Stockfish 17

Official release version of Stockfish 17

Bench: 1484730

---

Stockfish 17

Today we have the pleasure to announce a new major release of Stockfish. As
always, you can freely download it at https://stockfishchess.org/download and
use it in the GUI of your choice.

Don’t forget to join our Discord server[1] to get in touch with the community
of developers and users of the project!

*Quality of chess play*

In tests against Stockfish 16, this release brings an Elo gain of up to 46
points[2] and wins up to 4.5 times more game pairs[3] than it loses. In
practice, high-quality moves are now found in less time, with a user upgrading
from Stockfish 14 being able to analyze games at least 6 times[4] faster with
Stockfish 17 while maintaining roughly the same quality.

During this development period, Stockfish won its 9th consecutive first place
in the main league of the Top Chess Engine Championship (TCEC)[5], and the 24th
consecutive first place in the main events (bullet, blitz, and rapid) of the
Computer Chess Championship (CCC)[6].

*Update highlights*

*Improved engine lines*

This release introduces principal variations (PVs) that are more informative
for mate and decisive table base (TB) scores. In both cases, the PV will
contain all moves up to checkmate. For mate scores, the PV shown is the best
variation known to the engine at that point, while for table base wins, it
follows, based on the TB, a sequence of moves that preserves the game outcome
to checkmate.

*NUMA performance optimization*

For high-end computers with multiple CPUs (typically a dual-socket architecture
with 100+ cores), this release automatically improves performance with a
`NumaPolicy` setting that optimizes non-uniform memory access (NUMA).  Although
typical consumer hardware will not benefit, speedups of up to 2.8x[7] have been
measured.

*Shoutouts*

*ChessDB*

During the past 1.5 years, hundreds of cores have been continuously running
Stockfish to grow a database of analyzed positions. This chess cloud
database[8] now contains well over 45 billion positions, providing excellent
coverage of all openings and commonly played lines. This database is already
integrated into GUIs such as En Croissant[9] and Nibbler[10], which access it
through the public API.

*Leela Chess Zero*

Generally considered to be the strongest GPU engine, it continues to provide
open data which is essential for training our NNUE networks. They released
version 0.31.1[11] of their engine a few weeks ago, check it out!

*Website redesign*

Our website has undergone a redesign in recent months, most notably in our home
page[12], now featuring a darker color scheme and a more modern aesthetic,
while still maintaining its core identity. We hope you'll like it as much as we
do!

*Thank you*

The Stockfish project builds on a thriving community of enthusiasts (thanks
everybody!) who contribute their expertise, time, and resources to build a free
and open-source chess engine that is robust, widely available, and very strong.

We would like to express our gratitude for the 11k stars[13] that light up our
GitHub project! Thank you for your support and encouragement – your recognition
means a lot to us.

We invite our chess fans to join the Fishtest testing framework[14] to
contribute compute resources needed for development. Programmers can contribute
to the project either directly to Stockfish[15] (C++), to Fishtest[16] (HTML,
CSS, JavaScript, and Python), to our trainer nnue-pytorch[17] (C++ and Python),
or to our website[18] (HTML, CSS/SCSS, and JavaScript).

The Stockfish team

[1] https://discord.gg/GWDRS3kU6R
[2] https://tests.stockfishchess.org/tests/view/66d738ba9de3e7f9b33d159a
[3] https://tests.stockfishchess.org/tests/view/66d738f39de3e7f9b33d15a0
[4] https://github.com/official-stockfish/Stockfish/wiki/Useful-data#equivalent-time-odds-and-normalized-game-pair-elo
[5] https://en.wikipedia.org/wiki/Stockfish_(chess)#Top_Chess_Engine_Championship
[6] https://en.wikipedia.org/wiki/Stockfish_(chess)#Chess.com_Computer_Chess_Championship
[7] https://github.com/official-stockfish/Stockfish/pull/5285
[8] https://chessdb.cn/queryc_en/
[9] https://encroissant.org/
[10] https://github.com/rooklift/nibbler
[11] https://github.com/LeelaChessZero/lc0/releases/tag/v0.31.1
[12] https://stockfishchess.org/
[13] https://github.com/official-stockfish/Stockfish/stargazers
[14] https://github.com/official-stockfish/fishtest/wiki/Running-the-worker
[15] https://github.com/official-stockfish/Stockfish
[16] https://github.com/official-stockfish/fishtest
[17] https://github.com/official-stockfish/nnue-pytorch
[18] https://github.com/official-stockfish/stockfish-web

.github/ci/libcxx17.imp

@@ -7,6 +7,7 @@
     { 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 ] },

.github/workflows/arm_compilation.yml

@@ -10,7 +10,7 @@ jobs:
     name: ${{ matrix.config.name }} ${{ matrix.binaries }}
     runs-on: ${{ matrix.config.os }}
     env:
-      COMPILER: ${{ matrix.config.compiler }}
+      COMPCXX: ${{ matrix.config.compiler }}
       COMP: ${{ matrix.config.comp }}
       EMU: ${{ matrix.config.emu }}
       EXT: ${{ matrix.config.ext }}
@@ -26,6 +26,7 @@ jobs:
       - uses: actions/checkout@v4
         with:
           fetch-depth: 0
+          persist-credentials: false
 
       - name: Download required linux packages
         if: runner.os == 'Linux'
@@ -62,7 +63,7 @@ jobs:
           if [ $COMP == ndk ]; then
             export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
           fi
-          $COMPILER -v
+          $COMPCXX -v
 
       - name: Test help target
         run: make help
@@ -91,4 +92,7 @@ jobs:
         uses: actions/upload-artifact@v4
         with:
           name: ${{ matrix.config.simple_name }} ${{ matrix.binaries }}
-          path: .
+          path: |
+            .
+            !.git
+            !.output

.github/workflows/clang-format.yml

@@ -11,6 +11,10 @@ on:
     paths:
       - "**.cpp"
       - "**.h"
+
+permissions:
+  pull-requests: write
+
 jobs:
   Clang-Format:
     name: Clang-Format
@@ -25,27 +29,29 @@ jobs:
         id: clang-format
         continue-on-error: true
         with:
-          clang-format-version: "17"
+          clang-format-version: "18"
           exclude-regex: "incbin"
 
       - name: Comment on PR
         if: steps.clang-format.outcome == 'failure'
-        uses: thollander/actions-comment-pull-request@1d3973dc4b8e1399c0620d3f2b1aa5e795465308 # @v2.4.3
+        uses: thollander/actions-comment-pull-request@fabd468d3a1a0b97feee5f6b9e499eab0dd903f6 # @v2.5.0
         with:
           message: |
-            clang-format 17 needs to be run on this PR.
+            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/mantic/clang-format-17.
+            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@1d3973dc4b8e1399c0620d3f2b1aa5e795465308 # @v2.4.3
+        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

@@ -30,6 +30,8 @@ jobs:
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
+        with:
+          persist-credentials: false
 
       # Initializes the CodeQL tools for scanning.
       - name: Initialize CodeQL

.github/workflows/compilation.yml

@@ -10,7 +10,7 @@ jobs:
     name: ${{ matrix.config.name }} ${{ matrix.binaries }}
     runs-on: ${{ matrix.config.os }}
     env:
-      COMPILER: ${{ matrix.config.compiler }}
+      COMPCXX: ${{ matrix.config.compiler }}
       COMP: ${{ matrix.config.comp }}
       EXT: ${{ matrix.config.ext }}
       NAME: ${{ matrix.config.simple_name }}
@@ -25,6 +25,8 @@ jobs:
         shell: ${{ matrix.config.shell }}
     steps:
       - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
 
       - name: Install fixed GCC on Linux
         if: runner.os == 'Linux'
@@ -50,7 +52,7 @@ jobs:
         run: make net
 
       - name: Check compiler
-        run: $COMPILER -v
+        run: $COMPCXX -v
 
       - name: Test help target
         run: make help
@@ -59,7 +61,7 @@ jobs:
         run: git --version
 
       - name: Check compiler
-        run: $COMPILER -v
+        run: $COMPCXX -v
 
       - name: Show g++ cpu info
         if: runner.os != 'macOS'
@@ -86,4 +88,7 @@ jobs:
         uses: actions/upload-artifact@v4
         with:
           name: ${{ matrix.config.simple_name }} ${{ matrix.binaries }}
-          path: .
+          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 fast-chess repo
+        uses: actions/checkout@v4
+        with:
+          repository: Disservin/fast-chess
+          path: fast-chess
+          ref: d54af1910d5479c669dc731f1f54f9108a251951
+          persist-credentials: false
+
+      - name: fast-chess build
+        working-directory: fast-chess
+        run: make -j
+
+      - name: Run games
+        working-directory: fast-chess
+        run: |
+          ./fast-chess  -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

@@ -14,6 +14,7 @@ jobs:
         uses: actions/checkout@v4
         with:
           path: Stockfish
+          persist-credentials: false
 
       - name: Checkout include-what-you-use
         uses: actions/checkout@v4
@@ -21,6 +22,7 @@ jobs:
           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: |

.github/workflows/matetrack.yml

@@ -0,0 +1,54 @@
+# 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: 814160f82e6428ed2f6522dc06c2a6fa539cd413
+          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

.github/workflows/sanitizers.yml

@@ -6,7 +6,7 @@ jobs:
     name: ${{ matrix.sanitizers.name }}
     runs-on: ${{ matrix.config.os }}
     env:
-      COMPILER: ${{ matrix.config.compiler }}
+      COMPCXX: ${{ matrix.config.compiler }}
       COMP: ${{ matrix.config.comp }}
       CXXFLAGS: "-Werror"
     strategy:
@@ -31,12 +31,17 @@ jobs:
           - name: Run under valgrind-thread
             make_option: ""
             instrumented_option: valgrind-thread
+          - name: Run non-instrumented
+            make_option: ""
+            instrumented_option: none
     defaults:
       run:
         working-directory: src
         shell: ${{ matrix.config.shell }}
     steps:
       - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
 
       - name: Download required linux packages
         run: |
@@ -47,7 +52,7 @@ jobs:
         run: make net
 
       - name: Check compiler
-        run: $COMPILER -v
+        run: $COMPCXX -v
 
       - name: Test help target
         run: make help
@@ -55,6 +60,14 @@ jobs:
       - 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 }}

.github/workflows/stockfish.yml

@@ -15,7 +15,13 @@ jobs:
   Prerelease:
     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@v4
+        with:
+          persist-credentials: false
+
       # returns null if no pre-release exists
       - name: Get Commit SHA of Latest Pre-release
         run: |
@@ -23,14 +29,40 @@ 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
+      # delete old previous pre-release and tag
-      - uses: actions/checkout@v4
+      - run: gh release delete ${{ env.COMMIT_SHA_TAG }} --cleanup-tag
-      - run: gh release delete ${{ env.COMMIT_SHA }} --cleanup-tag
+        if: env.COMMIT_SHA_TAG != 'null'
-        if: env.COMMIT_SHA != 'null'
         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:
@@ -38,6 +70,8 @@ jobs:
       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) )
@@ -62,15 +96,27 @@ jobs:
     uses: ./.github/workflows/sanitizers.yml
   Tests:
     uses: ./.github/workflows/tests.yml
+  Matetrack:
+    uses: ./.github/workflows/matetrack.yml
+  Games:
+    uses: ./.github/workflows/games.yml
   Binaries:
     if: github.repository == 'official-stockfish/Stockfish'
     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.repository == 'official-stockfish/Stockfish'
     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/tests.yml

@@ -6,7 +6,7 @@ jobs:
     name: ${{ matrix.config.name }}
     runs-on: ${{ matrix.config.os }}
     env:
-      COMPILER: ${{ matrix.config.compiler }}
+      COMPCXX: ${{ matrix.config.compiler }}
       COMP: ${{ matrix.config.comp }}
       CXXFLAGS: "-Werror"
     strategy:
@@ -106,6 +106,7 @@ jobs:
       - uses: actions/checkout@v4
         with:
           fetch-depth: 0
+          persist-credentials: false
 
       - name: Download required linux packages
         if: runner.os == 'Linux'
@@ -147,7 +148,7 @@ jobs:
 
       - name: Download required macOS packages
         if: runner.os == 'macOS'
-        run: brew install coreutils
+        run: brew install coreutils gcc@11
 
       - name: Setup msys and install required packages
         if: runner.os == 'Windows'
@@ -172,9 +173,9 @@ jobs:
             if [ $COMP == ndk ]; then
               export PATH=${{ env.ANDROID_NDK_BIN }}:$PATH
             fi
-            $COMPILER -v
+            $COMPCXX -v
           else
-            echo "$COMPILER -v" > script.sh
+            echo "$COMPCXX -v" > script.sh
             docker run --rm --platform ${{ matrix.config.platform }} -v ${{ github.workspace }}/src:/app sf_builder
           fi
 

.github/workflows/upload_binaries.yml

@@ -5,13 +5,16 @@ on:
       matrix:
         type: string
         required: true
+    secrets:
+      token:
+        required: true
 
 jobs:
   Artifacts:
     name: ${{ matrix.config.name }} ${{ matrix.binaries }}
     runs-on: ${{ matrix.config.os }}
     env:
-      COMPILER: ${{ matrix.config.compiler }}
+      COMPCXX: ${{ matrix.config.compiler }}
       COMP: ${{ matrix.config.comp }}
       EXT: ${{ matrix.config.ext }}
       NAME: ${{ matrix.config.simple_name }}
@@ -25,6 +28,8 @@ jobs:
         shell: ${{ matrix.config.shell }}
     steps:
       - uses: actions/checkout@v4
+        with:
+          persist-credentials: false
 
       - name: Download artifact from compilation
         uses: actions/download-artifact@v4
@@ -65,6 +70,7 @@ jobs:
       - 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
@@ -77,6 +83,7 @@ jobs:
         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
@@ -97,9 +104,10 @@ jobs:
       - name: Prerelease
         if: github.ref_name == 'master' && env.CHANGES == '0'
         continue-on-error: true
-        uses: softprops/action-gh-release@de2c0eb89ae2a093876385947365aca7b0e5f844 # @v1
+        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 }}

AUTHORS

@@ -20,6 +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)
@@ -46,6 +47,7 @@ Bryan Cross (crossbr)
 candirufish
 Chess13234
 Chris Cain (ceebo)
+Ciekce
 clefrks
 Clemens L. (rn5f107s2)
 Cody Ho (aesrentai)
@@ -67,9 +69,11 @@ 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)
@@ -126,6 +130,7 @@ Kojirion
 Krystian Kuzniarek (kuzkry)
 Leonardo Ljubičić (ICCF World Champion)
 Leonid Pechenik (lp--)
+Li Ying (yl25946)
 Liam Keegan (lkeegan)
 Linmiao Xu (linrock)
 Linus Arver (listx)
@@ -166,6 +171,7 @@ Niklas Fiekas (niklasf)
 Nikolay Kostov (NikolayIT)
 Norman Schmidt (FireFather)
 notruck
+Nour Berakdar (Nonlinear)
 Ofek Shochat (OfekShochat, ghostway)
 Ondrej Mosnáček (WOnder93)
 Ondřej Mišina (AndrovT)
@@ -204,6 +210,7 @@ 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)

CONTRIBUTING.md

@@ -59,7 +59,7 @@ discussion._
 
 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 17 to be installed on your system.
+`make format`. This requires clang-format version 18 to be installed on your system.
 
 ## Navigate
 

Top CPU Contributors.txt

@@ -1,106 +1,109 @@
-Contributors to Fishtest with >10,000 CPU hours, as of 2024-02-24.
+Contributors to Fishtest with >10,000 CPU hours, as of 2024-08-31.
 Thank you!
 
 Username                                CPU Hours     Games played
 ------------------------------------------------------------------
-noobpwnftw                               39302472       3055513453
+noobpwnftw                               40428649       3164740143
-technologov                              20845762        994893444
+technologov                              23581394       1076895482
-linrock                                   8616428        560281417
+vdv                                      19425375        718302718
+linrock                                  10034115        643194527
 mlang                                     3026000        200065824
-okrout                                    2332151        222639518
+okrout                                    2572676        237511408
-pemo                                      1800019         60274069
+pemo                                      1836785         62226157
 dew                                       1689162        100033738
-TueRens                                   1474943         75121774
+TueRens                                   1648780         77891164
-grandphish2                               1463002         91616949
+sebastronomy                              1468328         60859092
-JojoM                                     1109702         72927902
+grandphish2                               1466110         91776075
-olafm                                      978631         71037944
+JojoM                                     1130625         73666098
-sebastronomy                               939955         44920556
+olafm                                     1067009         74807270
 tvijlbrief                                 796125         51897690
-gvreuls                                    711320         49142318
+oz                                         781847         53910686
+rpngn                                      768460         49812975
+gvreuls                                    751085         52177668
 mibere                                     703840         46867607
-oz                                         646268         46293638
+leszek                                     566598         42024615
-rpngn                                      572571         38928563
+cw                                         519601         34988161
-leszek                                     531858         39316505
-cw                                         518116         34894291
 fastgm                                     503862         30260818
 CSU_Dynasty                                468784         31385034
-ctoks                                      434591         28520597
+maximmasiutin                              439192         27893522
-maximmasiutin                              429983         27066286
+ctoks                                      435148         28541909
 crunchy                                    427414         27371625
 bcross                                     415724         29061187
+robal                                      371112         24642270
+mgrabiak                                   367963         26464704
 velislav                                   342588         22140902
-mgrabiak                                   338763         23999170
+ncfish1                                    329039         20624527
 Fisherman                                  327231         21829379
-robal                                      299836         20213182
 Dantist                                    296386         18031762
-ncfish1                                    267604         17881149
+tolkki963                                  262050         22049676
+Sylvain27                                  255595          8864404
 nordlandia                                 249322         16420192
+Fifis                                      237657         13065577
 marrco                                     234581         17714473
-tolkki963                                  233490         19773930
+Calis007                                   217537         14450582
 glinscott                                  208125         13277240
 drabel                                     204167         13930674
 mhoram                                     202894         12601997
 bking_US                                   198894         11876016
-Calis007                                   188631         12795784
 Thanar                                     179852         12365359
-Fifis                                      176209         10638245
+javran                                     169679         13481966
-vdv                                        175544          9904472
+armo9494                                   162863         10937118
 spams                                      157128         10319326
-DesolatedDodo                              156659         10210328
+DesolatedDodo                              156683         10211206
-armo9494                                   155355         10566898
+Wencey                                     152308          8375444
 sqrt2                                      147963          9724586
+vdbergh                                    140311          9225125
 jcAEie                                     140086         10603658
-vdbergh                                    139746          9172061
 CoffeeOne                                  137100          5024116
 malala                                     136182          8002293
 xoto                                       133759          9159372
+Dubslow                                    129614          8519312
 davar                                      129023          8376525
 DMBK                                       122960          8980062
 dsmith                                     122059          7570238
-javran                                     121564         10144656
+CypressChess                               120784          8672620
+sschnee                                    120526          7547722
+maposora                                   119734         10749710
 amicic                                     119661          7938029
-sschnee                                    118107          7389266
+Wolfgang                                   115713          8159062
-Wolfgang                                   114616          8070494
 Data                                       113305          8220352
 BrunoBanani                                112960          7436849
-Wencey                                     111502          5991676
+markkulix                                  112897          9133168
-cuistot                                    108503          7006992
+cuistot                                    109802          7121030
-CypressChess                               108331          7759788
 skiminki                                   107583          7218170
+sterni1971                                 104431          5938282
 MaZePallas                                 102823          6633619
-sterni1971                                 100532          5880772
 sunu                                       100167          7040199
 zeryl                                       99331          6221261
 thirdlife                                   99156          2245320
 ElbertoOne                                  99028          7023771
-Dubslow                                     98600          6903242
+megaman7de                                  98456          6675076
-markkulix                                   97010          7643900
+Goatminola                                  96765          8257832
-bigpen0r                                    94809          6529203
+bigpen0r                                    94825          6529241
 brabos                                      92118          6186135
 Maxim                                       90818          3283364
 psk                                         89957          5984901
-megaman7de                                  88822          6052132
 racerschmacer                               85805          6122790
-maposora                                    85710          7778146
 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
 Pking_cda                                   73776          5293873
-yurikvelo                                   73516          5036928
+yurikvelo                                   73611          5046822
-MarcusTullius                               71053          4803477
+Mineta                                      71130          4711422
 Bobo1239                                    70579          4794999
 solarlight                                  70517          5028306
 dv8silencer                                 70287          3883992
-Spprtr                                      69646          4806763
-Mineta                                      66325          4537742
 manap                                       66273          4121774
-szupaw                                      65468          5669742
 tinker                                      64333          4268790
 qurashee                                    61208          3429862
-woutboat                                    59496          4906352
 AGI                                         58195          4329580
 robnjr                                      57262          4053117
 Freja                                       56938          3733019
@@ -108,39 +111,45 @@ MaxKlaxxMiner                               56879          3423958
 ttruscott                                   56010          3680085
 rkl                                         55132          4164467
 jmdana                                      54697          4012593
+notchris                                    53936          4184018
 renouve                                     53811          3501516
-notchris                                    52433          4044590
 finfish                                     51360          3370515
 eva42                                       51272          3599691
 eastorwest                                  51117          3454811
-Goatminola                                  51004          4432492
 rap                                         49985          3219146
 pb00067                                     49733          3298934
 GPUex                                       48686          3684998
 OuaisBla                                    48626          3445134
 ronaldjerum                                 47654          3240695
 biffhero                                    46564          3111352
-oryx                                        45533          3539290
+oryx                                        45639          3546530
 VoyagerOne                                  45476          3452465
 speedycpu                                   43842          3003273
 jbwiebe                                     43305          2805433
 Antihistamine                               41788          2761312
 mhunt                                       41735          2691355
+jibarbosa                                   41640          4145702
 homyur                                      39893          2850481
 gri                                         39871          2515779
+DeepnessFulled                              39020          3323102
 Garf                                        37741          2999686
 SC                                          37299          2731694
-Sylvain27                                   36520          1467082
+Gaster319                                   37118          3279678
+naclosagc                                   36562          1279618
 csnodgrass                                  36207          2688994
-Gaster319                                   35655          3149442
 strelock                                    34716          2074055
+gopeto                                      33717          2245606
 EthanOConnor                                33370          2090311
 slakovv                                     32915          2021889
-gopeto                                      31884          2076712
+jojo2357                                    32890          2826662
+shawnxu                                     32019          2802552
 Gelma                                       31771          1551204
+vidar808                                    31560          1351810
 kdave                                       31157          2198362
 manapbk                                     30987          1810399
-ZacHFX                                      30551          2238078
+ZacHFX                                      30966          2272416
+TataneSan                                   30713          1513402
+votoanthuan                                 30691          2460856
 Prcuvu                                      30377          2170122
 anst                                        30301          2190091
 jkiiski                                     30136          1904470
@@ -149,14 +158,15 @@ hyperbolic.tom                              29840          2017394
 chuckstablers                               29659          2093438
 Pyafue                                      29650          1902349
 belzedar94                                  28846          1811530
-votoanthuan                                 27978          2285818
+mecevdimitar                                27610          1721382
-shawnxu                                     27438          2465810
 chriswk                                     26902          1868317
 xwziegtm                                    26897          2124586
 achambord                                   26582          1767323
+somethingintheshadows                       26496          2186404
 Patrick_G                                   26276          1801617
 yorkman                                     26193          1992080
-Ulysses                                     25397          1701264
+srowen                                      25743          1490684
+Ulysses                                     25413          1702830
 Jopo12321                                   25227          1652482
 SFTUser                                     25182          1675689
 nabildanial                                 25068          1531665
@@ -164,66 +174,69 @@ Sharaf_DG                                   24765          1786697
 rodneyc                                     24376          1416402
 jsys14                                      24297          1721230
 agg177                                      23890          1395014
-srowen                                      23842          1342508
+AndreasKrug                                 23754          1890115
 Ente                                        23752          1678188
-jojo2357                                    23479          2061238
 JanErik                                     23408          1703875
 Isidor                                      23388          1680691
 Norabor                                     23371          1603244
+WoodMan777                                  23253          2023048
+Nullvalue                                   23155          2022752
 cisco2015                                   22920          1763301
 Zirie                                       22542          1472937
-Nullvalue                                   22490          1970374
-AndreasKrug                                 22485          1769491
 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
-qoo_charly_cai                              21135          1514907
+qoo_charly_cai                              21136          1514927
+IslandLambda                                21062          1220838
 jjoshua2                                    21001          1423089
 Zake9298                                    20938          1565848
 horst.prack                                 20878          1465656
+fishtester                                  20729          1348888
 0xB00B1ES                                   20590          1208666
-Serpensin                                   20487          1729674
+ols                                         20477          1195945
-Dinde                                       20440          1292390
+Dinde                                       20459          1292774
 j3corre                                     20405           941444
 Adrian.Schmidt123                           20316          1281436
 wei                                         19973          1745989
-fishtester                                  19617          1257388
+teenychess                                  19819          1762006
 rstoesser                                   19569          1293588
 eudhan                                      19274          1283717
 vulcan                                      18871          1729392
+wizardassassin                              18795          1376884
 Karpovbot                                   18766          1053178
-WoodMan777                                  18556          1628264
 jundery                                     18445          1115855
+mkstockfishtester                           18350          1690676
 ville                                       17883          1384026
 chris                                       17698          1487385
 purplefishies                               17595          1092533
 dju                                         17414           981289
-ols                                         17291          1042003
 iisiraider                                  17275          1049015
-Skiff84                                     17111           950248
 DragonLord                                  17014          1162790
+Karby                                       17008          1013160
+pirt                                        16965          1271519
 redstone59                                  16842          1461780
-Karby                                       16839          1010124
 Alb11747                                    16787          1213990
-pirt                                        16493          1237199
 Naven94                                     16414           951718
-wizardassassin                              16392          1148672
+scuzzi                                      16115           994341
 IgorLeMasson                                16064          1147232
-scuzzi                                      15757           968735
 ako027ako                                   15671          1173203
+infinigon                                   15285           965966
 Nikolay.IT                                  15154          1068349
 Andrew Grant                                15114           895539
 OssumOpossum                                14857          1007129
 LunaticBFF57                                14525          1190310
 enedene                                     14476           905279
-IslandLambda                                14393           958196
+Hjax                                        14394          1005013
 bpfliegel                                   14233           882523
 YELNAMRON                                   14230          1128094
 mpx86                                       14019           759568
@@ -233,54 +246,56 @@ Nesa92                                      13806          1116101
 crocogoat                                   13803          1117422
 joster                                      13710           946160
 mbeier                                      13650          1044928
-Hjax                                        13535           915487
+Pablohn26                                   13552          1088532
+wxt9861                                     13550          1312306
 Dark_wizzie                                 13422          1007152
 Rudolphous                                  13244           883140
 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
-tsim67                                      12570           890180
-Jackfish                                    12553           836958
 fatmurphy                                   12547           853210
-Oakwen                                      12503           853105
+Oakwen                                      12532           855759
+icewulf                                     12447           854878
 SapphireBrand                               12416           969604
 deflectooor                                 12386           579392
 modolief                                    12386           896470
-TataneSan                                   12358           609332
 Farseer                                     12249           694108
+Hongildong                                  12201           648712
 pgontarz                                    12151           848794
 dbernier                                    12103           860824
-FormazChar                                  11989           907809
+szczur90                                    12035           942376
+FormazChar                                  12019           910409
+rensonthemove                               11999           971993
 stocky                                      11954           699440
-somethingintheshadows                       11940           989472
+MooTheCow                                   11923           779432
-MooTheCow                                   11892           776126
 3cho                                        11842          1036786
-whelanh                                     11557           245188
+ckaz                                        11792           732276
 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
-ckaz                                        11170           680866
+ImperiumAeternum                            11155           952000
 snicolet                                    11106           869170
 dapper                                      11032           771402
 Ethnikoi                                    10993           945906
 Snuuka                                      10938           435504
-Karmatron                                   10859           678058
+Karmatron                                   10871           678306
 basepi                                      10637           744851
-jibarbosa                                   10628           857100
 Cubox                                       10621           826448
-mecevdimitar                                10609           787318
+gerbil                                      10519           971688
 michaelrpg                                  10509           739239
-Def9Infinity                                10427           686978
 OIVAS7572                                   10420           995586
-wxt9861                                     10412          1013864
 Garruk                                      10365           706465
 dzjp                                        10343           732529
+RickGroszkiewicz                            10263           990798

src/Makefile

@@ -55,15 +55,15 @@ PGOBENCH = $(WINE_PATH) ./$(EXE) bench
 SRCS = benchmark.cpp bitboard.cpp evaluate.cpp main.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 \
-		nnue/evaluate_nnue.h nnue/features/half_ka_v2_hm.h nnue/layers/affine_transform.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
+		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))
 
@@ -153,8 +153,8 @@ dotprod = no
 arm_version = 0
 STRIP = strip
 
-ifneq ($(shell which clang-format-17 2> /dev/null),)
+ifneq ($(shell which clang-format-18 2> /dev/null),)
-	CLANG-FORMAT = clang-format-17
+	CLANG-FORMAT = clang-format-18
 else
 	CLANG-FORMAT = clang-format
 endif
@@ -489,8 +489,8 @@ 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)
@@ -546,11 +546,6 @@ 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)
@@ -1056,14 +1051,14 @@ 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

@@ -23,8 +23,6 @@
 #include <iostream>
 #include <vector>
 
-#include "position.h"
-
 namespace {
 
 // clang-format off
@@ -95,7 +93,7 @@ const std::vector<std::string> Defaults = {
 
 }  // namespace
 
-namespace Stockfish {
+namespace Stockfish::Benchmark {
 
 // Builds a list of UCI commands to be run by bench. There
 // are five parameters: TT size in MB, number of search threads that
@@ -108,7 +106,7 @@ namespace Stockfish {
 // 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 Position& current, std::istream& is) {
+std::vector<std::string> setup_bench(const std::string& currentFen, std::istream& is) {
 
     std::vector<std::string> fens, list;
     std::string              go, token;
@@ -126,7 +124,7 @@ std::vector<std::string> setup_bench(const Position& current, std::istream& is)
         fens = Defaults;
 
     else if (fenFile == "current")
-        fens.push_back(current.fen());
+        fens.push_back(currentFen);
 
     else
     {

src/benchmark.h

@@ -23,11 +23,9 @@
 #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&);
 
 }  // namespace Stockfish
 

src/bitboard.cpp

@@ -124,8 +124,14 @@ Bitboard sliding_attack(PieceType pt, Square sq, Bitboard occupied) {
     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;

src/engine.cpp

@@ -0,0 +1,335 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2024 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::string path) :
+    binaryDirectory(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());
+    capSq = SQ_NONE;
+
+    options["Debug Log File"] << Option("", [](const Option& o) {
+        start_logger(o);
+        return std::nullopt;
+    });
+
+    options["NumaPolicy"] << Option("auto", [this](const Option& o) {
+        set_numa_config_from_option(o);
+        return numa_config_information_as_string() + "\n" + thread_binding_information_as_string();
+    });
+
+    options["Threads"] << Option(1, 1, 1024, [this](const Option&) {
+        resize_threads();
+        return thread_binding_information_as_string();
+    });
+
+    options["Hash"] << Option(16, 1, MaxHashMB, [this](const Option& o) {
+        set_tt_size(o);
+        return std::nullopt;
+    });
+
+    options["Clear Hash"] << Option([this](const Option&) {
+        search_clear();
+        return std::nullopt;
+    });
+    options["Ponder"] << Option(false);
+    options["MultiPV"] << Option(1, 1, MAX_MOVES);
+    options["Skill Level"] << Option(20, 0, 20);
+    options["Move Overhead"] << Option(10, 0, 5000);
+    options["nodestime"] << Option(0, 0, 10000);
+    options["UCI_Chess960"] << Option(false);
+    options["UCI_LimitStrength"] << Option(false);
+    options["UCI_Elo"] << Option(Stockfish::Search::Skill::LowestElo,
+                                 Stockfish::Search::Skill::LowestElo,
+                                 Stockfish::Search::Skill::HighestElo);
+    options["UCI_ShowWDL"] << Option(false);
+    options["SyzygyPath"] << Option("", [](const Option& o) {
+        Tablebases::init(o);
+        return std::nullopt;
+    });
+    options["SyzygyProbeDepth"] << Option(1, 1, 100);
+    options["Syzygy50MoveRule"] << Option(true);
+    options["SyzygyProbeLimit"] << Option(7, 0, 7);
+    options["EvalFile"] << Option(EvalFileDefaultNameBig, [this](const Option& o) {
+        load_big_network(o);
+        return std::nullopt;
+    });
+    options["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();
+    limits.capSq = capSq;
+
+    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::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());
+
+    capSq = SQ_NONE;
+    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());
+
+        capSq          = SQ_NONE;
+        DirtyPiece& dp = states->back().dirtyPiece;
+        if (dp.dirty_num > 1 && dp.to[1] == SQ_NONE)
+            capSq = m.to_sq();
+    }
+}
+
+// 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"]);
+    networks->small.verify(options["EvalFileSmall"]);
+}
+
+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();
+}
+
+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;
+
+    size_t threadsSize = threads.size();
+    ss << "Using " << threadsSize << (threadsSize > 1 ? " threads" : " thread");
+
+    if (boundThreadsByNode.empty())
+        return ss.str();
+
+    ss << " with NUMA node thread binding: ";
+
+    bool isFirst = true;
+
+    for (auto&& [current, total] : boundThreadsByNode)
+    {
+        if (!isFirst)
+            ss << ":";
+        ss << current << "/" << total;
+        isFirst = false;
+    }
+
+    return ss.str();
+}
+
+}

src/engine.h

@@ -0,0 +1,128 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2024 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 {
+
+enum Square : int;
+
+class Engine {
+   public:
+    using InfoShort = Search::InfoShort;
+    using InfoFull  = Search::InfoFull;
+    using InfoIter  = Search::InfoIteration;
+
+    Engine(std::string path = "");
+
+    // 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)>&&);
+
+    // 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();
+
+    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_binding_information_as_string() const;
+
+   private:
+    const std::string binaryDirectory;
+
+    NumaReplicationContext numaContext;
+
+    Position     pos;
+    StateListPtr states;
+    Square       capSq;
+
+    OptionsMap                               options;
+    ThreadPool                               threads;
+    TranspositionTable                       tt;
+    LazyNumaReplicated<Eval::NNUE::Networks> networks;
+
+    Search::SearchManager::UpdateContext updateContext;
+};
+
+}  // namespace Stockfish
+
+
+#endif  // #ifndef ENGINE_H_INCLUDED

src/evaluate.cpp

@@ -22,161 +22,21 @@
 #include <cassert>
 #include <cmath>
 #include <cstdlib>
-#include <fstream>
 #include <iomanip>
 #include <iostream>
-#include <optional>
+#include <memory>
 #include <sstream>
-#include <unordered_map>
+#include <tuple>
-#include <vector>
 
-#include "incbin/incbin.h"
+#include "nnue/network.h"
-#include "misc.h"
+#include "nnue/nnue_misc.h"
-#include "nnue/evaluate_nnue.h"
-#include "nnue/nnue_architecture.h"
 #include "position.h"
 #include "types.h"
 #include "uci.h"
-#include "ucioption.h"
+#include "nnue/nnue_accumulator.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 Stockfish {
 
-namespace Eval {
-
-
-// Tries to load a NNUE network at startup time, or when the engine
-// receives a UCI command "setoption name EvalFile value nn-[a-z0-9]{12}.nnue"
-// The name of the NNUE network is always retrieved from the EvalFile option.
-// We search the given network in three locations: internally (the default
-// network may be embedded in the binary), in the active working directory and
-// in the engine directory. Distro packagers may define the DEFAULT_NNUE_DIRECTORY
-// variable to have the engine search in a special directory in their distro.
-NNUE::EvalFiles NNUE::load_networks(const std::string& rootDirectory,
-                                    const OptionsMap&  options,
-                                    NNUE::EvalFiles    evalFiles) {
-
-    for (auto& [netSize, evalFile] : evalFiles)
-    {
-        std::string user_eval_file = options[evalFile.optionName];
-
-        if (user_eval_file.empty())
-            user_eval_file = evalFile.defaultName;
-
-#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
-
-        for (const std::string& directory : dirs)
-        {
-            if (evalFile.current != user_eval_file)
-            {
-                if (directory != "<internal>")
-                {
-                    std::ifstream stream(directory + user_eval_file, std::ios::binary);
-                    auto          description = NNUE::load_eval(stream, netSize);
-
-                    if (description.has_value())
-                    {
-                        evalFile.current        = user_eval_file;
-                        evalFile.netDescription = description.value();
-                    }
-                }
-
-                if (directory == "<internal>" && user_eval_file == evalFile.defaultName)
-                {
-                    // 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);
-                        }
-                    };
-
-                    MemoryBuffer buffer(
-                      const_cast<char*>(reinterpret_cast<const char*>(
-                        netSize == Small ? gEmbeddedNNUESmallData : gEmbeddedNNUEBigData)),
-                      size_t(netSize == Small ? gEmbeddedNNUESmallSize : gEmbeddedNNUEBigSize));
-                    (void) gEmbeddedNNUEBigEnd;  // Silence warning on unused variable
-                    (void) gEmbeddedNNUESmallEnd;
-
-                    std::istream stream(&buffer);
-                    auto         description = NNUE::load_eval(stream, netSize);
-
-                    if (description.has_value())
-                    {
-                        evalFile.current        = user_eval_file;
-                        evalFile.netDescription = description.value();
-                    }
-                }
-            }
-        }
-    }
-
-    return evalFiles;
-}
-
-// Verifies that the last net used was loaded successfully
-void NNUE::verify(const OptionsMap&                                        options,
-                  const std::unordered_map<Eval::NNUE::NetSize, EvalFile>& evalFiles) {
-
-    for (const auto& [netSize, evalFile] : evalFiles)
-    {
-        std::string user_eval_file = options[evalFile.optionName];
-
-        if (user_eval_file.empty())
-            user_eval_file = evalFile.defaultName;
-
-        if (evalFile.current != user_eval_file)
-        {
-            std::string msg1 =
-              "Network evaluation parameters compatible with the engine must be available.";
-            std::string msg2 =
-              "The network file " + user_eval_file + " 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.";
-
-            sync_cout << "info string ERROR: " << msg1 << sync_endl;
-            sync_cout << "info string ERROR: " << msg2 << sync_endl;
-            sync_cout << "info string ERROR: " << msg3 << sync_endl;
-            sync_cout << "info string ERROR: " << msg4 << sync_endl;
-            sync_cout << "info string ERROR: " << msg5 << sync_endl;
-
-            exit(EXIT_FAILURE);
-        }
-
-        sync_cout << "info string NNUE evaluation using " << user_eval_file << sync_endl;
-    }
-}
-}
-
 // Returns a static, purely materialistic evaluation of the position from
 // the point of view of the given color. It can be divided by PawnValue to get
 // an approximation of the material advantage on the board in terms of pawns.
@@ -185,31 +45,49 @@ int Eval::simple_eval(const Position& pos, Color c) {
          + (pos.non_pawn_material(c) - pos.non_pawn_material(~c));
 }
 
+bool Eval::use_smallnet(const Position& pos) {
+    int simpleEval = simple_eval(pos, pos.side_to_move());
+    return std::abs(simpleEval) > 962;
+}
 
 // Evaluate is the evaluator for the outer world. It returns a static evaluation
 // of the position from the point of view of the side to move.
-Value Eval::evaluate(const Position& pos, int optimism) {
+Value Eval::evaluate(const Eval::NNUE::Networks&    networks,
+                     const Position&                pos,
+                     Eval::NNUE::AccumulatorCaches& caches,
+                     int                            optimism) {
 
     assert(!pos.checkers());
 
-    int  simpleEval = simple_eval(pos, pos.side_to_move());
+    bool smallNet = use_smallnet(pos);
-    bool smallNet   = std::abs(simpleEval) > 1050;
+    int  v;
 
-    int nnueComplexity;
+    auto [psqt, positional] = smallNet ? networks.small.evaluate(pos, &caches.small)
+                                       : networks.big.evaluate(pos, &caches.big);
 
-    Value nnue = smallNet ? NNUE::evaluate<NNUE::Small>(pos, true, &nnueComplexity)
+    Value nnue = (125 * psqt + 131 * positional) / 128;
-                          : NNUE::evaluate<NNUE::Big>(pos, true, &nnueComplexity);
 
-    // Blend optimism and eval with nnue complexity and material imbalance
+    // Re-evaluate the position when higher eval accuracy is worth the time spent
-    optimism += optimism * (nnueComplexity + std::abs(simpleEval - nnue)) / 512;
+    if (smallNet && (nnue * psqt < 0 || std::abs(nnue) < 227))
-    nnue -= nnue * (nnueComplexity + std::abs(simpleEval - nnue)) / 32768;
+    {
+        std::tie(psqt, positional) = networks.big.evaluate(pos, &caches.big);
+        nnue                       = (125 * psqt + 131 * positional) / 128;
+        smallNet                   = false;
+    }
 
-    int npm = pos.non_pawn_material() / 64;
+    // Blend optimism and eval with nnue complexity
-    int v   = (nnue * (915 + npm + 9 * pos.count<PAWN>()) + optimism * (154 + npm)) / 1024;
+    int nnueComplexity = std::abs(psqt - positional);
+    optimism += optimism * nnueComplexity / (smallNet ? 433 : 453);
+    nnue -= nnue * nnueComplexity / (smallNet ? 18815 : 17864);
+
+    int material = (smallNet ? 553 : 532) * pos.count<PAWN>() + pos.non_pawn_material();
+    v = (nnue * (73921 + material) + optimism * (8112 + material)) / (smallNet ? 68104 : 74715);
+
+    // Evaluation grain (to get more alpha-beta cuts) with randomization (for robustness)
+    v = (v / 16) * 16 - 1 + (pos.key() & 0x2);
 
     // Damp down the evaluation linearly when shuffling
-    int shuffling = pos.rule50_count();
+    v -= v * pos.rule50_count() / 212;
-    v             = v * (200 - shuffling) / 214;
 
     // Guarantee evaluation does not hit the tablebase range
     v = std::clamp(v, VALUE_TB_LOSS_IN_MAX_PLY + 1, VALUE_TB_WIN_IN_MAX_PLY - 1);
@@ -221,25 +99,27 @@ Value Eval::evaluate(const Position& pos, int optimism) {
 // a string (suitable for outputting to stdout) that contains the detailed
 // descriptions and values of each evaluation term. Useful for debugging.
 // Trace scores are from white's point of view
-std::string Eval::trace(Position& pos) {
+std::string Eval::trace(Position& pos, const Eval::NNUE::Networks& networks) {
 
     if (pos.checkers())
         return "Final evaluation: none (in check)";
 
+    auto caches = std::make_unique<Eval::NNUE::AccumulatorCaches>(networks);
+
     std::stringstream ss;
     ss << std::showpoint << std::noshowpos << std::fixed << std::setprecision(2);
-    ss << '\n' << NNUE::trace(pos) << '\n';
+    ss << '\n' << NNUE::trace(pos, networks, *caches) << '\n';
 
     ss << std::showpoint << std::showpos << std::fixed << std::setprecision(2) << std::setw(15);
 
-    Value v;
+    auto [psqt, positional] = networks.big.evaluate(pos, &caches->big);
-    v = NNUE::evaluate<NNUE::Big>(pos, false);
+    Value v                 = psqt + positional;
-    v = pos.side_to_move() == WHITE ? v : -v;
+    v                       = pos.side_to_move() == WHITE ? v : -v;
-    ss << "NNUE evaluation        " << 0.01 * UCI::to_cp(v) << " (white side)\n";
+    ss << "NNUE evaluation        " << 0.01 * UCIEngine::to_cp(v, pos) << " (white side)\n";
 
-    v = evaluate(pos, VALUE_ZERO);
+    v = evaluate(networks, pos, *caches, VALUE_ZERO);
     v = pos.side_to_move() == WHITE ? v : -v;
-    ss << "Final evaluation       " << 0.01 * UCI::to_cp(v) << " (white side)";
+    ss << "Final evaluation       " << 0.01 * UCIEngine::to_cp(v, pos) << " (white side)";
     ss << " [with scaled NNUE, ...]";
     ss << "\n";
 

src/evaluate.h

@@ -20,50 +20,35 @@
 #define EVALUATE_H_INCLUDED
 
 #include <string>
-#include <unordered_map>
 
 #include "types.h"
 
 namespace Stockfish {
 
 class Position;
-class OptionsMap;
 
 namespace Eval {
 
-std::string trace(Position& pos);
-
-int   simple_eval(const Position& pos, Color c);
-Value evaluate(const Position& pos, int optimism);
-
 // The default net name MUST follow the format nn-[SHA256 first 12 digits].nnue
 // for the build process (profile-build and fishtest) to work. Do not change the
-// name of the macro, as it is used in the Makefile.
+// name of the macro or the location where this macro is defined, as it is used
-#define EvalFileDefaultNameBig "nn-b1a57edbea57.nnue"
+// in the Makefile/Fishtest.
-#define EvalFileDefaultNameSmall "nn-baff1ede1f90.nnue"
+#define EvalFileDefaultNameBig "nn-1111cefa1111.nnue"
-
+#define EvalFileDefaultNameSmall "nn-37f18f62d772.nnue"
-struct EvalFile {
-    // UCI option name
-    std::string optionName;
-    // Default net name, will use one of the macros above
-    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;
-};
 
 namespace NNUE {
+struct Networks;
+struct AccumulatorCaches;
+}
 
-enum NetSize : int;
+std::string trace(Position& pos, const Eval::NNUE::Networks& networks);
-
-using EvalFiles = std::unordered_map<Eval::NNUE::NetSize, EvalFile>;
-
-EvalFiles load_networks(const std::string&, const OptionsMap&, EvalFiles);
-void      verify(const OptionsMap&, const EvalFiles&);
-
-}  // 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
 
 }  // namespace Stockfish

src/main.cpp

@@ -17,15 +17,13 @@
 */
 
 #include <iostream>
-#include <unordered_map>
 
 #include "bitboard.h"
-#include "evaluate.h"
 #include "misc.h"
 #include "position.h"
-#include "tune.h"
 #include "types.h"
 #include "uci.h"
+#include "tune.h"
 
 using namespace Stockfish;
 
@@ -36,11 +34,9 @@ int main(int argc, char* argv[]) {
     Bitboards::init();
     Position::init();
 
-    UCI uci(argc, argv);
+    UCIEngine uci(argc, argv);
 
-    Tune::init(uci.options);
+    Tune::init(uci.engine_options());
-
-    uci.evalFiles = Eval::NNUE::load_networks(uci.workingDirectory(), uci.options, uci.evalFiles);
 
     uci.loop();
 

src/memory.cpp

@@ -0,0 +1,237 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2024 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
+
+
+// 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 +1,216 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2024 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);
+
+// 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.cpp

@@ -18,64 +18,27 @@
 
 #include "misc.h"
 
-#ifdef _WIN32
-    #if _WIN32_WINNT < 0x0601
-        #undef _WIN32_WINNT
-        #define _WIN32_WINNT 0x0601  // Force to include needed API prototypes
-    #endif
-
-    #ifndef NOMINMAX
-        #define NOMINMAX
-    #endif
-
-    #include <windows.h>
-// The needed Windows API for processor groups could be missed from old Windows
-// versions, so instead of calling them directly (forcing the linker to resolve
-// the calls at compile time), try to load them at runtime. To do this we need
-// first to define the corresponding function pointers.
-extern "C" {
-using fun1_t = bool (*)(LOGICAL_PROCESSOR_RELATIONSHIP,
-                        PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX,
-                        PDWORD);
-using fun2_t = bool (*)(USHORT, PGROUP_AFFINITY);
-using fun3_t = bool (*)(HANDLE, CONST GROUP_AFFINITY*, PGROUP_AFFINITY);
-using fun4_t = bool (*)(USHORT, PGROUP_AFFINITY, USHORT, PUSHORT);
-using fun5_t = WORD (*)();
-using fun6_t = bool (*)(HANDLE, DWORD, PHANDLE);
-using fun7_t = bool (*)(LPCSTR, LPCSTR, PLUID);
-using fun8_t = bool (*)(HANDLE, BOOL, PTOKEN_PRIVILEGES, DWORD, PTOKEN_PRIVILEGES, PDWORD);
-}
-#endif
-
 #include <atomic>
+#include <cctype>
 #include <cmath>
 #include <cstdlib>
 #include <fstream>
 #include <iomanip>
 #include <iostream>
+#include <iterator>
+#include <limits>
 #include <mutex>
 #include <sstream>
 #include <string_view>
 
 #include "types.h"
 
-#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
-
 namespace Stockfish {
 
 namespace {
 
 // Version number or dev.
-constexpr std::string_view version = "16.1";
+constexpr std::string_view version = "17";
 
 // Our fancy logging facility. The trick here is to replace cin.rdbuf() and
 // cout.rdbuf() with two Tie objects that tie cin and cout to a file stream. We
@@ -149,14 +112,16 @@ class Logger {
 
 
 // Returns the full name of the current Stockfish version.
-// For local dev compiles we try to append the commit sha and commit date
+//
-// from git if that fails only the local compilation date is set and "nogit" is specified:
+// For local dev compiles we try to append the commit SHA and
-// Stockfish dev-YYYYMMDD-SHA
+// commit date from git. If that fails only the local compilation
-// or
+// date is set and "nogit" is specified:
-// Stockfish dev-YYYYMMDD-nogit
+//      Stockfish dev-YYYYMMDD-SHA
+//      or
+//      Stockfish dev-YYYYMMDD-nogit
 //
 // For releases (non-dev builds) we only include the version number:
-// Stockfish version
+//      Stockfish version
 std::string engine_info(bool to_uci) {
     std::stringstream ss;
     ss << "Stockfish " << version << std::setfill('0');
@@ -168,8 +133,9 @@ std::string engine_info(bool to_uci) {
         ss << stringify(GIT_DATE);
 #else
         constexpr std::string_view months("Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec");
-        std::string                month, day, year;
+
-        std::stringstream          date(__DATE__);  // From compiler, format is "Sep 21 2008"
+        std::string       month, day, year;
+        std::stringstream date(__DATE__);  // From compiler, format is "Sep 21 2008"
 
         date >> month >> day >> year;
         ss << year << std::setw(2) << std::setfill('0') << (1 + months.find(month) / 4)
@@ -318,13 +284,21 @@ template<size_t N>
 struct DebugInfo {
     std::atomic<int64_t> data[N] = {0};
 
-    constexpr inline std::atomic<int64_t>& operator[](int index) { return data[index]; }
+    constexpr std::atomic<int64_t>& operator[](int index) { return data[index]; }
 };
 
-DebugInfo<2> hit[MaxDebugSlots];
+struct DebugExtremes: public DebugInfo<3> {
-DebugInfo<2> mean[MaxDebugSlots];
+    DebugExtremes() {
-DebugInfo<3> stdev[MaxDebugSlots];
+        data[1] = std::numeric_limits<int64_t>::min();
-DebugInfo<6> correl[MaxDebugSlots];
+        data[2] = std::numeric_limits<int64_t>::max();
+    }
+};
+
+DebugInfo<2>  hit[MaxDebugSlots];
+DebugInfo<2>  mean[MaxDebugSlots];
+DebugInfo<3>  stdev[MaxDebugSlots];
+DebugInfo<6>  correl[MaxDebugSlots];
+DebugExtremes extremes[MaxDebugSlots];
 
 }  // namespace
 
@@ -348,6 +322,18 @@ void dbg_stdev_of(int64_t value, int slot) {
     stdev[slot][2] += value * value;
 }
 
+void dbg_extremes_of(int64_t value, int slot) {
+    ++extremes[slot][0];
+
+    int64_t current_max = extremes[slot][1].load();
+    while (current_max < value && !extremes[slot][1].compare_exchange_weak(current_max, value))
+    {}
+
+    int64_t current_min = extremes[slot][2].load();
+    while (current_min > value && !extremes[slot][2].compare_exchange_weak(current_min, value))
+    {}
+}
+
 void dbg_correl_of(int64_t value1, int64_t value2, int slot) {
 
     ++correl[slot][0];
@@ -382,6 +368,13 @@ void dbg_print() {
             std::cerr << "Stdev #" << i << ": Total " << n << " Stdev " << r << std::endl;
         }
 
+    for (int i = 0; i < MaxDebugSlots; ++i)
+        if ((n = extremes[i][0]))
+        {
+            std::cerr << "Extremity #" << i << ": Total " << n << " Min " << extremes[i][2]
+                      << " Max " << extremes[i][1] << std::endl;
+        }
+
     for (int i = 0; i < MaxDebugSlots; ++i)
         if ((n = correl[i][0]))
         {
@@ -408,6 +401,8 @@ std::ostream& operator<<(std::ostream& os, SyncCout sc) {
     return os;
 }
 
+void sync_cout_start() { std::cout << IO_LOCK; }
+void sync_cout_end() { std::cout << IO_UNLOCK; }
 
 // Trampoline helper to avoid moving Logger to misc.h
 void start_logger(const std::string& fname) { Logger::start(fname); }
@@ -415,14 +410,14 @@ void start_logger(const std::string& fname) { Logger::start(fname); }
 
 #ifdef NO_PREFETCH
 
-void prefetch(void*) {}
+void prefetch(const void*) {}
 
 #else
 
-void prefetch(void* addr) {
+void prefetch(const void* addr) {
 
     #if defined(_MSC_VER)
-    _mm_prefetch((char*) addr, _MM_HINT_T0);
+    _mm_prefetch((char const*) addr, _MM_HINT_T0);
     #else
     __builtin_prefetch(addr);
     #endif
@@ -430,289 +425,6 @@ void prefetch(void* addr) {
 
 #endif
 
-
-// Wrapper for systems where the c++17 implementation
-// does not guarantee the availability of aligned_alloc(). Memory allocated with
-// std_aligned_alloc() must be freed with std_aligned_free().
-void* std_aligned_alloc(size_t alignment, size_t size) {
-
-#if defined(POSIXALIGNEDALLOC)
-    void* mem;
-    return posix_memalign(&mem, alignment, size) ? nullptr : mem;
-#elif defined(_WIN32) && !defined(_M_ARM) && !defined(_M_ARM64)
-    return _mm_malloc(size, alignment);
-#elif defined(_WIN32)
-    return _aligned_malloc(size, alignment);
-#else
-    return std::aligned_alloc(alignment, size);
-#endif
-}
-
-void std_aligned_free(void* ptr) {
-
-#if defined(POSIXALIGNEDALLOC)
-    free(ptr);
-#elif defined(_WIN32) && !defined(_M_ARM) && !defined(_M_ARM64)
-    _mm_free(ptr);
-#elif defined(_WIN32)
-    _aligned_free(ptr);
-#else
-    free(ptr);
-#endif
-}
-
-// aligned_large_pages_alloc() will return suitably aligned memory, if possible using large pages.
-
-#if defined(_WIN32)
-
-static void* aligned_large_pages_alloc_windows([[maybe_unused]] size_t allocSize) {
-
-    #if !defined(_WIN64)
-    return nullptr;
-    #else
-
-    HANDLE hProcessToken{};
-    LUID   luid{};
-    void*  mem = nullptr;
-
-    const size_t largePageSize = GetLargePageMinimum();
-    if (!largePageSize)
-        return nullptr;
-
-    // Dynamically link OpenProcessToken, LookupPrivilegeValue and AdjustTokenPrivileges
-
-    HMODULE hAdvapi32 = GetModuleHandle(TEXT("advapi32.dll"));
-
-    if (!hAdvapi32)
-        hAdvapi32 = LoadLibrary(TEXT("advapi32.dll"));
-
-    auto fun6 = fun6_t((void (*)()) GetProcAddress(hAdvapi32, "OpenProcessToken"));
-    if (!fun6)
-        return nullptr;
-    auto fun7 = fun7_t((void (*)()) GetProcAddress(hAdvapi32, "LookupPrivilegeValueA"));
-    if (!fun7)
-        return nullptr;
-    auto fun8 = fun8_t((void (*)()) GetProcAddress(hAdvapi32, "AdjustTokenPrivileges"));
-    if (!fun8)
-        return nullptr;
-
-    // We need SeLockMemoryPrivilege, so try to enable it for the process
-    if (!fun6(  // OpenProcessToken()
-          GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES | TOKEN_QUERY, &hProcessToken))
-        return nullptr;
-
-    if (fun7(  // LookupPrivilegeValue(nullptr, SE_LOCK_MEMORY_NAME, &luid)
-          nullptr, "SeLockMemoryPrivilege", &luid))
-    {
-        TOKEN_PRIVILEGES tp{};
-        TOKEN_PRIVILEGES prevTp{};
-        DWORD            prevTpLen = 0;
-
-        tp.PrivilegeCount           = 1;
-        tp.Privileges[0].Luid       = luid;
-        tp.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED;
-
-        // Try to enable SeLockMemoryPrivilege. Note that even if AdjustTokenPrivileges() succeeds,
-        // we still need to query GetLastError() to ensure that the privileges were actually obtained.
-        if (fun8(  // AdjustTokenPrivileges()
-              hProcessToken, FALSE, &tp, sizeof(TOKEN_PRIVILEGES), &prevTp, &prevTpLen)
-            && GetLastError() == ERROR_SUCCESS)
-        {
-            // Round up size to full pages and allocate
-            allocSize = (allocSize + largePageSize - 1) & ~size_t(largePageSize - 1);
-            mem       = VirtualAlloc(nullptr, allocSize, MEM_RESERVE | MEM_COMMIT | MEM_LARGE_PAGES,
-                                     PAGE_READWRITE);
-
-            // Privilege no longer needed, restore previous state
-            fun8(  // AdjustTokenPrivileges ()
-              hProcessToken, FALSE, &prevTp, 0, nullptr, nullptr);
-        }
-    }
-
-    CloseHandle(hProcessToken);
-
-    return mem;
-
-    #endif
-}
-
-void* aligned_large_pages_alloc(size_t allocSize) {
-
-    // Try to allocate large pages
-    void* mem = aligned_large_pages_alloc_windows(allocSize);
-
-    // Fall back to regular, page-aligned, allocation if necessary
-    if (!mem)
-        mem = VirtualAlloc(nullptr, allocSize, MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE);
-
-    return mem;
-}
-
-#else
-
-void* aligned_large_pages_alloc(size_t allocSize) {
-
-    #if defined(__linux__)
-    constexpr size_t alignment = 2 * 1024 * 1024;  // assumed 2MB page size
-    #else
-    constexpr size_t alignment = 4096;  // assumed small page size
-    #endif
-
-    // Round up to multiples of alignment
-    size_t size = ((allocSize + alignment - 1) / alignment) * alignment;
-    void*  mem  = std_aligned_alloc(alignment, size);
-    #if defined(MADV_HUGEPAGE)
-    madvise(mem, size, MADV_HUGEPAGE);
-    #endif
-    return mem;
-}
-
-#endif
-
-
-// aligned_large_pages_free() will free the previously allocated ttmem
-
-#if defined(_WIN32)
-
-void aligned_large_pages_free(void* mem) {
-
-    if (mem && !VirtualFree(mem, 0, MEM_RELEASE))
-    {
-        DWORD err = GetLastError();
-        std::cerr << "Failed to free large page memory. Error code: 0x" << std::hex << err
-                  << std::dec << std::endl;
-        exit(EXIT_FAILURE);
-    }
-}
-
-#else
-
-void aligned_large_pages_free(void* mem) { std_aligned_free(mem); }
-
-#endif
-
-
-namespace WinProcGroup {
-
-#ifndef _WIN32
-
-void bindThisThread(size_t) {}
-
-#else
-
-// Retrieves logical processor information using Windows-specific
-// API and returns the best node id for the thread with index idx. Original
-// code from Texel by Peter Österlund.
-static int best_node(size_t idx) {
-
-    int   threads      = 0;
-    int   nodes        = 0;
-    int   cores        = 0;
-    DWORD returnLength = 0;
-    DWORD byteOffset   = 0;
-
-    // Early exit if the needed API is not available at runtime
-    HMODULE k32  = GetModuleHandle(TEXT("Kernel32.dll"));
-    auto    fun1 = (fun1_t) (void (*)()) GetProcAddress(k32, "GetLogicalProcessorInformationEx");
-    if (!fun1)
-        return -1;
-
-    // First call to GetLogicalProcessorInformationEx() to get returnLength.
-    // We expect the call to fail due to null buffer.
-    if (fun1(RelationAll, nullptr, &returnLength))
-        return -1;
-
-    // Once we know returnLength, allocate the buffer
-    SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *buffer, *ptr;
-    ptr = buffer = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*) malloc(returnLength);
-
-    // Second call to GetLogicalProcessorInformationEx(), now we expect to succeed
-    if (!fun1(RelationAll, buffer, &returnLength))
-    {
-        free(buffer);
-        return -1;
-    }
-
-    while (byteOffset < returnLength)
-    {
-        if (ptr->Relationship == RelationNumaNode)
-            nodes++;
-
-        else if (ptr->Relationship == RelationProcessorCore)
-        {
-            cores++;
-            threads += (ptr->Processor.Flags == LTP_PC_SMT) ? 2 : 1;
-        }
-
-        assert(ptr->Size);
-        byteOffset += ptr->Size;
-        ptr = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*) (((char*) ptr) + ptr->Size);
-    }
-
-    free(buffer);
-
-    std::vector<int> groups;
-
-    // Run as many threads as possible on the same node until the core limit is
-    // reached, then move on to filling the next node.
-    for (int n = 0; n < nodes; n++)
-        for (int i = 0; i < cores / nodes; i++)
-            groups.push_back(n);
-
-    // In case a core has more than one logical processor (we assume 2) and we
-    // still have threads to allocate, spread them evenly across available nodes.
-    for (int t = 0; t < threads - cores; t++)
-        groups.push_back(t % nodes);
-
-    // If we still have more threads than the total number of logical processors
-    // then return -1 and let the OS to decide what to do.
-    return idx < groups.size() ? groups[idx] : -1;
-}
-
-
-// Sets the group affinity of the current thread
-void bindThisThread(size_t idx) {
-
-    // Use only local variables to be thread-safe
-    int node = best_node(idx);
-
-    if (node == -1)
-        return;
-
-    // Early exit if the needed API are not available at runtime
-    HMODULE k32  = GetModuleHandle(TEXT("Kernel32.dll"));
-    auto    fun2 = fun2_t((void (*)()) GetProcAddress(k32, "GetNumaNodeProcessorMaskEx"));
-    auto    fun3 = fun3_t((void (*)()) GetProcAddress(k32, "SetThreadGroupAffinity"));
-    auto    fun4 = fun4_t((void (*)()) GetProcAddress(k32, "GetNumaNodeProcessorMask2"));
-    auto    fun5 = fun5_t((void (*)()) GetProcAddress(k32, "GetMaximumProcessorGroupCount"));
-
-    if (!fun2 || !fun3)
-        return;
-
-    if (!fun4 || !fun5)
-    {
-        GROUP_AFFINITY affinity;
-        if (fun2(node, &affinity))                         // GetNumaNodeProcessorMaskEx
-            fun3(GetCurrentThread(), &affinity, nullptr);  // SetThreadGroupAffinity
-    }
-    else
-    {
-        // If a numa node has more than one processor group, we assume they are
-        // sized equal and we spread threads evenly across the groups.
-        USHORT elements, returnedElements;
-        elements                 = fun5();  // GetMaximumProcessorGroupCount
-        GROUP_AFFINITY* affinity = (GROUP_AFFINITY*) malloc(elements * sizeof(GROUP_AFFINITY));
-        if (fun4(node, affinity, elements, &returnedElements))  // GetNumaNodeProcessorMask2
-            fun3(GetCurrentThread(), &affinity[idx % returnedElements],
-                 nullptr);  // SetThreadGroupAffinity
-        free(affinity);
-    }
-}
-
-#endif
-
-}  // namespace WinProcGroup
-
 #ifdef _WIN32
     #include <direct.h>
     #define GETCWD _getcwd
@@ -721,13 +433,30 @@ void bindThisThread(size_t idx) {
     #define GETCWD getcwd
 #endif
 
-CommandLine::CommandLine(int _argc, char** _argv) :
+size_t str_to_size_t(const std::string& s) {
-    argc(_argc),
+    unsigned long long value = std::stoull(s);
-    argv(_argv) {
+    if (value > std::numeric_limits<size_t>::max())
-    std::string pathSeparator;
+        std::exit(EXIT_FAILURE);
+    return static_cast<size_t>(value);
+}
 
-    // Extract the path+name of the executable binary
+std::optional<std::string> read_file_to_string(const std::string& path) {
-    std::string argv0 = argv[0];
+    std::ifstream f(path, std::ios_base::binary);
+    if (!f)
+        return std::nullopt;
+    return std::string(std::istreambuf_iterator<char>(f), std::istreambuf_iterator<char>());
+}
+
+void remove_whitespace(std::string& s) {
+    s.erase(std::remove_if(s.begin(), s.end(), [](char c) { return std::isspace(c); }), s.end());
+}
+
+bool is_whitespace(const std::string& s) {
+    return std::all_of(s.begin(), s.end(), [](char c) { return std::isspace(c); });
+}
+
+std::string CommandLine::get_binary_directory(std::string argv0) {
+    std::string pathSeparator;
 
 #ifdef _WIN32
     pathSeparator = "\\";
@@ -743,15 +472,11 @@ CommandLine::CommandLine(int _argc, char** _argv) :
 #endif
 
     // Extract the working directory
-    workingDirectory = "";
+    auto workingDirectory = CommandLine::get_working_directory();
-    char  buff[40000];
-    char* cwd = GETCWD(buff, 40000);
-    if (cwd)
-        workingDirectory = cwd;
 
     // Extract the binary directory path from argv0
-    binaryDirectory = argv0;
+    auto   binaryDirectory = argv0;
-    size_t pos      = binaryDirectory.find_last_of("\\/");
+    size_t pos             = binaryDirectory.find_last_of("\\/");
     if (pos == std::string::npos)
         binaryDirectory = "." + pathSeparator;
     else
@@ -760,6 +485,19 @@ CommandLine::CommandLine(int _argc, char** _argv) :
     // Pattern replacement: "./" at the start of path is replaced by the working directory
     if (binaryDirectory.find("." + pathSeparator) == 0)
         binaryDirectory.replace(0, 1, workingDirectory);
+
+    return binaryDirectory;
 }
 
+std::string CommandLine::get_working_directory() {
+    std::string workingDirectory = "";
+    char        buff[40000];
+    char*       cwd = GETCWD(buff, 40000);
+    if (cwd)
+        workingDirectory = cwd;
+
+    return workingDirectory;
+}
+
+
 }  // namespace Stockfish

src/misc.h

@@ -24,7 +24,9 @@
 #include <chrono>
 #include <cstddef>
 #include <cstdint>
+#include <cstdio>
 #include <iosfwd>
+#include <optional>
 #include <string>
 #include <vector>
 
@@ -39,19 +41,33 @@ std::string compiler_info();
 // 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(void* addr);
+void prefetch(const void* addr);
 
-void  start_logger(const std::string& fname);
+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);
-// memory aligned by page size, min alignment: 4096 bytes
+
-void* aligned_large_pages_alloc(size_t size);
+#if defined(__linux__)
-// nop if mem == nullptr
+
-void aligned_large_pages_free(void* mem);
+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();
 
@@ -63,6 +79,30 @@ inline TimePoint now() {
       .count();
 }
 
+inline std::vector<std::string> split(const std::string& s, const std::string& delimiter) {
+    std::vector<std::string> res;
+
+    if (s.empty())
+        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(const std::string& s);
 
 enum SyncCout {
     IO_LOCK,
@@ -73,19 +113,8 @@ std::ostream& operator<<(std::ostream&, SyncCout);
 #define sync_cout std::cout << IO_LOCK
 #define sync_endl std::endl << IO_UNLOCK
 
-
+void sync_cout_start();
-// Get the first aligned element of an array.
+void sync_cout_end();
-// 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));
-}
-
 
 // True if and only if the binary is compiled on a little-endian machine
 static inline const union {
@@ -169,25 +198,18 @@ inline uint64_t mul_hi64(uint64_t a, uint64_t b) {
 #endif
 }
 
-// Under Windows it is not possible for a process to run on more than one
-// logical processor group. This usually means being limited to using max 64
-// cores. To overcome this, some special platform-specific API should be
-// called to set group affinity for each thread. Original code from Texel by
-// Peter Österlund.
-namespace WinProcGroup {
-void bindThisThread(size_t idx);
-}
-
 
 struct CommandLine {
    public:
-    CommandLine(int, char**);
+    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;
-
-    std::string binaryDirectory;   // path of the executable directory
-    std::string workingDirectory;  // path of the working directory
 };
 
 namespace Utility {

src/movegen.cpp

@@ -75,17 +75,6 @@ ExtMove* generate_pawn_moves(const Position& pos, ExtMove* moveList, Bitboard ta
             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);
@@ -158,7 +147,7 @@ ExtMove* generate_pawn_moves(const Position& pos, ExtMove* moveList, Bitboard ta
 }
 
 
-template<Color Us, PieceType Pt, bool Checks>
+template<Color Us, PieceType Pt>
 ExtMove* generate_moves(const Position& pos, ExtMove* moveList, Bitboard target) {
 
     static_assert(Pt != KING && Pt != PAWN, "Unsupported piece type in generate_moves()");
@@ -170,10 +159,6 @@ ExtMove* generate_moves(const Position& pos, ExtMove* moveList, Bitboard target)
         Square   from = pop_lsb(bb);
         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++ = Move(from, pop_lsb(b));
     }
@@ -187,9 +172,8 @@ 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);
-    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()))
@@ -197,29 +181,24 @@ ExtMove* generate_all(const Position& pos, ExtMove* moveList) {
         target = Type == EVASIONS     ? between_bb(ksq, lsb(pos.checkers()))
                : Type == NON_EVASIONS ? ~pos.pieces(Us)
                : Type == CAPTURES     ? pos.pieces(~Us)
-                                      : ~pos.pieces();  // QUIETS || QUIET_CHECKS
+                                      : ~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++ = 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++ = Move::make<CASTLING>(ksq, pos.castling_rook_square(cr));
+                *moveList++ = Move::make<CASTLING>(ksq, pos.castling_rook_square(cr));
-    }
 
     return moveList;
 }
@@ -231,8 +210,6 @@ ExtMove* generate_all(const Position& pos, ExtMove* moveList) {
 // <QUIETS>       Generates all pseudo-legal non-captures and underpromotions
 // <EVASIONS>     Generates all pseudo-legal check evasions
 // <NON_EVASIONS> Generates all pseudo-legal captures and non-captures
-// <QUIET_CHECKS> Generates all pseudo-legal non-captures giving check,
-//                except castling and promotions
 //
 // Returns a pointer to the end of the move list.
 template<GenType Type>
@@ -251,7 +228,6 @@ ExtMove* generate(const Position& pos, ExtMove* moveList) {
 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*);
 
 

src/movegen.h

@@ -31,7 +31,6 @@ class Position;
 enum GenType {
     CAPTURES,
     QUIETS,
-    QUIET_CHECKS,
     EVASIONS,
     NON_EVASIONS,
     LEGAL

src/movepick.cpp

@@ -20,7 +20,6 @@
 
 #include <algorithm>
 #include <cassert>
-#include <iterator>
 #include <utility>
 
 #include "bitboard.h"
@@ -35,7 +34,6 @@ enum Stages {
     MAIN_TT,
     CAPTURE_INIT,
     GOOD_CAPTURE,
-    REFUTATION,
     QUIET_INIT,
     GOOD_QUIET,
     BAD_CAPTURE,
@@ -54,13 +52,11 @@ enum Stages {
     // generate qsearch moves
     QSEARCH_TT,
     QCAPTURE_INIT,
-    QCAPTURE,
+    QCAPTURE
-    QCHECK_INIT,
-    QCHECK
 };
 
-// Sort moves in descending order up to and including
+// Sort moves in descending order up to and including a given limit.
-// a given limit. The order of moves smaller than the limit is left unspecified.
+// 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)
@@ -78,35 +74,10 @@ void partial_insertion_sort(ExtMove* begin, ExtMove* end, int limit) {
 
 
 // Constructors of the MovePicker class. As arguments, we pass information
-// to help it return the (presumably) good moves first, to decide which
+// to decide which class of moves to emit, to help sorting the (presumably)
-// moves to return (in the quiescence search, for instance, we only want to
+// good moves first, and how important move ordering is at the current node.
-// search captures, promotions, and some checks) and how important a good
-// move ordering is at the current node.
 
-// MovePicker constructor for the main search
+// MovePicker constructor for the main search and for the quiescence search
-MovePicker::MovePicker(const Position&              p,
-                       Move                         ttm,
-                       Depth                        d,
-                       const ButterflyHistory*      mh,
-                       const CapturePieceToHistory* cph,
-                       const PieceToHistory**       ch,
-                       const PawnHistory*           ph,
-                       Move                         cm,
-                       const Move*                  killers) :
-    pos(p),
-    mainHistory(mh),
-    captureHistory(cph),
-    continuationHistory(ch),
-    pawnHistory(ph),
-    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));
-}
-
-// Constructor for quiescence search
 MovePicker::MovePicker(const Position&              p,
                        Move                         ttm,
                        Depth                        d,
@@ -121,13 +92,16 @@ MovePicker::MovePicker(const Position&              p,
     pawnHistory(ph),
     ttMove(ttm),
     depth(d) {
-    assert(d <= 0);
 
-    stage = (pos.checkers() ? EVASION_TT : QSEARCH_TT) + !(ttm && pos.pseudo_legal(ttm));
+    if (pos.checkers())
+        stage = EVASION_TT + !(ttm && pos.pseudo_legal(ttm));
+
+    else
+        stage = (depth > 0 ? MAIN_TT : QSEARCH_TT) + !(ttm && pos.pseudo_legal(ttm));
 }
 
-// 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, int th, const CapturePieceToHistory* cph) :
     pos(p),
     captureHistory(cph),
@@ -139,9 +113,9 @@ MovePicker::MovePicker(const Position& p, Move ttm, int th, const CapturePieceTo
           + !(ttm && pos.capture_stage(ttm) && pos.pseudo_legal(ttm) && pos.see_ge(ttm, threshold));
 }
 
-// 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() {
 
@@ -178,11 +152,11 @@ void MovePicker::score() {
             Square    to   = m.to_sq();
 
             // histories
-            m.value = 2 * (*mainHistory)[pos.side_to_move()][m.from_to()];
+            m.value = (*mainHistory)[pos.side_to_move()][m.from_to()];
             m.value += 2 * (*pawnHistory)[pawn_structure_index(pos)][pc][to];
             m.value += 2 * (*continuationHistory[0])[pc][to];
             m.value += (*continuationHistory[1])[pc][to];
-            m.value += (*continuationHistory[2])[pc][to] / 4;
+            m.value += (*continuationHistory[2])[pc][to] / 3;
             m.value += (*continuationHistory[3])[pc][to];
             m.value += (*continuationHistory[5])[pc][to];
 
@@ -190,20 +164,16 @@ void MovePicker::score() {
             m.value += bool(pos.check_squares(pt) & to) * 16384;
 
             // bonus for escaping from capture
-            m.value += threatenedPieces & from ? (pt == QUEEN && !(to & threatenedByRook)   ? 50000
+            m.value += threatenedPieces & from ? (pt == QUEEN && !(to & threatenedByRook)   ? 51700
-                                                  : pt == ROOK && !(to & threatenedByMinor) ? 25000
+                                                  : pt == ROOK && !(to & threatenedByMinor) ? 25600
-                                                  : !(to & threatenedByPawn)                ? 15000
+                                                  : !(to & threatenedByPawn)                ? 14450
                                                                                             : 0)
                                                : 0;
 
             // malus for putting piece en prise
-            m.value -= !(threatenedPieces & from)
+            m.value -= (pt == QUEEN  ? bool(to & threatenedByRook) * 49000
-                       ? (pt == QUEEN ? bool(to & threatenedByRook) * 50000
+                        : pt == ROOK ? bool(to & threatenedByMinor) * 24335
-                                          + bool(to & threatenedByMinor) * 10000
+                                     : bool(to & threatenedByPawn) * 14900);
-                          : pt == ROOK ? bool(to & threatenedByMinor) * 25000
-                          : pt != PAWN ? bool(to & threatenedByPawn) * 15000
-                                       : 0)
-                       : 0;
         }
 
         else  // Type == EVASIONS
@@ -219,7 +189,7 @@ void MovePicker::score() {
 }
 
 // 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>
 Move MovePicker::select(Pred filter) {
 
@@ -236,12 +206,12 @@ Move MovePicker::select(Pred filter) {
     return Move::none();
 }
 
-// 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) {
 
-    auto quiet_threshold = [](Depth d) { return -3330 * d; };
+    auto quiet_threshold = [](Depth d) { return -3560 * d; };
 
 top:
     switch (stage)
@@ -273,22 +243,6 @@ top:
             }))
             return *(cur - 1);
 
-        // Prepare the pointers to loop over the refutations array
-        cur      = std::begin(refutations);
-        endMoves = std::end(refutations);
-
-        // If the countermove is the same as a killer, skip it
-        if (refutations[0] == refutations[2] || refutations[1] == refutations[2])
-            --endMoves;
-
-        ++stage;
-        [[fallthrough]];
-
-    case REFUTATION :
-        if (select<Next>([&]() {
-                return *cur != Move::none() && !pos.capture_stage(*cur) && pos.pseudo_legal(*cur);
-            }))
-            return *(cur - 1);
         ++stage;
         [[fallthrough]];
 
@@ -306,11 +260,9 @@ top:
         [[fallthrough]];
 
     case GOOD_QUIET :
-        if (!skipQuiets && select<Next>([&]() {
+        if (!skipQuiets && select<Next>([]() { return true; }))
-                return *cur != refutations[0] && *cur != refutations[1] && *cur != refutations[2];
-            }))
         {
-            if ((cur - 1)->value > -8000 || (cur - 1)->value <= quiet_threshold(depth))
+            if ((cur - 1)->value > -7998 || (cur - 1)->value <= quiet_threshold(depth))
                 return *(cur - 1);
 
             // Remaining quiets are bad
@@ -337,9 +289,7 @@ top:
 
     case BAD_QUIET :
         if (!skipQuiets)
-            return select<Next>([&]() {
+            return select<Next>([]() { return true; });
-                return *cur != refutations[0] && *cur != refutations[1] && *cur != refutations[2];
-            });
 
         return Move::none();
 
@@ -358,24 +308,6 @@ top:
         return select<Next>([&]() { return pos.see_ge(*cur, threshold); });
 
     case QCAPTURE :
-        if (select<Next>([]() { return true; }))
-            return *(cur - 1);
-
-        // If we did not find any move and we do not try checks, we have finished
-        if (depth != DEPTH_QS_CHECKS)
-            return Move::none();
-
-        ++stage;
-        [[fallthrough]];
-
-    case QCHECK_INIT :
-        cur      = moves;
-        endMoves = generate<QUIET_CHECKS>(pos, cur);
-
-        ++stage;
-        [[fallthrough]];
-
-    case QCHECK :
         return select<Next>([]() { return true; });
     }
 

src/movepick.h

@@ -19,6 +19,7 @@
 #ifndef MOVEPICK_H_INCLUDED
 #define MOVEPICK_H_INCLUDED
 
+#include <algorithm>
 #include <array>
 #include <cassert>
 #include <cmath>
@@ -28,8 +29,8 @@
 #include <type_traits>  // IWYU pragma: keep
 
 #include "movegen.h"
-#include "types.h"
 #include "position.h"
+#include "types.h"
 
 namespace Stockfish {
 
@@ -69,10 +70,11 @@ class StatsEntry {
     operator const T&() const { return entry; }
 
     void operator<<(int bonus) {
-        assert(std::abs(bonus) <= D);  // Ensure range is [-D, D]
         static_assert(D <= std::numeric_limits<T>::max(), "D overflows T");
 
-        entry += bonus - entry * std::abs(bonus) / D;
+        // 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);
     }
@@ -116,10 +118,6 @@ enum StatsType {
 // 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>;
 
@@ -139,12 +137,12 @@ using PawnHistory = Stats<int16_t, 8192, PAWN_HISTORY_SIZE, PIECE_NB, SQUARE_NB>
 using CorrectionHistory =
   Stats<int16_t, CORRECTION_HISTORY_LIMIT, COLOR_NB, CORRECTION_HISTORY_SIZE>;
 
-// MovePicker class is used to pick one pseudo-legal move at a time from the
+// The MovePicker class is used to pick one pseudo-legal move at a time from the
-// current position. The most important method is next_move(), which returns a
+// current position. The most important method is next_move(), which emits one
-// new pseudo-legal move each time it is called, until there are no moves left,
+// new pseudo-legal move on every call, until there are no moves left, when
-// when Move::none() is returned. In order to improve the efficiency of the
+// Move::none() is returned. In order to improve the efficiency of the alpha-beta
-// alpha-beta algorithm, MovePicker attempts to return the moves which are most
+// algorithm, MovePicker attempts to return the moves which are most likely to get
-// likely to get a cut-off first.
+// a cut-off first.
 class MovePicker {
 
     enum PickType {
@@ -155,15 +153,6 @@ class MovePicker {
    public:
     MovePicker(const MovePicker&)            = delete;
     MovePicker& operator=(const MovePicker&) = delete;
-    MovePicker(const Position&,
-               Move,
-               Depth,
-               const ButterflyHistory*,
-               const CapturePieceToHistory*,
-               const PieceToHistory**,
-               const PawnHistory*,
-               Move,
-               const Move*);
     MovePicker(const Position&,
                Move,
                Depth,
@@ -188,11 +177,11 @@ class MovePicker {
     const PieceToHistory**       continuationHistory;
     const PawnHistory*           pawnHistory;
     Move                         ttMove;
-    ExtMove refutations[3], *cur, *endMoves, *endBadCaptures, *beginBadQuiets, *endBadQuiets;
+    ExtMove *                    cur, *endMoves, *endBadCaptures, *beginBadQuiets, *endBadQuiets;
-    int     stage;
+    int                          stage;
-    int     threshold;
+    int                          threshold;
-    Depth   depth;
+    Depth                        depth;
-    ExtMove moves[MAX_MOVES];
+    ExtMove                      moves[MAX_MOVES];
 };
 
 }  // namespace Stockfish

src/nnue/evaluate_nnue.cpp

@@ -1,482 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2024 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 "evaluate_nnue.h"
-
-#include <cmath>
-#include <cstdlib>
-#include <cstring>
-#include <fstream>
-#include <iomanip>
-#include <iostream>
-#include <optional>
-#include <sstream>
-#include <string_view>
-#include <type_traits>
-#include <unordered_map>
-
-#include "../evaluate.h"
-#include "../misc.h"
-#include "../position.h"
-#include "../types.h"
-#include "../uci.h"
-#include "nnue_accumulator.h"
-#include "nnue_common.h"
-
-namespace Stockfish::Eval::NNUE {
-
-// Input feature converter
-LargePagePtr<FeatureTransformer<TransformedFeatureDimensionsBig, &StateInfo::accumulatorBig>>
-  featureTransformerBig;
-LargePagePtr<FeatureTransformer<TransformedFeatureDimensionsSmall, &StateInfo::accumulatorSmall>>
-  featureTransformerSmall;
-
-// Evaluation function
-AlignedPtr<Network<TransformedFeatureDimensionsBig, L2Big, L3Big>>       networkBig[LayerStacks];
-AlignedPtr<Network<TransformedFeatureDimensionsSmall, L2Small, L3Small>> networkSmall[LayerStacks];
-
-// Evaluation function file names
-
-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(NetSize netSize) {
-
-    if (netSize == Small)
-    {
-        Detail::initialize(featureTransformerSmall);
-        for (std::size_t i = 0; i < LayerStacks; ++i)
-            Detail::initialize(networkSmall[i]);
-    }
-    else
-    {
-        Detail::initialize(featureTransformerBig);
-        for (std::size_t i = 0; i < LayerStacks; ++i)
-            Detail::initialize(networkBig[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, NetSize netSize, std::string& netDescription) {
-
-    std::uint32_t hashValue;
-    if (!read_header(stream, &hashValue, &netDescription))
-        return false;
-    if (hashValue != HashValue[netSize])
-        return false;
-    if (netSize == Big && !Detail::read_parameters(stream, *featureTransformerBig))
-        return false;
-    if (netSize == Small && !Detail::read_parameters(stream, *featureTransformerSmall))
-        return false;
-    for (std::size_t i = 0; i < LayerStacks; ++i)
-    {
-        if (netSize == Big && !Detail::read_parameters(stream, *(networkBig[i])))
-            return false;
-        if (netSize == Small && !Detail::read_parameters(stream, *(networkSmall[i])))
-            return false;
-    }
-    return stream && stream.peek() == std::ios::traits_type::eof();
-}
-
-// Write network parameters
-static bool
-write_parameters(std::ostream& stream, NetSize netSize, const std::string& netDescription) {
-
-    if (!write_header(stream, HashValue[netSize], netDescription))
-        return false;
-    if (netSize == Big && !Detail::write_parameters(stream, *featureTransformerBig))
-        return false;
-    if (netSize == Small && !Detail::write_parameters(stream, *featureTransformerSmall))
-        return false;
-    for (std::size_t i = 0; i < LayerStacks; ++i)
-    {
-        if (netSize == Big && !Detail::write_parameters(stream, *(networkBig[i])))
-            return false;
-        if (netSize == Small && !Detail::write_parameters(stream, *(networkSmall[i])))
-            return false;
-    }
-    return bool(stream);
-}
-
-void hint_common_parent_position(const Position& pos) {
-
-    int simpleEval = simple_eval(pos, pos.side_to_move());
-    if (std::abs(simpleEval) > 1050)
-        featureTransformerSmall->hint_common_access(pos);
-    else
-        featureTransformerBig->hint_common_access(pos);
-}
-
-// Evaluation function. Perform differential calculation.
-template<NetSize Net_Size>
-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 < Net_Size == Small ? TransformedFeatureDimensionsSmall
-                                              : TransformedFeatureDimensionsBig,
-       nullptr > ::BufferSize + alignment / sizeof(TransformedFeatureType)];
-
-    auto* transformedFeatures = align_ptr_up<alignment>(&transformedFeaturesUnaligned[0]);
-#else
-
-    alignas(alignment) TransformedFeatureType
-      transformedFeatures[FeatureTransformer < Net_Size == Small ? TransformedFeatureDimensionsSmall
-                                                                 : TransformedFeatureDimensionsBig,
-                          nullptr > ::BufferSize];
-#endif
-
-    ASSERT_ALIGNED(transformedFeatures, alignment);
-
-    const int  bucket     = (pos.count<ALL_PIECES>() - 1) / 4;
-    const auto psqt       = Net_Size == Small
-                            ? featureTransformerSmall->transform(pos, transformedFeatures, bucket)
-                            : featureTransformerBig->transform(pos, transformedFeatures, bucket);
-    const auto positional = Net_Size == Small ? networkSmall[bucket]->propagate(transformedFeatures)
-                                              : networkBig[bucket]->propagate(transformedFeatures);
-
-    if (complexity)
-        *complexity = std::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);
-}
-
-template Value evaluate<Big>(const Position& pos, bool adjusted, int* complexity);
-template Value evaluate<Small>(const Position& pos, bool adjusted, int* complexity);
-
-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<TransformedFeatureDimensionsBig, nullptr>::BufferSize
-       + alignment / sizeof(TransformedFeatureType)];
-
-    auto* transformedFeatures = align_ptr_up<alignment>(&transformedFeaturesUnaligned[0]);
-#else
-    alignas(alignment) TransformedFeatureType
-      transformedFeatures[FeatureTransformer<TransformedFeatureDimensionsBig, 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 = featureTransformerBig->transform(pos, transformedFeatures, bucket);
-        const auto positional  = networkBig[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");
-
-
-// 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(UCI::to_cp(v));
-    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
-static void format_cp_aligned_dot(Value v, std::stringstream& stream) {
-
-    const double pawns = std::abs(0.01 * UCI::to_cp(v));
-
-    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) {
-
-    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<NNUE::Big>(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->accumulatorBig.computed[WHITE] = false;
-                st->accumulatorBig.computed[BLACK] = false;
-
-                Value eval = evaluate<NNUE::Big>(pos);
-                eval       = pos.side_to_move() == WHITE ? eval : -eval;
-                v          = base - eval;
-
-                pos.put_piece(pc, sq);
-                st->accumulatorBig.computed[WHITE] = false;
-                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 = 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
-std::optional<std::string> load_eval(std::istream& stream, NetSize netSize) {
-
-    initialize(netSize);
-    std::string netDescription;
-    return read_parameters(stream, netSize, netDescription) ? std::make_optional(netDescription)
-                                                            : std::nullopt;
-}
-
-// Save eval, to a file stream or a memory stream
-bool save_eval(std::ostream&      stream,
-               NetSize            netSize,
-               const std::string& name,
-               const std::string& netDescription) {
-
-    if (name.empty() || name == "None")
-        return false;
-
-    return write_parameters(stream, netSize, netDescription);
-}
-
-// Save eval, to a file given by its name
-bool save_eval(const std::optional<std::string>&                              filename,
-               NetSize                                                        netSize,
-               const std::unordered_map<Eval::NNUE::NetSize, Eval::EvalFile>& evalFiles) {
-
-    std::string actualFilename;
-    std::string msg;
-
-    if (filename.has_value())
-        actualFilename = filename.value();
-    else
-    {
-        if (evalFiles.at(netSize).current
-            != (netSize == Small ? EvalFileDefaultNameSmall : EvalFileDefaultNameBig))
-        {
-            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 = (netSize == Small ? EvalFileDefaultNameSmall : EvalFileDefaultNameBig);
-    }
-
-    std::ofstream stream(actualFilename, std::ios_base::binary);
-    bool          saved = save_eval(stream, netSize, evalFiles.at(netSize).current,
-                                    evalFiles.at(netSize).netDescription);
-
-    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

@@ -1,93 +0,0 @@
-/*
-  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
-  Copyright (C) 2004-2024 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 <cstdint>
-#include <iosfwd>
-#include <memory>
-#include <optional>
-#include <string>
-#include <unordered_map>
-
-#include "../misc.h"
-#include "../types.h"
-#include "nnue_architecture.h"
-#include "nnue_feature_transformer.h"
-
-namespace Stockfish {
-class Position;
-
-namespace Eval {
-struct EvalFile;
-}
-
-}
-
-namespace Stockfish::Eval::NNUE {
-
-// Hash value of evaluation function structure
-constexpr std::uint32_t HashValue[2] = {
-  FeatureTransformer<TransformedFeatureDimensionsBig, nullptr>::get_hash_value()
-    ^ Network<TransformedFeatureDimensionsBig, L2Big, L3Big>::get_hash_value(),
-  FeatureTransformer<TransformedFeatureDimensionsSmall, nullptr>::get_hash_value()
-    ^ Network<TransformedFeatureDimensionsSmall, L2Small, L3Small>::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);
-template<NetSize Net_Size>
-Value evaluate(const Position& pos, bool adjusted = false, int* complexity = nullptr);
-void  hint_common_parent_position(const Position& pos);
-
-std::optional<std::string> load_eval(std::istream& stream, NetSize netSize);
-bool                       save_eval(std::ostream&      stream,
-                                     NetSize            netSize,
-                                     const std::string& name,
-                                     const std::string& netDescription);
-bool                       save_eval(const std::optional<std::string>& filename,
-                                     NetSize                           netSize,
-                                     const std::unordered_map<Eval::NNUE::NetSize, Eval::EvalFile>&);
-
-}  // namespace Stockfish::Eval::NNUE
-
-#endif  // #ifndef NNUE_EVALUATE_NNUE_H_INCLUDED

src/nnue/features/half_ka_v2_hm.cpp

@@ -23,7 +23,7 @@
 #include "../../bitboard.h"
 #include "../../position.h"
 #include "../../types.h"
-#include "../nnue_common.h"
+#include "../nnue_accumulator.h"
 
 namespace Stockfish::Eval::NNUE::Features {
 
@@ -49,6 +49,8 @@ void HalfKAv2_hm::append_active_indices(const Position& pos, IndexList& active)
 // Explicit template instantiations
 template void HalfKAv2_hm::append_active_indices<WHITE>(const Position& pos, IndexList& active);
 template void HalfKAv2_hm::append_active_indices<BLACK>(const Position& pos, IndexList& active);
+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);
 
 // Get a list of indices for recently changed features
 template<Color Perspective>

src/nnue/features/half_ka_v2_hm.h

@@ -63,10 +63,6 @@ class HalfKAv2_hm {
       {PS_NONE, PS_B_PAWN, PS_B_KNIGHT, PS_B_BISHOP, PS_B_ROOK, PS_B_QUEEN, PS_KING, PS_NONE,
        PS_NONE, PS_W_PAWN, PS_W_KNIGHT, PS_W_BISHOP, PS_W_ROOK, PS_W_QUEEN, PS_KING, PS_NONE}};
 
-    // Index of a feature for a given king position and another piece on some square
-    template<Color Perspective>
-    static IndexType make_index(Square s, Piece pc, Square ksq);
-
    public:
     // Feature name
     static constexpr const char* Name = "HalfKAv2_hm(Friend)";
@@ -126,6 +122,10 @@ class HalfKAv2_hm {
     static constexpr IndexType MaxActiveDimensions = 32;
     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(const Position& pos, IndexList& active);

src/nnue/layers/affine_transform.h

@@ -39,25 +39,26 @@
 
 namespace Stockfish::Eval::NNUE::Layers {
 
+#if defined(USE_SSSE3) || defined(USE_NEON_DOTPROD)
+    #define ENABLE_SEQ_OPT
+#endif
+
 // Fallback implementation for older/other architectures.
 // 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) {
-    #if defined(USE_SSE2) || defined(USE_NEON_DOTPROD) || defined(USE_NEON)
+    #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;
     const __m128i       Zeros       = _mm_setzero_si128();
     const auto          inputVector = reinterpret_cast<const __m128i*>(input);
 
-        #elif defined(USE_NEON_DOTPROD)
-    constexpr IndexType NumChunks   = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
-    const auto          inputVector = reinterpret_cast<const int8x16_t*>(input);
-
         #elif defined(USE_NEON)
     constexpr IndexType NumChunks   = ceil_to_multiple<IndexType>(InputDimensions, 16) / 16;
     const auto          inputVector = reinterpret_cast<const int8x8_t*>(input);
@@ -91,16 +92,8 @@ static void affine_transform_non_ssse3(std::int32_t*       output,
         sum                   = _mm_add_epi32(sum, sum_second_32);
         output[i]             = _mm_cvtsi128_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)
@@ -127,7 +120,8 @@ static void affine_transform_non_ssse3(std::int32_t*       output,
         }
     #endif
 }
-#endif
+
+#endif  // !ENABLE_SEQ_OPT
 
 template<IndexType InDims, IndexType OutDims>
 class AffineTransform {
@@ -162,7 +156,7 @@ class AffineTransform {
     }
 
     static constexpr IndexType get_weight_index(IndexType i) {
-#if defined(USE_SSSE3)
+#ifdef ENABLE_SEQ_OPT
         return get_weight_index_scrambled(i);
 #else
         return i;
@@ -190,29 +184,28 @@ class AffineTransform {
     // Forward propagation
     void propagate(const InputType* input, OutputType* output) const {
 
-#if defined(USE_SSSE3)
+#ifdef ENABLE_SEQ_OPT
 
         if constexpr (OutputDimensions > 1)
         {
-
     #if defined(USE_AVX512)
             using vec_t = __m512i;
-        #define vec_setzero _mm512_setzero_si512
         #define vec_set_32 _mm512_set1_epi32
         #define vec_add_dpbusd_32 Simd::m512_add_dpbusd_epi32
-        #define vec_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_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_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);
@@ -242,28 +235,33 @@ class AffineTransform {
             for (IndexType k = 0; k < NumRegs; ++k)
                 outptr[k] = acc[k];
 
-    #undef vec_setzero
     #undef vec_set_32
     #undef vec_add_dpbusd_32
-    #undef vec_hadd
         }
         else if constexpr (OutputDimensions == 1)
         {
-
     // We cannot use AVX512 for the last layer because there are only 32 inputs
     // and the buffer is not padded to 64 elements.
     #if defined(USE_AVX2)
             using vec_t = __m256i;
-        #define vec_setzero _mm256_setzero_si256
+        #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_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
 
             const auto inputVector = reinterpret_cast<const vec_t*>(input);

src/nnue/layers/clipped_relu.h

@@ -65,41 +65,37 @@ class ClippedReLU {
         if constexpr (InputDimensions % SimdWidth == 0)
         {
             constexpr IndexType NumChunks = InputDimensions / SimdWidth;
-            const __m256i       Zero      = _mm256_setzero_si256();
             const __m256i       Offsets   = _mm256_set_epi32(7, 3, 6, 2, 5, 1, 4, 0);
             const auto          in        = reinterpret_cast<const __m256i*>(input);
             const auto          out       = reinterpret_cast<__m256i*>(output);
             for (IndexType i = 0; i < NumChunks; ++i)
             {
                 const __m256i words0 =
-                  _mm256_srai_epi16(_mm256_packs_epi32(_mm256_load_si256(&in[i * 4 + 0]),
+                  _mm256_srli_epi16(_mm256_packus_epi32(_mm256_load_si256(&in[i * 4 + 0]),
-                                                       _mm256_load_si256(&in[i * 4 + 1])),
+                                                        _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_srli_epi16(_mm256_packus_epi32(_mm256_load_si256(&in[i * 4 + 2]),
-                                                       _mm256_load_si256(&in[i * 4 + 3])),
+                                                        _mm256_load_si256(&in[i * 4 + 3])),
                                     WeightScaleBits);
-                _mm256_store_si256(
+                _mm256_store_si256(&out[i], _mm256_permutevar8x32_epi32(
-                  &out[i], _mm256_permutevar8x32_epi32(
+                                              _mm256_packs_epi16(words0, words1), Offsets));
-                             _mm256_max_epi8(_mm256_packs_epi16(words0, words1), Zero), Offsets));
             }
         }
         else
         {
             constexpr IndexType NumChunks = InputDimensions / (SimdWidth / 2);
-            const __m128i       Zero      = _mm_setzero_si128();
             const auto          in        = reinterpret_cast<const __m128i*>(input);
             const auto          out       = reinterpret_cast<__m128i*>(output);
             for (IndexType i = 0; i < NumChunks; ++i)
             {
-                const __m128i words0 = _mm_srai_epi16(
+                const __m128i words0 = _mm_srli_epi16(
-                  _mm_packs_epi32(_mm_load_si128(&in[i * 4 + 0]), _mm_load_si128(&in[i * 4 + 1])),
+                  _mm_packus_epi32(_mm_load_si128(&in[i * 4 + 0]), _mm_load_si128(&in[i * 4 + 1])),
                   WeightScaleBits);
-                const __m128i words1 = _mm_srai_epi16(
+                const __m128i words1 = _mm_srli_epi16(
-                  _mm_packs_epi32(_mm_load_si128(&in[i * 4 + 2]), _mm_load_si128(&in[i * 4 + 3])),
+                  _mm_packus_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_packs_epi16(words0, words1));
-                _mm_store_si128(&out[i], _mm_max_epi8(packedbytes, Zero));
             }
         }
         constexpr IndexType Start = InputDimensions % SimdWidth == 0
@@ -109,9 +105,7 @@ class ClippedReLU {
 #elif defined(USE_SSE2)
         constexpr IndexType NumChunks = InputDimensions / SimdWidth;
 
-    #ifdef USE_SSE41
+    #ifndef USE_SSE41
-        const __m128i Zero = _mm_setzero_si128();
-    #else
         const __m128i k0x80s = _mm_set1_epi8(-128);
     #endif
 
@@ -119,6 +113,15 @@ class ClippedReLU {
         const auto out = reinterpret_cast<__m128i*>(output);
         for (IndexType i = 0; i < NumChunks; ++i)
         {
+    #if defined(USE_SSE41)
+            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));
+    #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);
@@ -126,15 +129,8 @@ class ClippedReLU {
               _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_store_si128(&out[i], _mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s));
-
-    #ifdef USE_SSE41
-                            _mm_max_epi8(packedbytes, Zero)
-    #else
-                            _mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s)
     #endif
-
-            );
         }
         constexpr IndexType Start = NumChunks * SimdWidth;
 

src/nnue/layers/simd.h

@@ -43,39 +43,6 @@ namespace Stockfish::Simd {
     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 void m512_add_dpbusd_epi32(__m512i& acc, __m512i a, __m512i b) {
 
     #if defined(USE_VNNI)

src/nnue/network.cpp

@@ -0,0 +1,455 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2024 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>
+
+#include "../evaluate.h"
+#include "../incbin/incbin.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"
+
+namespace {
+// 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
+
+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, const 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 {
+    if (evalfilePath.empty())
+        evalfilePath = evalFile.defaultName;
+
+    if (evalFile.current != evalfilePath)
+    {
+        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.";
+
+        sync_cout << "info string ERROR: " << msg1 << sync_endl;
+        sync_cout << "info string ERROR: " << msg2 << sync_endl;
+        sync_cout << "info string ERROR: " << msg3 << sync_endl;
+        sync_cout << "info string ERROR: " << msg4 << sync_endl;
+        sync_cout << "info string ERROR: " << msg5 << sync_endl;
+        exit(EXIT_FAILURE);
+    }
+
+    size_t size = sizeof(*featureTransformer) + sizeof(Arch) * LayerStacks;
+    sync_cout << "info string NNUE evaluation using " << evalfilePath << " ("
+              << size / (1024 * 1024) << "MiB, (" << featureTransformer->InputDimensions << ", "
+              << network[0].TransformedFeatureDimensions << ", " << network[0].FC_0_OUTPUTS << ", "
+              << network[0].FC_1_OUTPUTS << ", 1))" << sync_endl;
+}
+
+
+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 +1,132 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2024 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 <iostream>
+#include <optional>
+#include <string>
+#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;
+    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

@@ -28,12 +28,76 @@
 
 namespace Stockfish::Eval::NNUE {
 
+using BiasType       = std::int16_t;
+using PSQTWeightType = std::int32_t;
+using IndexType      = std::uint32_t;
+
 // Class that holds the result of affine transformation of input features
 template<IndexType Size>
 struct alignas(CacheLineSize) Accumulator {
-    std::int16_t accumulation[2][Size];
+    std::int16_t accumulation[COLOR_NB][Size];
-    std::int32_t psqtAccumulation[2][PSQTBuckets];
+    std::int32_t psqtAccumulation[COLOR_NB][PSQTBuckets];
-    bool         computed[2];
+    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);
+        }
+
+        void clear(const BiasType* biases) {
+            for (auto& entry : entries)
+                entry.clear(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

src/nnue/nnue_architecture.h

@@ -37,13 +37,8 @@ namespace Stockfish::Eval::NNUE {
 // Input features used in evaluation function
 using FeatureSet = Features::HalfKAv2_hm;
 
-enum NetSize : int {
-    Big,
-    Small
-};
-
 // Number of input feature dimensions after conversion
-constexpr IndexType TransformedFeatureDimensionsBig = 2560;
+constexpr IndexType TransformedFeatureDimensionsBig = 3072;
 constexpr int       L2Big                           = 15;
 constexpr int       L3Big                           = 32;
 
@@ -55,7 +50,7 @@ constexpr IndexType PSQTBuckets = 8;
 constexpr IndexType LayerStacks = 8;
 
 template<IndexType L1, int L2, int L3>
-struct Network {
+struct NetworkArchitecture {
     static constexpr IndexType TransformedFeatureDimensions = L1;
     static constexpr int       FC_0_OUTPUTS                 = L2;
     static constexpr int       FC_1_OUTPUTS                 = L3;

src/nnue/nnue_feature_transformer.h

@@ -55,15 +55,14 @@ using psqt_vec_t = __m256i;
     #define vec_store(a, b) _mm512_store_si512(a, b)
     #define vec_add_16(a, b) _mm512_add_epi16(a, b)
     #define vec_sub_16(a, b) _mm512_sub_epi16(a, b)
-    #define vec_mul_16(a, b) _mm512_mullo_epi16(a, b)
+    #define vec_mulhi_16(a, b) _mm512_mulhi_epi16(a, b)
     #define vec_zero() _mm512_setzero_epi32()
     #define vec_set_16(a) _mm512_set1_epi16(a)
     #define vec_max_16(a, b) _mm512_max_epi16(a, b)
     #define vec_min_16(a, b) _mm512_min_epi16(a, b)
-inline vec_t vec_msb_pack_16(vec_t a, vec_t b) {
+    #define vec_slli_16(a, b) _mm512_slli_epi16(a, b)
-    vec_t compacted = _mm512_packs_epi16(_mm512_srli_epi16(a, 7), _mm512_srli_epi16(b, 7));
+    // Inverse permuted at load time
-    return _mm512_permutexvar_epi64(_mm512_setr_epi64(0, 2, 4, 6, 1, 3, 5, 7), compacted);
+    #define vec_packus_16(a, b) _mm512_packus_epi16(a, b)
-}
     #define vec_load_psqt(a) _mm256_load_si256(a)
     #define vec_store_psqt(a, b) _mm256_store_si256(a, b)
     #define vec_add_psqt_32(a, b) _mm256_add_epi32(a, b)
@@ -79,15 +78,14 @@ using psqt_vec_t = __m256i;
     #define vec_store(a, b) _mm256_store_si256(a, b)
     #define vec_add_16(a, b) _mm256_add_epi16(a, b)
     #define vec_sub_16(a, b) _mm256_sub_epi16(a, b)
-    #define vec_mul_16(a, b) _mm256_mullo_epi16(a, b)
+    #define vec_mulhi_16(a, b) _mm256_mulhi_epi16(a, b)
     #define vec_zero() _mm256_setzero_si256()
     #define vec_set_16(a) _mm256_set1_epi16(a)
     #define vec_max_16(a, b) _mm256_max_epi16(a, b)
     #define vec_min_16(a, b) _mm256_min_epi16(a, b)
-inline vec_t vec_msb_pack_16(vec_t a, vec_t b) {
+    #define vec_slli_16(a, b) _mm256_slli_epi16(a, b)
-    vec_t compacted = _mm256_packs_epi16(_mm256_srli_epi16(a, 7), _mm256_srli_epi16(b, 7));
+    // Inverse permuted at load time
-    return _mm256_permute4x64_epi64(compacted, 0b11011000);
+    #define vec_packus_16(a, b) _mm256_packus_epi16(a, b)
-}
     #define vec_load_psqt(a) _mm256_load_si256(a)
     #define vec_store_psqt(a, b) _mm256_store_si256(a, b)
     #define vec_add_psqt_32(a, b) _mm256_add_epi32(a, b)
@@ -103,12 +101,13 @@ using psqt_vec_t = __m128i;
     #define vec_store(a, b) *(a) = (b)
     #define vec_add_16(a, b) _mm_add_epi16(a, b)
     #define vec_sub_16(a, b) _mm_sub_epi16(a, b)
-    #define vec_mul_16(a, b) _mm_mullo_epi16(a, b)
+    #define vec_mulhi_16(a, b) _mm_mulhi_epi16(a, b)
     #define vec_zero() _mm_setzero_si128()
     #define vec_set_16(a) _mm_set1_epi16(a)
     #define vec_max_16(a, b) _mm_max_epi16(a, b)
     #define vec_min_16(a, b) _mm_min_epi16(a, b)
-    #define vec_msb_pack_16(a, b) _mm_packs_epi16(_mm_srli_epi16(a, 7), _mm_srli_epi16(b, 7))
+    #define vec_slli_16(a, b) _mm_slli_epi16(a, b)
+    #define vec_packus_16(a, b) _mm_packus_epi16(a, b)
     #define vec_load_psqt(a) (*(a))
     #define vec_store_psqt(a, b) *(a) = (b)
     #define vec_add_psqt_32(a, b) _mm_add_epi32(a, b)
@@ -124,18 +123,14 @@ using psqt_vec_t = int32x4_t;
     #define vec_store(a, b) *(a) = (b)
     #define vec_add_16(a, b) vaddq_s16(a, b)
     #define vec_sub_16(a, b) vsubq_s16(a, b)
-    #define vec_mul_16(a, b) vmulq_s16(a, b)
+    #define vec_mulhi_16(a, b) vqdmulhq_s16(a, b)
     #define vec_zero() \
         vec_t { 0 }
     #define vec_set_16(a) vdupq_n_s16(a)
     #define vec_max_16(a, b) vmaxq_s16(a, b)
     #define vec_min_16(a, b) vminq_s16(a, b)
-inline vec_t vec_msb_pack_16(vec_t a, vec_t b) {
+    #define vec_slli_16(a, b) vshlq_s16(a, vec_set_16(b))
-    const int8x8_t  shifta    = vshrn_n_s16(a, 7);
+    #define vec_packus_16(a, b) reinterpret_cast<vec_t>(vcombine_u8(vqmovun_s16(a), vqmovun_s16(b)))
-    const int8x8_t  shiftb    = vshrn_n_s16(b, 7);
-    const int8x16_t compacted = vcombine_s8(shifta, shiftb);
-    return *reinterpret_cast<const vec_t*>(&compacted);
-}
     #define vec_load_psqt(a) (*(a))
     #define vec_store_psqt(a, b) *(a) = (b)
     #define vec_add_psqt_32(a, b) vaddq_s32(a, b)
@@ -197,10 +192,10 @@ template<IndexType                                 TransformedFeatureDimensions,
          Accumulator<TransformedFeatureDimensions> StateInfo::*accPtr>
 class FeatureTransformer {
 
-   private:
     // Number of output dimensions for one side
     static constexpr IndexType HalfDimensions = TransformedFeatureDimensions;
 
+   private:
 #ifdef VECTOR
     static constexpr int NumRegs =
       BestRegisterCount<vec_t, WeightType, TransformedFeatureDimensions, NumRegistersSIMD>();
@@ -229,6 +224,73 @@ class FeatureTransformer {
         return FeatureSet::HashValue ^ (OutputDimensions * 2);
     }
 
+    static constexpr void order_packs([[maybe_unused]] uint64_t* v) {
+#if defined(USE_AVX512)  // _mm512_packs_epi16 ordering
+        uint64_t tmp0 = v[2], tmp1 = v[3];
+        v[2] = v[8], v[3] = v[9];
+        v[8] = v[4], v[9] = v[5];
+        v[4] = tmp0, v[5] = tmp1;
+        tmp0 = v[6], tmp1 = v[7];
+        v[6] = v[10], v[7] = v[11];
+        v[10] = v[12], v[11] = v[13];
+        v[12] = tmp0, v[13] = tmp1;
+#elif defined(USE_AVX2)  // _mm256_packs_epi16 ordering
+        std::swap(v[2], v[4]);
+        std::swap(v[3], v[5]);
+#endif
+    }
+
+    static constexpr void inverse_order_packs([[maybe_unused]] uint64_t* v) {
+#if defined(USE_AVX512)  // Inverse _mm512_packs_epi16 ordering
+        uint64_t tmp0 = v[2], tmp1 = v[3];
+        v[2] = v[4], v[3] = v[5];
+        v[4] = v[8], v[5] = v[9];
+        v[8] = tmp0, v[9] = tmp1;
+        tmp0 = v[6], tmp1 = v[7];
+        v[6] = v[12], v[7] = v[13];
+        v[12] = v[10], v[13] = v[11];
+        v[10] = tmp0, v[11] = tmp1;
+#elif defined(USE_AVX2)  // Inverse _mm256_packs_epi16 ordering
+        std::swap(v[2], v[4]);
+        std::swap(v[3], v[5]);
+#endif
+    }
+
+    void permute_weights([[maybe_unused]] void (*order_fn)(uint64_t*)) const {
+#if defined(USE_AVX2)
+    #if defined(USE_AVX512)
+        constexpr IndexType di = 16;
+    #else
+        constexpr IndexType di = 8;
+    #endif
+        uint64_t* b = reinterpret_cast<uint64_t*>(const_cast<BiasType*>(&biases[0]));
+        for (IndexType i = 0; i < HalfDimensions * sizeof(BiasType) / sizeof(uint64_t); i += di)
+            order_fn(&b[i]);
+
+        for (IndexType j = 0; j < InputDimensions; ++j)
+        {
+            uint64_t* w =
+              reinterpret_cast<uint64_t*>(const_cast<WeightType*>(&weights[j * HalfDimensions]));
+            for (IndexType i = 0; i < HalfDimensions * sizeof(WeightType) / sizeof(uint64_t);
+                 i += di)
+                order_fn(&w[i]);
+        }
+#endif
+    }
+
+    inline void scale_weights(bool read) const {
+        for (IndexType j = 0; j < InputDimensions; ++j)
+        {
+            WeightType* w = const_cast<WeightType*>(&weights[j * HalfDimensions]);
+            for (IndexType i = 0; i < HalfDimensions; ++i)
+                w[i] = read ? w[i] * 2 : w[i] / 2;
+        }
+
+        BiasType* b = const_cast<BiasType*>(biases);
+        for (IndexType i = 0; i < HalfDimensions; ++i)
+            b[i] = read ? b[i] * 2 : b[i] / 2;
+    }
+
     // Read network parameters
     bool read_parameters(std::istream& stream) {
 
@@ -236,32 +298,41 @@ class FeatureTransformer {
         read_leb_128<WeightType>(stream, weights, HalfDimensions * InputDimensions);
         read_leb_128<PSQTWeightType>(stream, psqtWeights, PSQTBuckets * InputDimensions);
 
+        permute_weights(inverse_order_packs);
+        scale_weights(true);
         return !stream.fail();
     }
 
     // Write network parameters
     bool write_parameters(std::ostream& stream) const {
 
+        permute_weights(order_packs);
+        scale_weights(false);
+
         write_leb_128<BiasType>(stream, biases, HalfDimensions);
         write_leb_128<WeightType>(stream, weights, HalfDimensions * InputDimensions);
         write_leb_128<PSQTWeightType>(stream, psqtWeights, PSQTBuckets * InputDimensions);
 
+        permute_weights(inverse_order_packs);
+        scale_weights(true);
         return !stream.fail();
     }
 
     // Convert input features
-    std::int32_t transform(const Position& pos, OutputType* output, int bucket) const {
+    std::int32_t transform(const Position&                           pos,
-        update_accumulator<WHITE>(pos);
+                           AccumulatorCaches::Cache<HalfDimensions>* cache,
-        update_accumulator<BLACK>(pos);
+                           OutputType*                               output,
+                           int                                       bucket) const {
+        update_accumulator<WHITE>(pos, cache);
+        update_accumulator<BLACK>(pos, cache);
 
         const Color perspectives[2]  = {pos.side_to_move(), ~pos.side_to_move()};
-        const auto& accumulation     = (pos.state()->*accPtr).accumulation;
         const auto& psqtAccumulation = (pos.state()->*accPtr).psqtAccumulation;
-
+        const auto  psqt =
-        const auto psqt =
           (psqtAccumulation[perspectives[0]][bucket] - psqtAccumulation[perspectives[1]][bucket])
           / 2;
 
+        const auto& accumulation = (pos.state()->*accPtr).accumulation;
 
         for (IndexType p = 0; p < 2; ++p)
         {
@@ -273,25 +344,87 @@ class FeatureTransformer {
             static_assert((HalfDimensions / 2) % OutputChunkSize == 0);
             constexpr IndexType NumOutputChunks = HalfDimensions / 2 / OutputChunkSize;
 
-            vec_t Zero = vec_zero();
+            const vec_t Zero = vec_zero();
-            vec_t One  = vec_set_16(127);
+            const vec_t One  = vec_set_16(127 * 2);
 
             const vec_t* in0 = reinterpret_cast<const vec_t*>(&(accumulation[perspectives[p]][0]));
             const vec_t* in1 =
               reinterpret_cast<const vec_t*>(&(accumulation[perspectives[p]][HalfDimensions / 2]));
             vec_t* out = reinterpret_cast<vec_t*>(output + offset);
 
+            // Per the NNUE architecture, here we want to multiply pairs of
+            // clipped elements and divide the product by 128. To do this,
+            // we can naively perform min/max operation to clip each of the
+            // four int16 vectors, mullo pairs together, then pack them into
+            // one int8 vector. However, there exists a faster way.
+
+            // The idea here is to use the implicit clipping from packus to
+            // save us two vec_max_16 instructions. This clipping works due
+            // to the fact that any int16 integer below zero will be zeroed
+            // on packus.
+
+            // Consider the case where the second element is negative.
+            // If we do standard clipping, that element will be zero, which
+            // means our pairwise product is zero. If we perform packus and
+            // remove the lower-side clip for the second element, then our
+            // product before packus will be negative, and is zeroed on pack.
+            // The two operation produce equivalent results, but the second
+            // one (using packus) saves one max operation per pair.
+
+            // But here we run into a problem: mullo does not preserve the
+            // sign of the multiplication. We can get around this by doing
+            // mulhi, which keeps the sign. But that requires an additional
+            // tweak.
+
+            // mulhi cuts off the last 16 bits of the resulting product,
+            // which is the same as performing a rightward shift of 16 bits.
+            // We can use this to our advantage. Recall that we want to
+            // divide the final product by 128, which is equivalent to a
+            // 7-bit right shift. Intuitively, if we shift the clipped
+            // value left by 9, and perform mulhi, which shifts the product
+            // right by 16 bits, then we will net a right shift of 7 bits.
+            // However, this won't work as intended. Since we clip the
+            // values to have a maximum value of 127, shifting it by 9 bits
+            // might occupy the signed bit, resulting in some positive
+            // values being interpreted as negative after the shift.
+
+            // There is a way, however, to get around this limitation. When
+            // loading the network, scale accumulator weights and biases by
+            // 2. To get the same pairwise multiplication result as before,
+            // we need to divide the product by 128 * 2 * 2 = 512, which
+            // amounts to a right shift of 9 bits. So now we only have to
+            // shift left by 7 bits, perform mulhi (shifts right by 16 bits)
+            // and net a 9 bit right shift. Since we scaled everything by
+            // two, the values are clipped at 127 * 2 = 254, which occupies
+            // 8 bits. Shifting it by 7 bits left will no longer occupy the
+            // signed bit, so we are safe.
+
+            // Note that on NEON processors, we shift left by 6 instead
+            // because the instruction "vqdmulhq_s16" also doubles the
+            // return value after the multiplication, adding an extra shift
+            // to the left by 1, so we compensate by shifting less before
+            // the multiplication.
+
+            constexpr int shift =
+    #if defined(USE_SSE2)
+              7;
+    #else
+              6;
+    #endif
+
             for (IndexType j = 0; j < NumOutputChunks; ++j)
             {
-                const vec_t sum0a = vec_max_16(vec_min_16(in0[j * 2 + 0], One), Zero);
+                const vec_t sum0a =
-                const vec_t sum0b = vec_max_16(vec_min_16(in0[j * 2 + 1], One), Zero);
+                  vec_slli_16(vec_max_16(vec_min_16(in0[j * 2 + 0], One), Zero), shift);
-                const vec_t sum1a = vec_max_16(vec_min_16(in1[j * 2 + 0], One), Zero);
+                const vec_t sum0b =
-                const vec_t sum1b = vec_max_16(vec_min_16(in1[j * 2 + 1], One), Zero);
+                  vec_slli_16(vec_max_16(vec_min_16(in0[j * 2 + 1], One), Zero), shift);
+                const vec_t sum1a = vec_min_16(in1[j * 2 + 0], One);
+                const vec_t sum1b = vec_min_16(in1[j * 2 + 1], One);
 
-                const vec_t pa = vec_mul_16(sum0a, sum1a);
+                const vec_t pa = vec_mulhi_16(sum0a, sum1a);
-                const vec_t pb = vec_mul_16(sum0b, sum1b);
+                const vec_t pb = vec_mulhi_16(sum0b, sum1b);
 
-                out[j] = vec_msb_pack_16(pa, pb);
+                out[j] = vec_packus_16(pa, pb);
             }
 
 #else
@@ -301,9 +434,9 @@ class FeatureTransformer {
                 BiasType sum0 = accumulation[static_cast<int>(perspectives[p])][j + 0];
                 BiasType sum1 =
                   accumulation[static_cast<int>(perspectives[p])][j + HalfDimensions / 2];
-                sum0               = std::clamp<BiasType>(sum0, 0, 127);
+                sum0               = std::clamp<BiasType>(sum0, 0, 127 * 2);
-                sum1               = std::clamp<BiasType>(sum1, 0, 127);
+                sum1               = std::clamp<BiasType>(sum1, 0, 127 * 2);
-                output[offset + j] = static_cast<OutputType>(unsigned(sum0 * sum1) / 128);
+                output[offset + j] = static_cast<OutputType>(unsigned(sum0 * sum1) / 512);
             }
 
 #endif
@@ -312,9 +445,10 @@ class FeatureTransformer {
         return psqt;
     }  // end of function transform()
 
-    void hint_common_access(const Position& pos) const {
+    void hint_common_access(const Position&                           pos,
-        hint_common_access_for_perspective<WHITE>(pos);
+                            AccumulatorCaches::Cache<HalfDimensions>* cache) const {
-        hint_common_access_for_perspective<BLACK>(pos);
+        hint_common_access_for_perspective<WHITE>(pos, cache);
+        hint_common_access_for_perspective<BLACK>(pos, cache);
     }
 
    private:
@@ -338,31 +472,33 @@ class FeatureTransformer {
         return {st, next};
     }
 
-    // NOTE: The parameter states_to_update is an array of position states, ending with nullptr.
+    // NOTE: The parameter states_to_update is an array of position states.
-    //       All states must be sequential, that is states_to_update[i] must either be reachable
+    //       All states must be sequential, that is states_to_update[i] must
-    //       by repeatedly applying ->previous from states_to_update[i+1] or
+    //       either be reachable by repeatedly applying ->previous from
-    //       states_to_update[i] == nullptr.
+    //       states_to_update[i+1], and computed_st must be reachable by
-    //       computed_st must be reachable by repeatedly applying ->previous on
+    //       repeatedly applying ->previous on states_to_update[0].
-    //       states_to_update[0], if not nullptr.
     template<Color Perspective, size_t N>
     void update_accumulator_incremental(const Position& pos,
                                         StateInfo*      computed_st,
                                         StateInfo*      states_to_update[N]) const {
         static_assert(N > 0);
-        assert(states_to_update[N - 1] == nullptr);
+        assert([&]() {
+            for (size_t i = 0; i < N; ++i)
+            {
+                if (states_to_update[i] == nullptr)
+                    return false;
+            }
+            return true;
+        }());
 
 #ifdef VECTOR
         // Gcc-10.2 unnecessarily spills AVX2 registers if this array
-        // is defined in the VECTOR code below, once in each branch
+        // is defined in the VECTOR code below, once in each branch.
         vec_t      acc[NumRegs];
         psqt_vec_t psqt[NumPsqtRegs];
 #endif
 
-        if (states_to_update[0] == nullptr)
-            return;
-
         // Update incrementally going back through states_to_update.
-
         // Gather all features to be updated.
         const Square ksq = pos.square<KING>(Perspective);
 
@@ -370,36 +506,26 @@ class FeatureTransformer {
         // That might depend on the feature set and generally relies on the
         // feature set's update cost calculation to be correct and never allow
         // updates with more added/removed features than MaxActiveDimensions.
-        FeatureSet::IndexList removed[N - 1], added[N - 1];
+        FeatureSet::IndexList removed[N], added[N];
 
+        for (int i = N - 1; i >= 0; --i)
         {
-            int i =
+            (states_to_update[i]->*accPtr).computed[Perspective] = true;
-              N
-              - 2;  // Last potential state to update. Skip last element because it must be nullptr.
-            while (states_to_update[i] == nullptr)
-                --i;
 
-            StateInfo* st2 = states_to_update[i];
+            const StateInfo* end_state = i == 0 ? computed_st : states_to_update[i - 1];
 
-            for (; i >= 0; --i)
+            for (StateInfo* st2 = states_to_update[i]; st2 != end_state; st2 = st2->previous)
-            {
+                FeatureSet::append_changed_indices<Perspective>(ksq, st2->dirtyPiece, removed[i],
-                (states_to_update[i]->*accPtr).computed[Perspective] = true;
+                                                                added[i]);
-
-                const StateInfo* end_state = i == 0 ? computed_st : states_to_update[i - 1];
-
-                for (; st2 != end_state; st2 = st2->previous)
-                    FeatureSet::append_changed_indices<Perspective>(ksq, st2->dirtyPiece,
-                                                                    removed[i], added[i]);
-            }
         }
 
         StateInfo* st = computed_st;
 
-        // Now update the accumulators listed in states_to_update[], where the last element is a sentinel.
+        // Now update the accumulators listed in states_to_update[],
+        // where the last element is a sentinel.
 #ifdef VECTOR
 
-        if (states_to_update[1] == nullptr && (removed[0].size() == 1 || removed[0].size() == 2)
+        if (N == 1 && (removed[0].size() == 1 || removed[0].size() == 2) && added[0].size() == 1)
-            && added[0].size() == 1)
         {
             assert(states_to_update[0]);
 
@@ -469,7 +595,7 @@ class FeatureTransformer {
                 for (IndexType k = 0; k < NumRegs; ++k)
                     acc[k] = vec_load(&accTileIn[k]);
 
-                for (IndexType i = 0; states_to_update[i]; ++i)
+                for (IndexType i = 0; i < N; ++i)
                 {
                     // Difference calculation for the deactivated features
                     for (const auto index : removed[i])
@@ -505,7 +631,7 @@ class FeatureTransformer {
                 for (std::size_t k = 0; k < NumPsqtRegs; ++k)
                     psqt[k] = vec_load_psqt(&accTilePsqtIn[k]);
 
-                for (IndexType i = 0; states_to_update[i]; ++i)
+                for (IndexType i = 0; i < N; ++i)
                 {
                     // Difference calculation for the deactivated features
                     for (const auto index : removed[i])
@@ -535,7 +661,7 @@ class FeatureTransformer {
             }
         }
 #else
-        for (IndexType i = 0; states_to_update[i]; ++i)
+        for (IndexType i = 0; i < N; ++i)
         {
             std::memcpy((states_to_update[i]->*accPtr).accumulation[Perspective],
                         (st->*accPtr).accumulation[Perspective], HalfDimensions * sizeof(BiasType));
@@ -550,7 +676,6 @@ class FeatureTransformer {
             for (const auto index : removed[i])
             {
                 const IndexType offset = HalfDimensions * index;
-
                 for (IndexType j = 0; j < HalfDimensions; ++j)
                     (st->*accPtr).accumulation[Perspective][j] -= weights[offset + j];
 
@@ -563,7 +688,6 @@ class FeatureTransformer {
             for (const auto index : added[i])
             {
                 const IndexType offset = HalfDimensions * index;
-
                 for (IndexType j = 0; j < HalfDimensions; ++j)
                     (st->*accPtr).accumulation[Perspective][j] += weights[offset + j];
 
@@ -576,31 +700,78 @@ class FeatureTransformer {
     }
 
     template<Color Perspective>
-    void update_accumulator_refresh(const Position& pos) const {
+    void update_accumulator_refresh_cache(const Position&                           pos,
-#ifdef VECTOR
+                                          AccumulatorCaches::Cache<HalfDimensions>* cache) const {
-        // Gcc-10.2 unnecessarily spills AVX2 registers if this array
+        assert(cache != nullptr);
-        // is defined in the VECTOR code below, once in each branch
+
-        vec_t      acc[NumRegs];
+        Square                ksq   = pos.square<KING>(Perspective);
-        psqt_vec_t psqt[NumPsqtRegs];
+        auto&                 entry = (*cache)[ksq][Perspective];
-#endif
+        FeatureSet::IndexList removed, added;
+
+        for (Color c : {WHITE, BLACK})
+        {
+            for (PieceType pt = PAWN; pt <= KING; ++pt)
+            {
+                const Piece    piece    = make_piece(c, pt);
+                const Bitboard oldBB    = entry.byColorBB[c] & entry.byTypeBB[pt];
+                const Bitboard newBB    = pos.pieces(c, pt);
+                Bitboard       toRemove = oldBB & ~newBB;
+                Bitboard       toAdd    = newBB & ~oldBB;
+
+                while (toRemove)
+                {
+                    Square sq = pop_lsb(toRemove);
+                    removed.push_back(FeatureSet::make_index<Perspective>(sq, piece, ksq));
+                }
+                while (toAdd)
+                {
+                    Square sq = pop_lsb(toAdd);
+                    added.push_back(FeatureSet::make_index<Perspective>(sq, piece, ksq));
+                }
+            }
+        }
 
-        // Refresh the accumulator
-        // Could be extracted to a separate function because it's done in 2 places,
-        // but it's unclear if compilers would correctly handle register allocation.
         auto& accumulator                 = pos.state()->*accPtr;
         accumulator.computed[Perspective] = true;
-        FeatureSet::IndexList active;
-        FeatureSet::append_active_indices<Perspective>(pos, active);
 
 #ifdef VECTOR
+        vec_t      acc[NumRegs];
+        psqt_vec_t psqt[NumPsqtRegs];
+
         for (IndexType j = 0; j < HalfDimensions / TileHeight; ++j)
         {
-            auto biasesTile = reinterpret_cast<const vec_t*>(&biases[j * TileHeight]);
+            auto accTile =
-            for (IndexType k = 0; k < NumRegs; ++k)
+              reinterpret_cast<vec_t*>(&accumulator.accumulation[Perspective][j * TileHeight]);
-                acc[k] = biasesTile[k];
+            auto entryTile = reinterpret_cast<vec_t*>(&entry.accumulation[j * TileHeight]);
 
-            for (const auto index : active)
+            for (IndexType k = 0; k < NumRegs; ++k)
+                acc[k] = entryTile[k];
+
+            int i = 0;
+            for (; i < int(std::min(removed.size(), added.size())); ++i)
             {
+                IndexType       indexR  = removed[i];
+                const IndexType offsetR = HalfDimensions * indexR + j * TileHeight;
+                auto            columnR = reinterpret_cast<const vec_t*>(&weights[offsetR]);
+                IndexType       indexA  = added[i];
+                const IndexType offsetA = HalfDimensions * indexA + j * TileHeight;
+                auto            columnA = reinterpret_cast<const vec_t*>(&weights[offsetA]);
+
+                for (unsigned k = 0; k < NumRegs; ++k)
+                    acc[k] = vec_add_16(acc[k], vec_sub_16(columnA[k], columnR[k]));
+            }
+            for (; i < int(removed.size()); ++i)
+            {
+                IndexType       index  = removed[i];
+                const IndexType offset = HalfDimensions * index + j * TileHeight;
+                auto            column = reinterpret_cast<const vec_t*>(&weights[offset]);
+
+                for (unsigned k = 0; k < NumRegs; ++k)
+                    acc[k] = vec_sub_16(acc[k], column[k]);
+            }
+            for (; i < int(added.size()); ++i)
+            {
+                IndexType       index  = added[i];
                 const IndexType offset = HalfDimensions * index + j * TileHeight;
                 auto            column = reinterpret_cast<const vec_t*>(&weights[offset]);
 
@@ -608,19 +779,34 @@ class FeatureTransformer {
                     acc[k] = vec_add_16(acc[k], column[k]);
             }
 
-            auto accTile =
+            for (IndexType k = 0; k < NumRegs; k++)
-              reinterpret_cast<vec_t*>(&accumulator.accumulation[Perspective][j * TileHeight]);
+                vec_store(&entryTile[k], acc[k]);
-            for (unsigned k = 0; k < NumRegs; k++)
+            for (IndexType k = 0; k < NumRegs; k++)
                 vec_store(&accTile[k], acc[k]);
         }
 
         for (IndexType j = 0; j < PSQTBuckets / PsqtTileHeight; ++j)
         {
-            for (std::size_t k = 0; k < NumPsqtRegs; ++k)
+            auto accTilePsqt = reinterpret_cast<psqt_vec_t*>(
-                psqt[k] = vec_zero_psqt();
+              &accumulator.psqtAccumulation[Perspective][j * PsqtTileHeight]);
+            auto entryTilePsqt =
+              reinterpret_cast<psqt_vec_t*>(&entry.psqtAccumulation[j * PsqtTileHeight]);
 
-            for (const auto index : active)
+            for (std::size_t k = 0; k < NumPsqtRegs; ++k)
+                psqt[k] = entryTilePsqt[k];
+
+            for (int i = 0; i < int(removed.size()); ++i)
             {
+                IndexType       index  = removed[i];
+                const IndexType offset = PSQTBuckets * index + j * PsqtTileHeight;
+                auto columnPsqt        = reinterpret_cast<const psqt_vec_t*>(&psqtWeights[offset]);
+
+                for (std::size_t k = 0; k < NumPsqtRegs; ++k)
+                    psqt[k] = vec_sub_psqt_32(psqt[k], columnPsqt[k]);
+            }
+            for (int i = 0; i < int(added.size()); ++i)
+            {
+                IndexType       index  = added[i];
                 const IndexType offset = PSQTBuckets * index + j * PsqtTileHeight;
                 auto columnPsqt        = reinterpret_cast<const psqt_vec_t*>(&psqtWeights[offset]);
 
@@ -628,35 +814,53 @@ class FeatureTransformer {
                     psqt[k] = vec_add_psqt_32(psqt[k], columnPsqt[k]);
             }
 
-            auto accTilePsqt = reinterpret_cast<psqt_vec_t*>(
+            for (std::size_t k = 0; k < NumPsqtRegs; ++k)
-              &accumulator.psqtAccumulation[Perspective][j * PsqtTileHeight]);
+                vec_store_psqt(&entryTilePsqt[k], psqt[k]);
             for (std::size_t k = 0; k < NumPsqtRegs; ++k)
                 vec_store_psqt(&accTilePsqt[k], psqt[k]);
         }
 
 #else
-        std::memcpy(accumulator.accumulation[Perspective], biases,
-                    HalfDimensions * sizeof(BiasType));
 
-        for (std::size_t k = 0; k < PSQTBuckets; ++k)
+        for (const auto index : removed)
-            accumulator.psqtAccumulation[Perspective][k] = 0;
-
-        for (const auto index : active)
         {
             const IndexType offset = HalfDimensions * index;
-
             for (IndexType j = 0; j < HalfDimensions; ++j)
-                accumulator.accumulation[Perspective][j] += weights[offset + j];
+                entry.accumulation[j] -= weights[offset + j];
 
             for (std::size_t k = 0; k < PSQTBuckets; ++k)
-                accumulator.psqtAccumulation[Perspective][k] +=
+                entry.psqtAccumulation[k] -= psqtWeights[index * PSQTBuckets + k];
-                  psqtWeights[index * PSQTBuckets + k];
         }
+        for (const auto index : added)
+        {
+            const IndexType offset = HalfDimensions * index;
+            for (IndexType j = 0; j < HalfDimensions; ++j)
+                entry.accumulation[j] += weights[offset + j];
+
+            for (std::size_t k = 0; k < PSQTBuckets; ++k)
+                entry.psqtAccumulation[k] += psqtWeights[index * PSQTBuckets + k];
+        }
+
+        // The accumulator of the refresh entry has been updated.
+        // Now copy its content to the actual accumulator we were refreshing.
+
+        std::memcpy(accumulator.accumulation[Perspective], entry.accumulation,
+                    sizeof(BiasType) * HalfDimensions);
+
+        std::memcpy(accumulator.psqtAccumulation[Perspective], entry.psqtAccumulation,
+                    sizeof(int32_t) * PSQTBuckets);
 #endif
+
+        for (Color c : {WHITE, BLACK})
+            entry.byColorBB[c] = pos.pieces(c);
+
+        for (PieceType pt = PAWN; pt <= KING; ++pt)
+            entry.byTypeBB[pt] = pos.pieces(pt);
     }
 
     template<Color Perspective>
-    void hint_common_access_for_perspective(const Position& pos) const {
+    void hint_common_access_for_perspective(const Position&                           pos,
+                                            AccumulatorCaches::Cache<HalfDimensions>* cache) const {
 
         // Works like update_accumulator, but performs less work.
         // Updates ONLY the accumulator for pos.
@@ -671,16 +875,17 @@ class FeatureTransformer {
 
         if ((oldest_st->*accPtr).computed[Perspective])
         {
-            // Only update current position accumulator to minimize work.
+            // Only update current position accumulator to minimize work
-            StateInfo* states_to_update[2] = {pos.state(), nullptr};
+            StateInfo* states_to_update[1] = {pos.state()};
-            update_accumulator_incremental<Perspective, 2>(pos, oldest_st, states_to_update);
+            update_accumulator_incremental<Perspective, 1>(pos, oldest_st, states_to_update);
         }
         else
-            update_accumulator_refresh<Perspective>(pos);
+            update_accumulator_refresh_cache<Perspective>(pos, cache);
     }
 
     template<Color Perspective>
-    void update_accumulator(const Position& pos) const {
+    void update_accumulator(const Position&                           pos,
+                            AccumulatorCaches::Cache<HalfDimensions>* cache) const {
 
         auto [oldest_st, next] = try_find_computed_accumulator<Perspective>(pos);
 
@@ -689,22 +894,31 @@ class FeatureTransformer {
             if (next == nullptr)
                 return;
 
-            // Now update the accumulators listed in states_to_update[], where the last element is a sentinel.
+            // Now update the accumulators listed in states_to_update[], where
-            // Currently we update 2 accumulators.
+            // the last element is a sentinel. Currently we update two accumulators:
             //     1. for the current position
             //     2. the next accumulator after the computed one
             // The heuristic may change in the future.
-            StateInfo* states_to_update[3] = {next, next == pos.state() ? nullptr : pos.state(),
+            if (next == pos.state())
-                                              nullptr};
+            {
+                StateInfo* states_to_update[1] = {next};
 
-            update_accumulator_incremental<Perspective, 3>(pos, oldest_st, states_to_update);
+                update_accumulator_incremental<Perspective, 1>(pos, oldest_st, states_to_update);
+            }
+            else
+            {
+                StateInfo* states_to_update[2] = {next, pos.state()};
+
+                update_accumulator_incremental<Perspective, 2>(pos, oldest_st, states_to_update);
+            }
         }
         else
-        {
+            update_accumulator_refresh_cache<Perspective>(pos, cache);
-            update_accumulator_refresh<Perspective>(pos);
-        }
     }
 
+    template<IndexType Size>
+    friend struct AccumulatorCaches::Cache;
+
     alignas(CacheLineSize) BiasType biases[HalfDimensions];
     alignas(CacheLineSize) WeightType weights[HalfDimensions * InputDimensions];
     alignas(CacheLineSize) PSQTWeightType psqtWeights[InputDimensions * PSQTBuckets];

src/nnue/nnue_misc.cpp

@@ -0,0 +1,203 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2024 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 (value != VALUE_NONE)
+            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) 2004-2024 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/perft.h

@@ -26,7 +26,7 @@
 #include "types.h"
 #include "uci.h"
 
-namespace Stockfish {
+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.
@@ -51,18 +51,17 @@ uint64_t perft(Position& pos, Depth depth) {
             pos.undo_move(m);
         }
         if (Root)
-            sync_cout << UCI::move(m, pos.is_chess960()) << ": " << cnt << sync_endl;
+            sync_cout << UCIEngine::move(m, pos.is_chess960()) << ": " << cnt << sync_endl;
     }
     return nodes;
 }
 
-inline void perft(const std::string& fen, Depth depth, bool isChess960) {
+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());
 
-    uint64_t nodes = perft<true>(p, depth);
+    return perft<true>(p, depth);
-    sync_cout << "\nNodes searched: " << nodes << "\n" << sync_endl;
 }
 }
 

src/position.cpp

@@ -78,7 +78,7 @@ std::ostream& operator<<(std::ostream& os, const Position& pos) {
        << std::setw(16) << pos.key() << std::setfill(' ') << std::dec << "\nCheckers: ";
 
     for (Bitboard b = pos.checkers(); b;)
-        os << UCI::square(pop_lsb(b)) << " ";
+        os << UCIEngine::square(pop_lsb(b)) << " ";
 
     if (int(Tablebases::MaxCardinality) >= popcount(pos.pieces()) && !pos.can_castle(ANY_CASTLING))
     {
@@ -431,8 +431,8 @@ string Position::fen() const {
     if (!can_castle(ANY_CASTLING))
         ss << '-';
 
-    ss << (ep_square() == SQ_NONE ? " - " : " " + UCI::square(ep_square()) + " ") << st->rule50
+    ss << (ep_square() == SQ_NONE ? " - " : " " + UCIEngine::square(ep_square()) + " ")
-       << " " << 1 + (gamePly - (sideToMove == BLACK)) / 2;
+       << st->rule50 << " " << 1 + (gamePly - (sideToMove == BLACK)) / 2;
 
     return ss.str();
 }
@@ -682,8 +682,9 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
     // Used by NNUE
     st->accumulatorBig.computed[WHITE]     = st->accumulatorBig.computed[BLACK] =
       st->accumulatorSmall.computed[WHITE] = st->accumulatorSmall.computed[BLACK] = false;
-    auto& dp                                                                      = st->dirtyPiece;
+
-    dp.dirty_num                                                                  = 1;
+    auto& dp     = st->dirtyPiece;
+    dp.dirty_num = 1;
 
     Color  us       = sideToMove;
     Color  them     = ~us;
@@ -740,7 +741,6 @@ void Position::do_move(Move m, StateInfo& newSt, bool givesCheck) {
         // Update board and piece lists
         remove_piece(capsq);
 
-        // Update material hash key and prefetch access to materialTable
         k ^= Zobrist::psq[captured][capsq];
         st->materialKey ^= Zobrist::psq[captured][pieceCount[captured]];
 
@@ -1156,9 +1156,9 @@ bool Position::has_repeated() const {
 }
 
 
-// Tests if the position has a move which draws by repetition,
+// Tests if the position has a move which draws by repetition.
-// or an earlier position has a move that directly reaches the current position.
+// This function accurately matches the outcome of is_draw() over all legal moves.
-bool Position::has_game_cycle(int ply) const {
+bool Position::upcoming_repetition(int ply) const {
 
     int j;
 
@@ -1169,10 +1169,16 @@ bool Position::has_game_cycle(int ply) const {
 
     Key        originalKey = st->key;
     StateInfo* stp         = st->previous;
+    Key        other       = originalKey ^ stp->key ^ Zobrist::side;
 
     for (int i = 3; i <= end; i += 2)
     {
-        stp = stp->previous->previous;
+        stp = stp->previous;
+        other ^= stp->key ^ stp->previous->key ^ Zobrist::side;
+        stp = stp->previous;
+
+        if (other != 0)
+            continue;
 
         Key moveKey = originalKey ^ stp->key;
         if ((j = H1(moveKey), cuckoo[j] == moveKey) || (j = H2(moveKey), cuckoo[j] == moveKey))
@@ -1188,12 +1194,6 @@ bool Position::has_game_cycle(int ply) const {
 
                 // For nodes before or at the root, check that the move is a
                 // repetition rather than a move to the current position.
-                // In the cuckoo table, both moves Rc1c5 and Rc5c1 are stored in
-                // the same location, so we have to select which square to check.
-                if (color_of(piece_on(empty(s1) ? s2 : s1)) != side_to_move())
-                    continue;
-
-                // For repetitions before or at the root, require one more
                 if (stp->repetition)
                     return true;
             }

src/position.h

@@ -156,7 +156,7 @@ class Position {
     int   game_ply() const;
     bool  is_chess960() const;
     bool  is_draw(int ply) const;
-    bool  has_game_cycle(int ply) const;
+    bool  upcoming_repetition(int ply) const;
     bool  has_repeated() const;
     int   rule50_count() const;
     Value non_pawn_material(Color c) const;
@@ -315,8 +315,8 @@ inline bool Position::capture(Move m) const {
 }
 
 // Returns true if a move is generated from the capture stage, having also
-// queen promotions covered, i.e. consistency with the capture stage move generation
+// 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(m.is_ok());
     return capture(m) || m.promotion_type() == QUEEN;

src/score.cpp

@@ -0,0 +1,48 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2024 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 "score.h"
+
+#include <cassert>
+#include <cmath>
+#include <cstdlib>
+
+#include "uci.h"
+
+namespace Stockfish {
+
+Score::Score(Value v, const Position& pos) {
+    assert(-VALUE_INFINITE < v && v < VALUE_INFINITE);
+
+    if (std::abs(v) < VALUE_TB_WIN_IN_MAX_PLY)
+    {
+        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};
+    }
+}
+
+}

src/score.h

@@ -0,0 +1,70 @@
+/*
+  Stockfish, a UCI chess playing engine derived from Glaurung 2.1
+  Copyright (C) 2004-2024 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

@@ -19,18 +19,25 @@
 #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 <vector>
 #include <string>
+#include <string_view>
+#include <vector>
 
 #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"
@@ -59,14 +66,12 @@ struct Stack {
     int             ply;
     Move            currentMove;
     Move            excludedMove;
-    Move            killers[2];
     Value           staticEval;
     int             statScore;
     int             moveCount;
     bool            inCheck;
     bool            ttPv;
     bool            ttHit;
-    int             multipleExtensions;
     int             cutoffCnt;
 };
 
@@ -85,6 +90,7 @@ struct RootMove {
         return m.score != score ? m.score < score : m.previousScore < previousScore;
     }
 
+    uint64_t          effort          = 0;
     Value             score           = -VALUE_INFINITE;
     Value             previousScore   = -VALUE_INFINITE;
     Value             averageScore    = -VALUE_INFINITE;
@@ -100,8 +106,7 @@ struct RootMove {
 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 {
 
     // Init explicitly due to broken value-initialization of non POD in MSVC
@@ -109,30 +114,36 @@ struct LimitsType {
         time[WHITE] = time[BLACK] = inc[WHITE] = inc[BLACK] = npmsec = movetime = TimePoint(0);
         movestogo = depth = mate = perft = infinite = 0;
         nodes                                       = 0;
+        ponderMode                                  = false;
     }
 
     bool use_time_management() const { 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;
-    uint64_t          nodes;
+    uint64_t                 nodes;
+    bool                     ponderMode;
+    Square                   capSq;
 };
 
 
 // The UCI stores the uci options, thread pool, and transposition table.
 // This struct is used to easily forward data to the Search::Worker class.
 struct SharedState {
-    SharedState(const OptionsMap&   optionsMap,
+    SharedState(const OptionsMap&                               optionsMap,
-                ThreadPool&         threadPool,
+                ThreadPool&                                     threadPool,
-                TranspositionTable& transpositionTable) :
+                TranspositionTable&                             transpositionTable,
+                const LazyNumaReplicated<Eval::NNUE::Networks>& nets) :
         options(optionsMap),
         threads(threadPool),
-        tt(transpositionTable) {}
+        tt(transpositionTable),
+        networks(nets) {}
 
-    const OptionsMap&   options;
+    const OptionsMap&                               options;
-    ThreadPool&         threads;
+    ThreadPool&                                     threads;
-    TranspositionTable& tt;
+    TranspositionTable&                             tt;
+    const LazyNumaReplicated<Eval::NNUE::Networks>& networks;
 };
 
 class Worker;
@@ -145,18 +156,87 @@ class 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;
 
-    std::string pv(const Search::Worker&     worker,
+    void pv(Search::Worker&           worker,
-                   const ThreadPool&         threads,
+            const ThreadPool&         threads,
-                   const TranspositionTable& tt,
+            const TranspositionTable& tt,
-                   Depth                     depth) const;
+            Depth                     depth);
 
     Stockfish::TimeManagement tm;
+    double                    originalTimeAdjust;
     int                       callsCnt;
     std::atomic_bool          ponder;
 
@@ -167,6 +247,8 @@ class SearchManager: public ISearchManager {
     bool                 stopOnPonderhit;
 
     size_t id;
+
+    const UpdateContext& updates;
 };
 
 class NullSearchManager: public ISearchManager {
@@ -174,25 +256,27 @@ class NullSearchManager: public ISearchManager {
     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);
+    Worker(SharedState&, std::unique_ptr<ISearchManager>, size_t, NumaReplicatedAccessToken);
 
-    // Called at instantiation to initialize Reductions tables
+    // Called at instantiation to initialize reductions tables.
-    // Reset histories, usually before a new game
+    // 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 thread_idx == 0; }
+    bool is_mainthread() const { return threadIdx == 0; }
+
+    void ensure_network_replicated();
 
     // Public because they need to be updatable by the stats
-    CounterMoveHistory    counterMoves;
     ButterflyHistory      mainHistory;
     CapturePieceToHistory captureHistory;
     ContinuationHistory   continuationHistory[2][2];
@@ -202,24 +286,24 @@ class Worker {
    private:
     void iterative_deepening();
 
-    // Main search function for both PV and non-PV nodes
+    // 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 depth = 0);
+    Value qsearch(Position& pos, Stack* ss, Value alpha, Value beta);
 
-    Depth reduction(bool i, Depth d, int mn, int delta);
+    Depth reduction(bool i, Depth d, int mn, int delta) const;
 
-    // Get a pointer to the search manager, only allowed to be called by the
+    // Pointer to the search manager, only allowed to be called by the main thread
-    // main thread.
     SearchManager* main_manager() const {
-        assert(thread_idx == 0);
+        assert(threadIdx == 0);
         return static_cast<SearchManager*>(manager.get());
     }
 
-    std::array<std::array<uint64_t, SQUARE_NB>, SQUARE_NB> effort;
+    TimePoint elapsed() const;
+    TimePoint elapsed_time() const;
 
     LimitsType limits;
 
@@ -235,7 +319,8 @@ class Worker {
     Depth     rootDepth, completedDepth;
     Value     rootDelta;
 
-    size_t thread_idx;
+    size_t                    threadIdx;
+    NumaReplicatedAccessToken numaAccessToken;
 
     // Reductions lookup table initialized at startup
     std::array<int, MAX_MOVES> reductions;  // [depth or moveNumber]
@@ -245,9 +330,13 @@ class Worker {
 
     Tablebases::Config tbConfig;
 
-    const OptionsMap&   options;
+    const OptionsMap&                               options;
-    ThreadPool&         threads;
+    ThreadPool&                                     threads;
-    TranspositionTable& tt;
+    TranspositionTable&                             tt;
+    const LazyNumaReplicated<Eval::NNUE::Networks>& networks;
+
+    // Used by NNUE
+    Eval::NNUE::AccumulatorCaches refreshTable;
 
     friend class Stockfish::ThreadPool;
     friend class SearchManager;

src/syzygy/tbprobe.cpp

@@ -66,7 +66,7 @@ namespace {
 
 constexpr int TBPIECES = 7;  // Max number of supported pieces
 constexpr int MAX_DTZ =
-  1 << 18;  // Max DTZ supported, large enough to deal with the syzygy TB limit.
+  1 << 18;  // Max DTZ supported times 2, large enough to deal with the syzygy TB limit.
 
 enum {
     BigEndian,
@@ -443,6 +443,8 @@ class TBTables {
 
     std::deque<TBTable<WDL>> wdlTable;
     std::deque<TBTable<DTZ>> dtzTable;
+    size_t                   foundDTZFiles = 0;
+    size_t                   foundWDLFiles = 0;
 
     void insert(Key key, TBTable<WDL>* wdl, TBTable<DTZ>* dtz) {
         uint32_t homeBucket = uint32_t(key) & (Size - 1);
@@ -486,9 +488,16 @@ class TBTables {
         memset(hashTable, 0, sizeof(hashTable));
         wdlTable.clear();
         dtzTable.clear();
+        foundDTZFiles = 0;
+        foundWDLFiles = 0;
     }
-    size_t size() const { return wdlTable.size(); }
+
-    void   add(const std::vector<PieceType>& pieces);
+    void info() const {
+        sync_cout << "info string Found " << foundWDLFiles << " WDL and " << foundDTZFiles
+                  << " DTZ tablebase files (up to " << MaxCardinality << "-man)." << sync_endl;
+    }
+
+    void add(const std::vector<PieceType>& pieces);
 };
 
 TBTables TBTables;
@@ -501,13 +510,22 @@ void TBTables::add(const std::vector<PieceType>& pieces) {
 
     for (PieceType pt : pieces)
         code += PieceToChar[pt];
+    code.insert(code.find('K', 1), "v");
 
-    TBFile file(code.insert(code.find('K', 1), "v") + ".rtbw");  // KRK -> KRvK
+    TBFile file_dtz(code + ".rtbz");  // KRK -> KRvK
+    if (file_dtz.is_open())
+    {
+        file_dtz.close();
+        foundDTZFiles++;
+    }
+
+    TBFile file(code + ".rtbw");  // KRK -> KRvK
 
     if (!file.is_open())  // Only WDL file is checked
         return;
 
     file.close();
+    foundWDLFiles++;
 
     MaxCardinality = std::max(int(pieces.size()), MaxCardinality);
 
@@ -1326,7 +1344,7 @@ void Tablebases::init(const std::string& paths) {
     MaxCardinality = 0;
     TBFile::Paths  = paths;
 
-    if (paths.empty() || paths == "<empty>")
+    if (paths.empty())
         return;
 
     // MapB1H1H7[] encodes a square below a1-h8 diagonal to 0..27
@@ -1466,7 +1484,7 @@ void Tablebases::init(const std::string& paths) {
         }
     }
 
-    sync_cout << "info string Found " << TBTables.size() << " tablebases" << sync_endl;
+    TBTables.info();
 }
 
 // Probe the WDL table for a particular position.
@@ -1574,7 +1592,10 @@ int Tablebases::probe_dtz(Position& pos, ProbeState* result) {
 // Use the DTZ tables to rank root moves.
 //
 // A return value false indicates that not all probes were successful.
-bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves, bool rule50) {
+bool Tablebases::root_probe(Position&          pos,
+                            Search::RootMoves& rootMoves,
+                            bool               rule50,
+                            bool               rankDTZ) {
 
     ProbeState result = OK;
     StateInfo  st;
@@ -1585,7 +1606,7 @@ bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves, bool ru
     // Check whether a position was repeated since the last zeroing move.
     bool rep = pos.has_repeated();
 
-    int dtz, bound = rule50 ? (MAX_DTZ - 100) : 1;
+    int dtz, bound = rule50 ? (MAX_DTZ / 2 - 100) : 1;
 
     // Probe and rank each move
     for (auto& m : rootMoves)
@@ -1624,8 +1645,10 @@ bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves, bool ru
 
         // Better moves are ranked higher. Certain wins are ranked equally.
         // Losing moves are ranked equally unless a 50-move draw is in sight.
-        int r    = dtz > 0 ? (dtz + cnt50 <= 99 && !rep ? MAX_DTZ : MAX_DTZ - (dtz + cnt50))
+        int r    = dtz > 0 ? (dtz + cnt50 <= 99 && !rep ? MAX_DTZ - (rankDTZ ? dtz : 0)
-                 : dtz < 0 ? (-dtz * 2 + cnt50 < 100 ? -MAX_DTZ : -MAX_DTZ + (-dtz + cnt50))
+                                                        : MAX_DTZ / 2 - (dtz + cnt50))
+                 : dtz < 0 ? (-dtz * 2 + cnt50 < 100 ? -MAX_DTZ - (rankDTZ ? dtz : 0)
+                                                     : -MAX_DTZ / 2 + (-dtz + cnt50))
                            : 0;
         m.tbRank = r;
 
@@ -1633,10 +1656,11 @@ bool Tablebases::root_probe(Position& pos, Search::RootMoves& rootMoves, bool ru
         // 1 cp to cursed wins and let it grow to 49 cp as the positions gets
         // closer to a real win.
         m.tbScore = r >= bound ? VALUE_MATE - MAX_PLY - 1
-                  : r > 0      ? Value((std::max(3, r - (MAX_DTZ - 200)) * int(PawnValue)) / 200)
+                  : r > 0  ? Value((std::max(3, r - (MAX_DTZ / 2 - 200)) * int(PawnValue)) / 200)
-                  : r == 0     ? VALUE_DRAW
+                  : r == 0 ? VALUE_DRAW
-                  : r > -bound ? Value((std::min(-3, r + (MAX_DTZ - 200)) * int(PawnValue)) / 200)
+                  : r > -bound
-                               : -VALUE_MATE + MAX_PLY + 1;
+                    ? Value((std::min(-3, r + (MAX_DTZ / 2 - 200)) * int(PawnValue)) / 200)
+                    : -VALUE_MATE + MAX_PLY + 1;
     }
 
     return true;
@@ -1683,7 +1707,8 @@ bool Tablebases::root_probe_wdl(Position& pos, Search::RootMoves& rootMoves, boo
 
 Config Tablebases::rank_root_moves(const OptionsMap&  options,
                                    Position&          pos,
-                                   Search::RootMoves& rootMoves) {
+                                   Search::RootMoves& rootMoves,
+                                   bool               rankDTZ) {
     Config config;
 
     if (rootMoves.empty())
@@ -1707,7 +1732,7 @@ Config Tablebases::rank_root_moves(const OptionsMap&  options,
     if (config.cardinality >= popcount(pos.pieces()) && !pos.can_castle(ANY_CASTLING))
     {
         // Rank moves using DTZ tables
-        config.rootInTB = root_probe(pos, rootMoves, options["Syzygy50MoveRule"]);
+        config.rootInTB = root_probe(pos, rootMoves, options["Syzygy50MoveRule"], rankDTZ);
 
         if (!config.rootInTB)
         {

src/syzygy/tbprobe.h

@@ -66,9 +66,12 @@ extern int MaxCardinality;
 void     init(const std::string& paths);
 WDLScore probe_wdl(Position& pos, ProbeState* result);
 int      probe_dtz(Position& pos, ProbeState* result);
-bool     root_probe(Position& pos, Search::RootMoves& rootMoves, bool rule50);
+bool     root_probe(Position& pos, Search::RootMoves& rootMoves, bool rule50, bool rankDTZ);
 bool     root_probe_wdl(Position& pos, Search::RootMoves& rootMoves, bool rule50);
-Config   rank_root_moves(const OptionsMap& options, Position& pos, Search::RootMoves& rootMoves);
+Config   rank_root_moves(const OptionsMap&  options,
+                         Position&          pos,
+                         Search::RootMoves& rootMoves,
+                         bool               rankDTZ = false);
 
 }  // namespace Stockfish::Tablebases
 

src/thread.cpp

@@ -22,17 +22,16 @@
 #include <cassert>
 #include <deque>
 #include <memory>
+#include <string>
 #include <unordered_map>
 #include <utility>
-#include <array>
 
-#include "misc.h"
 #include "movegen.h"
 #include "search.h"
 #include "syzygy/tbprobe.h"
 #include "timeman.h"
-#include "tt.h"
 #include "types.h"
+#include "uci.h"
 #include "ucioption.h"
 
 namespace Stockfish {
@@ -41,13 +40,24 @@ namespace Stockfish {
 // 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) :
+               size_t                                  n,
-    worker(std::make_unique<Search::Worker>(sharedState, std::move(sm), n)),
+               OptionalThreadToNumaNodeBinder          binder) :
     idx(n),
     nthreads(sharedState.options["Threads"]),
     stdThread(&Thread::idle_loop, this) {
 
     wait_for_search_finished();
+
+    run_custom_job([this, &binder, &sharedState, &sm, n]() {
+        // 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);
+    });
+
+    wait_for_search_finished();
 }
 
 
@@ -64,35 +74,40 @@ Thread::~Thread() {
 
 // 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(); });
+}
 
-// 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);
     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 gets parked here, 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() {
-
-    // If OS already scheduled us on a different group than 0 then don't overwrite
-    // the choice, eventually we are one of many one-threaded processes running on
-    // some Windows NUMA hardware, for instance in fishtest. To make it simple,
-    // just check if running threads are below a threshold, in this case, all this
-    // NUMA machinery is not needed.
-    if (nthreads > 8)
-        WinProcGroup::bindThisThread(idx);
-
     while (true)
     {
         std::unique_lock<std::mutex> lk(mutex);
@@ -103,81 +118,150 @@ void Thread::idle_loop() {
         if (exit)
             return;
 
+        std::function<void()> job = std::move(jobFunc);
+        jobFunc                   = nullptr;
+
         lk.unlock();
 
-        worker->start_searching();
+        if (job)
+            job();
     }
 }
 
+Search::SearchManager* ThreadPool::main_manager() { return main_thread()->worker->main_manager(); }
+
+uint64_t ThreadPool::nodes_searched() const { return accumulate(&Search::Worker::nodes); }
+uint64_t ThreadPool::tb_hits() const { return accumulate(&Search::Worker::tbHits); }
+
 // Creates/destroys threads to match the requested number.
 // Created and launched threads will immediately go to sleep in idle_loop.
 // Upon resizing, threads are recreated to allow for binding if necessary.
-void ThreadPool::set(Search::SharedState sharedState) {
+void ThreadPool::set(const NumaConfig&                           numaConfig,
+                     Search::SharedState                         sharedState,
+                     const Search::SearchManager::UpdateContext& updateContext) {
 
     if (threads.size() > 0)  // destroy any existing thread(s)
     {
         main_thread()->wait_for_search_finished();
 
-        while (threads.size() > 0)
+        threads.clear();
-            delete threads.back(), threads.pop_back();
+
+        boundThreadToNumaNode.clear();
     }
 
     const size_t requested = sharedState.options["Threads"];
 
     if (requested > 0)  // create new thread(s)
     {
-        threads.push_back(new Thread(
+        // Binding threads may be problematic when there's multiple NUMA nodes and
-          sharedState, std::unique_ptr<Search::ISearchManager>(new Search::SearchManager()), 0));
+        // 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)
-            threads.push_back(new Thread(
+        {
-              sharedState, std::unique_ptr<Search::ISearchManager>(new Search::NullSearchManager()),
+            const size_t    threadId = threads.size();
-              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();
-
-        // Reallocate the hash with the new threadpool size
-        sharedState.tt.resize(sharedState.options["Hash"], requested);
     }
 }
 
 
 // Sets threadPool data to initial values
 void ThreadPool::clear() {
+    if (threads.size() == 0)
+        return;
 
-    for (Thread* th : threads)
+    for (auto&& th : threads)
-        th->worker->clear();
+        th->clear_worker();
 
-    main_manager()->callsCnt                 = 0;
+    for (auto&& th : threads)
-    main_manager()->bestPreviousScore        = VALUE_INFINITE;
+        th->wait_for_search_finished();
+
+    // These two affect the time taken on the first move of a game:
     main_manager()->bestPreviousAverageScore = VALUE_INFINITE;
-    main_manager()->previousTimeReduction    = 1.0;
+    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));
+}
 
-// 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();
+}
+
+size_t ThreadPool::num_threads() const { return threads.size(); }
+
+
+// Wakes up main thread waiting in idle_loop() and returns immediately.
+// Main thread will wake up other threads and start the search.
 void ThreadPool::start_thinking(const OptionsMap&  options,
                                 Position&          pos,
                                 StateListPtr&      states,
-                                Search::LimitsType limits,
+                                Search::LimitsType limits) {
-                                bool               ponderMode) {
 
     main_thread()->wait_for_search_finished();
 
     main_manager()->stopOnPonderhit = stop = abortedSearch = false;
-    main_manager()->ponder                                 = ponderMode;
+    main_manager()->ponder                                 = limits.ponderMode;
 
     increaseDepth = true;
 
     Search::RootMoves rootMoves;
+    const auto        legalmoves = MoveList<LEGAL>(pos);
 
-    for (const auto& m : MoveList<LEGAL>(pos))
+    for (const auto& uciMove : limits.searchmoves)
-        if (limits.searchmoves.empty()
+    {
-            || std::count(limits.searchmoves.begin(), limits.searchmoves.end(), m))
+        auto move = UCIEngine::to_move(pos, uciMove);
+
+        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);
@@ -192,34 +276,38 @@ void ThreadPool::start_thinking(const OptionsMap&  options,
     // 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
+    // setupStates->back() later. The rootState is per thread, earlier states are
-    // since they are read-only.
+    // shared since they are read-only.
-    for (Thread* th : threads)
+    for (auto&& th : threads)
     {
-        th->worker->limits = limits;
+        th->run_custom_job([&]() {
-        th->worker->nodes = th->worker->tbHits = th->worker->nmpMinPly =
+            th->worker->limits = limits;
-          th->worker->bestMoveChanges          = 0;
+            th->worker->nodes = th->worker->tbHits = th->worker->nmpMinPly =
-        th->worker->rootDepth = th->worker->completedDepth = 0;
+              th->worker->bestMoveChanges          = 0;
-        th->worker->rootMoves                              = rootMoves;
+            th->worker->rootDepth = th->worker->completedDepth = 0;
-        th->worker->rootPos.set(pos.fen(), pos.is_chess960(), &th->worker->rootState);
+            th->worker->rootMoves                              = rootMoves;
-        th->worker->rootState = setupStates->back();
+            th->worker->rootPos.set(pos.fen(), pos.is_chess960(), &th->worker->rootState);
-        th->worker->tbConfig  = tbConfig;
+            th->worker->rootState = setupStates->back();
-        th->worker->effort    = {};
+            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();
     Value   minScore   = VALUE_NONE;
 
     std::unordered_map<Move, int64_t, Move::MoveHash> votes(
       2 * std::min(size(), bestThread->worker->rootMoves.size()));
 
     // Find the minimum score of all threads
-    for (Thread* th : 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
@@ -227,10 +315,10 @@ Thread* ThreadPool::get_best_thread() const {
         return (th->worker->rootMoves[0].score - minScore + 14) * int(th->worker->completedDepth);
     };
 
-    for (Thread* th : threads)
+    for (auto&& th : threads)
-        votes[th->worker->rootMoves[0].pv[0]] += thread_voting_value(th);
+        votes[th->worker->rootMoves[0].pv[0]] += thread_voting_value(th.get());
 
-    for (Thread* th : threads)
+    for (auto&& th : threads)
     {
         const auto bestThreadScore = bestThread->worker->rootMoves[0].score;
         const auto newThreadScore  = th->worker->rootMoves[0].score;
@@ -249,51 +337,74 @@ Thread* ThreadPool::get_best_thread() const {
         const bool newThreadInProvenLoss =
           newThreadScore != -VALUE_INFINITE && newThreadScore <= VALUE_TB_LOSS_IN_MAX_PLY;
 
-        // Note that we make sure not to pick a thread with truncated-PV for better viewer experience.
+        // We make sure not to pick a thread with truncated principal variation
         const bool betterVotingValue =
-          thread_voting_value(th) * int(newThreadPV.size() > 2)
+          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
             if (newThreadScore > bestThreadScore)
-                bestThread = th;
+                bestThread = th.get();
         }
         else if (bestThreadInProvenLoss)
         {
             // Make sure we pick the shortest mated / TB conversion
             if (newThreadInProvenLoss && newThreadScore < bestThreadScore)
-                bestThread = th;
+                bestThread = th.get();
         }
         else if (newThreadInProvenWin || newThreadInProvenLoss
                  || (newThreadScore > VALUE_TB_LOSS_IN_MAX_PLY
                      && (newThreadMoveVote > bestThreadMoveVote
                          || (newThreadMoveVote == bestThreadMoveVote && betterVotingValue))))
-            bestThread = th;
+            bestThread = th.get();
     }
 
     return bestThread;
 }
 
 
-// Start non-main threads
+// Start non-main threads.
-// Will be invoked by main thread after it has started searching
+// 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
-
 void ThreadPool::wait_for_search_finished() const {
 
-    for (Thread* th : threads)
+    for (auto&& th : threads)
         if (th != threads.front())
             th->wait_for_search_finished();
 }
 
+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

@@ -23,19 +23,48 @@
 #include <condition_variable>
 #include <cstddef>
 #include <cstdint>
+#include <functional>
 #include <memory>
 #include <mutex>
 #include <vector>
 
+#include "numa.h"
 #include "position.h"
 #include "search.h"
 #include "thread_win32_osx.h"
 
 namespace Stockfish {
 
+
 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.
@@ -43,22 +72,37 @@ using Value = int;
 // 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);
+    Thread(Search::SharedState&,
+           std::unique_ptr<Search::ISearchManager>,
+           size_t,
+           OptionalThreadToNumaNodeBinder);
     virtual ~Thread();
 
-    void   idle_loop();
+    void idle_loop();
-    void   start_searching();
+    void start_searching();
+    void clear_worker();
+    void run_custom_job(std::function<void()> f);
+
+    void ensure_network_replicated();
+
+    // Thread has been slightly altered to allow running custom jobs, so
+    // this name is no longer correct. However, this class (and ThreadPool)
+    // require further work to make them properly generic while maintaining
+    // appropriate specificity regarding search, from the point of view of an
+    // outside user, so renaming of this function is left for whenever that happens.
     void   wait_for_search_finished();
     size_t id() const { return idx; }
 
     std::unique_ptr<Search::Worker> worker;
+    std::function<void()>           jobFunc;
 
    private:
-    std::mutex              mutex;
+    std::mutex                mutex;
-    std::condition_variable cv;
+    std::condition_variable   cv;
-    size_t                  idx, nthreads;
+    size_t                    idx, nthreads;
-    bool                    exit = false, searching = true;  // Set before starting std::thread
+    bool                      exit = false, searching = true;  // Set before starting std::thread
-    NativeThread            stdThread;
+    NativeThread              stdThread;
+    NumaReplicatedAccessToken numaAccessToken;
 };
 
 
@@ -66,33 +110,45 @@ class Thread {
 // parking and, most importantly, launching a thread. All the access to threads
 // is done through this class.
 class ThreadPool {
-
    public:
+    ThreadPool() {}
+
     ~ThreadPool() {
         // destroy any existing thread(s)
         if (threads.size() > 0)
         {
             main_thread()->wait_for_search_finished();
 
-            while (threads.size() > 0)
+            threads.clear();
-                delete threads.back(), threads.pop_back();
         }
     }
 
-    void
+    ThreadPool(const ThreadPool&) = delete;
-    start_thinking(const OptionsMap&, Position&, StateListPtr&, Search::LimitsType, bool = false);
+    ThreadPool(ThreadPool&&)      = delete;
-    void clear();
-    void set(Search::SharedState);
 
-    Search::SearchManager* main_manager() const {
+    ThreadPool& operator=(const ThreadPool&) = delete;
-        return static_cast<Search::SearchManager*>(main_thread()->worker.get()->manager.get());
+    ThreadPool& operator=(ThreadPool&&)      = delete;
-    };
+
-    Thread*  main_thread() const { return threads.front(); }
+    void   start_thinking(const OptionsMap&, Position&, StateListPtr&, Search::LimitsType);
-    uint64_t nodes_searched() const { return accumulate(&Search::Worker::nodes); }
+    void   run_on_thread(size_t threadId, std::function<void()> f);
-    uint64_t tb_hits() const { return accumulate(&Search::Worker::tbHits); }
+    void   wait_on_thread(size_t threadId);
-    Thread*  get_best_thread() const;
+    size_t num_threads() const;
-    void     start_searching();
+    void   clear();
-    void     wait_for_search_finished() const;
+    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;
 
@@ -104,13 +160,14 @@ class ThreadPool {
     auto empty() const noexcept { return threads.empty(); }
 
    private:
-    StateListPtr         setupStates;
+    StateListPtr                         setupStates;
-    std::vector<Thread*> threads;
+    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 (Thread* th : threads)
+        for (auto&& th : threads)
             sum += (th->worker.get()->*member).load(std::memory_order_relaxed);
         return sum;
     }

src/timeman.cpp

@@ -30,17 +30,14 @@ namespace Stockfish {
 
 TimePoint TimeManagement::optimum() const { return optimumTime; }
 TimePoint TimeManagement::maximum() const { return maximumTime; }
-TimePoint TimeManagement::elapsed(size_t nodes) const {
-    return useNodesTime ? TimePoint(nodes) : now() - startTime;
-}
 
 void TimeManagement::clear() {
-    availableNodes = 0;  // When in 'nodes as time' mode
+    availableNodes = -1;  // When in 'nodes as time' mode
 }
 
 void TimeManagement::advance_nodes_time(std::int64_t nodes) {
     assert(useNodesTime);
-    availableNodes += nodes;
+    availableNodes = std::max(int64_t(0), availableNodes - nodes);
 }
 
 // Called at the beginning of the search and calculates
@@ -50,15 +47,19 @@ void TimeManagement::advance_nodes_time(std::int64_t nodes) {
 void TimeManagement::init(Search::LimitsType& limits,
                           Color               us,
                           int                 ply,
-                          const OptionsMap&   options) {
+                          const OptionsMap&   options,
-    // If we have no time, no need to initialize TM, except for the start time,
+                          double&             originalTimeAdjust) {
-    // which is used by movetime.
+    TimePoint npmsec = TimePoint(options["nodestime"]);
-    startTime = limits.startTime;
+
+    // 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"]);
-    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.
@@ -68,56 +69,69 @@ void TimeManagement::init(Search::LimitsType& limits,
     // 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 (useNodesTime)
     {
-        useNodesTime = true;
+        if (availableNodes == -1)                       // Only once at game start
-
-        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;
+        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 of 50 moves
     int mtg = limits.movestogo ? std::min(limits.movestogo, 50) : 50;
 
+    // If less than one second, gradually reduce mtg
+    if (scaledTime < 1000 && double(mtg) / scaledInc > 0.05)
+    {
+        mtg = scaledTime * 0.05;
+    }
+
     // 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));
 
     // x basetime (+ z increment)
-    // If there is a healthy increment, timeLeft can exceed actual available
+    // If there is a healthy increment, timeLeft can exceed the actual available
-    // game time for the current move, so also cap to 20% of available game time.
+    // game time for the current move, so also cap to a percentage of available game time.
     if (limits.movestogo == 0)
     {
-        // Use extra time with larger increments
+        // Extra time according to timeLeft
-        double optExtra = std::clamp(1.0 + 12.5 * limits.inc[us] / limits.time[us], 1.0, 1.11);
+        if (originalTimeAdjust < 0)
+            originalTimeAdjust = 0.3285 * std::log10(timeLeft) - 0.4830;
 
         // Calculate time constants based on current time left.
-        double optConstant =
+        double logTimeInSec = std::log10(scaledTime / 1000.0);
-          std::min(0.00334 + 0.0003 * std::log10(limits.time[us] / 1000.0), 0.0049);
+        double optConstant  = std::min(0.00308 + 0.000319 * logTimeInSec, 0.00506);
-        double maxConstant = std::max(3.4 + 3.0 * std::log10(limits.time[us] / 1000.0), 2.76);
+        double maxConstant  = std::max(3.39 + 3.01 * logTimeInSec, 2.93);
 
-        optScale = std::min(0.0120 + std::pow(ply + 3.1, 0.44) * optConstant,
+        optScale = std::min(0.0122 + std::pow(ply + 2.95, 0.462) * optConstant,
-                            0.21 * limits.time[us] / double(timeLeft))
+                            0.213 * limits.time[us] / timeLeft)
-                 * optExtra;
+                 * originalTimeAdjust;
-        maxScale = std::min(6.9, maxConstant + ply / 12.2);
+
+        maxScale = std::min(6.64, maxConstant + 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));
+        optScale = std::min((0.88 + ply / 116.4) / mtg, 0.88 * limits.time[us] / timeLeft);
         maxScale = std::min(6.3, 1.5 + 0.11 * mtg);
     }
 
     // Limit the maximum possible time for this move
     optimumTime = TimePoint(optScale * timeLeft);
     maximumTime =
-      TimePoint(std::min(0.84 * limits.time[us] - moveOverhead, maxScale * optimumTime)) - 10;
+      TimePoint(std::min(0.825 * limits.time[us] - moveOverhead, maxScale * optimumTime)) - 10;
 
     if (options["Ponder"])
         optimumTime += optimumTime / 4;

src/timeman.h

@@ -19,7 +19,6 @@
 #ifndef TIMEMAN_H_INCLUDED
 #define TIMEMAN_H_INCLUDED
 
-#include <cstddef>
 #include <cstdint>
 
 #include "misc.h"
@@ -37,11 +36,19 @@ struct LimitsType;
 // the maximum available time, the game move number, and other parameters.
 class TimeManagement {
    public:
-    void init(Search::LimitsType& limits, Color us, int ply, const OptionsMap& options);
+    void init(Search::LimitsType& limits,
+              Color               us,
+              int                 ply,
+              const OptionsMap&   options,
+              double&             originalTimeAdjust);
 
     TimePoint optimum() const;
     TimePoint maximum() const;
-    TimePoint elapsed(std::size_t nodes) const;
+    template<typename FUNC>
+    TimePoint elapsed(FUNC nodes) const {
+        return useNodesTime ? TimePoint(nodes()) : elapsed_time();
+    }
+    TimePoint elapsed_time() const { return now() - startTime; };
 
     void clear();
     void advance_nodes_time(std::int64_t nodes);
@@ -51,7 +58,7 @@ class TimeManagement {
     TimePoint optimumTime;
     TimePoint maximumTime;
 
-    std::int64_t availableNodes = 0;      // When in 'nodes as time' mode
+    std::int64_t availableNodes = -1;     // When in 'nodes as time' mode
     bool         useNodesTime   = false;  // True if we are in 'nodes as time' mode
 };
 

src/tt.cpp

@@ -19,33 +19,93 @@
 #include "tt.h"
 
 #include <cassert>
+#include <cstdint>
 #include <cstdlib>
 #include <cstring>
 #include <iostream>
-#include <thread>
-#include <vector>
 
+#include "memory.h"
 #include "misc.h"
+#include "syzygy/tbprobe.h"
+#include "thread.h"
 
 namespace Stockfish {
 
+
+// TTEntry struct is the 10 bytes transposition table entry, defined as below:
+//
+// key        16 bit
+// depth       8 bit
+// generation  5 bit
+// pv node     1 bit
+// bound type  2 bit
+// move       16 bit
+// value      16 bit
+// 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.
+
+struct TTEntry {
+
+    // Convert internal bitfields to external types
+    TTData read() const {
+        return TTData{Move(move16),           Value(value16),
+                      Value(eval16),          Depth(depth8 + DEPTH_ENTRY_OFFSET),
+                      Bound(genBound8 & 0x3), bool(genBound8 & 0x4)};
+    }
+
+    bool is_occupied() const;
+    void save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev, uint8_t generation8);
+    // The returned age is a multiple of TranspositionTable::GENERATION_DELTA
+    uint8_t relative_age(const uint8_t generation8) const;
+
+   private:
+    friend class TranspositionTable;
+
+    uint16_t key16;
+    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 any existing move for the same position
+    // 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_OFFSET + 2 * pv > depth8 - 4)
+    if (b == BOUND_EXACT || uint16_t(k) != key16 || d - DEPTH_ENTRY_OFFSET + 2 * pv > depth8 - 4
+        || relative_age(generation8))
     {
-        assert(d > DEPTH_OFFSET);
+        assert(d > DEPTH_ENTRY_OFFSET);
-        assert(d < 256 + DEPTH_OFFSET);
+        assert(d < 256 + DEPTH_ENTRY_OFFSET);
 
         key16     = uint16_t(k);
-        depth8    = uint8_t(d - DEPTH_OFFSET);
+        depth8    = uint8_t(d - DEPTH_ENTRY_OFFSET);
         genBound8 = uint8_t(generation8 | uint8_t(pv) << 2 | b);
         value16   = int16_t(v);
         eval16    = int16_t(ev);
@@ -53,100 +113,137 @@ void TTEntry::save(
 }
 
 
+uint8_t TTEntry::relative_age(const uint8_t generation8) const {
+    // Due to our packed storage format for generation and its cyclic
+    // nature we add GENERATION_CYCLE (256 is the modulus, plus what
+    // is needed to keep the unrelated lowest n bits from affecting
+    // the result) to calculate the entry age correctly even after
+    // generation8 overflows into the next cycle.
+    return (GENERATION_CYCLE + generation8 - genBound8) & GENERATION_MASK;
+}
+
+
+// TTWriter is but a very thin wrapper around the pointer
+TTWriter::TTWriter(TTEntry* tte) :
+    entry(tte) {}
+
+void TTWriter::write(
+  Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev, uint8_t generation8) {
+    entry->save(k, v, pv, b, d, m, ev, generation8);
+}
+
+
+// A TranspositionTable is an array of Cluster, of size clusterCount. Each cluster consists of ClusterSize number
+// of TTEntry. Each non-empty TTEntry contains information on exactly one position. The size of a Cluster should
+// divide the size of a cache line for best performance, as the cacheline is prefetched when possible.
+
+static constexpr int ClusterSize = 3;
+
+struct Cluster {
+    TTEntry entry[ClusterSize];
+    char    padding[2];  // Pad to 32 bytes
+};
+
+static_assert(sizeof(Cluster) == 32, "Suboptimal Cluster size");
+
+
 // Sets the size of the transposition table,
-// measured in megabytes. Transposition table consists of a power of 2 number
+// measured in megabytes. Transposition table consists
 // of clusters and each cluster consists of ClusterSize number of TTEntry.
-void TranspositionTable::resize(size_t mbSize, int threadCount) {
+void TranspositionTable::resize(size_t mbSize, ThreadPool& threads) {
     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(threadCount);
+    clear(threads);
 }
 
 
 // Initializes the entire transposition table to zero,
 // in a multi-threaded way.
-void TranspositionTable::clear(size_t threadCount) {
+void TranspositionTable::clear(ThreadPool& threads) {
-    std::vector<std::thread> threads;
+    generation8              = 0;
+    const size_t threadCount = threads.num_threads();
 
-    for (size_t idx = 0; idx < size_t(threadCount); ++idx)
+    for (size_t i = 0; i < threadCount; ++i)
     {
-        threads.emplace_back([this, idx, threadCount]() {
+        threads.run_on_thread(i, [this, i, threadCount]() {
-            // Thread binding gives faster search on systems with a first-touch policy
-            if (threadCount > 8)
-                WinProcGroup::bindThisThread(idx);
-
             // Each thread will zero its part of the hash table
-            const size_t stride = size_t(clusterCount / threadCount), start = size_t(stride * idx),
+            const size_t stride = clusterCount / threadCount;
-                         len = idx != size_t(threadCount) - 1 ? stride : clusterCount - start;
+            const size_t start  = stride * i;
+            const size_t len    = i + 1 != threadCount ? stride : clusterCount - start;
 
             std::memset(&table[start], 0, len * sizeof(Cluster));
         });
     }
 
-    for (std::thread& th : threads)
+    for (size_t i = 0; i < threadCount; ++i)
-        th.join();
+        threads.wait_on_thread(i);
-}
-
-
-// Looks up the current position in the transposition
-// table. It returns true and a pointer to the TTEntry if the position is found.
-// Otherwise, it returns false and a pointer to an empty or least valuable TTEntry
-// to be replaced later. The replace value of an entry is calculated as its depth
-// minus 8 times its relative age. TTEntry t1 is considered more valuable than
-// TTEntry t2 if its replace value is greater than that of t2.
-TTEntry* TranspositionTable::probe(const Key key, bool& found) 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 || !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
-    TTEntry* replace = tte;
-    for (int i = 1; i < ClusterSize; ++i)
-        // Due to our packed storage format for generation and its cyclic
-        // nature we add GENERATION_CYCLE (256 is the modulus, plus what
-        // is needed to keep the unrelated lowest n bits from affecting
-        // the result) to calculate the entry age correctly even after
-        // generation8 overflows into the next cycle.
-        if (replace->depth8
-              - ((GENERATION_CYCLE + generation8 - replace->genBound8) & GENERATION_MASK)
-            > tte[i].depth8
-                - ((GENERATION_CYCLE + generation8 - tte[i].genBound8) & GENERATION_MASK))
-            replace = &tte[i];
-
-    return found = false, replace;
 }
 
 
 // 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() const {
 
     int cnt = 0;
     for (int i = 0; i < 1000; ++i)
         for (int j = 0; j < ClusterSize; ++j)
-            cnt += table[i].entry[j].depth8
+            cnt += table[i].entry[j].is_occupied()
                 && (table[i].entry[j].genBound8 & GENERATION_MASK) == generation8;
 
     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(), TTWriter(replace)};
+}
+
+
+TTEntry* TranspositionTable::first_entry(const Key key) const {
+    return &table[mul_hi64(key, clusterCount)].entry[0];
+}
+
 }  // namespace Stockfish

src/tt.h

@@ -21,88 +21,76 @@
 
 #include <cstddef>
 #include <cstdint>
+#include <tuple>
 
-#include "misc.h"
+#include "memory.h"
 #include "types.h"
 
 namespace Stockfish {
 
-// TTEntry struct is the 10 bytes transposition table entry, defined as below:
+class ThreadPool;
+struct TTEntry;
+struct Cluster;
+
+// 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.
 //
-// key        16 bit
+// `probe` is the primary method: given a board position, we lookup its entry in the table, and return a tuple of:
-// depth       8 bit
+//   1) whether the entry already has this position
-// generation  5 bit
+//   2) a copy of the prior data (if any) (may be inconsistent due to read races)
-// pv node     1 bit
+//   3) a writer object to this entry
-// bound type  2 bit
+// The copied data and the writer are separated to maintain clear boundaries between local vs global objects.
-// move       16 bit
-// value      16 bit
-// eval value 16 bit
-struct TTEntry {
 
-    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, uint8_t generation8);
 
-   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;
-    uint16_t key16;
+    Value value, eval;
-    uint8_t  depth8;
+    Depth depth;
-    uint8_t  genBound8;
+    Bound bound;
-    Move     move16;
+    bool  is_pv;
-    int16_t  value16;
+};
-    int16_t  eval16;
+
+
+// This is used to make racy writes to the global TT.
+struct TTWriter {
+   public:
+    void write(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev, uint8_t generation8);
+
+   private:
+    friend class TranspositionTable;
+    TTEntry* entry;
+    TTWriter(TTEntry* tte);
 };
 
 
-// A TranspositionTable is an array of Cluster, of size clusterCount. Each
-// cluster consists of ClusterSize number of TTEntry. Each non-empty TTEntry
-// contains information on exactly one position. The size of a Cluster should
-// divide the size of a cache line for best performance, as the cacheline is
-// prefetched when possible.
 class TranspositionTable {
 
-    static constexpr int ClusterSize = 3;
-
-    struct Cluster {
-        TTEntry entry[ClusterSize];
-        char    padding[2];  // Pad to 32 bytes
-    };
-
-    static_assert(sizeof(Cluster) == 32, "Unexpected Cluster size");
-
-    // Constants used to refresh the hash table periodically
-    static constexpr unsigned GENERATION_BITS = 3;  // nb of bits reserved for other things
-    static constexpr int      GENERATION_DELTA =
-      (1 << GENERATION_BITS);  // increment for generation field
-    static constexpr int GENERATION_CYCLE = 255 + (1 << GENERATION_BITS);  // cycle length
-    static constexpr int GENERATION_MASK =
-      (0xFF << GENERATION_BITS) & 0xFF;  // mask to pull out generation number
-
    public:
     ~TranspositionTable() { aligned_large_pages_free(table); }
-    void new_search() { generation8 += GENERATION_DELTA; }  // Lower bits are used for other things
-    TTEntry* probe(const Key key, bool& found) const;
-    int      hashfull() const;
-    void     resize(size_t mbSize, int threadCount);
-    void     clear(size_t threadCount);
 
-    TTEntry* first_entry(const Key key) const {
+    void resize(size_t mbSize, ThreadPool& threads);  // Set TT size
-        return &table[mul_hi64(key, clusterCount)].entry[0];
+    void clear(ThreadPool& threads);                  // Re-initialize memory, multithreaded
-    }
+    int  hashfull()
+      const;  // Approximate what fraction of entries (permille) have been written to during this root search
 
-    uint8_t generation() const { return generation8; }
+    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.
 
    private:
     friend struct TTEntry;
 
     size_t   clusterCount;
-    Cluster* table       = nullptr;
+    Cluster* table = nullptr;
-    uint8_t  generation8 = 0;  // Size must be not bigger than TTEntry::genBound8
+
+    uint8_t generation8 = 0;  // Size must be not bigger than TTEntry::genBound8
 };
 
 }  // namespace Stockfish

src/tune.cpp

@@ -21,6 +21,7 @@
 #include <algorithm>
 #include <iostream>
 #include <map>
+#include <optional>
 #include <sstream>
 #include <string>
 
@@ -30,10 +31,41 @@ using std::string;
 
 namespace Stockfish {
 
-bool                              Tune::update_on_last;
+bool          Tune::update_on_last;
-const Option*                     LastOption = nullptr;
+const Option* LastOption = nullptr;
-OptionsMap*                       Tune::options;
+OptionsMap*   Tune::options;
-static std::map<std::string, int> TuneResults;
+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)[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) {
 
@@ -54,29 +86,6 @@ string Tune::next(string& names, bool pop) {
     return name;
 }
 
-static void on_tune(const Option& o) {
-
-    if (!Tune::update_on_last || LastOption == &o)
-        Tune::read_options();
-}
-
-static void make_option(OptionsMap* options, const string& n, int v, const SetRange& r) {
-
-    // Do not generate option when there is nothing to tune (ie. min = max)
-    if (r(v).first == r(v).second)
-        return;
-
-    if (TuneResults.count(n))
-        v = TuneResults[n];
-
-    (*options)[n] << 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() {
@@ -112,7 +121,6 @@ void Tune::Entry<Tune::PostUpdate>::read_option() {
 
 namespace Stockfish {
 
-void Tune::read_results() { /* ...insert your values here... */
+void Tune::read_results() { /* ...insert your values here... */ }
-}
 
 }  // namespace Stockfish

src/tune.h

@@ -145,6 +145,8 @@ class Tune {
         return add(value, (next(names), std::move(names)), args...);
     }
 
+    static void make_option(OptionsMap* options, const std::string& n, int v, const SetRange& r);
+
     std::vector<std::unique_ptr<EntryBase>> list;
 
    public:
@@ -158,7 +160,7 @@ class Tune {
         for (auto& e : instance().list)
             e->init_option();
         read_options();
-    }  // Deferred, due to UCI::Options access
+    }  // Deferred, due to UCIEngine::Options access
     static void read_options() {
         for (auto& e : instance().list)
             e->read_option();

src/types.h

@@ -137,9 +137,9 @@ enum Bound {
     BOUND_EXACT = BOUND_UPPER | BOUND_LOWER
 };
 
-// Value is used as an alias for int16_t, this is done to differentiate between
+// Value is used as an alias for int, this is done to differentiate between a search
-// a search value and any other integer value. The values used in search are always
+// value and any other integer value. The values used in search are always supposed
-// supposed to be in the range (-VALUE_NONE, VALUE_NONE] and should not exceed this range.
+// to be in the range (-VALUE_NONE, VALUE_NONE] and should not exceed this range.
 using Value = int;
 
 constexpr Value VALUE_ZERO     = 0;
@@ -187,12 +187,21 @@ constexpr Value PieceValue[PIECE_NB] = {
 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 stored in the transposition table is literal: the search depth
-    DEPTH_NONE = -6,
+    // (effort) used to make the corresponding transposition table value. In
-
+    // quiescence search, however, the transposition table entries only store
-    DEPTH_OFFSET = -7  // value used only for TT entry occupancy check
+    // the current quiescence move generation stage (which should thus compare
+    // 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
@@ -351,9 +360,10 @@ enum MoveType {
 // 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
+// Special cases are Move::none() and Move::null(). We can sneak these in because
-// any normal move destination square is always different from origin square
+// in any normal move the destination square and origin square are always different,
-// while Move::none() and Move::null() have the same origin and destination square.
+// but Move::none() and Move::null() have the same origin and destination square.
+
 class Move {
    public:
     Move() = default;

src/uci.cpp

@@ -19,86 +19,65 @@
 #include "uci.h"
 
 #include <algorithm>
-#include <cassert>
 #include <cctype>
 #include <cmath>
-#include <cstdlib>
+#include <cstdint>
-#include <deque>
-#include <memory>
 #include <optional>
 #include <sstream>
+#include <string_view>
+#include <utility>
 #include <vector>
-#include <cstdint>
 
 #include "benchmark.h"
-#include "evaluate.h"
+#include "engine.h"
 #include "movegen.h"
-#include "nnue/evaluate_nnue.h"
-#include "nnue/nnue_architecture.h"
 #include "position.h"
+#include "score.h"
 #include "search.h"
-#include "syzygy/tbprobe.h"
 #include "types.h"
 #include "ucioption.h"
-#include "perft.h"
 
 namespace Stockfish {
 
-constexpr auto StartFEN             = "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1";
+constexpr auto StartFEN = "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1";
-constexpr int  NormalizeToPawnValue = 356;
+template<typename... Ts>
-constexpr int  MaxHashMB            = Is64Bit ? 33554432 : 2048;
+struct overload: Ts... {
+    using Ts::operator()...;
+};
 
-UCI::UCI(int argc, char** argv) :
+template<typename... Ts>
-    cli(argc, argv) {
+overload(Ts...) -> overload<Ts...>;
 
-    evalFiles = {{Eval::NNUE::Big, {"EvalFile", EvalFileDefaultNameBig, "None", ""}},
+void UCIEngine::print_info_string(const std::string& str) {
-                 {Eval::NNUE::Small, {"EvalFileSmall", EvalFileDefaultNameSmall, "None", ""}}};
+    sync_cout_start();
-
+    for (auto& line : split(str, "\n"))
-
+    {
-    options["Debug Log File"] << Option("", [](const Option& o) { start_logger(o); });
+        if (!is_whitespace(line))
-
+        {
-    options["Threads"] << Option(1, 1, 1024, [this](const Option&) {
+            std::cout << "info string " << line << '\n';
-        threads.set({options, threads, tt});
+        }
-    });
+    }
-
+    sync_cout_end();
-    options["Hash"] << Option(16, 1, MaxHashMB, [this](const Option& o) {
-        threads.main_thread()->wait_for_search_finished();
-        tt.resize(o, options["Threads"]);
-    });
-
-    options["Clear Hash"] << Option([this](const Option&) { search_clear(); });
-    options["Ponder"] << Option(false);
-    options["MultiPV"] << Option(1, 1, MAX_MOVES);
-    options["Skill Level"] << Option(20, 0, 20);
-    options["Move Overhead"] << Option(10, 0, 5000);
-    options["nodestime"] << Option(0, 0, 10000);
-    options["UCI_Chess960"] << Option(false);
-    options["UCI_LimitStrength"] << Option(false);
-    options["UCI_Elo"] << Option(1320, 1320, 3190);
-    options["UCI_ShowWDL"] << Option(false);
-    options["SyzygyPath"] << Option("<empty>", [](const Option& o) { Tablebases::init(o); });
-    options["SyzygyProbeDepth"] << Option(1, 1, 100);
-    options["Syzygy50MoveRule"] << Option(true);
-    options["SyzygyProbeLimit"] << Option(7, 0, 7);
-    options["EvalFile"] << Option(EvalFileDefaultNameBig, [this](const Option&) {
-        evalFiles = Eval::NNUE::load_networks(cli.binaryDirectory, options, evalFiles);
-    });
-    options["EvalFileSmall"] << Option(EvalFileDefaultNameSmall, [this](const Option&) {
-        evalFiles = Eval::NNUE::load_networks(cli.binaryDirectory, options, evalFiles);
-    });
-
-    threads.set({options, threads, tt});
-
-    search_clear();  // After threads are up
 }
 
-void UCI::loop() {
+UCIEngine::UCIEngine(int argc, char** argv) :
+    engine(argv[0]),
+    cli(argc, argv) {
 
-    Position     pos;
+    engine.get_options().add_info_listener([](const std::optional<std::string>& str) {
-    std::string  token, cmd;
+        if (str.has_value())
-    StateListPtr states(new std::deque<StateInfo>(1));
+            print_info_string(*str);
+    });
 
-    pos.set(StartFEN, false, &states->back());
+    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); });
+}
+
+void UCIEngine::loop() {
+    std::string token, cmd;
 
     for (int i = 1; i < cli.argc; ++i)
         cmd += std::string(cli.argv[i]) + " ";
@@ -115,49 +94,62 @@ void UCI::loop() {
         is >> std::skipws >> token;
 
         if (token == "quit" || token == "stop")
-            threads.stop = true;
+            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")
-            threads.main_manager()->ponder = false;  // Switch to the normal search
+            engine.set_ponderhit(false);
 
         else if (token == "uci")
+        {
             sync_cout << "id name " << engine_info(true) << "\n"
-                      << options << "\nuciok" << sync_endl;
+                      << engine.get_options() << sync_endl;
+
+            sync_cout << "uciok" << sync_endl;
+        }
 
         else if (token == "setoption")
             setoption(is);
         else if (token == "go")
-            go(pos, is, states);
+        {
+            // 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_binding_information_as_string());
+            go(is);
+        }
         else if (token == "position")
-            position(pos, is, states);
+            position(is);
         else if (token == "ucinewgame")
-            search_clear();
+            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")
-            pos.flip();
+            engine.flip();
         else if (token == "bench")
-            bench(pos, is, states);
+            bench(is);
         else if (token == "d")
-            sync_cout << pos << sync_endl;
+            sync_cout << engine.visualize() << sync_endl;
         else if (token == "eval")
-            trace_eval(pos);
+            engine.trace_eval();
         else if (token == "compiler")
             sync_cout << compiler_info() << sync_endl;
         else if (token == "export_net")
         {
-            std::optional<std::string> filename;
+            std::pair<std::optional<std::string>, std::string> files[2];
-            std::string                f;
+
-            if (is >> std::skipws >> f)
+            if (is >> std::skipws >> files[0].second)
-                filename = f;
+                files[0].first = files[0].second;
-            Eval::NNUE::save_eval(filename, Eval::NNUE::Big, evalFiles);
+
+            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
@@ -175,18 +167,16 @@ void UCI::loop() {
     } while (token != "quit" && cli.argc == 1);  // The command-line arguments are one-shot
 }
 
-void UCI::go(Position& pos, std::istringstream& is, StateListPtr& states) {
+Search::LimitsType UCIEngine::parse_limits(std::istream& is) {
-
     Search::LimitsType limits;
     std::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(to_move(pos, token));
+                limits.searchmoves.push_back(to_lower(token));
 
         else if (token == "wtime")
             is >> limits.time[WHITE];
@@ -211,24 +201,33 @@ void UCI::go(Position& pos, std::istringstream& is, StateListPtr& states) {
         else if (token == "infinite")
             limits.infinite = 1;
         else if (token == "ponder")
-            ponderMode = true;
+            limits.ponderMode = true;
 
-    Eval::NNUE::verify(options, evalFiles);
+    return limits;
-
-    if (limits.perft)
-    {
-        perft(pos.fen(), limits.perft, options["UCI_Chess960"]);
-        return;
-    }
-
-    threads.start_thinking(options, pos, states, limits, ponderMode);
 }
 
-void UCI::bench(Position& pos, std::istream& args, StateListPtr& states) {
+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();
 
-    std::vector<std::string> list = setup_bench(pos, args);
+    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; });
@@ -242,24 +241,33 @@ void UCI::bench(Position& pos, std::istream& args, StateListPtr& states) {
 
         if (token == "go" || token == "eval")
         {
-            std::cerr << "\nPosition: " << cnt++ << '/' << num << " (" << pos.fen() << ")"
+            std::cerr << "\nPosition: " << cnt++ << '/' << num << " (" << engine.fen() << ")"
                       << std::endl;
             if (token == "go")
             {
-                go(pos, is, states);
+                Search::LimitsType limits = parse_limits(is);
-                threads.main_thread()->wait_for_search_finished();
+
-                nodes += threads.nodes_searched();
+                if (limits.perft)
+                    nodesSearched = perft(limits);
+                else
+                {
+                    engine.go(limits);
+                    engine.wait_for_search_finished();
+                }
+
+                nodes += nodesSearched;
+                nodesSearched = 0;
             }
             else
-                trace_eval(pos);
+                engine.trace_eval();
         }
         else if (token == "setoption")
             setoption(is);
         else if (token == "position")
-            position(pos, is, states);
+            position(is);
         else if (token == "ucinewgame")
         {
-            search_clear();  // Search::clear() may take a while
+            engine.search_clear();  // search_clear may take a while
             elapsed = now();
         }
     }
@@ -268,36 +276,28 @@ void UCI::bench(Position& pos, std::istream& args, StateListPtr& states) {
 
     dbg_print();
 
-    std::cerr << "\n==========================="
+    std::cerr << "\n==========================="    //
-              << "\nTotal time (ms) : " << elapsed << "\nNodes searched  : " << nodes
+              << "\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 UCI::trace_eval(Position& pos) {
-    StateListPtr states(new std::deque<StateInfo>(1));
-    Position     p;
-    p.set(pos.fen(), options["UCI_Chess960"], &states->back());
 
-    Eval::NNUE::verify(options, evalFiles);
+void UCIEngine::setoption(std::istringstream& is) {
-
+    engine.wait_for_search_finished();
-    sync_cout << "\n" << Eval::trace(p) << sync_endl;
+    engine.get_options().setoption(is);
 }
 
-void UCI::search_clear() {
+std::uint64_t UCIEngine::perft(const Search::LimitsType& limits) {
-    threads.main_thread()->wait_for_search_finished();
+    auto nodes = engine.perft(engine.fen(), limits.perft, engine.get_options()["UCI_Chess960"]);
-
+    sync_cout << "\nNodes searched: " << nodes << "\n" << sync_endl;
-    tt.clear(options["Threads"]);
+    return nodes;
-    threads.clear();
-    Tablebases::init(options["SyzygyPath"]);  // Free mapped files
 }
 
-void UCI::setoption(std::istringstream& is) {
+void UCIEngine::position(std::istringstream& is) {
-    threads.main_thread()->wait_for_search_finished();
-    options.setoption(is);
-}
-
-void UCI::position(Position& pos, std::istringstream& is, StateListPtr& states) {
-    Move        m;
     std::string token, fen;
 
     is >> token;
@@ -313,42 +313,99 @@ void UCI::position(Position& pos, std::istringstream& is, StateListPtr& states)
     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());
 
-    // Parse the move list, if any
+    while (is >> token)
-    while (is >> token && (m = to_move(pos, token)) != Move::none())
     {
-        states->emplace_back();
+        moves.push_back(token);
-        pos.do_move(m, states->back());
     }
+
+    engine.set_position(fen, moves);
 }
 
-int UCI::to_cp(Value v) { return 100 * v / NormalizeToPawnValue; }
+namespace {
 
-std::string UCI::value(Value v) {
+struct WinRateParams {
-    assert(-VALUE_INFINITE < v && v < VALUE_INFINITE);
+    double a;
+    double b;
+};
 
+WinRateParams win_rate_params(const Position& pos) {
+
+    int material = pos.count<PAWN>() + 3 * pos.count<KNIGHT>() + 3 * pos.count<BISHOP>()
+                 + 5 * pos.count<ROOK>() + 9 * pos.count<QUEEN>();
+
+    // The fitted model only uses data for material counts in [17, 78], and is anchored at count 58.
+    double m = std::clamp(material, 17, 78) / 58.0;
+
+    // Return a = p_a(material) and b = p_b(material), see github.com/official-stockfish/WDL_model
+    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 b = (((bs[0] * m + bs[1]) * m + bs[2]) * m) + bs[3];
+
+    return {a, b};
+}
+
+// The win rate model is 1 / (1 + exp((a - eval) / b)), where a = p_a(material) and b = p_b(material).
+// It fits the LTC fishtest statistics rather accurately.
+int win_rate_model(Value v, const Position& pos) {
+
+    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)));
+}
+}
+
+std::string UCIEngine::format_score(const Score& s) {
+    constexpr int TB_CP = 20000;
+    const auto    format =
+      overload{[](Score::Mate mate) -> std::string {
+                   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);
+               }};
+
+    return s.visit(format);
+}
+
+// Turns a Value to an integer centipawn number,
+// without treatment of mate and similar special scores.
+int UCIEngine::to_cp(Value v, const Position& pos) {
+
+    // 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.
+
+    auto [a, b] = win_rate_params(pos);
+
+    return std::round(100 * int(v) / a);
+}
+
+std::string UCIEngine::wdl(Value v, const Position& pos) {
     std::stringstream ss;
 
-    if (std::abs(v) < VALUE_TB_WIN_IN_MAX_PLY)
+    int wdl_w = win_rate_model(v, pos);
-        ss << "cp " << to_cp(v);
+    int wdl_l = win_rate_model(-v, pos);
-    else if (std::abs(v) <= VALUE_TB)
+    int wdl_d = 1000 - wdl_w - wdl_l;
-    {
+    ss << wdl_w << " " << wdl_d << " " << wdl_l;
-        const int ply = VALUE_TB - 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();
 }
 
-std::string UCI::square(Square s) {
+std::string UCIEngine::square(Square s) {
     return std::string{char('a' + file_of(s)), char('1' + rank_of(s))};
 }
 
-std::string UCI::move(Move m, bool chess960) {
+std::string UCIEngine::move(Move m, bool chess960) {
     if (m == Move::none())
         return "(none)";
 
@@ -369,45 +426,15 @@ std::string UCI::move(Move m, bool chess960) {
     return move;
 }
 
-namespace {
-// The win rate model returns the probability of winning (in per mille units) given an
-// eval and a game ply. It fits the LTC fishtest statistics rather accurately.
-int win_rate_model(Value v, int ply) {
 
-    // The fitted model only uses data for moves in [8, 120], and is anchored at move 32.
+std::string UCIEngine::to_lower(std::string str) {
-    double m = std::clamp(ply / 2 + 1, 8, 120) / 32.0;
+    std::transform(str.begin(), str.end(), str.begin(), [](auto c) { return std::tolower(c); });
 
-    // The coefficients of a third-order polynomial fit is based on the fishtest data
+    return str;
-    // for two parameters that need to transform eval to the argument of a logistic
-    // function.
-    constexpr double as[] = {-1.06249702, 7.42016937, 0.89425629, 348.60356174};
-    constexpr double bs[] = {-5.33122190, 39.57831533, -90.84473771, 123.40620748};
-
-    // Enforce that NormalizeToPawnValue corresponds to a 50% win rate at move 32.
-    static_assert(NormalizeToPawnValue == int(0.5 + as[0] + as[1] + as[2] + as[3]));
-
-    double a = (((as[0] * m + as[1]) * m + as[2]) * m) + as[3];
-    double b = (((bs[0] * m + bs[1]) * m + bs[2]) * m) + bs[3];
-
-    // 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)));
-}
 }
 
-std::string UCI::wdl(Value v, int ply) {
+Move UCIEngine::to_move(const Position& pos, std::string str) {
-    std::stringstream ss;
+    str = to_lower(str);
-
-    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();
-}
-
-Move UCI::to_move(const Position& pos, std::string& str) {
-    if (str.length() == 5)
-        str[4] = char(tolower(str[4]));  // The promotion piece character must be lowercased
 
     for (const auto& m : MoveList<LEGAL>(pos))
         if (str == move(m, pos.is_chess960()))
@@ -416,4 +443,51 @@ Move UCI::to_move(const Position& pos, std::string& str) {
     return Move::none();
 }
 
+void UCIEngine::on_update_no_moves(const Engine::InfoShort& info) {
+    sync_cout << "info depth " << info.depth << " score " << format_score(info.score) << sync_endl;
+}
+
+void UCIEngine::on_update_full(const Engine::InfoFull& info, bool showWDL) {
+    std::stringstream ss;
+
+    ss << "info";
+    ss << " depth " << info.depth                 //
+       << " seldepth " << info.selDepth           //
+       << " multipv " << info.multiPV             //
+       << " score " << format_score(info.score);  //
+
+    if (showWDL)
+        ss << " wdl " << info.wdl;
+
+    if (!info.bound.empty())
+        ss << " " << info.bound;
+
+    ss << " nodes " << info.nodes        //
+       << " nps " << info.nps            //
+       << " hashfull " << info.hashfull  //
+       << " tbhits " << info.tbHits      //
+       << " time " << info.timeMs        //
+       << " pv " << info.pv;             //
+
+    sync_cout << ss.str() << sync_endl;
+}
+
+void UCIEngine::on_iter(const Engine::InfoIter& info) {
+    std::stringstream ss;
+
+    ss << "info";
+    ss << " depth " << info.depth                     //
+       << " currmove " << info.currmove               //
+       << " currmovenumber " << info.currmovenumber;  //
+
+    sync_cout << ss.str() << sync_endl;
+}
+
+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

@@ -19,57 +19,57 @@
 #ifndef UCI_H_INCLUDED
 #define UCI_H_INCLUDED
 
+#include <cstdint>
 #include <iostream>
 #include <string>
-#include <unordered_map>
+#include <string_view>
 
-#include "evaluate.h"
+#include "engine.h"
 #include "misc.h"
-#include "position.h"
+#include "search.h"
-#include "thread.h"
-#include "tt.h"
-#include "ucioption.h"
 
 namespace Stockfish {
 
-namespace Eval::NNUE {
+class Position;
-enum NetSize : int;
-}
-
 class Move;
+class Score;
 enum Square : int;
 using Value = int;
 
-class UCI {
+class UCIEngine {
    public:
-    UCI(int argc, char** argv);
+    UCIEngine(int argc, char** argv);
 
     void loop();
 
-    static int         to_cp(Value v);
+    static int         to_cp(Value v, const Position& pos);
-    static std::string value(Value v);
+    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, int ply);
+    static std::string wdl(Value v, const Position& pos);
-    static Move        to_move(const Position& pos, std::string& str);
+    static std::string to_lower(std::string str);
+    static Move        to_move(const Position& pos, std::string str);
 
-    const std::string& workingDirectory() const { return cli.workingDirectory; }
+    static Search::LimitsType parse_limits(std::istream& is);
 
-    OptionsMap options;
+    auto& engine_options() { return engine.get_options(); }
-
-    std::unordered_map<Eval::NNUE::NetSize, Eval::EvalFile> evalFiles;
 
    private:
-    TranspositionTable tt;
+    Engine      engine;
-    ThreadPool         threads;
+    CommandLine cli;
-    CommandLine        cli;
 
-    void go(Position& pos, std::istringstream& is, StateListPtr& states);
+    static void print_info_string(const std::string& str);
-    void bench(Position& pos, std::istream& args, StateListPtr& states);
+
-    void position(Position& pos, std::istringstream& is, StateListPtr& states);
+    void          go(std::istringstream& is);
-    void trace_eval(Position& pos);
+    void          bench(std::istream& args);
-    void search_clear();
+    void          position(std::istringstream& is);
-    void setoption(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);
 };
 
 }  // namespace Stockfish

src/ucioption.cpp

@@ -36,6 +36,8 @@ bool CaseInsensitiveLess::operator()(const std::string& s1, const std::string& s
       [](char c1, char c2) { return std::tolower(c1) < std::tolower(c2); });
 }
 
+void OptionsMap::add_info_listener(InfoListener&& message_func) { info = std::move(message_func); }
+
 void OptionsMap::setoption(std::istringstream& is) {
     std::string token, name, value;
 
@@ -57,13 +59,20 @@ void OptionsMap::setoption(std::istringstream& is) {
 
 Option OptionsMap::operator[](const std::string& name) const {
     auto it = options_map.find(name);
-    return it != options_map.end() ? it->second : Option();
+    return it != options_map.end() ? it->second : Option(this);
 }
 
-Option& OptionsMap::operator[](const std::string& name) { return options_map[name]; }
+Option& OptionsMap::operator[](const std::string& name) {
+    if (!options_map.count(name))
+        options_map[name] = Option(this);
+    return options_map[name];
+}
 
 std::size_t OptionsMap::count(const std::string& name) const { return options_map.count(name); }
 
+Option::Option(const OptionsMap* map) :
+    parent(map) {}
+
 Option::Option(const char* v, OnChange f) :
     type("string"),
     min(0),
@@ -118,6 +127,8 @@ bool Option::operator==(const char* s) const {
     return !CaseInsensitiveLess()(currentValue, s) && !CaseInsensitiveLess()(s, currentValue);
 }
 
+bool Option::operator!=(const char* s) const { return !(*this == s); }
+
 
 // Inits options and assigns idx in the correct printing order
 
@@ -125,10 +136,12 @@ void Option::operator<<(const Option& o) {
 
     static size_t insert_order = 0;
 
-    *this = o;
+    auto p = this->parent;
-    idx   = insert_order++;
+    *this  = o;
-}
 
+    this->parent = p;
+    idx          = insert_order++;
+}
 
 // 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
@@ -153,11 +166,18 @@ Option& Option::operator=(const std::string& v) {
             return *this;
     }
 
-    if (type != "button")
+    if (type == "string")
+        currentValue = v == "<empty>" ? "" : v;
+    else if (type != "button")
         currentValue = v;
 
     if (on_change)
-        on_change(*this);
+    {
+        const auto ret = on_change(*this);
+
+        if (ret && parent != nullptr && parent->info != nullptr)
+            parent->info(ret);
+    }
 
     return *this;
 }
@@ -170,10 +190,16 @@ std::ostream& operator<<(std::ostream& os, const OptionsMap& om) {
                 const Option& o = it.second;
                 os << "\noption name " << it.first << " type " << o.type;
 
-                if (o.type == "string" || o.type == "check" || o.type == "combo")
+                if (o.type == "check" || o.type == "combo")
                     os << " default " << o.defaultValue;
 
-                if (o.type == "spin")
+                else if (o.type == "string")
+                {
+                    std::string defaultValue = o.defaultValue.empty() ? "<empty>" : o.defaultValue;
+                    os << " default " << defaultValue;
+                }
+
+                else if (o.type == "spin")
                     os << " default " << int(stof(o.defaultValue)) << " min " << o.min << " max "
                        << o.max;
 

src/ucioption.h

@@ -23,6 +23,7 @@
 #include <functional>
 #include <iosfwd>
 #include <map>
+#include <optional>
 #include <string>
 
 namespace Stockfish {
@@ -31,31 +32,14 @@ struct CaseInsensitiveLess {
     bool operator()(const std::string&, const std::string&) const;
 };
 
-class Option;
+class OptionsMap;
-
-class OptionsMap {
-   public:
-    void setoption(std::istringstream&);
-
-    friend std::ostream& operator<<(std::ostream&, const OptionsMap&);
-
-    Option  operator[](const std::string&) const;
-    Option& operator[](const std::string&);
-
-    std::size_t count(const std::string&) const;
-
-   private:
-    // The options container is defined as a std::map
-    using OptionsStore = std::map<std::string, Option, CaseInsensitiveLess>;
-
-    OptionsStore options_map;
-};
 
 // The Option class implements each option as specified by the UCI protocol
 class Option {
    public:
-    using OnChange = std::function<void(const Option&)>;
+    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);
@@ -63,18 +47,57 @@ class Option {
     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;
+    bool operator!=(const char*) const;
 
     friend std::ostream& operator<<(std::ostream&, const OptionsMap&);
 
    private:
-    std::string defaultValue, currentValue, type;
+    friend class OptionsMap;
-    int         min, max;
+    friend class Engine;
-    size_t      idx;
+    friend class Tune;
-    OnChange    on_change;
+
+    void operator<<(const Option&);
+
+    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&);
+
+    Option  operator[](const std::string&) const;
+    Option& operator[](const std::string&);
+
+    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;
 };
 
 }

tests/instrumented.sh

@@ -14,14 +14,14 @@ case $1 in
     echo "valgrind testing started"
     prefix=''
     exeprefix='valgrind --error-exitcode=42 --errors-for-leak-kinds=all --leak-check=full'
-    postfix='1>/dev/null'
+    postfix=''
     threads="1"
   ;;
   --valgrind-thread)
     echo "valgrind-thread testing started"
     prefix=''
     exeprefix='valgrind --fair-sched=try --error-exitcode=42'
-    postfix='1>/dev/null'
+    postfix=''
     threads="2"
   ;;
   --sanitizer-undefined)
@@ -39,13 +39,8 @@ case $1 in
     threads="2"
 
 cat << EOF > tsan.supp
-race:Stockfish::TTEntry::move
+race:Stockfish::TTEntry::read
-race:Stockfish::TTEntry::depth
-race:Stockfish::TTEntry::bound
 race:Stockfish::TTEntry::save
-race:Stockfish::TTEntry::value
-race:Stockfish::TTEntry::eval
-race:Stockfish::TTEntry::is_pv
 
 race:Stockfish::TranspositionTable::probe
 race:Stockfish::TranspositionTable::hashfull
@@ -105,7 +100,12 @@ diff $network verify.nnue
 # more general testing, following an uci protocol exchange
 cat << EOF > game.exp
  set timeout 240
+ # to correctly catch eof we need the following line
+ # expect_before timeout { exit 2 } eof { exit 3 }
+ expect_before timeout { exit 2 }
+
  spawn $exeprefix ./stockfish
+ expect "Stockfish"
 
  send "uci\n"
  expect "uciok"
@@ -118,27 +118,106 @@ cat << EOF > game.exp
  send "go nodes 1000\n"
  expect "bestmove"
 
+ send "ucinewgame\n"
  send "position startpos moves e2e4 e7e6\n"
  send "go nodes 1000\n"
  expect "bestmove"
 
+ send "ucinewgame\n"
  send "position fen 5rk1/1K4p1/8/8/3B4/8/8/8 b - - 0 1\n"
  send "go depth 10\n"
  expect "bestmove"
 
- send "setoption name UCI_ShowWDL value true\n"
+ send "ucinewgame\n"
- send "position startpos\n"
+ send "position fen 5rk1/1K4p1/8/8/3B4/8/8/8 b - - 0 1\n"
  send "flip\n"
- send "go depth 5\n"
+ send "go depth 10\n"
  expect "bestmove"
 
- send "setoption name Skill Level value 10\n"
+ send "ucinewgame\n"
  send "position startpos\n"
  send "go depth 5\n"
+ expect -re {info depth \d+ seldepth \d+ multipv \d+ score cp \d+ nodes \d+ nps \d+ hashfull \d+ tbhits \d+ time \d+ pv}
+ expect "bestmove"
+
+ send "ucinewgame\n"
+ send "setoption name UCI_ShowWDL value true\n"
+ send "position startpos\n"
+ send "go depth 9\n"
+ expect -re {info depth 1 seldepth \d+ multipv \d+ score cp \d+ wdl \d+ \d+ \d+ nodes \d+ nps \d+ hashfull \d+ tbhits \d+ time \d+ pv}
+ expect -re {info depth 2 seldepth \d+ multipv \d+ score cp \d+ wdl \d+ \d+ \d+ nodes \d+ nps \d+ hashfull \d+ tbhits \d+ time \d+ pv}
+ expect -re {info depth 3 seldepth \d+ multipv \d+ score cp \d+ wdl \d+ \d+ \d+ nodes \d+ nps \d+ hashfull \d+ tbhits \d+ time \d+ pv}
+ expect -re {info depth 4 seldepth \d+ multipv \d+ score cp \d+ wdl \d+ \d+ \d+ nodes \d+ nps \d+ hashfull \d+ tbhits \d+ time \d+ pv}
+ expect -re {info depth 5 seldepth \d+ multipv \d+ score cp \d+ wdl \d+ \d+ \d+ nodes \d+ nps \d+ hashfull \d+ tbhits \d+ time \d+ pv}
+ expect -re {info depth 6 seldepth \d+ multipv \d+ score cp \d+ wdl \d+ \d+ \d+ nodes \d+ nps \d+ hashfull \d+ tbhits \d+ time \d+ pv}
+ expect -re {info depth 7 seldepth \d+ multipv \d+ score cp \d+ wdl \d+ \d+ \d+ nodes \d+ nps \d+ hashfull \d+ tbhits \d+ time \d+ pv}
+ expect -re {info depth 8 seldepth \d+ multipv \d+ score cp \d+ wdl \d+ \d+ \d+ nodes \d+ nps \d+ hashfull \d+ tbhits \d+ time \d+ pv}
+ expect -re {info depth 9 seldepth \d+ multipv \d+ score cp \d+ wdl \d+ \d+ \d+ nodes \d+ nps \d+ hashfull \d+ tbhits \d+ time \d+ pv}
  expect "bestmove"
 
  send "setoption name Clear Hash\n"
 
+ send "ucinewgame\n"
+ send "position fen 5K2/8/2qk4/2nPp3/3r4/6B1/B7/3R4 w - e6\n"
+ send "go depth 18\n"
+ expect "score mate 1"
+ expect "pv d5e6"
+ expect "bestmove d5e6"
+
+ send "ucinewgame\n"
+ send "position fen 2brrb2/8/p7/Q7/1p1kpPp1/1P1pN1K1/3P4/8 b - -\n"
+ send "go depth 18\n"
+ expect "score mate -1"
+ expect "bestmove"
+
+ send "ucinewgame\n"
+ send "position fen 7K/P1p1p1p1/2P1P1Pk/6pP/3p2P1/1P6/3P4/8 w - - 0 1\n"
+ send "go nodes 500000\n"
+ expect "bestmove"
+
+ send "ucinewgame\n"
+ send "position fen 8/5R2/2K1P3/4k3/8/b1PPpp1B/5p2/8 w - -\n"
+ send "go depth 18 searchmoves c6d7\n"
+ expect "score mate 2 * pv c6d7 * f7f5"
+ expect "bestmove c6d7"
+
+ send "ucinewgame\n"
+ send "position fen 8/5R2/2K1P3/4k3/8/b1PPpp1B/5p2/8 w - -\n"
+ send "go mate 2 searchmoves c6d7\n"
+ expect "score mate 2 * pv c6d7"
+ expect "bestmove c6d7"
+
+ send "ucinewgame\n"
+ send "position fen 8/5R2/2K1P3/4k3/8/b1PPpp1B/5p2/8 w - -\n"
+ send "go nodes 500000 searchmoves c6d7\n"
+ expect "score mate 2 * pv c6d7 * f7f5"
+ expect "bestmove c6d7"
+
+ send "ucinewgame\n"
+ send "position fen 1NR2B2/5p2/5p2/1p1kpp2/1P2rp2/2P1pB2/2P1P1K1/8 b - - \n"
+ send "go depth 27\n"
+ expect "score mate -2"
+ expect "pv d5e6 c8d8"
+ expect "bestmove d5e6"
+
+ send "ucinewgame\n"
+ send "position fen 8/5R2/2K1P3/4k3/8/b1PPpp1B/5p2/8 w - - moves c6d7 f2f1q\n"
+ send "go depth 18\n"
+ expect "score mate 1 * pv f7f5"
+ expect "bestmove f7f5"
+ 
+ send "ucinewgame\n"
+ send "position fen 8/5R2/2K1P3/4k3/8/b1PPpp1B/5p2/8 w - -\n"
+ send "go depth 18 searchmoves c6d7\n"
+ expect "score mate 2 * pv c6d7 * f7f5"
+ expect "bestmove c6d7"
+ 
+ send "ucinewgame\n"
+ send "position fen 8/5R2/2K1P3/4k3/8/b1PPpp1B/5p2/8 w - - moves c6d7\n"
+ send "go depth 18 searchmoves e3e2\n"
+ expect "score mate -1 * pv e3e2 f7f5"
+ expect "bestmove e3e2"
+
  send "setoption name EvalFile value verify.nnue\n"
  send "position startpos\n"
  send "go depth 5\n"
@@ -147,6 +226,13 @@ cat << EOF > game.exp
  send "setoption name MultiPV value 4\n"
  send "position startpos\n"
  send "go depth 5\n"
+ expect "bestmove"
+
+ send "setoption name Skill Level value 10\n"
+ send "position startpos\n"
+ send "go depth 5\n"
+ expect "bestmove"
+ send "setoption name Skill Level value 20\n"
 
  send "quit\n"
  expect eof
@@ -164,17 +250,30 @@ fi
 
 cat << EOF > syzygy.exp
  set timeout 240
+ # to correctly catch eof we need the following line
+ # expect_before timeout { exit 2 } eof { exit 3 }
+ expect_before timeout { exit 2 }
  spawn $exeprefix ./stockfish
+ expect "Stockfish"
  send "uci\n"
  send "setoption name SyzygyPath value ../tests/syzygy/\n"
- expect "info string Found 35 tablebases" {} timeout {exit 1}
+ expect "info string Found 35 WDL and 35 DTZ tablebase files (up to 4-man)."
  send "bench 128 1 8 default depth\n"
+ expect "Nodes searched  :"
  send "ucinewgame\n"
  send "position fen 4k3/PP6/8/8/8/8/8/4K3 w - - 0 1\n"
  send "go depth 5\n"
+ expect -re {score cp 20000|score mate}
  expect "bestmove"
+ send "ucinewgame\n"
  send "position fen 8/1P6/2B5/8/4K3/8/6k1/8 w - - 0 1\n"
  send "go depth 5\n"
+ expect -re {score cp 20000|score mate}
+ expect "bestmove"
+ send "ucinewgame\n"
+ send "position fen 8/1P6/2B5/8/4K3/8/6k1/8 b - - 0 1\n"
+ send "go depth 5\n"
+ expect -re {score cp -20000|score mate}
  expect "bestmove"
  send "quit\n"
  expect eof
@@ -187,6 +286,9 @@ EOF
 for exp in game.exp syzygy.exp
 do
 
+  echo "======== $exp =============="
+  cat $exp
+  echo "============================"
   echo "$prefix expect $exp $postfix"
   eval "$prefix expect $exp $postfix"