
 
-#include "position.h"
-
-using namespace std;
-
 namespace {
 
-const vector<string> Defaults = {
+// clang-format off
+const std::vector<std::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",
@@ -92,86 +90,423 @@ const vector<string> Defaults = {
   "nqbnrkrb/pppppppp/8/8/8/8/PPPPPPPP/NQBNRKRB w KQkq - 0 1",
   "setoption name UCI_Chess960 value false"
 };
+// clang-format on
 
-} // namespace
+// clang-format off
+// human-randomly picked 5 games with <60 moves from
+// https://tests.stockfishchess.org/tests/view/665c71f9fd45fb0f907c21e0
+// only moves for one side
+const std::vector<std::vector<std::string>> BenchmarkPositions = {
+    {
+        "rnbq1k1r/ppp1bppp/4pn2/8/2B5/2NP1N2/PPP2PPP/R1BQR1K1 b - - 2 8",
+        "rnbq1k1r/pp2bppp/4pn2/2p5/2B2B2/2NP1N2/PPP2PPP/R2QR1K1 b - - 1 9",
+        "r1bq1k1r/pp2bppp/2n1pn2/2p5/2B1NB2/3P1N2/PPP2PPP/R2QR1K1 b - - 3 10",
+        "r1bq1k1r/pp2bppp/2n1p3/2p5/2B1PB2/5N2/PPP2PPP/R2QR1K1 b - - 0 11",
+        "r1b2k1r/pp2bppp/2n1p3/2p5/2B1PB2/5N2/PPP2PPP/3RR1K1 b - - 0 12",
+        "r1b1k2r/pp2bppp/2n1p3/2p5/2B1PB2/2P2N2/PP3PPP/3RR1K1 b - - 0 13",
+        "r1b1k2r/1p2bppp/p1n1p3/2p5/4PB2/2P2N2/PP2BPPP/3RR1K1 b - - 1 14",
+        "r1b1k2r/4bppp/p1n1p3/1pp5/P3PB2/2P2N2/1P2BPPP/3RR1K1 b - - 0 15",
+        "r1b1k2r/4bppp/p1n1p3/1P6/2p1PB2/2P2N2/1P2BPPP/3RR1K1 b - - 0 16",
+        "r1b1k2r/4bppp/2n1p3/1p6/2p1PB2/1PP2N2/4BPPP/3RR1K1 b - - 0 17",
+        "r3k2r/3bbppp/2n1p3/1p6/2P1PB2/2P2N2/4BPPP/3RR1K1 b - - 0 18",
+        "r3k2r/3bbppp/2n1p3/8/1pP1P3/2P2N2/3BBPPP/3RR1K1 b - - 1 19",
+        "1r2k2r/3bbppp/2n1p3/8/1pPNP3/2P5/3BBPPP/3RR1K1 b - - 3 20",
+        "1r2k2r/3bbppp/2n1p3/8/2PNP3/2B5/4BPPP/3RR1K1 b - - 0 21",
+        "1r2k2r/3bb1pp/2n1pp2/1N6/2P1P3/2B5/4BPPP/3RR1K1 b - - 1 22",
+        "1r2k2r/3b2pp/2n1pp2/1N6/1BP1P3/8/4BPPP/3RR1K1 b - - 0 23",
+        "1r2k2r/3b2pp/4pp2/1N6/1nP1P3/8/3RBPPP/4R1K1 b - - 1 24",
+        "1r5r/3bk1pp/4pp2/1N6/1nP1PP2/8/3RB1PP/4R1K1 b - - 0 25",
+        "1r5r/3bk1pp/2n1pp2/1N6/2P1PP2/8/3RBKPP/4R3 b - - 2 26",
+        "1r5r/3bk1pp/2n2p2/1N2p3/2P1PP2/6P1/3RBK1P/4R3 b - - 0 27",
+        "1r1r4/3bk1pp/2n2p2/1N2p3/2P1PP2/6P1/3RBK1P/R7 b - - 2 28",
+        "1r1r4/N3k1pp/2n1bp2/4p3/2P1PP2/6P1/3RBK1P/R7 b - - 4 29",
+        "1r1r4/3bk1pp/2N2p2/4p3/2P1PP2/6P1/3RBK1P/R7 b - - 0 30",
+        "1r1R4/4k1pp/2b2p2/4p3/2P1PP2/6P1/4BK1P/R7 b - - 0 31",
+        "3r4/4k1pp/2b2p2/4P3/2P1P3/6P1/4BK1P/R7 b - - 0 32",
+        "3r4/R3k1pp/2b5/4p3/2P1P3/6P1/4BK1P/8 b - - 1 33",
+        "8/3rk1pp/2b5/R3p3/2P1P3/6P1/4BK1P/8 b - - 3 34",
+        "8/3r2pp/2bk4/R1P1p3/4P3/6P1/4BK1P/8 b - - 0 35",
+        "8/2kr2pp/2b5/R1P1p3/4P3/4K1P1/4B2P/8 b - - 2 36",
+        "1k6/3r2pp/2b5/RBP1p3/4P3/4K1P1/7P/8 b - - 4 37",
+        "8/1k1r2pp/2b5/R1P1p3/4P3/3BK1P1/7P/8 b - - 6 38",
+        "1k6/3r2pp/2b5/2P1p3/4P3/3BK1P1/7P/R7 b - - 8 39",
+        "1k6/r5pp/2b5/2P1p3/4P3/3BK1P1/7P/5R2 b - - 10 40",
+        "1k3R2/6pp/2b5/2P1p3/4P3/r2BK1P1/7P/8 b - - 12 41",
+        "5R2/2k3pp/2b5/2P1p3/4P3/r2B2P1/3K3P/8 b - - 14 42",
+        "5R2/2k3pp/2b5/2P1p3/4P3/3BK1P1/r6P/8 b - - 16 43",
+        "5R2/2k3pp/2b5/2P1p3/4P3/r2B2P1/4K2P/8 b - - 18 44",
+        "5R2/2k3pp/2b5/2P1p3/4P3/3B1KP1/r6P/8 b - - 20 45",
+        "8/2k2Rpp/2b5/2P1p3/4P3/r2B1KP1/7P/8 b - - 22 46",
+        "3k4/5Rpp/2b5/2P1p3/4P3/r2B2P1/4K2P/8 b - - 24 47",
+        "3k4/5Rpp/2b5/2P1p3/4P3/3B1KP1/r6P/8 b - - 26 48",
+        "3k4/5Rpp/2b5/2P1p3/4P3/r2B2P1/4K2P/8 b - - 28 49",
+        "3k4/5Rpp/2b5/2P1p3/4P3/3BK1P1/r6P/8 b - - 30 50",
+        "3k4/5Rpp/2b5/2P1p3/4P3/r2B2P1/3K3P/8 b - - 32 51",
+        "3k4/5Rpp/2b5/2P1p3/4P3/2KB2P1/r6P/8 b - - 34 52",
+        "3k4/5Rpp/2b5/2P1p3/4P3/r2B2P1/2K4P/8 b - - 36 53",
+        "3k4/5Rpp/2b5/2P1p3/4P3/1K1B2P1/r6P/8 b - - 38 54",
+        "3k4/6Rp/2b5/2P1p3/4P3/1K1B2P1/7r/8 b - - 0 55",
+        "3k4/8/2b3Rp/2P1p3/4P3/1K1B2P1/7r/8 b - - 1 56",
+        "8/2k3R1/2b4p/2P1p3/4P3/1K1B2P1/7r/8 b - - 3 57",
+        "3k4/8/2b3Rp/2P1p3/4P3/1K1B2P1/7r/8 b - - 5 58",
+        "8/2k5/2b3Rp/2P1p3/1K2P3/3B2P1/7r/8 b - - 7 59",
+        "8/2k5/2b3Rp/2P1p3/4P3/2KB2P1/3r4/8 b - - 9 60",
+        "8/2k5/2b3Rp/2P1p3/1K2P3/3B2P1/6r1/8 b - - 11 61",
+        "8/2k5/2b3Rp/2P1p3/4P3/2KB2P1/3r4/8 b - - 13 62",
+        "8/2k5/2b3Rp/2P1p3/2K1P3/3B2P1/6r1/8 b - - 15 63",
+        "4b3/2k3R1/7p/2P1p3/2K1P3/3B2P1/6r1/8 b - - 17 64",
+    },
+    {
+        "r1bqkbnr/npp1pppp/p7/3P4/4pB2/2N5/PPP2PPP/R2QKBNR w KQkq - 1 6",
+        "r1bqkb1r/npp1pppp/p4n2/3P4/4pB2/2N5/PPP1QPPP/R3KBNR w KQkq - 3 7",
+        "r2qkb1r/npp1pppp/p4n2/3P1b2/4pB2/2N5/PPP1QPPP/2KR1BNR w kq - 5 8",
+        "r2qkb1r/1pp1pppp/p4n2/1n1P1b2/4pB2/2N4P/PPP1QPP1/2KR1BNR w kq - 1 9",
+        "r2qkb1r/1pp1pppp/5n2/1p1P1b2/4pB2/7P/PPP1QPP1/2KR1BNR w kq - 0 10",
+        "r2qkb1r/1ppbpppp/5n2/1Q1P4/4pB2/7P/PPP2PP1/2KR1BNR w kq - 1 11",
+        "3qkb1r/1Qpbpppp/5n2/3P4/4pB2/7P/rPP2PP1/2KR1BNR w k - 0 12",
+        "q3kb1r/1Qpbpppp/5n2/3P4/4pB2/7P/rPP2PP1/1K1R1BNR w k - 2 13",
+        "r3kb1r/2pbpppp/5n2/3P4/4pB2/7P/1PP2PP1/1K1R1BNR w k - 0 14",
+        "r3kb1r/2Bb1ppp/4pn2/3P4/4p3/7P/1PP2PP1/1K1R1BNR w k - 0 15",
+        "r3kb1r/2Bb2pp/4pn2/8/4p3/7P/1PP2PP1/1K1R1BNR w k - 0 16",
+        "r3k2r/2Bb2pp/4pn2/2b5/4p3/7P/1PP1NPP1/1K1R1B1R w k - 2 17",
+        "r6r/2Bbk1pp/4pn2/2b5/3Np3/7P/1PP2PP1/1K1R1B1R w - - 4 18",
+        "r6r/b2bk1pp/4pn2/4B3/3Np3/7P/1PP2PP1/1K1R1B1R w - - 6 19",
+        "r1r5/b2bk1pp/4pn2/4B3/2BNp3/7P/1PP2PP1/1K1R3R w - - 8 20",
+        "r7/b2bk1pp/4pn2/2r1B3/2BNp3/1P5P/2P2PP1/1K1R3R w - - 1 21",
+        "rb6/3bk1pp/4pn2/2r1B3/2BNpP2/1P5P/2P3P1/1K1R3R w - - 1 22",
+        "1r6/3bk1pp/4pn2/2r5/2BNpP2/1P5P/2P3P1/1K1R3R w - - 0 23",
+        "1r6/3bk1p1/4pn1p/2r5/2BNpP2/1P5P/2P3P1/2KR3R w - - 0 24",
+        "8/3bk1p1/1r2pn1p/2r5/2BNpP1P/1P6/2P3P1/2KR3R w - - 1 25",
+        "8/3bk3/1r2pnpp/2r5/2BNpP1P/1P6/2P3P1/2K1R2R w - - 0 26",
+        "2b5/4k3/1r2pnpp/2r5/2BNpP1P/1P4P1/2P5/2K1R2R w - - 1 27",
+        "8/1b2k3/1r2pnpp/2r5/2BNpP1P/1P4P1/2P5/2K1R1R1 w - - 3 28",
+        "8/1b1nk3/1r2p1pp/2r5/2BNpPPP/1P6/2P5/2K1R1R1 w - - 1 29",
+        "8/1b2k3/1r2p1pp/2r1nP2/2BNp1PP/1P6/2P5/2K1R1R1 w - - 1 30",
+        "8/1b2k3/1r2p1p1/2r1nPp1/2BNp2P/1P6/2P5/2K1R1R1 w - - 0 31",
+        "8/1b2k3/1r2p1n1/2r3p1/2BNp2P/1P6/2P5/2K1R1R1 w - - 0 32",
+        "8/1b2k3/1r2p1n1/6r1/2BNp2P/1P6/2P5/2K1R3 w - - 0 33",
+        "8/1b2k3/1r2p3/4n1P1/2BNp3/1P6/2P5/2K1R3 w - - 1 34",
+        "8/1b2k3/1r2p3/4n1P1/2BN4/1P2p3/2P5/2K4R w - - 0 35",
+        "8/1b2k3/1r2p2R/6P1/2nN4/1P2p3/2P5/2K5 w - - 0 36",
+        "8/1b2k3/3rp2R/6P1/2PN4/4p3/2P5/2K5 w - - 1 37",
+        "8/4k3/3rp2R/6P1/2PN4/2P1p3/6b1/2K5 w - - 1 38",
+        "8/4k3/r3p2R/2P3P1/3N4/2P1p3/6b1/2K5 w - - 1 39",
+        "8/3k4/r3p2R/2P2NP1/8/2P1p3/6b1/2K5 w - - 3 40",
+        "8/3k4/4p2R/2P3P1/8/2P1N3/6b1/r1K5 w - - 1 41",
+        "8/3k4/4p2R/2P3P1/8/2P1N3/3K2b1/6r1 w - - 3 42",
+        "8/3k4/4p2R/2P3P1/8/2PKNb2/8/6r1 w - - 5 43",
+        "8/4k3/4p1R1/2P3P1/8/2PKNb2/8/6r1 w - - 7 44",
+        "8/4k3/4p1R1/2P3P1/3K4/2P1N3/8/6rb w - - 9 45",
+        "8/3k4/4p1R1/2P1K1P1/8/2P1N3/8/6rb w - - 11 46",
+        "8/3k4/4p1R1/2P3P1/5K2/2P1N3/8/4r2b w - - 13 47",
+        "8/3k4/2b1p2R/2P3P1/5K2/2P1N3/8/4r3 w - - 15 48",
+        "8/3k4/2b1p3/2P3P1/5K2/2P1N2R/8/6r1 w - - 17 49",
+        "2k5/7R/2b1p3/2P3P1/5K2/2P1N3/8/6r1 w - - 19 50",
+        "2k5/7R/4p3/2P3P1/b1P2K2/4N3/8/6r1 w - - 1 51",
+        "2k5/3bR3/4p3/2P3P1/2P2K2/4N3/8/6r1 w - - 3 52",
+        "3k4/3b2R1/4p3/2P3P1/2P2K2/4N3/8/6r1 w - - 5 53",
+        "3kb3/6R1/4p1P1/2P5/2P2K2/4N3/8/6r1 w - - 1 54",
+        "3kb3/6R1/4p1P1/2P5/2P2KN1/8/8/2r5 w - - 3 55",
+        "3kb3/6R1/4p1P1/2P1N3/2P2K2/8/8/5r2 w - - 5 56",
+        "3kb3/6R1/4p1P1/2P1N3/2P5/4K3/8/4r3 w - - 7 57",
+    },
+    {
+        "rnbq1rk1/ppp1npb1/4p1p1/3P3p/3PP3/2N2N2/PP2BPPP/R1BQ1RK1 b - - 0 8",
+        "rnbq1rk1/ppp1npb1/6p1/3pP2p/3P4/2N2N2/PP2BPPP/R1BQ1RK1 b - - 0 9",
+        "rn1q1rk1/ppp1npb1/6p1/3pP2p/3P2b1/2N2N2/PP2BPPP/R1BQR1K1 b - - 2 10",
+        "r2q1rk1/ppp1npb1/2n3p1/3pP2p/3P2bN/2N5/PP2BPPP/R1BQR1K1 b - - 4 11",
+        "r4rk1/pppqnpb1/2n3p1/3pP2p/3P2bN/2N4P/PP2BPP1/R1BQR1K1 b - - 0 12",
+        "r4rk1/pppqnpb1/2n3p1/3pP2p/3P3N/7P/PP2NPP1/R1BQR1K1 b - - 0 13",
+        "r4rk1/pppq1pb1/2n3p1/3pPN1p/3P4/7P/PP2NPP1/R1BQR1K1 b - - 0 14",
+        "r4rk1/ppp2pb1/2n3p1/3pPq1p/3P1N2/7P/PP3PP1/R1BQR1K1 b - - 1 15",
+        "r4rk1/pppq1pb1/2n3p1/3pP2p/P2P1N2/7P/1P3PP1/R1BQR1K1 b - - 0 16",
+        "r2n1rk1/pppq1pb1/6p1/3pP2p/P2P1N2/R6P/1P3PP1/2BQR1K1 b - - 2 17",
+        "r4rk1/pppq1pb1/4N1p1/3pP2p/P2P4/R6P/1P3PP1/2BQR1K1 b - - 0 18",
+        "r4rk1/ppp2pb1/4q1p1/3pP1Bp/P2P4/R6P/1P3PP1/3QR1K1 b - - 1 19",
+        "r3r1k1/ppp2pb1/4q1p1/3pP1Bp/P2P1P2/R6P/1P4P1/3QR1K1 b - - 0 20",
+        "r3r1k1/ppp3b1/4qpp1/3pP2p/P2P1P1B/R6P/1P4P1/3QR1K1 b - - 1 21",
+        "r3r1k1/ppp3b1/4q1p1/3pP2p/P4P1B/R6P/1P4P1/3QR1K1 b - - 0 22",
+        "r4rk1/ppp3b1/4q1p1/3pP1Bp/P4P2/R6P/1P4P1/3QR1K1 b - - 2 23",
+        "r4rk1/pp4b1/4q1p1/2ppP1Bp/P4P2/3R3P/1P4P1/3QR1K1 b - - 1 24",
+        "r4rk1/pp4b1/4q1p1/2p1P1Bp/P2p1PP1/3R3P/1P6/3QR1K1 b - - 0 25",
+        "r4rk1/pp4b1/4q1p1/2p1P1B1/P2p1PP1/3R4/1P6/3QR1K1 b - - 0 26",
+        "r5k1/pp3rb1/4q1p1/2p1P1B1/P2p1PP1/6R1/1P6/3QR1K1 b - - 2 27",
+        "5rk1/pp3rb1/4q1p1/2p1P1B1/P2pRPP1/6R1/1P6/3Q2K1 b - - 4 28",
+        "5rk1/1p3rb1/p3q1p1/P1p1P1B1/3pRPP1/6R1/1P6/3Q2K1 b - - 0 29",
+        "4r1k1/1p3rb1/p3q1p1/P1p1P1B1/3pRPP1/1P4R1/8/3Q2K1 b - - 0 30",
+        "4r1k1/5rb1/pP2q1p1/2p1P1B1/3pRPP1/1P4R1/8/3Q2K1 b - - 0 31",
+        "4r1k1/5rb1/pq4p1/2p1P1B1/3pRPP1/1P4R1/4Q3/6K1 b - - 1 32",
+        "4r1k1/1r4b1/pq4p1/2p1P1B1/3pRPP1/1P4R1/2Q5/6K1 b - - 3 33",
+        "4r1k1/1r4b1/1q4p1/p1p1P1B1/3p1PP1/1P4R1/2Q5/4R1K1 b - - 1 34",
+        "4r1k1/3r2b1/1q4p1/p1p1P1B1/2Qp1PP1/1P4R1/8/4R1K1 b - - 3 35",
+        "4r1k1/3r2b1/4q1p1/p1p1P1B1/2Qp1PP1/1P4R1/5K2/4R3 b - - 5 36",
+        "4r1k1/3r2b1/6p1/p1p1P1B1/2Pp1PP1/6R1/5K2/4R3 b - - 0 37",
+        "4r1k1/3r2b1/6p1/p1p1P1B1/2P2PP1/3p2R1/5K2/3R4 b - - 1 38",
+        "5rk1/3r2b1/6p1/p1p1P1B1/2P2PP1/3p2R1/8/3RK3 b - - 3 39",
+        "5rk1/6b1/6p1/p1p1P1B1/2Pr1PP1/3R4/8/3RK3 b - - 0 40",
+        "5rk1/3R2b1/6p1/p1p1P1B1/2r2PP1/8/8/3RK3 b - - 1 41",
+        "5rk1/3R2b1/6p1/p1p1P1B1/4rPP1/8/3K4/3R4 b - - 3 42",
+        "1r4k1/3R2b1/6p1/p1p1P1B1/4rPP1/2K5/8/3R4 b - - 5 43",
+        "1r4k1/3R2b1/6p1/p1p1P1B1/2K2PP1/4r3/8/3R4 b - - 7 44",
+        "1r3bk1/8/3R2p1/p1p1P1B1/2K2PP1/4r3/8/3R4 b - - 9 45",
+        "1r3bk1/8/6R1/2p1P1B1/p1K2PP1/4r3/8/3R4 b - - 0 46",
+        "1r3b2/5k2/R7/2p1P1B1/p1K2PP1/4r3/8/3R4 b - - 2 47",
+        "5b2/1r3k2/R7/2p1P1B1/p1K2PP1/4r3/8/7R b - - 4 48",
+        "5b2/5k2/R7/2pKP1B1/pr3PP1/4r3/8/7R b - - 6 49",
+        "5b2/5k2/R1K5/2p1P1B1/p2r1PP1/4r3/8/7R b - - 8 50",
+        "8/R4kb1/2K5/2p1P1B1/p2r1PP1/4r3/8/7R b - - 10 51",
+        "8/R5b1/2K3k1/2p1PPB1/p2r2P1/4r3/8/7R b - - 0 52",
+        "8/6R1/2K5/2p1PPk1/p2r2P1/4r3/8/7R b - - 0 53",
+        "8/6R1/2K5/2p1PP2/p2r1kP1/4r3/8/5R2 b - - 2 54",
+        "8/6R1/2K2P2/2p1P3/p2r2P1/4r1k1/8/5R2 b - - 0 55",
+        "8/5PR1/2K5/2p1P3/p2r2P1/4r3/6k1/5R2 b - - 0 56",
+    },
+    {
+        "rn1qkb1r/p1pbpppp/5n2/8/2pP4/2N5/1PQ1PPPP/R1B1KBNR w KQkq - 0 7",
+        "r2qkb1r/p1pbpppp/2n2n2/8/2pP4/2N2N2/1PQ1PPPP/R1B1KB1R w KQkq - 2 8",
+        "r2qkb1r/p1pbpppp/5n2/8/1npPP3/2N2N2/1PQ2PPP/R1B1KB1R w KQkq - 1 9",
+        "r2qkb1r/p1pb1ppp/4pn2/8/1npPP3/2N2N2/1P3PPP/R1BQKB1R w KQkq - 0 10",
+        "r2qk2r/p1pbbppp/4pn2/8/1nBPP3/2N2N2/1P3PPP/R1BQK2R w KQkq - 1 11",
+        "r2q1rk1/p1pbbppp/4pn2/8/1nBPP3/2N2N2/1P3PPP/R1BQ1RK1 w - - 3 12",
+        "r2q1rk1/2pbbppp/p3pn2/8/1nBPPB2/2N2N2/1P3PPP/R2Q1RK1 w - - 0 13",
+        "r2q1rk1/2p1bppp/p3pn2/1b6/1nBPPB2/2N2N2/1P3PPP/R2QR1K1 w - - 2 14",
+        "r2q1rk1/4bppp/p1p1pn2/1b6/1nBPPB2/1PN2N2/5PPP/R2QR1K1 w - - 0 15",
+        "r4rk1/3qbppp/p1p1pn2/1b6/1nBPPB2/1PN2N2/3Q1PPP/R3R1K1 w - - 2 16",
+        "r4rk1/1q2bppp/p1p1pn2/1b6/1nBPPB2/1PN2N1P/3Q1PP1/R3R1K1 w - - 1 17",
+        "r3r1k1/1q2bppp/p1p1pn2/1b6/1nBPPB2/1PN2N1P/4QPP1/R3R1K1 w - - 3 18",
+        "r3r1k1/1q1nbppp/p1p1p3/1b6/1nBPPB2/1PN2N1P/4QPP1/3RR1K1 w - - 5 19",
+        "r3rbk1/1q1n1ppp/p1p1p3/1b6/1nBPPB2/1PN2N1P/3RQPP1/4R1K1 w - - 7 20",
+        "r3rbk1/1q3ppp/pnp1p3/1b6/1nBPPB2/1PN2N1P/3RQPP1/4R2K w - - 9 21",
+        "2r1rbk1/1q3ppp/pnp1p3/1b6/1nBPPB2/1PN2N1P/3RQPP1/1R5K w - - 11 22",
+        "2r1rbk1/1q4pp/pnp1pp2/1b6/1nBPPB2/1PN2N1P/4QPP1/1R1R3K w - - 0 23",
+        "2r1rbk1/5qpp/pnp1pp2/1b6/1nBPP3/1PN1BN1P/4QPP1/1R1R3K w - - 2 24",
+        "2r1rbk1/5qp1/pnp1pp1p/1b6/1nBPP3/1PN1BN1P/4QPP1/1R1R2K1 w - - 0 25",
+        "2r1rbk1/5qp1/pnp1pp1p/1b6/2BPP3/1P2BN1P/n3QPP1/1R1R2K1 w - - 0 26",
+        "r3rbk1/5qp1/pnp1pp1p/1b6/2BPP3/1P2BN1P/Q4PP1/1R1R2K1 w - - 1 27",
+        "rr3bk1/5qp1/pnp1pp1p/1b6/2BPP3/1P2BN1P/Q4PP1/R2R2K1 w - - 3 28",
+        "rr2qbk1/6p1/pnp1pp1p/1b6/2BPP3/1P2BN1P/4QPP1/R2R2K1 w - - 5 29",
+        "rr2qbk1/6p1/1np1pp1p/pb6/2BPP3/1P1QBN1P/5PP1/R2R2K1 w - - 0 30",
+        "rr2qbk1/6p1/1n2pp1p/pp6/3PP3/1P1QBN1P/5PP1/R2R2K1 w - - 0 31",
+        "rr2qbk1/6p1/1n2pp1p/1p1P4/p3P3/1P1QBN1P/5PP1/R2R2K1 w - - 0 32",
+        "rr2qbk1/3n2p1/3Ppp1p/1p6/p3P3/1P1QBN1P/5PP1/R2R2K1 w - - 1 33",
+        "rr3bk1/3n2p1/3Ppp1p/1p5q/pP2P3/3QBN1P/5PP1/R2R2K1 w - - 1 34",
+        "rr3bk1/3n2p1/3Ppp1p/1p5q/1P2P3/p2QBN1P/5PP1/2RR2K1 w - - 0 35",
+        "1r3bk1/3n2p1/r2Ppp1p/1p5q/1P2P3/pQ2BN1P/5PP1/2RR2K1 w - - 2 36",
+        "1r2qbk1/2Rn2p1/r2Ppp1p/1p6/1P2P3/pQ2BN1P/5PP1/3R2K1 w - - 4 37",
+        "1r2qbk1/2Rn2p1/r2Ppp1p/1pB5/1P2P3/1Q3N1P/p4PP1/3R2K1 w - - 0 38",
+        "1r2q1k1/2Rn2p1/r2bpp1p/1pB5/1P2P3/1Q3N1P/p4PP1/R5K1 w - - 0 39",
+        "1r2q1k1/2Rn2p1/3rpp1p/1p6/1P2P3/1Q3N1P/p4PP1/R5K1 w - - 0 40",
+        "2r1q1k1/2Rn2p1/3rpp1p/1p6/1P2P3/5N1P/Q4PP1/R5K1 w - - 1 41",
+        "1r2q1k1/1R1n2p1/3rpp1p/1p6/1P2P3/5N1P/Q4PP1/R5K1 w - - 3 42",
+        "2r1q1k1/2Rn2p1/3rpp1p/1p6/1P2P3/5N1P/Q4PP1/R5K1 w - - 5 43",
+        "1r2q1k1/1R1n2p1/3rpp1p/1p6/1P2P3/5N1P/Q4PP1/R5K1 w - - 7 44",
+        "1rq3k1/R2n2p1/3rpp1p/1p6/1P2P3/5N1P/Q4PP1/R5K1 w - - 9 45",
+        "2q3k1/Rr1n2p1/3rpp1p/1p6/1P2P3/5N1P/4QPP1/R5K1 w - - 11 46",
+        "Rrq3k1/3n2p1/3rpp1p/1p6/1P2P3/5N1P/4QPP1/R5K1 w - - 13 47",
+    },
+    {
+        "rn1qkb1r/1pp2ppp/p4p2/3p1b2/5P2/1P2PN2/P1PP2PP/RN1QKB1R b KQkq - 1 6",
+        "r2qkb1r/1pp2ppp/p1n2p2/3p1b2/3P1P2/1P2PN2/P1P3PP/RN1QKB1R b KQkq - 0 7",
+        "r2qkb1r/1pp2ppp/p4p2/3p1b2/1n1P1P2/1P1BPN2/P1P3PP/RN1QK2R b KQkq - 2 8",
+        "r2qkb1r/1pp2ppp/p4p2/3p1b2/3P1P2/1P1PPN2/P5PP/RN1QK2R b KQkq - 0 9",
+        "r2qk2r/1pp2ppp/p2b1p2/3p1b2/3P1P2/1PNPPN2/P5PP/R2QK2R b KQkq - 2 10",
+        "r2qk2r/1p3ppp/p1pb1p2/3p1b2/3P1P2/1PNPPN2/P5PP/R2Q1RK1 b kq - 1 11",
+        "r2q1rk1/1p3ppp/p1pb1p2/3p1b2/3P1P2/1PNPPN2/P2Q2PP/R4RK1 b - - 3 12",
+        "r2qr1k1/1p3ppp/p1pb1p2/3p1b2/3P1P2/1P1PPN2/P2QN1PP/R4RK1 b - - 5 13",
+        "r3r1k1/1p3ppp/pqpb1p2/3p1b2/3P1P2/1P1PPNN1/P2Q2PP/R4RK1 b - - 7 14",
+        "r3r1k1/1p3ppp/pqp2p2/3p1b2/1b1P1P2/1P1PPNN1/P1Q3PP/R4RK1 b - - 9 15",
+        "r3r1k1/1p1b1ppp/pqp2p2/3p4/1b1P1P2/1P1PPNN1/P4QPP/R4RK1 b - - 11 16",
+        "2r1r1k1/1p1b1ppp/pqp2p2/3p4/1b1PPP2/1P1P1NN1/P4QPP/R4RK1 b - - 0 17",
+        "2r1r1k1/1p1b1ppp/pq3p2/2pp4/1b1PPP2/PP1P1NN1/5QPP/R4RK1 b - - 0 18",
+        "2r1r1k1/1p1b1ppp/pq3p2/2Pp4/4PP2/PPbP1NN1/5QPP/R4RK1 b - - 0 19",
+        "2r1r1k1/1p1b1ppp/p4p2/2Pp4/4PP2/PqbP1NN1/5QPP/RR4K1 b - - 1 20",
+        "2r1r1k1/1p1b1ppp/p4p2/2Pp4/q3PP2/P1bP1NN1/R4QPP/1R4K1 b - - 3 21",
+        "2r1r1k1/1p3ppp/p4p2/1bPP4/q4P2/P1bP1NN1/R4QPP/1R4K1 b - - 0 22",
+        "2r1r1k1/1p3ppp/p4p2/2PP4/q4P2/P1bb1NN1/R4QPP/2R3K1 b - - 1 23",
+        "2r1r1k1/1p3ppp/p2P1p2/2P5/2q2P2/P1bb1NN1/R4QPP/2R3K1 b - - 0 24",
+        "2rr2k1/1p3ppp/p2P1p2/2P5/2q2P2/P1bb1NN1/R4QPP/2R4K b - - 2 25",
+        "2rr2k1/1p3ppp/p2P1p2/2Q5/5P2/P1bb1NN1/R5PP/2R4K b - - 0 26",
+        "3r2k1/1p3ppp/p2P1p2/2r5/5P2/P1bb1N2/R3N1PP/2R4K b - - 1 27",
+        "3r2k1/1p3ppp/p2P1p2/2r5/5P2/P1b2N2/4R1PP/2R4K b - - 0 28",
+        "3r2k1/1p3ppp/p2P1p2/2r5/1b3P2/P4N2/4R1PP/3R3K b - - 2 29",
+        "3r2k1/1p2Rppp/p2P1p2/b1r5/5P2/P4N2/6PP/3R3K b - - 4 30",
+        "3r2k1/1R3ppp/p1rP1p2/b7/5P2/P4N2/6PP/3R3K b - - 0 31",
+        "3r2k1/1R3ppp/p2R1p2/b7/5P2/P4N2/6PP/7K b - - 0 32",
+        "6k1/1R3ppp/p2r1p2/b7/5P2/P4NP1/7P/7K b - - 0 33",
+        "6k1/1R3p1p/p2r1pp1/b7/5P1P/P4NP1/8/7K b - - 0 34",
+        "6k1/3R1p1p/pr3pp1/b7/5P1P/P4NP1/8/7K b - - 2 35",
+        "6k1/5p2/pr3pp1/b2R3p/5P1P/P4NP1/8/7K b - - 1 36",
+        "6k1/5p2/pr3pp1/7p/5P1P/P1bR1NP1/8/7K b - - 3 37",
+        "6k1/5p2/p1r2pp1/7p/5P1P/P1bR1NP1/6K1/8 b - - 5 38",
+        "6k1/5p2/p1r2pp1/b2R3p/5P1P/P4NP1/6K1/8 b - - 7 39",
+        "6k1/5p2/p4pp1/b2R3p/5P1P/P4NPK/2r5/8 b - - 9 40",
+        "6k1/2b2p2/p4pp1/7p/5P1P/P2R1NPK/2r5/8 b - - 11 41",
+        "6k1/2b2p2/5pp1/p6p/3N1P1P/P2R2PK/2r5/8 b - - 1 42",
+        "6k1/2b2p2/5pp1/p6p/3N1P1P/P1R3PK/r7/8 b - - 3 43",
+        "6k1/5p2/1b3pp1/p6p/5P1P/P1R3PK/r1N5/8 b - - 5 44",
+        "8/5pk1/1bR2pp1/p6p/5P1P/P5PK/r1N5/8 b - - 7 45",
+        "3b4/5pk1/2R2pp1/p4P1p/7P/P5PK/r1N5/8 b - - 0 46",
+        "8/4bpk1/2R2pp1/p4P1p/6PP/P6K/r1N5/8 b - - 0 47",
+        "8/5pk1/2R2pP1/p6p/6PP/b6K/r1N5/8 b - - 0 48",
+        "8/6k1/2R2pp1/p6P/7P/b6K/r1N5/8 b - - 0 49",
+        "8/6k1/2R2p2/p6p/7P/b5K1/r1N5/8 b - - 1 50",
+        "8/8/2R2pk1/p6p/7P/b4K2/r1N5/8 b - - 3 51",
+        "8/8/2R2pk1/p6p/7P/4NK2/rb6/8 b - - 5 52",
+        "2R5/8/5pk1/7p/p6P/4NK2/rb6/8 b - - 1 53",
+        "6R1/8/5pk1/7p/p6P/4NK2/1b6/r7 b - - 3 54",
+        "R7/5k2/5p2/7p/p6P/4NK2/1b6/r7 b - - 5 55",
+        "R7/5k2/5p2/7p/7P/p3N3/1b2K3/r7 b - - 1 56",
+        "8/R4k2/5p2/7p/7P/p3N3/1b2K3/7r b - - 3 57",
+        "8/8/5pk1/7p/R6P/p3N3/1b2K3/7r b - - 5 58",
+        "8/8/5pk1/7p/R6P/p7/4K3/2bN3r b - - 7 59",
+        "8/8/5pk1/7p/R6P/p7/4KN1r/2b5 b - - 9 60",
+        "8/8/5pk1/7p/R6P/p3K3/1b3N1r/8 b - - 11 61",
+        "8/8/R4pk1/7p/7P/p1b1K3/5N1r/8 b - - 13 62",
+        "8/8/5pk1/7p/7P/2b1K3/R4N1r/8 b - - 0 63",
+        "8/8/5pk1/7p/3K3P/8/R4N1r/4b3 b - - 2 64",
+    }
+};
+// clang-format on
 
-namespace Stockfish {
+}  // namespace
 
-/// setup_bench() builds a list of UCI commands to be run by bench. There
+namespace Stockfish::Benchmark {
-/// 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
 
-vector<string> setup_bench(const Position& current, istream& is) {
+// Builds a list of UCI commands to be run by bench. There
+// are five parameters: TT size in MB, number of search threads that
+// should be used, the limit value spent for each position, a file name
+// where to look for positions in FEN format, and the type of the limit:
+// depth, perft, nodes and movetime (in milliseconds). Examples:
+//
+// bench                            : search default positions up to depth 13
+// bench 64 1 15                    : search default positions up to depth 15 (TT = 64MB)
+// bench 64 1 100000 default nodes  : search default positions for 100K nodes each
+// bench 64 4 5000 current movetime : search current position with 4 threads for 5 sec
+// bench 16 1 5 blah perft          : run a perft 5 on positions in file "blah"
+std::vector<std::string> setup_bench(const std::string& currentFen, std::istream& is) {
 
-  vector<string> fens, list;
+    std::vector<std::string> fens, list;
-  string go, token;
+    std::string              go, token;
 
-  // Assign default values to missing arguments
+    // Assign default values to missing arguments
-  string ttSize    = (is >> token) ? token : "16";
+    std::string ttSize    = (is >> token) ? token : "16";
-  string threads   = (is >> token) ? token : "1";
+    std::string threads   = (is >> token) ? token : "1";
-  string limit     = (is >> token) ? token : "13";
+    std::string limit     = (is >> token) ? token : "13";
-  string fenFile   = (is >> token) ? token : "default";
+    std::string fenFile   = (is >> token) ? token : "default";
-  string limitType = (is >> token) ? token : "depth";
+    std::string limitType = (is >> token) ? token : "depth";
-  string evalType  = (is >> token) ? token : "mixed";
 
-  go = limitType == "eval" ? "eval" : "go " + limitType + " " + limit;
+    go = limitType == "eval" ? "eval" : "go " + limitType + " " + limit;
 
-  if (fenFile == "default")
+    if (fenFile == "default")
-      fens = Defaults;
+        fens = Defaults;
 
-  else if (fenFile == "current")
+    else if (fenFile == "current")
-      fens.push_back(current.fen());
+        fens.push_back(currentFen);
 
-  else
+    else
-  {
+    {
-      string fen;
+        std::string   fen;
-      ifstream file(fenFile);
+        std::ifstream file(fenFile);
 
-      if (!file.is_open())
+        if (!file.is_open())
-      {
+        {
-          cerr << "Unable to open file " << fenFile << endl;
+            std::cerr << "Unable to open file " << fenFile << std::endl;
-          exit(EXIT_FAILURE);
+            exit(EXIT_FAILURE);
-      }
+        }
 
-      while (getline(file, fen))
+        while (getline(file, fen))
-          if (!fen.empty())
+            if (!fen.empty())
-              fens.push_back(fen);
+                fens.push_back(fen);
 
-      file.close();
+        file.close();
-  }
+    }
 
-  list.emplace_back("setoption name Threads value " + threads);
+    list.emplace_back("setoption name Threads value " + threads);
-  list.emplace_back("setoption name Hash value " + ttSize);
+    list.emplace_back("setoption name Hash value " + ttSize);
-  list.emplace_back("ucinewgame");
+    list.emplace_back("ucinewgame");
 
-  size_t posCounter = 0;
+    for (const std::string& fen : fens)
+        if (fen.find("setoption") != std::string::npos)
+            list.emplace_back(fen);
+        else
+        {
+            list.emplace_back("position fen " + fen);
+            list.emplace_back(go);
+        }
 
-  for (const string& fen : fens)
+    return list;
-      if (fen.find("setoption") != string::npos)
-          list.emplace_back(fen);
-      else
-      {
-          if (evalType == "classical" || (evalType == "mixed" && posCounter % 2 == 0))
-              list.emplace_back("setoption name Use NNUE value false");
-          else if (evalType == "NNUE" || (evalType == "mixed" && posCounter % 2 != 0))
-              list.emplace_back("setoption name Use NNUE value true");
-          list.emplace_back("position fen " + fen);
-          list.emplace_back(go);
-          ++posCounter;
-      }
-
-  list.emplace_back("setoption name Use NNUE value true");
-
-  return list;
 }
 
-} // namespace Stockfish
+BenchmarkSetup setup_benchmark(std::istream& is) {
+    // TT_SIZE_PER_THREAD is chosen such that roughly half of the hash is used all positions
+    // for the current sequence have been searched.
+    static constexpr int TT_SIZE_PER_THREAD = 128;
+
+    static constexpr int DEFAULT_DURATION_S = 150;
+
+    BenchmarkSetup setup{};
+
+    // Assign default values to missing arguments
+    int desiredTimeS;
+
+    if (!(is >> setup.threads))
+        setup.threads = get_hardware_concurrency();
+    else
+        setup.originalInvocation += std::to_string(setup.threads);
+
+    if (!(is >> setup.ttSize))
+        setup.ttSize = TT_SIZE_PER_THREAD * setup.threads;
+    else
+        setup.originalInvocation += " " + std::to_string(setup.ttSize);
+
+    if (!(is >> desiredTimeS))
+        desiredTimeS = DEFAULT_DURATION_S;
+    else
+        setup.originalInvocation += " " + std::to_string(desiredTimeS);
+
+    setup.filledInvocation += std::to_string(setup.threads) + " " + std::to_string(setup.ttSize)
+                            + " " + std::to_string(desiredTimeS);
+
+    auto getCorrectedTime = [&](int ply) {
+        // time per move is fit roughly based on LTC games
+        // seconds = 50/{ply+15}
+        // ms = 50000/{ply+15}
+        // with this fit 10th move gets 2000ms
+        // adjust for desired 10th move time
+        return 50000.0 / (static_cast<double>(ply) + 15.0);
+    };
+
+    float totalTime = 0;
+    for (const auto& game : BenchmarkPositions)
+    {
+        setup.commands.emplace_back("ucinewgame");
+        int ply = 1;
+        for (int i = 0; i < static_cast<int>(game.size()); ++i)
+        {
+            const float correctedTime = getCorrectedTime(ply);
+            totalTime += correctedTime;
+            ply += 1;
+        }
+    }
+
+    float timeScaleFactor = static_cast<float>(desiredTimeS * 1000) / totalTime;
+
+    for (const auto& game : BenchmarkPositions)
+    {
+        setup.commands.emplace_back("ucinewgame");
+        int ply = 1;
+        for (const std::string& fen : game)
+        {
+            setup.commands.emplace_back("position fen " + fen);
+
+            const int correctedTime = static_cast<int>(getCorrectedTime(ply) * timeScaleFactor);
+            setup.commands.emplace_back("go movetime " + std::to_string(correctedTime));
+
+            ply += 1;
+        }
+    }
+
+    return setup;
+}
+
+}  // namespace Stockfish

src/benchmark.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -23,12 +23,20 @@
 #include <string>
 #include <vector>
 
-namespace Stockfish {
+namespace Stockfish::Benchmark {
 
-class Position;
+std::vector<std::string> setup_bench(const std::string&, std::istream&);
 
-std::vector<std::string> setup_bench(const Position&, std::istream&);
+struct BenchmarkSetup {
+    int                      ttSize;
+    int                      threads;
+    std::vector<std::string> commands;
+    std::string              originalInvocation;
+    std::string              filledInvocation;
+};
 
-} // namespace Stockfish
+BenchmarkSetup setup_benchmark(std::istream&);
 
-#endif // #ifndef BENCHMARK_H_INCLUDED
+}  // namespace Stockfish
+
+#endif  // #ifndef BENCHMARK_H_INCLUDED

src/bitbase.cpp

@@ -1,172 +0,0 @@
-/*
-  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 <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,
-    DRAW    = 2,
-    WIN     = 4
-  };
-
-  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);
-
-    Color stm;
-    Square ksq[COLOR_NB], psq;
-    Result result;
-  };
-
-} // namespace
-
-bool Bitbases::probe(Square wksq, Square wpsq, Square bksq, Color stm) {
-
-  assert(file_of(wpsq) <= FILE_D);
-
-  return KPKBitbase[index(stm, bksq, wksq, wpsq)];
-}
-
-
-void Bitbases::init() {
-
-  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);
-
-  // 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 < MAX_INDEX; ++idx)
-          repeat |= (db[idx] == UNKNOWN && db[idx].classify(db) != UNKNOWN);
-
-  // Fill the bitbase with the decisive results
-  for (idx = 0; idx < MAX_INDEX; ++idx)
-      if (db[idx] == WIN)
-          KPKBitbase.set(idx);
-}
-
-namespace {
-
-  KPKPosition::KPKPosition(unsigned idx) {
-
-    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)));
-
-    // 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;
-
-    // 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;
-  }
-
-  Result KPKPosition::classify(const std::vector<KPKPosition>& db) {
-
-    // 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.
-    //
-    // 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 = 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)
-    {
-        if (rank_of(psq) < RANK_7)      // Single push
-            r |= db[index(BLACK, ksq[BLACK], ksq[WHITE], psq + NORTH)];
-
-        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)];
-    }
-
-    return result = r & Good  ? Good  : r & UNKNOWN ? UNKNOWN : Bad;
-  }
-
-} // namespace
-
-} // namespace Stockfish

src/bitboard.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,10 +16,12 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#include "bitboard.h"
+
 #include <algorithm>
 #include <bitset>
+#include <initializer_list>
 
-#include "bitboard.h"
 #include "misc.h"
 
 namespace Stockfish {
@@ -32,145 +34,154 @@ Bitboard BetweenBB[SQUARE_NB][SQUARE_NB];
 Bitboard PseudoAttacks[PIECE_TYPE_NB][SQUARE_NB];
 Bitboard PawnAttacks[COLOR_NB][SQUARE_NB];
 
-Magic RookMagics[SQUARE_NB];
+alignas(64) Magic Magics[SQUARE_NB][2];
-Magic BishopMagics[SQUARE_NB];
 
 namespace {
 
-  Bitboard RookTable[0x19000];  // To store rook attacks
+Bitboard RookTable[0x19000];   // To store rook attacks
-  Bitboard BishopTable[0x1480]; // To store bishop attacks
+Bitboard BishopTable[0x1480];  // To store bishop attacks
 
-  void init_magics(PieceType pt, Bitboard table[], Magic magics[]);
+void init_magics(PieceType pt, Bitboard table[], Magic magics[][2]);
 
-}
+// Returns the bitboard of target square for the given step
-
+// from the given square. If the step is off the board, returns empty bitboard.
-/// safe_destination() returns the bitboard of target square for the given step
+Bitboard safe_destination(Square s, int step) {
-/// from the given square. If the step is off the board, returns empty bitboard.
-
-inline Bitboard safe_destination(Square s, int step) {
     Square to = Square(s + step);
     return is_ok(to) && distance(s, to) <= 2 ? square_bb(to) : Bitboard(0);
 }
+}
 
-
+// Returns an ASCII representation of a bitboard suitable
-/// Bitboards::pretty() returns an ASCII representation of a bitboard suitable
+// to be printed to standard output. Useful for debugging.
-/// to be printed to standard output. Useful for debugging.
-
 std::string Bitboards::pretty(Bitboard b) {
 
-  std::string s = "+---+---+---+---+---+---+---+---+\n";
+    std::string s = "+---+---+---+---+---+---+---+---+\n";
 
-  for (Rank r = RANK_8; r >= RANK_1; --r)
+    for (Rank r = RANK_8; r >= RANK_1; --r)
-  {
+    {
-      for (File f = FILE_A; f <= FILE_H; ++f)
+        for (File f = FILE_A; f <= FILE_H; ++f)
-          s += b & make_square(f, r) ? "| X " : "|   ";
+            s += b & make_square(f, r) ? "| X " : "|   ";
 
-      s += "| " + std::to_string(1 + r) + "\n+---+---+---+---+---+---+---+---+\n";
+        s += "| " + std::to_string(1 + r) + "\n+---+---+---+---+---+---+---+---+\n";
-  }
+    }
-  s += "  a   b   c   d   e   f   g   h\n";
+    s += "  a   b   c   d   e   f   g   h\n";
 
-  return s;
+    return s;
 }
 
 
-/// Bitboards::init() initializes various bitboard tables. It is called at
+// Initializes various bitboard tables. It is called at
-/// startup and relies on global objects to be already zero-initialized.
+// startup and relies on global objects to be already zero-initialized.
-
 void Bitboards::init() {
 
-  for (unsigned i = 0; i < (1 << 16); ++i)
+    for (unsigned i = 0; i < (1 << 16); ++i)
-      PopCnt16[i] = uint8_t(std::bitset<16>(i).count());
+        PopCnt16[i] = uint8_t(std::bitset<16>(i).count());
 
-  for (Square s1 = SQ_A1; s1 <= SQ_H8; ++s1)
+    for (Square s1 = SQ_A1; s1 <= SQ_H8; ++s1)
-      for (Square s2 = SQ_A1; s2 <= SQ_H8; ++s2)
+        for (Square s2 = SQ_A1; s2 <= SQ_H8; ++s2)
-          SquareDistance[s1][s2] = std::max(distance<File>(s1, s2), distance<Rank>(s1, s2));
+            SquareDistance[s1][s2] = std::max(distance<File>(s1, s2), distance<Rank>(s1, s2));
 
-  init_magics(ROOK, RookTable, RookMagics);
+    init_magics(ROOK, RookTable, Magics);
-  init_magics(BISHOP, BishopTable, BishopMagics);
+    init_magics(BISHOP, BishopTable, Magics);
 
-  for (Square s1 = SQ_A1; s1 <= SQ_H8; ++s1)
+    for (Square s1 = SQ_A1; s1 <= SQ_H8; ++s1)
-  {
+    {
-      PawnAttacks[WHITE][s1] = pawn_attacks_bb<WHITE>(square_bb(s1));
+        PawnAttacks[WHITE][s1] = pawn_attacks_bb<WHITE>(square_bb(s1));
-      PawnAttacks[BLACK][s1] = pawn_attacks_bb<BLACK>(square_bb(s1));
+        PawnAttacks[BLACK][s1] = pawn_attacks_bb<BLACK>(square_bb(s1));
 
-      for (int step : {-9, -8, -7, -1, 1, 7, 8, 9} )
+        for (int step : {-9, -8, -7, -1, 1, 7, 8, 9})
-         PseudoAttacks[KING][s1] |= safe_destination(s1, step);
+            PseudoAttacks[KING][s1] |= safe_destination(s1, step);
 
-      for (int step : {-17, -15, -10, -6, 6, 10, 15, 17} )
+        for (int step : {-17, -15, -10, -6, 6, 10, 15, 17})
-         PseudoAttacks[KNIGHT][s1] |= safe_destination(s1, step);
+            PseudoAttacks[KNIGHT][s1] |= safe_destination(s1, step);
 
-      PseudoAttacks[QUEEN][s1]  = PseudoAttacks[BISHOP][s1] = attacks_bb<BISHOP>(s1, 0);
+        PseudoAttacks[QUEEN][s1] = PseudoAttacks[BISHOP][s1] = attacks_bb<BISHOP>(s1, 0);
-      PseudoAttacks[QUEEN][s1] |= PseudoAttacks[  ROOK][s1] = attacks_bb<  ROOK>(s1, 0);
+        PseudoAttacks[QUEEN][s1] |= PseudoAttacks[ROOK][s1]  = attacks_bb<ROOK>(s1, 0);
 
-      for (PieceType pt : { BISHOP, ROOK })
+        for (PieceType pt : {BISHOP, ROOK})
-          for (Square s2 = SQ_A1; s2 <= SQ_H8; ++s2)
+            for (Square s2 = SQ_A1; s2 <= SQ_H8; ++s2)
-          {
+            {
-              if (PseudoAttacks[pt][s1] & s2)
+                if (PseudoAttacks[pt][s1] & s2)
-              {
+                {
-                  LineBB[s1][s2]    = (attacks_bb(pt, s1, 0) & attacks_bb(pt, s2, 0)) | s1 | s2;
+                    LineBB[s1][s2] = (attacks_bb(pt, s1, 0) & attacks_bb(pt, s2, 0)) | s1 | s2;
-                  BetweenBB[s1][s2] = (attacks_bb(pt, s1, square_bb(s2)) & attacks_bb(pt, s2, square_bb(s1)));
+                    BetweenBB[s1][s2] =
-              }
+                      (attacks_bb(pt, s1, square_bb(s2)) & attacks_bb(pt, s2, square_bb(s1)));
-              BetweenBB[s1][s2] |= s2;
+                }
-          }
+                BetweenBB[s1][s2] |= s2;
-  }
+            }
+    }
 }
 
 namespace {
 
-  Bitboard sliding_attack(PieceType pt, Square sq, Bitboard occupied) {
+Bitboard sliding_attack(PieceType pt, Square sq, Bitboard occupied) {
 
-    Bitboard attacks = 0;
+    Bitboard  attacks             = 0;
-    Direction   RookDirections[4] = {NORTH, SOUTH, EAST, WEST};
+    Direction RookDirections[4]   = {NORTH, SOUTH, EAST, WEST};
     Direction BishopDirections[4] = {NORTH_EAST, SOUTH_EAST, SOUTH_WEST, NORTH_WEST};
 
     for (Direction d : (pt == ROOK ? RookDirections : BishopDirections))
     {
         Square s = sq;
-        while (safe_destination(s, d) && !(occupied & s))
+        while (safe_destination(s, d))
+        {
             attacks |= (s += d);
+            if (occupied & s)
+            {
+                break;
+            }
+        }
     }
 
     return attacks;
-  }
+}
 
 
-  // init_magics() computes all rook and bishop attacks at startup. Magic
+// Computes all rook and bishop attacks at startup. Magic
-  // bitboards are used to look up attacks of sliding pieces. As a reference see
+// bitboards are used to look up attacks of sliding pieces. As a reference see
-  // www.chessprogramming.org/Magic_Bitboards. In particular, here we use the so
+// https://www.chessprogramming.org/Magic_Bitboards. In particular, here we use
-  // called "fancy" approach.
+// the so called "fancy" approach.
-
+void init_magics(PieceType pt, Bitboard table[], Magic magics[][2]) {
-  void init_magics(PieceType pt, Bitboard table[], Magic magics[]) {
 
+#ifndef USE_PEXT
     // Optimal PRNG seeds to pick the correct magics in the shortest time
-    int seeds[][RANK_NB] = { { 8977, 44560, 54343, 38998,  5731, 95205, 104912, 17020 },
+    int seeds[][RANK_NB] = {{8977, 44560, 54343, 38998, 5731, 95205, 104912, 17020},
-                             {  728, 10316, 55013, 32803, 12281, 15100,  16645,   255 } };
+                            {728, 10316, 55013, 32803, 12281, 15100, 16645, 255}};
 
-    Bitboard occupancy[4096], reference[4096], edges, b;
+    Bitboard occupancy[4096];
-    int epoch[4096] = {}, cnt = 0, size = 0;
+    int      epoch[4096] = {}, cnt = 0;
+#endif
+    Bitboard reference[4096];
+    int      size = 0;
 
     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));
+        Bitboard edges = ((Rank1BB | Rank8BB) & ~rank_bb(s)) | ((FileABB | FileHBB) & ~file_bb(s));
 
         // Given a square 's', the mask is the bitboard of sliding attacks from
         // 's' computed on an empty board. The index must be big enough to contain
         // 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.
-        Magic& m = magics[s];
+        Magic& m = magics[s][pt - BISHOP];
-        m.mask  = sliding_attack(pt, s, 0) & ~edges;
+        m.mask   = sliding_attack(pt, s, 0) & ~edges;
+#ifndef USE_PEXT
         m.shift = (Is64Bit ? 64 : 32) - popcount(m.mask);
-
+#endif
         // 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;
+        m.attacks = s == SQ_A1 ? table : magics[s - 1][pt - BISHOP].attacks + size;
+        size      = 0;
 
         // Use Carry-Rippler trick to enumerate all subsets of masks[s] and
         // store the corresponding sliding attack bitboard in reference[].
-        b = size = 0;
+        Bitboard b = 0;
-        do {
+        do
+        {
+#ifndef USE_PEXT
             occupancy[size] = b;
+#endif
             reference[size] = sliding_attack(pt, s, b);
 
             if (HasPext)
@@ -180,16 +191,14 @@ namespace {
             b = (b - m.mask) & m.mask;
         } while (b);
 
-        if (HasPext)
+#ifndef USE_PEXT
-            continue;
-
         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.
-        for (int i = 0; i < size; )
+        for (int i = 0; i < size;)
         {
-            for (m.magic = 0; popcount((m.magic * m.mask) >> 56) < 6; )
+            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
@@ -204,15 +213,16 @@ namespace {
 
                 if (epoch[idx] < cnt)
                 {
-                    epoch[idx] = cnt;
+                    epoch[idx]     = cnt;
                     m.attacks[idx] = reference[i];
                 }
                 else if (m.attacks[idx] != reference[i])
                     break;
             }
         }
+#endif
     }
-  }
+}
 }
 
-} // namespace Stockfish
+}  // namespace Stockfish

src/bitboard.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,28 +19,24 @@
 #ifndef BITBOARD_H_INCLUDED
 #define BITBOARD_H_INCLUDED
 
+#include <algorithm>
+#include <cassert>
+#include <cmath>
+#include <cstring>
+#include <cstdint>
+#include <cstdlib>
 #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        init();
 std::string pretty(Bitboard b);
 
-} // namespace Stockfish::Bitboards
+}  // namespace Stockfish::Bitboards
-
-constexpr Bitboard AllSquares = ~Bitboard(0);
-constexpr Bitboard DarkSquares = 0xAA55AA55AA55AA55ULL;
 
 constexpr Bitboard FileABB = 0x0101010101010101ULL;
 constexpr Bitboard FileBBB = FileABB << 1;
@@ -60,17 +56,6 @@ 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];
 
@@ -80,371 +65,310 @@ 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
+// Magic holds all magic bitboards relevant data for a single square
 struct Magic {
-  Bitboard  mask;
+    Bitboard  mask;
-  Bitboard  magic;
+    Bitboard* attacks;
-  Bitboard* attacks;
+#ifndef USE_PEXT
-  unsigned  shift;
+    Bitboard magic;
+    unsigned shift;
+#endif
 
-  // Compute the attack's index using the 'magic bitboards' approach
+    // Compute the attack's index using the 'magic bitboards' approach
-  unsigned index(Bitboard occupied) const {
+    unsigned index(Bitboard occupied) const {
 
-    if (HasPext)
+#ifdef USE_PEXT
         return unsigned(pext(occupied, mask));
+#else
+        if (Is64Bit)
+            return unsigned(((occupied & mask) * magic) >> shift);
 
-    if (Is64Bit)
+        unsigned lo = unsigned(occupied) & unsigned(mask);
-        return unsigned(((occupied & mask) * magic) >> shift);
+        unsigned hi = unsigned(occupied >> 32) & unsigned(mask >> 32);
+        return (lo * unsigned(magic) ^ hi * unsigned(magic >> 32)) >> shift;
+#endif
+    }
 
-    unsigned lo = unsigned(occupied) & unsigned(mask);
+    Bitboard attacks_bb(Bitboard occupied) const { return attacks[index(occupied)]; }
-    unsigned hi = unsigned(occupied >> 32) & unsigned(mask >> 32);
-    return (lo * unsigned(magic) ^ hi * unsigned(magic >> 32)) >> shift;
-  }
 };
 
-extern Magic RookMagics[SQUARE_NB];
+extern Magic Magics[SQUARE_NB][2];
-extern Magic BishopMagics[SQUARE_NB];
 
-inline Bitboard square_bb(Square s) {
+constexpr Bitboard square_bb(Square s) {
-  assert(is_ok(s));
+    assert(is_ok(s));
-  return (1ULL << s);
+    return (1ULL << s);
 }
 
 
-/// Overloads of bitwise operators between a Bitboard and a Square for testing
+// 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.
+// whether a given bit is set in a bitboard, and for setting and clearing bits.
 
-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 ^ 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&(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|(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|(Square s1, Square s2) { return square_bb(s1) | s2; }
+inline Bitboard operator|(Square s1, Square s2) { return square_bb(s1) | s2; }
 
-constexpr bool more_than_one(Bitboard b) {
+constexpr bool more_than_one(Bitboard b) { return b & (b - 1); }
-  return b & (b - 1);
-}
 
 
-constexpr bool opposite_colors(Square s1, Square s2) {
+// rank_bb() and file_bb() return a bitboard representing all the squares on
-  return (s1 + rank_of(s1) + s2 + rank_of(s2)) & 1;
+// the given file or rank.
-}
+
+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)); }
 
 
-/// rank_bb() and file_bb() return a bitboard representing all the squares on
+// Moves a bitboard one or two steps as specified by the direction D
-/// the given file or rank.
-
-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));
-}
-
-
-/// shift() moves a bitboard one or two steps as specified by the direction D
-
 template<Direction D>
 constexpr Bitboard shift(Bitboard b) {
-  return  D == NORTH      ?  b             << 8 : D == SOUTH      ?  b             >> 8
+    return D == NORTH         ? b << 8
-        : D == NORTH+NORTH?  b             <<16 : D == SOUTH+SOUTH?  b             >>16
+         : D == SOUTH         ? b >> 8
-        : D == EAST       ? (b & ~FileHBB) << 1 : D == WEST       ? (b & ~FileABB) >> 1
+         : D == NORTH + NORTH ? b << 16
-        : D == NORTH_EAST ? (b & ~FileHBB) << 9 : D == NORTH_WEST ? (b & ~FileABB) << 7
+         : D == SOUTH + SOUTH ? b >> 16
-        : D == SOUTH_EAST ? (b & ~FileHBB) >> 7 : D == SOUTH_WEST ? (b & ~FileABB) >> 9
+         : D == EAST          ? (b & ~FileHBB) << 1
-        : 0;
+         : 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;
 }
 
 
-/// pawn_attacks_bb() returns the squares attacked by pawns of the given color
+// Returns the squares attacked by pawns of the given color
-/// from the squares in the given bitboard.
+// from the squares in the given bitboard.
-
 template<Color C>
 constexpr Bitboard pawn_attacks_bb(Bitboard b) {
-  return C == WHITE ? shift<NORTH_WEST>(b) | shift<NORTH_EAST>(b)
+    return C == WHITE ? shift<NORTH_WEST>(b) | shift<NORTH_EAST>(b)
-                    : shift<SOUTH_WEST>(b) | shift<SOUTH_EAST>(b);
+                      : shift<SOUTH_WEST>(b) | shift<SOUTH_EAST>(b);
 }
 
 inline Bitboard pawn_attacks_bb(Color c, Square s) {
 
-  assert(is_ok(s));
+    assert(is_ok(s));
-  return PawnAttacks[c][s];
+    return PawnAttacks[c][s];
 }
 
-
+// Returns a bitboard representing an entire line (from board edge
-/// pawn_double_attacks_bb() returns the squares doubly attacked by pawns of the
+// to board edge) that intersects the two given squares. If the given squares
-/// given color from the squares in the given bitboard.
+// 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.
-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));
+    assert(is_ok(s1) && is_ok(s2));
-
+    return LineBB[s1][s2];
-  return LineBB[s1][s2];
 }
 
 
-/// between_bb(s1, s2) returns a bitboard representing the squares in the semi-open
+// Returns a bitboard representing the squares in the semi-open
-/// segment between the squares s1 and s2 (excluding s1 but including s2). If the
+// 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,
+// 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_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
+// 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
+// allows to generate non-king evasion moves faster: the defending piece must either
-/// interpose itself to cover the check or capture the checking piece.
+// 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));
+    assert(is_ok(s1) && is_ok(s2));
-
+    return BetweenBB[s1][s2];
-  return BetweenBB[s1][s2];
 }
 
+// Returns true if the squares s1, s2 and s3 are aligned either on a
+// straight or on a diagonal line.
+inline bool aligned(Square s1, Square s2, Square s3) { return line_bb(s1, s2) & s3; }
 
-/// forward_ranks_bb() returns a bitboard representing the squares on the ranks in
-/// front of the given one, from the point of view of the given color. For instance,
-/// forward_ranks_bb(BLACK, SQ_D3) will return the 16 squares on ranks 1 and 2.
 
-constexpr Bitboard forward_ranks_bb(Color c, Square s) {
+// distance() functions return the distance between x and y, defined as the
-  return c == WHITE ? ~Rank1BB << 8 * relative_rank(WHITE, s)
+// number of steps for a king in x to reach y.
-                    : ~Rank8BB >> 8 * relative_rank(BLACK, s);
+
+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<>
-/// forward_file_bb() returns a bitboard representing all the squares along the
+inline int distance<Rank>(Square x, Square y) {
-/// line in front of the given one, from the point of view of the given color.
+    return std::abs(rank_of(x) - rank_of(y));
-
-constexpr Bitboard forward_file_bb(Color c, Square s) {
-  return forward_ranks_bb(c, s) & file_bb(s);
 }
 
-
+template<>
-/// pawn_attack_span() returns a bitboard representing all the squares that can
+inline int distance<Square>(Square x, Square y) {
-/// be attacked by a pawn of the given color when it moves along its file, starting
+    return SquareDistance[x][y];
-/// from the given square.
-
-constexpr Bitboard pawn_attack_span(Color c, Square s) {
-  return forward_ranks_bb(c, s) & adjacent_files_bb(s);
 }
 
-
-/// 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.
-
-constexpr Bitboard passed_pawn_span(Color c, Square s) {
-  return pawn_attack_span(c, s) | forward_file_bb(c, s);
-}
-
-
-/// aligned() returns true if the squares s1, s2 and s3 are aligned either on a
-/// straight or on a diagonal line.
-
-inline bool aligned(Square s1, Square s2, Square s3) {
-  return line_bb(s1, s2) & s3;
-}
-
-
-/// distance() functions return the distance between x and y, defined as the
-/// number of steps for a king in x to reach y.
-
-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]; }
-
 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)); }
-
-
-/// attacks_bb(Square) returns the pseudo attacks of the give piece type
-/// assuming an empty board.
 
+// Returns the pseudo attacks of the given piece type
+// assuming an empty board.
 template<PieceType Pt>
 inline Bitboard attacks_bb(Square s) {
 
-  assert((Pt != PAWN) && (is_ok(s)));
+    assert((Pt != PAWN) && (is_ok(s)));
-
+    return PseudoAttacks[Pt][s];
-  return PseudoAttacks[Pt][s];
 }
 
 
-/// attacks_bb(Square, Bitboard) returns the attacks by the given piece
+// Returns the attacks by the given piece
-/// assuming the board is occupied according to the passed Bitboard.
+// assuming the board is occupied according to the passed Bitboard.
-/// Sliding piece attacks do not continue passed an occupied square.
+// Sliding piece attacks do not continue passed an occupied square.
-
 template<PieceType Pt>
 inline Bitboard attacks_bb(Square s, Bitboard occupied) {
 
-  assert((Pt != PAWN) && (is_ok(s)));
+    assert((Pt != PAWN) && (is_ok(s)));
 
-  switch (Pt)
+    switch (Pt)
-  {
+    {
-  case BISHOP: return BishopMagics[s].attacks[BishopMagics[s].index(occupied)];
+    case BISHOP :
-  case ROOK  : return   RookMagics[s].attacks[  RookMagics[s].index(occupied)];
+    case ROOK :
-  case QUEEN : return attacks_bb<BISHOP>(s, occupied) | attacks_bb<ROOK>(s, occupied);
+        return Magics[s][Pt - BISHOP].attacks_bb(occupied);
-  default    : return PseudoAttacks[Pt][s];
+    case QUEEN :
-  }
+        return attacks_bb<BISHOP>(s, occupied) | attacks_bb<ROOK>(s, occupied);
+    default :
+        return PseudoAttacks[Pt][s];
+    }
 }
 
+// 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.
 inline Bitboard attacks_bb(PieceType pt, Square s, Bitboard occupied) {
 
-  assert((pt != PAWN) && (is_ok(s)));
+    assert((pt != PAWN) && (is_ok(s)));
 
-  switch (pt)
+    switch (pt)
-  {
+    {
-  case BISHOP: return attacks_bb<BISHOP>(s, occupied);
+    case BISHOP :
-  case ROOK  : return attacks_bb<  ROOK>(s, occupied);
+        return attacks_bb<BISHOP>(s, occupied);
-  case QUEEN : return attacks_bb<BISHOP>(s, occupied) | attacks_bb<ROOK>(s, occupied);
+    case ROOK :
-  default    : return PseudoAttacks[pt][s];
+        return attacks_bb<ROOK>(s, occupied);
-  }
+    case QUEEN :
+        return attacks_bb<BISHOP>(s, occupied) | attacks_bb<ROOK>(s, occupied);
+    default :
+        return PseudoAttacks[pt][s];
+    }
 }
 
 
-/// popcount() counts the number of non-zero bits in a bitboard
+// 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 };
+    std::uint16_t indices[4];
-  return PopCnt16[v.u[0]] + PopCnt16[v.u[1]] + PopCnt16[v.u[2]] + PopCnt16[v.u[3]];
+    std::memcpy(indices, &b, sizeof(b));
+    return PopCnt16[indices[0]] + PopCnt16[indices[1]] + PopCnt16[indices[2]]
+         + PopCnt16[indices[3]];
 
-#elif defined(_MSC_VER) || defined(__INTEL_COMPILER)
+#elif defined(_MSC_VER)
 
-  return (int)_mm_popcnt_u64(b);
+    return int(_mm_popcnt_u64(b));
 
-#else // Assumed gcc or compatible compiler
+#else  // Assumed gcc or compatible compiler
 
-  return __builtin_popcountll(b);
+    return __builtin_popcountll(b);
 
 #endif
 }
 
-
+// Returns the least significant bit in a non-zero bitboard.
-/// 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);
+    assert(b);
-  return Square(__builtin_ctzll(b));
-}
 
-inline Square msb(Bitboard b) {
+#if defined(__GNUC__)  // GCC, Clang, ICX
-  assert(b);
-  return Square(63 ^ __builtin_clzll(b));
-}
 
-#elif defined(_MSC_VER)  // MSVC
+    return Square(__builtin_ctzll(b));
 
-#ifdef _WIN64  // MSVC, WIN64
+#elif defined(_MSC_VER)
+    #ifdef _WIN64  // MSVC, WIN64
 
-inline Square lsb(Bitboard b) {
+    unsigned long idx;
-  assert(b);
+    _BitScanForward64(&idx, b);
-  unsigned long idx;
+    return Square(idx);
-  _BitScanForward64(&idx, b);
-  return (Square) idx;
-}
 
-inline Square msb(Bitboard b) {
+    #else  // MSVC, WIN32
-  assert(b);
+    unsigned long idx;
-  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
 
+    if (b & 0xffffffff)
+    {
+        _BitScanForward(&idx, int32_t(b));
+        return Square(idx);
+    }
+    else
+    {
+        _BitScanForward(&idx, int32_t(b >> 32));
+        return Square(idx + 32);
+    }
+    #endif
 #else  // Compiler is neither GCC nor MSVC compatible
-
+    #error "Compiler not supported."
-#error "Compiler not supported."
-
 #endif
+}
 
-/// least_significant_square_bb() returns the bitboard of the least significant
+// Returns the most significant bit in a non-zero bitboard.
-/// square of a non-zero bitboard. It is equivalent to square_bb(lsb(bb)).
+inline Square msb(Bitboard b) {
+    assert(b);
 
+#if defined(__GNUC__)  // GCC, Clang, ICX
+
+    return Square(63 ^ __builtin_clzll(b));
+
+#elif defined(_MSC_VER)
+    #ifdef _WIN64  // MSVC, WIN64
+
+    unsigned long idx;
+    _BitScanReverse64(&idx, b);
+    return Square(idx);
+
+    #else  // MSVC, WIN32
+
+    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
+#else  // Compiler is neither GCC nor MSVC compatible
+    #error "Compiler not supported."
+#endif
+}
+
+// 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);
+    assert(b);
-  return b & -b;
+    return b & -b;
 }
 
-/// pop_lsb() finds and clears the least significant bit in a non-zero bitboard
+// Finds and clears the least significant bit in a non-zero bitboard.
-
 inline Square pop_lsb(Bitboard& b) {
-  assert(b);
+    assert(b);
-  const Square s = lsb(b);
+    const Square s = lsb(b);
-  b &= b - 1;
+    b &= b - 1;
-  return s;
+    return s;
 }
 
+}  // namespace Stockfish
 
-/// frontmost_sq() returns the most advanced square for the given color,
+#endif  // #ifndef BITBOARD_H_INCLUDED
-/// 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/endgame.cpp

@@ -1,747 +0,0 @@
-/*
-  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 <cassert>
-
-#include "bitboard.h"
-#include "endgame.h"
-#include "movegen.h"
-
-namespace Stockfish {
-
-namespace {
-
-  // Used to drive the king towards the edge of the board
-  // in KX vs K and KQ vs KR endgames.
-  // Values range from 27 (center squares) to 90 (in the corners)
-  inline int push_to_edge(Square s) {
-      int rd = edge_distance(rank_of(s)), fd = edge_distance(file_of(s));
-      return 90 - (7 * fd * fd / 2 + 7 * rd * rd / 2);
-  }
-
-  // Used to drive the king towards A1H8 corners in KBN vs K endgames.
-  // Values range from 0 on A8H1 diagonal to 7 in A1H8 corners
-  inline int push_to_corner(Square s) {
-      return abs(7 - rank_of(s) - file_of(s));
-  }
-
-  // Drive a piece close to or away from another piece
-  inline int push_close(Square s1, Square s2) { return 140 - 20 * distance(s1, s2); }
-  inline int push_away(Square s1, Square s2) { return 120 - push_close(s1, s2); }
-
-#ifndef NDEBUG
-  bool verify_material(const Position& pos, Color c, Value npm, int pawnsCnt) {
-    return pos.non_pawn_material(c) == npm && pos.count<PAWN>(c) == pawnsCnt;
-  }
-#endif
-
-  // Map the square as if strongSide is white and strongSide's only pawn
-  // is on the left half of the board.
-  Square normalize(const Position& pos, Color strongSide, Square sq) {
-
-    assert(pos.count<PAWN>(strongSide) == 1);
-
-    if (file_of(pos.square<PAWN>(strongSide)) >= FILE_E)
-        sq = flip_file(sq);
-
-    return strongSide == WHITE ? sq : flip_rank(sq);
-  }
-
-} // namespace
-
-
-namespace Endgames {
-
-  std::pair<Map<Value>, Map<ScaleFactor>> maps;
-
-  void init() {
-
-    add<KPK>("KPK");
-    add<KNNK>("KNNK");
-    add<KBNK>("KBNK");
-    add<KRKP>("KRKP");
-    add<KRKB>("KRKB");
-    add<KRKN>("KRKN");
-    add<KQKP>("KQKP");
-    add<KQKR>("KQKR");
-    add<KNNKP>("KNNKP");
-
-    add<KRPKR>("KRPKR");
-    add<KRPKB>("KRPKB");
-    add<KBPKB>("KBPKB");
-    add<KBPKN>("KBPKN");
-    add<KBPPKB>("KBPPKB");
-    add<KRPPKRP>("KRPPKRP");
-  }
-}
-
-
-/// Mate with KX vs K. This function is used to evaluate positions with
-/// king and plenty of material vs a lone king. It simply gives the
-/// attacking side a bonus for driving the defending king towards the edge
-/// of the board, and for keeping the distance between the two kings small.
-template<>
-Value Endgame<KXK>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, weakSide, VALUE_ZERO, 0));
-  assert(!pos.checkers()); // Eval is never called when in check
-
-  // Stalemate detection with lone king
-  if (pos.side_to_move() == weakSide && !MoveList<LEGAL>(pos).size())
-      return VALUE_DRAW;
-
-  Square strongKing = pos.square<KING>(strongSide);
-  Square weakKing   = pos.square<KING>(weakSide);
-
-  Value result =  pos.non_pawn_material(strongSide)
-                + pos.count<PAWN>(strongSide) * PawnValueEg
-                + push_to_edge(weakKing)
-                + push_close(strongKing, weakKing);
-
-  if (   pos.count<QUEEN>(strongSide)
-      || pos.count<ROOK>(strongSide)
-      ||(pos.count<BISHOP>(strongSide) && pos.count<KNIGHT>(strongSide))
-      || (   (pos.pieces(strongSide, BISHOP) & ~DarkSquares)
-          && (pos.pieces(strongSide, BISHOP) &  DarkSquares)))
-      result = std::min(result + VALUE_KNOWN_WIN, VALUE_TB_WIN_IN_MAX_PLY - 1);
-
-  return strongSide == pos.side_to_move() ? result : -result;
-}
-
-
-/// Mate with KBN vs K. This is similar to KX vs K, but we have to drive the
-/// defending king towards a corner square that our bishop attacks.
-template<>
-Value Endgame<KBNK>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, KnightValueMg + BishopValueMg, 0));
-  assert(verify_material(pos, weakSide, VALUE_ZERO, 0));
-
-  Square strongKing   = pos.square<KING>(strongSide);
-  Square strongBishop = pos.square<BISHOP>(strongSide);
-  Square weakKing     = pos.square<KING>(weakSide);
-
-  // If our bishop does not attack A1/H8, we flip the enemy king square
-  // to drive to opposite corners (A8/H1).
-
-  Value result =  (VALUE_KNOWN_WIN + 3520)
-                + push_close(strongKing, weakKing)
-                + 420 * push_to_corner(opposite_colors(strongBishop, SQ_A1) ? flip_file(weakKing) : weakKing);
-
-  assert(abs(result) < VALUE_TB_WIN_IN_MAX_PLY);
-  return strongSide == pos.side_to_move() ? result : -result;
-}
-
-
-/// KP vs K. This endgame is evaluated with the help of a bitbase
-template<>
-Value Endgame<KPK>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, VALUE_ZERO, 1));
-  assert(verify_material(pos, weakSide, VALUE_ZERO, 0));
-
-  // Assume strongSide is white and the pawn is on files A-D
-  Square strongKing = normalize(pos, strongSide, pos.square<KING>(strongSide));
-  Square strongPawn = normalize(pos, strongSide, pos.square<PAWN>(strongSide));
-  Square weakKing   = normalize(pos, strongSide, pos.square<KING>(weakSide));
-
-  Color us = strongSide == pos.side_to_move() ? WHITE : BLACK;
-
-  if (!Bitbases::probe(strongKing, strongPawn, weakKing, us))
-      return VALUE_DRAW;
-
-  Value result = VALUE_KNOWN_WIN + PawnValueEg + Value(rank_of(strongPawn));
-
-  return strongSide == pos.side_to_move() ? result : -result;
-}
-
-
-/// KR vs KP. This is a somewhat tricky endgame to evaluate precisely without
-/// a bitbase. The function below returns drawish scores when the pawn is
-/// far advanced with support of the king, while the attacking king is far
-/// away.
-template<>
-Value Endgame<KRKP>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, RookValueMg, 0));
-  assert(verify_material(pos, weakSide, VALUE_ZERO, 1));
-
-  Square strongKing = pos.square<KING>(strongSide);
-  Square weakKing   = pos.square<KING>(weakSide);
-  Square strongRook = pos.square<ROOK>(strongSide);
-  Square weakPawn   = pos.square<PAWN>(weakSide);
-  Square queeningSquare = make_square(file_of(weakPawn), relative_rank(weakSide, RANK_8));
-  Value result;
-
-  // If the stronger side's king is in front of the pawn, it's a win
-  if (forward_file_bb(strongSide, strongKing) & weakPawn)
-      result = RookValueEg - distance(strongKing, weakPawn);
-
-  // If the weaker side's king is too far from the pawn and the rook,
-  // it's a win.
-  else if (   distance(weakKing, weakPawn) >= 3 + (pos.side_to_move() == weakSide)
-           && distance(weakKing, strongRook) >= 3)
-      result = RookValueEg - distance(strongKing, weakPawn);
-
-  // If the pawn is far advanced and supported by the defending king,
-  // the position is drawish
-  else if (   relative_rank(strongSide, weakKing) <= RANK_3
-           && distance(weakKing, weakPawn) == 1
-           && relative_rank(strongSide, strongKing) >= RANK_4
-           && distance(strongKing, weakPawn) > 2 + (pos.side_to_move() == strongSide))
-      result = Value(80) - 8 * distance(strongKing, weakPawn);
-
-  else
-      result =  Value(200) - 8 * (  distance(strongKing, weakPawn + pawn_push(weakSide))
-                                  - distance(weakKing, weakPawn + pawn_push(weakSide))
-                                  - distance(weakPawn, queeningSquare));
-
-  return strongSide == pos.side_to_move() ? result : -result;
-}
-
-
-/// KR vs KB. This is very simple, and always returns drawish scores. The
-/// score is slightly bigger when the defending king is close to the edge.
-template<>
-Value Endgame<KRKB>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, RookValueMg, 0));
-  assert(verify_material(pos, weakSide, BishopValueMg, 0));
-
-  Value result = Value(push_to_edge(pos.square<KING>(weakSide)));
-  return strongSide == pos.side_to_move() ? result : -result;
-}
-
-
-/// KR vs KN. The attacking side has slightly better winning chances than
-/// in KR vs KB, particularly if the king and the knight are far apart.
-template<>
-Value Endgame<KRKN>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, RookValueMg, 0));
-  assert(verify_material(pos, weakSide, KnightValueMg, 0));
-
-  Square weakKing   = pos.square<KING>(weakSide);
-  Square weakKnight = pos.square<KNIGHT>(weakSide);
-  Value result = Value(push_to_edge(weakKing) + push_away(weakKing, weakKnight));
-  return strongSide == pos.side_to_move() ? result : -result;
-}
-
-
-/// KQ vs KP. In general, this is a win for the stronger side, but there are a
-/// few important exceptions. A pawn on 7th rank and on the A,C,F or H files
-/// with a king positioned next to it can be a draw, so in that case, we only
-/// use the distance between the kings.
-template<>
-Value Endgame<KQKP>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, QueenValueMg, 0));
-  assert(verify_material(pos, weakSide, VALUE_ZERO, 1));
-
-  Square strongKing = pos.square<KING>(strongSide);
-  Square weakKing   = pos.square<KING>(weakSide);
-  Square weakPawn   = pos.square<PAWN>(weakSide);
-
-  Value result = Value(push_close(strongKing, weakKing));
-
-  if (   relative_rank(weakSide, weakPawn) != RANK_7
-      || distance(weakKing, weakPawn) != 1
-      || ((FileBBB | FileDBB | FileEBB | FileGBB) & weakPawn))
-      result += QueenValueEg - PawnValueEg;
-
-  return strongSide == pos.side_to_move() ? result : -result;
-}
-
-
-/// KQ vs KR. This is almost identical to KX vs K: we give the attacking
-/// king a bonus for having the kings close together, and for forcing the
-/// defending king towards the edge. If we also take care to avoid null move for
-/// the defending side in the search, this is usually sufficient to win KQ vs KR.
-template<>
-Value Endgame<KQKR>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, QueenValueMg, 0));
-  assert(verify_material(pos, weakSide, RookValueMg, 0));
-
-  Square strongKing = pos.square<KING>(strongSide);
-  Square weakKing   = pos.square<KING>(weakSide);
-
-  Value result =  QueenValueEg
-                - RookValueEg
-                + push_to_edge(weakKing)
-                + push_close(strongKing, weakKing);
-
-  return strongSide == pos.side_to_move() ? result : -result;
-}
-
-
-/// KNN vs KP. Very drawish, but there are some mate opportunities if we can
-/// press the weakSide King to a corner before the pawn advances too much.
-template<>
-Value Endgame<KNNKP>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, 2 * KnightValueMg, 0));
-  assert(verify_material(pos, weakSide, VALUE_ZERO, 1));
-
-  Square weakKing = pos.square<KING>(weakSide);
-  Square weakPawn = pos.square<PAWN>(weakSide);
-
-  Value result =      PawnValueEg
-               +  2 * push_to_edge(weakKing)
-               - 10 * relative_rank(weakSide, weakPawn);
-
-  return strongSide == pos.side_to_move() ? result : -result;
-}
-
-
-/// Some cases of trivial draws
-template<> Value Endgame<KNNK>::operator()(const Position&) const { return VALUE_DRAW; }
-
-
-/// KB and one or more pawns vs K. It checks for draws with rook pawns and
-/// a bishop of the wrong color. If such a draw is detected, SCALE_FACTOR_DRAW
-/// is returned. If not, the return value is SCALE_FACTOR_NONE, i.e. no scaling
-/// will be used.
-template<>
-ScaleFactor Endgame<KBPsK>::operator()(const Position& pos) const {
-
-  assert(pos.non_pawn_material(strongSide) == BishopValueMg);
-  assert(pos.count<PAWN>(strongSide) >= 1);
-
-  // No assertions about the material of weakSide, because we want draws to
-  // be detected even when the weaker side has some pawns.
-
-  Bitboard strongPawns = pos.pieces(strongSide, PAWN);
-  Bitboard allPawns = pos.pieces(PAWN);
-
-  Square strongBishop = pos.square<BISHOP>(strongSide);
-  Square weakKing = pos.square<KING>(weakSide);
-  Square strongKing = pos.square<KING>(strongSide);
-
-  // All strongSide pawns are on a single rook file?
-  if (!(strongPawns & ~FileABB) || !(strongPawns & ~FileHBB))
-  {
-      Square queeningSquare = relative_square(strongSide, make_square(file_of(lsb(strongPawns)), RANK_8));
-
-      if (   opposite_colors(queeningSquare, strongBishop)
-          && distance(queeningSquare, weakKing) <= 1)
-          return SCALE_FACTOR_DRAW;
-  }
-
-  // If all the pawns are on the same B or G file, then it's potentially a draw
-  if ((!(allPawns & ~FileBBB) || !(allPawns & ~FileGBB))
-      && pos.non_pawn_material(weakSide) == 0
-      && pos.count<PAWN>(weakSide) >= 1)
-  {
-      // Get the least advanced weakSide pawn
-      Square weakPawn = frontmost_sq(strongSide, pos.pieces(weakSide, PAWN));
-
-      // There's potential for a draw if our pawn is blocked on the 7th rank,
-      // the bishop cannot attack it or they only have one pawn left.
-      if (   relative_rank(strongSide, weakPawn) == RANK_7
-          && (strongPawns & (weakPawn + pawn_push(weakSide)))
-          && (opposite_colors(strongBishop, weakPawn) || !more_than_one(strongPawns)))
-      {
-          int strongKingDist = distance(weakPawn, strongKing);
-          int weakKingDist = distance(weakPawn, weakKing);
-
-          // It's a draw if the weak king is on its back two ranks, within 2
-          // squares of the blocking pawn and the strong king is not
-          // closer. (I think this rule only fails in practically
-          // unreachable positions such as 5k1K/6p1/6P1/8/8/3B4/8/8 w
-          // and positions where qsearch will immediately correct the
-          // problem such as 8/4k1p1/6P1/1K6/3B4/8/8/8 w).
-          if (   relative_rank(strongSide, weakKing) >= RANK_7
-              && weakKingDist <= 2
-              && weakKingDist <= strongKingDist)
-              return SCALE_FACTOR_DRAW;
-      }
-  }
-
-  return SCALE_FACTOR_NONE;
-}
-
-
-/// KQ vs KR and one or more pawns. It tests for fortress draws with a rook on
-/// the third rank defended by a pawn.
-template<>
-ScaleFactor Endgame<KQKRPs>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, QueenValueMg, 0));
-  assert(pos.count<ROOK>(weakSide) == 1);
-  assert(pos.count<PAWN>(weakSide) >= 1);
-
-  Square strongKing = pos.square<KING>(strongSide);
-  Square weakKing   = pos.square<KING>(weakSide);
-  Square weakRook   = pos.square<ROOK>(weakSide);
-
-  if (    relative_rank(weakSide,   weakKing) <= RANK_2
-      &&  relative_rank(weakSide, strongKing) >= RANK_4
-      &&  relative_rank(weakSide,   weakRook) == RANK_3
-      && (  pos.pieces(weakSide, PAWN)
-          & attacks_bb<KING>(weakKing)
-          & pawn_attacks_bb(strongSide, weakRook)))
-          return SCALE_FACTOR_DRAW;
-
-  return SCALE_FACTOR_NONE;
-}
-
-
-/// KRP vs KR. This function knows a handful of the most important classes of
-/// drawn positions, but is far from perfect. It would probably be a good idea
-/// to add more knowledge in the future.
-///
-/// It would also be nice to rewrite the actual code for this function,
-/// which is mostly copied from Glaurung 1.x, and isn't very pretty.
-template<>
-ScaleFactor Endgame<KRPKR>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, RookValueMg, 1));
-  assert(verify_material(pos, weakSide,   RookValueMg, 0));
-
-  // Assume strongSide is white and the pawn is on files A-D
-  Square strongKing = normalize(pos, strongSide, pos.square<KING>(strongSide));
-  Square strongRook = normalize(pos, strongSide, pos.square<ROOK>(strongSide));
-  Square strongPawn = normalize(pos, strongSide, pos.square<PAWN>(strongSide));
-  Square weakKing = normalize(pos, strongSide, pos.square<KING>(weakSide));
-  Square weakRook = normalize(pos, strongSide, pos.square<ROOK>(weakSide));
-
-  File pawnFile = file_of(strongPawn);
-  Rank pawnRank = rank_of(strongPawn);
-  Square queeningSquare = make_square(pawnFile, RANK_8);
-  int tempo = (pos.side_to_move() == strongSide);
-
-  // If the pawn is not too far advanced and the defending king defends the
-  // queening square, use the third-rank defence.
-  if (   pawnRank <= RANK_5
-      && distance(weakKing, queeningSquare) <= 1
-      && strongKing <= SQ_H5
-      && (rank_of(weakRook) == RANK_6 || (pawnRank <= RANK_3 && rank_of(strongRook) != RANK_6)))
-      return SCALE_FACTOR_DRAW;
-
-  // The defending side saves a draw by checking from behind in case the pawn
-  // has advanced to the 6th rank with the king behind.
-  if (   pawnRank == RANK_6
-      && distance(weakKing, queeningSquare) <= 1
-      && rank_of(strongKing) + tempo <= RANK_6
-      && (rank_of(weakRook) == RANK_1 || (!tempo && distance<File>(weakRook, strongPawn) >= 3)))
-      return SCALE_FACTOR_DRAW;
-
-  if (   pawnRank >= RANK_6
-      && weakKing == queeningSquare
-      && rank_of(weakRook) == RANK_1
-      && (!tempo || distance(strongKing, strongPawn) >= 2))
-      return SCALE_FACTOR_DRAW;
-
-  // White pawn on a7 and rook on a8 is a draw if black's king is on g7 or h7
-  // and the black rook is behind the pawn.
-  if (   strongPawn == SQ_A7
-      && strongRook == SQ_A8
-      && (weakKing == SQ_H7 || weakKing == SQ_G7)
-      && file_of(weakRook) == FILE_A
-      && (rank_of(weakRook) <= RANK_3 || file_of(strongKing) >= FILE_D || rank_of(strongKing) <= RANK_5))
-      return SCALE_FACTOR_DRAW;
-
-  // If the defending king blocks the pawn and the attacking king is too far
-  // away, it's a draw.
-  if (   pawnRank <= RANK_5
-      && weakKing == strongPawn + NORTH
-      && distance(strongKing, strongPawn) - tempo >= 2
-      && distance(strongKing, weakRook) - tempo >= 2)
-      return SCALE_FACTOR_DRAW;
-
-  // Pawn on the 7th rank supported by the rook from behind usually wins if the
-  // attacking king is closer to the queening square than the defending king,
-  // and the defending king cannot gain tempi by threatening the attacking rook.
-  if (   pawnRank == RANK_7
-      && pawnFile != FILE_A
-      && file_of(strongRook) == pawnFile
-      && strongRook != queeningSquare
-      && (distance(strongKing, queeningSquare) < distance(weakKing, queeningSquare) - 2 + tempo)
-      && (distance(strongKing, queeningSquare) < distance(weakKing, strongRook) + tempo))
-      return ScaleFactor(SCALE_FACTOR_MAX - 2 * distance(strongKing, queeningSquare));
-
-  // Similar to the above, but with the pawn further back
-  if (   pawnFile != FILE_A
-      && file_of(strongRook) == pawnFile
-      && strongRook < strongPawn
-      && (distance(strongKing, queeningSquare) < distance(weakKing, queeningSquare) - 2 + tempo)
-      && (distance(strongKing, strongPawn + NORTH) < distance(weakKing, strongPawn + NORTH) - 2 + tempo)
-      && (  distance(weakKing, strongRook) + tempo >= 3
-          || (    distance(strongKing, queeningSquare) < distance(weakKing, strongRook) + tempo
-              && (distance(strongKing, strongPawn + NORTH) < distance(weakKing, strongPawn) + tempo))))
-      return ScaleFactor(  SCALE_FACTOR_MAX
-                         - 8 * distance(strongPawn, queeningSquare)
-                         - 2 * distance(strongKing, queeningSquare));
-
-  // If the pawn is not far advanced and the defending king is somewhere in
-  // the pawn's path, it's probably a draw.
-  if (pawnRank <= RANK_4 && weakKing > strongPawn)
-  {
-      if (file_of(weakKing) == file_of(strongPawn))
-          return ScaleFactor(10);
-      if (   distance<File>(weakKing, strongPawn) == 1
-          && distance(strongKing, weakKing) > 2)
-          return ScaleFactor(24 - 2 * distance(strongKing, weakKing));
-  }
-  return SCALE_FACTOR_NONE;
-}
-
-template<>
-ScaleFactor Endgame<KRPKB>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, RookValueMg, 1));
-  assert(verify_material(pos, weakSide, BishopValueMg, 0));
-
-  // Test for a rook pawn
-  if (pos.pieces(PAWN) & (FileABB | FileHBB))
-  {
-      Square weakKing = pos.square<KING>(weakSide);
-      Square weakBishop = pos.square<BISHOP>(weakSide);
-      Square strongKing = pos.square<KING>(strongSide);
-      Square strongPawn = pos.square<PAWN>(strongSide);
-      Rank pawnRank = relative_rank(strongSide, strongPawn);
-      Direction push = pawn_push(strongSide);
-
-      // If the pawn is on the 5th rank and the pawn (currently) is on
-      // the same color square as the bishop then there is a chance of
-      // a fortress. Depending on the king position give a moderate
-      // reduction or a stronger one if the defending king is near the
-      // corner but not trapped there.
-      if (pawnRank == RANK_5 && !opposite_colors(weakBishop, strongPawn))
-      {
-          int d = distance(strongPawn + 3 * push, weakKing);
-
-          if (d <= 2 && !(d == 0 && weakKing == strongKing + 2 * push))
-              return ScaleFactor(24);
-          else
-              return ScaleFactor(48);
-      }
-
-      // When the pawn has moved to the 6th rank we can be fairly sure
-      // it's drawn if the bishop attacks the square in front of the
-      // pawn from a reasonable distance and the defending king is near
-      // the corner
-      if (   pawnRank == RANK_6
-          && distance(strongPawn + 2 * push, weakKing) <= 1
-          && (attacks_bb<BISHOP>(weakBishop) & (strongPawn + push))
-          && distance<File>(weakBishop, strongPawn) >= 2)
-          return ScaleFactor(8);
-  }
-
-  return SCALE_FACTOR_NONE;
-}
-
-/// KRPP vs KRP. There is just a single rule: if the stronger side has no passed
-/// pawns and the defending king is actively placed, the position is drawish.
-template<>
-ScaleFactor Endgame<KRPPKRP>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, RookValueMg, 2));
-  assert(verify_material(pos, weakSide,   RookValueMg, 1));
-
-  Square strongPawn1 = lsb(pos.pieces(strongSide, PAWN));
-  Square strongPawn2 = msb(pos.pieces(strongSide, PAWN));
-  Square weakKing = pos.square<KING>(weakSide);
-
-  // Does the stronger side have a passed pawn?
-  if (pos.pawn_passed(strongSide, strongPawn1) || pos.pawn_passed(strongSide, strongPawn2))
-      return SCALE_FACTOR_NONE;
-
-  Rank pawnRank = std::max(relative_rank(strongSide, strongPawn1), relative_rank(strongSide, strongPawn2));
-
-  if (   distance<File>(weakKing, strongPawn1) <= 1
-      && distance<File>(weakKing, strongPawn2) <= 1
-      && relative_rank(strongSide, weakKing) > pawnRank)
-  {
-      assert(pawnRank > RANK_1 && pawnRank < RANK_7);
-      return ScaleFactor(7 * pawnRank);
-  }
-  return SCALE_FACTOR_NONE;
-}
-
-
-/// K and two or more pawns vs K. There is just a single rule here: if all pawns
-/// are on the same rook file and are blocked by the defending king, it's a draw.
-template<>
-ScaleFactor Endgame<KPsK>::operator()(const Position& pos) const {
-
-  assert(pos.non_pawn_material(strongSide) == VALUE_ZERO);
-  assert(pos.count<PAWN>(strongSide) >= 2);
-  assert(verify_material(pos, weakSide, VALUE_ZERO, 0));
-
-  Square weakKing = pos.square<KING>(weakSide);
-  Bitboard strongPawns = pos.pieces(strongSide, PAWN);
-
-  // If all pawns are ahead of the king on a single rook file, it's a draw.
-  if (   !(strongPawns & ~(FileABB | FileHBB))
-      && !(strongPawns & ~passed_pawn_span(weakSide, weakKing)))
-      return SCALE_FACTOR_DRAW;
-
-  return SCALE_FACTOR_NONE;
-}
-
-
-/// KBP vs KB. There are two rules: if the defending king is somewhere along the
-/// path of the pawn, and the square of the king is not of the same color as the
-/// stronger side's bishop, it's a draw. If the two bishops have opposite color,
-/// it's almost always a draw.
-template<>
-ScaleFactor Endgame<KBPKB>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, BishopValueMg, 1));
-  assert(verify_material(pos, weakSide,   BishopValueMg, 0));
-
-  Square strongPawn = pos.square<PAWN>(strongSide);
-  Square strongBishop = pos.square<BISHOP>(strongSide);
-  Square weakBishop = pos.square<BISHOP>(weakSide);
-  Square weakKing = pos.square<KING>(weakSide);
-
-  // Case 1: Defending king blocks the pawn, and cannot be driven away
-  if (   (forward_file_bb(strongSide, strongPawn) & weakKing)
-      && (   opposite_colors(weakKing, strongBishop)
-          || relative_rank(strongSide, weakKing) <= RANK_6))
-      return SCALE_FACTOR_DRAW;
-
-  // Case 2: Opposite colored bishops
-  if (opposite_colors(strongBishop, weakBishop))
-      return SCALE_FACTOR_DRAW;
-
-  return SCALE_FACTOR_NONE;
-}
-
-
-/// KBPP vs KB. It detects a few basic draws with opposite-colored bishops
-template<>
-ScaleFactor Endgame<KBPPKB>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, BishopValueMg, 2));
-  assert(verify_material(pos, weakSide,   BishopValueMg, 0));
-
-  Square strongBishop = pos.square<BISHOP>(strongSide);
-  Square weakBishop   = pos.square<BISHOP>(weakSide);
-
-  if (!opposite_colors(strongBishop, weakBishop))
-      return SCALE_FACTOR_NONE;
-
-  Square weakKing = pos.square<KING>(weakSide);
-  Square strongPawn1 = lsb(pos.pieces(strongSide, PAWN));
-  Square strongPawn2 = msb(pos.pieces(strongSide, PAWN));
-  Square blockSq1, blockSq2;
-
-  if (relative_rank(strongSide, strongPawn1) > relative_rank(strongSide, strongPawn2))
-  {
-      blockSq1 = strongPawn1 + pawn_push(strongSide);
-      blockSq2 = make_square(file_of(strongPawn2), rank_of(strongPawn1));
-  }
-  else
-  {
-      blockSq1 = strongPawn2 + pawn_push(strongSide);
-      blockSq2 = make_square(file_of(strongPawn1), rank_of(strongPawn2));
-  }
-
-  switch (distance<File>(strongPawn1, strongPawn2))
-  {
-  case 0:
-    // Both pawns are on the same file. It's an easy draw if the defender firmly
-    // controls some square in the frontmost pawn's path.
-    if (   file_of(weakKing) == file_of(blockSq1)
-        && relative_rank(strongSide, weakKing) >= relative_rank(strongSide, blockSq1)
-        && opposite_colors(weakKing, strongBishop))
-        return SCALE_FACTOR_DRAW;
-    else
-        return SCALE_FACTOR_NONE;
-
-  case 1:
-    // Pawns on adjacent files. It's a draw if the defender firmly controls the
-    // square in front of the frontmost pawn's path, and the square diagonally
-    // behind this square on the file of the other pawn.
-    if (   weakKing == blockSq1
-        && opposite_colors(weakKing, strongBishop)
-        && (   weakBishop == blockSq2
-            || (attacks_bb<BISHOP>(blockSq2, pos.pieces()) & pos.pieces(weakSide, BISHOP))
-            || distance<Rank>(strongPawn1, strongPawn2) >= 2))
-        return SCALE_FACTOR_DRAW;
-
-    else if (   weakKing == blockSq2
-             && opposite_colors(weakKing, strongBishop)
-             && (   weakBishop == blockSq1
-                 || (attacks_bb<BISHOP>(blockSq1, pos.pieces()) & pos.pieces(weakSide, BISHOP))))
-        return SCALE_FACTOR_DRAW;
-    else
-        return SCALE_FACTOR_NONE;
-
-  default:
-    // The pawns are not on the same file or adjacent files. No scaling.
-    return SCALE_FACTOR_NONE;
-  }
-}
-
-
-/// KBP vs KN. There is a single rule: if the defending king is somewhere along
-/// the path of the pawn, and the square of the king is not of the same color as
-/// the stronger side's bishop, it's a draw.
-template<>
-ScaleFactor Endgame<KBPKN>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, BishopValueMg, 1));
-  assert(verify_material(pos, weakSide, KnightValueMg, 0));
-
-  Square strongPawn = pos.square<PAWN>(strongSide);
-  Square strongBishop = pos.square<BISHOP>(strongSide);
-  Square weakKing = pos.square<KING>(weakSide);
-
-  if (   file_of(weakKing) == file_of(strongPawn)
-      && relative_rank(strongSide, strongPawn) < relative_rank(strongSide, weakKing)
-      && (   opposite_colors(weakKing, strongBishop)
-          || relative_rank(strongSide, weakKing) <= RANK_6))
-      return SCALE_FACTOR_DRAW;
-
-  return SCALE_FACTOR_NONE;
-}
-
-
-/// KP vs KP. This is done by removing the weakest side's pawn and probing the
-/// KP vs K bitbase: if the weakest side has a draw without the pawn, it probably
-/// has at least a draw with the pawn as well. The exception is when the stronger
-/// side's pawn is far advanced and not on a rook file; in this case it is often
-/// possible to win (e.g. 8/4k3/3p4/3P4/6K1/8/8/8 w - - 0 1).
-template<>
-ScaleFactor Endgame<KPKP>::operator()(const Position& pos) const {
-
-  assert(verify_material(pos, strongSide, VALUE_ZERO, 1));
-  assert(verify_material(pos, weakSide,   VALUE_ZERO, 1));
-
-  // Assume strongSide is white and the pawn is on files A-D
-  Square strongKing = normalize(pos, strongSide, pos.square<KING>(strongSide));
-  Square weakKing   = normalize(pos, strongSide, pos.square<KING>(weakSide));
-  Square strongPawn = normalize(pos, strongSide, pos.square<PAWN>(strongSide));
-
-  Color us = strongSide == pos.side_to_move() ? WHITE : BLACK;
-
-  // If the pawn has advanced to the fifth rank or further, and is not a
-  // rook pawn, it's too dangerous to assume that it's at least a draw.
-  if (rank_of(strongPawn) >= RANK_5 && file_of(strongPawn) != FILE_A)
-      return SCALE_FACTOR_NONE;
-
-  // Probe the KPK bitbase with the weakest side's pawn removed. If it's a draw,
-  // it's probably at least a draw even with the pawn.
-  return Bitbases::probe(strongKing, strongPawn, weakKing, us) ? SCALE_FACTOR_NONE : SCALE_FACTOR_DRAW;
-}
-
-} // namespace Stockfish

src/endgame.h

@@ -1,126 +0,0 @@
-/*
-  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 ENDGAME_H_INCLUDED
-#define ENDGAME_H_INCLUDED
-
-#include <memory>
-#include <string>
-#include <type_traits>
-#include <unordered_map>
-#include <utility>
-
-#include "position.h"
-#include "types.h"
-
-namespace Stockfish {
-
-/// EndgameCode lists all supported endgame functions by corresponding codes
-
-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
-
-  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,    // K and pawns vs K
-  KBPKB,   // KBP vs KB
-  KBPPKB,  // KBPP vs KB
-  KBPKN,   // KBP vs KN
-  KPKP     // KP vs KP
-};
-
-
-/// Endgame functions can be of two types depending on whether they return a
-/// Value or a ScaleFactor.
-
-template<EndgameCode E> using
-eg_type = typename std::conditional<(E < SCALING_FUNCTIONS), Value, ScaleFactor>::type;
-
-
-/// Base and derived functors for endgame evaluation and scaling functions
-
-template<typename T>
-struct EndgameBase {
-
-  explicit EndgameBase(Color c) : strongSide(c), weakSide(~c) {}
-  virtual ~EndgameBase() = default;
-  virtual T operator()(const Position&) const = 0;
-
-  const Color strongSide, weakSide;
-};
-
-
-template<EndgameCode E, typename T = eg_type<E>>
-struct Endgame : public EndgameBase<T> {
-
-  explicit Endgame(Color c) : EndgameBase<T>(c) {}
-  T operator()(const Position&) const override;
-};
-
-
-/// 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().
-
-namespace Endgames {
-
-  template<typename T> using Ptr = std::unique_ptr<EndgameBase<T>>;
-  template<typename T> using Map = std::unordered_map<Key, Ptr<T>>;
-
-  extern std::pair<Map<Value>, Map<ScaleFactor>> maps;
-
-  void init();
-
-  template<typename T>
-  Map<T>& map() {
-    return std::get<std::is_same<T, ScaleFactor>::value>(maps);
-  }
-
-  template<EndgameCode E, typename T = eg_type<E>>
-  void add(const std::string& code) {
-
-    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));
-  }
-
-  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/engine.cpp

@@ -0,0 +1,369 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 "engine.h"
+
+#include <cassert>
+#include <deque>
+#include <iosfwd>
+#include <memory>
+#include <ostream>
+#include <sstream>
+#include <string_view>
+#include <utility>
+#include <vector>
+
+#include "evaluate.h"
+#include "misc.h"
+#include "nnue/network.h"
+#include "nnue/nnue_common.h"
+#include "perft.h"
+#include "position.h"
+#include "search.h"
+#include "syzygy/tbprobe.h"
+#include "types.h"
+#include "uci.h"
+#include "ucioption.h"
+
+namespace Stockfish {
+
+namespace NN = Eval::NNUE;
+
+constexpr auto StartFEN  = "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1";
+constexpr int  MaxHashMB = Is64Bit ? 33554432 : 2048;
+
+Engine::Engine(std::optional<std::string> path) :
+    binaryDirectory(path ? CommandLine::get_binary_directory(*path) : ""),
+    numaContext(NumaConfig::from_system()),
+    states(new std::deque<StateInfo>(1)),
+    threads(),
+    networks(
+      numaContext,
+      NN::Networks(
+        NN::NetworkBig({EvalFileDefaultNameBig, "None", ""}, NN::EmbeddedNNUEType::BIG),
+        NN::NetworkSmall({EvalFileDefaultNameSmall, "None", ""}, NN::EmbeddedNNUEType::SMALL))) {
+    pos.set(StartFEN, false, &states->back());
+
+
+    options.add(  //
+      "Debug Log File", Option("", [](const Option& o) {
+          start_logger(o);
+          return std::nullopt;
+      }));
+
+    options.add(  //
+      "NumaPolicy", Option("auto", [this](const Option& o) {
+          set_numa_config_from_option(o);
+          return numa_config_information_as_string() + "\n"
+               + thread_allocation_information_as_string();
+      }));
+
+    options.add(  //
+      "Threads", Option(1, 1, 1024, [this](const Option&) {
+          resize_threads();
+          return thread_allocation_information_as_string();
+      }));
+
+    options.add(  //
+      "Hash", Option(16, 1, MaxHashMB, [this](const Option& o) {
+          set_tt_size(o);
+          return std::nullopt;
+      }));
+
+    options.add(  //
+      "Clear Hash", Option([this](const Option&) {
+          search_clear();
+          return std::nullopt;
+      }));
+
+    options.add(  //
+      "Ponder", Option(false));
+
+    options.add(  //
+      "MultiPV", Option(1, 1, MAX_MOVES));
+
+    options.add("Skill Level", Option(20, 0, 20));
+
+    options.add("Move Overhead", Option(10, 0, 5000));
+
+    options.add("nodestime", Option(0, 0, 10000));
+
+    options.add("UCI_Chess960", Option(false));
+
+    options.add("UCI_LimitStrength", Option(false));
+
+    options.add("UCI_Elo",
+                Option(Stockfish::Search::Skill::LowestElo, Stockfish::Search::Skill::LowestElo,
+                       Stockfish::Search::Skill::HighestElo));
+
+    options.add("UCI_ShowWDL", Option(false));
+
+    options.add(  //
+      "SyzygyPath", Option("", [](const Option& o) {
+          Tablebases::init(o);
+          return std::nullopt;
+      }));
+
+    options.add("SyzygyProbeDepth", Option(1, 1, 100));
+
+    options.add("Syzygy50MoveRule", Option(true));
+
+    options.add("SyzygyProbeLimit", Option(7, 0, 7));
+
+    options.add(  //
+      "EvalFile", Option(EvalFileDefaultNameBig, [this](const Option& o) {
+          load_big_network(o);
+          return std::nullopt;
+      }));
+
+    options.add(  //
+      "EvalFileSmall", Option(EvalFileDefaultNameSmall, [this](const Option& o) {
+          load_small_network(o);
+          return std::nullopt;
+      }));
+
+    load_networks();
+    resize_threads();
+}
+
+std::uint64_t Engine::perft(const std::string& fen, Depth depth, bool isChess960) {
+    verify_networks();
+
+    return Benchmark::perft(fen, depth, isChess960);
+}
+
+void Engine::go(Search::LimitsType& limits) {
+    assert(limits.perft == 0);
+    verify_networks();
+
+    threads.start_thinking(options, pos, states, limits);
+}
+void Engine::stop() { threads.stop = true; }
+
+void Engine::search_clear() {
+    wait_for_search_finished();
+
+    tt.clear(threads);
+    threads.clear();
+
+    // @TODO wont work with multiple instances
+    Tablebases::init(options["SyzygyPath"]);  // Free mapped files
+}
+
+void Engine::set_on_update_no_moves(std::function<void(const Engine::InfoShort&)>&& f) {
+    updateContext.onUpdateNoMoves = std::move(f);
+}
+
+void Engine::set_on_update_full(std::function<void(const Engine::InfoFull&)>&& f) {
+    updateContext.onUpdateFull = std::move(f);
+}
+
+void Engine::set_on_iter(std::function<void(const Engine::InfoIter&)>&& f) {
+    updateContext.onIter = std::move(f);
+}
+
+void Engine::set_on_bestmove(std::function<void(std::string_view, std::string_view)>&& f) {
+    updateContext.onBestmove = std::move(f);
+}
+
+void Engine::set_on_verify_networks(std::function<void(std::string_view)>&& f) {
+    onVerifyNetworks = std::move(f);
+}
+
+void Engine::wait_for_search_finished() { threads.main_thread()->wait_for_search_finished(); }
+
+void Engine::set_position(const std::string& fen, const std::vector<std::string>& moves) {
+    // Drop the old state and create a new one
+    states = StateListPtr(new std::deque<StateInfo>(1));
+    pos.set(fen, options["UCI_Chess960"], &states->back());
+
+    for (const auto& move : moves)
+    {
+        auto m = UCIEngine::to_move(pos, move);
+
+        if (m == Move::none())
+            break;
+
+        states->emplace_back();
+        pos.do_move(m, states->back());
+    }
+}
+
+// modifiers
+
+void Engine::set_numa_config_from_option(const std::string& o) {
+    if (o == "auto" || o == "system")
+    {
+        numaContext.set_numa_config(NumaConfig::from_system());
+    }
+    else if (o == "hardware")
+    {
+        // Don't respect affinity set in the system.
+        numaContext.set_numa_config(NumaConfig::from_system(false));
+    }
+    else if (o == "none")
+    {
+        numaContext.set_numa_config(NumaConfig{});
+    }
+    else
+    {
+        numaContext.set_numa_config(NumaConfig::from_string(o));
+    }
+
+    // Force reallocation of threads in case affinities need to change.
+    resize_threads();
+    threads.ensure_network_replicated();
+}
+
+void Engine::resize_threads() {
+    threads.wait_for_search_finished();
+    threads.set(numaContext.get_numa_config(), {options, threads, tt, networks}, updateContext);
+
+    // Reallocate the hash with the new threadpool size
+    set_tt_size(options["Hash"]);
+    threads.ensure_network_replicated();
+}
+
+void Engine::set_tt_size(size_t mb) {
+    wait_for_search_finished();
+    tt.resize(mb, threads);
+}
+
+void Engine::set_ponderhit(bool b) { threads.main_manager()->ponder = b; }
+
+// network related
+
+void Engine::verify_networks() const {
+    networks->big.verify(options["EvalFile"], onVerifyNetworks);
+    networks->small.verify(options["EvalFileSmall"], onVerifyNetworks);
+}
+
+void Engine::load_networks() {
+    networks.modify_and_replicate([this](NN::Networks& networks_) {
+        networks_.big.load(binaryDirectory, options["EvalFile"]);
+        networks_.small.load(binaryDirectory, options["EvalFileSmall"]);
+    });
+    threads.clear();
+    threads.ensure_network_replicated();
+}
+
+void Engine::load_big_network(const std::string& file) {
+    networks.modify_and_replicate(
+      [this, &file](NN::Networks& networks_) { networks_.big.load(binaryDirectory, file); });
+    threads.clear();
+    threads.ensure_network_replicated();
+}
+
+void Engine::load_small_network(const std::string& file) {
+    networks.modify_and_replicate(
+      [this, &file](NN::Networks& networks_) { networks_.small.load(binaryDirectory, file); });
+    threads.clear();
+    threads.ensure_network_replicated();
+}
+
+void Engine::save_network(const std::pair<std::optional<std::string>, std::string> files[2]) {
+    networks.modify_and_replicate([&files](NN::Networks& networks_) {
+        networks_.big.save(files[0].first);
+        networks_.small.save(files[1].first);
+    });
+}
+
+// utility functions
+
+void Engine::trace_eval() const {
+    StateListPtr trace_states(new std::deque<StateInfo>(1));
+    Position     p;
+    p.set(pos.fen(), options["UCI_Chess960"], &trace_states->back());
+
+    verify_networks();
+
+    sync_cout << "\n" << Eval::trace(p, *networks) << sync_endl;
+}
+
+const OptionsMap& Engine::get_options() const { return options; }
+OptionsMap&       Engine::get_options() { return options; }
+
+std::string Engine::fen() const { return pos.fen(); }
+
+void Engine::flip() { pos.flip(); }
+
+std::string Engine::visualize() const {
+    std::stringstream ss;
+    ss << pos;
+    return ss.str();
+}
+
+int Engine::get_hashfull(int maxAge) const { return tt.hashfull(maxAge); }
+
+std::vector<std::pair<size_t, size_t>> Engine::get_bound_thread_count_by_numa_node() const {
+    auto                                   counts = threads.get_bound_thread_count_by_numa_node();
+    const NumaConfig&                      cfg    = numaContext.get_numa_config();
+    std::vector<std::pair<size_t, size_t>> ratios;
+    NumaIndex                              n = 0;
+    for (; n < counts.size(); ++n)
+        ratios.emplace_back(counts[n], cfg.num_cpus_in_numa_node(n));
+    if (!counts.empty())
+        for (; n < cfg.num_numa_nodes(); ++n)
+            ratios.emplace_back(0, cfg.num_cpus_in_numa_node(n));
+    return ratios;
+}
+
+std::string Engine::get_numa_config_as_string() const {
+    return numaContext.get_numa_config().to_string();
+}
+
+std::string Engine::numa_config_information_as_string() const {
+    auto cfgStr = get_numa_config_as_string();
+    return "Available processors: " + cfgStr;
+}
+
+std::string Engine::thread_binding_information_as_string() const {
+    auto              boundThreadsByNode = get_bound_thread_count_by_numa_node();
+    std::stringstream ss;
+    if (boundThreadsByNode.empty())
+        return ss.str();
+
+    bool isFirst = true;
+
+    for (auto&& [current, total] : boundThreadsByNode)
+    {
+        if (!isFirst)
+            ss << ":";
+        ss << current << "/" << total;
+        isFirst = false;
+    }
+
+    return ss.str();
+}
+
+std::string Engine::thread_allocation_information_as_string() const {
+    std::stringstream ss;
+
+    size_t threadsSize = threads.size();
+    ss << "Using " << threadsSize << (threadsSize > 1 ? " threads" : " thread");
+
+    auto boundThreadsByNodeStr = thread_binding_information_as_string();
+    if (boundThreadsByNodeStr.empty())
+        return ss.str();
+
+    ss << " with NUMA node thread binding: ";
+    ss << boundThreadsByNodeStr;
+
+    return ss.str();
+}
+}

src/engine.h

@@ -0,0 +1,130 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 ENGINE_H_INCLUDED
+#define ENGINE_H_INCLUDED
+
+#include <cstddef>
+#include <cstdint>
+#include <functional>
+#include <optional>
+#include <string>
+#include <string_view>
+#include <utility>
+#include <vector>
+
+#include "nnue/network.h"
+#include "numa.h"
+#include "position.h"
+#include "search.h"
+#include "syzygy/tbprobe.h"  // for Stockfish::Depth
+#include "thread.h"
+#include "tt.h"
+#include "ucioption.h"
+
+namespace Stockfish {
+
+class Engine {
+   public:
+    using InfoShort = Search::InfoShort;
+    using InfoFull  = Search::InfoFull;
+    using InfoIter  = Search::InfoIteration;
+
+    Engine(std::optional<std::string> path = std::nullopt);
+
+    // Cannot be movable due to components holding backreferences to fields
+    Engine(const Engine&)            = delete;
+    Engine(Engine&&)                 = delete;
+    Engine& operator=(const Engine&) = delete;
+    Engine& operator=(Engine&&)      = delete;
+
+    ~Engine() { wait_for_search_finished(); }
+
+    std::uint64_t perft(const std::string& fen, Depth depth, bool isChess960);
+
+    // non blocking call to start searching
+    void go(Search::LimitsType&);
+    // non blocking call to stop searching
+    void stop();
+
+    // blocking call to wait for search to finish
+    void wait_for_search_finished();
+    // set a new position, moves are in UCI format
+    void set_position(const std::string& fen, const std::vector<std::string>& moves);
+
+    // modifiers
+
+    void set_numa_config_from_option(const std::string& o);
+    void resize_threads();
+    void set_tt_size(size_t mb);
+    void set_ponderhit(bool);
+    void search_clear();
+
+    void set_on_update_no_moves(std::function<void(const InfoShort&)>&&);
+    void set_on_update_full(std::function<void(const InfoFull&)>&&);
+    void set_on_iter(std::function<void(const InfoIter&)>&&);
+    void set_on_bestmove(std::function<void(std::string_view, std::string_view)>&&);
+    void set_on_verify_networks(std::function<void(std::string_view)>&&);
+
+    // network related
+
+    void verify_networks() const;
+    void load_networks();
+    void load_big_network(const std::string& file);
+    void load_small_network(const std::string& file);
+    void save_network(const std::pair<std::optional<std::string>, std::string> files[2]);
+
+    // utility functions
+
+    void trace_eval() const;
+
+    const OptionsMap& get_options() const;
+    OptionsMap&       get_options();
+
+    int get_hashfull(int maxAge = 0) const;
+
+    std::string                            fen() const;
+    void                                   flip();
+    std::string                            visualize() const;
+    std::vector<std::pair<size_t, size_t>> get_bound_thread_count_by_numa_node() const;
+    std::string                            get_numa_config_as_string() const;
+    std::string                            numa_config_information_as_string() const;
+    std::string                            thread_allocation_information_as_string() const;
+    std::string                            thread_binding_information_as_string() const;
+
+   private:
+    const std::string binaryDirectory;
+
+    NumaReplicationContext numaContext;
+
+    Position     pos;
+    StateListPtr states;
+
+    OptionsMap                               options;
+    ThreadPool                               threads;
+    TranspositionTable                       tt;
+    LazyNumaReplicated<Eval::NNUE::Networks> networks;
+
+    Search::SearchManager::UpdateContext  updateContext;
+    std::function<void(std::string_view)> onVerifyNetworks;
+};
+
+}  // namespace Stockfish
+
+
+#endif  // #ifndef ENGINE_H_INCLUDED

src/evaluate.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -20,7 +20,6 @@
 #define EVALUATE_H_INCLUDED
 
 #include <string>
-#include <optional>
 
 #include "types.h"
 
@@ -30,26 +29,28 @@ class Position;
 
 namespace Eval {
 
-  std::string trace(Position& pos);
+// The default net name MUST follow the format nn-[SHA256 first 12 digits].nnue
-  Value evaluate(const Position& pos);
+// for the build process (profile-build and fishtest) to work. Do not change the
+// name of the macro or the location where this macro is defined, as it is used
+// in the Makefile/Fishtest.
+#define EvalFileDefaultNameBig "nn-287197aae49d.nnue"
+#define EvalFileDefaultNameSmall "nn-287197aae49d.nnue"
 
-  extern bool useNNUE;
+namespace NNUE {
-  extern std::string currentEvalFileName;
+struct Networks;
+struct AccumulatorCaches;
+}
 
-  // The default net name MUST follow the format nn-[SHA256 first 12 digits].nnue
+std::string trace(Position& pos, const Eval::NNUE::Networks& networks);
-  // 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"
 
-  namespace NNUE {
+int   simple_eval(const Position& pos, Color c);
+bool  use_smallnet(const Position& pos);
+Value evaluate(const NNUE::Networks&          networks,
+               const Position&                pos,
+               Eval::NNUE::AccumulatorCaches& caches,
+               int                            optimism);
+}  // namespace Eval
 
-    void init();
+}  // namespace Stockfish
-    void verify();
 
-  } // namespace NNUE
+#endif  // #ifndef EVALUATE_H_INCLUDED
-
-} // namespace Eval
-
-} // namespace Stockfish
-
-#endif // #ifndef EVALUATE_H_INCLUDED

src/history.h

@@ -0,0 +1,165 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 HISTORY_H_INCLUDED
+#define HISTORY_H_INCLUDED
+
+#include <algorithm>
+#include <array>
+#include <cassert>
+#include <cmath>
+#include <cstdint>
+#include <cstdlib>
+#include <limits>
+#include <type_traits>  // IWYU pragma: keep
+
+#include "misc.h"
+#include "position.h"
+
+namespace Stockfish {
+
+constexpr int PAWN_HISTORY_SIZE        = 512;    // has to be a power of 2
+constexpr int CORRECTION_HISTORY_SIZE  = 32768;  // has to be a power of 2
+constexpr int CORRECTION_HISTORY_LIMIT = 1024;
+constexpr int LOW_PLY_HISTORY_SIZE     = 4;
+
+static_assert((PAWN_HISTORY_SIZE & (PAWN_HISTORY_SIZE - 1)) == 0,
+              "PAWN_HISTORY_SIZE has to be a power of 2");
+
+static_assert((CORRECTION_HISTORY_SIZE & (CORRECTION_HISTORY_SIZE - 1)) == 0,
+              "CORRECTION_HISTORY_SIZE has to be a power of 2");
+
+enum PawnHistoryType {
+    Normal,
+    Correction
+};
+
+template<PawnHistoryType T = Normal>
+inline int pawn_structure_index(const Position& pos) {
+    return pos.pawn_key() & ((T == Normal ? PAWN_HISTORY_SIZE : CORRECTION_HISTORY_SIZE) - 1);
+}
+
+inline int minor_piece_index(const Position& pos) {
+    return pos.minor_piece_key() & (CORRECTION_HISTORY_SIZE - 1);
+}
+
+template<Color c>
+inline int non_pawn_index(const Position& pos) {
+    return pos.non_pawn_key(c) & (CORRECTION_HISTORY_SIZE - 1);
+}
+
+// StatsEntry is the container of various numerical statistics. We use a class
+// instead of a naked value to directly call history update operator<<() on
+// the entry. The first template parameter T is the base type of the array,
+// and the second template parameter D limits the range of updates in [-D, D]
+// when we update values with the << operator
+template<typename T, int D>
+class StatsEntry {
+
+    static_assert(std::is_arithmetic<T>::value, "Not an arithmetic type");
+    static_assert(D <= std::numeric_limits<T>::max(), "D overflows T");
+
+    T entry;
+
+   public:
+    StatsEntry& operator=(const T& v) {
+        entry = v;
+        return *this;
+    }
+    operator const T&() const { return entry; }
+
+    void operator<<(int bonus) {
+        // Make sure that bonus is in range [-D, D]
+        int clampedBonus = std::clamp(bonus, -D, D);
+        entry += clampedBonus - entry * std::abs(clampedBonus) / D;
+
+        assert(std::abs(entry) <= D);
+    }
+};
+
+enum StatsType {
+    NoCaptures,
+    Captures
+};
+
+template<typename T, int D, std::size_t... Sizes>
+using Stats = MultiArray<StatsEntry<T, D>, Sizes...>;
+
+// 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 https://www.chessprogramming.org/Butterfly_Boards (~11 elo)
+using ButterflyHistory = Stats<std::int16_t, 7183, COLOR_NB, int(SQUARE_NB) * int(SQUARE_NB)>;
+
+// LowPlyHistory is adressed by play and move's from and to squares, used
+// to improve move ordering near the root
+using LowPlyHistory =
+  Stats<std::int16_t, 7183, LOW_PLY_HISTORY_SIZE, int(SQUARE_NB) * int(SQUARE_NB)>;
+
+// CapturePieceToHistory is addressed by a move's [piece][to][captured piece type]
+using CapturePieceToHistory = Stats<std::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<std::int16_t, 30000, 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 = MultiArray<PieceToHistory, PIECE_NB, SQUARE_NB>;
+
+// PawnHistory is addressed by the pawn structure and a move's [piece][to]
+using PawnHistory = Stats<std::int16_t, 8192, PAWN_HISTORY_SIZE, PIECE_NB, SQUARE_NB>;
+
+// Correction histories record differences between the static evaluation of
+// positions and their search score. It is used to improve the static evaluation
+// used by some search heuristics.
+// see https://www.chessprogramming.org/Static_Evaluation_Correction_History
+enum CorrHistType {
+    Pawn,          // By color and pawn structure
+    Minor,         // By color and positions of minor pieces (Knight, Bishop)
+    NonPawn,       // By non-pawn material positions and color
+    PieceTo,       // By [piece][to] move
+    Continuation,  // Combined history of move pairs
+};
+
+namespace Detail {
+
+template<CorrHistType>
+struct CorrHistTypedef {
+    using type = Stats<std::int16_t, CORRECTION_HISTORY_LIMIT, CORRECTION_HISTORY_SIZE, COLOR_NB>;
+};
+
+template<>
+struct CorrHistTypedef<PieceTo> {
+    using type = Stats<std::int16_t, CORRECTION_HISTORY_LIMIT, PIECE_NB, SQUARE_NB>;
+};
+
+template<>
+struct CorrHistTypedef<Continuation> {
+    using type = MultiArray<CorrHistTypedef<PieceTo>::type, PIECE_NB, SQUARE_NB>;
+};
+
+}
+
+template<CorrHistType T>
+using CorrectionHistory = typename Detail::CorrHistTypedef<T>::type;
+
+}  // namespace Stockfish
+
+#endif  // #ifndef HISTORY_H_INCLUDED

src/incbin/incbin.h

@@ -26,7 +26,9 @@
       defined(__SSSE3__)    || \
       defined(__SSE4_1__)   || \
       defined(__SSE4_2__)   || \
-      defined(__neon__)
+      defined(__neon__)     || \
+      defined(__ARM_NEON)   || \
+      defined(__ALTIVEC__)
 # define INCBIN_ALIGNMENT_INDEX 4
 #elif ULONG_MAX != 0xffffffffu
 # define INCBIN_ALIGNMENT_INDEX 3
@@ -64,6 +66,9 @@
     X
 #define INCBIN_INVOKE(N, ...) \
     INCBIN_EVAL(N(__VA_ARGS__))
+/* Variable argument count for overloading by arity */
+#define INCBIN_VA_ARG_COUNTER(_1, _2, _3, N, ...) N
+#define INCBIN_VA_ARGC(...) INCBIN_VA_ARG_COUNTER(__VA_ARGS__, 3, 2, 1, 0)
 
 /* Green Hills uses a different directive for including binary data */
 #if defined(__ghs__)
@@ -117,28 +122,49 @@
 #endif
 
 /**
- * @brief Optionally override the linker section into which data is emitted.
+ * @brief Optionally override the linker section into which size and data is
- *
+ * emitted.
- * @warning If you use this facility, you'll have to deal with platform-specific linker output
+ * 
- * section naming on your own
+ * @warning If you use this facility, you might 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"
+#    define INCBIN_OUTPUT_SECTION ".const_data"
 #  else
-#    define INCBIN_OUTPUT_SECTION         ".rodata"
+#    define INCBIN_OUTPUT_SECTION ".rodata"
 #  endif
 #endif
 
+/**
+ * @brief Optionally override the linker section into which data is emitted.
+ *
+ * @warning If you use this facility, you might have to deal with
+ * platform-specific linker output section naming on your own.
+ */
+#if !defined(INCBIN_OUTPUT_DATA_SECTION)
+#  define INCBIN_OUTPUT_DATA_SECTION INCBIN_OUTPUT_SECTION
+#endif
+
+/**
+ * @brief Optionally override the linker section into which size is emitted.
+ *
+ * @warning If you use this facility, you might have to deal with
+ * platform-specific linker output section naming on your own.
+ * 
+ * @note This is useful for Harvard architectures where program memory cannot
+ * be directly read from the program without special instructions. With this you
+ * can chose to put the size variable in RAM rather than ROM.
+ */
+#if !defined(INCBIN_OUTPUT_SIZE_SECTION)
+#  define INCBIN_OUTPUT_SIZE_SECTION INCBIN_OUTPUT_SECTION
+#endif
+
 #if defined(__APPLE__)
+#  include "TargetConditionals.h"
+#  if defined(TARGET_OS_IPHONE) && !defined(INCBIN_SILENCE_BITCODE_WARNING)
+#    warning "incbin is incompatible with bitcode. Using the library will break upload to App Store if you have bitcode enabled. Add `#define INCBIN_SILENCE_BITCODE_WARNING` before including this header to silence this warning."
+#  endif
 /* 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"
@@ -179,27 +205,17 @@
 /**
  * @brief Specify the prefix to use for symbol names.
  *
- * By default this is `g', producing symbols of the form:
+ * @note By default this is "g".
- * @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 incbinFoo<data>[];
- * // const unsigned char *const incbinFooEnd;
+ * // const unsigned char *const incbinFoo<end>;
- * // const unsigned int incbinFooSize;
+ * // const unsigned int incbinFoo<size>;
  * @endcode
  */
 #if !defined(INCBIN_PREFIX)
@@ -213,18 +229,8 @@
  * - INCBIN_STYLE_CAMEL "CamelCase"
  * - INCBIN_STYLE_SNAKE "snake_case"
  *
- * Default option is *INCBIN_STYLE_CAMEL* producing symbols of the form:
+ * @note By default this is INCBIN_STYLE_CAMEL
- * @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"
@@ -288,23 +294,38 @@
  * 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 TYPE Optional array type. Omitting this picks a default of `unsigned char`.
  * @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 <prefix>Foo<data>[];
- * // extern const unsigned char *const <prefix>FooEnd;
+ * // extern const unsigned char *const <prefix>Foo<end>;
- * // extern const unsigned int <prefix>FooSize;
+ * // extern const unsigned int <prefix>Foo<size>;
+ * @endcode
+ * 
+ * You may specify a custom optional data type as well as the first argument.
+ * @code
+ * INCBIN_EXTERN(custom_type, Foo);
+ * 
+ * // Now you have the following symbols:
+ * // extern const custom_type <prefix>Foo<data>[];
+ * // extern const custom_type *const <prefix>Foo<end>;
+ * // extern const unsigned int <prefix>Foo<size>;
  * @endcode
  */
-#define INCBIN_EXTERN(NAME) \
+#define INCBIN_EXTERN(...) \
-    INCBIN_EXTERNAL const INCBIN_ALIGN unsigned char \
+    INCBIN_CONCATENATE(INCBIN_EXTERN_, INCBIN_VA_ARGC(__VA_ARGS__))(__VA_ARGS__)
+#define INCBIN_EXTERN_1(NAME, ...) \
+    INCBIN_EXTERN_2(unsigned char, NAME)
+#define INCBIN_EXTERN_2(TYPE, NAME) \
+    INCBIN_EXTERNAL const INCBIN_ALIGN TYPE \
         INCBIN_CONCATENATE( \
             INCBIN_CONCATENATE(INCBIN_PREFIX, NAME), \
             INCBIN_STYLE_IDENT(DATA))[]; \
-    INCBIN_EXTERNAL const INCBIN_ALIGN unsigned char *const \
+    INCBIN_EXTERNAL const INCBIN_ALIGN TYPE *const \
     INCBIN_CONCATENATE( \
         INCBIN_CONCATENATE(INCBIN_PREFIX, NAME), \
         INCBIN_STYLE_IDENT(END)); \
@@ -313,6 +334,29 @@
             INCBIN_CONCATENATE(INCBIN_PREFIX, NAME), \
             INCBIN_STYLE_IDENT(SIZE))
 
+/**
+ * @brief Externally reference textual data included in another translation unit.
+ *
+ * Produces three external symbols that reference the textual 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 textual data
+ *
+ * @code
+ * INCBIN_EXTERN(Foo);
+ *
+ * // Now you have the following symbols:
+ * // extern const char <prefix>Foo<data>[];
+ * // extern const char *const <prefix>Foo<end>;
+ * // extern const unsigned int <prefix>Foo<size>;
+ * @endcode
+ */
+#define INCTXT_EXTERN(NAME) \
+    INCBIN_EXTERN_2(char, NAME)
+
 /**
  * @brief Include a binary file into the current translation unit.
  *
@@ -322,6 +366,7 @@
  * 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 TYPE Optional array type. Omitting this picks a default of `unsigned char`.
  * @param NAME The name to associate with this binary data (as an identifier.)
  * @param FILENAME The file to include (as a string literal.)
  *
@@ -329,9 +374,20 @@
  * INCBIN(Icon, "icon.png");
  *
  * // Now you have the following symbols:
- * // const unsigned char <prefix>IconData[];
+ * // const unsigned char <prefix>Icon<data>[];
- * // const unsigned char *const <prefix>IconEnd;
+ * // const unsigned char *const <prefix>Icon<end>;
- * // const unsigned int <prefix>IconSize;
+ * // const unsigned int <prefix>Icon<size>;
+ * @endcode
+ * 
+ * You may specify a custom optional data type as well as the first argument.
+ * These macros are specialized by arity.
+ * @code
+ * INCBIN(custom_type, Icon, "icon.png");
+ *
+ * // Now you have the following symbols:
+ * // const custom_type <prefix>Icon<data>[];
+ * // const custom_type *const <prefix>Icon<end>;
+ * // const unsigned int <prefix>Icon<size>;
  * @endcode
  *
  * @warning This must be used in global scope
@@ -341,15 +397,28 @@
  * please @see INCBIN_EXTERN.
  */
 #ifdef _MSC_VER
-#define INCBIN(NAME, FILENAME) \
+#  define INCBIN(NAME, FILENAME) \
-    INCBIN_EXTERN(NAME)
+      INCBIN_EXTERN(NAME)
 #else
-#define INCBIN(NAME, FILENAME) \
+#  define INCBIN(...) \
+     INCBIN_CONCATENATE(INCBIN_, INCBIN_VA_ARGC(__VA_ARGS__))(__VA_ARGS__)
+#  if defined(__GNUC__)
+#    define INCBIN_1(...) _Pragma("GCC error \"Single argument INCBIN not allowed\"")
+#  elif defined(__clang__)
+#    define INCBIN_1(...) _Pragma("clang error \"Single argument INCBIN not allowed\"")
+#  else
+#    define INCBIN_1(...) /* Cannot do anything here */
+#  endif
+#  define INCBIN_2(NAME, FILENAME) \
+      INCBIN_3(unsigned char, NAME, FILENAME)
+#  define INCBIN_3(TYPE, NAME, FILENAME) INCBIN_COMMON(TYPE, NAME, FILENAME, /* No terminator for binary data */)
+#  define INCBIN_COMMON(TYPE, NAME, FILENAME, TERMINATOR) \
     __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" \
+                TERMINATOR \
             INCBIN_GLOBAL_LABELS(NAME, END) \
             INCBIN_ALIGN_BYTE \
             INCBIN_MANGLE INCBIN_STRINGIZE(INCBIN_PREFIX) #NAME INCBIN_STYLE_STRING(END) ":\n" \
@@ -362,7 +431,46 @@
             INCBIN_ALIGN_HOST \
             ".text\n" \
     ); \
-    INCBIN_EXTERN(NAME)
+    INCBIN_EXTERN(TYPE, NAME)
+#endif
 
+/**
+ * @brief Include a textual file into the current translation unit.
+ * 
+ * This behaves the same as INCBIN except it produces char compatible arrays
+ * and implicitly adds a null-terminator byte, thus the size of data included
+ * by this is one byte larger than that of INCBIN.
+ *
+ * Includes a textual 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
+ * INCTXT(Readme, "readme.txt");
+ *
+ * // Now you have the following symbols:
+ * // const char <prefix>Readme<data>[];
+ * // const char *const <prefix>Readme<end>;
+ * // const unsigned int <prefix>Readme<size>;
+ * @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.
+ */
+#if defined(_MSC_VER)
+#  define INCTXT(NAME, FILENAME) \
+     INCBIN_EXTERN(NAME)
+#else
+#  define INCTXT(NAME, FILENAME) \
+     INCBIN_COMMON(char, NAME, FILENAME, INCBIN_BYTE "0\n")
 #endif
-#endif
+
+#endif

src/main.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,35 +19,26 @@
 #include <iostream>
 
 #include "bitboard.h"
-#include "endgame.h"
+#include "misc.h"
 #include "position.h"
-#include "psqt.h"
+#include "types.h"
-#include "search.h"
-#include "syzygy/tbprobe.h"
-#include "thread.h"
-#include "tt.h"
 #include "uci.h"
+#include "tune.h"
 
 using namespace Stockfish;
 
 int main(int argc, char* argv[]) {
 
-  std::cout << engine_info() << std::endl;
+    std::cout << engine_info() << std::endl;
 
-  CommandLine::init(argc, argv);
+    Bitboards::init();
-  UCI::init(Options);
+    Position::init();
-  Tune::init();
-  PSQT::init();
-  Bitboards::init();
-  Position::init();
-  Bitbases::init();
-  Endgames::init();
-  Threads.set(size_t(Options["Threads"]));
-  Search::clear(); // After threads are up
-  Eval::NNUE::init();
 
-  UCI::loop(argc, argv);
+    UCIEngine uci(argc, argv);
 
-  Threads.set(0);
+    Tune::init(uci.engine_options());
-  return 0;
+
+    uci.loop();
+
+    return 0;
 }

src/material.cpp

@@ -1,229 +0,0 @@
-/*
-  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 <cassert>
-#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)
-
-  // Polynomial material imbalance parameters
-
-  // 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
-  };
-
-  // One Score parameter for each pair (our piece, their piece)
-  constexpr Score QuadraticTheirs[][PIECE_TYPE_NB] = {
-    // THEIR PIECE
-    // bishop pair   pawn         knight       bishop       rook         queen
-    {                                                                               }, // Bishop pair
-    {S(  33,  30)                                                                   }, // Pawn
-    {S(  46,  18), S(106,  84)                                                      }, // Knight      OUR PIECE
-    {S(  75,  35), S( 59,  44), S( 60,  15)                                         }, // Bishop
-    {S(  26,  35), S(  6,  22), S( 38,  39), S(-12,  -2)                            }, // Rook
-    {S(  97,  93), S(100, 163), S(-58, -91), S(112, 192), S(276, 225)               }  // Queen
-  };
-
-  #undef S
-
-  // Endgame evaluation and scaling functions are accessed directly and not through
-  // the function maps because they correspond to more than one material hash key.
-  Endgame<KXK>    EvaluateKXK[] = { Endgame<KXK>(WHITE),    Endgame<KXK>(BLACK) };
-
-  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 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;
-  }
-
-  bool is_KBPsK(const Position& pos, Color us) {
-    return   pos.non_pawn_material(us) == BishopValueMg
-          && pos.count<PAWN>(us) >= 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 {
-
-
-/// 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();
-  Entry* e = pos.this_thread()->materialTable[key];
-
-  if (e->key == key)
-      return e;
-
-  std::memset(e, 0, sizeof(Entry));
-  e->key = key;
-  e->factor[WHITE] = e->factor[BLACK] = (uint8_t)SCALE_FACTOR_NORMAL;
-
-  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);
-
-  // 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))
-      {
-          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.count<PAWN>(WHITE))
-      {
-          assert(pos.count<PAWN>(BLACK) >= 2);
-
-          e->scalingFunction[BLACK] = &ScaleKPsK[BLACK];
-      }
-      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.
-          e->scalingFunction[WHITE] = &ScaleKPKP[WHITE];
-          e->scalingFunction[BLACK] = &ScaleKPKP[BLACK];
-      }
-  }
-
-  // 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.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", which allows us to be more flexible
-  // in defining bishop pair bonuses.
-  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->score = (imbalance<WHITE>(pieceCount) - imbalance<BLACK>(pieceCount)) / 16;
-  return e;
-}
-
-} // namespace Material
-
-} // namespace Stockfish

src/material.h

@@ -1,71 +0,0 @@
-/*
-  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 MATERIAL_H_INCLUDED
-#define MATERIAL_H_INCLUDED
-
-#include "endgame.h"
-#include "misc.h"
-#include "position.h"
-#include "types.h"
-
-namespace Stockfish::Material {
-
-/// 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.
-
-struct Entry {
-
-  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); }
-
-  // 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]);
-  }
-
-  Key key;
-  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>;
-
-Entry* probe(const Position& pos);
-
-} // namespace Stockfish::Material
-
-#endif // #ifndef MATERIAL_H_INCLUDED

src/memory.cpp

@@ -0,0 +1,268 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 "memory.h"
+
+#include <cstdlib>
+
+#if __has_include("features.h")
+    #include <features.h>
+#endif
+
+#if defined(__linux__) && !defined(__ANDROID__)
+    #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
+
+#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 <ios>       // std::hex, std::dec
+    #include <iostream>  // std::cerr
+    #include <ostream>   // std::endl
+    #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 OpenProcessToken_t      = bool (*)(HANDLE, DWORD, PHANDLE);
+using LookupPrivilegeValueA_t = bool (*)(LPCSTR, LPCSTR, PLUID);
+using AdjustTokenPrivileges_t =
+  bool (*)(HANDLE, BOOL, PTOKEN_PRIVILEGES, DWORD, PTOKEN_PRIVILEGES, PDWORD);
+}
+#endif
+
+
+namespace Stockfish {
+
+// Wrappers 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(_ISOC11_SOURCE)
+    return aligned_alloc(alignment, size);
+#elif defined(POSIXALIGNEDALLOC)
+    void* mem = nullptr;
+    posix_memalign(&mem, alignment, size);
+    return 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 OpenProcessToken_f =
+      OpenProcessToken_t((void (*)()) GetProcAddress(hAdvapi32, "OpenProcessToken"));
+    if (!OpenProcessToken_f)
+        return nullptr;
+    auto LookupPrivilegeValueA_f =
+      LookupPrivilegeValueA_t((void (*)()) GetProcAddress(hAdvapi32, "LookupPrivilegeValueA"));
+    if (!LookupPrivilegeValueA_f)
+        return nullptr;
+    auto AdjustTokenPrivileges_f =
+      AdjustTokenPrivileges_t((void (*)()) GetProcAddress(hAdvapi32, "AdjustTokenPrivileges"));
+    if (!AdjustTokenPrivileges_f)
+        return nullptr;
+
+    // We need SeLockMemoryPrivilege, so try to enable it for the process
+
+    if (!OpenProcessToken_f(  // OpenProcessToken()
+          GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES | TOKEN_QUERY, &hProcessToken))
+        return nullptr;
+
+    if (LookupPrivilegeValueA_f(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 (AdjustTokenPrivileges_f(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
+            AdjustTokenPrivileges_f(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;  // 2MB page size assumed
+    #else
+    constexpr size_t alignment = 4096;  // small page size assumed
+    #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
+
+bool has_large_pages() {
+
+#if defined(_WIN32)
+
+    constexpr size_t page_size = 2 * 1024 * 1024;  // 2MB page size assumed
+    void*            mem       = aligned_large_pages_alloc_windows(page_size);
+    if (mem == nullptr)
+    {
+        return false;
+    }
+    else
+    {
+        aligned_large_pages_free(mem);
+        return true;
+    }
+
+#elif defined(__linux__)
+
+    #if defined(MADV_HUGEPAGE)
+    return true;
+    #else
+    return false;
+    #endif
+
+#else
+
+    return false;
+
+#endif
+}
+
+
+// aligned_large_pages_free() will free the previously memory allocated
+// by aligned_large_pages_alloc(). The effect is a nop if mem == nullptr.
+
+#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 Stockfish

src/memory.h

@@ -0,0 +1,218 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 MEMORY_H_INCLUDED
+#define MEMORY_H_INCLUDED
+
+#include <algorithm>
+#include <cstddef>
+#include <cstdint>
+#include <memory>
+#include <new>
+#include <type_traits>
+#include <utility>
+
+#include "types.h"
+
+namespace Stockfish {
+
+void* std_aligned_alloc(size_t alignment, size_t size);
+void  std_aligned_free(void* ptr);
+
+// Memory aligned by page size, min alignment: 4096 bytes
+void* aligned_large_pages_alloc(size_t size);
+void  aligned_large_pages_free(void* mem);
+
+bool has_large_pages();
+
+// Frees memory which was placed there with placement new.
+// Works for both single objects and arrays of unknown bound.
+template<typename T, typename FREE_FUNC>
+void memory_deleter(T* ptr, FREE_FUNC free_func) {
+    if (!ptr)
+        return;
+
+    // Explicitly needed to call the destructor
+    if constexpr (!std::is_trivially_destructible_v<T>)
+        ptr->~T();
+
+    free_func(ptr);
+    return;
+}
+
+// Frees memory which was placed there with placement new.
+// Works for both single objects and arrays of unknown bound.
+template<typename T, typename FREE_FUNC>
+void memory_deleter_array(T* ptr, FREE_FUNC free_func) {
+    if (!ptr)
+        return;
+
+
+    // Move back on the pointer to where the size is allocated
+    const size_t array_offset = std::max(sizeof(size_t), alignof(T));
+    char*        raw_memory   = reinterpret_cast<char*>(ptr) - array_offset;
+
+    if constexpr (!std::is_trivially_destructible_v<T>)
+    {
+        const size_t size = *reinterpret_cast<size_t*>(raw_memory);
+
+        // Explicitly call the destructor for each element in reverse order
+        for (size_t i = size; i-- > 0;)
+            ptr[i].~T();
+    }
+
+    free_func(raw_memory);
+}
+
+// Allocates memory for a single object and places it there with placement new
+template<typename T, typename ALLOC_FUNC, typename... Args>
+inline std::enable_if_t<!std::is_array_v<T>, T*> memory_allocator(ALLOC_FUNC alloc_func,
+                                                                  Args&&... args) {
+    void* raw_memory = alloc_func(sizeof(T));
+    ASSERT_ALIGNED(raw_memory, alignof(T));
+    return new (raw_memory) T(std::forward<Args>(args)...);
+}
+
+// Allocates memory for an array of unknown bound and places it there with placement new
+template<typename T, typename ALLOC_FUNC>
+inline std::enable_if_t<std::is_array_v<T>, std::remove_extent_t<T>*>
+memory_allocator(ALLOC_FUNC alloc_func, size_t num) {
+    using ElementType = std::remove_extent_t<T>;
+
+    const size_t array_offset = std::max(sizeof(size_t), alignof(ElementType));
+
+    // Save the array size in the memory location
+    char* raw_memory =
+      reinterpret_cast<char*>(alloc_func(array_offset + num * sizeof(ElementType)));
+    ASSERT_ALIGNED(raw_memory, alignof(T));
+
+    new (raw_memory) size_t(num);
+
+    for (size_t i = 0; i < num; ++i)
+        new (raw_memory + array_offset + i * sizeof(ElementType)) ElementType();
+
+    // Need to return the pointer at the start of the array so that
+    // the indexing in unique_ptr<T[]> works.
+    return reinterpret_cast<ElementType*>(raw_memory + array_offset);
+}
+
+//
+//
+// aligned large page unique ptr
+//
+//
+
+template<typename T>
+struct LargePageDeleter {
+    void operator()(T* ptr) const { return memory_deleter<T>(ptr, aligned_large_pages_free); }
+};
+
+template<typename T>
+struct LargePageArrayDeleter {
+    void operator()(T* ptr) const { return memory_deleter_array<T>(ptr, aligned_large_pages_free); }
+};
+
+template<typename T>
+using LargePagePtr =
+  std::conditional_t<std::is_array_v<T>,
+                     std::unique_ptr<T, LargePageArrayDeleter<std::remove_extent_t<T>>>,
+                     std::unique_ptr<T, LargePageDeleter<T>>>;
+
+// make_unique_large_page for single objects
+template<typename T, typename... Args>
+std::enable_if_t<!std::is_array_v<T>, LargePagePtr<T>> make_unique_large_page(Args&&... args) {
+    static_assert(alignof(T) <= 4096,
+                  "aligned_large_pages_alloc() may fail for such a big alignment requirement of T");
+
+    T* obj = memory_allocator<T>(aligned_large_pages_alloc, std::forward<Args>(args)...);
+
+    return LargePagePtr<T>(obj);
+}
+
+// make_unique_large_page for arrays of unknown bound
+template<typename T>
+std::enable_if_t<std::is_array_v<T>, LargePagePtr<T>> make_unique_large_page(size_t num) {
+    using ElementType = std::remove_extent_t<T>;
+
+    static_assert(alignof(ElementType) <= 4096,
+                  "aligned_large_pages_alloc() may fail for such a big alignment requirement of T");
+
+    ElementType* memory = memory_allocator<T>(aligned_large_pages_alloc, num);
+
+    return LargePagePtr<T>(memory);
+}
+
+//
+//
+// aligned unique ptr
+//
+//
+
+template<typename T>
+struct AlignedDeleter {
+    void operator()(T* ptr) const { return memory_deleter<T>(ptr, std_aligned_free); }
+};
+
+template<typename T>
+struct AlignedArrayDeleter {
+    void operator()(T* ptr) const { return memory_deleter_array<T>(ptr, std_aligned_free); }
+};
+
+template<typename T>
+using AlignedPtr =
+  std::conditional_t<std::is_array_v<T>,
+                     std::unique_ptr<T, AlignedArrayDeleter<std::remove_extent_t<T>>>,
+                     std::unique_ptr<T, AlignedDeleter<T>>>;
+
+// make_unique_aligned for single objects
+template<typename T, typename... Args>
+std::enable_if_t<!std::is_array_v<T>, AlignedPtr<T>> make_unique_aligned(Args&&... args) {
+    const auto func = [](size_t size) { return std_aligned_alloc(alignof(T), size); };
+    T*         obj  = memory_allocator<T>(func, std::forward<Args>(args)...);
+
+    return AlignedPtr<T>(obj);
+}
+
+// make_unique_aligned for arrays of unknown bound
+template<typename T>
+std::enable_if_t<std::is_array_v<T>, AlignedPtr<T>> make_unique_aligned(size_t num) {
+    using ElementType = std::remove_extent_t<T>;
+
+    const auto   func   = [](size_t size) { return std_aligned_alloc(alignof(ElementType), size); };
+    ElementType* memory = memory_allocator<T>(func, num);
+
+    return AlignedPtr<T>(memory);
+}
+
+
+// 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));
+}
+
+
+}  // namespace Stockfish
+
+#endif  // #ifndef MEMORY_H_INCLUDED

src/misc.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,158 +19,298 @@
 #ifndef MISC_H_INCLUDED
 #define MISC_H_INCLUDED
 
+#include <algorithm>
+#include <array>
 #include <cassert>
 #include <chrono>
-#include <ostream>
+#include <cstddef>
-#include <string>
-#include <vector>
 #include <cstdint>
-
+#include <cstdio>
-#include "types.h"
+#include <iosfwd>
+#include <optional>
+#include <string>
+#include <string_view>
+#include <vector>
 
 #define stringify2(x) #x
 #define stringify(x) stringify2(x)
 
 namespace Stockfish {
 
+std::string engine_version_info();
 std::string engine_info(bool to_uci = false);
 std::string compiler_info();
-void prefetch(void* addr);
+
+// 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.
+void prefetch(const 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);
+size_t str_to_size_t(const std::string& s);
-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
+#if defined(__linux__)
+
+struct PipeDeleter {
+    void operator()(FILE* file) const {
+        if (file != nullptr)
+        {
+            pclose(file);
+        }
+    }
+};
+
+#endif
+
+// Reads the file as bytes.
+// Returns std::nullopt if the file does not exist.
+std::optional<std::string> read_file_to_string(const std::string& path);
 
 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_extremes_of(int64_t value, int slot = 0);
 void dbg_correl_of(int64_t value1, int64_t value2, int slot = 0);
 void dbg_print();
 
-using TimePoint = std::chrono::milliseconds::rep; // A value in milliseconds
+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>
+    return std::chrono::duration_cast<std::chrono::milliseconds>(
-        (std::chrono::steady_clock::now().time_since_epoch()).count();
+             std::chrono::steady_clock::now().time_since_epoch())
+      .count();
 }
 
-template<class Entry, int Size>
+inline std::vector<std::string_view> split(std::string_view s, std::string_view delimiter) {
-struct HashTable {
+    std::vector<std::string_view> res;
-  Entry* operator[](Key key) { return &table[(uint32_t)key & (Size - 1)]; }
 
-private:
+    if (s.empty())
-  std::vector<Entry> table = std::vector<Entry>(Size); // Allocate on the heap
+        return res;
+
+    size_t begin = 0;
+    for (;;)
+    {
+        const size_t end = s.find(delimiter, begin);
+        if (end == std::string::npos)
+            break;
+
+        res.emplace_back(s.substr(begin, end - begin));
+        begin = end + delimiter.size();
+    }
+
+    res.emplace_back(s.substr(begin));
+
+    return res;
+}
+
+void remove_whitespace(std::string& s);
+bool is_whitespace(std::string_view s);
+
+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
 
+void sync_cout_start();
+void sync_cout_end();
 
-// align_ptr_up() : get the first aligned element of an array.
+// True if and only if the binary is compiled on a little-endian machine
-// ptr must point to an array of size at least `sizeof(T) * N + alignment` bytes,
+static inline const std::uint16_t Le             = 1;
-// where N is the number of elements in the array.
+static inline const bool          IsLittleEndian = *reinterpret_cast<const char*>(&Le) == 1;
-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
+template<typename T, std::size_t MaxSize>
-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:
+   public:
-  std::size_t size() const { return size_; }
+    std::size_t size() const { return size_; }
-  void push_back(const T& value) { values_[size_++] = value; }
+    void        push_back(const T& value) { values_[size_++] = value; }
-  const T* begin() const { return values_; }
+    const T*    begin() const { return values_; }
-  const T* end() const { return values_ + size_; }
+    const T*    end() const { return values_ + size_; }
+    const T&    operator[](int index) const { return values_[index]; }
 
-private:
+   private:
-  T values_[MaxSize];
+    T           values_[MaxSize];
-  std::size_t size_ = 0;
+    std::size_t size_ = 0;
 };
 
 
-/// xorshift64star Pseudo-Random Number Generator
+template<typename T, std::size_t Size, std::size_t... Sizes>
-/// This class is based on original code written and dedicated
+class MultiArray;
-/// to the public domain by Sebastiano Vigna (2014).
+
-/// It has the following characteristics:
+namespace Detail {
-///
+
-///  -  Outputs 64-bit numbers
+template<typename T, std::size_t Size, std::size_t... Sizes>
-///  -  Passes Dieharder and SmallCrush test batteries
+struct MultiArrayHelper {
-///  -  Does not require warm-up, no zeroland to escape
+    using ChildType = MultiArray<T, Sizes...>;
-///  -  Internal state is a single 64-bit integer
+};
-///  -  Period is 2^64 - 1
+
-///  -  Speed: 1.60 ns/call (Core i7 @3.40GHz)
+template<typename T, std::size_t Size>
-///
+struct MultiArrayHelper<T, Size> {
-/// For further analysis see
+    using ChildType = T;
-///   <http://vigna.di.unimi.it/ftp/papers/xorshift.pdf>
+};
+
+}
+
+// MultiArray is a generic N-dimensional array.
+// The template parameters (Size and Sizes) encode the dimensions of the array.
+template<typename T, std::size_t Size, std::size_t... Sizes>
+class MultiArray {
+    using ChildType = typename Detail::MultiArrayHelper<T, Size, Sizes...>::ChildType;
+    using ArrayType = std::array<ChildType, Size>;
+    ArrayType data_;
+
+   public:
+    using value_type             = typename ArrayType::value_type;
+    using size_type              = typename ArrayType::size_type;
+    using difference_type        = typename ArrayType::difference_type;
+    using reference              = typename ArrayType::reference;
+    using const_reference        = typename ArrayType::const_reference;
+    using pointer                = typename ArrayType::pointer;
+    using const_pointer          = typename ArrayType::const_pointer;
+    using iterator               = typename ArrayType::iterator;
+    using const_iterator         = typename ArrayType::const_iterator;
+    using reverse_iterator       = typename ArrayType::reverse_iterator;
+    using const_reverse_iterator = typename ArrayType::const_reverse_iterator;
+
+    constexpr auto&       at(size_type index) noexcept { return data_.at(index); }
+    constexpr const auto& at(size_type index) const noexcept { return data_.at(index); }
+
+    constexpr auto&       operator[](size_type index) noexcept { return data_[index]; }
+    constexpr const auto& operator[](size_type index) const noexcept { return data_[index]; }
+
+    constexpr auto&       front() noexcept { return data_.front(); }
+    constexpr const auto& front() const noexcept { return data_.front(); }
+    constexpr auto&       back() noexcept { return data_.back(); }
+    constexpr const auto& back() const noexcept { return data_.back(); }
+
+    auto*       data() { return data_.data(); }
+    const auto* data() const { return data_.data(); }
+
+    constexpr auto begin() noexcept { return data_.begin(); }
+    constexpr auto end() noexcept { return data_.end(); }
+    constexpr auto begin() const noexcept { return data_.begin(); }
+    constexpr auto end() const noexcept { return data_.end(); }
+    constexpr auto cbegin() const noexcept { return data_.cbegin(); }
+    constexpr auto cend() const noexcept { return data_.cend(); }
+
+    constexpr auto rbegin() noexcept { return data_.rbegin(); }
+    constexpr auto rend() noexcept { return data_.rend(); }
+    constexpr auto rbegin() const noexcept { return data_.rbegin(); }
+    constexpr auto rend() const noexcept { return data_.rend(); }
+    constexpr auto crbegin() const noexcept { return data_.crbegin(); }
+    constexpr auto crend() const noexcept { return data_.crend(); }
+
+    constexpr bool      empty() const noexcept { return data_.empty(); }
+    constexpr size_type size() const noexcept { return data_.size(); }
+    constexpr size_type max_size() const noexcept { return data_.max_size(); }
+
+    template<typename U>
+    void fill(const U& v) {
+        static_assert(std::is_assignable_v<T, U>, "Cannot assign fill value to entry type");
+        for (auto& ele : data_)
+        {
+            if constexpr (sizeof...(Sizes) == 0)
+                ele = v;
+            else
+                ele.fill(v);
+        }
+    }
+
+    constexpr void swap(MultiArray<T, Size, Sizes...>& other) noexcept { data_.swap(other.data_); }
+};
+
+
+// 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 s;
 
-  uint64_t rand64() {
+    uint64_t rand64() {
 
-    s ^= s >> 12, s ^= s << 25, s ^= s >> 27;
+        s ^= s >> 12, s ^= s << 25, s ^= s >> 27;
-    return s * 2685821657736338717LL;
+        return s * 2685821657736338717LL;
-  }
+    }
 
-public:
+   public:
-  PRNG(uint64_t seed) : s(seed) { assert(seed); }
+    PRNG(uint64_t seed) :
+        s(seed) {
+        assert(seed);
+    }
 
-  template<typename T> T rand() { return T(rand64()); }
+    template<typename T>
+    T rand() {
+        return T(rand64());
+    }
 
-  /// Special generator used to fast init magic numbers.
+    // Special generator used to fast init magic numbers.
-  /// Output values only have 1/8th of their bits set on average.
+    // Output values only have 1/8th of their bits set on average.
-  template<typename T> T sparse_rand()
+    template<typename T>
-  { return T(rand64() & rand64() & rand64()); }
+    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;
+    return (uint128(a) * uint128(b)) >> 64;
 #else
-    uint64_t aL = (uint32_t)a, aH = a >> 32;
+    uint64_t aL = uint32_t(a), aH = a >> 32;
-    uint64_t bL = (uint32_t)b, bH = b >> 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;
+    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 {
+struct CommandLine {
-  void bindThisThread(size_t idx);
+   public:
+    CommandLine(int _argc, char** _argv) :
+        argc(_argc),
+        argv(_argv) {}
+
+    static std::string get_binary_directory(std::string argv0);
+    static std::string get_working_directory();
+
+    int    argc;
+    char** argv;
+};
+
+namespace Utility {
+
+template<typename T, typename Predicate>
+void move_to_front(std::vector<T>& vec, Predicate pred) {
+    auto it = std::find_if(vec.begin(), vec.end(), pred);
+
+    if (it != vec.end())
+    {
+        std::rotate(vec.begin(), it, it + 1);
+    }
+}
 }
 
-namespace CommandLine {
+}  // namespace Stockfish
-  void init(int argc, char* argv[]);
 
-  extern std::string binaryDirectory;  // path of the executable directory
+#endif  // #ifndef MISC_H_INCLUDED
-  extern std::string workingDirectory; // path of the working directory
-}
-
-} // namespace Stockfish
-
-#endif // #ifndef MISC_H_INCLUDED

src/movegen.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,90 +16,75 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#include <cassert>
-
 #include "movegen.h"
+
+#include <cassert>
+#include <initializer_list>
+
+#include "bitboard.h"
 #include "position.h"
 
 namespace Stockfish {
 
 namespace {
 
-  template<GenType Type, Direction D, bool Enemy>
+template<GenType Type, Direction D, bool Enemy>
-  ExtMove* make_promotions(ExtMove* moveList, [[maybe_unused]] Square to) {
+ExtMove* make_promotions(ExtMove* moveList, [[maybe_unused]] Square to) {
 
-    if constexpr (Type == CAPTURES || Type == EVASIONS || Type == NON_EVASIONS)
+    constexpr bool all = Type == EVASIONS || Type == NON_EVASIONS;
-    {
-        *moveList++ = make<PROMOTION>(to - D, to, QUEEN);
-        if constexpr (Enemy && Type == CAPTURES)
-        {
-            *moveList++ = make<PROMOTION>(to - D, to, ROOK);
-            *moveList++ = make<PROMOTION>(to - D, to, BISHOP);
-            *moveList++ = make<PROMOTION>(to - D, to, KNIGHT);
-        }
-    }
 
-    if constexpr ((Type == QUIETS && !Enemy) || Type == EVASIONS || Type == NON_EVASIONS)
+    if constexpr (Type == CAPTURES || all)
+        *moveList++ = Move::make<PROMOTION>(to - D, to, QUEEN);
+
+    if constexpr ((Type == CAPTURES && Enemy) || (Type == QUIETS && !Enemy) || all)
     {
-        *moveList++ = make<PROMOTION>(to - D, to, ROOK);
+        *moveList++ = Move::make<PROMOTION>(to - D, to, ROOK);
-        *moveList++ = make<PROMOTION>(to - D, to, BISHOP);
+        *moveList++ = Move::make<PROMOTION>(to - D, to, BISHOP);
-        *moveList++ = make<PROMOTION>(to - D, to, KNIGHT);
+        *moveList++ = Move::make<PROMOTION>(to - D, to, KNIGHT);
     }
 
     return moveList;
-  }
+}
 
 
-  template<Color Us, GenType Type>
+template<Color Us, GenType Type>
-  ExtMove* generate_pawn_moves(const Position& pos, ExtMove* moveList, Bitboard target) {
+ExtMove* generate_pawn_moves(const Position& pos, ExtMove* moveList, Bitboard target) {
 
     constexpr Color     Them     = ~Us;
-    constexpr Bitboard  TRank7BB = (Us == WHITE ? Rank7BB    : Rank2BB);
+    constexpr Bitboard  TRank7BB = (Us == WHITE ? Rank7BB : Rank2BB);
-    constexpr Bitboard  TRank3BB = (Us == WHITE ? Rank3BB    : Rank6BB);
+    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);
 
     const Bitboard emptySquares = ~pos.pieces();
-    const Bitboard enemies      =  Type == EVASIONS ? pos.checkers()
+    const Bitboard enemies      = Type == EVASIONS ? pos.checkers() : pos.pieces(Them);
-                                                    : pos.pieces(Them);
 
-    Bitboard pawnsOn7    = pos.pieces(Us, PAWN) &  TRank7BB;
+    Bitboard pawnsOn7    = pos.pieces(Us, PAWN) & TRank7BB;
     Bitboard pawnsNotOn7 = pos.pieces(Us, PAWN) & ~TRank7BB;
 
     // Single and double pawn pushes, no promotions
     if constexpr (Type != CAPTURES)
     {
-        Bitboard b1 = shift<Up>(pawnsNotOn7)   & emptySquares;
+        Bitboard b1 = shift<Up>(pawnsNotOn7) & emptySquares;
         Bitboard b2 = shift<Up>(b1 & TRank3BB) & emptySquares;
 
-        if constexpr (Type == EVASIONS) // Consider only blocking squares
+        if constexpr (Type == EVASIONS)  // Consider only blocking squares
         {
             b1 &= target;
             b2 &= target;
         }
 
-        if constexpr (Type == QUIET_CHECKS)
-        {
-            // To make a quiet check, you either make a direct check by pushing a pawn
-            // or push a blocker pawn that is not on the same file as the enemy king.
-            // Discovered check promotion has been already generated amongst the captures.
-            Square ksq = pos.square<KING>(Them);
-            Bitboard dcCandidatePawns = pos.blockers_for_king(Them) & ~file_bb(ksq);
-            b1 &= pawn_attacks_bb(Them, ksq) | shift<   Up>(dcCandidatePawns);
-            b2 &= pawn_attacks_bb(Them, ksq) | shift<Up+Up>(dcCandidatePawns);
-        }
-
         while (b1)
         {
-            Square to = pop_lsb(b1);
+            Square to   = pop_lsb(b1);
-            *moveList++ = make_move(to - Up, to);
+            *moveList++ = Move(to - Up, to);
         }
 
         while (b2)
         {
-            Square to = pop_lsb(b2);
+            Square to   = pop_lsb(b2);
-            *moveList++ = make_move(to - Up - Up, to);
+            *moveList++ = Move(to - Up - Up, to);
         }
     }
 
@@ -107,8 +92,8 @@ namespace {
     if (pawnsOn7)
     {
         Bitboard b1 = shift<UpRight>(pawnsOn7) & enemies;
-        Bitboard b2 = shift<UpLeft >(pawnsOn7) & enemies;
+        Bitboard b2 = shift<UpLeft>(pawnsOn7) & enemies;
-        Bitboard b3 = shift<Up     >(pawnsOn7) & emptySquares;
+        Bitboard b3 = shift<Up>(pawnsOn7) & emptySquares;
 
         if constexpr (Type == EVASIONS)
             b3 &= target;
@@ -120,25 +105,25 @@ namespace {
             moveList = make_promotions<Type, UpLeft, true>(moveList, pop_lsb(b2));
 
         while (b3)
-            moveList = make_promotions<Type, Up,    false>(moveList, pop_lsb(b3));
+            moveList = make_promotions<Type, Up, false>(moveList, pop_lsb(b3));
     }
 
     // Standard and en passant captures
     if constexpr (Type == CAPTURES || Type == EVASIONS || Type == NON_EVASIONS)
     {
         Bitboard b1 = shift<UpRight>(pawnsNotOn7) & enemies;
-        Bitboard b2 = shift<UpLeft >(pawnsNotOn7) & enemies;
+        Bitboard b2 = shift<UpLeft>(pawnsNotOn7) & enemies;
 
         while (b1)
         {
-            Square to = pop_lsb(b1);
+            Square to   = pop_lsb(b1);
-            *moveList++ = make_move(to - UpRight, to);
+            *moveList++ = Move(to - UpRight, to);
         }
 
         while (b2)
         {
-            Square to = pop_lsb(b2);
+            Square to   = pop_lsb(b2);
-            *moveList++ = make_move(to - UpLeft, to);
+            *moveList++ = Move(to - UpLeft, to);
         }
 
         if (pos.ep_square() != SQ_NONE)
@@ -154,16 +139,16 @@ namespace {
             assert(b1);
 
             while (b1)
-                *moveList++ = make<EN_PASSANT>(pop_lsb(b1), pos.ep_square());
+                *moveList++ = Move::make<EN_PASSANT>(pop_lsb(b1), pos.ep_square());
         }
     }
 
     return moveList;
-  }
+}
 
 
-  template<Color Us, PieceType Pt, bool Checks>
+template<Color Us, PieceType Pt>
-  ExtMove* generate_moves(const Position& pos, ExtMove* moveList, Bitboard target) {
+ExtMove* generate_moves(const Position& pos, ExtMove* moveList, Bitboard target) {
 
     static_assert(Pt != KING && Pt != PAWN, "Unsupported piece type in generate_moves()");
 
@@ -171,114 +156,101 @@ namespace {
 
     while (bb)
     {
-        Square from = pop_lsb(bb);
+        Square   from = pop_lsb(bb);
-        Bitboard b = attacks_bb<Pt>(from, pos.pieces()) & target;
+        Bitboard b    = attacks_bb<Pt>(from, pos.pieces()) & target;
-
-        // To check, you either move freely a blocker or make a direct check.
-        if (Checks && (Pt == QUEEN || !(pos.blockers_for_king(~Us) & from)))
-            b &= pos.check_squares(Pt);
 
         while (b)
-            *moveList++ = make_move(from, pop_lsb(b));
+            *moveList++ = Move(from, pop_lsb(b));
     }
 
     return moveList;
-  }
+}
 
 
-  template<Color Us, GenType Type>
+template<Color Us, GenType Type>
-  ExtMove* generate_all(const Position& pos, ExtMove* moveList) {
+ExtMove* generate_all(const Position& pos, ExtMove* moveList) {
 
     static_assert(Type != LEGAL, "Unsupported type in generate_all()");
 
-    constexpr bool Checks = Type == QUIET_CHECKS; // Reduce template instantiations
     const Square ksq = pos.square<KING>(Us);
-    Bitboard target;
+    Bitboard     target;
 
     // Skip generating non-king moves when in double check
     if (Type != EVASIONS || !more_than_one(pos.checkers()))
     {
-        target = Type == EVASIONS     ?  between_bb(ksq, lsb(pos.checkers()))
+        target = Type == EVASIONS     ? between_bb(ksq, lsb(pos.checkers()))
-               : Type == NON_EVASIONS ? ~pos.pieces( Us)
+               : Type == NON_EVASIONS ? ~pos.pieces(Us)
-               : Type == CAPTURES     ?  pos.pieces(~Us)
+               : Type == CAPTURES     ? pos.pieces(~Us)
-                                      : ~pos.pieces(   ); // QUIETS || QUIET_CHECKS
+                                      : ~pos.pieces();  // QUIETS
 
         moveList = generate_pawn_moves<Us, Type>(pos, moveList, target);
-        moveList = generate_moves<Us, KNIGHT, Checks>(pos, moveList, target);
+        moveList = generate_moves<Us, KNIGHT>(pos, moveList, target);
-        moveList = generate_moves<Us, BISHOP, Checks>(pos, moveList, target);
+        moveList = generate_moves<Us, BISHOP>(pos, moveList, target);
-        moveList = generate_moves<Us,   ROOK, Checks>(pos, moveList, target);
+        moveList = generate_moves<Us, ROOK>(pos, moveList, target);
-        moveList = generate_moves<Us,  QUEEN, Checks>(pos, moveList, target);
+        moveList = generate_moves<Us, QUEEN>(pos, moveList, target);
     }
 
-    if (!Checks || pos.blockers_for_king(~Us) & ksq)
+    Bitboard b = attacks_bb<KING>(ksq) & (Type == EVASIONS ? ~pos.pieces(Us) : target);
-    {
-        Bitboard b = attacks_bb<KING>(ksq) & (Type == EVASIONS ? ~pos.pieces(Us) : target);
-        if (Checks)
-            b &= ~attacks_bb<QUEEN>(pos.square<KING>(~Us));
 
-        while (b)
+    while (b)
-            *moveList++ = make_move(ksq, pop_lsb(b));
+        *moveList++ = Move(ksq, pop_lsb(b));
 
-        if ((Type == QUIETS || Type == NON_EVASIONS) && pos.can_castle(Us & ANY_CASTLING))
+    if ((Type == QUIETS || Type == NON_EVASIONS) && pos.can_castle(Us & ANY_CASTLING))
-            for (CastlingRights cr : { Us & KING_SIDE, Us & QUEEN_SIDE } )
+        for (CastlingRights cr : {Us & KING_SIDE, Us & QUEEN_SIDE})
-                if (!pos.castling_impeded(cr) && pos.can_castle(cr))
+            if (!pos.castling_impeded(cr) && pos.can_castle(cr))
-                    *moveList++ = make<CASTLING>(ksq, pos.castling_rook_square(cr));
+                *moveList++ = Move::make<CASTLING>(ksq, pos.castling_rook_square(cr));
-    }
 
     return moveList;
-  }
+}
 
-} // namespace
+}  // namespace
 
 
-/// <CAPTURES>     Generates all pseudo-legal captures plus queen promotions
+// <CAPTURES>     Generates all pseudo-legal captures plus queen promotions
-/// <QUIETS>       Generates all pseudo-legal non-captures and underpromotions
+// <QUIETS>       Generates all pseudo-legal non-captures and underpromotions
-/// <EVASIONS>     Generates all pseudo-legal check evasions when the side to move is in check
+// <EVASIONS>     Generates all pseudo-legal check evasions
-/// <QUIET_CHECKS> Generates all pseudo-legal non-captures giving check, except castling and promotions
+// <NON_EVASIONS> Generates all pseudo-legal captures and non-captures
-/// <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>
 ExtMove* generate(const Position& pos, ExtMove* moveList) {
 
-  static_assert(Type != LEGAL, "Unsupported type in generate()");
+    static_assert(Type != LEGAL, "Unsupported type in generate()");
-  assert((Type == EVASIONS) == (bool)pos.checkers());
+    assert((Type == EVASIONS) == bool(pos.checkers()));
 
-  Color us = pos.side_to_move();
+    Color us = pos.side_to_move();
 
-  return us == WHITE ? generate_all<WHITE, Type>(pos, moveList)
+    return us == WHITE ? generate_all<WHITE, Type>(pos, moveList)
-                     : generate_all<BLACK, Type>(pos, moveList);
+                       : generate_all<BLACK, Type>(pos, moveList);
 }
 
 // Explicit template instantiations
 template ExtMove* generate<CAPTURES>(const Position&, ExtMove*);
 template ExtMove* generate<QUIETS>(const Position&, ExtMove*);
 template ExtMove* generate<EVASIONS>(const Position&, ExtMove*);
-template ExtMove* generate<QUIET_CHECKS>(const Position&, ExtMove*);
 template ExtMove* generate<NON_EVASIONS>(const Position&, ExtMove*);
 
 
-/// generate<LEGAL> generates all the legal moves in the given position
+// generate<LEGAL> generates all the legal moves in the given position
 
 template<>
 ExtMove* generate<LEGAL>(const Position& pos, ExtMove* moveList) {
 
-  Color us = pos.side_to_move();
+    Color    us     = pos.side_to_move();
-  Bitboard pinned = pos.blockers_for_king(us) & pos.pieces(us);
+    Bitboard pinned = pos.blockers_for_king(us) & pos.pieces(us);
-  Square ksq = pos.square<KING>(us);
+    Square   ksq    = pos.square<KING>(us);
-  ExtMove* cur = moveList;
+    ExtMove* cur    = moveList;
 
-  moveList = pos.checkers() ? generate<EVASIONS    >(pos, moveList)
+    moveList =
-                            : generate<NON_EVASIONS>(pos, moveList);
+      pos.checkers() ? generate<EVASIONS>(pos, moveList) : generate<NON_EVASIONS>(pos, moveList);
-  while (cur != moveList)
+    while (cur != moveList)
-      if (  ((pinned & from_sq(*cur)) || from_sq(*cur) == ksq || type_of(*cur) == EN_PASSANT)
+        if (((pinned & cur->from_sq()) || cur->from_sq() == ksq || cur->type_of() == EN_PASSANT)
-          && !pos.legal(*cur))
+            && !pos.legal(*cur))
-          *cur = (--moveList)->move;
+            *cur = *(--moveList);
-      else
+        else
-          ++cur;
+            ++cur;
 
-  return moveList;
+    return moveList;
 }
 
-} // namespace Stockfish
+}  // namespace Stockfish

src/movegen.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,7 +19,8 @@
 #ifndef MOVEGEN_H_INCLUDED
 #define MOVEGEN_H_INCLUDED
 
-#include <algorithm>
+#include <algorithm>  // IWYU pragma: keep
+#include <cstddef>
 
 #include "types.h"
 
@@ -28,50 +29,45 @@ namespace Stockfish {
 class Position;
 
 enum GenType {
-  CAPTURES,
+    CAPTURES,
-  QUIETS,
+    QUIETS,
-  QUIET_CHECKS,
+    EVASIONS,
-  EVASIONS,
+    NON_EVASIONS,
-  NON_EVASIONS,
+    LEGAL
-  LEGAL
 };
 
-struct ExtMove {
+struct ExtMove: public Move {
-  Move move;
+    int value;
-  int value;
 
-  operator Move() const { return move; }
+    void operator=(Move m) { data = m.raw(); }
-  void operator=(Move m) { move = m; }
 
-  // Inhibit unwanted implicit conversions to Move
+    // Inhibit unwanted implicit conversions to Move
-  // with an ambiguity that yields to a compile error.
+    // with an ambiguity that yields to a compile error.
-  operator float() const = delete;
+    operator float() const = delete;
 };
 
-inline bool operator<(const ExtMove& f, const ExtMove& s) {
+inline bool operator<(const ExtMove& f, const ExtMove& s) { return f.value < s.value; }
-  return f.value < s.value;
-}
 
 template<GenType>
 ExtMove* generate(const Position& pos, ExtMove* moveList);
 
-/// The MoveList struct is a simple wrapper around generate(). It sometimes comes
+// The MoveList struct wraps the generate() function and returns a convenient
-/// in handy to use this class instead of the low level generate() function.
+// list of moves. Using MoveList is sometimes preferable to directly calling
+// the lower level generate() function.
 template<GenType T>
 struct MoveList {
 
-  explicit MoveList(const Position& pos) : last(generate<T>(pos, moveList)) {}
+    explicit MoveList(const Position& pos) :
-  const ExtMove* begin() const { return moveList; }
+        last(generate<T>(pos, moveList)) {}
-  const ExtMove* end() const { return last; }
+    const ExtMove* begin() const { return moveList; }
-  size_t size() const { return last - moveList; }
+    const ExtMove* end() const { return last; }
-  bool contains(Move move) const {
+    size_t         size() const { return last - moveList; }
-    return std::find(begin(), end(), move) != end();
+    bool           contains(Move move) const { return std::find(begin(), end(), move) != end(); }
-  }
 
-private:
+   private:
-  ExtMove moveList[MAX_MOVES], *last;
+    ExtMove moveList[MAX_MOVES], *last;
 };
 
-} // namespace Stockfish
+}  // namespace Stockfish
 
-#endif // #ifndef MOVEGEN_H_INCLUDED
+#endif  // #ifndef MOVEGEN_H_INCLUDED

src/movepick.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,280 +16,305 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#include "movepick.h"
+
 #include <cassert>
+#include <limits>
 
 #include "bitboard.h"
-#include "movepick.h"
+#include "misc.h"
+#include "position.h"
 
 namespace Stockfish {
 
 namespace {
 
-  enum Stages {
+enum Stages {
-    MAIN_TT, CAPTURE_INIT, GOOD_CAPTURE, REFUTATION, QUIET_INIT, QUIET, BAD_CAPTURE,
+    // generate main search moves
-    EVASION_TT, EVASION_INIT, EVASION,
+    MAIN_TT,
-    PROBCUT_TT, PROBCUT_INIT, PROBCUT,
+    CAPTURE_INIT,
-    QSEARCH_TT, QCAPTURE_INIT, QCAPTURE, QCHECK_INIT, QCHECK
+    GOOD_CAPTURE,
-  };
+    QUIET_INIT,
+    GOOD_QUIET,
+    BAD_CAPTURE,
+    BAD_QUIET,
 
-  // partial_insertion_sort() sorts moves in descending order up to and including
+    // generate evasion moves
-  // a given limit. The order of moves smaller than the limit is left unspecified.
+    EVASION_TT,
-  void partial_insertion_sort(ExtMove* begin, ExtMove* end, int limit) {
+    EVASION_INIT,
+    EVASION,
+
+    // generate probcut moves
+    PROBCUT_TT,
+    PROBCUT_INIT,
+    PROBCUT,
+
+    // generate qsearch moves
+    QSEARCH_TT,
+    QCAPTURE_INIT,
+    QCAPTURE
+};
+
+// Sort 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) {
 
     for (ExtMove *sortedEnd = begin, *p = begin + 1; p < end; ++p)
         if (p->value >= limit)
         {
             ExtMove tmp = *p, *q;
-            *p = *++sortedEnd;
+            *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
-/// moves to return (in the quiescence search, for instance, we only want to
-/// search captures, promotions, and some checks) and how important good move
-/// ordering is at the current node.
-
-/// 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);
-
-  stage = (pos.checkers() ? EVASION_TT : MAIN_TT) +
-          !(ttm && pos.pseudo_legal(ttm));
 }
 
-/// MovePicker constructor for quiescence search
+}  // namespace
-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);
 
-  stage = (pos.checkers() ? EVASION_TT : QSEARCH_TT) +
+
-          !(   ttm
+// Constructors of the MovePicker class. As arguments, we pass information
-            && pos.pseudo_legal(ttm));
+// to decide which class of moves to emit, to help sorting the (presumably)
+// good moves first, and how important move ordering is at the current node.
+
+// MovePicker constructor for the main search and for the quiescence search
+MovePicker::MovePicker(const Position&              p,
+                       Move                         ttm,
+                       Depth                        d,
+                       const ButterflyHistory*      mh,
+                       const LowPlyHistory*         lph,
+                       const CapturePieceToHistory* cph,
+                       const PieceToHistory**       ch,
+                       const PawnHistory*           ph,
+                       int                          pl) :
+    pos(p),
+    mainHistory(mh),
+    lowPlyHistory(lph),
+    captureHistory(cph),
+    continuationHistory(ch),
+    pawnHistory(ph),
+    ttMove(ttm),
+    depth(d),
+    ply(pl) {
+
+    if (pos.checkers())
+        stage = EVASION_TT + !(ttm && pos.pseudo_legal(ttm));
+
+    else
+        stage = (depth > 0 ? MAIN_TT : QSEARCH_TT) + !(ttm && pos.pseudo_legal(ttm));
 }
 
-/// MovePicker constructor for ProbCut: we generate captures with SEE greater
+// MovePicker constructor for ProbCut: we generate captures with Static Exchange
-/// than or equal to the given threshold.
+// Evaluation (SEE) greater than or equal to the given threshold.
-MovePicker::MovePicker(const Position& p, Move ttm, Value th, const CapturePieceToHistory* cph)
+MovePicker::MovePicker(const Position& p, Move ttm, int th, const CapturePieceToHistory* cph) :
-           : pos(p), captureHistory(cph), ttMove(ttm), threshold(th)
+    pos(p),
-{
+    captureHistory(cph),
-  assert(!pos.checkers());
+    ttMove(ttm),
+    threshold(th) {
+    assert(!pos.checkers());
 
-  stage = PROBCUT_TT + !(ttm && pos.capture_stage(ttm)
+    stage = PROBCUT_TT
-                             && pos.pseudo_legal(ttm)
+          + !(ttm && pos.capture_stage(ttm) && pos.pseudo_legal(ttm) && pos.see_ge(ttm, threshold));
-                             && pos.see_ge(ttm, threshold));
 }
 
-/// MovePicker::score() assigns a numerical value to each move in a list, used
+// Assigns a numerical value to each move in a list, used for sorting.
-/// for sorting. Captures are ordered by Most Valuable Victim (MVV), preferring
+// Captures are ordered by Most Valuable Victim (MVV), preferring captures
-/// captures with a good history. Quiets moves are ordered using the history tables.
+// with a good history. Quiets moves are ordered using the history tables.
 template<GenType Type>
 void MovePicker::score() {
 
-  static_assert(Type == CAPTURES || Type == QUIETS || Type == EVASIONS, "Wrong type");
+    static_assert(Type == CAPTURES || Type == QUIETS || Type == EVASIONS, "Wrong type");
 
-  [[maybe_unused]] Bitboard threatenedByPawn, threatenedByMinor, threatenedByRook, threatenedPieces;
+    [[maybe_unused]] Bitboard threatenedByPawn, threatenedByMinor, threatenedByRook,
-  if constexpr (Type == QUIETS)
+      threatenedPieces;
-  {
+    if constexpr (Type == QUIETS)
-      Color us = pos.side_to_move();
+    {
+        Color us = pos.side_to_move();
 
-      threatenedByPawn  = pos.attacks_by<PAWN>(~us);
+        threatenedByPawn = pos.attacks_by<PAWN>(~us);
-      threatenedByMinor = pos.attacks_by<KNIGHT>(~us) | pos.attacks_by<BISHOP>(~us) | threatenedByPawn;
+        threatenedByMinor =
-      threatenedByRook  = pos.attacks_by<ROOK>(~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
+        // Pieces threatened by pieces of lesser material value
-      threatenedPieces = (pos.pieces(us, QUEEN) & threatenedByRook)
+        threatenedPieces = (pos.pieces(us, QUEEN) & threatenedByRook)
-                       | (pos.pieces(us, ROOK)  & threatenedByMinor)
+                         | (pos.pieces(us, ROOK) & threatenedByMinor)
-                       | (pos.pieces(us, KNIGHT, BISHOP) & threatenedByPawn);
+                         | (pos.pieces(us, KNIGHT, BISHOP) & threatenedByPawn);
-  }
+    }
 
-  for (auto& m : *this)
+    for (auto& m : *this)
-      if constexpr (Type == CAPTURES)
+        if constexpr (Type == CAPTURES)
-          m.value =  (7 * int(PieceValue[MG][pos.piece_on(to_sq(m))])
+            m.value =
-                   +     (*captureHistory)[pos.moved_piece(m)][to_sq(m)][type_of(pos.piece_on(to_sq(m)))]) / 16;
+              7 * int(PieceValue[pos.piece_on(m.to_sq())])
+              + (*captureHistory)[pos.moved_piece(m)][m.to_sq()][type_of(pos.piece_on(m.to_sq()))];
 
-      else if constexpr (Type == QUIETS)
+        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)]
+            Piece     pc   = pos.moved_piece(m);
-                   +     (*continuationHistory[1])[pos.moved_piece(m)][to_sq(m)]
+            PieceType pt   = type_of(pc);
-                   +     (*continuationHistory[3])[pos.moved_piece(m)][to_sq(m)]
+            Square    from = m.from_sq();
-                   +     (*continuationHistory[5])[pos.moved_piece(m)][to_sq(m)]
+            Square    to   = m.to_sq();
-                   +     (threatenedPieces & from_sq(m) ?
+
-                           (type_of(pos.moved_piece(m)) == QUEEN && !(to_sq(m) & threatenedByRook)  ? 50000
+            // histories
-                          : type_of(pos.moved_piece(m)) == ROOK  && !(to_sq(m) & threatenedByMinor) ? 25000
+            m.value = 2 * (*mainHistory)[pos.side_to_move()][m.from_to()];
-                          :                                         !(to_sq(m) & threatenedByPawn)  ? 15000
+            m.value += 2 * (*pawnHistory)[pawn_structure_index(pos)][pc][to];
-                          :                                                                           0)
+            m.value += (*continuationHistory[0])[pc][to];
-                          :                                                                           0)
+            m.value += (*continuationHistory[1])[pc][to];
-                   +     bool(pos.check_squares(type_of(pos.moved_piece(m))) & to_sq(m)) * 16384;
+            m.value += (*continuationHistory[2])[pc][to];
-      else // Type == EVASIONS
+            m.value += (*continuationHistory[3])[pc][to];
-      {
+            m.value += (*continuationHistory[4])[pc][to] / 3;
-          if (pos.capture_stage(m))
+            m.value += (*continuationHistory[5])[pc][to];
-              m.value =  PieceValue[MG][pos.piece_on(to_sq(m))]
+
-                       - Value(type_of(pos.moved_piece(m)))
+            // bonus for checks
-                       + (1 << 28);
+            m.value += bool(pos.check_squares(pt) & to) * 16384;
-          else
+
-              m.value =  (*mainHistory)[pos.side_to_move()][from_to(m)]
+            // bonus for escaping from capture
-                       + (*continuationHistory[0])[pos.moved_piece(m)][to_sq(m)];
+            m.value += threatenedPieces & from ? (pt == QUEEN && !(to & threatenedByRook)   ? 51700
-      }
+                                                  : pt == ROOK && !(to & threatenedByMinor) ? 25600
+                                                  : !(to & threatenedByPawn)                ? 14450
+                                                                                            : 0)
+                                               : 0;
+
+            // malus for putting piece en prise
+            m.value -= (pt == QUEEN ? bool(to & threatenedByRook) * 49000
+                        : pt == ROOK && bool(to & threatenedByMinor) ? 24335
+                                                                     : 0);
+
+            if (ply < LOW_PLY_HISTORY_SIZE)
+                m.value += 8 * (*lowPlyHistory)[ply][m.from_to()] / (1 + 2 * ply);
+        }
+
+        else  // Type == EVASIONS
+        {
+            if (pos.capture_stage(m))
+                m.value = PieceValue[pos.piece_on(m.to_sq())] + (1 << 28);
+            else
+                m.value = (*mainHistory)[pos.side_to_move()][m.from_to()]
+                        + (*continuationHistory[0])[pos.moved_piece(m)][m.to_sq()]
+                        + (*pawnHistory)[pawn_structure_index(pos)][pos.moved_piece(m)][m.to_sq()];
+        }
 }
 
-/// MovePicker::select() returns the next move satisfying a predicate function.
+// Returns the next move satisfying a predicate function.
-/// It never returns the TT move.
+// This never returns the TT move, as it was emitted before.
-template<MovePicker::PickType T, typename Pred>
+template<typename Pred>
 Move MovePicker::select(Pred filter) {
 
-  while (cur < endMoves)
+    for (; cur < endMoves; ++cur)
-  {
+        if (*cur != ttMove && filter())
-      if constexpr (T == Best)
+            return *cur++;
-          std::swap(*cur, *std::max_element(cur, endMoves));
 
-      if (*cur != ttMove && filter())
+    return Move::none();
-          return *cur++;
-
-      cur++;
-  }
-  return MOVE_NONE;
 }
 
-/// MovePicker::next_move() is the most important method of the MovePicker class. It
+// This is the most important method of the MovePicker class. We emit one
-/// returns a new pseudo-legal move every time it is called until there are no more
+// new pseudo-legal move on every call until there are no more moves left,
-/// moves left, picking the move with the highest score from a list of generated moves.
+// picking the move with the highest score from a list of generated moves.
-Move MovePicker::next_move(bool skipQuiets) {
+Move MovePicker::next_move() {
+
+    auto quiet_threshold = [](Depth d) { return -3560 * d; };
 
 top:
-  switch (stage) {
+    switch (stage)
+    {
 
-  case MAIN_TT:
+    case MAIN_TT :
-  case EVASION_TT:
+    case EVASION_TT :
-  case QSEARCH_TT:
+    case QSEARCH_TT :
-  case PROBCUT_TT:
+    case PROBCUT_TT :
-      ++stage;
+        ++stage;
-      return ttMove;
+        return ttMove;
 
-  case CAPTURE_INIT:
+    case CAPTURE_INIT :
-  case PROBCUT_INIT:
+    case PROBCUT_INIT :
-  case QCAPTURE_INIT:
+    case QCAPTURE_INIT :
-      cur = endBadCaptures = moves;
+        cur = endBadCaptures = moves;
-      endMoves = generate<CAPTURES>(pos, cur);
+        endMoves             = generate<CAPTURES>(pos, cur);
 
-      score<CAPTURES>();
+        score<CAPTURES>();
-      partial_insertion_sort(cur, endMoves, std::numeric_limits<int>::min());
+        partial_insertion_sort(cur, endMoves, std::numeric_limits<int>::min());
-      ++stage;
+        ++stage;
-      goto top;
+        goto top;
 
-  case GOOD_CAPTURE:
+    case GOOD_CAPTURE :
-      if (select<Next>([&](){
+        if (select([&]() {
-                       return pos.see_ge(*cur, Value(-cur->value)) ?
+                // Move losing capture to endBadCaptures to be tried later
-                              // Move losing capture to endBadCaptures to be tried later
+                return pos.see_ge(*cur, -cur->value / 18) ? true
-                              true : (*endBadCaptures++ = *cur, false); }))
+                                                          : (*endBadCaptures++ = *cur, false);
-          return *(cur - 1);
+            }))
+            return *(cur - 1);
 
-      // Prepare the pointers to loop over the refutations array
+        ++stage;
-      cur = std::begin(refutations);
+        [[fallthrough]];
-      endMoves = std::end(refutations);
 
-      // If the countermove is the same as a killer, skip it
+    case QUIET_INIT :
-      if (   refutations[0].move == refutations[2].move
+        if (!skipQuiets)
-          || refutations[1].move == refutations[2].move)
+        {
-          --endMoves;
+            cur      = endBadCaptures;
+            endMoves = beginBadQuiets = endBadQuiets = generate<QUIETS>(pos, cur);
 
-      ++stage;
+            score<QUIETS>();
-      [[fallthrough]];
+            partial_insertion_sort(cur, endMoves, quiet_threshold(depth));
+        }
 
-  case REFUTATION:
+        ++stage;
-      if (select<Next>([&](){ return    *cur != MOVE_NONE
+        [[fallthrough]];
-                                    && !pos.capture_stage(*cur)
-                                    &&  pos.pseudo_legal(*cur); }))
-          return *(cur - 1);
-      ++stage;
-      [[fallthrough]];
 
-  case QUIET_INIT:
+    case GOOD_QUIET :
-      if (!skipQuiets)
+        if (!skipQuiets && select([]() { return true; }))
-      {
+        {
-          cur = endBadCaptures;
+            if ((cur - 1)->value > -7998 || (cur - 1)->value <= quiet_threshold(depth))
-          endMoves = generate<QUIETS>(pos, cur);
+                return *(cur - 1);
 
-          score<QUIETS>();
+            // Remaining quiets are bad
-          partial_insertion_sort(cur, endMoves, -3000 * depth);
+            beginBadQuiets = cur - 1;
-      }
+        }
 
-      ++stage;
+        // Prepare the pointers to loop over the bad captures
-      [[fallthrough]];
+        cur      = moves;
+        endMoves = endBadCaptures;
 
-  case QUIET:
+        ++stage;
-      if (   !skipQuiets
+        [[fallthrough]];
-          && select<Next>([&](){return   *cur != refutations[0].move
-                                      && *cur != refutations[1].move
-                                      && *cur != refutations[2].move;}))
-          return *(cur - 1);
 
-      // Prepare the pointers to loop over the bad captures
+    case BAD_CAPTURE :
-      cur = moves;
+        if (select([]() { return true; }))
-      endMoves = endBadCaptures;
+            return *(cur - 1);
 
-      ++stage;
+        // Prepare the pointers to loop over the bad quiets
-      [[fallthrough]];
+        cur      = beginBadQuiets;
+        endMoves = endBadQuiets;
 
-  case BAD_CAPTURE:
+        ++stage;
-      return select<Next>([](){ return true; });
+        [[fallthrough]];
 
-  case EVASION_INIT:
+    case BAD_QUIET :
-      cur = moves;
+        if (!skipQuiets)
-      endMoves = generate<EVASIONS>(pos, cur);
+            return select([]() { return true; });
 
-      score<EVASIONS>();
+        return Move::none();
-      ++stage;
-      [[fallthrough]];
 
-  case EVASION:
+    case EVASION_INIT :
-      return select<Best>([](){ return true; });
+        cur      = moves;
+        endMoves = generate<EVASIONS>(pos, cur);
 
-  case PROBCUT:
+        score<EVASIONS>();
-      return select<Next>([&](){ return pos.see_ge(*cur, threshold); });
+        partial_insertion_sort(cur, endMoves, std::numeric_limits<int>::min());
+        ++stage;
+        [[fallthrough]];
 
-  case QCAPTURE:
+    case EVASION :
-      if (select<Next>([&](){ return   depth > DEPTH_QS_RECAPTURES
+    case QCAPTURE :
-                                    || to_sq(*cur) == recaptureSquare; }))
+        return select([]() { return true; });
-          return *(cur - 1);
 
-      // If we did not find any move and we do not try checks, we have finished
+    case PROBCUT :
-      if (depth != DEPTH_QS_CHECKS)
+        return select([&]() { return pos.see_ge(*cur, threshold); });
-          return MOVE_NONE;
+    }
 
-      ++stage;
+    assert(false);
-      [[fallthrough]];
+    return Move::none();  // Silence warning
-
-  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
 }
 
-} // namespace Stockfish
+void MovePicker::skip_quiet_moves() { skipQuiets = true; }
+
+}  // namespace Stockfish

src/movepick.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,137 +19,62 @@
 #ifndef MOVEPICK_H_INCLUDED
 #define MOVEPICK_H_INCLUDED
 
-#include <array>
+#include "history.h"
-#include <limits>
-#include <type_traits>
-
 #include "movegen.h"
-#include "position.h"
 #include "types.h"
 
 namespace Stockfish {
 
-/// StatsEntry stores the stat table value. It is usually a number but could
+class Position;
-/// 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 {
 
-  T entry;
+// The MovePicker class is used to pick one pseudo-legal move at a time from the
-
+// current position. The most important method is next_move(), which emits one
-public:
+// new pseudo-legal move on every call, until there are no moves left, when
-  void operator=(const T& v) { entry = v; }
+// Move::none() is returned. In order to improve the efficiency of the alpha-beta
-  T* operator&() { return &entry; }
+// algorithm, MovePicker attempts to return the moves which are most likely to get
-  T* operator->() { return &entry; }
+// a cut-off first.
-  operator const T&() const { return entry; }
-
-  void operator<<(int bonus) {
-    assert(abs(bonus) <= D); // Ensure range is [-D, D]
-    static_assert(D <= std::numeric_limits<T>::max(), "D overflows T");
-
-    entry += bonus - entry * abs(bonus) / D;
-
-    assert(abs(entry) <= D);
-  }
-};
-
-/// 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 LowPlyHistory*,
+               const CapturePieceToHistory*,
+               const PieceToHistory**,
+               const PawnHistory*,
+               int);
+    MovePicker(const Position&, Move, int, const CapturePieceToHistory*);
+    Move next_move();
+    void skip_quiet_moves();
 
-public:
+   private:
-  MovePicker(const MovePicker&) = delete;
+    template<typename Pred>
-  MovePicker& operator=(const MovePicker&) = delete;
+    Move select(Pred);
-  MovePicker(const Position&, Move, Depth, const ButterflyHistory*,
+    template<GenType>
-                                           const CapturePieceToHistory*,
+    void     score();
-                                           const PieceToHistory**,
+    ExtMove* begin() { return cur; }
-                                           Move,
+    ExtMove* end() { return endMoves; }
-                                           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:
+    const Position&              pos;
-  template<PickType T, typename Pred> Move select(Pred);
+    const ButterflyHistory*      mainHistory;
-  template<GenType> void score();
+    const LowPlyHistory*         lowPlyHistory;
-  ExtMove* begin() { return cur; }
+    const CapturePieceToHistory* captureHistory;
-  ExtMove* end() { return endMoves; }
+    const PieceToHistory**       continuationHistory;
-
+    const PawnHistory*           pawnHistory;
-  const Position& pos;
+    Move                         ttMove;
-  const ButterflyHistory* mainHistory;
+    ExtMove *                    cur, *endMoves, *endBadCaptures, *beginBadQuiets, *endBadQuiets;
-  const CapturePieceToHistory* captureHistory;
+    int                          stage;
-  const PieceToHistory** continuationHistory;
+    int                          threshold;
-  Move ttMove;
+    Depth                        depth;
-  ExtMove refutations[3], *cur, *endMoves, *endBadCaptures;
+    int                          ply;
-  int stage;
+    bool                         skipQuiets = false;
-  Square recaptureSquare;
+    ExtMove                      moves[MAX_MOVES];
-  Value threshold;
-  Depth depth;
-  ExtMove moves[MAX_MOVES];
 };
 
-} // namespace Stockfish
+}  // namespace Stockfish
 
-#endif // #ifndef MOVEPICK_H_INCLUDED
+#endif  // #ifndef MOVEPICK_H_INCLUDED

src/nnue/evaluate_nnue.cpp

@@ -1,405 +0,0 @@
-/*
-  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

@@ -1,68 +0,0 @@
-/*
-  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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -20,65 +20,69 @@
 
 #include "half_ka_v2_hm.h"
 
+#include "../../bitboard.h"
 #include "../../position.h"
+#include "../../types.h"
+#include "../nnue_accumulator.h"
 
 namespace Stockfish::Eval::NNUE::Features {
 
-  // Index of a feature for a given king position and another piece on some square
+// Index of a feature for a given king position and another piece on some square
-  template<Color Perspective>
+template<Color Perspective>
-  inline IndexType HalfKAv2_hm::make_index(Square s, Piece pc, Square ksq) {
+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]);
+    return IndexType((int(s) ^ OrientTBL[Perspective][ksq]) + PieceSquareIndex[Perspective][pc]
-  }
+                     + KingBuckets[Perspective][ksq]);
+}
 
-  // Get a list of indices for active features
+// Get a list of indices for active features
-  template<Color Perspective>
+template<Color Perspective>
-  void HalfKAv2_hm::append_active_indices(
+void HalfKAv2_hm::append_active_indices(const Position& pos, IndexList& active) {
-    const Position& pos,
+    Square   ksq = pos.square<KING>(Perspective);
-    IndexList& active
+    Bitboard bb  = pos.pieces();
-  ) {
-    Square ksq = pos.square<KING>(Perspective);
-    Bitboard bb = pos.pieces();
     while (bb)
     {
-      Square s = pop_lsb(bb);
+        Square s = pop_lsb(bb);
-      active.push_back(make_index<Perspective>(s, pos.piece_on(s), ksq));
+        active.push_back(make_index<Perspective>(s, pos.piece_on(s), ksq));
     }
-  }
+}
 
-  // Explicit template instantiations
+// Explicit template instantiations
-  template void HalfKAv2_hm::append_active_indices<WHITE>(const Position& pos, IndexList& active);
+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);
+template void HalfKAv2_hm::append_active_indices<BLACK>(const Position& pos, IndexList& active);
+template IndexType HalfKAv2_hm::make_index<WHITE>(Square s, Piece pc, Square ksq);
+template IndexType HalfKAv2_hm::make_index<BLACK>(Square s, Piece pc, Square ksq);
 
-  // append_changed_indices() : get a list of indices for recently changed features
+// Get a list of indices for recently changed features
-  template<Color Perspective>
+template<Color Perspective>
-  void HalfKAv2_hm::append_changed_indices(
+void HalfKAv2_hm::append_changed_indices(Square            ksq,
-    Square ksq,
+                                         const DirtyPiece& dp,
-    const DirtyPiece& dp,
+                                         IndexList&        removed,
-    IndexList& removed,
+                                         IndexList&        added) {
-    IndexList& added
+    for (int i = 0; i < dp.dirty_num; ++i)
-  ) {
+    {
-    for (int i = 0; i < dp.dirty_num; ++i) {
+        if (dp.from[i] != SQ_NONE)
-      if (dp.from[i] != SQ_NONE)
+            removed.push_back(make_index<Perspective>(dp.from[i], dp.piece[i], ksq));
-        removed.push_back(make_index<Perspective>(dp.from[i], dp.piece[i], ksq));
+        if (dp.to[i] != SQ_NONE)
-      if (dp.to[i] != SQ_NONE)
+            added.push_back(make_index<Perspective>(dp.to[i], dp.piece[i], ksq));
-        added.push_back(make_index<Perspective>(dp.to[i], dp.piece[i], ksq));
     }
-  }
+}
 
-  // Explicit template instantiations
+// 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<WHITE>(Square            ksq,
-  template void HalfKAv2_hm::append_changed_indices<BLACK>(Square ksq, const DirtyPiece& dp, IndexList& removed, IndexList& added);
+                                                         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) {
+int HalfKAv2_hm::update_cost(const StateInfo* st) { return st->dirtyPiece.dirty_num; }
-    return st->dirtyPiece.dirty_num;
-  }
 
-  int HalfKAv2_hm::refresh_cost(const Position& pos) {
+int HalfKAv2_hm::refresh_cost(const Position& pos) { return pos.count<ALL_PIECES>(); }
-    return pos.count<ALL_PIECES>();
-  }
 
-  bool HalfKAv2_hm::requires_refresh(const StateInfo* st, Color perspective) {
+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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,50 +21,47 @@
 #ifndef NNUE_FEATURES_HALF_KA_V2_HM_H_INCLUDED
 #define NNUE_FEATURES_HALF_KA_V2_HM_H_INCLUDED
 
+#include <cstdint>
+
+#include "../../misc.h"
+#include "../../types.h"
 #include "../nnue_common.h"
 
-#include "../../evaluate.h"
-#include "../../misc.h"
-
 namespace Stockfish {
-  struct StateInfo;
+struct StateInfo;
+class Position;
 }
 
 namespace Stockfish::Eval::NNUE::Features {
 
-  // Feature HalfKAv2_hm: Combination of the position of own king
+// Feature HalfKAv2_hm: Combination of the position of own king and the
-  // and the position of pieces. Position mirrored such that king always on e..h files.
+// position of pieces. Position mirrored such that king is always on e..h files.
-  class HalfKAv2_hm {
+class HalfKAv2_hm {
 
-    // unique number for each piece type on each square
+    // Unique number for each piece type on each square
     enum {
-      PS_NONE     =  0,
+        PS_NONE     = 0,
-      PS_W_PAWN   =  0,
+        PS_W_PAWN   = 0,
-      PS_B_PAWN   =  1 * SQUARE_NB,
+        PS_B_PAWN   = 1 * SQUARE_NB,
-      PS_W_KNIGHT =  2 * SQUARE_NB,
+        PS_W_KNIGHT = 2 * SQUARE_NB,
-      PS_B_KNIGHT =  3 * SQUARE_NB,
+        PS_B_KNIGHT = 3 * SQUARE_NB,
-      PS_W_BISHOP =  4 * SQUARE_NB,
+        PS_W_BISHOP = 4 * SQUARE_NB,
-      PS_B_BISHOP =  5 * SQUARE_NB,
+        PS_B_BISHOP = 5 * SQUARE_NB,
-      PS_W_ROOK   =  6 * SQUARE_NB,
+        PS_W_ROOK   = 6 * SQUARE_NB,
-      PS_B_ROOK   =  7 * SQUARE_NB,
+        PS_B_ROOK   = 7 * SQUARE_NB,
-      PS_W_QUEEN  =  8 * SQUARE_NB,
+        PS_W_QUEEN  = 8 * SQUARE_NB,
-      PS_B_QUEEN  =  9 * SQUARE_NB,
+        PS_B_QUEEN  = 9 * SQUARE_NB,
-      PS_KING     = 10 * SQUARE_NB,
+        PS_KING     = 10 * SQUARE_NB,
-      PS_NB       = 11 * SQUARE_NB
+        PS_NB       = 11 * SQUARE_NB
     };
 
     static constexpr IndexType PieceSquareIndex[COLOR_NB][PIECE_NB] = {
-      // convention: W - us, B - them
+      // Convention: W - us, B - them
-      // viewed from other side, W and B are reversed
+      // 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_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_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 }
+       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
@@ -75,9 +72,10 @@ namespace Stockfish::Eval::NNUE::Features {
 
     // Number of feature dimensions
     static constexpr IndexType Dimensions =
-        static_cast<IndexType>(SQUARE_NB) * static_cast<IndexType>(PS_NB) / 2;
+      static_cast<IndexType>(SQUARE_NB) * static_cast<IndexType>(PS_NB) / 2;
 
 #define B(v) (v * PS_NB)
+    // clang-format off
     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),
@@ -96,8 +94,9 @@ namespace Stockfish::Eval::NNUE::Features {
         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) }
     };
+    // clang-format on
 #undef B
-
+    // clang-format off
     // 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,
@@ -117,36 +116,35 @@ namespace Stockfish::Eval::NNUE::Features {
         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 }
     };
+    // clang-format on
 
     // Maximum number of simultaneously active features.
     static constexpr IndexType MaxActiveDimensions = 32;
-    using IndexList = ValueList<IndexType, MaxActiveDimensions>;
+    using IndexList                                = ValueList<IndexType, MaxActiveDimensions>;
+
+    // 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);
 
     // Get a list of indices for active features
     template<Color Perspective>
-    static void append_active_indices(
+    static void append_active_indices(const Position& pos, IndexList& active);
-      const Position& pos,
-      IndexList& active);
 
     // Get a list of indices for recently changed features
     template<Color Perspective>
-    static void append_changed_indices(
+    static void
-      Square ksq,
+    append_changed_indices(Square ksq, const DirtyPiece& dp, IndexList& removed, IndexList& added);
-      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
+    // Returns whether the change stored in this StateInfo means
-    // a full accumulator refresh is required.
+    // 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
+#endif  // #ifndef NNUE_FEATURES_HALF_KA_V2_HM_H_INCLUDED

src/nnue/layers/affine_transform.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,163 +21,117 @@
 #ifndef NNUE_LAYERS_AFFINE_TRANSFORM_H_INCLUDED
 #define NNUE_LAYERS_AFFINE_TRANSFORM_H_INCLUDED
 
+#include <cstdint>
 #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 {
 
+#if defined(USE_SSSE3) || defined(USE_NEON_DOTPROD)
+    #define ENABLE_SEQ_OPT
+#endif
+
 // Fallback implementation for older/other architectures.
-// Identical for both approaches. Requires the input to be padded to at least 16 values.
+// Requires the input to be padded to at least 16 values.
-#if !defined(USE_SSSE3)
+#ifndef ENABLE_SEQ_OPT
-  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)
+template<IndexType InputDimensions, IndexType PaddedInputDimensions, IndexType OutputDimensions>
-  {
+static void affine_transform_non_ssse3(std::int32_t*       output,
-# if defined(USE_SSE2)
+                                       const std::int8_t*  weights,
+                                       const std::int32_t* biases,
+                                       const std::uint8_t* input) {
+    #if defined(USE_SSE2) || defined(USE_NEON)
+        #if defined(USE_SSE2)
     // At least a multiple of 16, with SSE2.
-    constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
+    constexpr IndexType NumChunks   = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
-    const __m128i Zeros = _mm_setzero_si128();
+    const __m128i       Zeros       = _mm_setzero_si128();
-    const auto inputVector = reinterpret_cast<const __m128i*>(input);
+    const auto          inputVector = reinterpret_cast<const __m128i*>(input);
 
-# elif defined(USE_MMX)
+        #elif defined(USE_NEON)
-    constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 8) / 8;
+    constexpr IndexType NumChunks   = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
-    const __m64 Zeros = _mm_setzero_si64();
+    const auto          inputVector = reinterpret_cast<const int8x8_t*>(input);
-    const auto inputVector = reinterpret_cast<const __m64*>(input);
+        #endif
 
-# elif defined(USE_NEON_DOTPROD)
+    for (IndexType i = 0; i < OutputDimensions; ++i)
-    constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
+    {
-    const auto inputVector = reinterpret_cast<const int8x16_t*>(input);
+        const IndexType offset = i * PaddedInputDimensions;
 
-# elif defined(USE_NEON)
+        #if defined(USE_SSE2)
-    constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
+        __m128i    sumLo = _mm_cvtsi32_si128(biases[i]);
-    const auto inputVector = reinterpret_cast<const int8x8_t*>(input);
+        __m128i    sumHi = Zeros;
-# endif
+        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);
 
-    for (IndexType i = 0; i < OutputDimensions; ++i) {
+        #elif defined(USE_NEON)
-      const IndexType offset = i * PaddedInputDimensions;
 
-# if defined(USE_SSE2)
+        int32x4_t  sum = {biases[i]};
-      __m128i sumLo = _mm_cvtsi32_si128(biases[i]);
+        const auto row = reinterpret_cast<const int8x8_t*>(&weights[offset]);
-      __m128i sumHi = Zeros;
+        for (IndexType j = 0; j < NumChunks; ++j)
-      const auto row = reinterpret_cast<const __m128i*>(&weights[offset]);
+        {
-      for (IndexType j = 0; j < NumChunks; ++j) {
+            int16x8_t product = vmull_s8(inputVector[j * 2], row[j * 2]);
-        __m128i row_j = _mm_load_si128(&row[j]);
+            product           = vmlal_s8(product, inputVector[j * 2 + 1], row[j * 2 + 1]);
-        __m128i input_j = _mm_load_si128(&inputVector[j]);
+            sum               = vpadalq_s16(sum, product);
-        __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);
+        output[i] = sum[0] + sum[1] + sum[2] + sum[3];
-        __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)
+        #endif
-      __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
     }
+    #else
+    std::memcpy(output, biases, sizeof(std::int32_t) * OutputDimensions);
 
-# if defined(USE_MMX)
+    // Traverse weights in transpose order to take advantage of input sparsity
-    _mm_empty();
+    for (IndexType i = 0; i < InputDimensions; ++i)
-# endif
+        if (input[i])
-  }
+        {
-#endif
+            const std::int8_t* w  = &weights[i];
+            const int          in = input[i];
+            for (IndexType j = 0; j < OutputDimensions; ++j)
+                output[j] += w[j * PaddedInputDimensions] * in;
+        }
+    #endif
+}
 
-  template <IndexType InDims, IndexType OutDims, typename Enabled = void>
+#endif  // !ENABLE_SEQ_OPT
-  class AffineTransform;
 
-#if defined (USE_AVX512)
+template<IndexType InDims, IndexType OutDims>
-  constexpr IndexType LargeInputSize = 2 * 64;
+class AffineTransform {
-#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 InputType  = std::uint8_t;
     using OutputType = std::int32_t;
 
     // Number of input/output dimensions
-    static constexpr IndexType InputDimensions = InDims;
+    static constexpr IndexType InputDimensions  = InDims;
     static constexpr IndexType OutputDimensions = OutDims;
 
     static constexpr IndexType PaddedInputDimensions =
@@ -187,377 +141,166 @@ namespace Stockfish::Eval::NNUE::Layers {
 
     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;
+        std::uint32_t hashValue = 0xCC03DAE4u;
-      hashValue += OutputDimensions;
+        hashValue += OutputDimensions;
-      hashValue ^= prevHash >> 1;
+        hashValue ^= prevHash >> 1;
-      hashValue ^= prevHash << 31;
+        hashValue ^= prevHash << 31;
-      return hashValue;
+        return hashValue;
     }
 
-    /*
+    static constexpr IndexType get_weight_index_scrambled(IndexType i) {
-      Transposes the small blocks within a block.
+        return (i / 4) % (PaddedInputDimensions / 4) * OutputDimensions * 4
-      Effectively means that weights can be traversed sequentially during inference.
+             + i / PaddedInputDimensions * 4 + i % 4;
-    */
-    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
+    static constexpr IndexType get_weight_index(IndexType i) {
-    bool read_parameters(std::istream& stream) {
+#ifdef ENABLE_SEQ_OPT
-      read_little_endian<BiasType>(stream, biases, OutputDimensions);
+        return get_weight_index_scrambled(i);
-
-      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.
+        return i;
-      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);
+        read_little_endian<BiasType>(stream, biases, OutputDimensions);
-      for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+        for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
-        weights[get_weight_index(i)] = read_little_endian<WeightType>(stream);
+            weights[get_weight_index(i)] = read_little_endian<WeightType>(stream);
 
-      return !stream.fail();
+        return !stream.fail();
     }
 
     // Write network parameters
     bool write_parameters(std::ostream& stream) const {
-      write_little_endian<BiasType>(stream, biases, OutputDimensions);
+        write_little_endian<BiasType>(stream, biases, OutputDimensions);
 
-      for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+        for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
-        write_little_endian<WeightType>(stream, weights[get_weight_index(i)]);
+            write_little_endian<WeightType>(stream, weights[get_weight_index(i)]);
 
-      return !stream.fail();
+        return !stream.fail();
     }
     // Forward propagation
-    const OutputType* propagate(
+    void propagate(const InputType* input, OutputType* output) const {
-        const InputType* input, OutputType* output) const {
 
-#if defined (USE_AVX512)
+#ifdef ENABLE_SEQ_OPT
-      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)
+        if constexpr (OutputDimensions > 1)
-      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]);
+    #if defined(USE_AVX512)
-          const vec_t in1 = vec_set_32(input32[i + 1]);
+            using vec_t = __m512i;
-          const auto col0 = reinterpret_cast<const vec_t*>(&weights[(i + 0) * OutputDimensions * 4]);
+        #define vec_set_32 _mm512_set1_epi32
-          const auto col1 = reinterpret_cast<const vec_t*>(&weights[(i + 1) * OutputDimensions * 4]);
+        #define vec_add_dpbusd_32 Simd::m512_add_dpbusd_epi32
-          for (IndexType k = 0; k < NumRegs; ++k)
+    #elif defined(USE_AVX2)
-            vec_add_dpbusd_32x2(acc[k], in0, col0[k], in1, col1[k]);
+            using vec_t = __m256i;
+        #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_set_32 _mm_set1_epi32
+        #define vec_add_dpbusd_32 Simd::m128_add_dpbusd_epi32
+    #elif defined(USE_NEON_DOTPROD)
+            using vec_t = int32x4_t;
+        #define vec_set_32 vdupq_n_s32
+        #define vec_add_dpbusd_32(acc, a, b) \
+            Simd::dotprod_m128_add_dpbusd_epi32(acc, vreinterpretq_s8_s32(a), \
+                                                vreinterpretq_s8_s32(b))
+    #endif
+
+            static constexpr IndexType OutputSimdWidth = sizeof(vec_t) / sizeof(OutputType);
+
+            static_assert(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)
+            {
+                const vec_t in0 = vec_set_32(input32[i]);
+                const auto  col0 =
+                  reinterpret_cast<const vec_t*>(&weights[i * OutputDimensions * 4]);
+
+                for (IndexType k = 0; k < NumRegs; ++k)
+                    vec_add_dpbusd_32(acc[k], in0, col0[k]);
+            }
+
+            vec_t* outptr = reinterpret_cast<vec_t*>(output);
+            for (IndexType k = 0; k < NumRegs; ++k)
+                outptr[k] = acc[k];
+
+    #undef vec_set_32
+    #undef vec_add_dpbusd_32
         }
-
+        else if constexpr (OutputDimensions == 1)
-        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];
+    // We cannot use AVX512 for the last layer because there are only 32 inputs
-          vec_add_dpbusd_32(sum0, in, row0[j]);
+    // and the buffer is not padded to 64 elements.
-        }
+    #if defined(USE_AVX2)
-        output[0] = vec_hadd(sum0, biases[0]);
+            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_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_hadd Simd::m128_hadd
+    #elif defined(USE_NEON_DOTPROD)
+            using vec_t = int32x4_t;
+        #define vec_setzero() vdupq_n_s32(0)
+        #define vec_set_32 vdupq_n_s32
+        #define vec_add_dpbusd_32(acc, a, b) \
+            Simd::dotprod_m128_add_dpbusd_epi32(acc, vreinterpretq_s8_s32(a), \
+                                                vreinterpretq_s8_s32(b))
+        #define vec_hadd Simd::neon_m128_hadd
+    #endif
 
-# undef vec_setzero
+            const auto inputVector = reinterpret_cast<const vec_t*>(input);
-# undef vec_set_32
+
-# undef vec_add_dpbusd_32
+            static constexpr IndexType InputSimdWidth = sizeof(vec_t) / sizeof(InputType);
-# undef vec_add_dpbusd_32x2
+
-# undef vec_hadd
+            static_assert(PaddedInputDimensions % InputSimdWidth == 0);
+
+            constexpr IndexType NumChunks = PaddedInputDimensions / InputSimdWidth;
+            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_hadd
+        }
 #else
-      // Use old implementation for the other architectures.
+        // Use old implementation for the other architectures.
-      affine_transform_non_ssse3<
+        affine_transform_non_ssse3<InputDimensions, PaddedInputDimensions, OutputDimensions>(
-        InputDimensions,
+          output, weights, biases, input);
-        PaddedInputDimensions,
-        OutputDimensions>(output, weights, biases, input);
 #endif
-
-      return output;
     }
 
    private:
-    using BiasType = OutputType;
+    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
+#endif  // #ifndef NNUE_LAYERS_AFFINE_TRANSFORM_H_INCLUDED

src/nnue/layers/affine_transform_sparse_input.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,10 +21,12 @@
 #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 <cstdint>
+#include <iostream>
+
+#include "../../bitboard.h"
 #include "../nnue_common.h"
 #include "affine_transform.h"
 #include "simd.h"
@@ -34,130 +36,147 @@
 */
 
 namespace Stockfish::Eval::NNUE::Layers {
-#if defined(__GNUC__)  // GCC, Clang, ICC
 
-  static inline IndexType lsb_(std::uint32_t b) {
+#if (USE_SSSE3 | (USE_NEON >= 8))
-    assert(b);
+static constexpr int lsb_index64[64] = {
-    return IndexType(__builtin_ctzl(b));
+  0,  47, 1,  56, 48, 27, 2,  60, 57, 49, 41, 37, 28, 16, 3,  61, 54, 58, 35, 52, 50, 42,
-  }
+  21, 44, 38, 32, 29, 23, 17, 11, 4,  62, 46, 55, 26, 59, 40, 36, 15, 53, 34, 51, 20, 43,
+  31, 22, 10, 45, 25, 39, 14, 33, 19, 30, 9,  24, 13, 18, 8,  12, 7,  6,  5,  63};
 
-#elif defined(_MSC_VER)  // MSVC
+constexpr int constexpr_lsb(uint64_t bb) {
+    assert(bb != 0);
+    constexpr uint64_t debruijn64 = 0x03F79D71B4CB0A89ULL;
+    return lsb_index64[((bb ^ (bb - 1)) * debruijn64) >> 58];
+}
 
-  static inline IndexType lsb_(std::uint32_t b) {
+alignas(CacheLineSize) static constexpr struct OffsetIndices {
-    assert(b);
-    unsigned long idx;
-    _BitScanForward(&idx, b);
-    return (IndexType) idx;
-  }
 
-#else  // Compiler is neither GCC nor MSVC compatible
+    #if (USE_SSE41)
+    std::uint8_t offset_indices[256][8];
+    #else
+    std::uint16_t offset_indices[256][8];
+    #endif
 
-#error "Compiler not supported."
+    constexpr OffsetIndices() :
-
+        offset_indices() {
-#endif
+        for (int i = 0; i < 256; ++i)
-
+        {
-
+            std::uint64_t j = i, k = 0;
-#if defined(USE_SSSE3)
+            while (j)
-  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{};
+                offset_indices[i][k++] = constexpr_lsb(j);
-    for (int i = 0; i < 256; ++i)
+                j &= j - 1;
-    {
+            }
-      int j = i;
+            while (k < 8)
-      int k = 0;
+                offset_indices[i][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
+} Lookup;
-  template<const IndexType InputDimensions>
+
-  void find_nnz(const std::int32_t* input, std::uint16_t* out, IndexType& count_out) {
+// Find indices of nonzero numbers in an int32_t array
-#if defined (USE_AVX512)
+template<const IndexType InputDimensions>
+void find_nnz(const std::int32_t* input, std::uint16_t* out, IndexType& count_out) {
+    #if defined(USE_SSSE3)
+        #if defined(USE_AVX512)
     using vec_t = __m512i;
-    #define vec_nnz(a) _mm512_cmpgt_epi32_mask(a, _mm512_setzero_si512())
+            #define vec_nnz(a) _mm512_cmpgt_epi32_mask(a, _mm512_setzero_si512())
-#elif defined (USE_AVX2)
+        #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())))
+            #if defined(USE_VNNI) && !defined(USE_AVXVNNI)
-#elif defined (USE_SSSE3)
+                #define vec_nnz(a) _mm256_cmpgt_epi32_mask(a, _mm256_setzero_si256())
+            #else
+                #define vec_nnz(a) \
+                    _mm256_movemask_ps( \
+                      _mm256_castsi256_ps(_mm256_cmpgt_epi32(a, _mm256_setzero_si256())))
+            #endif
+        #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())))
+            #define vec_nnz(a) \
-#endif
+                _mm_movemask_ps(_mm_castsi128_ps(_mm_cmpgt_epi32(a, _mm_setzero_si128())))
+        #endif
+    using vec128_t = __m128i;
+        #define vec128_zero _mm_setzero_si128()
+        #define vec128_set_16(a) _mm_set1_epi16(a)
+        #if (USE_SSE41)
+            #define vec128_load(a) _mm_cvtepu8_epi16(_mm_loadl_epi64(a))
+        #else
+            #define vec128_load(a) _mm_load_si128(a)
+        #endif
+        #define vec128_storeu(a, b) _mm_storeu_si128(a, b)
+        #define vec128_add(a, b) _mm_add_epi16(a, b)
+    #elif defined(USE_NEON)
+    using vec_t                        = uint32x4_t;
+    static const std::uint32_t Mask[4] = {1, 2, 4, 8};
+        #define vec_nnz(a) vaddvq_u32(vandq_u32(vtstq_u32(a, a), vld1q_u32(Mask)))
+    using vec128_t                     = uint16x8_t;
+        #define vec128_zero vdupq_n_u16(0)
+        #define vec128_set_16(a) vdupq_n_u16(a)
+        #define vec128_load(a) vld1q_u16(reinterpret_cast<const std::uint16_t*>(a))
+        #define vec128_storeu(a, b) vst1q_u16(reinterpret_cast<std::uint16_t*>(a), b)
+        #define vec128_add(a, b) vaddq_u16(a, b)
+    #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 ChunkSize       = std::max<IndexType>(InputSimdWidth, 8);
-    constexpr IndexType NumChunks = InputDimensions / ChunkSize;
+    constexpr IndexType NumChunks       = InputDimensions / ChunkSize;
-    constexpr IndexType InputsPerChunk = ChunkSize / InputSimdWidth;
+    constexpr IndexType InputsPerChunk  = ChunkSize / InputSimdWidth;
     constexpr IndexType OutputsPerChunk = ChunkSize / 8;
 
-    const auto inputVector = reinterpret_cast<const vec_t*>(input);
+    const auto     inputVector = reinterpret_cast<const vec_t*>(input);
-    IndexType count = 0;
+    IndexType      count       = 0;
-    __m128i base = _mm_set1_epi16(0);
+    vec128_t       base        = vec128_zero;
-    __m128i increment = _mm_set1_epi16(8);
+    const vec128_t increment   = vec128_set_16(8);
     for (IndexType i = 0; i < NumChunks; ++i)
     {
-      // bitmask of nonzero values in this chunk
+        // bitmask of nonzero values in this chunk
-      unsigned nnz = 0;
+        unsigned nnz = 0;
-      for (IndexType j = 0; j < InputsPerChunk; ++j)
+        for (IndexType j = 0; j < InputsPerChunk; ++j)
-      {
+        {
-        const vec_t inputChunk = inputVector[i * InputsPerChunk + j];
+            const vec_t inputChunk = inputVector[i * InputsPerChunk + j];
-        nnz |= (unsigned)vec_nnz(inputChunk) << (j * InputSimdWidth);
+            nnz |= unsigned(vec_nnz(inputChunk)) << (j * InputSimdWidth);
-      }
+        }
-      for (IndexType j = 0; j < OutputsPerChunk; ++j)
+        for (IndexType j = 0; j < OutputsPerChunk; ++j)
-      {
+        {
-        const auto lookup = (nnz >> (j * 8)) & 0xFF;
+            const unsigned lookup = (nnz >> (j * 8)) & 0xFF;
-        const auto offsets = _mm_loadu_si128(reinterpret_cast<const __m128i*>(&lookup_indices[lookup]));
+            const vec128_t offsets =
-        _mm_storeu_si128(reinterpret_cast<__m128i*>(out + count), _mm_add_epi16(base, offsets));
+              vec128_load(reinterpret_cast<const vec128_t*>(&Lookup.offset_indices[lookup]));
-        count += lookup_count[lookup];
+            vec128_storeu(reinterpret_cast<vec128_t*>(out + count), vec128_add(base, offsets));
-        base = _mm_add_epi16(base, increment);
+            count += popcount(lookup);
-      }
+            base = vec128_add(base, increment);
+        }
     }
     count_out = count;
-  }
+}
-# undef vec_nnz
+    #undef vec_nnz
+    #undef vec128_zero
+    #undef vec128_set_16
+    #undef vec128_load
+    #undef vec128_storeu
+    #undef vec128_add
 #endif
 
-  // Sparse input implementation
+// Sparse input implementation
-  template <IndexType InDims, IndexType OutDims>
+template<IndexType InDims, IndexType OutDims>
-  class AffineTransformSparseInput {
+class AffineTransformSparseInput {
    public:
     // Input/output type
-    // Input/output type
+    using InputType  = std::uint8_t;
-    using InputType = std::uint8_t;
     using OutputType = std::int32_t;
 
     // Number of input/output dimensions
-    static constexpr IndexType InputDimensions = InDims;
+    static constexpr IndexType InputDimensions  = InDims;
     static constexpr IndexType OutputDimensions = OutDims;
 
-    static_assert(OutputDimensions % 16 == 0, "Only implemented for OutputDimensions divisible by 16.");
+    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)
+#if (USE_SSSE3 | (USE_NEON >= 8))
     static constexpr IndexType ChunkSize = 4;
 #else
     static constexpr IndexType ChunkSize = 1;
@@ -167,120 +186,121 @@ namespace Stockfish::Eval::NNUE::Layers {
 
     // Hash value embedded in the evaluation file
     static constexpr std::uint32_t get_hash_value(std::uint32_t prevHash) {
-      std::uint32_t hashValue = 0xCC03DAE4u;
+        std::uint32_t hashValue = 0xCC03DAE4u;
-      hashValue += OutputDimensions;
+        hashValue += OutputDimensions;
-      hashValue ^= prevHash >> 1;
+        hashValue ^= prevHash >> 1;
-      hashValue ^= prevHash << 31;
+        hashValue ^= prevHash << 31;
-      return hashValue;
+        return hashValue;
     }
 
-    static IndexType get_weight_index_scrambled(IndexType i)
+    static constexpr IndexType get_weight_index_scrambled(IndexType i) {
-    {
+        return (i / ChunkSize) % (PaddedInputDimensions / ChunkSize) * OutputDimensions * ChunkSize
-      return
+             + i / PaddedInputDimensions * ChunkSize + i % ChunkSize;
-        (i / ChunkSize) % (PaddedInputDimensions / ChunkSize) * OutputDimensions * ChunkSize +
-        i / PaddedInputDimensions * ChunkSize +
-        i % ChunkSize;
     }
 
-    static IndexType get_weight_index(IndexType i)
+    static constexpr IndexType get_weight_index(IndexType i) {
-    {
+#if (USE_SSSE3 | (USE_NEON >= 8))
-#if defined (USE_SSSE3)
+        return get_weight_index_scrambled(i);
-      return get_weight_index_scrambled(i);
 #else
-      return i;
+        return i;
 #endif
     }
 
     // Read network parameters
     bool read_parameters(std::istream& stream) {
-      read_little_endian<BiasType>(stream, biases, OutputDimensions);
+        read_little_endian<BiasType>(stream, biases, OutputDimensions);
-      for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+        for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
-        weights[get_weight_index(i)] = read_little_endian<WeightType>(stream);
+            weights[get_weight_index(i)] = read_little_endian<WeightType>(stream);
 
-      return !stream.fail();
+        return !stream.fail();
     }
 
     // Write network parameters
     bool write_parameters(std::ostream& stream) const {
-      write_little_endian<BiasType>(stream, biases, OutputDimensions);
+        write_little_endian<BiasType>(stream, biases, OutputDimensions);
 
-      for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
+        for (IndexType i = 0; i < OutputDimensions * PaddedInputDimensions; ++i)
-        write_little_endian<WeightType>(stream, weights[get_weight_index(i)]);
+            write_little_endian<WeightType>(stream, weights[get_weight_index(i)]);
 
-      return !stream.fail();
+        return !stream.fail();
     }
     // Forward propagation
-    const OutputType* propagate(
+    void propagate(const InputType* input, OutputType* output) const {
-        const InputType* input, OutputType* output) const {
 
-#if defined (USE_SSSE3)
+#if (USE_SSSE3 | (USE_NEON >= 8))
-#if defined (USE_AVX512)
+    #if defined(USE_AVX512)
-      using vec_t = __m512i;
+        using invec_t  = __m512i;
-      #define vec_setzero _mm512_setzero_si512
+        using outvec_t = __m512i;
-      #define vec_set_32 _mm512_set1_epi32
+        #define vec_set_32 _mm512_set1_epi32
-      #define vec_add_dpbusd_32 Simd::m512_add_dpbusd_epi32
+        #define vec_add_dpbusd_32 Simd::m512_add_dpbusd_epi32
-#elif defined (USE_AVX2)
+    #elif defined(USE_AVX2)
-      using vec_t = __m256i;
+        using invec_t  = __m256i;
-      #define vec_setzero _mm256_setzero_si256
+        using outvec_t = __m256i;
-      #define vec_set_32 _mm256_set1_epi32
+        #define vec_set_32 _mm256_set1_epi32
-      #define vec_add_dpbusd_32 Simd::m256_add_dpbusd_epi32
+        #define vec_add_dpbusd_32 Simd::m256_add_dpbusd_epi32
-#elif defined (USE_SSSE3)
+    #elif defined(USE_SSSE3)
-      using vec_t = __m128i;
+        using invec_t  = __m128i;
-      #define vec_setzero _mm_setzero_si128
+        using outvec_t = __m128i;
-      #define vec_set_32 _mm_set1_epi32
+        #define vec_set_32 _mm_set1_epi32
-      #define vec_add_dpbusd_32 Simd::m128_add_dpbusd_epi32
+        #define vec_add_dpbusd_32 Simd::m128_add_dpbusd_epi32
-#endif
+    #elif defined(USE_NEON_DOTPROD)
-      static constexpr IndexType OutputSimdWidth = sizeof(vec_t) / sizeof(OutputType);
+        using invec_t  = int8x16_t;
+        using outvec_t = int32x4_t;
+        #define vec_set_32(a) vreinterpretq_s8_u32(vdupq_n_u32(a))
+        #define vec_add_dpbusd_32 Simd::dotprod_m128_add_dpbusd_epi32
+    #elif defined(USE_NEON)
+        using invec_t  = int8x16_t;
+        using outvec_t = int32x4_t;
+        #define vec_set_32(a) vreinterpretq_s8_u32(vdupq_n_u32(a))
+        #define vec_add_dpbusd_32 Simd::neon_m128_add_dpbusd_epi32
+    #endif
+        static constexpr IndexType OutputSimdWidth = sizeof(outvec_t) / sizeof(OutputType);
 
-      constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 8) / ChunkSize;
+        constexpr IndexType NumChunks = ceil_to_multiple<IndexType>(InputDimensions, 8) / ChunkSize;
-      constexpr IndexType NumRegs = OutputDimensions / OutputSimdWidth;
+        constexpr IndexType NumRegs   = OutputDimensions / OutputSimdWidth;
-      std::uint16_t nnz[NumChunks];
+        std::uint16_t       nnz[NumChunks];
-      IndexType count;
+        IndexType           count;
 
-      const auto input32 = reinterpret_cast<const std::int32_t*>(input);
+        const auto input32 = reinterpret_cast<const std::int32_t*>(input);
 
-      // Find indices of nonzero 32bit blocks
+        // Find indices of nonzero 32-bit blocks
-      find_nnz<NumChunks>(input32, nnz, count);
+        find_nnz<NumChunks>(input32, nnz, count);
 
-      const vec_t* biasvec = reinterpret_cast<const vec_t*>(biases);
+        const outvec_t* biasvec = reinterpret_cast<const outvec_t*>(biases);
-      vec_t acc[NumRegs];
+        outvec_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]);
+            acc[k] = biasvec[k];
-      }
 
-      vec_t* outptr = reinterpret_cast<vec_t*>(output);
+        for (IndexType j = 0; j < count; ++j)
-      for (IndexType k = 0; k < NumRegs; ++k)
+        {
-        outptr[k] = acc[k];
+            const auto    i  = nnz[j];
-# undef vec_setzero
+            const invec_t in = vec_set_32(input32[i]);
-# undef vec_set_32
+            const auto    col =
-# undef vec_add_dpbusd_32
+              reinterpret_cast<const invec_t*>(&weights[i * OutputDimensions * ChunkSize]);
+            for (IndexType k = 0; k < NumRegs; ++k)
+                vec_add_dpbusd_32(acc[k], in, col[k]);
+        }
+
+        outvec_t* outptr = reinterpret_cast<outvec_t*>(output);
+        for (IndexType k = 0; k < NumRegs; ++k)
+            outptr[k] = acc[k];
+    #undef vec_set_32
+    #undef vec_add_dpbusd_32
 #else
-      // Use dense implementation for the other architectures.
+        // Use dense implementation for the other architectures.
-      affine_transform_non_ssse3<
+        affine_transform_non_ssse3<InputDimensions, PaddedInputDimensions, OutputDimensions>(
-        InputDimensions,
+          output, weights, biases, input);
-        PaddedInputDimensions,
-        OutputDimensions>(output, weights, biases, input);
 #endif
-
-      return output;
     }
 
    private:
-    using BiasType = OutputType;
+    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
+#endif  // #ifndef NNUE_LAYERS_AFFINE_TRANSFORM_SPARSE_INPUT_H_INCLUDED

src/nnue/layers/clipped_relu.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,160 +21,144 @@
 #ifndef NNUE_LAYERS_CLIPPED_RELU_H_INCLUDED
 #define NNUE_LAYERS_CLIPPED_RELU_H_INCLUDED
 
+#include <algorithm>
+#include <cstdint>
+#include <iosfwd>
+
 #include "../nnue_common.h"
 
 namespace Stockfish::Eval::NNUE::Layers {
 
-  // Clipped ReLU
+// Clipped ReLU
-  template <IndexType InDims>
+template<IndexType InDims>
-  class ClippedReLU {
+class ClippedReLU {
    public:
     // Input/output type
-    using InputType = std::int32_t;
+    using InputType  = std::int32_t;
     using OutputType = std::uint8_t;
 
     // Number of input/output dimensions
-    static constexpr IndexType InputDimensions = InDims;
+    static constexpr IndexType InputDimensions  = InDims;
     static constexpr IndexType OutputDimensions = InputDimensions;
     static constexpr IndexType PaddedOutputDimensions =
-        ceil_to_multiple<IndexType>(OutputDimensions, 32);
+      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;
+        std::uint32_t hashValue = 0x538D24C7u;
-      hashValue += prevHash;
+        hashValue += prevHash;
-      return hashValue;
+        return hashValue;
     }
 
     // Read network parameters
-    bool read_parameters(std::istream&) {
+    bool read_parameters(std::istream&) { return true; }
-      return true;
-    }
 
     // Write network parameters
-    bool write_parameters(std::ostream&) const {
+    bool write_parameters(std::ostream&) const { return true; }
-      return true;
-    }
 
     // Forward propagation
-    const OutputType* propagate(
+    void propagate(const InputType* input, OutputType* output) const {
-        const InputType* input, OutputType* output) const {
 
-  #if defined(USE_AVX2)
+#if defined(USE_AVX2)
-      if constexpr (InputDimensions % SimdWidth == 0) {
+        if constexpr (InputDimensions % SimdWidth == 0)
+        {
+            constexpr IndexType NumChunks = InputDimensions / SimdWidth;
+            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_srli_epi16(_mm256_packus_epi32(_mm256_load_si256(&in[i * 4 + 0]),
+                                                        _mm256_load_si256(&in[i * 4 + 1])),
+                                    WeightScaleBits);
+                const __m256i words1 =
+                  _mm256_srli_epi16(_mm256_packus_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_packs_epi16(words0, words1), Offsets));
+            }
+        }
+        else
+        {
+            constexpr IndexType NumChunks = InputDimensions / (SimdWidth / 2);
+            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_srli_epi16(
+                  _mm_packus_epi32(_mm_load_si128(&in[i * 4 + 0]), _mm_load_si128(&in[i * 4 + 1])),
+                  WeightScaleBits);
+                const __m128i words1 = _mm_srli_epi16(
+                  _mm_packus_epi32(_mm_load_si128(&in[i * 4 + 2]), _mm_load_si128(&in[i * 4 + 3])),
+                  WeightScaleBits);
+                _mm_store_si128(&out[i], _mm_packs_epi16(words0, words1));
+            }
+        }
+        constexpr IndexType Start = InputDimensions % SimdWidth == 0
+                                    ? InputDimensions / SimdWidth * SimdWidth
+                                    : InputDimensions / (SimdWidth / 2) * (SimdWidth / 2);
+
+#elif defined(USE_SSE2)
         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);
+    #ifndef USE_SSE41
-        const auto in = reinterpret_cast<const __m256i*>(input);
+        const __m128i k0x80s = _mm_set1_epi8(-128);
-        const auto out = reinterpret_cast<__m256i*>(output);
+    #endif
-        for (IndexType i = 0; i < NumChunks; ++i) {
+
-          const __m256i words0 = _mm256_srai_epi16(_mm256_packs_epi32(
+        const auto in  = reinterpret_cast<const __m128i*>(input);
-              _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) {
+        for (IndexType i = 0; i < NumChunks; ++i)
-          const __m128i words0 = _mm_srai_epi16(_mm_packs_epi32(
+        {
-              _mm_load_si128(&in[i * 4 + 0]),
+    #if defined(USE_SSE41)
-              _mm_load_si128(&in[i * 4 + 1])), WeightScaleBits);
+            const __m128i words0 = _mm_srli_epi16(
-          const __m128i words1 = _mm_srai_epi16(_mm_packs_epi32(
+              _mm_packus_epi32(_mm_load_si128(&in[i * 4 + 0]), _mm_load_si128(&in[i * 4 + 1])),
-              _mm_load_si128(&in[i * 4 + 2]),
+              WeightScaleBits);
-              _mm_load_si128(&in[i * 4 + 3])), WeightScaleBits);
+            const __m128i words1 = _mm_srli_epi16(
-          const __m128i packedbytes = _mm_packs_epi16(words0, words1);
+              _mm_packus_epi32(_mm_load_si128(&in[i * 4 + 2]), _mm_load_si128(&in[i * 4 + 3])),
-          _mm_store_si128(&out[i], _mm_max_epi8(packedbytes, Zero));
+              WeightScaleBits);
+            _mm_store_si128(&out[i], _mm_packs_epi16(words0, words1));
+    #else
+            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_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s));
+    #endif
         }
-      }
+        constexpr IndexType Start = NumChunks * SimdWidth;
-      constexpr IndexType Start =
-        InputDimensions % SimdWidth == 0
-        ? InputDimensions / SimdWidth * SimdWidth
-        : InputDimensions / (SimdWidth / 2) * (SimdWidth / 2);
 
-  #elif defined(USE_SSE2)
+#elif defined(USE_NEON)
-      constexpr IndexType NumChunks = InputDimensions / SimdWidth;
+        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
 
-  #ifdef USE_SSE41
+        for (IndexType i = Start; i < InputDimensions; ++i)
-      const __m128i Zero = _mm_setzero_si128();
+        {
-  #else
+            output[i] = static_cast<OutputType>(std::clamp(input[i] >> WeightScaleBits, 0, 127));
-      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
+#endif  // NNUE_LAYERS_CLIPPED_RELU_H_INCLUDED

src/nnue/layers/simd.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -20,384 +20,115 @@
 #define STOCKFISH_SIMD_H_INCLUDED
 
 #if defined(USE_AVX2)
-# include <immintrin.h>
+    #include <immintrin.h>
 
 #elif defined(USE_SSE41)
-# include <smmintrin.h>
+    #include <smmintrin.h>
 
 #elif defined(USE_SSSE3)
-# include <tmmintrin.h>
+    #include <tmmintrin.h>
 
 #elif defined(USE_SSE2)
-# include <emmintrin.h>
+    #include <emmintrin.h>
-
-#elif defined(USE_MMX)
-# include <mmintrin.h>
 
 #elif defined(USE_NEON)
-# include <arm_neon.h>
+    #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)
+#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
 
+[[maybe_unused]] static int m512_hadd(__m512i sum, int bias) {
+    return _mm512_reduce_add_epi32(sum) + bias;
 }
 
-#endif // STOCKFISH_SIMD_H_INCLUDED
+[[maybe_unused]] static void m512_add_dpbusd_epi32(__m512i& acc, __m512i a, __m512i b) {
+
+    #if defined(USE_VNNI)
+    acc = _mm512_dpbusd_epi32(acc, a, b);
+    #else
+    __m512i product0 = _mm512_maddubs_epi16(a, b);
+    product0         = _mm512_madd_epi16(product0, _mm512_set1_epi16(1));
+    acc              = _mm512_add_epi32(acc, product0);
+    #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 void m256_add_dpbusd_epi32(__m256i& acc, __m256i a, __m256i b) {
+
+    #if defined(USE_VNNI)
+    acc = _mm256_dpbusd_epi32(acc, a, b);
+    #else
+    __m256i product0 = _mm256_maddubs_epi16(a, b);
+    product0         = _mm256_madd_epi16(product0, _mm256_set1_epi16(1));
+    acc              = _mm256_add_epi32(acc, product0);
+    #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 void m128_add_dpbusd_epi32(__m128i& acc, __m128i a, __m128i b) {
+
+    __m128i product0 = _mm_maddubs_epi16(a, b);
+    product0         = _mm_madd_epi16(product0, _mm_set1_epi16(1));
+    acc              = _mm_add_epi32(acc, product0);
+}
+
+#endif
+
+#if defined(USE_NEON_DOTPROD)
+
+[[maybe_unused]] static void
+dotprod_m128_add_dpbusd_epi32(int32x4_t& acc, int8x16_t a, int8x16_t b) {
+
+    acc = vdotq_s32(acc, a, b);
+}
+#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;
+}
+
+#endif
+
+#if USE_NEON >= 8
+[[maybe_unused]] static void neon_m128_add_dpbusd_epi32(int32x4_t& acc, int8x16_t a, int8x16_t b) {
+
+    int16x8_t product0 = vmull_s8(vget_low_s8(a), vget_low_s8(b));
+    int16x8_t product1 = vmull_high_s8(a, b);
+    int16x8_t sum      = vpaddq_s16(product0, product1);
+    acc                = vpadalq_s16(acc, sum);
+}
+#endif
+}
+
+#endif  // STOCKFISH_SIMD_H_INCLUDED

src/nnue/layers/sqr_clipped_relu.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,100 +21,83 @@
 #ifndef NNUE_LAYERS_SQR_CLIPPED_RELU_H_INCLUDED
 #define NNUE_LAYERS_SQR_CLIPPED_RELU_H_INCLUDED
 
+#include <algorithm>
+#include <cstdint>
+#include <iosfwd>
+
 #include "../nnue_common.h"
 
 namespace Stockfish::Eval::NNUE::Layers {
 
-  // Clipped ReLU
+// Clipped ReLU
-  template <IndexType InDims>
+template<IndexType InDims>
-  class SqrClippedReLU {
+class SqrClippedReLU {
    public:
     // Input/output type
-    using InputType = std::int32_t;
+    using InputType  = std::int32_t;
     using OutputType = std::uint8_t;
 
     // Number of input/output dimensions
-    static constexpr IndexType InputDimensions = InDims;
+    static constexpr IndexType InputDimensions  = InDims;
     static constexpr IndexType OutputDimensions = InputDimensions;
     static constexpr IndexType PaddedOutputDimensions =
-        ceil_to_multiple<IndexType>(OutputDimensions, 32);
+      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;
+        std::uint32_t hashValue = 0x538D24C7u;
-      hashValue += prevHash;
+        hashValue += prevHash;
-      return hashValue;
+        return hashValue;
     }
 
     // Read network parameters
-    bool read_parameters(std::istream&) {
+    bool read_parameters(std::istream&) { return true; }
-      return true;
-    }
 
     // Write network parameters
-    bool write_parameters(std::ostream&) const {
+    bool write_parameters(std::ostream&) const { return true; }
-      return true;
-    }
 
     // Forward propagation
-    const OutputType* propagate(
+    void propagate(const InputType* input, OutputType* output) const {
-        const InputType* input, OutputType* output) const {
 
-  #if defined(USE_SSE2)
+#if defined(USE_SSE2)
-      constexpr IndexType NumChunks = InputDimensions / 16;
+        constexpr IndexType NumChunks = InputDimensions / 16;
 
-  #ifdef USE_SSE41
+        static_assert(WeightScaleBits == 6);
-      const __m128i Zero = _mm_setzero_si128();
+        const auto in  = reinterpret_cast<const __m128i*>(input);
-  #else
+        const auto out = reinterpret_cast<__m128i*>(output);
-      const __m128i k0x80s = _mm_set1_epi8(-128);
+        for (IndexType i = 0; i < NumChunks; ++i)
-  #endif
+        {
+            __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]));
 
-      static_assert(WeightScaleBits == 6);
+            // We shift by WeightScaleBits * 2 = 12 and divide by 128
-      const auto in = reinterpret_cast<const __m128i*>(input);
+            // which is an additional shift-right of 7, meaning 19 in total.
-      const auto out = reinterpret_cast<__m128i*>(output);
+            // MulHi strips the lower 16 bits so we need to shift out 3 more to match.
-      for (IndexType i = 0; i < NumChunks; ++i) {
+            words0 = _mm_srli_epi16(_mm_mulhi_epi16(words0, words0), 3);
-        __m128i words0 = _mm_packs_epi32(
+            words1 = _mm_srli_epi16(_mm_mulhi_epi16(words1, words1), 3);
-            _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
+            _mm_store_si128(&out[i], _mm_packs_epi16(words0, words1));
-        words0 = _mm_srli_epi16(_mm_mulhi_epi16(words0, words0), 3);
+        }
-        words1 = _mm_srli_epi16(_mm_mulhi_epi16(words1, words1), 3);
+        constexpr IndexType Start = NumChunks * 16;
 
-        const __m128i packedbytes = _mm_packs_epi16(words0, words1);
+#else
+        constexpr IndexType Start = 0;
+#endif
 
-        _mm_store_si128(&out[i],
+        for (IndexType i = Start; i < InputDimensions; ++i)
-
+        {
-  #ifdef USE_SSE41
+            output[i] = static_cast<OutputType>(
-          _mm_max_epi8(packedbytes, Zero)
+              // Really should be /127 but we need to make it fast so we right-shift
-  #else
+              // by an extra 7 bits instead. Needs to be accounted for in the trainer.
-          _mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s)
+              std::min(127ll, ((long long) (input[i]) * input[i]) >> (2 * WeightScaleBits + 7)));
-  #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
+#endif  // NNUE_LAYERS_SQR_CLIPPED_RELU_H_INCLUDED

src/nnue/network.cpp

@@ -0,0 +1,467 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 "network.h"
+
+#include <cstdlib>
+#include <fstream>
+#include <iostream>
+#include <memory>
+#include <optional>
+#include <type_traits>
+#include <vector>
+
+#define INCBIN_SILENCE_BITCODE_WARNING
+#include "../incbin/incbin.h"
+
+#include "../evaluate.h"
+#include "../memory.h"
+#include "../misc.h"
+#include "../position.h"
+#include "../types.h"
+#include "nnue_architecture.h"
+#include "nnue_common.h"
+#include "nnue_misc.h"
+
+// Macro to embed the default efficiently updatable neural network (NNUE) file
+// data in the engine binary (using incbin.h, by Dale Weiler).
+// This macro invocation will declare the following three variables
+//     const unsigned char        gEmbeddedNNUEData[];  // a pointer to the embedded data
+//     const unsigned char *const gEmbeddedNNUEEnd;     // a marker to the end
+//     const unsigned int         gEmbeddedNNUESize;    // the size of the embedded file
+// Note that this does not work in Microsoft Visual Studio.
+#if !defined(_MSC_VER) && !defined(NNUE_EMBEDDING_OFF)
+INCBIN(EmbeddedNNUEBig, EvalFileDefaultNameBig);
+INCBIN(EmbeddedNNUESmall, EvalFileDefaultNameSmall);
+#else
+const unsigned char        gEmbeddedNNUEBigData[1]   = {0x0};
+const unsigned char* const gEmbeddedNNUEBigEnd       = &gEmbeddedNNUEBigData[1];
+const unsigned int         gEmbeddedNNUEBigSize      = 1;
+const unsigned char        gEmbeddedNNUESmallData[1] = {0x0};
+const unsigned char* const gEmbeddedNNUESmallEnd     = &gEmbeddedNNUESmallData[1];
+const unsigned int         gEmbeddedNNUESmallSize    = 1;
+#endif
+
+namespace {
+
+struct EmbeddedNNUE {
+    EmbeddedNNUE(const unsigned char* embeddedData,
+                 const unsigned char* embeddedEnd,
+                 const unsigned int   embeddedSize) :
+        data(embeddedData),
+        end(embeddedEnd),
+        size(embeddedSize) {}
+    const unsigned char* data;
+    const unsigned char* end;
+    const unsigned int   size;
+};
+
+using namespace Stockfish::Eval::NNUE;
+
+EmbeddedNNUE get_embedded(EmbeddedNNUEType type) {
+    if (type == EmbeddedNNUEType::BIG)
+        return EmbeddedNNUE(gEmbeddedNNUEBigData, gEmbeddedNNUEBigEnd, gEmbeddedNNUEBigSize);
+    else
+        return EmbeddedNNUE(gEmbeddedNNUESmallData, gEmbeddedNNUESmallEnd, gEmbeddedNNUESmallSize);
+}
+
+}
+
+
+namespace Stockfish::Eval::NNUE {
+
+
+namespace Detail {
+
+// 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, T& reference) {
+
+    write_little_endian<std::uint32_t>(stream, T::get_hash_value());
+    return reference.write_parameters(stream);
+}
+
+}  // namespace Detail
+
+template<typename Arch, typename Transformer>
+Network<Arch, Transformer>::Network(const Network<Arch, Transformer>& other) :
+    evalFile(other.evalFile),
+    embeddedType(other.embeddedType) {
+
+    if (other.featureTransformer)
+        featureTransformer = make_unique_large_page<Transformer>(*other.featureTransformer);
+
+    network = make_unique_aligned<Arch[]>(LayerStacks);
+
+    if (!other.network)
+        return;
+
+    for (std::size_t i = 0; i < LayerStacks; ++i)
+        network[i] = other.network[i];
+}
+
+template<typename Arch, typename Transformer>
+Network<Arch, Transformer>&
+Network<Arch, Transformer>::operator=(const Network<Arch, Transformer>& other) {
+    evalFile     = other.evalFile;
+    embeddedType = other.embeddedType;
+
+    if (other.featureTransformer)
+        featureTransformer = make_unique_large_page<Transformer>(*other.featureTransformer);
+
+    network = make_unique_aligned<Arch[]>(LayerStacks);
+
+    if (!other.network)
+        return *this;
+
+    for (std::size_t i = 0; i < LayerStacks; ++i)
+        network[i] = other.network[i];
+
+    return *this;
+}
+
+template<typename Arch, typename Transformer>
+void Network<Arch, Transformer>::load(const std::string& rootDirectory, std::string evalfilePath) {
+#if defined(DEFAULT_NNUE_DIRECTORY)
+    std::vector<std::string> dirs = {"<internal>", "", rootDirectory,
+                                     stringify(DEFAULT_NNUE_DIRECTORY)};
+#else
+    std::vector<std::string> dirs = {"<internal>", "", rootDirectory};
+#endif
+
+    if (evalfilePath.empty())
+        evalfilePath = evalFile.defaultName;
+
+    for (const auto& directory : dirs)
+    {
+        if (evalFile.current != evalfilePath)
+        {
+            if (directory != "<internal>")
+            {
+                load_user_net(directory, evalfilePath);
+            }
+
+            if (directory == "<internal>" && evalfilePath == evalFile.defaultName)
+            {
+                load_internal();
+            }
+        }
+    }
+}
+
+
+template<typename Arch, typename Transformer>
+bool Network<Arch, Transformer>::save(const std::optional<std::string>& filename) const {
+    std::string actualFilename;
+    std::string msg;
+
+    if (filename.has_value())
+        actualFilename = filename.value();
+    else
+    {
+        if (evalFile.current != evalFile.defaultName)
+        {
+            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 = evalFile.defaultName;
+    }
+
+    std::ofstream stream(actualFilename, std::ios_base::binary);
+    bool          saved = save(stream, evalFile.current, evalFile.netDescription);
+
+    msg = saved ? "Network saved successfully to " + actualFilename : "Failed to export a net";
+
+    sync_cout << msg << sync_endl;
+    return saved;
+}
+
+
+template<typename Arch, typename Transformer>
+NetworkOutput
+Network<Arch, Transformer>::evaluate(const Position&                         pos,
+                                     AccumulatorCaches::Cache<FTDimensions>* cache) const {
+    // 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<FTDimensions, nullptr>::BufferSize
+                                   + alignment / sizeof(TransformedFeatureType)];
+
+    auto* transformedFeatures = align_ptr_up<alignment>(&transformedFeaturesUnaligned[0]);
+#else
+    alignas(alignment) TransformedFeatureType
+      transformedFeatures[FeatureTransformer<FTDimensions, nullptr>::BufferSize];
+#endif
+
+    ASSERT_ALIGNED(transformedFeatures, alignment);
+
+    const int  bucket     = (pos.count<ALL_PIECES>() - 1) / 4;
+    const auto psqt       = featureTransformer->transform(pos, cache, transformedFeatures, bucket);
+    const auto positional = network[bucket].propagate(transformedFeatures);
+    return {static_cast<Value>(psqt / OutputScale), static_cast<Value>(positional / OutputScale)};
+}
+
+
+template<typename Arch, typename Transformer>
+void Network<Arch, Transformer>::verify(std::string                                  evalfilePath,
+                                        const std::function<void(std::string_view)>& f) const {
+    if (evalfilePath.empty())
+        evalfilePath = evalFile.defaultName;
+
+    if (evalFile.current != evalfilePath)
+    {
+        if (f)
+        {
+            std::string msg1 =
+              "Network evaluation parameters compatible with the engine must be available.";
+            std::string msg2 = "The network file " + evalfilePath + " was not loaded successfully.";
+            std::string msg3 = "The UCI option EvalFile might need to specify the full path, "
+                               "including the directory name, to the network file.";
+            std::string msg4 = "The default net can be downloaded from: "
+                               "https://tests.stockfishchess.org/api/nn/"
+                             + evalFile.defaultName;
+            std::string msg5 = "The engine will be terminated now.";
+
+            std::string msg = "ERROR: " + msg1 + '\n' + "ERROR: " + msg2 + '\n' + "ERROR: " + msg3
+                            + '\n' + "ERROR: " + msg4 + '\n' + "ERROR: " + msg5 + '\n';
+
+            f(msg);
+        }
+
+        exit(EXIT_FAILURE);
+    }
+
+    if (f)
+    {
+        size_t size = sizeof(*featureTransformer) + sizeof(Arch) * LayerStacks;
+        f("info string NNUE evaluation using " + evalfilePath + " ("
+          + std::to_string(size / (1024 * 1024)) + "MiB, ("
+          + std::to_string(featureTransformer->InputDimensions) + ", "
+          + std::to_string(network[0].TransformedFeatureDimensions) + ", "
+          + std::to_string(network[0].FC_0_OUTPUTS) + ", " + std::to_string(network[0].FC_1_OUTPUTS)
+          + ", 1))");
+    }
+}
+
+
+template<typename Arch, typename Transformer>
+void Network<Arch, Transformer>::hint_common_access(
+  const Position& pos, AccumulatorCaches::Cache<FTDimensions>* cache) const {
+    featureTransformer->hint_common_access(pos, cache);
+}
+
+template<typename Arch, typename Transformer>
+NnueEvalTrace
+Network<Arch, Transformer>::trace_evaluate(const Position&                         pos,
+                                           AccumulatorCaches::Cache<FTDimensions>* cache) const {
+    // 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<FTDimensions, nullptr>::BufferSize
+                                   + alignment / sizeof(TransformedFeatureType)];
+
+    auto* transformedFeatures = align_ptr_up<alignment>(&transformedFeaturesUnaligned[0]);
+#else
+    alignas(alignment) TransformedFeatureType
+      transformedFeatures[FeatureTransformer<FTDimensions, nullptr>::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, cache, 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;
+}
+
+
+template<typename Arch, typename Transformer>
+void Network<Arch, Transformer>::load_user_net(const std::string& dir,
+                                               const std::string& evalfilePath) {
+    std::ifstream stream(dir + evalfilePath, std::ios::binary);
+    auto          description = load(stream);
+
+    if (description.has_value())
+    {
+        evalFile.current        = evalfilePath;
+        evalFile.netDescription = description.value();
+    }
+}
+
+
+template<typename Arch, typename Transformer>
+void Network<Arch, Transformer>::load_internal() {
+    // C++ way to prepare a buffer for a memory stream
+    class MemoryBuffer: public std::basic_streambuf<char> {
+       public:
+        MemoryBuffer(char* p, size_t n) {
+            setg(p, p, p + n);
+            setp(p, p + n);
+        }
+    };
+
+    const auto embedded = get_embedded(embeddedType);
+
+    MemoryBuffer buffer(const_cast<char*>(reinterpret_cast<const char*>(embedded.data)),
+                        size_t(embedded.size));
+
+    std::istream stream(&buffer);
+    auto         description = load(stream);
+
+    if (description.has_value())
+    {
+        evalFile.current        = evalFile.defaultName;
+        evalFile.netDescription = description.value();
+    }
+}
+
+
+template<typename Arch, typename Transformer>
+void Network<Arch, Transformer>::initialize() {
+    featureTransformer = make_unique_large_page<Transformer>();
+    network            = make_unique_aligned<Arch[]>(LayerStacks);
+}
+
+
+template<typename Arch, typename Transformer>
+bool Network<Arch, Transformer>::save(std::ostream&      stream,
+                                      const std::string& name,
+                                      const std::string& netDescription) const {
+    if (name.empty() || name == "None")
+        return false;
+
+    return write_parameters(stream, netDescription);
+}
+
+
+template<typename Arch, typename Transformer>
+std::optional<std::string> Network<Arch, Transformer>::load(std::istream& stream) {
+    initialize();
+    std::string description;
+
+    return read_parameters(stream, description) ? std::make_optional(description) : std::nullopt;
+}
+
+
+// Read network header
+template<typename Arch, typename Transformer>
+bool Network<Arch, Transformer>::read_header(std::istream&  stream,
+                                             std::uint32_t* hashValue,
+                                             std::string*   desc) const {
+    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
+template<typename Arch, typename Transformer>
+bool Network<Arch, Transformer>::write_header(std::ostream&      stream,
+                                              std::uint32_t      hashValue,
+                                              const std::string& desc) const {
+    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();
+}
+
+
+template<typename Arch, typename Transformer>
+bool Network<Arch, Transformer>::read_parameters(std::istream& stream,
+                                                 std::string&  netDescription) const {
+    std::uint32_t hashValue;
+    if (!read_header(stream, &hashValue, &netDescription))
+        return false;
+    if (hashValue != Network::hash)
+        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();
+}
+
+
+template<typename Arch, typename Transformer>
+bool Network<Arch, Transformer>::write_parameters(std::ostream&      stream,
+                                                  const std::string& netDescription) const {
+    if (!write_header(stream, Network::hash, 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);
+}
+
+// Explicit template instantiation
+
+template class Network<
+  NetworkArchitecture<TransformedFeatureDimensionsBig, L2Big, L3Big>,
+  FeatureTransformer<TransformedFeatureDimensionsBig, &StateInfo::accumulatorBig>>;
+
+template class Network<
+  NetworkArchitecture<TransformedFeatureDimensionsSmall, L2Small, L3Small>,
+  FeatureTransformer<TransformedFeatureDimensionsSmall, &StateInfo::accumulatorSmall>>;
+
+}  // namespace Stockfish::Eval::NNUE

src/nnue/network.h

@@ -0,0 +1,134 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 NETWORK_H_INCLUDED
+#define NETWORK_H_INCLUDED
+
+#include <cstdint>
+#include <functional>
+#include <iostream>
+#include <optional>
+#include <string>
+#include <string_view>
+#include <tuple>
+#include <utility>
+
+#include "../memory.h"
+#include "../position.h"
+#include "../types.h"
+#include "nnue_accumulator.h"
+#include "nnue_architecture.h"
+#include "nnue_feature_transformer.h"
+#include "nnue_misc.h"
+
+namespace Stockfish::Eval::NNUE {
+
+enum class EmbeddedNNUEType {
+    BIG,
+    SMALL,
+};
+
+using NetworkOutput = std::tuple<Value, Value>;
+
+template<typename Arch, typename Transformer>
+class Network {
+    static constexpr IndexType FTDimensions = Arch::TransformedFeatureDimensions;
+
+   public:
+    Network(EvalFile file, EmbeddedNNUEType type) :
+        evalFile(file),
+        embeddedType(type) {}
+
+    Network(const Network& other);
+    Network(Network&& other) = default;
+
+    Network& operator=(const Network& other);
+    Network& operator=(Network&& other) = default;
+
+    void load(const std::string& rootDirectory, std::string evalfilePath);
+    bool save(const std::optional<std::string>& filename) const;
+
+    NetworkOutput evaluate(const Position&                         pos,
+                           AccumulatorCaches::Cache<FTDimensions>* cache) const;
+
+
+    void hint_common_access(const Position&                         pos,
+                            AccumulatorCaches::Cache<FTDimensions>* cache) const;
+
+    void verify(std::string evalfilePath, const std::function<void(std::string_view)>&) const;
+    NnueEvalTrace trace_evaluate(const Position&                         pos,
+                                 AccumulatorCaches::Cache<FTDimensions>* cache) const;
+
+   private:
+    void load_user_net(const std::string&, const std::string&);
+    void load_internal();
+
+    void initialize();
+
+    bool                       save(std::ostream&, const std::string&, const std::string&) const;
+    std::optional<std::string> load(std::istream&);
+
+    bool read_header(std::istream&, std::uint32_t*, std::string*) const;
+    bool write_header(std::ostream&, std::uint32_t, const std::string&) const;
+
+    bool read_parameters(std::istream&, std::string&) const;
+    bool write_parameters(std::ostream&, const std::string&) const;
+
+    // Input feature converter
+    LargePagePtr<Transformer> featureTransformer;
+
+    // Evaluation function
+    AlignedPtr<Arch[]> network;
+
+    EvalFile         evalFile;
+    EmbeddedNNUEType embeddedType;
+
+    // Hash value of evaluation function structure
+    static constexpr std::uint32_t hash = Transformer::get_hash_value() ^ Arch::get_hash_value();
+
+    template<IndexType Size>
+    friend struct AccumulatorCaches::Cache;
+};
+
+// Definitions of the network types
+using SmallFeatureTransformer =
+  FeatureTransformer<TransformedFeatureDimensionsSmall, &StateInfo::accumulatorSmall>;
+using SmallNetworkArchitecture =
+  NetworkArchitecture<TransformedFeatureDimensionsSmall, L2Small, L3Small>;
+
+using BigFeatureTransformer =
+  FeatureTransformer<TransformedFeatureDimensionsBig, &StateInfo::accumulatorBig>;
+using BigNetworkArchitecture = NetworkArchitecture<TransformedFeatureDimensionsBig, L2Big, L3Big>;
+
+using NetworkBig   = Network<BigNetworkArchitecture, BigFeatureTransformer>;
+using NetworkSmall = Network<SmallNetworkArchitecture, SmallFeatureTransformer>;
+
+
+struct Networks {
+    Networks(NetworkBig&& nB, NetworkSmall&& nS) :
+        big(std::move(nB)),
+        small(std::move(nS)) {}
+
+    NetworkBig   big;
+    NetworkSmall small;
+};
+
+
+}  // namespace Stockfish
+
+#endif

src/nnue/nnue_accumulator.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,17 +21,80 @@
 #ifndef NNUE_ACCUMULATOR_H_INCLUDED
 #define NNUE_ACCUMULATOR_H_INCLUDED
 
+#include <cstdint>
+
 #include "nnue_architecture.h"
+#include "nnue_common.h"
 
 namespace Stockfish::Eval::NNUE {
 
-  // Class that holds the result of affine transformation of input features
+using BiasType       = std::int16_t;
-  struct alignas(CacheLineSize) Accumulator {
+using PSQTWeightType = std::int32_t;
-    std::int16_t accumulation[2][TransformedFeatureDimensions];
+using IndexType      = std::uint32_t;
-    std::int32_t psqtAccumulation[2][PSQTBuckets];
+
-    bool computed[2];
+// Class that holds the result of affine transformation of input features
-  };
+template<IndexType Size>
+struct alignas(CacheLineSize) Accumulator {
+    std::int16_t accumulation[COLOR_NB][Size];
+    std::int32_t psqtAccumulation[COLOR_NB][PSQTBuckets];
+    bool         computed[COLOR_NB];
+};
+
+
+// AccumulatorCaches struct provides per-thread accumulator caches, where each
+// cache contains multiple entries for each of the possible king squares.
+// When the accumulator needs to be refreshed, the cached entry is used to more
+// efficiently update the accumulator, instead of rebuilding it from scratch.
+// This idea, was first described by Luecx (author of Koivisto) and
+// is commonly referred to as "Finny Tables".
+struct AccumulatorCaches {
+
+    template<typename Networks>
+    AccumulatorCaches(const Networks& networks) {
+        clear(networks);
+    }
+
+    template<IndexType Size>
+    struct alignas(CacheLineSize) Cache {
+
+        struct alignas(CacheLineSize) Entry {
+            BiasType       accumulation[Size];
+            PSQTWeightType psqtAccumulation[PSQTBuckets];
+            Bitboard       byColorBB[COLOR_NB];
+            Bitboard       byTypeBB[PIECE_TYPE_NB];
+
+            // To initialize a refresh entry, we set all its bitboards empty,
+            // so we put the biases in the accumulation, without any weights on top
+            void clear(const BiasType* biases) {
+
+                std::memcpy(accumulation, biases, sizeof(accumulation));
+                std::memset((uint8_t*) this + offsetof(Entry, psqtAccumulation), 0,
+                            sizeof(Entry) - offsetof(Entry, psqtAccumulation));
+            }
+        };
+
+        template<typename Network>
+        void clear(const Network& network) {
+            for (auto& entries1D : entries)
+                for (auto& entry : entries1D)
+                    entry.clear(network.featureTransformer->biases);
+        }
+
+        std::array<Entry, COLOR_NB>& operator[](Square sq) { return entries[sq]; }
+
+        std::array<std::array<Entry, COLOR_NB>, SQUARE_NB> entries;
+    };
+
+    template<typename Networks>
+    void clear(const Networks& networks) {
+        big.clear(networks.big);
+        small.clear(networks.small);
+    }
+
+    Cache<TransformedFeatureDimensionsBig>   big;
+    Cache<TransformedFeatureDimensionsSmall> small;
+};
 
 }  // namespace Stockfish::Eval::NNUE
 
-#endif // NNUE_ACCUMULATOR_H_INCLUDED
+#endif  // NNUE_ACCUMULATOR_H_INCLUDED

src/nnue/nnue_architecture.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,18 +21,16 @@
 #ifndef NNUE_ARCHITECTURE_H_INCLUDED
 #define NNUE_ARCHITECTURE_H_INCLUDED
 
-#include <memory>
+#include <cstdint>
-
+#include <cstring>
-#include "nnue_common.h"
+#include <iosfwd>
 
 #include "features/half_ka_v2_hm.h"
-
-#include "layers/affine_transform_sparse_input.h"
 #include "layers/affine_transform.h"
+#include "layers/affine_transform_sparse_input.h"
 #include "layers/clipped_relu.h"
 #include "layers/sqr_clipped_relu.h"
-
+#include "nnue_common.h"
-#include "../misc.h"
 
 namespace Stockfish::Eval::NNUE {
 
@@ -40,98 +38,100 @@ namespace Stockfish::Eval::NNUE {
 using FeatureSet = Features::HalfKAv2_hm;
 
 // Number of input feature dimensions after conversion
-constexpr IndexType TransformedFeatureDimensions = 1536;
+constexpr IndexType TransformedFeatureDimensionsBig = 3072;
+constexpr int       L2Big                           = 15;
+constexpr int       L3Big                           = 32;
+
+constexpr IndexType TransformedFeatureDimensionsSmall = 128;
+constexpr int       L2Small                           = 15;
+constexpr int       L3Small                           = 32;
+
 constexpr IndexType PSQTBuckets = 8;
 constexpr IndexType LayerStacks = 8;
 
-struct Network
+template<IndexType L1, int L2, int L3>
-{
+struct NetworkArchitecture {
-  static constexpr int FC_0_OUTPUTS = 15;
+    static constexpr IndexType TransformedFeatureDimensions = L1;
-  static constexpr int FC_1_OUTPUTS = 32;
+    static constexpr int       FC_0_OUTPUTS                 = L2;
+    static constexpr int       FC_1_OUTPUTS                 = L3;
 
-  Layers::AffineTransformSparseInput<TransformedFeatureDimensions, FC_0_OUTPUTS + 1> fc_0;
+    Layers::AffineTransformSparseInput<TransformedFeatureDimensions, FC_0_OUTPUTS + 1> fc_0;
-  Layers::SqrClippedReLU<FC_0_OUTPUTS + 1> ac_sqr_0;
+    Layers::SqrClippedReLU<FC_0_OUTPUTS + 1>                                           ac_sqr_0;
-  Layers::ClippedReLU<FC_0_OUTPUTS + 1> ac_0;
+    Layers::ClippedReLU<FC_0_OUTPUTS + 1>                                              ac_0;
-  Layers::AffineTransform<FC_0_OUTPUTS * 2, FC_1_OUTPUTS> fc_1;
+    Layers::AffineTransform<FC_0_OUTPUTS * 2, FC_1_OUTPUTS>                            fc_1;
-  Layers::ClippedReLU<FC_1_OUTPUTS> ac_1;
+    Layers::ClippedReLU<FC_1_OUTPUTS>                                                  ac_1;
-  Layers::AffineTransform<FC_1_OUTPUTS, 1> fc_2;
+    Layers::AffineTransform<FC_1_OUTPUTS, 1>                                           fc_2;
 
-  // Hash value embedded in the evaluation file
+    // Hash value embedded in the evaluation file
-  static constexpr std::uint32_t get_hash_value() {
+    static constexpr std::uint32_t get_hash_value() {
-    // input slice hash
+        // input slice hash
-    std::uint32_t hashValue = 0xEC42E90Du;
+        std::uint32_t hashValue = 0xEC42E90Du;
-    hashValue ^= TransformedFeatureDimensions * 2;
+        hashValue ^= TransformedFeatureDimensions * 2;
 
-    hashValue = decltype(fc_0)::get_hash_value(hashValue);
+        hashValue = decltype(fc_0)::get_hash_value(hashValue);
-    hashValue = decltype(ac_0)::get_hash_value(hashValue);
+        hashValue = decltype(ac_0)::get_hash_value(hashValue);
-    hashValue = decltype(fc_1)::get_hash_value(hashValue);
+        hashValue = decltype(fc_1)::get_hash_value(hashValue);
-    hashValue = decltype(ac_1)::get_hash_value(hashValue);
+        hashValue = decltype(ac_1)::get_hash_value(hashValue);
-    hashValue = decltype(fc_2)::get_hash_value(hashValue);
+        hashValue = decltype(fc_2)::get_hash_value(hashValue);
 
-    return hashValue;
+        return hashValue;
-  }
+    }
 
-  // Read network parameters
+    // Read network parameters
-  bool read_parameters(std::istream& stream) {
+    bool read_parameters(std::istream& stream) {
-    return   fc_0.read_parameters(stream)
+        return fc_0.read_parameters(stream) && ac_0.read_parameters(stream)
-          && ac_0.read_parameters(stream)
+            && fc_1.read_parameters(stream) && ac_1.read_parameters(stream)
-          && fc_1.read_parameters(stream)
+            && fc_2.read_parameters(stream);
-          && ac_1.read_parameters(stream)
+    }
-          && fc_2.read_parameters(stream);
-  }
 
-  // Write network parameters
+    // Write network parameters
-  bool write_parameters(std::ostream& stream) const {
+    bool write_parameters(std::ostream& stream) const {
-    return   fc_0.write_parameters(stream)
+        return fc_0.write_parameters(stream) && ac_0.write_parameters(stream)
-          && ac_0.write_parameters(stream)
+            && fc_1.write_parameters(stream) && ac_1.write_parameters(stream)
-          && fc_1.write_parameters(stream)
+            && fc_2.write_parameters(stream);
-          && ac_1.write_parameters(stream)
+    }
-          && fc_2.write_parameters(stream);
-  }
 
-  std::int32_t propagate(const TransformedFeatureType* transformedFeatures)
+    std::int32_t propagate(const TransformedFeatureType* transformedFeatures) {
-  {
+        struct alignas(CacheLineSize) Buffer {
-    struct alignas(CacheLineSize) Buffer
+            alignas(CacheLineSize) typename decltype(fc_0)::OutputBuffer fc_0_out;
-    {
+            alignas(CacheLineSize) typename decltype(ac_sqr_0)::OutputType
-      alignas(CacheLineSize) decltype(fc_0)::OutputBuffer fc_0_out;
+              ac_sqr_0_out[ceil_to_multiple<IndexType>(FC_0_OUTPUTS * 2, 32)];
-      alignas(CacheLineSize) decltype(ac_sqr_0)::OutputType ac_sqr_0_out[ceil_to_multiple<IndexType>(FC_0_OUTPUTS * 2, 32)];
+            alignas(CacheLineSize) typename decltype(ac_0)::OutputBuffer ac_0_out;
-      alignas(CacheLineSize) decltype(ac_0)::OutputBuffer ac_0_out;
+            alignas(CacheLineSize) typename decltype(fc_1)::OutputBuffer fc_1_out;
-      alignas(CacheLineSize) decltype(fc_1)::OutputBuffer fc_1_out;
+            alignas(CacheLineSize) typename decltype(ac_1)::OutputBuffer ac_1_out;
-      alignas(CacheLineSize) decltype(ac_1)::OutputBuffer ac_1_out;
+            alignas(CacheLineSize) typename decltype(fc_2)::OutputBuffer fc_2_out;
-      alignas(CacheLineSize) decltype(fc_2)::OutputBuffer fc_2_out;
 
-      Buffer()
+            Buffer() { std::memset(this, 0, sizeof(*this)); }
-      {
+        };
-          std::memset(this, 0, sizeof(*this));
-      }
-    };
 
 #if defined(__clang__) && (__APPLE__)
-    // workaround for a bug reported with xcode 12
+        // workaround for a bug reported with xcode 12
-    static thread_local auto tlsBuffer = std::make_unique<Buffer>();
+        static thread_local auto tlsBuffer = std::make_unique<Buffer>();
-    // Access TLS only once, cache result.
+        // Access TLS only once, cache result.
-    Buffer& buffer = *tlsBuffer;
+        Buffer& buffer = *tlsBuffer;
 #else
-    alignas(CacheLineSize) static thread_local Buffer buffer;
+        alignas(CacheLineSize) static thread_local Buffer buffer;
 #endif
 
-    fc_0.propagate(transformedFeatures, buffer.fc_0_out);
+        fc_0.propagate(transformedFeatures, buffer.fc_0_out);
-    ac_sqr_0.propagate(buffer.fc_0_out, buffer.ac_sqr_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);
+        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));
+        std::memcpy(buffer.ac_sqr_0_out + FC_0_OUTPUTS, buffer.ac_0_out,
-    fc_1.propagate(buffer.ac_sqr_0_out, buffer.fc_1_out);
+                    FC_0_OUTPUTS * sizeof(typename decltype(ac_0)::OutputType));
-    ac_1.propagate(buffer.fc_1_out, buffer.ac_1_out);
+        fc_1.propagate(buffer.ac_sqr_0_out, buffer.fc_1_out);
-    fc_2.propagate(buffer.ac_1_out, buffer.fc_2_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
+        // buffer.fc_0_out[FC_0_OUTPUTS] is such that 1.0 is equal to 127*(1<<WeightScaleBits) in
-    // but we want 1.0 to be equal to 600*OutputScale
+        // 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 fwdOut =
-    std::int32_t outputValue = buffer.fc_2_out[0] + fwdOut;
+          (buffer.fc_0_out[FC_0_OUTPUTS]) * (600 * OutputScale) / (127 * (1 << WeightScaleBits));
+        std::int32_t outputValue = buffer.fc_2_out[0] + fwdOut;
 
-    return outputValue;
+        return outputValue;
-  }
+    }
 };
 
 }  // namespace Stockfish::Eval::NNUE
 
-#endif // #ifndef NNUE_ARCHITECTURE_H_INCLUDED
+#endif  // #ifndef NNUE_ARCHITECTURE_H_INCLUDED

src/nnue/nnue_common.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,221 +21,264 @@
 #ifndef NNUE_COMMON_H_INCLUDED
 #define NNUE_COMMON_H_INCLUDED
 
+#include <algorithm>
+#include <cassert>
+#include <cstdint>
 #include <cstring>
 #include <iostream>
+#include <type_traits>
 
-#include "../misc.h"  // for IsLittleEndian
+#include "../misc.h"
 
 #if defined(USE_AVX2)
-#include <immintrin.h>
+    #include <immintrin.h>
 
 #elif defined(USE_SSE41)
-#include <smmintrin.h>
+    #include <smmintrin.h>
 
 #elif defined(USE_SSSE3)
-#include <tmmintrin.h>
+    #include <tmmintrin.h>
 
 #elif defined(USE_SSE2)
-#include <emmintrin.h>
+    #include <emmintrin.h>
-
-#elif defined(USE_MMX)
-#include <mmintrin.h>
 
 #elif defined(USE_NEON)
-#include <arm_neon.h>
+    #include <arm_neon.h>
 #endif
 
 namespace Stockfish::Eval::NNUE {
 
-  // Version of the evaluation file
+// Version of the evaluation file
-  constexpr std::uint32_t Version = 0x7AF32F20u;
+constexpr std::uint32_t Version = 0x7AF32F20u;
 
-  // Constant used in evaluation value calculation
+// Constant used in evaluation value calculation
-  constexpr int OutputScale = 16;
+constexpr int OutputScale     = 16;
-  constexpr int WeightScaleBits = 6;
+constexpr int WeightScaleBits = 6;
 
-  // Size of cache line (in bytes)
+// Size of cache line (in bytes)
-  constexpr std::size_t CacheLineSize = 64;
+constexpr std::size_t CacheLineSize = 64;
 
-  constexpr const char Leb128MagicString[] = "COMPRESSED_LEB128";
+constexpr const char        Leb128MagicString[]   = "COMPRESSED_LEB128";
-  constexpr const std::size_t Leb128MagicStringSize = sizeof(Leb128MagicString) - 1;
+constexpr const std::size_t Leb128MagicStringSize = sizeof(Leb128MagicString) - 1;
 
-  // SIMD width (in bytes)
+// SIMD width (in bytes)
-  #if defined(USE_AVX2)
+#if defined(USE_AVX2)
-  constexpr std::size_t SimdWidth = 32;
+constexpr std::size_t SimdWidth = 32;
 
-  #elif defined(USE_SSE2)
+#elif defined(USE_SSE2)
-  constexpr std::size_t SimdWidth = 16;
+constexpr std::size_t SimdWidth = 16;
 
-  #elif defined(USE_MMX)
+#elif defined(USE_NEON)
-  constexpr std::size_t SimdWidth = 8;
+constexpr std::size_t SimdWidth = 16;
+#endif
 
-  #elif defined(USE_NEON)
+constexpr std::size_t MaxSimdWidth = 32;
-  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;
 
-  // Type of input feature after conversion
+// Round n up to be a multiple of base
-  using TransformedFeatureType = std::uint8_t;
+template<typename IntType>
-  using IndexType = std::uint32_t;
+constexpr IntType ceil_to_multiple(IntType n, IntType base) {
+    return (n + base - 1) / base * base;
+}
 
-  // 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)
+// 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
+// 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.
+// necessary to return a result with the byte ordering of the compiling machine.
-  template <typename IntType>
+template<typename IntType>
-  inline IntType read_little_endian(std::istream& stream) {
+inline IntType read_little_endian(std::istream& stream) {
-      IntType result;
+    IntType result;
 
-      if (IsLittleEndian)
+    if (IsLittleEndian)
-          stream.read(reinterpret_cast<char*>(&result), sizeof(IntType));
+        stream.read(reinterpret_cast<char*>(&result), sizeof(IntType));
-      else
+    else
-      {
+    {
-          std::uint8_t u[sizeof(IntType)];
+        std::uint8_t                  u[sizeof(IntType)];
-          typename std::make_unsigned<IntType>::type v = 0;
+        std::make_unsigned_t<IntType> v = 0;
 
-          stream.read(reinterpret_cast<char*>(u), sizeof(IntType));
+        stream.read(reinterpret_cast<char*>(u), sizeof(IntType));
-          for (std::size_t i = 0; i < sizeof(IntType); ++i)
+        for (std::size_t i = 0; i < sizeof(IntType); ++i)
-              v = (v << 8) | u[sizeof(IntType) - i - 1];
+            v = (v << 8) | u[sizeof(IntType) - i - 1];
 
-          std::memcpy(&result, &v, sizeof(IntType));
+        std::memcpy(&result, &v, sizeof(IntType));
-      }
+    }
 
-      return result;
+    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)
+// Utility to write an integer (signed or unsigned, any size)
-          stream.write(reinterpret_cast<const char*>(&value), sizeof(IntType));
+// to a stream in little-endian order. We swap the byte order before the write if
-      else
+// necessary to always write in little-endian order, independently of the byte
-      {
+// ordering of the compiling machine.
-          std::uint8_t u[sizeof(IntType)];
+template<typename IntType>
-          typename std::make_unsigned<IntType>::type v = value;
+inline void write_little_endian(std::ostream& stream, IntType value) {
 
-          std::size_t i = 0;
+    if (IsLittleEndian)
-          // if constexpr to silence the warning about shift by 8
+        stream.write(reinterpret_cast<const char*>(&value), sizeof(IntType));
-          if constexpr (sizeof(IntType) > 1)
+    else
-          {
+    {
+        std::uint8_t                  u[sizeof(IntType)];
+        std::make_unsigned_t<IntType> 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;
+                u[i] = std::uint8_t(v);
                 v >>= 8;
             }
-          }
+        }
-          u[i] = (std::uint8_t)v;
+        u[i] = std::uint8_t(v);
 
-          stream.write(reinterpret_cast<char*>(u), sizeof(IntType));
+        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.
+// Read integers in bulk from a little-endian stream.
-  // This takes N integers from array values and writes them on stream s.
+// This reads N integers from stream s and puts them in array out.
-  template <typename IntType>
+template<typename IntType>
-  inline void write_little_endian(std::ostream& stream, const IntType* values, std::size_t count) {
+inline void read_little_endian(std::istream& stream, IntType* out, std::size_t count) {
-      if (IsLittleEndian)
+    if (IsLittleEndian)
-          stream.write(reinterpret_cast<const char*>(values), sizeof(IntType) * count);
+        stream.read(reinterpret_cast<char*>(out), sizeof(IntType) * count);
-      else
+    else
-          for (std::size_t i = 0; i < count; ++i)
+        for (std::size_t i = 0; i < count; ++i)
-              write_little_endian<IntType>(stream, values[i]);
+            out[i] = read_little_endian<IntType>(stream);
-  }
+}
 
-  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>
+// Write integers in bulk to a little-endian stream.
-  inline void write_leb_128(std::ostream& stream, const IntType* values, std::size_t count) {
+// This takes N integers from array values and writes them on stream s.
-      static_assert(std::is_signed_v<IntType>, "Not implemented for unsigned types");
+template<typename IntType>
-      stream.write(Leb128MagicString, Leb128MagicStringSize);
+inline void write_little_endian(std::ostream& stream, const IntType* values, std::size_t count) {
-      std::uint32_t byte_count = 0;
+    if (IsLittleEndian)
-      for (std::size_t i = 0; i < count; ++i) {
+        stream.write(reinterpret_cast<const char*>(values), sizeof(IntType) * count);
-          IntType value = values[i];
+    else
-          std::uint8_t byte;
+        for (std::size_t i = 0; i < count; ++i)
-          do {
+            write_little_endian<IntType>(stream, values[i]);
-              byte = value & 0x7f;
+}
-              value >>= 7;
+
-              ++byte_count;
+
-          } while ((byte & 0x40) == 0 ? value != 0 : value != -1);
+// Read N signed integers from the stream s, putting them in the array out.
-      }
+// The stream is assumed to be compressed using the signed LEB128 format.
-      write_little_endian(stream, byte_count);
+// See https://en.wikipedia.org/wiki/LEB128 for a description of the compression scheme.
-      const std::uint32_t BUF_SIZE = 4096;
+template<typename IntType>
-      std::uint8_t buf[BUF_SIZE];
+inline void read_leb_128(std::istream& stream, IntType* out, std::size_t count) {
-      std::uint32_t buf_pos = 0;
+
-      auto flush = [&]() {
+    // Check the presence of our LEB128 magic string
-          if (buf_pos > 0) {
+    char leb128MagicString[Leb128MagicStringSize];
-              stream.write(reinterpret_cast<char*>(buf), buf_pos);
+    stream.read(leb128MagicString, Leb128MagicStringSize);
-              buf_pos = 0;
+    assert(strncmp(Leb128MagicString, leb128MagicString, Leb128MagicStringSize) == 0);
-          }
+
-      };
+    static_assert(std::is_signed_v<IntType>, "Not implemented for unsigned types");
-      auto write = [&](std::uint8_t byte) {
+
-          buf[buf_pos++] = byte;
+    const std::uint32_t BUF_SIZE = 4096;
-          if (buf_pos == BUF_SIZE) flush();
+    std::uint8_t        buf[BUF_SIZE];
-      };
+
-      for (std::size_t i = 0; i < count; ++i) {
+    auto bytes_left = read_little_endian<std::uint32_t>(stream);
-          IntType value = values[i];
+
-          while (true) {
+    std::uint32_t buf_pos = BUF_SIZE;
-              std::uint8_t byte = value & 0x7f;
+    for (std::size_t i = 0; i < count; ++i)
-              value >>= 7;
+    {
-              if ((byte & 0x40) == 0 ? value == 0 : value == -1) {
+        IntType result = 0;
-                  write(byte);
+        size_t  shift  = 0;
-                  break;
+        do
-              }
+        {
-              write(byte | 0x80);
+            if (buf_pos == BUF_SIZE)
-          }
+            {
-      }
+                stream.read(reinterpret_cast<char*>(buf), std::min(bytes_left, BUF_SIZE));
-      flush();
+                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);
+}
+
+
+// Write signed integers to a stream with LEB128 compression.
+// This takes N integers from array values, compresses them with
+// the LEB128 algorithm and writes the result on the stream s.
+// See https://en.wikipedia.org/wiki/LEB128 for a description of the compression scheme.
+template<typename IntType>
+inline void write_leb_128(std::ostream& stream, const IntType* values, std::size_t count) {
+
+    // Write our LEB128 magic string
+    stream.write(Leb128MagicString, Leb128MagicStringSize);
+
+    static_assert(std::is_signed_v<IntType>, "Not implemented for unsigned types");
+
+    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
+#endif  // #ifndef NNUE_COMMON_H_INCLUDED

src/nnue/nnue_misc.cpp

@@ -0,0 +1,203 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 "nnue_misc.h"
+
+#include <cmath>
+#include <cstdlib>
+#include <cstring>
+#include <iomanip>
+#include <iosfwd>
+#include <iostream>
+#include <sstream>
+#include <string_view>
+#include <tuple>
+
+#include "../evaluate.h"
+#include "../position.h"
+#include "../types.h"
+#include "../uci.h"
+#include "network.h"
+#include "nnue_accumulator.h"
+
+namespace Stockfish::Eval::NNUE {
+
+
+constexpr std::string_view PieceToChar(" PNBRQK  pnbrqk");
+
+
+void hint_common_parent_position(const Position&    pos,
+                                 const Networks&    networks,
+                                 AccumulatorCaches& caches) {
+    if (Eval::use_smallnet(pos))
+        networks.small.hint_common_access(pos, &caches.small);
+    else
+        networks.big.hint_common_access(pos, &caches.big);
+}
+
+namespace {
+// Converts a Value into (centi)pawns and writes it in a buffer.
+// The buffer must have capacity for at least 5 chars.
+void format_cp_compact(Value v, char* buffer, const Position& pos) {
+
+    buffer[0] = (v < 0 ? '-' : v > 0 ? '+' : ' ');
+
+    int cp = std::abs(UCIEngine::to_cp(v, pos));
+    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;
+    }
+}
+
+
+// Converts a Value into pawns, always keeping two decimals
+void format_cp_aligned_dot(Value v, std::stringstream& stream, const Position& pos) {
+
+    const double pawns = std::abs(0.01 * UCIEngine::to_cp(v, pos));
+
+    stream << (v < 0   ? '-'
+               : v > 0 ? '+'
+                       : ' ')
+           << std::setiosflags(std::ios::fixed) << std::setw(6) << std::setprecision(2) << pawns;
+}
+}
+
+
+// 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, const Eval::NNUE::Networks& networks, Eval::NNUE::AccumulatorCaches& caches) {
+
+    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, &pos](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 (is_valid(value))
+            format_cp_compact(value, &board[y + 2][x + 2], pos);
+    };
+
+    // 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.
+    auto [psqt, positional] = networks.big.evaluate(pos, &caches.big);
+    Value base              = psqt + positional;
+    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->accumulatorBig.computed[WHITE] = st->accumulatorBig.computed[BLACK] = false;
+
+                std::tie(psqt, positional) = networks.big.evaluate(pos, &caches.big);
+                Value eval                 = psqt + positional;
+                eval                       = pos.side_to_move() == WHITE ? eval : -eval;
+                v                          = base - eval;
+
+                pos.put_piece(pc, sq);
+                st->accumulatorBig.computed[WHITE] = st->accumulatorBig.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 = networks.big.trace_evaluate(pos, &caches.big);
+
+    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 << "        "  //
+           << " |  ";
+        format_cp_aligned_dot(t.psqt[bucket], ss, pos);
+        ss << "  "  //
+           << " |  ";
+        format_cp_aligned_dot(t.positional[bucket], ss, pos);
+        ss << "  "  //
+           << " |  ";
+        format_cp_aligned_dot(t.psqt[bucket] + t.positional[bucket], ss, pos);
+        ss << "  "  //
+           << " |";
+        if (bucket == t.correctBucket)
+            ss << " <-- this bucket is used";
+        ss << '\n';
+    }
+
+    ss << "+------------+------------+------------+------------+\n";
+
+    return ss.str();
+}
+
+
+}  // namespace Stockfish::Eval::NNUE

src/nnue/nnue_misc.h

@@ -0,0 +1,64 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 NNUE_MISC_H_INCLUDED
+#define NNUE_MISC_H_INCLUDED
+
+#include <cstddef>
+#include <string>
+
+#include "../types.h"
+#include "nnue_architecture.h"
+
+namespace Stockfish {
+
+class Position;
+
+namespace Eval::NNUE {
+
+struct EvalFile {
+    // Default net name, will use one of the EvalFileDefaultName* macros defined
+    // in evaluate.h
+    std::string defaultName;
+    // Selected net name, either via uci option or default
+    std::string current;
+    // Net description extracted from the net file
+    std::string netDescription;
+};
+
+
+struct NnueEvalTrace {
+    static_assert(LayerStacks == PSQTBuckets);
+
+    Value       psqt[LayerStacks];
+    Value       positional[LayerStacks];
+    std::size_t correctBucket;
+};
+
+struct Networks;
+struct AccumulatorCaches;
+
+std::string trace(Position& pos, const Networks& networks, AccumulatorCaches& caches);
+void        hint_common_parent_position(const Position&    pos,
+                                        const Networks&    networks,
+                                        AccumulatorCaches& caches);
+
+}  // namespace Stockfish::Eval::NNUE
+}  // namespace Stockfish
+
+#endif  // #ifndef NNUE_MISC_H_INCLUDED

src/pawns.cpp

@@ -1,305 +0,0 @@
-/*
-  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 <cassert>
-
-#include "bitboard.h"
-#include "pawns.h"
-#include "position.h"
-#include "thread.h"
-
-namespace Stockfish {
-
-namespace {
-
-  #define V Value
-  #define S(mg, eg) make_score(mg, eg)
-
-  // Pawn penalties
-  constexpr Score Backward      = S( 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);
-
-  // Bonus for blocked pawns at 5th or 6th rank
-  constexpr Score BlockedPawn[2] = { S(-19, -8), S(-7, 3) };
-
-  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)
-  };
-
-  // 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) }
-  };
-
-  // 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) }
-  };
-
-
-  // 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
-
-
-  /// 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.
-
-  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();
-  Entry* e = pos.this_thread()->pawnsTable[key];
-
-  if (e->key == key)
-      return e;
-
-  e->key = key;
-  e->blockedCount = 0;
-  e->scores[WHITE] = evaluate<WHITE>(pos, e);
-  e->scores[BLACK] = evaluate<BLACK>(pos, e);
-
-  return e;
-}
-
-
-/// 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 Entry::evaluate_shelter(const Position& pos, Square ksq) const {
-
-  constexpr Color Them = ~Us;
-
-  Bitboard b = pos.pieces(PAWN) & ~forward_ranks_bb(Them, ksq);
-  Bitboard ourPawns = b & pos.pieces(Us) & ~pawnAttacks[Them];
-  Bitboard theirPawns = b & pos.pieces(Them);
-
-  Score bonus = make_score(5, 5);
-
-  File center = std::clamp(file_of(ksq), FILE_B, FILE_G);
-  for (File f = File(center - 1); f <= File(center + 1); ++f)
-  {
-      b = ourPawns & file_bb(f);
-      int ourRank = b ? relative_rank(Us, frontmost_sq(Them, b)) : 0;
-
-      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);
-  }
-
-  // King On File
-  bonus -= KingOnFile[pos.is_on_semiopen_file(Us, ksq)][pos.is_on_semiopen_file(Them, ksq)];
-
-  return bonus;
-}
-
-
-/// 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 Entry::do_king_safety(const Position& pos) {
-
-  Square ksq = pos.square<KING>(Us);
-  kingSquares[Us] = ksq;
-  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);
-  int minPawnDist = 6;
-
-  if (pawns & attacks_bb<KING>(ksq))
-      minPawnDist = 1;
-  else while (pawns)
-      minPawnDist = std::min(minPawnDist, distance(ksq, pop_lsb(pawns)));
-
-  return shelter - make_score(0, 16 * minPawnDist);
-}
-
-// Explicit template instantiation
-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

@@ -1,70 +0,0 @@
-/*
-  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 PAWNS_H_INCLUDED
-#define PAWNS_H_INCLUDED
-
-#include "misc.h"
-#include "position.h"
-#include "types.h"
-
-namespace Stockfish::Pawns {
-
-/// Pawns::Entry contains various information about a pawn structure. A lookup
-/// to the pawn hash table (performed by calling the probe function) returns a
-/// pointer to an Entry object.
-
-struct Entry {
-
-  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) {
-    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>
-  Score do_king_safety(const Position& pos);
-
-  template<Color Us>
-  Score evaluate_shelter(const Position& pos, Square ksq) const;
-
-  Key key;
-  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>;
-
-Entry* probe(const Position& pos);
-
-} // namespace Stockfish::Pawns
-
-#endif // #ifndef PAWNS_H_INCLUDED

src/perft.h

@@ -0,0 +1,68 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 PERFT_H_INCLUDED
+#define PERFT_H_INCLUDED
+
+#include <cstdint>
+
+#include "movegen.h"
+#include "position.h"
+#include "types.h"
+#include "uci.h"
+
+namespace Stockfish::Benchmark {
+
+// Utility to verify move generation. All the leaf nodes up
+// to the given depth are generated and counted, and the sum is returned.
+template<bool Root>
+uint64_t perft(Position& pos, Depth depth) {
+
+    StateInfo st;
+    ASSERT_ALIGNED(&st, Eval::NNUE::CacheLineSize);
+
+    uint64_t   cnt, nodes = 0;
+    const bool leaf = (depth == 2);
+
+    for (const auto& m : MoveList<LEGAL>(pos))
+    {
+        if (Root && depth <= 1)
+            cnt = 1, nodes++;
+        else
+        {
+            pos.do_move(m, st);
+            cnt = leaf ? MoveList<LEGAL>(pos).size() : perft<false>(pos, depth - 1);
+            nodes += cnt;
+            pos.undo_move(m);
+        }
+        if (Root)
+            sync_cout << UCIEngine::move(m, pos.is_chess960()) << ": " << cnt << sync_endl;
+    }
+    return nodes;
+}
+
+inline uint64_t perft(const std::string& fen, Depth depth, bool isChess960) {
+    StateListPtr states(new std::deque<StateInfo>(1));
+    Position     p;
+    p.set(fen, isChess960, &states->back());
+
+    return perft<true>(p, depth);
+}
+}
+
+#endif  // PERFT_H_INCLUDED

src/position.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,430 +21,356 @@
 
 #include <cassert>
 #include <deque>
-#include <memory> // For std::unique_ptr
+#include <iosfwd>
+#include <memory>
 #include <string>
 
 #include "bitboard.h"
-#include "evaluate.h"
-#include "psqt.h"
-#include "types.h"
-
 #include "nnue/nnue_accumulator.h"
+#include "nnue/nnue_architecture.h"
+#include "types.h"
 
 namespace Stockfish {
 
-/// StateInfo struct stores information needed to restore a Position object to
+class TranspositionTable;
-/// 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.
+// 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 {
 
-  // Copied when making a move
+    // Copied when making a move
-  Key    pawnKey;
+    Key    materialKey;
-  Key    materialKey;
+    Key    pawnKey;
-  Value  nonPawnMaterial[COLOR_NB];
+    Key    minorPieceKey;
-  int    castlingRights;
+    Key    nonPawnKey[COLOR_NB];
-  int    rule50;
+    Value  nonPawnMaterial[COLOR_NB];
-  int    pliesFromNull;
+    int    castlingRights;
-  Square epSquare;
+    int    rule50;
+    int    pliesFromNull;
+    Square epSquare;
 
-  // Not copied when making a move (will be recomputed anyhow)
+    // Not copied when making a move (will be recomputed anyhow)
-  Key        key;
+    Key        key;
-  Bitboard   checkersBB;
+    Bitboard   checkersBB;
-  StateInfo* previous;
+    StateInfo* previous;
-  Bitboard   blockersForKing[COLOR_NB];
+    StateInfo* next;
-  Bitboard   pinners[COLOR_NB];
+    Bitboard   blockersForKing[COLOR_NB];
-  Bitboard   checkSquares[PIECE_TYPE_NB];
+    Bitboard   pinners[COLOR_NB];
-  Piece      capturedPiece;
+    Bitboard   checkSquares[PIECE_TYPE_NB];
-  int        repetition;
+    Piece      capturedPiece;
+    int        repetition;
 
-  // Used by NNUE
+    // Used by NNUE
-  Eval::NNUE::Accumulator accumulator;
+    Eval::NNUE::Accumulator<Eval::NNUE::TransformedFeatureDimensionsBig>   accumulatorBig;
-  DirtyPiece dirtyPiece;
+    Eval::NNUE::Accumulator<Eval::NNUE::TransformedFeatureDimensionsSmall> accumulatorSmall;
+    DirtyPiece                                                             dirtyPiece;
 };
 
 
-/// A list to keep track of the position states along the setup moves (from the
+// 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
+// start position to the position just before the search starts). Needed by
-/// 'draw by repetition' detection. Use a std::deque because pointers to
+// 'draw by repetition' detection. Use a std::deque because pointers to
-/// elements are not invalidated upon list resizing.
+// elements are not invalidated upon list resizing.
 using StateListPtr = std::unique_ptr<std::deque<StateInfo>>;
 
 
-/// Position class stores information regarding the board representation as
+// Position class stores information regarding the board representation as
-/// pieces, side to move, hash keys, castling info, etc. Important methods are
+// 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
+// do_move() and undo_move(), used by the search to update node info when
-/// traversing the search tree.
+// traversing the search tree.
-class Thread;
-
 class Position {
-public:
+   public:
-  static void init();
+    static void init();
 
-  Position() = default;
+    Position()                           = default;
-  Position(const Position&) = delete;
+    Position(const Position&)            = delete;
-  Position& operator=(const Position&) = delete;
+    Position& operator=(const Position&) = delete;
 
-  // FEN string input/output
+    // FEN string input/output
-  Position& set(const std::string& fenStr, bool isChess960, StateInfo* si, Thread* th);
+    Position&   set(const std::string& fenStr, bool isChess960, StateInfo* si);
-  Position& set(const std::string& code, Color c, StateInfo* si);
+    Position&   set(const std::string& code, Color c, StateInfo* si);
-  std::string fen() const;
+    std::string fen() const;
 
-  // Position representation
+    // Position representation
-  Bitboard pieces(PieceType pt) const;
+    Bitboard pieces(PieceType pt = ALL_PIECES) const;
-  template<typename ...PieceTypes> Bitboard pieces(PieceType pt, PieceTypes... pts) const;
+    template<typename... PieceTypes>
-  Bitboard pieces(Color c) const;
+    Bitboard pieces(PieceType pt, PieceTypes... pts) const;
-  template<typename ...PieceTypes> Bitboard pieces(Color c, PieceTypes... pts) const;
+    Bitboard pieces(Color c) const;
-  Piece piece_on(Square s) const;
+    template<typename... PieceTypes>
-  Square ep_square() const;
+    Bitboard pieces(Color c, PieceTypes... pts) const;
-  bool empty(Square s) const;
+    Piece    piece_on(Square s) const;
-  template<PieceType Pt> int count(Color c) const;
+    Square   ep_square() const;
-  template<PieceType Pt> int count() const;
+    bool     empty(Square s) const;
-  template<PieceType Pt> Square square(Color c) const;
+    template<PieceType Pt>
-  bool is_on_semiopen_file(Color c, Square s) const;
+    int count(Color c) const;
+    template<PieceType Pt>
+    int count() const;
+    template<PieceType Pt>
+    Square square(Color c) const;
 
-  // Castling
+    // Castling
-  CastlingRights castling_rights(Color c) const;
+    CastlingRights castling_rights(Color c) const;
-  bool can_castle(CastlingRights cr) const;
+    bool           can_castle(CastlingRights cr) const;
-  bool castling_impeded(CastlingRights cr) const;
+    bool           castling_impeded(CastlingRights cr) const;
-  Square castling_rook_square(CastlingRights cr) const;
+    Square         castling_rook_square(CastlingRights cr) const;
 
-  // Checking
+    // Checking
-  Bitboard checkers() const;
+    Bitboard checkers() const;
-  Bitboard blockers_for_king(Color c) const;
+    Bitboard blockers_for_king(Color c) const;
-  Bitboard check_squares(PieceType pt) const;
+    Bitboard check_squares(PieceType pt) const;
-  Bitboard pinners(Color c) const;
+    Bitboard pinners(Color c) const;
 
-  // Attacks to/from a given square
+    // Attacks to/from a given square
-  Bitboard attackers_to(Square s) const;
+    Bitboard attackers_to(Square s) const;
-  Bitboard attackers_to(Square s, Bitboard occupied) const;
+    Bitboard attackers_to(Square s, Bitboard occupied) const;
-  Bitboard slider_blockers(Bitboard sliders, Square s, Bitboard& pinners) const;
+    bool     attackers_to_exist(Square s, Bitboard occupied, Color c) const;
-  template<PieceType Pt> Bitboard attacks_by(Color c) const;
+    void     update_slider_blockers(Color c) const;
+    template<PieceType Pt>
+    Bitboard attacks_by(Color c) const;
 
-  // Properties of moves
+    // Properties of moves
-  bool legal(Move m) const;
+    bool  legal(Move m) const;
-  bool pseudo_legal(const Move m) const;
+    bool  pseudo_legal(const Move m) const;
-  bool capture(Move m) const;
+    bool  capture(Move m) const;
-  bool capture_stage(Move m) const;
+    bool  capture_stage(Move m) const;
-  bool gives_check(Move m) const;
+    bool  gives_check(Move m) const;
-  Piece moved_piece(Move m) const;
+    Piece moved_piece(Move m) const;
-  Piece captured_piece() const;
+    Piece captured_piece() const;
 
-  // Piece specific
+    // Doing and undoing moves
-  bool pawn_passed(Color c, Square s) const;
+    void do_move(Move m, StateInfo& newSt, const TranspositionTable* tt);
-  bool opposite_bishops() const;
+    void do_move(Move m, StateInfo& newSt, bool givesCheck, const TranspositionTable* tt);
-  int  pawns_on_same_color_squares(Color c, Square s) const;
+    void undo_move(Move m);
+    void do_null_move(StateInfo& newSt, const TranspositionTable& tt);
+    void undo_null_move();
 
-  // Doing and undoing moves
+    // Static Exchange Evaluation
-  void do_move(Move m, StateInfo& newSt);
+    bool see_ge(Move m, int threshold = 0) const;
-  void do_move(Move m, StateInfo& newSt, bool givesCheck);
-  void undo_move(Move m);
-  void do_null_move(StateInfo& newSt);
-  void undo_null_move();
 
-  // Static Exchange Evaluation
+    // Accessing hash keys
-  bool see_ge(Move m, Value threshold = VALUE_ZERO) const;
+    Key key() const;
-  bool see_ge(Move m, Bitboard& occupied, Value threshold = VALUE_ZERO) const;
+    Key material_key() const;
+    Key pawn_key() const;
+    Key minor_piece_key() const;
+    Key non_pawn_key(Color c) const;
 
-  // Accessing hash keys
+    // Other properties of the position
-  Key key() const;
+    Color side_to_move() const;
-  Key key_after(Move m) const;
+    int   game_ply() const;
-  Key material_key() const;
+    bool  is_chess960() const;
-  Key pawn_key() const;
+    bool  is_draw(int ply) const;
+    bool  is_repetition(int ply) const;
+    bool  upcoming_repetition(int ply) const;
+    bool  has_repeated() const;
+    int   rule50_count() const;
+    Value non_pawn_material(Color c) const;
+    Value non_pawn_material() const;
 
-  // Other properties of the position
+    // Position consistency check, for debugging
-  Color side_to_move() const;
+    bool pos_is_ok() const;
-  int game_ply() const;
+    void flip();
-  bool is_chess960() const;
-  Thread* this_thread() 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
+    // Used by NNUE
-  bool pos_is_ok() const;
+    StateInfo* state() const;
-  void flip();
 
-  // Used by NNUE
+    void put_piece(Piece pc, Square s);
-  StateInfo* state() const;
+    void remove_piece(Square s);
 
-  void put_piece(Piece pc, Square s);
+   private:
-  void remove_piece(Square s);
+    // Initialization helpers (used while setting up a position)
+    void set_castling_right(Color c, Square rfrom);
+    void set_state() const;
+    void set_check_info() const;
 
-private:
+    // Other helpers
-  // Initialization helpers (used while setting up a position)
+    void move_piece(Square from, Square to);
-  void set_castling_right(Color c, Square rfrom);
+    template<bool Do>
-  void set_state() const;
+    void do_castling(Color us, Square from, Square& to, Square& rfrom, Square& rto);
-  void set_check_info() const;
+    template<bool AfterMove>
+    Key adjust_key50(Key k) const;
 
-  // Other helpers
+    // Data members
-  void move_piece(Square from, Square to);
+    Piece      board[SQUARE_NB];
-  template<bool Do>
+    Bitboard   byTypeBB[PIECE_TYPE_NB];
-  void do_castling(Color us, Square from, Square& to, Square& rfrom, Square& rto);
+    Bitboard   byColorBB[COLOR_NB];
-  template<bool AfterMove>
+    int        pieceCount[PIECE_NB];
-  Key adjust_key50(Key k) const;
+    int        castlingRightsMask[SQUARE_NB];
-
+    Square     castlingRookSquare[CASTLING_RIGHT_NB];
-  // Data members
+    Bitboard   castlingPath[CASTLING_RIGHT_NB];
-  Piece board[SQUARE_NB];
+    StateInfo* st;
-  Bitboard byTypeBB[PIECE_TYPE_NB];
+    int        gamePly;
-  Bitboard byColorBB[COLOR_NB];
+    Color      sideToMove;
-  int pieceCount[PIECE_NB];
+    bool       chess960;
-  int castlingRightsMask[SQUARE_NB];
-  Square castlingRookSquare[CASTLING_RIGHT_NB];
-  Bitboard castlingPath[CASTLING_RIGHT_NB];
-  Thread* thisThread;
-  StateInfo* st;
-  int gamePly;
-  Color sideToMove;
-  Score psq;
-  bool chess960;
 };
 
 std::ostream& operator<<(std::ostream& os, const Position& pos);
 
-inline Color Position::side_to_move() const {
+inline Color Position::side_to_move() const { return sideToMove; }
-  return sideToMove;
-}
 
 inline Piece Position::piece_on(Square s) const {
-  assert(is_ok(s));
+    assert(is_ok(s));
-  return board[s];
+    return board[s];
 }
 
-inline bool Position::empty(Square s) const {
+inline bool Position::empty(Square s) const { return piece_on(s) == NO_PIECE; }
-  return piece_on(s) == NO_PIECE;
-}
 
-inline Piece Position::moved_piece(Move m) const {
+inline Piece Position::moved_piece(Move m) const { return piece_on(m.from_sq()); }
-  return piece_on(from_sq(m));
-}
 
-inline Bitboard Position::pieces(PieceType pt = ALL_PIECES) const {
+inline Bitboard Position::pieces(PieceType pt) const { return byTypeBB[pt]; }
-  return byTypeBB[pt];
-}
 
-template<typename ...PieceTypes>
+template<typename... PieceTypes>
 inline Bitboard Position::pieces(PieceType pt, PieceTypes... pts) const {
-  return pieces(pt) | pieces(pts...);
+    return pieces(pt) | pieces(pts...);
 }
 
-inline Bitboard Position::pieces(Color c) const {
+inline Bitboard Position::pieces(Color c) const { return byColorBB[c]; }
-  return byColorBB[c];
-}
 
-template<typename ...PieceTypes>
+template<typename... PieceTypes>
 inline Bitboard Position::pieces(Color c, PieceTypes... pts) const {
-  return pieces(c) & pieces(pts...);
+    return pieces(c) & pieces(pts...);
 }
 
-template<PieceType Pt> inline int Position::count(Color c) const {
+template<PieceType Pt>
-  return pieceCount[make_piece(c, Pt)];
+inline int Position::count(Color c) const {
+    return pieceCount[make_piece(c, Pt)];
 }
 
-template<PieceType Pt> inline int Position::count() const {
+template<PieceType Pt>
-  return count<Pt>(WHITE) + count<Pt>(BLACK);
+inline int Position::count() const {
+    return count<Pt>(WHITE) + count<Pt>(BLACK);
 }
 
-template<PieceType Pt> inline Square Position::square(Color c) const {
+template<PieceType Pt>
-  assert(count<Pt>(c) == 1);
+inline Square Position::square(Color c) const {
-  return lsb(pieces(c, Pt));
+    assert(count<Pt>(c) == 1);
+    return lsb(pieces(c, Pt));
 }
 
-inline Square Position::ep_square() const {
+inline Square Position::ep_square() const { return st->epSquare; }
-  return st->epSquare;
-}
 
-inline bool Position::is_on_semiopen_file(Color c, Square s) const {
+inline bool Position::can_castle(CastlingRights cr) const { return st->castlingRights & cr; }
-  return !(pieces(c, PAWN) & file_bb(s));
-}
-
-inline bool Position::can_castle(CastlingRights cr) const {
-  return st->castlingRights & cr;
-}
 
 inline CastlingRights Position::castling_rights(Color c) const {
-  return c & CastlingRights(st->castlingRights);
+    return c & CastlingRights(st->castlingRights);
 }
 
 inline bool Position::castling_impeded(CastlingRights cr) const {
-  assert(cr == WHITE_OO || cr == WHITE_OOO || cr == BLACK_OO || cr == BLACK_OOO);
+    assert(cr == WHITE_OO || cr == WHITE_OOO || cr == BLACK_OO || cr == BLACK_OOO);
-
+    return pieces() & castlingPath[cr];
-  return pieces() & castlingPath[cr];
 }
 
 inline Square Position::castling_rook_square(CastlingRights cr) const {
-  assert(cr == WHITE_OO || cr == WHITE_OOO || cr == BLACK_OO || cr == BLACK_OOO);
+    assert(cr == WHITE_OO || cr == WHITE_OOO || cr == BLACK_OO || cr == BLACK_OOO);
-
+    return castlingRookSquare[cr];
-  return castlingRookSquare[cr];
 }
 
-inline Bitboard Position::attackers_to(Square s) const {
+inline Bitboard Position::attackers_to(Square s) const { return attackers_to(s, pieces()); }
-  return attackers_to(s, pieces());
-}
 
 template<PieceType Pt>
 inline Bitboard Position::attacks_by(Color c) const {
 
-  if constexpr (Pt == PAWN)
+    if constexpr (Pt == PAWN)
-      return c == WHITE ? pawn_attacks_bb<WHITE>(pieces(WHITE, PAWN))
+        return c == WHITE ? pawn_attacks_bb<WHITE>(pieces(WHITE, PAWN))
-                        : pawn_attacks_bb<BLACK>(pieces(BLACK, PAWN));
+                          : pawn_attacks_bb<BLACK>(pieces(BLACK, PAWN));
-  else
+    else
-  {
+    {
-      Bitboard threats = 0;
+        Bitboard threats   = 0;
-      Bitboard attackers = pieces(c, Pt);
+        Bitboard attackers = pieces(c, Pt);
-      while (attackers)
+        while (attackers)
-          threats |= attacks_bb<Pt>(pop_lsb(attackers), pieces());
+            threats |= attacks_bb<Pt>(pop_lsb(attackers), pieces());
-      return threats;
+        return threats;
-  }
+    }
 }
 
-inline Bitboard Position::checkers() const {
+inline Bitboard Position::checkers() const { return st->checkersBB; }
-  return st->checkersBB;
-}
 
-inline Bitboard Position::blockers_for_king(Color c) const {
+inline Bitboard Position::blockers_for_king(Color c) const { return st->blockersForKing[c]; }
-  return st->blockersForKing[c];
-}
 
-inline Bitboard Position::pinners(Color c) const {
+inline Bitboard Position::pinners(Color c) const { return st->pinners[c]; }
-  return st->pinners[c];
-}
 
-inline Bitboard Position::check_squares(PieceType pt) const {
+inline Bitboard Position::check_squares(PieceType pt) const { return st->checkSquares[pt]; }
-  return st->checkSquares[pt];
-}
 
-inline bool Position::pawn_passed(Color c, Square s) const {
+inline Key Position::key() const { return adjust_key50<false>(st->key); }
-  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 adjust_key50<false>(st->key);
-}
 
 template<bool AfterMove>
-inline Key Position::adjust_key50(Key k) const
+inline Key Position::adjust_key50(Key k) const {
-{
+    return st->rule50 < 14 - AfterMove ? k : k ^ make_key((st->rule50 - (14 - AfterMove)) / 8);
-  return st->rule50 < 14 - AfterMove
-      ? k : k ^ make_key((st->rule50 - (14 - AfterMove)) / 8);
 }
 
-inline Key Position::pawn_key() const {
+inline Key Position::pawn_key() const { return st->pawnKey; }
-  return st->pawnKey;
-}
 
-inline Key Position::material_key() const {
+inline Key Position::material_key() const { return st->materialKey; }
-  return st->materialKey;
-}
 
-inline Score Position::psq_score() const {
+inline Key Position::minor_piece_key() const { return st->minorPieceKey; }
-  return psq;
-}
 
-inline Value Position::psq_eg_stm() const {
+inline Key Position::non_pawn_key(Color c) const { return st->nonPawnKey[c]; }
-  return (sideToMove == WHITE ? 1 : -1) * eg_value(psq);
-}
 
-inline Value Position::non_pawn_material(Color c) const {
+inline Value Position::non_pawn_material(Color c) const { return st->nonPawnMaterial[c]; }
-  return st->nonPawnMaterial[c];
-}
 
 inline Value Position::non_pawn_material() const {
-  return non_pawn_material(WHITE) + non_pawn_material(BLACK);
+    return non_pawn_material(WHITE) + non_pawn_material(BLACK);
 }
 
-inline int Position::game_ply() const {
+inline int Position::game_ply() const { return gamePly; }
-  return gamePly;
-}
 
-inline int Position::rule50_count() const {
+inline int Position::rule50_count() const { return st->rule50; }
-  return st->rule50;
-}
 
-inline bool Position::opposite_bishops() const {
+inline bool Position::is_chess960() const { return chess960; }
-  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::capture(Move m) const {
-  assert(is_ok(m));
+    assert(m.is_ok());
-  return     (!empty(to_sq(m)) && type_of(m) != CASTLING)
+    return (!empty(m.to_sq()) && m.type_of() != CASTLING) || m.type_of() == EN_PASSANT;
-          ||  type_of(m) == EN_PASSANT;
 }
 
-// returns true if a move is generated from the capture stage
+// Returns true if a move is generated from the capture stage, having also
-// having also queen promotions covered, i.e. consistency with the capture stage move generation
+// queen promotions covered, i.e. consistency with the capture stage move
-// is needed to avoid the generation of duplicate moves.
+// generation is needed to avoid the generation of duplicate moves.
 inline bool Position::capture_stage(Move m) const {
-  assert(is_ok(m));
+    assert(m.is_ok());
-  return  capture(m) || promotion_type(m) == QUEEN;
+    return capture(m) || m.promotion_type() == QUEEN;
 }
 
-inline Piece Position::captured_piece() const {
+inline Piece Position::captured_piece() const { return st->capturedPiece; }
-  return st->capturedPiece;
-}
-
-inline Thread* Position::this_thread() const {
-  return thisThread;
-}
 
 inline void Position::put_piece(Piece pc, Square s) {
 
-  board[s] = pc;
+    board[s] = pc;
-  byTypeBB[ALL_PIECES] |= byTypeBB[type_of(pc)] |= s;
+    byTypeBB[ALL_PIECES] |= byTypeBB[type_of(pc)] |= s;
-  byColorBB[color_of(pc)] |= s;
+    byColorBB[color_of(pc)] |= s;
-  pieceCount[pc]++;
+    pieceCount[pc]++;
-  pieceCount[make_piece(color_of(pc), ALL_PIECES)]++;
+    pieceCount[make_piece(color_of(pc), ALL_PIECES)]++;
-  psq += PSQT::psq[pc][s];
 }
 
 inline void Position::remove_piece(Square s) {
 
-  Piece pc = board[s];
+    Piece pc = board[s];
-  byTypeBB[ALL_PIECES] ^= s;
+    byTypeBB[ALL_PIECES] ^= s;
-  byTypeBB[type_of(pc)] ^= s;
+    byTypeBB[type_of(pc)] ^= s;
-  byColorBB[color_of(pc)] ^= s;
+    byColorBB[color_of(pc)] ^= s;
-  board[s] = NO_PIECE;
+    board[s] = NO_PIECE;
-  pieceCount[pc]--;
+    pieceCount[pc]--;
-  pieceCount[make_piece(color_of(pc), ALL_PIECES)]--;
+    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];
+    Piece    pc     = board[from];
-  Bitboard fromTo = from | to;
+    Bitboard fromTo = from | to;
-  byTypeBB[ALL_PIECES] ^= fromTo;
+    byTypeBB[ALL_PIECES] ^= fromTo;
-  byTypeBB[type_of(pc)] ^= fromTo;
+    byTypeBB[type_of(pc)] ^= fromTo;
-  byColorBB[color_of(pc)] ^= fromTo;
+    byColorBB[color_of(pc)] ^= fromTo;
-  board[from] = NO_PIECE;
+    board[from] = NO_PIECE;
-  board[to] = pc;
+    board[to]   = pc;
-  psq += PSQT::psq[pc][to] - PSQT::psq[pc][from];
 }
 
-inline void Position::do_move(Move m, StateInfo& newSt) {
+inline void Position::do_move(Move m, StateInfo& newSt, const TranspositionTable* tt = nullptr) {
-  do_move(m, newSt, gives_check(m));
+    do_move(m, newSt, gives_check(m), tt);
 }
 
-inline StateInfo* Position::state() const {
+inline StateInfo* Position::state() const { return st; }
 
-  return st;
+}  // namespace Stockfish
-}
 
-} // namespace Stockfish
+#endif  // #ifndef POSITION_H_INCLUDED
-
-#endif // #ifndef POSITION_H_INCLUDED

src/psqt.cpp

@@ -1,131 +0,0 @@
-/*
-  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/score.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,23 +16,33 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#include "score.h"
 
-#ifndef PSQT_H_INCLUDED
+#include <cassert>
-#define PSQT_H_INCLUDED
+#include <cmath>
+#include <cstdlib>
 
+#include "uci.h"
 
-#include "types.h"
+namespace Stockfish {
 
+Score::Score(Value v, const Position& pos) {
+    assert(-VALUE_INFINITE < v && v < VALUE_INFINITE);
 
-namespace Stockfish::PSQT
+    if (!is_decisive(v))
-{
+    {
+        score = InternalUnits{UCIEngine::to_cp(v, pos)};
+    }
+    else if (std::abs(v) <= VALUE_TB)
+    {
+        auto distance = VALUE_TB - std::abs(v);
+        score         = (v > 0) ? Tablebase{distance, true} : Tablebase{-distance, false};
+    }
+    else
+    {
+        auto distance = VALUE_MATE - std::abs(v);
+        score         = (v > 0) ? Mate{distance} : Mate{-distance};
+    }
+}
 
-extern Score psq[PIECE_NB][SQUARE_NB];
+}
-
-// Fill psqt array from a set of internally linked parameters
-void init();
-
-} // namespace Stockfish::PSQT
-
-
-#endif // PSQT_H_INCLUDED

src/score.h

@@ -0,0 +1,70 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 SCORE_H_INCLUDED
+#define SCORE_H_INCLUDED
+
+#include <variant>
+#include <utility>
+
+#include "types.h"
+
+namespace Stockfish {
+
+class Position;
+
+class Score {
+   public:
+    struct Mate {
+        int plies;
+    };
+
+    struct Tablebase {
+        int  plies;
+        bool win;
+    };
+
+    struct InternalUnits {
+        int value;
+    };
+
+    Score() = default;
+    Score(Value v, const Position& pos);
+
+    template<typename T>
+    bool is() const {
+        return std::holds_alternative<T>(score);
+    }
+
+    template<typename T>
+    T get() const {
+        return std::get<T>(score);
+    }
+
+    template<typename F>
+    decltype(auto) visit(F&& f) const {
+        return std::visit(std::forward<F>(f), score);
+    }
+
+   private:
+    std::variant<Mate, Tablebase, InternalUnits> score;
+};
+
+}
+
+#endif  // #ifndef SCORE_H_INCLUDED

src/search.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,98 +19,348 @@
 #ifndef SEARCH_H_INCLUDED
 #define SEARCH_H_INCLUDED
 
+#include <algorithm>
+#include <array>
+#include <atomic>
+#include <cassert>
+#include <cstddef>
+#include <cstdint>
+#include <functional>
+#include <memory>
+#include <string>
+#include <string_view>
 #include <vector>
 
+#include "history.h"
 #include "misc.h"
-#include "movepick.h"
+#include "nnue/network.h"
+#include "nnue/nnue_accumulator.h"
+#include "numa.h"
+#include "position.h"
+#include "score.h"
+#include "syzygy/tbprobe.h"
+#include "timeman.h"
 #include "types.h"
 
 namespace Stockfish {
 
-class Position;
+// Different node types, used as a template parameter
+enum NodeType {
+    NonPV,
+    PV,
+    Root
+};
+
+class TranspositionTable;
+class ThreadPool;
+class OptionsMap;
 
 namespace Search {
 
-
+// Stack struct keeps track of the information we need to remember from nodes
-/// 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
-/// shallower and deeper in the tree during the search. Each search thread has
+// its own array of Stack objects, indexed by the current ply.
-/// its own array of Stack objects, indexed by the current ply.
-
 struct Stack {
-  Move* pv;
+    Move*                       pv;
-  PieceToHistory* continuationHistory;
+    PieceToHistory*             continuationHistory;
-  int ply;
+    CorrectionHistory<PieceTo>* continuationCorrectionHistory;
-  Move currentMove;
+    int                         ply;
-  Move excludedMove;
+    Move                        currentMove;
-  Move killers[2];
+    Move                        excludedMove;
-  Value staticEval;
+    Value                       staticEval;
-  int statScore;
+    int                         statScore;
-  int moveCount;
+    int                         moveCount;
-  bool inCheck;
+    bool                        inCheck;
-  bool ttPv;
+    bool                        ttPv;
-  bool ttHit;
+    bool                        ttHit;
-  int doubleExtensions;
+    int                         cutoffCnt;
-  int cutoffCnt;
+    int                         reduction;
+    bool                        isTTMove;
 };
 
 
-/// RootMove struct is used for moves at the root of the tree. For each root move
+// 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
+// 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.
+// fail low). Score is normally set at -VALUE_INFINITE for all non-pv moves.
-
 struct RootMove {
 
-  explicit RootMove(Move m) : pv(1, m) {}
+    explicit RootMove(Move m) :
-  bool extract_ponder_from_tt(Position& pos);
+        pv(1, m) {}
-  bool operator==(const Move& m) const { return pv[0] == m; }
+    bool extract_ponder_from_tt(const TranspositionTable& tt, Position& pos);
-  bool operator<(const RootMove& m) const { // Sort in descending order
+    bool operator==(const Move& m) const { return pv[0] == m; }
-    return m.score != score ? m.score < score
+    // Sort in descending order
-                            : m.previousScore < previousScore;
+    bool operator<(const RootMove& m) const {
-  }
+        return m.score != score ? m.score < score : m.previousScore < previousScore;
+    }
 
-  Value score = -VALUE_INFINITE;
+    uint64_t          effort           = 0;
-  Value previousScore = -VALUE_INFINITE;
+    Value             score            = -VALUE_INFINITE;
-  Value averageScore = -VALUE_INFINITE;
+    Value             previousScore    = -VALUE_INFINITE;
-  Value uciScore = -VALUE_INFINITE;
+    Value             averageScore     = -VALUE_INFINITE;
-  bool scoreLowerbound = false;
+    Value             meanSquaredScore = -VALUE_INFINITE * VALUE_INFINITE;
-  bool scoreUpperbound = false;
+    Value             uciScore         = -VALUE_INFINITE;
-  int selDepth = 0;
+    bool              scoreLowerbound  = false;
-  int tbRank = 0;
+    bool              scoreUpperbound  = false;
-  Value tbScore;
+    int               selDepth         = 0;
-  std::vector<Move> pv;
+    int               tbRank           = 0;
+    Value             tbScore;
+    std::vector<Move> pv;
 };
 
 using RootMoves = std::vector<RootMove>;
 
 
-/// LimitsType struct stores information sent by GUI about available time to
+// LimitsType struct stores information sent by the caller about the analysis required.
-/// search the current move, maximum depth/time, or if we are in analysis mode.
-
 struct LimitsType {
 
-  LimitsType() { // Init explicitly due to broken value-initialization of non POD in MSVC
+    // Init explicitly due to broken value-initialization of non POD in MSVC
-    time[WHITE] = time[BLACK] = inc[WHITE] = inc[BLACK] = npmsec = movetime = TimePoint(0);
+    LimitsType() {
-    movestogo = depth = mate = perft = infinite = 0;
+        time[WHITE] = time[BLACK] = inc[WHITE] = inc[BLACK] = npmsec = movetime = TimePoint(0);
-    nodes = 0;
+        movestogo = depth = mate = perft = infinite = 0;
-  }
+        nodes                                       = 0;
+        ponderMode                                  = false;
+    }
 
-  bool use_time_management() const {
+    bool use_time_management() const { return time[WHITE] || time[BLACK]; }
-    return time[WHITE] || time[BLACK];
-  }
 
-  std::vector<Move> searchmoves;
+    std::vector<std::string> searchmoves;
-  TimePoint time[COLOR_NB], inc[COLOR_NB], npmsec, movetime, startTime;
+    TimePoint                time[COLOR_NB], inc[COLOR_NB], npmsec, movetime, startTime;
-  int movestogo, depth, mate, perft, infinite;
+    int                      movestogo, depth, mate, perft, infinite;
-  int64_t nodes;
+    uint64_t                 nodes;
+    bool                     ponderMode;
 };
 
-extern LimitsType Limits;
 
-void init();
+// The UCI stores the uci options, thread pool, and transposition table.
-void clear();
+// This struct is used to easily forward data to the Search::Worker class.
+struct SharedState {
+    SharedState(const OptionsMap&                               optionsMap,
+                ThreadPool&                                     threadPool,
+                TranspositionTable&                             transpositionTable,
+                const LazyNumaReplicated<Eval::NNUE::Networks>& nets) :
+        options(optionsMap),
+        threads(threadPool),
+        tt(transpositionTable),
+        networks(nets) {}
 
-} // namespace Search
+    const OptionsMap&                               options;
+    ThreadPool&                                     threads;
+    TranspositionTable&                             tt;
+    const LazyNumaReplicated<Eval::NNUE::Networks>& networks;
+};
 
-} // namespace Stockfish
+class Worker;
 
-#endif // #ifndef SEARCH_H_INCLUDED
+// Null Object Pattern, implement a common interface for the SearchManagers.
+// A Null Object will be given to non-mainthread workers.
+class ISearchManager {
+   public:
+    virtual ~ISearchManager() {}
+    virtual void check_time(Search::Worker&) = 0;
+};
+
+struct InfoShort {
+    int   depth;
+    Score score;
+};
+
+struct InfoFull: InfoShort {
+    int              selDepth;
+    size_t           multiPV;
+    std::string_view wdl;
+    std::string_view bound;
+    size_t           timeMs;
+    size_t           nodes;
+    size_t           nps;
+    size_t           tbHits;
+    std::string_view pv;
+    int              hashfull;
+};
+
+struct InfoIteration {
+    int              depth;
+    std::string_view currmove;
+    size_t           currmovenumber;
+};
+
+// Skill structure is used to implement strength limit. If we have a UCI_Elo,
+// we convert it to an appropriate skill level, anchored to the Stash engine.
+// This method is based on a fit of the Elo results for games played between
+// Stockfish at various skill levels and various versions of the Stash engine.
+// Skill 0 .. 19 now covers CCRL Blitz Elo from 1320 to 3190, approximately
+// Reference: https://github.com/vondele/Stockfish/commit/a08b8d4e9711c2
+struct Skill {
+    // Lowest and highest Elo ratings used in the skill level calculation
+    constexpr static int LowestElo  = 1320;
+    constexpr static int HighestElo = 3190;
+
+    Skill(int skill_level, int uci_elo) {
+        if (uci_elo)
+        {
+            double e = double(uci_elo - LowestElo) / (HighestElo - LowestElo);
+            level = std::clamp((((37.2473 * e - 40.8525) * e + 22.2943) * e - 0.311438), 0.0, 19.0);
+        }
+        else
+            level = double(skill_level);
+    }
+    bool enabled() const { return level < 20.0; }
+    bool time_to_pick(Depth depth) const { return depth == 1 + int(level); }
+    Move pick_best(const RootMoves&, size_t multiPV);
+
+    double level;
+    Move   best = Move::none();
+};
+
+// SearchManager manages the search from the main thread. It is responsible for
+// keeping track of the time, and storing data strictly related to the main thread.
+class SearchManager: public ISearchManager {
+   public:
+    using UpdateShort    = std::function<void(const InfoShort&)>;
+    using UpdateFull     = std::function<void(const InfoFull&)>;
+    using UpdateIter     = std::function<void(const InfoIteration&)>;
+    using UpdateBestmove = std::function<void(std::string_view, std::string_view)>;
+
+    struct UpdateContext {
+        UpdateShort    onUpdateNoMoves;
+        UpdateFull     onUpdateFull;
+        UpdateIter     onIter;
+        UpdateBestmove onBestmove;
+    };
+
+
+    SearchManager(const UpdateContext& updateContext) :
+        updates(updateContext) {}
+
+    void check_time(Search::Worker& worker) override;
+
+    void pv(Search::Worker&           worker,
+            const ThreadPool&         threads,
+            const TranspositionTable& tt,
+            Depth                     depth);
+
+    Stockfish::TimeManagement tm;
+    double                    originalTimeAdjust;
+    int                       callsCnt;
+    std::atomic_bool          ponder;
+
+    std::array<Value, 4> iterValue;
+    double               previousTimeReduction;
+    Value                bestPreviousScore;
+    Value                bestPreviousAverageScore;
+    bool                 stopOnPonderhit;
+
+    size_t id;
+
+    const UpdateContext& updates;
+};
+
+class NullSearchManager: public ISearchManager {
+   public:
+    void check_time(Search::Worker&) override {}
+};
+
+
+// Search::Worker is the class that does the actual search.
+// It is instantiated once per thread, and it is responsible for keeping track
+// of the search history, and storing data required for the search.
+class Worker {
+   public:
+    Worker(SharedState&, std::unique_ptr<ISearchManager>, size_t, NumaReplicatedAccessToken);
+
+    // Called at instantiation to initialize reductions tables.
+    // Reset histories, usually before a new game.
+    void clear();
+
+    // Called when the program receives the UCI 'go' command.
+    // It searches from the root position and outputs the "bestmove".
+    void start_searching();
+
+    bool is_mainthread() const { return threadIdx == 0; }
+
+    void ensure_network_replicated();
+
+    // Public because they need to be updatable by the stats
+    ButterflyHistory mainHistory;
+    LowPlyHistory    lowPlyHistory;
+
+    CapturePieceToHistory captureHistory;
+    ContinuationHistory   continuationHistory[2][2];
+    PawnHistory           pawnHistory;
+
+    CorrectionHistory<Pawn>         pawnCorrectionHistory;
+    CorrectionHistory<Minor>        minorPieceCorrectionHistory;
+    CorrectionHistory<NonPawn>      nonPawnCorrectionHistory[COLOR_NB];
+    CorrectionHistory<Continuation> continuationCorrectionHistory;
+
+   private:
+    void iterative_deepening();
+
+    // This is the main search function, for both PV and non-PV nodes
+    template<NodeType nodeType>
+    Value search(Position& pos, Stack* ss, Value alpha, Value beta, Depth depth, bool cutNode);
+
+    // Quiescence search function, which is called by the main search
+    template<NodeType nodeType>
+    Value qsearch(Position& pos, Stack* ss, Value alpha, Value beta);
+
+    Depth reduction(bool i, Depth d, int mn, int delta) const;
+
+    // Pointer to the search manager, only allowed to be called by the main thread
+    SearchManager* main_manager() const {
+        assert(threadIdx == 0);
+        return static_cast<SearchManager*>(manager.get());
+    }
+
+    TimePoint elapsed() const;
+    TimePoint elapsed_time() const;
+
+    Value evaluate(const Position&);
+
+    LimitsType limits;
+
+    size_t                pvIdx, pvLast;
+    std::atomic<uint64_t> nodes, tbHits, bestMoveChanges;
+    int                   selDepth, nmpMinPly;
+
+    Value optimism[COLOR_NB];
+
+    Position  rootPos;
+    StateInfo rootState;
+    RootMoves rootMoves;
+    Depth     rootDepth, completedDepth;
+    Value     rootDelta;
+
+    size_t                    threadIdx;
+    NumaReplicatedAccessToken numaAccessToken;
+
+    // Reductions lookup table initialized at startup
+    std::array<int, MAX_MOVES> reductions;  // [depth or moveNumber]
+
+    // The main thread has a SearchManager, the others have a NullSearchManager
+    std::unique_ptr<ISearchManager> manager;
+
+    Tablebases::Config tbConfig;
+
+    const OptionsMap&                               options;
+    ThreadPool&                                     threads;
+    TranspositionTable&                             tt;
+    const LazyNumaReplicated<Eval::NNUE::Networks>& networks;
+
+    // Used by NNUE
+    Eval::NNUE::AccumulatorCaches refreshTable;
+
+    friend class Stockfish::ThreadPool;
+    friend class SearchManager;
+};
+
+struct ConthistBonus {
+    int index;
+    int weight;
+};
+
+
+}  // namespace Search
+
+}  // namespace Stockfish
+
+#endif  // #ifndef SEARCH_H_INCLUDED

src/syzygy/tbprobe.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,58 +19,60 @@
 #ifndef TBPROBE_H
 #define TBPROBE_H
 
-#include <ostream>
+#include <string>
+#include <vector>
 
-#include "../search.h"
+
+namespace Stockfish {
+class Position;
+class OptionsMap;
+
+using Depth = int;
+
+namespace Search {
+struct RootMove;
+using RootMoves = std::vector<RootMove>;
+}
+}
 
 namespace Stockfish::Tablebases {
 
+struct Config {
+    int   cardinality = 0;
+    bool  rootInTB    = false;
+    bool  useRule50   = false;
+    Depth probeDepth  = 0;
+};
+
 enum WDLScore {
-    WDLLoss        = -2, // Loss
+    WDLLoss        = -2,  // Loss
-    WDLBlessedLoss = -1, // Loss, but draw under 50-move rule
+    WDLBlessedLoss = -1,  // Loss, but draw under 50-move rule
-    WDLDraw        =  0, // Draw
+    WDLDraw        = 0,   // Draw
-    WDLCursedWin   =  1, // Win, but draw under 50-move rule
+    WDLCursedWin   = 1,   // Win, but draw under 50-move rule
-    WDLWin         =  2, // Win
+    WDLWin         = 2,   // Win
 };
 
 // Possible states after a probing operation
 enum ProbeState {
-    FAIL              =  0, // Probe failed (missing file table)
+    FAIL              = 0,   // Probe failed (missing file table)
-    OK                =  1, // Probe successful
+    OK                = 1,   // Probe successful
-    CHANGE_STM        = -1, // DTZ should check the other side
+    CHANGE_STM        = -1,  // DTZ should check the other side
-    ZEROING_BEST_MOVE =  2  // Best move zeroes DTZ (capture or pawn move)
+    ZEROING_BEST_MOVE = 2    // Best move zeroes DTZ (capture or pawn move)
 };
 
 extern int MaxCardinality;
 
-void init(const std::string& paths);
+
+void     init(const std::string& paths);
 WDLScore probe_wdl(Position& pos, ProbeState* result);
-int probe_dtz(Position& pos, ProbeState* result);
+int      probe_dtz(Position& pos, ProbeState* result);
-bool root_probe(Position& pos, Search::RootMoves& rootMoves);
+bool     root_probe(Position& pos, Search::RootMoves& rootMoves, bool rule50, bool rankDTZ);
-bool root_probe_wdl(Position& pos, Search::RootMoves& rootMoves);
+bool     root_probe_wdl(Position& pos, Search::RootMoves& rootMoves, bool rule50);
-void rank_root_moves(Position& pos, Search::RootMoves& rootMoves);
+Config   rank_root_moves(const OptionsMap&  options,
+                         Position&          pos,
+                         Search::RootMoves& rootMoves,
+                         bool               rankDTZ = false);
 
-inline std::ostream& operator<<(std::ostream& os, const WDLScore v) {
+}  // namespace Stockfish::Tablebases
-
-    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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,253 +16,395 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#include <cassert>
+#include "thread.h"
+
+#include <algorithm>
+#include <cassert>
+#include <deque>
+#include <memory>
+#include <string>
+#include <unordered_map>
+#include <utility>
 
-#include <algorithm> // For std::count
 #include "movegen.h"
 #include "search.h"
-#include "thread.h"
-#include "uci.h"
 #include "syzygy/tbprobe.h"
-#include "tt.h"
+#include "timeman.h"
+#include "types.h"
+#include "uci.h"
+#include "ucioption.h"
 
 namespace Stockfish {
 
-ThreadPool Threads; // Global object
+// Constructor launches the thread and waits until it goes to sleep
+// in idle_loop(). Note that 'searching' and 'exit' should be already set.
+Thread::Thread(Search::SharedState&                    sharedState,
+               std::unique_ptr<Search::ISearchManager> sm,
+               size_t                                  n,
+               OptionalThreadToNumaNodeBinder          binder) :
+    idx(n),
+    nthreads(sharedState.options["Threads"]),
+    stdThread(&Thread::idle_loop, this) {
 
+    wait_for_search_finished();
 
-/// Thread constructor launches the thread and waits until it goes to sleep
+    run_custom_job([this, &binder, &sharedState, &sm, n]() {
-/// in idle_loop(). Note that 'searching' and 'exit' should be already set.
+        // Use the binder to [maybe] bind the threads to a NUMA node before doing
+        // the Worker allocation. Ideally we would also allocate the SearchManager
+        // here, but that's minor.
+        this->numaAccessToken = binder();
+        this->worker =
+          std::make_unique<Search::Worker>(sharedState, std::move(sm), n, this->numaAccessToken);
+    });
 
-Thread::Thread(size_t n) : idx(n), stdThread(&Thread::idle_loop, this) {
+    wait_for_search_finished();
-
-  wait_for_search_finished();
 }
 
 
-/// Thread destructor wakes up the thread in idle_loop() and waits
+// Destructor wakes up the thread in idle_loop() and waits
-/// for its termination. Thread should be already waiting.
+// for its termination. Thread should be already waiting.
-
 Thread::~Thread() {
 
-  assert(!searching);
+    assert(!searching);
 
-  exit = true;
+    exit = true;
-  start_searching();
+    start_searching();
-  stdThread.join();
+    stdThread.join();
 }
 
-
+// Wakes up the thread that will start the search
-/// Thread::clear() reset histories, usually before a new game
-
-void Thread::clear() {
-
-  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::start_searching() wakes up the thread that will start the search
-
 void Thread::start_searching() {
-  mutex.lock();
+    assert(worker != nullptr);
-  searching = true;
+    run_custom_job([this]() { worker->start_searching(); });
-  mutex.unlock(); // Unlock before notifying saves a few CPU-cycles
-  cv.notify_one(); // Wake up the thread in idle_loop()
 }
 
+// Clears the histories for the thread worker (usually before a new game)
+void Thread::clear_worker() {
+    assert(worker != nullptr);
+    run_custom_job([this]() { worker->clear(); });
+}
 
-/// Thread::wait_for_search_finished() blocks on the condition variable
+// Blocks on the condition variable until the thread has finished searching
-/// until the thread has finished searching.
-
 void Thread::wait_for_search_finished() {
 
-  std::unique_lock<std::mutex> lk(mutex);
+    std::unique_lock<std::mutex> lk(mutex);
-  cv.wait(lk, [&]{ return !searching; });
+    cv.wait(lk, [&] { return !searching; });
 }
 
+// Launching a function in the thread
+void Thread::run_custom_job(std::function<void()> f) {
+    {
+        std::unique_lock<std::mutex> lk(mutex);
+        cv.wait(lk, [&] { return !searching; });
+        jobFunc   = std::move(f);
+        searching = true;
+    }
+    cv.notify_one();
+}
 
-/// Thread::idle_loop() is where the thread is parked, blocked on the
+void Thread::ensure_network_replicated() { worker->ensure_network_replicated(); }
-/// condition variable, when it has no work to do.
+
+// Thread gets parked here, blocked on the condition variable
+// when the thread has no work to do.
 
 void Thread::idle_loop() {
+    while (true)
+    {
+        std::unique_lock<std::mutex> lk(mutex);
+        searching = false;
+        cv.notify_one();  // Wake up anyone waiting for search finished
+        cv.wait(lk, [&] { return searching; });
 
-  // If OS already scheduled us on a different group than 0 then don't overwrite
+        if (exit)
-  // the choice, eventually we are one of many one-threaded processes running on
+            return;
-  // 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)
+        std::function<void()> job = std::move(jobFunc);
-  {
+        jobFunc                   = nullptr;
-      std::unique_lock<std::mutex> lk(mutex);
-      searching = false;
-      cv.notify_one(); // Wake up anyone waiting for search finished
-      cv.wait(lk, [&]{ return searching; });
 
-      if (exit)
+        lk.unlock();
-          return;
 
-      lk.unlock();
+        if (job)
-
+            job();
-      search();
+    }
-  }
 }
 
-/// ThreadPool::set() creates/destroys threads to match the requested number.
+Search::SearchManager* ThreadPool::main_manager() { return main_thread()->worker->main_manager(); }
-/// Created and launched threads will immediately go to sleep in idle_loop.
-/// Upon resizing, threads are recreated to allow for binding if necessary.
 
-void ThreadPool::set(size_t requested) {
+uint64_t ThreadPool::nodes_searched() const { return accumulate(&Search::Worker::nodes); }
+uint64_t ThreadPool::tb_hits() const { return accumulate(&Search::Worker::tbHits); }
 
-  if (threads.size() > 0)   // destroy any existing thread(s)
+// Creates/destroys threads to match the requested number.
-  {
+// Created and launched threads will immediately go to sleep in idle_loop.
-      main()->wait_for_search_finished();
+// Upon resizing, threads are recreated to allow for binding if necessary.
+void ThreadPool::set(const NumaConfig&                           numaConfig,
+                     Search::SharedState                         sharedState,
+                     const Search::SearchManager::UpdateContext& updateContext) {
 
-      while (threads.size() > 0)
+    if (threads.size() > 0)  // destroy any existing thread(s)
-          delete threads.back(), threads.pop_back();
+    {
-  }
+        main_thread()->wait_for_search_finished();
 
-  if (requested > 0)   // create new thread(s)
+        threads.clear();
-  {
-      threads.push_back(new MainThread(0));
 
-      while (threads.size() < requested)
+        boundThreadToNumaNode.clear();
-          threads.push_back(new Thread(threads.size()));
+    }
-      clear();
 
-      // Reallocate the hash with the new threadpool size
+    const size_t requested = sharedState.options["Threads"];
-      TT.resize(size_t(Options["Hash"]));
 
-      // Init thread number dependent search params.
+    if (requested > 0)  // create new thread(s)
-      Search::init();
+    {
-  }
+        // Binding threads may be problematic when there's multiple NUMA nodes and
+        // multiple Stockfish instances running. In particular, if each instance
+        // runs a single thread then they would all be mapped to the first NUMA node.
+        // This is undesirable, and so the default behaviour (i.e. when the user does not
+        // change the NumaConfig UCI setting) is to not bind the threads to processors
+        // unless we know for sure that we span NUMA nodes and replication is required.
+        const std::string numaPolicy(sharedState.options["NumaPolicy"]);
+        const bool        doBindThreads = [&]() {
+            if (numaPolicy == "none")
+                return false;
+
+            if (numaPolicy == "auto")
+                return numaConfig.suggests_binding_threads(requested);
+
+            // numaPolicy == "system", or explicitly set by the user
+            return true;
+        }();
+
+        boundThreadToNumaNode = doBindThreads
+                                ? numaConfig.distribute_threads_among_numa_nodes(requested)
+                                : std::vector<NumaIndex>{};
+
+        while (threads.size() < requested)
+        {
+            const size_t    threadId = threads.size();
+            const NumaIndex numaId   = doBindThreads ? boundThreadToNumaNode[threadId] : 0;
+            auto            manager  = threadId == 0 ? std::unique_ptr<Search::ISearchManager>(
+                                             std::make_unique<Search::SearchManager>(updateContext))
+                                                     : std::make_unique<Search::NullSearchManager>();
+
+            // When not binding threads we want to force all access to happen
+            // from the same NUMA node, because in case of NUMA replicated memory
+            // accesses we don't want to trash cache in case the threads get scheduled
+            // on the same NUMA node.
+            auto binder = doBindThreads ? OptionalThreadToNumaNodeBinder(numaConfig, numaId)
+                                        : OptionalThreadToNumaNodeBinder(numaId);
+
+            threads.emplace_back(
+              std::make_unique<Thread>(sharedState, std::move(manager), threadId, binder));
+        }
+
+        clear();
+
+        main_thread()->wait_for_search_finished();
+    }
 }
 
 
-/// ThreadPool::clear() sets threadPool data to initial values
+// Sets threadPool data to initial values
-
 void ThreadPool::clear() {
+    if (threads.size() == 0)
+        return;
 
-  for (Thread* th : threads)
+    for (auto&& th : threads)
-      th->clear();
+        th->clear_worker();
 
-  main()->callsCnt = 0;
+    for (auto&& th : threads)
-  main()->bestPreviousScore = VALUE_INFINITE;
+        th->wait_for_search_finished();
-  main()->bestPreviousAverageScore = VALUE_INFINITE;
+
-  main()->previousTimeReduction = 1.0;
+    // These two affect the time taken on the first move of a game:
+    main_manager()->bestPreviousAverageScore = VALUE_INFINITE;
+    main_manager()->previousTimeReduction    = 0.85;
+
+    main_manager()->callsCnt           = 0;
+    main_manager()->bestPreviousScore  = VALUE_INFINITE;
+    main_manager()->originalTimeAdjust = -1;
+    main_manager()->tm.clear();
 }
 
+void ThreadPool::run_on_thread(size_t threadId, std::function<void()> f) {
+    assert(threads.size() > threadId);
+    threads[threadId]->run_custom_job(std::move(f));
+}
 
-/// ThreadPool::start_thinking() wakes up main thread waiting in idle_loop() and
+void ThreadPool::wait_on_thread(size_t threadId) {
-/// returns immediately. Main thread will wake up other threads and start the search.
+    assert(threads.size() > threadId);
+    threads[threadId]->wait_for_search_finished();
+}
 
-void ThreadPool::start_thinking(Position& pos, StateListPtr& states,
+size_t ThreadPool::num_threads() const { return threads.size(); }
-                                const Search::LimitsType& limits, bool ponderMode) {
 
-  main()->wait_for_search_finished();
 
-  main()->stopOnPonderhit = stop = false;
+// Wakes up main thread waiting in idle_loop() and returns immediately.
-  increaseDepth = true;
+// Main thread will wake up other threads and start the search.
-  main()->ponder = ponderMode;
+void ThreadPool::start_thinking(const OptionsMap&  options,
-  Search::Limits = limits;
+                                Position&          pos,
-  Search::RootMoves rootMoves;
+                                StateListPtr&      states,
+                                Search::LimitsType limits) {
 
-  for (const auto& m : MoveList<LEGAL>(pos))
+    main_thread()->wait_for_search_finished();
-      if (   limits.searchmoves.empty()
-          || std::count(limits.searchmoves.begin(), limits.searchmoves.end(), m))
-          rootMoves.emplace_back(m);
 
-  if (!rootMoves.empty())
+    main_manager()->stopOnPonderhit = stop = abortedSearch = false;
-      Tablebases::rank_root_moves(pos, rootMoves);
+    main_manager()->ponder                                 = limits.ponderMode;
 
-  // After ownership transfer 'states' becomes empty, so if we stop the search
+    increaseDepth = true;
-  // and call 'go' again without setting a new position states.get() == nullptr.
-  assert(states.get() || setupStates.get());
 
-  if (states.get())
+    Search::RootMoves rootMoves;
-      setupStates = std::move(states); // Ownership transfer, states is now empty
+    const auto        legalmoves = MoveList<LEGAL>(pos);
 
-  // We use Position::set() to set root position across threads. But there are
+    for (const auto& uciMove : limits.searchmoves)
-  // some StateInfo fields (previous, pliesFromNull, capturedPiece) that cannot
+    {
-  // be deduced from a fen string, so set() clears them and they are set from
+        auto move = UCIEngine::to_move(pos, uciMove);
-  // setupStates->back() later. The rootState is per thread, earlier states are shared
-  // since they are read-only.
-  for (Thread* th : threads)
-  {
-      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();
-  }
 
-  main()->start_searching();
+        if (std::find(legalmoves.begin(), legalmoves.end(), move) != legalmoves.end())
+            rootMoves.emplace_back(move);
+    }
+
+    if (rootMoves.empty())
+        for (const auto& m : legalmoves)
+            rootMoves.emplace_back(m);
+
+    Tablebases::Config tbConfig = Tablebases::rank_root_moves(options, pos, rootMoves);
+
+    // 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());
+
+    if (states.get())
+        setupStates = std::move(states);  // Ownership transfer, states is now empty
+
+    // 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 (auto&& th : threads)
+    {
+        th->run_custom_job([&]() {
+            th->worker->limits = limits;
+            th->worker->nodes = th->worker->tbHits = th->worker->nmpMinPly =
+              th->worker->bestMoveChanges          = 0;
+            th->worker->rootDepth = th->worker->completedDepth = 0;
+            th->worker->rootMoves                              = rootMoves;
+            th->worker->rootPos.set(pos.fen(), pos.is_chess960(), &th->worker->rootState);
+            th->worker->rootState = setupStates->back();
+            th->worker->tbConfig  = tbConfig;
+        });
+    }
+
+    for (auto&& th : threads)
+        th->wait_for_search_finished();
+
+    main_thread()->start_searching();
 }
 
 Thread* ThreadPool::get_best_thread() const {
 
-    Thread* bestThread = threads.front();
+    Thread* bestThread = threads.front().get();
-    std::map<Move, int64_t> votes;
+    Value   minScore   = VALUE_NONE;
-    Value minScore = VALUE_NONE;
 
-    // Find minimum score of all threads
+    std::unordered_map<Move, int64_t, Move::MoveHash> votes(
-    for (Thread* th: threads)
+      2 * std::min(size(), bestThread->worker->rootMoves.size()));
-        minScore = std::min(minScore, th->rootMoves[0].score);
+
+    // Find the minimum score of all threads
+    for (auto&& th : threads)
+        minScore = std::min(minScore, th->worker->rootMoves[0].score);
 
     // Vote according to score and depth, and select the best thread
-    auto thread_value = [minScore](Thread* th) {
+    auto thread_voting_value = [minScore](Thread* th) {
-            return (th->rootMoves[0].score - minScore + 14) * int(th->completedDepth);
+        return (th->worker->rootMoves[0].score - minScore + 14) * int(th->worker->completedDepth);
-        };
+    };
 
-    for (Thread* th : threads)
+    for (auto&& th : threads)
-        votes[th->rootMoves[0].pv[0]] += thread_value(th);
+        votes[th->worker->rootMoves[0].pv[0]] += thread_voting_value(th.get());
 
-    for (Thread* th : threads)
+    for (auto&& th : threads)
-        if (abs(bestThread->rootMoves[0].score) >= VALUE_TB_WIN_IN_MAX_PLY)
+    {
+        const auto bestThreadScore = bestThread->worker->rootMoves[0].score;
+        const auto newThreadScore  = th->worker->rootMoves[0].score;
+
+        const auto& bestThreadPV = bestThread->worker->rootMoves[0].pv;
+        const auto& newThreadPV  = th->worker->rootMoves[0].pv;
+
+        const auto bestThreadMoveVote = votes[bestThreadPV[0]];
+        const auto newThreadMoveVote  = votes[newThreadPV[0]];
+
+        const bool bestThreadInProvenWin = is_win(bestThreadScore);
+        const bool newThreadInProvenWin  = is_win(newThreadScore);
+
+        const bool bestThreadInProvenLoss =
+          bestThreadScore != -VALUE_INFINITE && is_loss(bestThreadScore);
+        const bool newThreadInProvenLoss =
+          newThreadScore != -VALUE_INFINITE && is_loss(newThreadScore);
+
+        // We make sure not to pick a thread with truncated principal variation
+        const bool betterVotingValue =
+          thread_voting_value(th.get()) * int(newThreadPV.size() > 2)
+          > thread_voting_value(bestThread) * int(bestThreadPV.size() > 2);
+
+        if (bestThreadInProvenWin)
         {
-            // Make sure we pick the shortest mate / TB conversion or stave off mate the longest
+            // Make sure we pick the shortest mate / TB conversion
-            if (th->rootMoves[0].score > bestThread->rootMoves[0].score)
+            if (newThreadScore > bestThreadScore)
-                bestThread = th;
+                bestThread = th.get();
         }
-        else if (   th->rootMoves[0].score >= VALUE_TB_WIN_IN_MAX_PLY
+        else if (bestThreadInProvenLoss)
-                 || (   th->rootMoves[0].score > VALUE_TB_LOSS_IN_MAX_PLY
+        {
-                     && (   votes[th->rootMoves[0].pv[0]] > votes[bestThread->rootMoves[0].pv[0]]
+            // Make sure we pick the shortest mated / TB conversion
-                         || (   votes[th->rootMoves[0].pv[0]] == votes[bestThread->rootMoves[0].pv[0]]
+            if (newThreadInProvenLoss && newThreadScore < bestThreadScore)
-                             &&   thread_value(th) * int(th->rootMoves[0].pv.size() > 2)
+                bestThread = th.get();
-                                > thread_value(bestThread) * int(bestThread->rootMoves[0].pv.size() > 2)))))
+        }
-            bestThread = th;
+        else if (newThreadInProvenWin || newThreadInProvenLoss
+                 || (!is_loss(newThreadScore)
+                     && (newThreadMoveVote > bestThreadMoveVote
+                         || (newThreadMoveVote == bestThreadMoveVote && betterVotingValue))))
+            bestThread = th.get();
+    }
 
     return bestThread;
 }
 
 
-/// Start non-main threads
+// Start non-main threads.
-
+// Will be invoked by main thread after it has started searching.
 void ThreadPool::start_searching() {
 
-    for (Thread* th : threads)
+    for (auto&& th : threads)
         if (th != threads.front())
             th->start_searching();
 }
 
 
-/// Wait for non-main threads
+// Wait for non-main threads
-
 void ThreadPool::wait_for_search_finished() const {
 
-    for (Thread* th : threads)
+    for (auto&& th : threads)
         if (th != threads.front())
             th->wait_for_search_finished();
 }
 
-} // namespace Stockfish
+std::vector<size_t> ThreadPool::get_bound_thread_count_by_numa_node() const {
+    std::vector<size_t> counts;
+
+    if (!boundThreadToNumaNode.empty())
+    {
+        NumaIndex highestNumaNode = 0;
+        for (NumaIndex n : boundThreadToNumaNode)
+            if (n > highestNumaNode)
+                highestNumaNode = n;
+
+        counts.resize(highestNumaNode + 1, 0);
+
+        for (NumaIndex n : boundThreadToNumaNode)
+            counts[n] += 1;
+    }
+
+    return counts;
+}
+
+void ThreadPool::ensure_network_replicated() {
+    for (auto&& th : threads)
+        th->ensure_network_replicated();
+}
+
+}  // namespace Stockfish

src/thread.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,121 +21,158 @@
 
 #include <atomic>
 #include <condition_variable>
+#include <cstddef>
+#include <cstdint>
+#include <functional>
+#include <memory>
 #include <mutex>
-#include <thread>
 #include <vector>
 
-#include "material.h"
+#include "numa.h"
-#include "movepick.h"
-#include "pawns.h"
 #include "position.h"
 #include "search.h"
 #include "thread_win32_osx.h"
 
 namespace Stockfish {
 
-/// Thread class keeps together all the thread-related stuff. We use
-/// per-thread pawn and material hash tables so that once we get a
-/// 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 OptionsMap;
+using Value = int;
+
+// Sometimes we don't want to actually bind the threads, but the recipient still
+// needs to think it runs on *some* NUMA node, such that it can access structures
+// that rely on NUMA node knowledge. This class encapsulates this optional process
+// such that the recipient does not need to know whether the binding happened or not.
+class OptionalThreadToNumaNodeBinder {
+   public:
+    OptionalThreadToNumaNodeBinder(NumaIndex n) :
+        numaConfig(nullptr),
+        numaId(n) {}
+
+    OptionalThreadToNumaNodeBinder(const NumaConfig& cfg, NumaIndex n) :
+        numaConfig(&cfg),
+        numaId(n) {}
+
+    NumaReplicatedAccessToken operator()() const {
+        if (numaConfig != nullptr)
+            return numaConfig->bind_current_thread_to_numa_node(numaId);
+        else
+            return NumaReplicatedAccessToken(numaId);
+    }
+
+   private:
+    const NumaConfig* numaConfig;
+    NumaIndex         numaId;
+};
+
+// Abstraction of a thread. It contains a pointer to the worker and a native thread.
+// After construction, the native thread is started with idle_loop()
+// waiting for a signal to start searching.
+// When the signal is received, the thread starts searching and when
+// the search is finished, it goes back to idle_loop() waiting for a new signal.
 class Thread {
+   public:
+    Thread(Search::SharedState&,
+           std::unique_ptr<Search::ISearchManager>,
+           size_t,
+           OptionalThreadToNumaNodeBinder);
+    virtual ~Thread();
 
-  std::mutex mutex;
+    void idle_loop();
-  std::condition_variable cv;
+    void start_searching();
-  size_t idx;
+    void clear_worker();
-  bool exit = false, searching = true; // Set before starting std::thread
+    void run_custom_job(std::function<void()> f);
-  NativeThread stdThread;
 
-public:
+    void ensure_network_replicated();
-  explicit Thread(size_t);
-  virtual ~Thread();
-  virtual void search();
-  void clear();
-  void idle_loop();
-  void start_searching();
-  void wait_for_search_finished();
-  size_t id() const { return idx; }
 
-  Pawns::Table pawnsTable;
+    // Thread has been slightly altered to allow running custom jobs, so
-  Material::Table materialTable;
+    // this name is no longer correct. However, this class (and ThreadPool)
-  size_t pvIdx, pvLast;
+    // require further work to make them properly generic while maintaining
-  std::atomic<uint64_t> nodes, tbHits, bestMoveChanges;
+    // appropriate specificity regarding search, from the point of view of an
-  int selDepth, nmpMinPly;
+    // outside user, so renaming of this function is left for whenever that happens.
-  Value bestValue, optimism[COLOR_NB];
+    void   wait_for_search_finished();
+    size_t id() const { return idx; }
 
-  Position rootPos;
+    std::unique_ptr<Search::Worker> worker;
-  StateInfo rootState;
+    std::function<void()>           jobFunc;
-  Search::RootMoves rootMoves;
+
-  Depth rootDepth, completedDepth;
+   private:
-  Value rootDelta;
+    std::mutex                mutex;
-  CounterMoveHistory counterMoves;
+    std::condition_variable   cv;
-  ButterflyHistory mainHistory;
+    size_t                    idx, nthreads;
-  CapturePieceToHistory captureHistory;
+    bool                      exit = false, searching = true;  // Set before starting std::thread
-  ContinuationHistory continuationHistory[2][2];
+    NativeThread              stdThread;
+    NumaReplicatedAccessToken numaAccessToken;
 };
 
 
-/// MainThread is a derived class specific for main thread
+// 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.
+class ThreadPool {
+   public:
+    ThreadPool() {}
 
-struct MainThread : public Thread {
+    ~ThreadPool() {
+        // destroy any existing thread(s)
+        if (threads.size() > 0)
+        {
+            main_thread()->wait_for_search_finished();
 
-  using Thread::Thread;
+            threads.clear();
+        }
+    }
 
-  void search() override;
+    ThreadPool(const ThreadPool&) = delete;
-  void check_time();
+    ThreadPool(ThreadPool&&)      = delete;
 
-  double previousTimeReduction;
+    ThreadPool& operator=(const ThreadPool&) = delete;
-  Value bestPreviousScore;
+    ThreadPool& operator=(ThreadPool&&)      = delete;
-  Value bestPreviousAverageScore;
+
-  Value iterValue[4];
+    void   start_thinking(const OptionsMap&, Position&, StateListPtr&, Search::LimitsType);
-  int callsCnt;
+    void   run_on_thread(size_t threadId, std::function<void()> f);
-  bool stopOnPonderhit;
+    void   wait_on_thread(size_t threadId);
-  std::atomic_bool ponder;
+    size_t num_threads() const;
+    void   clear();
+    void   set(const NumaConfig& numaConfig,
+               Search::SharedState,
+               const Search::SearchManager::UpdateContext&);
+
+    Search::SearchManager* main_manager();
+    Thread*                main_thread() const { return threads.front().get(); }
+    uint64_t               nodes_searched() const;
+    uint64_t               tb_hits() const;
+    Thread*                get_best_thread() const;
+    void                   start_searching();
+    void                   wait_for_search_finished() const;
+
+    std::vector<size_t> get_bound_thread_count_by_numa_node() const;
+
+    void ensure_network_replicated();
+
+    std::atomic_bool stop, abortedSearch, 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(); }
+
+   private:
+    StateListPtr                         setupStates;
+    std::vector<std::unique_ptr<Thread>> threads;
+    std::vector<NumaIndex>               boundThreadToNumaNode;
+
+    uint64_t accumulate(std::atomic<uint64_t> Search::Worker::*member) const {
+
+        uint64_t sum = 0;
+        for (auto&& th : threads)
+            sum += (th->worker.get()->*member).load(std::memory_order_relaxed);
+        return sum;
+    }
 };
 
+}  // namespace Stockfish
 
-/// ThreadPool struct handles all the threads-related stuff like init, starting,
+#endif  // #ifndef THREAD_H_INCLUDED
-/// 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(); }
-
-private:
-  StateListPtr setupStates;
-  std::vector<Thread*> threads;
-
-  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;
-
-} // namespace Stockfish
-
-#endif // #ifndef THREAD_H_INCLUDED

src/thread_win32_osx.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -21,54 +21,58 @@
 
 #include <thread>
 
-/// On OSX threads other than the main thread are created with a reduced stack
+// 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
+// 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.
+// somewhat more than 1MB stack, so adjust it to TH_STACK_SIZE.
-/// The implementation calls pthread_create() with the stack size parameter
+// The implementation calls pthread_create() with the stack size parameter
-/// equal to the linux 8MB default, on platforms that support it.
+// equal to the Linux 8MB default, on platforms that support it.
 
 #if defined(__APPLE__) || defined(__MINGW32__) || defined(__MINGW64__) || defined(USE_PTHREADS)
 
-#include <pthread.h>
+    #include <pthread.h>
+    #include <functional>
 
 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;
 
-   pthread_t thread;
+    static constexpr size_t TH_STACK_SIZE = 8 * 1024 * 1024;
 
-public:
+   public:
-  template<class T, class P = std::pair<T*, void(T::*)()>>
+    template<class Function, class... Args>
-  explicit NativeThread(void(T::*fun)(), T* obj) {
+    explicit NativeThread(Function&& fun, Args&&... args) {
-    pthread_attr_t attr_storage, *attr = &attr_storage;
+        auto func = new std::function<void()>(
-    pthread_attr_init(attr);
+          std::bind(std::forward<Function>(fun), std::forward<Args>(args)...));
-    pthread_attr_setstacksize(attr, TH_STACK_SIZE);
+
-    pthread_create(&thread, attr, start_routine<T>, new P(obj, fun));
+        pthread_attr_t attr_storage, *attr = &attr_storage;
-  }
+        pthread_attr_init(attr);
-  void join() { pthread_join(thread, nullptr); }
+        pthread_attr_setstacksize(attr, TH_STACK_SIZE);
+
+        auto start_routine = [](void* ptr) -> void* {
+            auto f = reinterpret_cast<std::function<void()>*>(ptr);
+            // Call the function
+            (*f)();
+            delete f;
+            return nullptr;
+        };
+
+        pthread_create(&thread, attr, start_routine, func);
+    }
+
+    void join() { pthread_join(thread, nullptr); }
 };
 
-} // namespace Stockfish
+}  // namespace Stockfish
 
-#else // Default case: use STL classes
+#else  // Default case: use STL classes
 
 namespace Stockfish {
 
 using NativeThread = std::thread;
 
-} // namespace Stockfish
+}  // namespace Stockfish
 
 #endif
 
-#endif // #ifndef THREAD_WIN32_OSX_H_INCLUDED
+#endif  // #ifndef THREAD_WIN32_OSX_H_INCLUDED

src/timeman.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,94 +16,128 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#include "timeman.h"
+
 #include <algorithm>
-#include <cfloat>
+#include <cassert>
 #include <cmath>
+#include <cstdint>
 
 #include "search.h"
-#include "timeman.h"
+#include "ucioption.h"
-#include "uci.h"
 
 namespace Stockfish {
 
-TimeManagement Time; // Our global time management object
+TimePoint TimeManagement::optimum() const { return optimumTime; }
+TimePoint TimeManagement::maximum() const { return maximumTime; }
 
-
+void TimeManagement::clear() {
-/// TimeManagement::init() is called at the beginning of the search and calculates
+    availableNodes = -1;  // When in 'nodes as time' mode
-/// 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)
-
-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;
-
-  TimePoint moveOverhead    = TimePoint(Options["Move Overhead"]);
-  TimePoint slowMover       = TimePoint(Options["Slow Mover"]);
-  TimePoint npmsec          = TimePoint(Options["nodestime"]);
-
-  // optScale is a percentage of available time to use for the current move.
-  // maxScale is a multiplier applied to optimumTime.
-  double optScale, maxScale;
-
-  // If we have to play in 'nodes as time' mode, then convert from time
-  // to nodes, and use resulting values in time management formulas.
-  // WARNING: to avoid time losses, the given npmsec (nodes per millisecond)
-  // must be much lower than the real engine speed.
-  if (npmsec)
-  {
-      if (!availableNodes) // Only once at game start
-          availableNodes = npmsec * limits.time[us]; // Time is in msec
-
-      // 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"])
-      optimumTime += optimumTime / 4;
 }
 
-} // namespace Stockfish
+void TimeManagement::advance_nodes_time(std::int64_t nodes) {
+    assert(useNodesTime);
+    availableNodes = std::max(int64_t(0), availableNodes - nodes);
+}
+
+// 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)
+void TimeManagement::init(Search::LimitsType& limits,
+                          Color               us,
+                          int                 ply,
+                          const OptionsMap&   options,
+                          double&             originalTimeAdjust) {
+    TimePoint npmsec = TimePoint(options["nodestime"]);
+
+    // If we have no time, we don't need to fully initialize TM.
+    // startTime is used by movetime and useNodesTime is used in elapsed calls.
+    startTime    = limits.startTime;
+    useNodesTime = npmsec != 0;
+
+    if (limits.time[us] == 0)
+        return;
+
+    TimePoint moveOverhead = TimePoint(options["Move Overhead"]);
+
+    // optScale is a percentage of available time to use for the current move.
+    // maxScale is a multiplier applied to optimumTime.
+    double optScale, maxScale;
+
+    // If we have to play in 'nodes as time' mode, then convert from time
+    // to nodes, and use resulting values in time management formulas.
+    // WARNING: to avoid time losses, the given npmsec (nodes per millisecond)
+    // must be much lower than the real engine speed.
+    if (useNodesTime)
+    {
+        if (availableNodes == -1)                       // Only once at game start
+            availableNodes = npmsec * limits.time[us];  // Time is in msec
+
+        // Convert from milliseconds to nodes
+        limits.time[us] = TimePoint(availableNodes);
+        limits.inc[us] *= npmsec;
+        limits.npmsec = npmsec;
+        moveOverhead *= npmsec;
+    }
+
+    // These numbers are used where multiplications, divisions or comparisons
+    // with constants are involved.
+    const int64_t   scaleFactor = useNodesTime ? npmsec : 1;
+    const TimePoint scaledTime  = limits.time[us] / scaleFactor;
+    const TimePoint scaledInc   = limits.inc[us] / scaleFactor;
+
+    // Maximum move horizon
+    int centiMTG = limits.movestogo ? std::min(limits.movestogo * 100, 5000) : 5051;
+
+    // If less than one second, gradually reduce mtg
+    if (scaledTime < 1000 && double(centiMTG) / scaledInc > 5.051)
+    {
+        centiMTG = scaledTime * 5.051;
+    }
+
+    // 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] * (centiMTG - 100) - moveOverhead * (200 + centiMTG)) / 100);
+
+    // x basetime (+ z increment)
+    // If there is a healthy increment, timeLeft can exceed the actual available
+    // game time for the current move, so also cap to a percentage of available game time.
+    if (limits.movestogo == 0)
+    {
+        // Extra time according to timeLeft
+        if (originalTimeAdjust < 0)
+            originalTimeAdjust = 0.3128 * std::log10(timeLeft) - 0.4354;
+
+        // Calculate time constants based on current time left.
+        double logTimeInSec = std::log10(scaledTime / 1000.0);
+        double optConstant  = std::min(0.0032116 + 0.000321123 * logTimeInSec, 0.00508017);
+        double maxConstant  = std::max(3.3977 + 3.03950 * logTimeInSec, 2.94761);
+
+        optScale = std::min(0.0121431 + std::pow(ply + 2.94693, 0.461073) * optConstant,
+                            0.213035 * limits.time[us] / timeLeft)
+                 * originalTimeAdjust;
+
+        maxScale = std::min(6.67704, maxConstant + ply / 11.9847);
+    }
+
+    // x moves in y seconds (+ z increment)
+    else
+    {
+        optScale =
+          std::min((0.88 + ply / 116.4) / (centiMTG / 100.0), 0.88 * limits.time[us] / timeLeft);
+        maxScale = 1.3 + 0.11 * (centiMTG / 100.0);
+    }
+
+    // Limit the maximum possible time for this move
+    optimumTime = TimePoint(optScale * timeLeft);
+    maximumTime =
+      TimePoint(std::min(0.825179 * limits.time[us] - moveOverhead, maxScale * optimumTime)) - 10;
+
+    if (options["Ponder"])
+        optimumTime += optimumTime / 4;
+}
+
+}  // namespace Stockfish

src/timeman.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,33 +19,49 @@
 #ifndef TIMEMAN_H_INCLUDED
 #define TIMEMAN_H_INCLUDED
 
+#include <cstdint>
+
 #include "misc.h"
-#include "search.h"
+#include "types.h"
-#include "thread.h"
 
 namespace Stockfish {
 
-/// The TimeManagement class computes the optimal time to think depending on
+class OptionsMap;
-/// the maximum available time, the game move number and other parameters.
 
+namespace Search {
+struct LimitsType;
+}
+
+// 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:
+   public:
-  void init(Search::LimitsType& limits, Color us, int ply);
+    void init(Search::LimitsType& limits,
-  TimePoint optimum() const { return optimumTime; }
+              Color               us,
-  TimePoint maximum() const { return maximumTime; }
+              int                 ply,
-  TimePoint elapsed() const { return Search::Limits.npmsec ?
+              const OptionsMap&   options,
-                                     TimePoint(Threads.nodes_searched()) : now() - startTime; }
+              double&             originalTimeAdjust);
 
-  int64_t availableNodes; // When in 'nodes as time' mode
+    TimePoint optimum() const;
+    TimePoint maximum() const;
+    template<typename FUNC>
+    TimePoint elapsed(FUNC nodes) const {
+        return useNodesTime ? TimePoint(nodes()) : elapsed_time();
+    }
+    TimePoint elapsed_time() const { return now() - startTime; };
 
-private:
+    void clear();
-  TimePoint startTime;
+    void advance_nodes_time(std::int64_t nodes);
-  TimePoint optimumTime;
+
-  TimePoint maximumTime;
+   private:
+    TimePoint startTime;
+    TimePoint optimumTime;
+    TimePoint maximumTime;
+
+    std::int64_t availableNodes = -1;     // When in 'nodes as time' mode
+    bool         useNodesTime   = false;  // True if we are in 'nodes as time' mode
 };
 
-extern TimeManagement Time;
+}  // namespace Stockfish
 
-} // namespace Stockfish
+#endif  // #ifndef TIMEMAN_H_INCLUDED
-
-#endif // #ifndef TIMEMAN_H_INCLUDED

src/tt.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,147 +16,236 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#include <cstring>   // For std::memset
-#include <iostream>
-#include <thread>
-
-#include "bitboard.h"
-#include "misc.h"
-#include "thread.h"
 #include "tt.h"
-#include "uci.h"
+
+#include <cassert>
+#include <cstdint>
+#include <cstdlib>
+#include <cstring>
+#include <iostream>
+
+#include "memory.h"
+#include "misc.h"
+#include "syzygy/tbprobe.h"
+#include "thread.h"
 
 namespace Stockfish {
 
-TranspositionTable TT; // Our global transposition table
 
-/// TTEntry::save() populates the TTEntry with a new node's data, possibly
+// TTEntry struct is the 10 bytes transposition table entry, defined as below:
-/// overwriting an old position. Update is not atomic and can be racy.
+//
+// key        16 bit
+// depth       8 bit
+// generation  5 bit
+// pv node     1 bit
+// bound type  2 bit
+// move       16 bit
+// value      16 bit
+// evaluation 16 bit
+//
+// These fields are in the same order as accessed by TT::probe(), since memory is fastest sequentially.
+// Equally, the store order in save() matches this order.
 
-void TTEntry::save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev) {
+struct TTEntry {
 
-  // Preserve any existing move for the same position
+    // Convert internal bitfields to external types
-  if (m || (uint16_t)k != key16)
+    TTData read() const {
-      move16 = (uint16_t)m;
+        return TTData{Move(move16),           Value(value16),
+                      Value(eval16),          Depth(depth8 + DEPTH_ENTRY_OFFSET),
+                      Bound(genBound8 & 0x3), bool(genBound8 & 0x4)};
+    }
 
-  // Overwrite less valuable entries (cheapest checks first)
+    bool is_occupied() const;
-  if (   b == BOUND_EXACT
+    void save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev, uint8_t generation8);
-      || (uint16_t)k != key16
+    // The returned age is a multiple of TranspositionTable::GENERATION_DELTA
-      || d - DEPTH_OFFSET + 2 * pv > depth8 - 4)
+    uint8_t relative_age(const uint8_t generation8) const;
-  {
-      assert(d > DEPTH_OFFSET);
-      assert(d < 256 + DEPTH_OFFSET);
 
-      key16     = (uint16_t)k;
+   private:
-      depth8    = (uint8_t)(d - DEPTH_OFFSET);
+    friend class TranspositionTable;
-      genBound8 = (uint8_t)(TT.generation8 | uint8_t(pv) << 2 | b);
+
-      value16   = (int16_t)v;
+    uint16_t key16;
-      eval16    = (int16_t)ev;
+    uint8_t  depth8;
-  }
+    uint8_t  genBound8;
+    Move     move16;
+    int16_t  value16;
+    int16_t  eval16;
+};
+
+// `genBound8` is where most of the details are. We use the following constants to manipulate 5 leading generation bits
+// and 3 trailing miscellaneous bits.
+
+// These bits are reserved for other things.
+static constexpr unsigned GENERATION_BITS = 3;
+// increment for generation field
+static constexpr int GENERATION_DELTA = (1 << GENERATION_BITS);
+// cycle length
+static constexpr int GENERATION_CYCLE = 255 + GENERATION_DELTA;
+// mask to pull out generation number
+static constexpr int GENERATION_MASK = (0xFF << GENERATION_BITS) & 0xFF;
+
+// DEPTH_ENTRY_OFFSET exists because 1) we use `bool(depth8)` as the occupancy check, but
+// 2) we need to store negative depths for QS. (`depth8` is the only field with "spare bits":
+// we sacrifice the ability to store depths greater than 1<<8 less the offset, as asserted in `save`.)
+bool TTEntry::is_occupied() const { return bool(depth8); }
+
+// Populates the TTEntry with a new node's data, possibly
+// overwriting an old position. The update is not atomic and can be racy.
+void TTEntry::save(
+  Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev, uint8_t generation8) {
+
+    // Preserve the old ttmove if we don't have a new one
+    if (m || uint16_t(k) != key16)
+        move16 = m;
+
+    // Overwrite less valuable entries (cheapest checks first)
+    if (b == BOUND_EXACT || uint16_t(k) != key16 || d - DEPTH_ENTRY_OFFSET + 2 * pv > depth8 - 4
+        || relative_age(generation8))
+    {
+        assert(d > DEPTH_ENTRY_OFFSET);
+        assert(d < 256 + DEPTH_ENTRY_OFFSET);
+
+        key16     = uint16_t(k);
+        depth8    = uint8_t(d - DEPTH_ENTRY_OFFSET);
+        genBound8 = uint8_t(generation8 | uint8_t(pv) << 2 | b);
+        value16   = int16_t(v);
+        eval16    = int16_t(ev);
+    }
 }
 
 
-/// TranspositionTable::resize() sets the size of the transposition table,
+uint8_t TTEntry::relative_age(const uint8_t generation8) const {
-/// measured in megabytes. Transposition table consists of a power of 2 number
+    // Due to our packed storage format for generation and its cyclic
-/// of clusters and each cluster consists of ClusterSize number of TTEntry.
+    // nature we add GENERATION_CYCLE (256 is the modulus, plus what
-
+    // is needed to keep the unrelated lowest n bits from affecting
-void TranspositionTable::resize(size_t mbSize) {
+    // the result) to calculate the entry age correctly even after
-
+    // generation8 overflows into the next cycle.
-  Threads.main()->wait_for_search_finished();
+    return (GENERATION_CYCLE + generation8 - genBound8) & GENERATION_MASK;
-
-  aligned_large_pages_free(table);
-
-  clusterCount = mbSize * 1024 * 1024 / sizeof(Cluster);
-
-  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();
 }
 
 
-/// TranspositionTable::clear() initializes the entire transposition table to zero,
+// TTWriter is but a very thin wrapper around the pointer
-//  in a multi-threaded way.
+TTWriter::TTWriter(TTEntry* tte) :
+    entry(tte) {}
 
-void TranspositionTable::clear() {
+void TTWriter::write(
-
+  Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev, uint8_t generation8) {
-  std::vector<std::thread> threads;
+    entry->save(k, v, pv, b, d, m, ev, generation8);
-
-  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::probe() looks up the current position in the transposition
+// A TranspositionTable is an array of Cluster, of size clusterCount. Each cluster consists of ClusterSize number
-/// table. It returns true and a pointer to the TTEntry if the position is found.
+// of TTEntry. Each non-empty TTEntry contains information on exactly one position. The size of a Cluster should
-/// Otherwise, it returns false and a pointer to an empty or least valuable TTEntry
+// divide the size of a cache line for best performance, as the cacheline is prefetched when possible.
-/// 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.
 
-TTEntry* TranspositionTable::probe(const Key key, bool& found) const {
+static constexpr int ClusterSize = 3;
 
-  TTEntry* const tte = first_entry(key);
+struct Cluster {
-  const uint16_t key16 = (uint16_t)key;  // Use the low 16 bits as key inside the cluster
+    TTEntry entry[ClusterSize];
+    char    padding[2];  // Pad to 32 bytes
+};
 
-  for (int i = 0; i < ClusterSize; ++i)
+static_assert(sizeof(Cluster) == 32, "Suboptimal Cluster size");
-      if (tte[i].key16 == key16 || !tte[i].depth8)
-      {
-          tte[i].genBound8 = uint8_t(generation8 | (tte[i].genBound8 & (GENERATION_DELTA - 1))); // Refresh
 
-          return found = (bool)tte[i].depth8, &tte[i];
-      }
 
-  // Find an entry to be replaced according to the replacement strategy
+// Sets the size of the transposition table,
-  TTEntry* replace = tte;
+// measured in megabytes. Transposition table consists
-  for (int i = 1; i < ClusterSize; ++i)
+// of clusters and each cluster consists of ClusterSize number of TTEntry.
-      // Due to our packed storage format for generation and its cyclic
+void TranspositionTable::resize(size_t mbSize, ThreadPool& threads) {
-      // nature we add GENERATION_CYCLE (256 is the modulus, plus what
+    aligned_large_pages_free(table);
-      // is needed to keep the unrelated lowest n bits from affecting
-      // the result) to calculate the entry age correctly even after
-      // generation8 overflows into the next cycle.
-      if (  replace->depth8 - ((GENERATION_CYCLE + generation8 - replace->genBound8) & GENERATION_MASK)
-          >   tte[i].depth8 - ((GENERATION_CYCLE + generation8 -   tte[i].genBound8) & GENERATION_MASK))
-          replace = &tte[i];
 
-  return found = false, replace;
+    clusterCount = mbSize * 1024 * 1024 / sizeof(Cluster);
+
+    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(threads);
 }
 
 
-/// TranspositionTable::hashfull() returns an approximation of the hashtable
+// Initializes the entire transposition table to zero,
-/// occupation during a search. The hash is x permill full, as per UCI protocol.
+// in a multi-threaded way.
+void TranspositionTable::clear(ThreadPool& threads) {
+    generation8              = 0;
+    const size_t threadCount = threads.num_threads();
 
-int TranspositionTable::hashfull() const {
+    for (size_t i = 0; i < threadCount; ++i)
+    {
+        threads.run_on_thread(i, [this, i, threadCount]() {
+            // Each thread will zero its part of the hash table
+            const size_t stride = clusterCount / threadCount;
+            const size_t start  = stride * i;
+            const size_t len    = i + 1 != threadCount ? stride : clusterCount - start;
 
-  int cnt = 0;
+            std::memset(&table[start], 0, len * sizeof(Cluster));
-  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;
 
-  return cnt / ClusterSize;
+    for (size_t i = 0; i < threadCount; ++i)
+        threads.wait_on_thread(i);
 }
 
-} // namespace Stockfish
+
+// Returns an approximation of the hashtable
+// occupation during a search. The hash is x permill full, as per UCI protocol.
+// Only counts entries which match the current generation.
+int TranspositionTable::hashfull(int maxAge) const {
+    int maxAgeInternal = maxAge << GENERATION_BITS;
+    int cnt            = 0;
+    for (int i = 0; i < 1000; ++i)
+        for (int j = 0; j < ClusterSize; ++j)
+            cnt += table[i].entry[j].is_occupied()
+                && table[i].entry[j].relative_age(generation8) <= maxAgeInternal;
+
+    return cnt / ClusterSize;
+}
+
+
+void TranspositionTable::new_search() {
+    // increment by delta to keep lower bits as is
+    generation8 += GENERATION_DELTA;
+}
+
+
+uint8_t TranspositionTable::generation() const { return generation8; }
+
+
+// Looks up the current position in the transposition
+// table. It returns true 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.
+std::tuple<bool, TTData, TTWriter> TranspositionTable::probe(const Key key) const {
+
+    TTEntry* const tte   = first_entry(key);
+    const uint16_t key16 = uint16_t(key);  // Use the low 16 bits as key inside the cluster
+
+    for (int i = 0; i < ClusterSize; ++i)
+        if (tte[i].key16 == key16)
+            // This gap is the main place for read races.
+            // After `read()` completes that copy is final, but may be self-inconsistent.
+            return {tte[i].is_occupied(), tte[i].read(), TTWriter(&tte[i])};
+
+    // Find an entry to be replaced according to the replacement strategy
+    TTEntry* replace = tte;
+    for (int i = 1; i < ClusterSize; ++i)
+        if (replace->depth8 - replace->relative_age(generation8) * 2
+            > tte[i].depth8 - tte[i].relative_age(generation8) * 2)
+            replace = &tte[i];
+
+    return {false,
+            TTData{Move::none(), VALUE_NONE, VALUE_NONE, DEPTH_ENTRY_OFFSET, BOUND_NONE, false},
+            TTWriter(replace)};
+}
+
+
+TTEntry* TranspositionTable::first_entry(const Key key) const {
+    return &table[mul_hi64(key, clusterCount)].entry[0];
+}
+
+}  // namespace Stockfish

src/tt.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,89 +19,92 @@
 #ifndef TT_H_INCLUDED
 #define TT_H_INCLUDED
 
-#include "misc.h"
+#include <cstddef>
+#include <cstdint>
+#include <tuple>
+
+#include "memory.h"
 #include "types.h"
 
 namespace Stockfish {
 
-/// TTEntry struct is the 10 bytes transposition table entry, defined as below:
+class ThreadPool;
-///
+struct TTEntry;
-/// key        16 bit
+struct Cluster;
-/// depth       8 bit
-/// generation  5 bit
-/// pv node     1 bit
-/// bound type  2 bit
-/// move       16 bit
-/// value      16 bit
-/// eval value 16 bit
 
-struct TTEntry {
+// There is only one global hash table for the engine and all its threads. For chess in particular, we even allow racy
+// updates between threads to and from the TT, as taking the time to synchronize access would cost thinking time and
+// thus elo. As a hash table, collisions are possible and may cause chess playing issues (bizarre blunders, faulty mate
+// reports, etc). Fixing these also loses elo; however such risk decreases quickly with larger TT size.
+//
+// `probe` is the primary method: given a board position, we lookup its entry in the table, and return a tuple of:
+//   1) whether the entry already has this position
+//   2) a copy of the prior data (if any) (may be inconsistent due to read races)
+//   3) a writer object to this entry
+// The copied data and the writer are separated to maintain clear boundaries between local vs global objects.
 
-  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:
+// A copy of the data already in the entry (possibly collided). `probe` may be racy, resulting in inconsistent data.
-  friend class TranspositionTable;
+struct TTData {
+    Move  move;
+    Value value, eval;
+    Depth depth;
+    Bound bound;
+    bool  is_pv;
 
-  uint16_t key16;
+    TTData() = delete;
-  uint8_t  depth8;
+
-  uint8_t  genBound8;
+    // clang-format off
-  uint16_t move16;
+    TTData(Move m, Value v, Value ev, Depth d, Bound b, bool pv) :
-  int16_t  value16;
+        move(m),
-  int16_t  eval16;
+        value(v),
+        eval(ev),
+        depth(d),
+        bound(b),
+        is_pv(pv) {};
+    // clang-format on
 };
 
 
-/// A TranspositionTable is an array of Cluster, of size clusterCount. Each
+// This is used to make racy writes to the global TT.
-/// cluster consists of ClusterSize number of TTEntry. Each non-empty TTEntry
+struct TTWriter {
-/// contains information on exactly one position. The size of a Cluster should
+   public:
-/// divide the size of a cache line for best performance, as the cacheline is
+    void write(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev, uint8_t generation8);
-/// prefetched when possible.
+
+   private:
+    friend class TranspositionTable;
+    TTEntry* entry;
+    TTWriter(TTEntry* tte);
+};
+
 
 class TranspositionTable {
 
-  static constexpr int ClusterSize = 3;
+   public:
+    ~TranspositionTable() { aligned_large_pages_free(table); }
 
-  struct Cluster {
+    void resize(size_t mbSize, ThreadPool& threads);  // Set TT size
-    TTEntry entry[ClusterSize];
+    void clear(ThreadPool& threads);                  // Re-initialize memory, multithreaded
-    char padding[2]; // Pad to 32 bytes
+    int  hashfull(int maxAge = 0)
-  };
+      const;  // Approximate what fraction of entries (permille) have been written to during this root search
 
-  static_assert(sizeof(Cluster) == 32, "Unexpected Cluster size");
+    void
+    new_search();  // This must be called at the beginning of each root search to track entry aging
+    uint8_t generation() const;  // The current age, used when writing new data to the TT
+    std::tuple<bool, TTData, TTWriter>
+    probe(const Key key) const;  // The main method, whose retvals separate local vs global objects
+    TTEntry* first_entry(const Key key)
+      const;  // This is the hash function; its only external use is memory prefetching.
 
-  // Constants used to refresh the hash table periodically
+   private:
-  static constexpr unsigned GENERATION_BITS  = 3;                                // nb of bits reserved for other things
+    friend struct TTEntry;
-  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:
+    size_t   clusterCount;
- ~TranspositionTable() { aligned_large_pages_free(table); }
+    Cluster* table = nullptr;
-  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();
 
-  TTEntry* first_entry(const Key key) const {
+    uint8_t generation8 = 0;  // Size must be not bigger than TTEntry::genBound8
-    return &table[mul_hi64(key, clusterCount)].entry[0];
-  }
-
-private:
-  friend struct TTEntry;
-
-  size_t clusterCount;
-  Cluster* table;
-  uint8_t generation8; // Size must be not bigger than TTEntry::genBound8
 };
 
-extern TranspositionTable TT;
+}  // namespace Stockfish
 
-} // namespace Stockfish
+#endif  // #ifndef TT_H_INCLUDED
-
-#endif // #ifndef TT_H_INCLUDED

src/tune.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,100 +16,97 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#include "tune.h"
+
 #include <algorithm>
 #include <iostream>
+#include <map>
+#include <optional>
 #include <sstream>
+#include <string>
 
-#include "types.h"
+#include "ucioption.h"
-#include "misc.h"
-#include "uci.h"
 
 using std::string;
 
 namespace Stockfish {
 
-bool Tune::update_on_last;
+bool          Tune::update_on_last;
-const UCI::Option* LastOption = nullptr;
+const Option* LastOption = nullptr;
-static std::map<std::string, int> TuneResults;
+OptionsMap*   Tune::options;
+namespace {
+std::map<std::string, int> TuneResults;
+
+std::optional<std::string> on_tune(const Option& o) {
+
+    if (!Tune::update_on_last || LastOption == &o)
+        Tune::read_options();
+
+    return std::nullopt;
+}
+}
+
+void Tune::make_option(OptionsMap* opts, 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];
+
+    opts->add(n, Option(v, r(v).first, r(v).second, on_tune));
+    LastOption = &((*opts)[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;
+}
 
 string Tune::next(string& names, bool pop) {
 
-  string name;
+    string name;
 
-  do {
+    do
-      string token = names.substr(0, names.find(','));
+    {
+        string token = names.substr(0, names.find(','));
 
-      if (pop)
+        if (pop)
-          names.erase(0, token.size() + 1);
+            names.erase(0, token.size() + 1);
 
-      std::stringstream ws(token);
+        std::stringstream ws(token);
-      name += (ws >> token, token); // Remove trailing whitespace
+        name += (ws >> token, token);  // Remove trailing whitespace
 
-  } while (  std::count(name.begin(), name.end(), '(')
+    } while (std::count(name.begin(), name.end(), '(') - std::count(name.begin(), name.end(), ')'));
-           - std::count(name.begin(), name.end(), ')'));
 
-  return name;
+    return name;
 }
 
-static void on_tune(const UCI::Option& o) {
 
-  if (!Tune::update_on_last || LastOption == &o)
+template<>
-      Tune::read_options();
+void Tune::Entry<int>::init_option() {
+    make_option(options, name, value, range);
 }
 
-static void make_option(const string& n, int v, const SetRange& r) {
+template<>
-
+void Tune::Entry<int>::read_option() {
-  // Do not generate option when there is nothing to tune (ie. min = max)
+    if (options->count(name))
-  if (r(v).first == r(v).second)
+        value = int((*options)[name]);
-      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<>
-template<> void Tune::Entry<Tune::PostUpdate>::read_option() { value(); }
+void Tune::Entry<Tune::PostUpdate>::init_option() {}
+template<>
+void Tune::Entry<Tune::PostUpdate>::read_option() {
+    value();
+}
 
-} // namespace Stockfish
+}  // namespace Stockfish
 
 
 // Init options with tuning session results instead of default values. Useful to
@@ -121,13 +118,9 @@ template<> void Tune::Entry<Tune::PostUpdate>::read_option() { value(); }
 //
 // Then paste the output below, as the function body
 
-#include <cmath>
 
 namespace Stockfish {
 
-void Tune::read_results() {
+void Tune::read_results() { /* ...insert your values here... */ }
 
-  /* ...insert your values here... */
+}  // namespace Stockfish
-}
-
-} // namespace Stockfish

src/tune.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,145 +19,165 @@
 #ifndef TUNE_H_INCLUDED
 #define TUNE_H_INCLUDED
 
+#include <cstddef>
 #include <memory>
 #include <string>
-#include <type_traits>
+#include <type_traits>  // IWYU pragma: keep
+#include <utility>
 #include <vector>
 
 namespace Stockfish {
 
-using Range = std::pair<int, int>; // Option's min-max values
+class OptionsMap;
-using RangeFun = Range (int);
+
+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) {
+inline Range default_range(int v) { return v > 0 ? Range(0, 2 * v) : Range(2 * v, 0); }
-  return v > 0 ? Range(0, 2 * v) : Range(2 * v, 0);
-}
 
 struct SetRange {
-  explicit SetRange(RangeFun f) : fun(f) {}
+    explicit SetRange(RangeFun f) :
-  SetRange(int min, int max) : fun(nullptr), range(min, max) {}
+        fun(f) {}
-  Range operator()(int v) const { return fun ? fun(v) : range; }
+    SetRange(int min, int max) :
+        fun(nullptr),
+        range(min, max) {}
+    Range operator()(int v) const { return fun ? fun(v) : range; }
 
-  RangeFun* fun;
+    RangeFun* fun;
-  Range range;
+    Range     range;
 };
 
 #define SetDefaultRange SetRange(default_range)
 
 
-/// Tune class implements the 'magic' code that makes the setup of a fishtest
+// Tune class implements the 'magic' code that makes the setup of a fishtest tuning
-/// tuning session as easy as it can be. Mainly you have just to remove const
+// session as easy as it can be. Mainly you have just to remove const qualifiers
-/// qualifiers from the variables you want to tune and flag them for tuning, so
+// from the variables you want to tune and flag them for tuning, so if you have:
-/// if you have:
+//
-///
+//   const Value myValue[][2] = { { V(100), V(20) }, { V(7), V(78) } };
-///   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
-/// If you have a my_post_update() function to run after values have been updated,
+// remove the 'const' qualifiers and write somewhere below in the file:
-/// 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), myValue, my_post_update);
-///
+//
-///   TUNE(SetRange(my_range), myScore, myValue, my_post_update);
+// You can also set the range directly, and restore the default at the end
-///
+//
-/// You can also set the range directly, and restore the default at the end
+//   TUNE(SetRange(-100, 100), myValue, SetDefaultRange);
-///
+//
-///   TUNE(SetRange(-100, 100), myScore, SetDefaultRange);
+// In case update function is slow and you have many parameters, you can add:
-///
+//
-/// In case update function is slow and you have many parameters, you can add:
+//   UPDATE_ON_LAST();
-///
+//
-///   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 the values update, including post update function call, will be done only
+// and created as the last one, so the GUI should send the options in the same
-/// once, after the engine receives the last UCI option, that is the one defined
+// order in which have been 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
+    using PostUpdate = void();  // Post-update function
 
-  Tune() { read_results(); }
+    Tune() { read_results(); }
-  Tune(const Tune&) = delete;
+    Tune(const Tune&)           = delete;
-  void operator=(const Tune&) = delete;
+    void operator=(const Tune&) = delete;
-  void read_results();
+    void read_results();
 
-  static Tune& instance() { static Tune t; return t; } // Singleton
+    static Tune& instance() {
+        static Tune t;
+        return t;
+    }  // Singleton
 
-  // Use polymorphism to accommodate Entry of different types in the same vector
+    // Use polymorphism to accommodate Entry of different types in the same vector
-  struct EntryBase {
+    struct EntryBase {
-    virtual ~EntryBase() = default;
+        virtual ~EntryBase()       = default;
-    virtual void init_option() = 0;
+        virtual void init_option() = 0;
-    virtual void read_option() = 0;
+        virtual void read_option() = 0;
-  };
+    };
 
-  template<typename T>
+    template<typename T>
-  struct Entry : public EntryBase {
+    struct Entry: public EntryBase {
 
-    static_assert(!std::is_const<T>::value, "Parameter cannot be const!");
+        static_assert(!std::is_const_v<T>, "Parameter cannot be const!");
 
-    static_assert(   std::is_same<T,   int>::value
+        static_assert(std::is_same_v<T, int> || std::is_same_v<T, PostUpdate>,
-                  || std::is_same<T, Value>::value
+                      "Parameter type not supported!");
-                  || 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) {}
+        Entry(const std::string& n, T& v, const SetRange& r) :
-    void operator=(const Entry&) = delete; // Because 'value' is a reference
+            name(n),
-    void init_option() override;
+            value(v),
-    void read_option() override;
+            range(r) {}
+        void operator=(const Entry&) = delete;  // Because 'value' is a reference
+        void init_option() override;
+        void read_option() override;
 
-    std::string name;
+        std::string name;
-    T& value;
+        T&          value;
-    SetRange range;
+        SetRange    range;
-  };
+    };
 
-  // Our facility to fill the container, each Entry corresponds to a parameter
+    // 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
+    // to tune. We use variadic templates to deal with an unspecified number of
-  // entries, each one of a possible different type.
+    // entries, each one of a possible different type.
-  static std::string next(std::string& names, bool pop = true);
+    static std::string next(std::string& names, bool pop = true);
 
-  int add(const SetRange&, std::string&&) { return 0; }
+    int add(const SetRange&, std::string&&) { return 0; }
 
-  template<typename T, typename... Args>
+    template<typename T, typename... Args>
-  int add(const SetRange& range, std::string&& names, T& value, Args&&... 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)));
+        list.push_back(std::unique_ptr<EntryBase>(new Entry<T>(next(names), value, range)));
-    return add(range, std::move(names), args...);
+        return add(range, std::move(names), args...);
-  }
+    }
 
-  // Template specialization for arrays: recursively handle multi-dimensional arrays
+    // Template specialization for arrays: recursively handle multi-dimensional arrays
-  template<typename T, size_t N, typename... Args>
+    template<typename T, size_t N, typename... Args>
-  int add(const SetRange& range, std::string&& names, T (&value)[N], Args&&... args) {
+    int add(const SetRange& range, std::string&& names, T (&value)[N], Args&&... args) {
-    for (size_t i = 0; i < N; i++)
+        for (size_t i = 0; i < N; i++)
-        add(range, next(names, i == N - 1) + "[" + std::to_string(i) + "]", value[i]);
+            add(range, next(names, i == N - 1) + "[" + std::to_string(i) + "]", value[i]);
-    return add(range, std::move(names), args...);
+        return add(range, std::move(names), args...);
-  }
+    }
 
-  // Template specialization for SetRange
+    // Template specialization for SetRange
-  template<typename... Args>
+    template<typename... Args>
-  int add(const SetRange&, std::string&& names, SetRange& value, Args&&... args) {
+    int add(const SetRange&, std::string&& names, SetRange& value, Args&&... args) {
-    return add(value, (next(names), std::move(names)), args...);
+        return add(value, (next(names), std::move(names)), args...);
-  }
+    }
 
-  std::vector<std::unique_ptr<EntryBase>> list;
+    static void make_option(OptionsMap* options, const std::string& n, int v, const SetRange& r);
 
-public:
+    std::vector<std::unique_ptr<EntryBase>> list;
-  template<typename... Args>
+
-  static int add(const std::string& names, Args&&... args) {
+   public:
-    return instance().add(SetDefaultRange, names.substr(1, names.size() - 2), args...); // Remove trailing parenthesis
+    template<typename... Args>
-  }
+    static int add(const std::string& names, Args&&... args) {
-  static void init() { for (auto& e : instance().list) e->init_option(); read_options(); } // Deferred, due to UCI::Options access
+        return instance().add(SetDefaultRange, names.substr(1, names.size() - 2),
-  static void read_options() { for (auto& e : instance().list) e->read_option(); }
+                              args...);  // Remove trailing parenthesis
-  static bool update_on_last;
+    }
+    static void init(OptionsMap& o) {
+        options = &o;
+        for (auto& e : instance().list)
+            e->init_option();
+        read_options();
+    }  // Deferred, due to UCIEngine::Options access
+    static void read_options() {
+        for (auto& e : instance().list)
+            e->read_option();
+    }
+
+    static bool        update_on_last;
+    static OptionsMap* options;
 };
 
-// Some macro magic :-) we define a dummy int variable that compiler initializes calling Tune::add()
+// Some macro magic :-) we define a dummy int variable that the compiler initializes calling Tune::add()
 #define STRINGIFY(x) #x
-#define UNIQUE2(x, y) x ## y
+#define UNIQUE2(x, y) x##y
-#define UNIQUE(x, y) UNIQUE2(x, y) // Two indirection levels to expand __LINE__
+#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
+}  // namespace Stockfish
 
-#endif // #ifndef TUNE_H_INCLUDED
+#endif  // #ifndef TUNE_H_INCLUDED

src/types.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -17,475 +17,420 @@
 */
 
 #ifndef TYPES_H_INCLUDED
-#define TYPES_H_INCLUDED
+    #define TYPES_H_INCLUDED
 
-/// When compiling with provided Makefile (e.g. for Linux and OSX), configuration
+// When compiling with provided Makefile (e.g. for Linux and OSX), configuration
-/// is done automatically. To get started type 'make help'.
+// is done automatically. To get started type 'make help'.
-///
+//
-/// When Makefile is not used (e.g. with Microsoft Visual Studio) some switches
+// When Makefile is not used (e.g. with Microsoft Visual Studio) some switches
-/// need to be set manually:
+// need to be set manually:
-///
+//
-/// -DNDEBUG      | Disable debugging mode. Always use this for release.
+// -DNDEBUG      | Disable debugging mode. Always use this for release.
-///
+//
-/// -DNO_PREFETCH | Disable use of prefetch asm-instruction. You may need this to
+// -DNO_PREFETCH | Disable use of prefetch asm-instruction. You may need this to
-///               | run on some very old machines.
+//               | run on some very old machines.
-///
+//
-/// -DUSE_POPCNT  | Add runtime support for use of popcnt asm-instruction. Works
+// -DUSE_POPCNT  | Add runtime support for use of popcnt asm-instruction. Works
-///               | only in 64-bit mode and 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
+// -DUSE_PEXT    | Add runtime support for use of pext asm-instruction. Works
-///               | only in 64-bit mode and requires hardware with pext support.
+//               | only in 64-bit mode and requires hardware with pext support.
 
-#include <cassert>
+    #include <cassert>
-#include <cctype>
+    #include <cstdint>
-#include <cstdint>
-#include <cstdlib>
-#include <algorithm>
 
-#if defined(_MSC_VER)
+    #if defined(_MSC_VER)
-// Disable some silly and noisy warning from MSVC compiler
+        // Disable some silly and noisy warnings from MSVC compiler
-#pragma warning(disable: 4127) // Conditional expression is constant
+        #pragma warning(disable: 4127)  // Conditional expression is constant
-#pragma warning(disable: 4146) // Unary minus operator applied to unsigned type
+        #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: 4800)  // Forcing value to bool 'true' or 'false'
-#endif
+    #endif
 
-/// Predefined macros hell:
+// Predefined macros hell:
-///
+//
-/// __GNUC__           Compiler is gcc, Clang or Intel on Linux
+// __GNUC__                Compiler is GCC, Clang or ICX
-/// __INTEL_COMPILER   Compiler is Intel
+// __clang__               Compiler is Clang or ICX
-/// _MSC_VER           Compiler is MSVC or Intel on Windows
+// __INTEL_LLVM_COMPILER   Compiler is ICX
-/// _WIN32             Building on Windows (any)
+// _MSC_VER                Compiler is MSVC
-/// _WIN64             Building on Windows 64 bit
+// _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__)
+    #if defined(__GNUC__) && (__GNUC__ < 9 || (__GNUC__ == 9 && __GNUC_MINOR__ <= 2)) \
-#define ALIGNAS_ON_STACK_VARIABLES_BROKEN
+      && defined(_WIN32) && !defined(__clang__)
-#endif
+        #define ALIGNAS_ON_STACK_VARIABLES_BROKEN
+    #endif
 
-#define ASSERT_ALIGNED(ptr, alignment) assert(reinterpret_cast<uintptr_t>(ptr) % alignment == 0)
+    #define ASSERT_ALIGNED(ptr, alignment) assert(reinterpret_cast<uintptr_t>(ptr) % alignment == 0)
 
-#if defined(_WIN64) && defined(_MSC_VER) // No Makefile used
+    #if defined(_WIN64) && defined(_MSC_VER)  // No Makefile used
-#  include <intrin.h> // Microsoft header for _BitScanForward64()
+        #include <intrin.h>                   // Microsoft header for _BitScanForward64()
-#  define IS_64BIT
+        #define IS_64BIT
-#endif
+    #endif
 
-#if defined(USE_POPCNT) && (defined(__INTEL_COMPILER) || defined(_MSC_VER))
+    #if defined(USE_POPCNT) && defined(_MSC_VER)
-#  include <nmmintrin.h> // Intel and Microsoft header for _mm_popcnt_u64()
+        #include <nmmintrin.h>  // Microsoft header for _mm_popcnt_u64()
-#endif
+    #endif
 
-#if !defined(NO_PREFETCH) && (defined(__INTEL_COMPILER) || defined(_MSC_VER))
+    #if !defined(NO_PREFETCH) && defined(_MSC_VER)
-#  include <xmmintrin.h> // Intel and Microsoft header for _mm_prefetch()
+        #include <xmmintrin.h>  // Microsoft header for _mm_prefetch()
-#endif
+    #endif
 
-#if defined(USE_PEXT)
+    #if defined(USE_PEXT)
-#  include <immintrin.h> // Header for _pext_u64() intrinsic
+        #include <immintrin.h>  // Header for _pext_u64() intrinsic
-#  define pext(b, m) _pext_u64(b, m)
+        #define pext(b, m) _pext_u64(b, m)
-#else
+    #else
-#  define pext(b, m) 0
+        #define pext(b, m) 0
-#endif
+    #endif
 
 namespace Stockfish {
 
-#ifdef USE_POPCNT
+    #ifdef USE_POPCNT
 constexpr bool HasPopCnt = true;
-#else
+    #else
 constexpr bool HasPopCnt = false;
-#endif
+    #endif
 
-#ifdef USE_PEXT
+    #ifdef USE_PEXT
 constexpr bool HasPext = true;
-#else
+    #else
 constexpr bool HasPext = false;
-#endif
+    #endif
 
-#ifdef IS_64BIT
+    #ifdef IS_64BIT
 constexpr bool Is64Bit = true;
-#else
+    #else
 constexpr bool Is64Bit = false;
-#endif
+    #endif
 
-using Key = uint64_t;
+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), 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 : int {
-  MOVE_NONE,
-  MOVE_NULL = 65
-};
-
-enum MoveType {
-  NORMAL,
-  PROMOTION = 1 << 14,
-  EN_PASSANT = 2 << 14,
-  CASTLING  = 3 << 14
-};
-
 enum Color {
-  WHITE, BLACK, COLOR_NB = 2
+    WHITE,
+    BLACK,
+    COLOR_NB = 2
 };
 
 enum CastlingRights {
-  NO_CASTLING,
+    NO_CASTLING,
-  WHITE_OO,
+    WHITE_OO,
-  WHITE_OOO = WHITE_OO << 1,
+    WHITE_OOO = WHITE_OO << 1,
-  BLACK_OO  = WHITE_OO << 2,
+    BLACK_OO  = WHITE_OO << 2,
-  BLACK_OOO = WHITE_OO << 3,
+    BLACK_OOO = WHITE_OO << 3,
 
-  KING_SIDE      = WHITE_OO  | BLACK_OO,
+    KING_SIDE      = WHITE_OO | BLACK_OO,
-  QUEEN_SIDE     = WHITE_OOO | BLACK_OOO,
+    QUEEN_SIDE     = WHITE_OOO | BLACK_OOO,
-  WHITE_CASTLING = WHITE_OO  | WHITE_OOO,
+    WHITE_CASTLING = WHITE_OO | WHITE_OOO,
-  BLACK_CASTLING = BLACK_OO  | BLACK_OOO,
+    BLACK_CASTLING = BLACK_OO | BLACK_OOO,
-  ANY_CASTLING   = WHITE_CASTLING | BLACK_CASTLING,
+    ANY_CASTLING   = WHITE_CASTLING | BLACK_CASTLING,
 
-  CASTLING_RIGHT_NB = 16
+    CASTLING_RIGHT_NB = 16
-};
-
-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_NONE,
-  BOUND_UPPER,
+    BOUND_UPPER,
-  BOUND_LOWER,
+    BOUND_LOWER,
-  BOUND_EXACT = BOUND_UPPER | BOUND_LOWER
+    BOUND_EXACT = BOUND_UPPER | BOUND_LOWER
 };
 
-enum Value : int {
+// Value is used as an alias for int, this is done to differentiate between a search
-  VALUE_ZERO      = 0,
+// value and any other integer value. The values used in search are always supposed
-  VALUE_DRAW      = 0,
+// to be in the range (-VALUE_NONE, VALUE_NONE] and should not exceed this range.
-  VALUE_KNOWN_WIN = 10000,
+using Value = int;
-  VALUE_MATE      = 32000,
-  VALUE_INFINITE  = 32001,
-  VALUE_NONE      = 32002,
 
-  VALUE_TB_WIN_IN_MAX_PLY  =  VALUE_MATE - 2 * MAX_PLY,
+constexpr Value VALUE_ZERO     = 0;
-  VALUE_TB_LOSS_IN_MAX_PLY = -VALUE_TB_WIN_IN_MAX_PLY,
+constexpr Value VALUE_DRAW     = 0;
-  VALUE_MATE_IN_MAX_PLY  =  VALUE_MATE - MAX_PLY,
+constexpr Value VALUE_NONE     = 32002;
-  VALUE_MATED_IN_MAX_PLY = -VALUE_MATE_IN_MAX_PLY,
+constexpr Value VALUE_INFINITE = 32001;
 
-  // In the code, we make the assumption that these values
+constexpr Value VALUE_MATE             = 32000;
-  // are such that non_pawn_material() can be used to uniquely
+constexpr Value VALUE_MATE_IN_MAX_PLY  = VALUE_MATE - MAX_PLY;
-  // identify the material on the board.
+constexpr Value VALUE_MATED_IN_MAX_PLY = -VALUE_MATE_IN_MAX_PLY;
-  PawnValueMg   = 126,   PawnValueEg   = 208,
-  KnightValueMg = 781,   KnightValueEg = 854,
-  BishopValueMg = 825,   BishopValueEg = 915,
-  RookValueMg   = 1276,  RookValueEg   = 1380,
-  QueenValueMg  = 2538,  QueenValueEg  = 2682,
 
-  MidgameLimit  = 15258, EndgameLimit  = 3915
+constexpr Value VALUE_TB                 = VALUE_MATE_IN_MAX_PLY - 1;
-};
+constexpr Value VALUE_TB_WIN_IN_MAX_PLY  = VALUE_TB - MAX_PLY;
+constexpr Value VALUE_TB_LOSS_IN_MAX_PLY = -VALUE_TB_WIN_IN_MAX_PLY;
 
+
+constexpr bool is_valid(Value value) { return value != VALUE_NONE; }
+
+constexpr bool is_win(Value value) {
+    assert(is_valid(value));
+    return value >= VALUE_TB_WIN_IN_MAX_PLY;
+}
+
+constexpr bool is_loss(Value value) {
+    assert(is_valid(value));
+    return value <= VALUE_TB_LOSS_IN_MAX_PLY;
+}
+
+constexpr bool is_decisive(Value value) { return is_win(value) || is_loss(value); }
+
+// 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.
+constexpr Value PawnValue   = 208;
+constexpr Value KnightValue = 781;
+constexpr Value BishopValue = 825;
+constexpr Value RookValue   = 1276;
+constexpr Value QueenValue  = 2538;
+
+
+// clang-format off
 enum PieceType {
-  NO_PIECE_TYPE, PAWN, KNIGHT, BISHOP, ROOK, QUEEN, KING,
+    NO_PIECE_TYPE, PAWN, KNIGHT, BISHOP, ROOK, QUEEN, KING,
-  ALL_PIECES = 0,
+    ALL_PIECES = 0,
-  PIECE_TYPE_NB = 8
+    PIECE_TYPE_NB = 8
 };
 
 enum Piece {
-  NO_PIECE,
+    NO_PIECE,
-  W_PAWN = PAWN,     W_KNIGHT, W_BISHOP, W_ROOK, W_QUEEN, W_KING,
+    W_PAWN = PAWN,     W_KNIGHT, W_BISHOP, W_ROOK, W_QUEEN, W_KING,
-  B_PAWN = PAWN + 8, B_KNIGHT, B_BISHOP, B_ROOK, B_QUEEN, B_KING,
+    B_PAWN = PAWN + 8, B_KNIGHT, B_BISHOP, B_ROOK, B_QUEEN, B_KING,
-  PIECE_NB = 16
+    PIECE_NB = 16
 };
+// clang-format on
 
-constexpr Value PieceValue[PHASE_NB][PIECE_NB] = {
+constexpr Value PieceValue[PIECE_NB] = {
-  { VALUE_ZERO, PawnValueMg, KnightValueMg, BishopValueMg, RookValueMg, QueenValueMg, VALUE_ZERO, VALUE_ZERO,
+  VALUE_ZERO, PawnValue, KnightValue, BishopValue, RookValue, QueenValue, VALUE_ZERO, VALUE_ZERO,
-    VALUE_ZERO, PawnValueMg, KnightValueMg, BishopValueMg, RookValueMg, QueenValueMg, VALUE_ZERO, VALUE_ZERO },
+  VALUE_ZERO, PawnValue, KnightValue, BishopValue, RookValue, QueenValue, 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,
+    // The following DEPTH_ constants are used for transposition table entries
-  DEPTH_QS_NO_CHECKS  = -1,
+    // and quiescence search move generation stages. In regular search, the
-  DEPTH_QS_RECAPTURES = -5,
+    // depth stored in the transposition table is literal: the search depth
-
+    // (effort) used to make the corresponding transposition table value. In
-  DEPTH_NONE   = -6,
+    // quiescence search, however, the transposition table entries only store
-
+    // the current quiescence move generation stage (which should thus compare
-  DEPTH_OFFSET = -7 // value used only for TT entry occupancy check
+    // lower than any regular search depth).
+    DEPTH_QS = 0,
+    // For transposition table entries where no searching at all was done
+    // (whether regular or qsearch) we use DEPTH_UNSEARCHED, which should thus
+    // compare lower than any quiescence or regular depth. DEPTH_ENTRY_OFFSET
+    // is used only for the transposition table entry occupancy check (see tt.cpp),
+    // and should thus be lower than DEPTH_UNSEARCHED.
+    DEPTH_UNSEARCHED   = -2,
+    DEPTH_ENTRY_OFFSET = -3
 };
 
+// clang-format off
 enum Square : int {
-  SQ_A1, SQ_B1, SQ_C1, SQ_D1, SQ_E1, SQ_F1, SQ_G1, SQ_H1,
+    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_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,
+    SQ_A3, SQ_B3, SQ_C3, SQ_D3, SQ_E3, SQ_F3, SQ_G3, SQ_H3,
-  SQ_A4, SQ_B4, SQ_C4, SQ_D4, SQ_E4, SQ_F4, SQ_G4, SQ_H4,
+    SQ_A4, SQ_B4, SQ_C4, SQ_D4, SQ_E4, SQ_F4, SQ_G4, SQ_H4,
-  SQ_A5, SQ_B5, SQ_C5, SQ_D5, SQ_E5, SQ_F5, SQ_G5, SQ_H5,
+    SQ_A5, SQ_B5, SQ_C5, SQ_D5, SQ_E5, SQ_F5, SQ_G5, SQ_H5,
-  SQ_A6, SQ_B6, SQ_C6, SQ_D6, SQ_E6, SQ_F6, SQ_G6, SQ_H6,
+    SQ_A6, SQ_B6, SQ_C6, SQ_D6, SQ_E6, SQ_F6, SQ_G6, SQ_H6,
-  SQ_A7, SQ_B7, SQ_C7, SQ_D7, SQ_E7, SQ_F7, SQ_G7, SQ_H7,
+    SQ_A7, SQ_B7, SQ_C7, SQ_D7, SQ_E7, SQ_F7, SQ_G7, SQ_H7,
-  SQ_A8, SQ_B8, SQ_C8, SQ_D8, SQ_E8, SQ_F8, SQ_G8, SQ_H8,
+    SQ_A8, SQ_B8, SQ_C8, SQ_D8, SQ_E8, SQ_F8, SQ_G8, SQ_H8,
-  SQ_NONE,
+    SQ_NONE,
 
-  SQUARE_ZERO = 0,
+    SQUARE_ZERO = 0,
-  SQUARE_NB   = 64
+    SQUARE_NB   = 64
 };
+// clang-format on
 
 enum Direction : int {
-  NORTH =  8,
+    NORTH = 8,
-  EAST  =  1,
+    EAST  = 1,
-  SOUTH = -NORTH,
+    SOUTH = -NORTH,
-  WEST  = -EAST,
+    WEST  = -EAST,
 
-  NORTH_EAST = NORTH + EAST,
+    NORTH_EAST = NORTH + EAST,
-  SOUTH_EAST = SOUTH + EAST,
+    SOUTH_EAST = SOUTH + EAST,
-  SOUTH_WEST = SOUTH + WEST,
+    SOUTH_WEST = SOUTH + WEST,
-  NORTH_WEST = NORTH + WEST
+    NORTH_WEST = NORTH + WEST
 };
 
 enum File : int {
-  FILE_A, FILE_B, FILE_C, FILE_D, FILE_E, FILE_F, FILE_G, FILE_H, FILE_NB
+    FILE_A,
+    FILE_B,
+    FILE_C,
+    FILE_D,
+    FILE_E,
+    FILE_F,
+    FILE_G,
+    FILE_H,
+    FILE_NB
 };
 
 enum Rank : int {
-  RANK_1, RANK_2, RANK_3, RANK_4, RANK_5, RANK_6, RANK_7, RANK_8, RANK_NB
+    RANK_1,
+    RANK_2,
+    RANK_3,
+    RANK_4,
+    RANK_5,
+    RANK_6,
+    RANK_7,
+    RANK_8,
+    RANK_NB
 };
 
 // Keep track of what a move changes on the board (used by NNUE)
 struct DirtyPiece {
 
-  // Number of changed pieces
+    // Number of changed pieces
-  int dirty_num;
+    int dirty_num;
 
-  // Max 3 pieces can change in one move. A promotion with capture moves
+    // 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
+    // both the pawn and the captured piece to SQ_NONE and the piece promoted
-  // to from SQ_NONE to the capture square.
+    // to from SQ_NONE to the capture square.
-  Piece piece[3];
+    Piece piece[3];
 
-  // From and to squares, which may be SQ_NONE
+    // From and to squares, which may be SQ_NONE
-  Square from[3];
+    Square from[3];
-  Square to[3];
+    Square to[3];
 };
 
-/// Score enum stores a middlegame and an endgame value in a single integer (enum).
+    #define ENABLE_INCR_OPERATORS_ON(T) \
-/// The least significant 16 bits are used to store the middlegame value and the
+        inline T& operator++(T& d) { return d = T(int(d) + 1); } \
-/// upper 16 bits are used to store the endgame value. We have to take care to
+        inline T& operator--(T& d) { return d = T(int(d) - 1); }
-/// avoid left-shifting a signed int to avoid undefined behavior.
-enum Score : int { SCORE_ZERO };
 
-constexpr Score make_score(int mg, int eg) {
-  return Score((int)((unsigned int)eg << 16) + mg);
-}
-
-/// 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);
-}
-
-inline Value mg_value(Score s) {
-  union { uint16_t u; int16_t s; } mg = { uint16_t(unsigned(s)) };
-  return Value(mg.s);
-}
-
-#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; }
-
-#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); }
-
-#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_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_INCR_OPERATORS_ON
 
-#undef ENABLE_FULL_OPERATORS_ON
+constexpr Direction operator+(Direction d1, Direction d2) { return Direction(int(d1) + int(d2)); }
-#undef ENABLE_INCR_OPERATORS_ON
+constexpr Direction operator*(int i, Direction d) { return Direction(i * int(d)); }
-#undef ENABLE_BASE_OPERATORS_ON
 
-/// Additional operators to add a Direction to a Square
+// 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; }
-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
+// Toggle color
-/// a very high risk of overflow. So user should explicitly convert to integer.
+constexpr Color operator~(Color c) { return Color(c ^ BLACK); }
-Score operator*(Score, Score) = delete;
 
-/// Division of a Score must be handled separately for each term
+// Swap A1 <-> A8
-inline Score operator/(Score s, int i) {
+constexpr Square flip_rank(Square s) { return Square(s ^ SQ_A8); }
-  return make_score(mg_value(s) / i, eg_value(s) / i);
-}
 
-/// Multiplication of a Score by an integer. We check for overflow in debug mode.
+// Swap A1 <-> H1
-inline Score operator*(Score s, int i) {
+constexpr Square flip_file(Square s) { return Square(s ^ SQ_H1); }
 
-  Score result = Score(int(s) * i);
+// Swap color of piece B_KNIGHT <-> W_KNIGHT
-
+constexpr Piece operator~(Piece pc) { return Piece(pc ^ 8); }
-  assert(eg_value(result) == (i * eg_value(s)));
-  assert(mg_value(result) == (i * mg_value(s)));
-  assert((i == 0) || (result / i) == s);
-
-  return result;
-}
-
-/// Multiplication of a Score by a boolean
-inline Score operator*(Score s, bool b) {
-  return b ? s : SCORE_ZERO;
-}
-
-constexpr Color operator~(Color c) {
-  return Color(c ^ BLACK); // Toggle color
-}
-
-constexpr Square flip_rank(Square s) { // Swap A1 <-> A8
-  return Square(s ^ SQ_A8);
-}
-
-constexpr Square flip_file(Square s) { // Swap A1 <-> H1
-  return Square(s ^ SQ_H1);
-}
-
-constexpr Piece operator~(Piece pc) {
-  return Piece(pc ^ 8); // Swap color of piece B_KNIGHT <-> W_KNIGHT
-}
 
 constexpr CastlingRights operator&(Color c, CastlingRights cr) {
-  return CastlingRights((c == WHITE ? WHITE_CASTLING : BLACK_CASTLING) & cr);
+    return CastlingRights((c == WHITE ? WHITE_CASTLING : BLACK_CASTLING) & cr);
 }
 
-constexpr Value mate_in(int ply) {
+constexpr Value mate_in(int ply) { return VALUE_MATE - ply; }
-  return VALUE_MATE - ply;
-}
 
-constexpr Value mated_in(int ply) {
+constexpr Value mated_in(int ply) { return -VALUE_MATE + ply; }
-  return -VALUE_MATE + ply;
-}
 
-constexpr Square make_square(File f, Rank r) {
+constexpr Square make_square(File f, Rank r) { return Square((r << 3) + f); }
-  return Square((r << 3) + f);
-}
 
-constexpr Piece make_piece(Color c, PieceType pt) {
+constexpr Piece make_piece(Color c, PieceType pt) { return Piece((c << 3) + pt); }
-  return Piece((c << 3) + pt);
-}
 
-constexpr PieceType type_of(Piece pc) {
+constexpr PieceType type_of(Piece pc) { return PieceType(pc & 7); }
-  return PieceType(pc & 7);
-}
 
 inline Color color_of(Piece pc) {
-  assert(pc != NO_PIECE);
+    assert(pc != NO_PIECE);
-  return Color(pc >> 3);
+    return Color(pc >> 3);
 }
 
-constexpr bool is_ok(Move m) {
+constexpr bool is_ok(Square s) { return s >= SQ_A1 && s <= SQ_H8; }
-  return m != MOVE_NONE && m != MOVE_NULL;
-}
 
-constexpr bool is_ok(Square s) {
+constexpr File file_of(Square s) { return File(s & 7); }
-  return s >= SQ_A1 && s <= SQ_H8;
-}
 
-constexpr File file_of(Square s) {
+constexpr Rank rank_of(Square s) { return Rank(s >> 3); }
-  return File(s & 7);
-}
 
-constexpr Rank rank_of(Square s) {
+constexpr Square relative_square(Color c, Square s) { return Square(s ^ (c * 56)); }
-  return Rank(s >> 3);
-}
 
-constexpr Square relative_square(Color c, Square s) {
+constexpr Rank relative_rank(Color c, Rank r) { return Rank(r ^ (c * 7)); }
-  return Square(s ^ (c * 56));
-}
 
-constexpr Rank relative_rank(Color c, Rank r) {
+constexpr Rank relative_rank(Color c, Square s) { return relative_rank(c, rank_of(s)); }
-  return Rank(r ^ (c * 7));
-}
 
-constexpr Rank relative_rank(Color c, Square s) {
+constexpr Direction pawn_push(Color c) { return c == WHITE ? NORTH : SOUTH; }
-  return relative_rank(c, rank_of(s));
-}
 
-constexpr Direction pawn_push(Color c) {
-  return c == WHITE ? NORTH : SOUTH;
-}
 
-constexpr Square from_sq(Move m) {
+// Based on a congruential pseudo-random number generator
-  assert(is_ok(m));
-  return Square((m >> 6) & 0x3F);
-}
-
-constexpr Square to_sq(Move m) {
-  assert(is_ok(m));
-  return Square(m & 0x3F);
-}
-
-constexpr int from_to(Move m) {
-  return m & 0xFFF;
-}
-
-constexpr MoveType type_of(Move m) {
-  return MoveType(m & (3 << 14));
-}
-
-constexpr PieceType promotion_type(Move m) {
-  return PieceType(((m >> 12) & 3) + KNIGHT);
-}
-
-constexpr Move make_move(Square from, Square to) {
-  return Move((from << 6) + to);
-}
-
-template<MoveType T>
-constexpr Move make(Square from, Square to, PieceType pt = KNIGHT) {
-  return Move(T + ((pt - KNIGHT) << 12) + (from << 6) + to);
-}
-
-/// Based on a congruential pseudo random number generator
 constexpr Key make_key(uint64_t seed) {
-  return seed * 6364136223846793005ULL + 1442695040888963407ULL;
+    return seed * 6364136223846793005ULL + 1442695040888963407ULL;
 }
 
-} // namespace Stockfish
 
-#endif // #ifndef TYPES_H_INCLUDED
+enum MoveType {
+    NORMAL,
+    PROMOTION  = 1 << 14,
+    EN_PASSANT = 2 << 14,
+    CASTLING   = 3 << 14
+};
 
-#include "tune.h" // Global visibility to tuning setup
+// 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), 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 the destination square and origin square are always different,
+// but Move::none() and Move::null() have the same origin and destination square.
+
+class Move {
+   public:
+    Move() = default;
+    constexpr explicit Move(std::uint16_t d) :
+        data(d) {}
+
+    constexpr Move(Square from, Square to) :
+        data((from << 6) + to) {}
+
+    template<MoveType T>
+    static constexpr Move make(Square from, Square to, PieceType pt = KNIGHT) {
+        return Move(T + ((pt - KNIGHT) << 12) + (from << 6) + to);
+    }
+
+    constexpr Square from_sq() const {
+        assert(is_ok());
+        return Square((data >> 6) & 0x3F);
+    }
+
+    constexpr Square to_sq() const {
+        assert(is_ok());
+        return Square(data & 0x3F);
+    }
+
+    constexpr int from_to() const { return data & 0xFFF; }
+
+    constexpr MoveType type_of() const { return MoveType(data & (3 << 14)); }
+
+    constexpr PieceType promotion_type() const { return PieceType(((data >> 12) & 3) + KNIGHT); }
+
+    constexpr bool is_ok() const { return none().data != data && null().data != data; }
+
+    static constexpr Move null() { return Move(65); }
+    static constexpr Move none() { return Move(0); }
+
+    constexpr bool operator==(const Move& m) const { return data == m.data; }
+    constexpr bool operator!=(const Move& m) const { return data != m.data; }
+
+    constexpr explicit operator bool() const { return data != 0; }
+
+    constexpr std::uint16_t raw() const { return data; }
+
+    struct MoveHash {
+        std::size_t operator()(const Move& m) const { return make_key(m.data); }
+    };
+
+   protected:
+    std::uint16_t data;
+};
+
+}  // namespace Stockfish
+
+#endif  // #ifndef TYPES_H_INCLUDED
+
+#include "tune.h"  // Global visibility to tuning setup

src/uci.cpp

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,50 +16,470 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
-#include <cassert>
+#include "uci.h"
+
+#include <algorithm>
+#include <cctype>
 #include <cmath>
-#include <iostream>
+#include <cstdint>
+#include <iterator>
+#include <optional>
 #include <sstream>
-#include <string>
+#include <string_view>
+#include <utility>
+#include <vector>
 
 #include "benchmark.h"
-#include "evaluate.h"
+#include "engine.h"
+#include "memory.h"
 #include "movegen.h"
 #include "position.h"
+#include "score.h"
 #include "search.h"
-#include "thread.h"
+#include "types.h"
-#include "timeman.h"
+#include "ucioption.h"
-#include "tt.h"
-#include "uci.h"
-#include "syzygy/tbprobe.h"
-#include "nnue/evaluate_nnue.h"
-
-using namespace std;
 
 namespace Stockfish {
 
-namespace {
+constexpr auto BenchmarkCommand = "speedtest";
 
-  // FEN string for the initial position in standard chess
+constexpr auto StartFEN = "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1";
-  const char* StartFEN = "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1";
+template<typename... Ts>
+struct overload: Ts... {
+    using Ts::operator()...;
+};
 
+template<typename... Ts>
+overload(Ts...) -> overload<Ts...>;
 
-  // position() is called when the engine receives the "position" UCI command.
+void UCIEngine::print_info_string(std::string_view str) {
-  // It sets up the position that is described in the given FEN string ("fen") or
+    sync_cout_start();
-  // the initial position ("startpos") and then makes the moves given in the following
+    for (auto& line : split(str, "\n"))
-  // move list ("moves").
+    {
+        if (!is_whitespace(line))
+        {
+            std::cout << "info string " << line << '\n';
+        }
+    }
+    sync_cout_end();
+}
 
-  void position(Position& pos, istringstream& is, StateListPtr& states) {
+UCIEngine::UCIEngine(int argc, char** argv) :
+    engine(argv[0]),
+    cli(argc, argv) {
 
-    Move m;
+    engine.get_options().add_info_listener([](const std::optional<std::string>& str) {
-    string token, fen;
+        if (str.has_value())
+            print_info_string(*str);
+    });
+
+    init_search_update_listeners();
+}
+
+void UCIEngine::init_search_update_listeners() {
+    engine.set_on_iter([](const auto& i) { on_iter(i); });
+    engine.set_on_update_no_moves([](const auto& i) { on_update_no_moves(i); });
+    engine.set_on_update_full(
+      [this](const auto& i) { on_update_full(i, engine.get_options()["UCI_ShowWDL"]); });
+    engine.set_on_bestmove([](const auto& bm, const auto& p) { on_bestmove(bm, p); });
+    engine.set_on_verify_networks([](const auto& s) { print_info_string(s); });
+}
+
+void UCIEngine::loop() {
+    std::string token, cmd;
+
+    for (int i = 1; i < cli.argc; ++i)
+        cmd += std::string(cli.argv[i]) + " ";
+
+    do
+    {
+        if (cli.argc == 1
+            && !getline(std::cin, cmd))  // Wait for an input or an end-of-file (EOF) indication
+            cmd = "quit";
+
+        std::istringstream is(cmd);
+
+        token.clear();  // Avoid a stale if getline() returns nothing or a blank line
+        is >> std::skipws >> token;
+
+        if (token == "quit" || token == "stop")
+            engine.stop();
+
+        // 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")
+            engine.set_ponderhit(false);
+
+        else if (token == "uci")
+        {
+            sync_cout << "id name " << engine_info(true) << "\n"
+                      << engine.get_options() << sync_endl;
+
+            sync_cout << "uciok" << sync_endl;
+        }
+
+        else if (token == "setoption")
+            setoption(is);
+        else if (token == "go")
+        {
+            // send info strings after the go command is sent for old GUIs and python-chess
+            print_info_string(engine.numa_config_information_as_string());
+            print_info_string(engine.thread_allocation_information_as_string());
+            go(is);
+        }
+        else if (token == "position")
+            position(is);
+        else if (token == "ucinewgame")
+            engine.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")
+            engine.flip();
+        else if (token == "bench")
+            bench(is);
+        else if (token == BenchmarkCommand)
+            benchmark(is);
+        else if (token == "d")
+            sync_cout << engine.visualize() << sync_endl;
+        else if (token == "eval")
+            engine.trace_eval();
+        else if (token == "compiler")
+            sync_cout << compiler_info() << sync_endl;
+        else if (token == "export_net")
+        {
+            std::pair<std::optional<std::string>, std::string> files[2];
+
+            if (is >> std::skipws >> files[0].second)
+                files[0].first = files[0].second;
+
+            if (is >> std::skipws >> files[1].second)
+                files[1].first = files[1].second;
+
+            engine.save_network(files);
+        }
+        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" && cli.argc == 1);  // The command-line arguments are one-shot
+}
+
+Search::LimitsType UCIEngine::parse_limits(std::istream& is) {
+    Search::LimitsType limits;
+    std::string        token;
+
+    limits.startTime = now();  // The search starts as early as possible
+
+    while (is >> token)
+        if (token == "searchmoves")  // Needs to be the last command on the line
+            while (is >> token)
+                limits.searchmoves.push_back(to_lower(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")
+            limits.ponderMode = true;
+
+    return limits;
+}
+
+void UCIEngine::go(std::istringstream& is) {
+
+    Search::LimitsType limits = parse_limits(is);
+
+    if (limits.perft)
+        perft(limits);
+    else
+        engine.go(limits);
+}
+
+void UCIEngine::bench(std::istream& args) {
+    std::string token;
+    uint64_t    num, nodes = 0, cnt = 1;
+    uint64_t    nodesSearched = 0;
+    const auto& options       = engine.get_options();
+
+    engine.set_on_update_full([&](const auto& i) {
+        nodesSearched = i.nodes;
+        on_update_full(i, options["UCI_ShowWDL"]);
+    });
+
+    std::vector<std::string> list = Benchmark::setup_bench(engine.fen(), args);
+
+    num = count_if(list.begin(), list.end(),
+                   [](const std::string& s) { return s.find("go ") == 0 || s.find("eval") == 0; });
+
+    TimePoint elapsed = now();
+
+    for (const auto& cmd : list)
+    {
+        std::istringstream is(cmd);
+        is >> std::skipws >> token;
+
+        if (token == "go" || token == "eval")
+        {
+            std::cerr << "\nPosition: " << cnt++ << '/' << num << " (" << engine.fen() << ")"
+                      << std::endl;
+            if (token == "go")
+            {
+                Search::LimitsType limits = parse_limits(is);
+
+                if (limits.perft)
+                    nodesSearched = perft(limits);
+                else
+                {
+                    engine.go(limits);
+                    engine.wait_for_search_finished();
+                }
+
+                nodes += nodesSearched;
+                nodesSearched = 0;
+            }
+            else
+                engine.trace_eval();
+        }
+        else if (token == "setoption")
+            setoption(is);
+        else if (token == "position")
+            position(is);
+        else if (token == "ucinewgame")
+        {
+            engine.search_clear();  // search_clear may take a while
+            elapsed = now();
+        }
+    }
+
+    elapsed = now() - elapsed + 1;  // Ensure positivity to avoid a 'divide by zero'
+
+    dbg_print();
+
+    std::cerr << "\n==========================="    //
+              << "\nTotal time (ms) : " << elapsed  //
+              << "\nNodes searched  : " << nodes    //
+              << "\nNodes/second    : " << 1000 * nodes / elapsed << std::endl;
+
+    // reset callback, to not capture a dangling reference to nodesSearched
+    engine.set_on_update_full([&](const auto& i) { on_update_full(i, options["UCI_ShowWDL"]); });
+}
+
+void UCIEngine::benchmark(std::istream& args) {
+    // Probably not very important for a test this long, but include for completeness and sanity.
+    static constexpr int NUM_WARMUP_POSITIONS = 3;
+
+    std::string token;
+    uint64_t    nodes = 0, cnt = 1;
+    uint64_t    nodesSearched = 0;
+
+    engine.set_on_update_full([&](const Engine::InfoFull& i) { nodesSearched = i.nodes; });
+
+    engine.set_on_iter([](const auto&) {});
+    engine.set_on_update_no_moves([](const auto&) {});
+    engine.set_on_bestmove([](const auto&, const auto&) {});
+    engine.set_on_verify_networks([](const auto&) {});
+
+    Benchmark::BenchmarkSetup setup = Benchmark::setup_benchmark(args);
+
+    const int numGoCommands = count_if(setup.commands.begin(), setup.commands.end(),
+                                       [](const std::string& s) { return s.find("go ") == 0; });
+
+    TimePoint totalTime = 0;
+
+    // Set options once at the start.
+    auto ss = std::istringstream("name Threads value " + std::to_string(setup.threads));
+    setoption(ss);
+    ss = std::istringstream("name Hash value " + std::to_string(setup.ttSize));
+    setoption(ss);
+    ss = std::istringstream("name UCI_Chess960 value false");
+    setoption(ss);
+
+    // Warmup
+    for (const auto& cmd : setup.commands)
+    {
+        std::istringstream is(cmd);
+        is >> std::skipws >> token;
+
+        if (token == "go")
+        {
+            // One new line is produced by the search, so omit it here
+            std::cerr << "\rWarmup position " << cnt++ << '/' << NUM_WARMUP_POSITIONS;
+
+            Search::LimitsType limits = parse_limits(is);
+
+            TimePoint elapsed = now();
+
+            // Run with silenced network verification
+            engine.go(limits);
+            engine.wait_for_search_finished();
+
+            totalTime += now() - elapsed;
+
+            nodes += nodesSearched;
+            nodesSearched = 0;
+        }
+        else if (token == "position")
+            position(is);
+        else if (token == "ucinewgame")
+        {
+            engine.search_clear();  // search_clear may take a while
+        }
+
+        if (cnt > NUM_WARMUP_POSITIONS)
+            break;
+    }
+
+    std::cerr << "\n";
+
+    cnt   = 1;
+    nodes = 0;
+
+    int           numHashfullReadings = 0;
+    constexpr int hashfullAges[]      = {0, 999};  // Only normal hashfull and touched hash.
+    int           totalHashfull[std::size(hashfullAges)] = {0};
+    int           maxHashfull[std::size(hashfullAges)]   = {0};
+
+    auto updateHashfullReadings = [&]() {
+        numHashfullReadings += 1;
+
+        for (int i = 0; i < static_cast<int>(std::size(hashfullAges)); ++i)
+        {
+            const int hashfull = engine.get_hashfull(hashfullAges[i]);
+            maxHashfull[i]     = std::max(maxHashfull[i], hashfull);
+            totalHashfull[i] += hashfull;
+        }
+    };
+
+    engine.search_clear();  // search_clear may take a while
+
+    for (const auto& cmd : setup.commands)
+    {
+        std::istringstream is(cmd);
+        is >> std::skipws >> token;
+
+        if (token == "go")
+        {
+            // One new line is produced by the search, so omit it here
+            std::cerr << "\rPosition " << cnt++ << '/' << numGoCommands;
+
+            Search::LimitsType limits = parse_limits(is);
+
+            TimePoint elapsed = now();
+
+            // Run with silenced network verification
+            engine.go(limits);
+            engine.wait_for_search_finished();
+
+            totalTime += now() - elapsed;
+
+            updateHashfullReadings();
+
+            nodes += nodesSearched;
+            nodesSearched = 0;
+        }
+        else if (token == "position")
+            position(is);
+        else if (token == "ucinewgame")
+        {
+            engine.search_clear();  // search_clear may take a while
+        }
+    }
+
+    totalTime = std::max<TimePoint>(totalTime, 1);  // Ensure positivity to avoid a 'divide by zero'
+
+    dbg_print();
+
+    std::cerr << "\n";
+
+    static_assert(
+      std::size(hashfullAges) == 2 && hashfullAges[0] == 0 && hashfullAges[1] == 999,
+      "Hardcoded for display. Would complicate the code needlessly in the current state.");
+
+    std::string threadBinding = engine.thread_binding_information_as_string();
+    if (threadBinding.empty())
+        threadBinding = "none";
+
+    // clang-format off
+
+    std::cerr << "==========================="
+              << "\nVersion                    : "
+              << engine_version_info()
+              // "\nCompiled by                : "
+              << compiler_info()
+              << "Large pages                : " << (has_large_pages() ? "yes" : "no")
+              << "\nUser invocation            : " << BenchmarkCommand << " "
+              << setup.originalInvocation << "\nFilled invocation          : " << BenchmarkCommand
+              << " " << setup.filledInvocation
+              << "\nAvailable processors       : " << engine.get_numa_config_as_string()
+              << "\nThread count               : " << setup.threads
+              << "\nThread binding             : " << threadBinding
+              << "\nTT size [MiB]              : " << setup.ttSize
+              << "\nHash max, avg [per mille]  : "
+              << "\n    single search          : " << maxHashfull[0] << ", "
+              << totalHashfull[0] / numHashfullReadings
+              << "\n    single game            : " << maxHashfull[1] << ", "
+              << totalHashfull[1] / numHashfullReadings
+              << "\nTotal nodes searched       : " << nodes
+              << "\nTotal search time [s]      : " << totalTime / 1000.0
+              << "\nNodes/second               : " << 1000 * nodes / totalTime << std::endl;
+
+    // clang-format on
+
+    init_search_update_listeners();
+}
+
+void UCIEngine::setoption(std::istringstream& is) {
+    engine.wait_for_search_finished();
+    engine.get_options().setoption(is);
+}
+
+std::uint64_t UCIEngine::perft(const Search::LimitsType& limits) {
+    auto nodes = engine.perft(engine.fen(), limits.perft, engine.get_options()["UCI_Chess960"]);
+    sync_cout << "\nNodes searched: " << nodes << "\n" << sync_endl;
+    return nodes;
+}
+
+void UCIEngine::position(std::istringstream& is) {
+    std::string token, fen;
 
     is >> token;
 
     if (token == "startpos")
     {
         fen = StartFEN;
-        is >> token; // Consume the "moves" token, if any
+        is >> token;  // Consume the "moves" token, if any
     }
     else if (token == "fen")
         while (is >> token && token != "moves")
@@ -67,332 +487,181 @@ namespace {
     else
         return;
 
-    states = StateListPtr(new std::deque<StateInfo>(1)); // Drop the old state and create a new one
+    std::vector<std::string> moves;
-    pos.set(fen, Options["UCI_Chess960"], &states->back(), Threads.main());
-
-    // Parse the move list, if any
-    while (is >> token && (m = UCI::to_move(pos, token)) != MOVE_NONE)
-    {
-        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.
-
-  void trace_eval(Position& pos) {
-
-    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 the "name" token
-
-    // Read the option name (can contain spaces)
-    while (is >> token && token != "value")
-        name += (name.empty() ? "" : " ") + token;
-
-    // Read the option value (can contain spaces)
-    while (is >> token)
-        value += (value.empty() ? "" : " ") + token;
-
-    if (Options.count(name))
-        Options[name] = value;
-    else
-        sync_cout << "No such option: " << name << sync_endl;
-  }
-
-
-  // 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, StateListPtr& states) {
-
-    Search::LimitsType limits;
-    string token;
-    bool ponderMode = false;
-
-    limits.startTime = now(); // The search starts as early as possible
 
     while (is >> token)
-        if (token == "searchmoves") // Needs to be the last command on the line
-            while (is >> 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);
+        moves.push_back(token);
-        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
     }
 
-    elapsed = now() - elapsed + 1; // Ensure positivity to avoid a 'divide by zero'
+    engine.set_position(fen, moves);
+}
 
-    dbg_print();
+namespace {
 
-    cerr << "\n==========================="
+struct WinRateParams {
-         << "\nTotal time (ms) : " << elapsed
+    double a;
-         << "\nNodes searched  : " << nodes
+    double b;
-         << "\nNodes/second    : " << 1000 * nodes / elapsed << endl;
+};
-  }
 
-  // The win rate model returns the probability of winning (in per mille units) given an
+WinRateParams win_rate_params(const Position& pos) {
-  // 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
+    int material = pos.count<PAWN>() + 3 * pos.count<KNIGHT>() + 3 * pos.count<BISHOP>()
-     double m = std::min(240, ply) / 64.0;
+                 + 5 * pos.count<ROOK>() + 9 * pos.count<QUEEN>();
 
-     // The coefficients of a third-order polynomial fit is based on the fishtest data
+    // The fitted model only uses data for material counts in [17, 78], and is anchored at count 58.
-     // for two parameters that need to transform eval to the argument of a logistic
+    double m = std::clamp(material, 17, 78) / 58.0;
-     // 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
+    // Return a = p_a(material) and b = p_b(material), see github.com/official-stockfish/WDL_model
-     static_assert(UCI::NormalizeToPawnValue == int(as[0] + as[1] + as[2] + as[3]));
+    constexpr double as[] = {-37.45051876, 121.19101539, -132.78783573, 420.70576692};
+    constexpr double bs[] = {90.26261072, -137.26549898, 71.10130540, 51.35259597};
 
-     double a = (((as[0] * m + as[1]) * m + as[2]) * m) + 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];
+    double b = (((bs[0] * m + bs[1]) * m + bs[2]) * m) + bs[3];
 
-     // Transform the eval to centipawns with limited range
+    return {a, b};
-     double x = std::clamp(double(v), -4000.0, 4000.0);
+}
 
-     // Return the win rate in per mille units rounded to the nearest value
+// The win rate model is 1 / (1 + exp((a - eval) / b)), where a = p_a(material) and b = p_b(material).
-     return int(0.5 + 1000 / (1 + std::exp((a - x) / b)));
+// It fits the LTC fishtest statistics rather accurately.
-  }
+int win_rate_model(Value v, const Position& pos) {
 
-} // namespace
+    auto [a, b] = win_rate_params(pos);
 
+    // Return the win rate in per mille units, rounded to the nearest integer.
+    return int(0.5 + 1000 / (1 + std::exp((a - double(v)) / b)));
+}
+}
 
-/// UCI::loop() waits for a command from the stdin, parses it and then calls the appropriate
+std::string UCIEngine::format_score(const Score& s) {
-/// function. It also intercepts an end-of-file (EOF) indication from the stdin to ensure a
+    constexpr int TB_CP = 20000;
-/// graceful exit if the GUI dies unexpectedly. When called with some command-line arguments,
+    const auto    format =
-/// like running 'bench', the function returns immediately after the command is executed.
+      overload{[](Score::Mate mate) -> std::string {
-/// In addition to the UCI ones, some additional debug commands are also supported.
+                   auto m = (mate.plies > 0 ? (mate.plies + 1) : mate.plies) / 2;
+                   return std::string("mate ") + std::to_string(m);
+               },
+               [](Score::Tablebase tb) -> std::string {
+                   return std::string("cp ")
+                        + std::to_string((tb.win ? TB_CP - tb.plies : -TB_CP - tb.plies));
+               },
+               [](Score::InternalUnits units) -> std::string {
+                   return std::string("cp ") + std::to_string(units.value);
+               }};
 
-void UCI::loop(int argc, char* argv[]) {
+    return s.visit(format);
+}
 
-  Position pos;
+// Turns a Value to an integer centipawn number,
-  string token, cmd;
+// without treatment of mate and similar special scores.
-  StateListPtr states(new std::deque<StateInfo>(1));
+int UCIEngine::to_cp(Value v, const Position& pos) {
 
-  pos.set(StartFEN, false, &states->back(), Threads.main());
+    // In general, the score can be defined via the WDL as
+    // (log(1/L - 1) - log(1/W - 1)) / (log(1/L - 1) + log(1/W - 1)).
+    // Based on our win_rate_model, this simply yields v / a.
 
-  for (int i = 1; i < argc; ++i)
+    auto [a, b] = win_rate_params(pos);
-      cmd += std::string(argv[i]) + " ";
 
-  do {
+    return std::round(100 * int(v) / a);
-      if (argc == 1 && !getline(cin, cmd)) // Wait for an input or an end-of-file (EOF) indication
+}
-          cmd = "quit";
 
-      istringstream is(cmd);
+std::string UCIEngine::wdl(Value v, const Position& pos) {
+    std::stringstream ss;
 
-      token.clear(); // Avoid a stale if getline() returns nothing or a blank line
+    int wdl_w = win_rate_model(v, pos);
-      is >> skipws >> token;
+    int wdl_l = win_rate_model(-v, pos);
+    int wdl_d = 1000 - wdl_w - wdl_l;
+    ss << wdl_w << " " << wdl_d << " " << wdl_l;
 
-      if (    token == "quit"
+    return ss.str();
-          ||  token == "stop")
+}
-          Threads.stop = true;
 
-      // The GUI sends 'ponderhit' to tell that the user has played the expected move.
+std::string UCIEngine::square(Square s) {
-      // So, 'ponderhit' is sent if pondering was done on the same move that the user
+    return std::string{char('a' + file_of(s)), char('1' + rank_of(s))};
-      // 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")
+std::string UCIEngine::move(Move m, bool chess960) {
-          sync_cout << "id name " << engine_info(true)
+    if (m == Move::none())
-                    << "\n"       << Options
+        return "(none)";
-                    << "\nuciok"  << sync_endl;
 
-      else if (token == "setoption")  setoption(is);
+    if (m == Move::null())
-      else if (token == "go")         go(pos, is, states);
+        return "0000";
-      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.
+    Square from = m.from_sq();
-      // These commands must not be used during a search!
+    Square to   = m.to_sq();
-      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
+    if (m.type_of() == CASTLING && !chess960)
+        to = make_square(to > from ? FILE_G : FILE_C, rank_of(from));
+
+    std::string move = square(from) + square(to);
+
+    if (m.type_of() == PROMOTION)
+        move += " pnbrqk"[m.promotion_type()];
+
+    return move;
 }
 
 
-/// UCI::value() converts a Value to a string by adhering to the UCI protocol specification:
+std::string UCIEngine::to_lower(std::string str) {
-///
+    std::transform(str.begin(), str.end(), str.begin(), [](auto c) { return std::tolower(c); });
-/// 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) {
+    return str;
-
-  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();
 }
 
+Move UCIEngine::to_move(const Position& pos, std::string str) {
+    str = to_lower(str);
 
-/// UCI::wdl() reports the win-draw-loss (WDL) statistics given an evaluation
+    for (const auto& m : MoveList<LEGAL>(pos))
-/// and a game ply based on the data gathered for fishtest LTC games.
+        if (str == move(m, pos.is_chess960()))
+            return m;
 
-string UCI::wdl(Value v, int ply) {
+    return Move::none();
-
-  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();
 }
 
-
+void UCIEngine::on_update_no_moves(const Engine::InfoShort& info) {
-/// UCI::square() converts a Square to a string in algebraic notation (g1, a7, etc.)
+    sync_cout << "info depth " << info.depth << " score " << format_score(info.score) << sync_endl;
-
-std::string UCI::square(Square s) {
-  return std::string{ char('a' + file_of(s)), char('1' + rank_of(s)) };
 }
 
+void UCIEngine::on_update_full(const Engine::InfoFull& info, bool showWDL) {
+    std::stringstream ss;
 
-/// UCI::move() converts a Move to a string in coordinate notation (g1f3, a7a8q).
+    ss << "info";
-/// The only special case is castling where the e1g1 notation is printed in
+    ss << " depth " << info.depth                 //
-/// standard chess mode and in e1h1 notation it is printed in Chess960 mode.
+       << " seldepth " << info.selDepth           //
-/// Internally, all castling moves are always encoded as 'king captures rook'.
+       << " multipv " << info.multiPV             //
+       << " score " << format_score(info.score);  //
 
-string UCI::move(Move m, bool chess960) {
+    if (showWDL)
+        ss << " wdl " << info.wdl;
 
-  if (m == MOVE_NONE)
+    if (!info.bound.empty())
-      return "(none)";
+        ss << " " << info.bound;
 
-  if (m == MOVE_NULL)
+    ss << " nodes " << info.nodes        //
-      return "0000";
+       << " nps " << info.nps            //
+       << " hashfull " << info.hashfull  //
+       << " tbhits " << info.tbHits      //
+       << " time " << info.timeMs        //
+       << " pv " << info.pv;             //
 
-  Square from = from_sq(m);
+    sync_cout << ss.str() << sync_endl;
-  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;
 }
 
+void UCIEngine::on_iter(const Engine::InfoIter& info) {
+    std::stringstream ss;
 
-/// UCI::to_move() converts a string representing a move in coordinate notation
+    ss << "info";
-/// (g1f3, a7a8q) to the corresponding legal Move, if any.
+    ss << " depth " << info.depth                     //
+       << " currmove " << info.currmove               //
+       << " currmovenumber " << info.currmovenumber;  //
 
-Move UCI::to_move(const Position& pos, string& str) {
+    sync_cout << ss.str() << sync_endl;
-
-  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
+void UCIEngine::on_bestmove(std::string_view bestmove, std::string_view ponder) {
+    sync_cout << "bestmove " << bestmove;
+    if (!ponder.empty())
+        std::cout << " ponder " << ponder;
+    std::cout << sync_endl;
+}
+
+}  // namespace Stockfish

src/uci.h

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 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
@@ -19,74 +19,62 @@
 #ifndef UCI_H_INCLUDED
 #define UCI_H_INCLUDED
 
-#include <map>
+#include <cstdint>
+#include <iostream>
 #include <string>
+#include <string_view>
 
-#include "types.h"
+#include "engine.h"
+#include "misc.h"
+#include "search.h"
 
 namespace Stockfish {
 
 class Position;
+class Move;
+class Score;
+enum Square : int;
+using Value = int;
 
-namespace UCI {
+class UCIEngine {
+   public:
+    UCIEngine(int argc, char** argv);
 
-// Normalizes the internal value as reported by evaluate or search
+    void loop();
-// 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;
+    static int         to_cp(Value v, const Position& pos);
+    static std::string format_score(const Score& s);
+    static std::string square(Square s);
+    static std::string move(Move m, bool chess960);
+    static std::string wdl(Value v, const Position& pos);
+    static std::string to_lower(std::string str);
+    static Move        to_move(const Position& pos, std::string str);
 
-/// Define a custom comparator, because the UCI options should be case-insensitive
+    static Search::LimitsType parse_limits(std::istream& is);
-struct CaseInsensitiveLess {
+
-  bool operator() (const std::string&, const std::string&) const;
+    auto& engine_options() { return engine.get_options(); }
+
+   private:
+    Engine      engine;
+    CommandLine cli;
+
+    static void print_info_string(std::string_view str);
+
+    void          go(std::istringstream& is);
+    void          bench(std::istream& args);
+    void          benchmark(std::istream& args);
+    void          position(std::istringstream& is);
+    void          setoption(std::istringstream& is);
+    std::uint64_t perft(const Search::LimitsType&);
+
+    static void on_update_no_moves(const Engine::InfoShort& info);
+    static void on_update_full(const Engine::InfoFull& info, bool showWDL);
+    static void on_iter(const Engine::InfoIter& info);
+    static void on_bestmove(std::string_view bestmove, std::string_view ponder);
+
+    void init_search_update_listeners();
 };
 
-/// The options container is defined as a std::map
+}  // namespace Stockfish
-using OptionsMap = std::map<std::string, Option, CaseInsensitiveLess>;
 
-/// The Option class implements each option as specified by the UCI protocol
+#endif  // #ifndef UCI_H_INCLUDED
-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

@@ -1,6 +1,6 @@
 /*
   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2023 The Stockfish developers (see AUTHORS file)
+  Copyright (C) 2004-2025 The Stockfish developers (see AUTHORS file)
 
   Stockfish is free software: you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
@@ -16,179 +16,198 @@
   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
 
+#include "ucioption.h"
+
 #include <algorithm>
 #include <cassert>
-#include <ostream>
+#include <cctype>
+#include <cstdlib>
+#include <iostream>
 #include <sstream>
+#include <utility>
 
-#include "evaluate.h"
 #include "misc.h"
-#include "search.h"
-#include "thread.h"
-#include "tt.h"
-#include "uci.h"
-#include "syzygy/tbprobe.h"
-
-using std::string;
 
 namespace Stockfish {
 
-UCI::OptionsMap Options; // Global object
+bool CaseInsensitiveLess::operator()(const std::string& s1, const std::string& s2) const {
 
-namespace UCI {
+    return std::lexicographical_compare(
+      s1.begin(), s1.end(), s2.begin(), s2.end(),
+      [](char c1, char c2) { return std::tolower(c1) < std::tolower(c2); });
+}
 
-/// 'On change' actions, triggered by an option's value change
+void OptionsMap::add_info_listener(InfoListener&& message_func) { info = std::move(message_func); }
-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
+void OptionsMap::setoption(std::istringstream& is) {
-bool CaseInsensitiveLess::operator() (const string& s1, const string& s2) const {
+    std::string token, name, value;
 
-  return std::lexicographical_compare(s1.begin(), s1.end(), s2.begin(), s2.end(),
+    is >> token;  // Consume the "name" token
-         [](char c1, char c2) { return tolower(c1) < tolower(c2); });
+
+    // Read the option name (can contain spaces)
+    while (is >> token && token != "value")
+        name += (name.empty() ? "" : " ") + token;
+
+    // Read the option value (can contain spaces)
+    while (is >> token)
+        value += (value.empty() ? "" : " ") + token;
+
+    if (options_map.count(name))
+        options_map[name] = value;
+    else
+        sync_cout << "No such option: " << name << sync_endl;
+}
+
+const Option& OptionsMap::operator[](const std::string& name) const {
+    auto it = options_map.find(name);
+    assert(it != options_map.end());
+    return it->second;
+}
+
+// Inits options and assigns idx in the correct printing order
+void OptionsMap::add(const std::string& name, const Option& option) {
+    if (!options_map.count(name))
+    {
+        static size_t insert_order = 0;
+
+        options_map[name] = option;
+
+        options_map[name].parent = this;
+        options_map[name].idx    = insert_order++;
+    }
+    else
+    {
+        std::cerr << "Option \"" << name << "\" was already added!" << std::endl;
+        std::exit(EXIT_FAILURE);
+    }
 }
 
 
-/// UCI::init() initializes the UCI options to their hard-coded default values
+std::size_t OptionsMap::count(const std::string& name) const { return options_map.count(name); }
 
-void init(OptionsMap& o) {
+Option::Option(const OptionsMap* map) :
+    parent(map) {}
 
-  constexpr int MaxHashMB = Is64Bit ? 33554432 : 2048;
+Option::Option(const char* v, OnChange f) :
-
+    type("string"),
-  o["Debug Log File"]        << Option("", on_logger);
+    min(0),
-  o["Threads"]               << Option(1, 1, 1024, on_threads);
+    max(0),
-  o["Hash"]                  << Option(16, 1, MaxHashMB, on_hash_size);
+    on_change(std::move(f)) {
-  o["Clear Hash"]            << Option(on_clear_hash);
+    defaultValue = currentValue = v;
-  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);
 }
 
-
+Option::Option(bool v, OnChange f) :
-/// operator<<() is used to print all the options default values in chronological
+    type("check"),
-/// insertion order (the idx field) and in the format defined by the UCI protocol.
+    min(0),
-
+    max(0),
-std::ostream& operator<<(std::ostream& os, const OptionsMap& om) {
+    on_change(std::move(f)) {
-
+    defaultValue = currentValue = (v ? "true" : "false");
-  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;
-
-              if (o.type == "string" || o.type == "check" || o.type == "combo")
-                  os << " default " << o.defaultValue;
-
-              if (o.type == "spin")
-                  os << " default " << int(stof(o.defaultValue))
-                     << " min "     << o.min
-                     << " max "     << o.max;
-
-              break;
-          }
-
-  return os;
 }
 
+Option::Option(OnChange f) :
+    type("button"),
+    min(0),
+    max(0),
+    on_change(std::move(f)) {}
 
-/// Option class constructors and conversion operators
+Option::Option(double v, int minv, int maxv, OnChange f) :
+    type("spin"),
+    min(minv),
+    max(maxv),
+    on_change(std::move(f)) {
+    defaultValue = currentValue = std::to_string(v);
+}
 
-Option::Option(const char* v, OnChange f) : type("string"), min(0), max(0), on_change(f)
+Option::Option(const char* v, const char* cur, OnChange f) :
-{ defaultValue = currentValue = v; }
+    type("combo"),
-
+    min(0),
-Option::Option(bool v, OnChange f) : type("check"), min(0), max(0), on_change(f)
+    max(0),
-{ defaultValue = currentValue = (v ? "true" : "false"); }
+    on_change(std::move(f)) {
-
+    defaultValue = v;
-Option::Option(OnChange f) : type("button"), min(0), max(0), on_change(f)
+    currentValue = cur;
-{}
+}
-
-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");
+    assert(type == "check" || type == "spin");
-  return (type == "spin" ? std::stoi(currentValue) : currentValue == "true");
+    return (type == "spin" ? std::stoi(currentValue) : currentValue == "true");
 }
 
 Option::operator std::string() const {
-  assert(type == "string");
+    assert(type == "string");
-  return currentValue;
+    return currentValue;
 }
 
 bool Option::operator==(const char* s) const {
-  assert(type == "combo");
+    assert(type == "combo");
-  return   !CaseInsensitiveLess()(currentValue, s)
+    return !CaseInsensitiveLess()(currentValue, s) && !CaseInsensitiveLess()(s, currentValue);
-        && !CaseInsensitiveLess()(s, currentValue);
 }
 
+bool Option::operator!=(const char* s) const { return !(*this == s); }
 
-/// operator<<() inits options and assigns idx in the correct printing order
 
-void Option::operator<<(const Option& o) {
+// 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.
+Option& Option::operator=(const std::string& v) {
 
-  static size_t insert_order = 0;
+    assert(!type.empty());
 
-  *this = o;
+    if ((type != "button" && type != "string" && v.empty())
-  idx = insert_order++;
+        || (type == "check" && v != "true" && v != "false")
+        || (type == "spin" && (std::stof(v) < min || std::stof(v) > max)))
+        return *this;
+
+    if (type == "combo")
+    {
+        OptionsMap         comboMap;  // To have case insensitive compare
+        std::string        token;
+        std::istringstream ss(defaultValue);
+        while (ss >> token)
+            comboMap.add(token, Option());
+        if (!comboMap.count(v) || v == "var")
+            return *this;
+    }
+
+    if (type == "string")
+        currentValue = v == "<empty>" ? "" : v;
+    else if (type != "button")
+        currentValue = v;
+
+    if (on_change)
+    {
+        const auto ret = on_change(*this);
+
+        if (ret && parent != nullptr && parent->info != nullptr)
+            parent->info(ret);
+    }
+
+    return *this;
 }
 
+std::ostream& operator<<(std::ostream& os, const OptionsMap& om) {
+    for (size_t idx = 0; idx < om.options_map.size(); ++idx)
+        for (const auto& it : om.options_map)
+            if (it.second.idx == idx)
+            {
+                const Option& o = it.second;
+                os << "\noption name " << it.first << " type " << o.type;
 
-/// operator=() updates currentValue and triggers on_change() action. It's up to
+                if (o.type == "check" || o.type == "combo")
-/// the GUI to check for option's limits, but we could receive the new value
+                    os << " default " << o.defaultValue;
-/// from the user by console window, so let's check the bounds anyway.
 
-Option& Option::operator=(const string& v) {
+                else if (o.type == "string")
+                {
+                    std::string defaultValue = o.defaultValue.empty() ? "<empty>" : o.defaultValue;
+                    os << " default " << defaultValue;
+                }
 
-  assert(!type.empty());
+                else if (o.type == "spin")
+                    os << " default " << int(stof(o.defaultValue)) << " min " << o.min << " max "
+                       << o.max;
 
-  if (   (type != "button" && type != "string" && v.empty())
+                break;
-      || (type == "check" && v != "true" && v != "false")
+            }
-      || (type == "spin" && (stof(v) < min || stof(v) > max)))
-      return *this;
 
-  if (type == "combo")
+    return os;
-  {
+}
-      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);
-
-  return *this;
 }
-
-} // namespace UCI
-
-} // namespace Stockfish

src/ucioption.h

@@ -0,0 +1,106 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2025 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 UCIOPTION_H_INCLUDED
+#define UCIOPTION_H_INCLUDED
+
+#include <cstddef>
+#include <functional>
+#include <iosfwd>
+#include <map>
+#include <optional>
+#include <string>
+
+namespace Stockfish {
+// Define a custom comparator, because the UCI options should be case-insensitive
+struct CaseInsensitiveLess {
+    bool operator()(const std::string&, const std::string&) const;
+};
+
+class OptionsMap;
+
+// The Option class implements each option as specified by the UCI protocol
+class Option {
+   public:
+    using OnChange = std::function<std::optional<std::string>(const Option&)>;
+
+    Option(const OptionsMap*);
+    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&);
+    operator int() const;
+    operator std::string() const;
+    bool operator==(const char*) const;
+    bool operator!=(const char*) const;
+
+    friend std::ostream& operator<<(std::ostream&, const OptionsMap&);
+
+    int operator<<(const Option&) = delete;
+
+   private:
+    friend class OptionsMap;
+    friend class Engine;
+    friend class Tune;
+
+
+    std::string       defaultValue, currentValue, type;
+    int               min, max;
+    size_t            idx;
+    OnChange          on_change;
+    const OptionsMap* parent = nullptr;
+};
+
+class OptionsMap {
+   public:
+    using InfoListener = std::function<void(std::optional<std::string>)>;
+
+    OptionsMap()                             = default;
+    OptionsMap(const OptionsMap&)            = delete;
+    OptionsMap(OptionsMap&&)                 = delete;
+    OptionsMap& operator=(const OptionsMap&) = delete;
+    OptionsMap& operator=(OptionsMap&&)      = delete;
+
+    void add_info_listener(InfoListener&&);
+
+    void setoption(std::istringstream&);
+
+    const Option& operator[](const std::string&) const;
+
+    void add(const std::string&, const Option& option);
+
+    std::size_t count(const std::string&) const;
+
+   private:
+    friend class Engine;
+    friend class Option;
+
+    friend std::ostream& operator<<(std::ostream&, const OptionsMap&);
+
+    // The options container is defined as a std::map
+    using OptionsStore = std::map<std::string, Option, CaseInsensitiveLess>;
+
+    OptionsStore options_map;
+    InfoListener info;
+};
+
+}
+#endif  // #ifndef UCIOPTION_H_INCLUDED