This is a version of the avx-512 kernel which uses ternary instructions and lower-level constructions. It is not faster.

Fix for ClangCL (Microsoft)
Updating the old CI tests.
2026-06-08 17:27:07 +00:00 · 2022-05-24 17:16:06 -04:00 · 2022-05-13 20:36:50 -04:00 · 2022-05-13 18:18:15 -04:00 · 2022-05-11 18:24:02 -04:00 · 2022-05-11 18:17:25 -04:00
113 changed files with 2712 additions and 814 deletions
@@ -1,46 +1,4 @@
 kind: pipeline
-name: i386-gcc # we do not support 32-bit systems, but we run tests
-platform: { os: linux, arch: amd64 }
-steps:
- name: Build and Test
-  image: i386/ubuntu
-  environment:
-    CC: gcc
-    CXX: g++
-    BUILD_FLAGS: -- -j
-    CMAKE_FLAGS: -DBUILD_SHARED_LIBS=OFF
-    CTEST_FLAGS: -j4 --output-on-failure -LE explicitonly 
-  commands:
-    - scripts/addcmakeppa.sh "$(env -i sh -c '. /etc/os-release; echo $VERSION_CODENAME')"
-    - apt-get install -y g++ cmake gcc git
-    - mkdir build
-    - cd build
-    - cmake $CMAKE_FLAGS ..
-    - cmake --build . $BUILD_FLAGS
-    - ctest $CTEST_FLAGS 
---
-kind: pipeline
-name: i386-clang # we do not support 32-bit systems, but we run tests
-platform: { os: linux, arch: amd64 }
-steps:
- name: Build and Test
-  image: i386/ubuntu
-  environment:
-    CC: clang-6.0
-    CXX: clang++-6.0
-    BUILD_FLAGS: -- -j
-    CMAKE_FLAGS: -DBUILD_SHARED_LIBS=OFF
-    CTEST_FLAGS: -j4 --output-on-failure -LE explicitonly 
-  commands:
-    - scripts/addcmakeppa.sh "$(env -i sh -c '. /etc/os-release; echo $VERSION_CODENAME')"
-    - apt-get install -y clang++-6.0 cmake git
-    - mkdir build
-    - cd build
-    - cmake $CMAKE_FLAGS ..
-    - cmake --build . $BUILD_FLAGS
-    - ctest $CTEST_FLAGS 
---
-kind: pipeline
 name: gcc9
 platform: { os: linux, arch: amd64 }
 steps:
@@ -50,7 +8,7 @@ steps:
    CC: gcc
    CXX: g++
    BUILD_FLAGS: -- -j
-    CMAKE_FLAGS: -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_IMPLEMENTATION=haswell;westmere;fallback
+    CMAKE_FLAGS: -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_IMPLEMENTATION=icelake;haswell;westmere;fallback
    CTEST_FLAGS: -j4 --output-on-failure -LE explicitonly
  commands:
    - echo "deb http://deb.debian.org/debian buster-backports main" >> /etc/apt/sources.list
@@ -61,6 +19,7 @@ steps:
    - cmake $CMAKE_FLAGS ..
    - cmake --build . $BUILD_FLAGS
    - ctest $CTEST_FLAGS -L acceptance -LE per_implementation
+    - SIMDJSON_FORCE_IMPLEMENTATION=icelake ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=haswell ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=westmere ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=fallback ctest $CTEST_FLAGS -L per_implementation
@@ -77,7 +36,7 @@ steps:
    CC: clang-6.0
    CXX: clang++-6.0
    BUILD_FLAGS: -- -j
-    CMAKE_FLAGS: -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_IMPLEMENTATION=haswell;westmere;fallback
+    CMAKE_FLAGS: -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_IMPLEMENTATION=icelake;haswell;westmere;fallback
    CTEST_FLAGS: -j4 --output-on-failure -LE explicitonly
  commands:
    - mkdir build
@@ -85,6 +44,7 @@ steps:
    - cmake $CMAKE_FLAGS ..
    - cmake --build . $BUILD_FLAGS
    - ctest $CTEST_FLAGS -L acceptance -LE per_implementation
+    - SIMDJSON_FORCE_IMPLEMENTATION=icelake ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=haswell ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=westmere ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=fallback ctest $CTEST_FLAGS -L per_implementation
@@ -142,7 +102,7 @@ steps:
    CC: gcc
    CXX: g++
    BUILD_FLAGS: -- -j
-    CMAKE_FLAGS: -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_IMPLEMENTATION=haswell;westmere;fallback
+    CMAKE_FLAGS: -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_IMPLEMENTATION=icelake;haswell;westmere;fallback
    CTEST_FLAGS: -j4 --output-on-failure -LE explicitonly
  commands:
    - echo "deb http://deb.debian.org/debian buster-backports main" >> /etc/apt/sources.list
@@ -153,6 +113,7 @@ steps:
    - cmake $CMAKE_FLAGS ..
    - cmake --build . $BUILD_FLAGS
    - ASAN_OPTIONS="detect_leaks=0" ctest $CTEST_FLAGS -L acceptance -LE per_implementation
+    - SIMDJSON_FORCE_IMPLEMENTATION=icelake ASAN_OPTIONS="detect_leaks=0" ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=haswell ASAN_OPTIONS="detect_leaks=0" ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=westmere ASAN_OPTIONS="detect_leaks=0" ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=fallback ASAN_OPTIONS="detect_leaks=0" ctest $CTEST_FLAGS -L per_implementation
@@ -168,7 +129,7 @@ steps:
  environment:
    CC: clang-9
    CXX: clang++-9
-    CMAKE_FLAGS: -DSIMDJSON_SANITIZE=ON -DSIMDJSON_IMPLEMENTATION=haswell;westmere;fallback
+    CMAKE_FLAGS: -DSIMDJSON_SANITIZE=ON -DSIMDJSON_IMPLEMENTATION=icelake;haswell;westmere;fallback
    BUILD_FLAGS: -- -j
    CTEST_FLAGS: -j4 --output-on-failure -LE explicitonly
  commands:
@@ -177,6 +138,7 @@ steps:
    - cmake $CMAKE_FLAGS ..
    - cmake --build . $BUILD_FLAGS
    - ASAN_OPTIONS="detect_leaks=0" ctest $CTEST_FLAGS -L acceptance -LE per_implementation
+    - SIMDJSON_FORCE_IMPLEMENTATION=icelake ASAN_OPTIONS="detect_leaks=0" ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=haswell ASAN_OPTIONS="detect_leaks=0" ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=westmere ASAN_OPTIONS="detect_leaks=0" ctest $CTEST_FLAGS -L per_implementation
    - SIMDJSON_FORCE_IMPLEMENTATION=fallback ASAN_OPTIONS="detect_leaks=0" ctest $CTEST_FLAGS -L per_implementation
@@ -0,0 +1,24 @@
+name: CIFuzz
+on: [pull_request]
+jobs:
+  Fuzzing:
+    runs-on: ubuntu-latest
+    steps:
+    - name: Build Fuzzers
+      id: build
+      uses: google/oss-fuzz/infra/cifuzz/actions/build_fuzzers@master
+      with:
+        oss-fuzz-project-name: 'simdjson'
+        dry-run: false
+    - name: Run Fuzzers
+      uses: google/oss-fuzz/infra/cifuzz/actions/run_fuzzers@master
+      with:
+        oss-fuzz-project-name: 'simdjson'
+        fuzz-seconds: 600
+        dry-run: false
+    - name: Upload Crash
+      uses: actions/upload-artifact@v1
+      if: failure() && steps.build.outcome == 'success'
+      with:
+        name: artifacts
+        path: ./out/artifacts
@@ -16,7 +16,7 @@ jobs:
          key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
      - name: Use cmake
        run: |
-          xcversion select 11.2.1
+          xcversion select 11.7
          mkdir builddebug &&
          cd builddebug &&
          cmake -DCMAKE_BUILD_TYPE=Debug -DSIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF ..  &&
@@ -1,63 +0,0 @@
-name: MinGW32-CI
-
-on: [push, pull_request]
-
-
-# Important: scoop will either install 32-bit GCC or 64-bit GCC, not both.
-
-# It is important to build static libraries because cmake is not smart enough under Windows/mingw to take care of the path. So
-# with a dynamic library, you could get failures due to the fact that the EXE can't find its DLL.
-
-jobs:
-  ci:
-    if: >-
-      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
-      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    name: windows-gcc
-    runs-on: windows-2016
-
-    env:
-      CMAKE_GENERATOR: Ninja # This is critical, try ' cmake  -GNinja-DBUILD_SHARED_LIBS=OFF .. ' if using the command line
-      CC: gcc
-      CXX: g++
-
-    steps: # To reproduce what is below, start a powershell with administrative rights, using scoop *is* a good idea
-      - uses: actions/checkout@v2
-
-      - uses: actions/cache@v2 # we cache the scoop setup with 32-bit GCC
-        id: cache
-        with:
-          path: |
-            C:\ProgramData\scoop
-          key: scoop32 # static key: should be good forever
-
-      - uses: actions/cache@v2
-        with:
-          path: dependencies/.cache
-          key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
-
-      - name: Setup Windows # This should almost never run if the cache works.
-        if: steps.cache.outputs.cache-hit != 'true'
-        shell: powershell
-        run: |
-          Invoke-Expression (New-Object System.Net.WebClient).DownloadString('https://get.scoop.sh')
-          scoop install sudo --global
-          sudo scoop install git --global
-          sudo scoop install ninja --global
-          sudo scoop install cmake --global
-          sudo scoop install gcc --arch 32bit --global
-          $env:path
-          Write-Host 'Everything has been installed, you are good!'
-      - name: Build and Test 32-bit x86
-        shell: powershell
-        run: |
-          $ENV:PATH="C:\ProgramData\scoop\shims;C:\ProgramData\scoop\apps\gcc\current\bin;C:\ProgramData\scoop\apps\ninja\current;$ENV:PATH"
-          g++ --version
-          cmake --version
-          ninja --version
-          git --version
-          mkdir build32
-          cd build32
-          cmake -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_COMPETITION=OFF -DSIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_ENABLE_THREADS=OFF ..
-          cmake --build . --target acceptance_tests --verbose
-          ctest -L acceptance --output-on-failure
@@ -1,69 +0,0 @@
-name: MinGW64-CI
-
-on: [push, pull_request]
-
-
-# Important: scoop will either install 32-bit GCC or 64-bit GCC, not both.
-
-# It is important to build static libraries because cmake is not smart enough under Windows/mingw to take care of the path. So
-# with a dynamic library, you could get failures due to the fact that the EXE can't find its DLL.
-
-jobs:
-  ci:
-    if: >-
-      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
-      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    name: windows-gcc
-    runs-on: windows-2016
-
-    env:
-      CMAKE_GENERATOR: Ninja # This is critical, try ' cmake  -GNinja-DBUILD_SHARED_LIBS=OFF .. ' if using the command line
-      CC: gcc
-      CXX: g++
-
-    steps: # To reproduce what is below, start a powershell with administrative rights, using scoop *is* a good idea
-      - uses: actions/checkout@v2
-
-      - uses: actions/cache@v2 # we cache the scoop setup with 64-bit GCC
-        id: cache
-        with:
-          path: |
-            C:\ProgramData\scoop
-          key: scoop64 # static key: should be good forever
-
-      - uses: actions/cache@v2
-        with:
-          path: dependencies/.cache
-          key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
-
-      - name: Setup Windows # This should almost never run if the cache works.
-        if: steps.cache.outputs.cache-hit != 'true'
-        shell: powershell
-        run: |
-          Invoke-Expression (New-Object System.Net.WebClient).DownloadString('https://get.scoop.sh')
-          scoop install sudo --global
-          sudo scoop install git --global
-          sudo scoop install ninja --global
-          sudo scoop install cmake --global
-          sudo scoop install gcc --arch 64bit --global
-          $env:path
-          Write-Host 'Everything has been installed, you are good!'
-      - name: Build and Test 64-bit x64
-        shell: powershell
-        run: |
-          $ENV:PATH="C:\ProgramData\scoop\shims;C:\ProgramData\scoop\apps\gcc\current\bin;C:\ProgramData\scoop\apps\ninja\current;$ENV:PATH"
-          g++ --version
-          cmake --version
-          ninja --version
-          git --version
-          mkdir build64
-          cd build64
-          cmake -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_COMPETITION=OFF -DSIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_ENABLE_THREADS=OFF ..
-          cmake --build . --target acceptance_tests --verbose
-          ctest -L acceptance --output-on-failure
-          cd ..
-          mkdir build64debug
-          cd build64debug
-          cmake -DSIMDJSON_DEVELOPER_MODE=ON -DCMAKE_BUILD_TYPE=Debug -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_COMPETITION=OFF -DSIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_ENABLE_THREADS=OFF ..
-          cmake --build . --target acceptance_tests --verbose
-          ctest -L acceptance --output-on-failure
@@ -17,15 +17,9 @@ jobs:
          - msystem: "MINGW64"
            install: mingw-w64-x86_64-libxml2 mingw-w64-x86_64-cmake mingw-w64-x86_64-ninja mingw-w64-x86_64-clang
            type: Release
-          - msystem: "MINGW32"
-            install: mingw-w64-i686-libxml2 mingw-w64-i686-cmake mingw-w64-i686-ninja mingw-w64-i686-clang
-            type: Release
          - msystem: "MINGW64"
            install: mingw-w64-x86_64-libxml2 mingw-w64-x86_64-cmake mingw-w64-x86_64-ninja mingw-w64-x86_64-clang
            type: Debug
-          - msystem: "MINGW32"
-            install: mingw-w64-i686-libxml2 mingw-w64-i686-cmake mingw-w64-i686-ninja mingw-w64-i686-clang
-            type: Debug
    env:
      CMAKE_GENERATOR: Ninja

@@ -44,6 +38,6 @@ jobs:
        run: |
          mkdir build
          cd build
-          cmake -DSIMDJSON_DEVELOPER_MODE=ON -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=${{ matrix.type }} -DSIMDJSON_BUILD_STATIC=ON -DSIMDJSON_DO_NOT_USE_THREADS_NO_MATTER_WHAT=ON ..
+          cmake -DSIMDJSON_DEVELOPER_MODE=ON -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=${{ matrix.type }} -DBUILD_SHARED_LIBS=OFF -DSIMDJSON_DO_NOT_USE_THREADS_NO_MATTER_WHAT=ON ..
          cmake --build . --verbose
          ctest -j4 --output-on-failure -LE explicitonly
@@ -19,15 +19,9 @@ jobs:
          - msystem: "MINGW64"
            install: mingw-w64-x86_64-cmake mingw-w64-x86_64-ninja mingw-w64-x86_64-gcc
            type: Release
-          - msystem: "MINGW32"
-            install: mingw-w64-i686-cmake mingw-w64-i686-ninja mingw-w64-i686-gcc
-            type: Release
          - msystem: "MINGW64"
            install: mingw-w64-x86_64-cmake mingw-w64-x86_64-ninja mingw-w64-x86_64-gcc
            type: Debug
-          - msystem: "MINGW32"
-            install: mingw-w64-i686-cmake mingw-w64-i686-ninja mingw-w64-i686-gcc
-            type: Debug
    env:
      CMAKE_GENERATOR: Ninja

@@ -0,0 +1,33 @@
+name: Ubuntu 22.04 CI (GCC 11)
+
+on: [push, pull_request]
+
+jobs:
+  ubuntu-build:
+    if: >-
+      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
+      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v2
+      - uses: actions/cache@v2
+        with:
+          path: dependencies/.cache
+          key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
+      - name: Use cmake
+        run: |
+          mkdir builddebug &&
+          cd builddebug &&
+          cmake -DCMAKE_BUILD_TYPE=Debug -DSIMDJSON_GOOGLE_BENCHMARKS=OFF -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF ..  &&
+          cmake --build .   &&
+          ctest -j --output-on-failure -LE explicitonly   &&
+          cd .. &&
+          mkdir build &&
+          cd build &&
+          cmake  -DSIMDJSON_GOOGLE_BENCHMARKS=ON -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF -DCMAKE_INSTALL_PREFIX:PATH=destination ..  &&
+          cmake --build .   &&
+          ctest -j --output-on-failure -LE explicitonly   &&
+          cmake --install . &&
+          echo -e '#include <simdjson.h>\nint main(int argc,char**argv) {simdjson::dom::parser parser;simdjson::dom::element tweets = parser.load(argv[1]); }' > tmp.cpp && c++ -Idestination/include -Ldestination/lib -std=c++17 -Wl,-rpath,destination/lib -o linkandrun tmp.cpp -lsimdjson && ./linkandrun jsonexamples/twitter.json &&
+          cd ../tests/installation_tests/find &&
+          mkdir build && cd build && cmake -DCMAKE_INSTALL_PREFIX:PATH=../../../build/destination .. &&  cmake --build .
@@ -1,31 +0,0 @@
-name: VS16-CLANG-CI
-
-on: [push, pull_request]
-
-jobs:
-  ci:
-    if: >-
-      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
-      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    name: windows-vs16
-    runs-on: windows-latest
-    steps:
-    - uses: actions/checkout@v2
-    - uses: actions/cache@v2
-      with:
-        path: dependencies/.cache
-        key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
-    - name: 'Run CMake with VS16 Clang'
-      uses: lukka/run-cmake@v3
-      with:
-        cmakeListsOrSettingsJson: CMakeListsTxtAdvanced
-        cmakeListsTxtPath: '${{ github.workspace }}/CMakeLists.txt'
-        buildDirectory: "${{ github.workspace }}/../../_temp/windows"
-        cmakeBuildType: Release
-        buildWithCMake: true
-        cmakeAppendedArgs:  -T ClangCL -DSIMDJSON_COMPETITION=OFF -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF
-        buildWithCMakeArgs: --config Release
-
-    - name: 'Run CTest'
-      run: ctest -C Release -LE explicitonly  --output-on-failure
-      working-directory: "${{ github.workspace }}/../../_temp/windows"
@@ -1,45 +0,0 @@
-name: VS16-Ninja-CI
-
-on: [push, pull_request]
-
-jobs:
-  ci:
-    if: >-
-      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
-      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    name: windows-vs16
-    runs-on: windows-latest
-    steps:
-    - uses: actions/checkout@v2
-    - uses: actions/cache@v2
-      with:
-        path: dependencies/.cache
-        key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
-    - name: 'Run CMake with VS16'
-      uses: lukka/run-cmake@v2
-      with:
-        cmakeListsOrSettingsJson: CMakeListsTxtAdvanced
-        cmakeListsTxtPath: '${{ github.workspace }}/CMakeLists.txt'
-        buildDirectory: "${{ github.workspace }}/../../_temp/windows"
-        cmakeBuildType: Release
-        buildWithCMake: true
-        cmakeAppendedArgs:  -G Ninja -DSIMDJSON_COMPETITION=OFF -DSIMDJSON_DEVELOPER_MODE=ON -DBUILD_SHARED_LIBS=OFF
-        buildWithCMakeArgs: --config Release
-    - name: 'Run CTest'
-      run: ctest -C Release  -LE explicitonly  --output-on-failure
-      working-directory: "${{ github.workspace }}/../../_temp/windows"
-    - name: 'Install with CMake'
-      uses: lukka/run-cmake@v3
-      with:
-        cmakeListsTxtPath: '${{ github.workspace }}/CMakeLists.txt'
-        buildWithCMakeArgs: '--target install'
-    - name: 'Test Installation with CMake'
-      uses: lukka/run-cmake@v3
-      with:
-        cmakeListsOrSettingsJson: CMakeListsTxtAdvanced
-        cmakeListsTxtPath: '${{ github.workspace }}/tests/installation_tests/find/CMakeLists.txt'
-        cmakeBuildType: Release
-        buildWithCMake: true
-        buildDirectory: '${{ github.workspace }}/tests/installation_tests/find/buildDirectory'
-        cmakeAppendedArgs:  -G Ninja
-        buildWithCMakeArgs: '--config Release --verbose'
@@ -1,10 +1,10 @@
-name: VS16-ARM-CI
+name: VS17-ARM-CI

 on: [push, pull_request]

 jobs:
   ci:
-     name: windows-vs16
+     name: windows-vs17
     runs-on: windows-latest
     strategy:
       fail-fast: false
@@ -1,4 +1,4 @@
-name: VS16-CI
+name: VS17-CI

 on: [push, pull_request]

@@ -7,22 +7,22 @@ jobs:
    if: >-
      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    name: windows-vs16
+    name: windows-vs17
    runs-on: windows-latest
    strategy:
      fail-fast: false
      matrix:
        include:
-          - {gen: Visual Studio 16 2019, arch: Win32, static: ON}
-          - {gen: Visual Studio 16 2019, arch: Win32, static: OFF}
-          - {gen: Visual Studio 16 2019, arch: x64, static: ON}
-          - {gen: Visual Studio 16 2019, arch: x64, static: OFF}
+          - {gen: Visual Studio 17 2022, arch: Win32, shared: ON}
+          - {gen: Visual Studio 17 2022, arch: Win32, shared: OFF}
+          - {gen: Visual Studio 17 2022, arch: x64, shared: ON}
+          - {gen: Visual Studio 17 2022, arch: x64, shared: OFF}
    steps:
    - name: checkout
      uses: actions/checkout@v2
    - name: Configure
      run: |
-        cmake -G "${{matrix.gen}}" -A ${{matrix.arch}} -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_COMPETITION=OFF -DSIMDJSON_BUILD_STATIC=${{matrix.static}} -B build
+        cmake -G "${{matrix.gen}}" -A ${{matrix.arch}} -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_COMPETITION=OFF -DBUILD_SHARED_LIBS=${{matrix.shared}} -B build
    - name: Build Debug
      run: cmake --build build --config Debug --verbose
    - name: Build Release
@@ -1,4 +1,4 @@
-name: VS15-CI
+name: VS17-CLANG-CI

 on: [push, pull_request]

@@ -7,22 +7,19 @@ jobs:
    if: >-
      ! contains(toJSON(github.event.commits.*.message), '[skip ci]') &&
      ! contains(toJSON(github.event.commits.*.message), '[skip github]')
-    name: windows-vs15
-    runs-on: windows-2016
+    name: windows-vs17
+    runs-on: windows-latest
    strategy:
      fail-fast: false
      matrix:
        include:
-          - {gen: Visual Studio 15 2017, arch: Win32, static: ON}
-          - {gen: Visual Studio 15 2017, arch: Win32, static: OFF}
-          - {gen: Visual Studio 15 2017, arch: x64, static: ON}
-          - {gen: Visual Studio 15 2017, arch: x64, static: OFF}
+          - {gen: Visual Studio 17 2022, arch: x64}
    steps:
    - name: checkout
      uses: actions/checkout@v2
    - name: Configure
      run: |
-        cmake -G "${{matrix.gen}}" -A ${{matrix.arch}} -DSIMDJSON_COMPETITION=OFF -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_BUILD_STATIC=${{matrix.static}} -B build
+        cmake -G "${{matrix.gen}}" -A ${{matrix.arch}}  -T ClangCL -DSIMDJSON_DEVELOPER_MODE=ON -DSIMDJSON_COMPETITION=OFF -B build
    - name: Build Debug
      run: cmake --build build --config Debug --verbose
    - name: Build Release
@@ -1,10 +1,10 @@
-name: VS16-NoExcept-CI
+name: VS17-NoExcept-CI

 on: [push, pull_request]

 jobs:
  ci:
-    name: windows-vs16
+    name: windows-vs17
    runs-on: windows-latest
    steps:
    - uses: actions/checkout@v2
@@ -12,7 +12,7 @@ jobs:
      with:
        path: dependencies/.cache
        key: ${{ hashFiles('dependencies/CMakeLists.txt') }}
-    - name: 'Run CMake with VS16'
+    - name: 'Run CMake with VS17'
      uses: lukka/run-cmake@v3
      with:
        cmakeListsOrSettingsJson: CMakeListsTxtAdvanced
@@ -3,7 +3,7 @@ cmake_minimum_required(VERSION 3.14)
 project(
    simdjson
    # The version number is modified by tools/release.py
-    VERSION 1.0.0
+    VERSION 1.0.2
    DESCRIPTION "Parsing gigabytes of JSON per second"
    HOMEPAGE_URL "https://simdjson.org/"
    LANGUAGES CXX C
@@ -49,19 +49,10 @@ endif()
 if(is_top_project)
  option(SIMDJSON_DEVELOPER_MODE "Enable targets for developing simdjson" OFF)
  option(BUILD_SHARED_LIBS "Build simdjson as a shared library" OFF)
-
-  if("$ENV{CI}")
-    set(SIMDJSON_DEVELOPER_MODE ON CACHE INTERNAL "")
-  endif()
 endif()

 include(cmake/handle-deprecations.cmake)
-
-if(SIMDJSON_DEVELOPER_MODE)
-  include(cmake/developer-options.cmake)
-else()
-  message(STATUS "Building only the library. Advanced users may want to turn SIMDJSON_DEVELOPER_MODE to ON, e.g., via -D SIMDJSON_DEVELOPER_MODE=ON.")
-endif()
+include(cmake/developer-options.cmake)

 # ---- simdjson library ----

@@ -180,6 +171,7 @@ endif()
 # ---- Developer mode extras ----

 if(NOT SIMDJSON_DEVELOPER_MODE)
+  message(STATUS "Building only the library. Advanced users may want to turn SIMDJSON_DEVELOPER_MODE to ON, e.g., via -D SIMDJSON_DEVELOPER_MODE=ON.")
  return()
 elseif(NOT is_top_project)
  message(AUTHOR_WARNING "Developer mode is intended for developers of simdjson")
@@ -37,5 +37,9 @@ Furkan Taşkale
 Brendan Knapp
 Danila Kutenin
 Pavel Pavlov
-# if you have contributed to the project and your name does not 
+Hao Chen
+Nicolas Boyer
+Kim Walisch and Jatin Bhateja (AVX-512 bitset decoder)
+Eric Zhang (AVX-512 kernel)
+# if you have contributed to the project and your name does not
 # appear in this list, please let us know!
@@ -38,7 +38,7 @@ PROJECT_NAME           = simdjson
 # could be handy for archiving the generated documentation or if some version
 # control system is used.

-PROJECT_NUMBER         = "1.0.0"
+PROJECT_NUMBER         = "1.0.2"

 # Using the PROJECT_BRIEF tag one can provide an optional one line description
 # for a project that appears at the top of each page and should give viewer a
@@ -138,6 +138,7 @@ We distinguish between "bindings" (which just wrap the C++ code) and a port to a
 - [rcppsimdjson](https://github.com/eddelbuettel/rcppsimdjson): R bindings.
 - [simdjson_erlang](https://github.com/ChomperT/simdjson_erlang): erlang bindings.
 - [lua-simdjson](https://github.com/FourierTransformer/lua-simdjson): lua bindings.
+- [hermes-json](https://hackage.haskell.org/package/hermes-json): haskell bindings.


 About simdjson
@@ -97,9 +97,9 @@ struct option_struct {
        verbose = true;
        break;
      case 'a': {
-          auto impl = simdjson::available_implementations[optarg];
+          auto impl = simdjson::get_available_implementations()[optarg];
          if(impl && impl->supported_by_runtime_system()) {
-            simdjson::active_implementation = impl;
+            simdjson::get_active_implementation() = impl;
          } else {
            std::cerr << "implementation " << optarg << " not found or not supported " << std::endl;
          }
@@ -66,7 +66,7 @@ void print_usage(ostream& out) {
  out << "-H           - Make the buffers hot (reduce page allocation and related OS tasks during parsing) [default]" << endl;
  out << "-a IMPL      - Use the given parser implementation. By default, detects the most advanced" << endl;
  out << "               implementation supported on the host machine." << endl;
-  for (auto impl : simdjson::available_implementations) {
+  for (auto impl : simdjson::get_available_implementations()) {
    if(impl->supported_by_runtime_system()) {
      out << "-a " << std::left << std::setw(9) << impl->name() << " - Use the " << impl->description() << " parser implementation." << endl;
    }
@@ -116,10 +116,10 @@ struct option_struct {
        verbose = true;
        break;
      case 'a': {
-        const implementation *impl = simdjson::available_implementations[optarg];
+        const implementation *impl = simdjson::get_available_implementations()[optarg];
        if ((!impl) || (!impl->supported_by_runtime_system())) {
          std::string exit_message = string("Unsupported option value -a ") + optarg + ": expected -a  with one of ";
-          for (auto imple : simdjson::available_implementations) {
+          for (auto imple : simdjson::get_available_implementations()) {
            if(imple->supported_by_runtime_system()) {
              exit_message += imple->name();
              exit_message += " ";
@@ -127,7 +127,7 @@ struct option_struct {
          }
          exit_usage(exit_message);
        }
-        simdjson::active_implementation = impl;
+        simdjson::get_active_implementation() = impl;
        break;
      }
      case 'C':
@@ -175,7 +175,7 @@ int main(int argc, char *argv[]) {
  option_struct options(argc, argv);
  if (options.verbose) {
    verbose_stream = &cout;
-    verbose() << "Implementation: " << simdjson::active_implementation->name() << endl;
+    verbose() << "Implementation: " << simdjson::get_active_implementation()->name() << endl;
  }

  // Start collecting events. We put this early so if it prints an error message, it's the
@@ -50,9 +50,9 @@ simdjson_really_inline void simdjson_process_atom(stat_t &s,
  } else if(element.is<double>()) {
    s.float_count++;
  } else if (element.is<bool>()) {
-    simdjson::error_code err;
    bool v;
-    err = element.get(v);
+    simdjson::error_code error;
+    if ((error = element.get(v))) { std::cerr << error << std::endl; abort(); }
    if (v) {
      s.true_count++;
    } else {
@@ -10,8 +10,8 @@ void maybe_display_implementation() {
  static bool displayed_implementation = false;
  if(!displayed_implementation) {
    displayed_implementation = true;
-    std::cout << "simdjson::dom implementation:      " << simdjson::active_implementation->name() << std::endl;
-    std::cout << "simdjson::ondemand implementation (stage 1): " << simdjson::active_implementation->name() << std::endl;
+    std::cout << "simdjson::dom implementation:      " << simdjson::get_active_implementation()->name() << std::endl;
+    std::cout << "simdjson::ondemand implementation (stage 1): " << simdjson::get_active_implementation()->name() << std::endl;
    std::cout << "simdjson::ondemand implementation (stage 2): " << simdjson::builtin_implementation()->name() << std::endl;
  }
 }
@@ -175,6 +175,18 @@ if(CMAKE_C_COMPILER_ID MATCHES "Intel")
  set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -static-intel")
 endif()

+option(
+    SIMDJSON_AVX512_ALLOWED
+    "Enable AVX-512 instructions (only affects processors and compilers with AVX-512 support)."
+    ON
+)
+if(SIMDJSON_AVX512_ALLOWED)
+  add_compile_definitions(SIMDJSON_AVX512_ALLOWED=1)
+else()
+  add_compile_definitions(SIMDJSON_AVX512_ALLOWED=0)
+  message(STATUS "AVX-512 instructions are not allowed.")
+endif()
+
 include(CheckSymbolExists)
 check_symbol_exists(fork unistd.h HAVE_POSIX_FORK)
 check_symbol_exists(wait sys/wait.h HAVE_POSIX_WAIT)
@@ -1,7 +1,7 @@
 #
 # Implementation selection
 #
-set(SIMDJSON_ALL_IMPLEMENTATIONS fallback westmere haswell arm64 ppc64)
+set(SIMDJSON_ALL_IMPLEMENTATIONS fallback westmere haswell icelake arm64 ppc64)

 set(
    SIMDJSON_IMPLEMENTATION ""
@@ -15,7 +15,7 @@ set(
    SIMDJSON_EXCLUDE_IMPLEMENTATION ""
    CACHE STRING "\
 Semicolon-separated list of implementations to exclude \
-(haswell/westmere/arm64/ppc64/fallback). By default, excludes any \
+(icelake/haswell/westmere/arm64/ppc64/fallback). By default, excludes any \
 implementations that are unsupported at compile time or cannot be selected at \
 runtime."
 )
@@ -13,8 +13,9 @@ cmake_dependent_option(SIMDJSON_GOOGLE_BENCHMARKS "compile the Google Benchmark
 if(SIMDJSON_GOOGLE_BENCHMARKS)
  set_off(BENCHMARK_ENABLE_TESTING)
  set_off(BENCHMARK_ENABLE_INSTALL)
+  set_off(BENCHMARK_ENABLE_WERROR)

-  import_dependency(google_benchmarks google/benchmark 8982e1e)
+  import_dependency(google_benchmarks google/benchmark f91b6b4)
  add_dependency(google_benchmarks)
 endif()

@@ -16,10 +16,10 @@ An overview of what you need to know to use simdjson, with examples.
 * [UTF-8 validation (alone)](#utf-8-validation-alone)
 * [JSON Pointer](#json-pointer)
 * [Error Handling](#error-handling)
-  * [Error Handling Example without Exceptions](#error-handling-example-without-exceptions)
+  * [Error Handling Example without Exceptions](#error-handling-examples-without-exceptions)
  * [Disabling Exceptions](#disabling-exceptions)
  * [Exceptions](#exceptions)
-  * [Current location in document](#current-location-in-documnet)
+  * [Current location in document](#current-location-in-document)
 * [Rewinding](#rewinding)
 * [Direct Access to the Raw String](#direct-access-to-the-raw-string)
 * [Newline-Delimited JSON (ndjson) and JSON lines](#newline-delimited-json-ndjson-and-json-lines)
@@ -54,6 +54,7 @@ c++ myproject.cpp simdjson.cpp

 Note:
 - Users on macOS and other platforms where default compilers do not provide C++11 compliant by default should request it with the appropriate flag (e.g., `c++ -std=c++17 myproject.cpp simdjson.cpp`).
+- The library relies on [runtime CPU detection](implementation-selection.md): avoid specifying an architecture at compile time (e.g., `-march-native`).

 Using simdjson with package managers
 ------------------
@@ -127,19 +128,23 @@ The Basics: Loading and Parsing JSON Documents
 ----------------------------------------------

 The simdjson library allows you to navigate and validate JSON documents ([RFC 8259](https://www.tbray.org/ongoing/When/201x/2017/12/14/rfc8259.html)).
-As required by the standard, your JSON document should be Unicode (UTF-8) strings.
+As required by the standard, your JSON document should be in a Unicode (UTF-8) string. The whole
+string, from the beginning to the end, needs to be valid: we do not attempt to tolerate bad
+inputs before or after a document.

-
-The simdjson library offers a tree-like [API](https://en.wikipedia.org/wiki/API), which you can access by creating a `ondemand::parser` and calling the `iterate()` method:
+The simdjson library offers a tree-like [API](https://en.wikipedia.org/wiki/API), which you can
+access by creating a `ondemand::parser` and calling the `iterate()` method. The iterate method
+quickly indexes the input string and may detect some errors. The following example illustrates
+how to get started with an input JSON file (`"twitter.json"`):

 ```c++
 ondemand::parser parser;
-auto json = padded_string::load("twitter.json");
+auto json = padded_string::load("twitter.json"); // load JSON file 'twitter.json'.
 ondemand::document doc = parser.iterate(json); // position a pointer at the beginning of the JSON data
 ```

-Or by creating a padded string---for efficiency reasons, simdjson requires a string with a few
- bytes (`simdjson::SIMDJSON_PADDING`) at the end---and calling `iterate()`:
+You can also create a padded string---for efficiency reasons, simdjson requires a string
+with a few bytes (`simdjson::SIMDJSON_PADDING`) at the end---and calling `iterate()`:

 ```c++
 ondemand::parser parser;
@@ -196,7 +201,8 @@ allocations during parsing when using simdjson. [See our performance notes for d
 C++11 Support and string_view
 -------------

-The simdjson library builds on compilers supporting the [C++11 standard](https://en.wikipedia.org/wiki/C%2B%2B11). It is also a strict requirement: we have no plan to support older C++ compilers.
+The simdjson library builds on compilers supporting the [C++11 standard](https://en.wikipedia.org/wiki/C%2B%2B11).
+It is also a strict requirement: we have no plan to support older C++ compilers.

 We represent parsed Unicode (UTF-8) strings in simdjson using the `std::string_view` class. It avoids
 the need to copy the data, as would be necessary with the `std::string` class. It also
@@ -214,7 +220,6 @@ is often best viewed as a temporary string value that is tied to the document yo
 At the cost of some memory allocation, you may convert your `std::string_view` instances for long-term storage into `std::string` instances:
 `std::string mycopy(view)` (C++17) or  `std::string mycopy(view.begin(), view.end())` (prior to C++17).

-
 The `std::string_view` class has become standard as part of C++17 but it is not always available
 on compilers which only supports C++11. When we detect that `string_view` is natively
 available, we define the macro `SIMDJSON_HAS_STRING_VIEW`.
@@ -232,24 +237,38 @@ transcode the UTF-8 strings produced by the simdjson library to other formats. S
 Using the Parsed JSON
 ---------------------

+
+
+We recommend that you first compile and run your code in Debug mode (with `NDEBUG`
+undefined). When you do so, the simdjson library runs additional sanity tests on
+your code to help ensure that you are using the library in a safe manner. Once
+your code has been tested, you can then run it in Release mode (with `NDEBUG`
+defined) for best performance. Alternatively, you can set the macro
+`SIMDJSON_DEVELOPMENT_CHECKS` to 1 prior to including the `simdjson.h` header
+to enable these additional checks: just make sure you remove the definition once your
+code has been tested.
+
 Once you have a document (`simdjson::ondemand::document`), you can navigate it with
-idiomatic C++ iterators, operators and casts. Besides the documents instances and
+idiomatic C++ iterators, operators and casts. Besides the document instances and
 native types (`double`, `uint64_t`, `int64_t`, `bool`), we also access
 Unicode (UTF-8) strings (`std::string_view`), objects (`simdjson::ondemand::object`)
 and arrays (`simdjson::ondemand::array`).
-We also have a generic type (`simdjson::ondemand::value`) which represent a potential
-array or object, or scalar type (`double`, `uint64_t`, `int64_t`, `bool`, `null`, string) inside an array or an object. Both generic types (`simdjson::ondemand::document` and `simdjson::ondemand::value`) have a `type()` method returning
-a `json_type` value describing the value (`json_type::array`, `json_type::object`, `json_type::number`, `json_type::string`, `json_type::boolean`, `json_type::null`).
+We also have a generic type (`simdjson::ondemand::value`) which represents a potential
+array or object, or scalar type (`double`, `uint64_t`, `int64_t`, `bool`, `null`, string) inside
+an array or an object. Both generic types (`simdjson::ondemand::document` and
+`simdjson::ondemand::value`) have a `type()` method returning a `json_type` value describing the
+value (`json_type::array`, `json_type::object`, `json_type::number`, `json_type::string`, `json_type::boolean`, `json_type::null`).

 Advanced users who need to determine the number types (integer or float) dynamically,
 should review our section [dynamic number types](#dynamic-number-types). Indeed,
 we have an additional `ondemand::number` type which may represent either integers
 or floating-point values, depending on how the numbers are formatted.
 floating-point values followed by an integer.
-While you are accessing the document, the `document` instance should remain in scope:
-it is your "iterator" which keeps track of where you are in the JSON document.
-By design, there is one and only one `document` instance per JSON document.

+We invite you to keep the following rules in mind:
+1. While you are accessing the document, the `document` instance should remain in scope: it is your "iterator" which keeps track of where you are in the JSON document. By design, there is one and only one `document` instance per JSON document.
+2. Because On Demand is really just an iterator, you must fully consume the current object or array before accessing a sibling object or array.
+3. Values can only be consumed once, you should get the values and store them if you plan to need them multiple times. You are expected to access the keys of an object just once. You are expected to go through the values of an array just once.

 The following specific instructions indicate how to use the JSON when exceptions are enabled, but simdjson has full, idiomatic
 support for users who avoid exceptions. See [the simdjson error handling documentation](basics.md#error-handling) for more.
@@ -259,16 +278,24 @@ support for users who avoid exceptions. See [the simdjson error handling documen
  However, it is not fully validated. On Demand only fully validates the values you use and the
  structure leading to it.
 * **Extracting Values:** You can cast a JSON element to a native type:
-  `double(element)` or `double x = json_element`. This works for `std::string_view`, double, uint64_t, int64_t, bool,
-  ondemand::object and ondemand::array. At this point, the number, string or boolean will be parsed,
-  or the initial `[` or `{` will be verified. An exception is thrown if the cast is not possible.
+  `double(element)`. This works for `std::string_view`, double, uint64_t, int64_t, bool,
+  ondemand::object and ondemand::array. We also have explicit methods such as `get_string()`, `get_double()`,
+  `get_uint64()`, `get_int64()`, `get_bool()`, `get_object()` and `get_array()`. After a cast or an explicit method,
+  the number, string or boolean will be parsed, or the initial `[` or `{` will be verified. An exception is thrown if
+  the cast is not possible.

  > IMPORTANT NOTE: values can only be parsed once. Since documents are *iterators*, once you have
-  > parsed a value (such as by casting to double), you cannot get at it again.
+  > parsed a value (such as by casting to double), you cannot get at it again. It is an error to call
+  > `get_string()` twice on an object (or to cast an object twice to `std::string_view`).
 * **Field Access:** To get the value of the "foo" field in an object, use `object["foo"]`. This will
  scan through the object looking for the field with the matching string, doing a character-by-character
  comparison. For efficiency reason, you should avoid looking up the same field repeatedly: e.g., do
-  not do `object["foo"]` followed by `object["foo"]` with the same `object` instance.
+  not do `object["foo"]` followed by `object["foo"]` with the same `object` instance. If you consume an
+  object twice: `std::string_view(object["foo"]` followed by `std::string_view(object["foo"]`, your code
+  is in error. Furthermore, you can only consume one field at a time, on the same object. Thus
+  if you have retrieved `content["bids"].get_array()` and you later call `content["asks"].get_array()`, then the
+  first array should no longer be accessed: it would be unsafe to do so. You can detect such mistakes by first
+  compiling and running the code in Debug mode: an OUT_OF_ORDER_ITERATION error is generated.

  > NOTE: JSON allows you to escape characters in keys. E.g., the key `"date"` may be written as
  > `"\u0064\u0061\u0074\u0065"`. By default, simdjson does *not* unescape keys when matching by default.
@@ -683,30 +710,23 @@ auto cars_json = R"( [
 ] )"_padded;

 ondemand::parser parser;
-ondemand::document cars;
 std::vector<double> measured;
-parser.iterate(cars_json).get(cars);
+ondemand::document cars = parser.iterate(cars_json);
 std::vector<car_type> content;
 for (int i = 0; i < 3; i++) {
-    ondemand::object obj;
    std::string json_pointer = "/" + std::to_string(i);
    // Each successive at_pointer call invalidates
    // previously parsed values, strings, objects and array.
-    cars.at_pointer(json_pointer).get(obj);
+    ondemand::object obj(cars.at_pointer(json_pointer).get_object());
    // We materialize the object.
-    std::string_view make;
-    ASSERT_SUCCESS(obj["make"].get(make));
-    std::string_view model;
-    ASSERT_SUCCESS(obj["model"].get(model));
-    uint64_t year;
-    ASSERT_SUCCESS(obj["year"].get(year));
+    std::string_view make = obj["make"];
+    std::string_view model = obj["model"];
+    uint64_t year(obj["year"]);
    // We materialize the array.
-    ondemand::array arr;
-    ASSERT_SUCCESS(obj["tire_pressure"].get(arr));
+    ondemand::array arr(obj["tire_pressure"].get_array());
    std::vector<double> values;
    for(auto x : arr) {
-        double value_double;
-        ASSERT_SUCCESS(x.get(value_double));
+        double value_double(x.get_double());
        values.push_back(value_double);
    }
    content.emplace_back(make, model, year, std::move(values));
@@ -741,12 +761,16 @@ The entire simdjson API is usable with and without exceptions. All simdjson APIs
 pair. You can retrieve the value with .get() without generating an exception, like so:

 ```c++
-ondemand::element doc;
+ondemand::document doc;
 auto error = parser.iterate(json).get(doc);
 if (error) { cerr << error << endl; exit(1); }
 ```

-When you use the code this way, it is your responsibility to check for error before using the
+When there is no error, the error code simdjson::SUCCESS is returned: it evaluates as false as a Boolean.
+We have several error codes to indicate errors, they all evaluate to true as a Boolean: your software should not generally not depend on exact
+error codes. We may change the error codes in future releases and the exact error codes could vary depending on your system.
+
+When you use the code without exceptions, it is your responsibility to check for error before using the
 result: if there is an error, the result value will not be valid and using it will caused undefined behavior. Most compilers should be able to help you if you activate the right
 set of warnings: they can identify variables that are written to but never otherwise accessed.

@@ -868,6 +892,9 @@ int main(void) {
 }
 ```

+The `at` method can only be called once on an array. It cannot be used
+to iterate through the values of an array.
+
 ### Error Handling Examples without Exceptions

 This is how the example in "Using the Parsed JSON" could be written using only error code checking (without exceptions):
@@ -1119,7 +1146,7 @@ Direct Access to the Raw String

 The simdjson library makes explicit assumptions about types. For examples, numbers
 must be integers (up to 64-bit integers) or binary64 floating-point numbers. Some users
-have different needs. For example, some users might want to support big infloating-point number followed by an integer.tegers.
+have different needs. For example, some users might want to support big integers.
 The library makes this possible by providing a `raw_json_token` method which returns
 a `std::string_view` instance containing the value as a string which you may then
 parse as you see fit.
@@ -1153,11 +1180,19 @@ The `raw_json_token()` should be fast and free of allocation.
 Newline-Delimited JSON (ndjson) and JSON lines
 ----------------------------------------------

-The simdjson library also supports multithreaded JSON streaming through a large file containing many
-smaller JSON documents in either [ndjson](http://ndjson.org) or [JSON lines](http://jsonlines.org)
-format. If your JSON documents all contain arrays or objects, we even support direct file
-concatenation without whitespace. The concatenated file has no size restrictions (including larger
-than 4GB), though each individual document must be no larger than 4 GB.
+When processing large inputs (e.g., in the context of data engineering), engineers commonly
+serialize data into streams of multiple JSON documents. That is, instead of one large
+(e.g., 2 GB) JSON document containing multiple records, it is often preferable to
+write out multiple records as independent JSON documents, to be read one-by-one.
+
+The simdjson library also supports multithreaded JSON streaming through a large file
+containing many smaller JSON documents in either [ndjson](http://ndjson.org)
+or [JSON lines](http://jsonlines.org) format. If your JSON documents all contain arrays
+or objects, we even support direct file concatenation without whitespace. However, if there
+is content between your JSON documents, it should be exclusively ASCII white-space characters.
+
+The concatenated file has no size restrictions (including larger than 4GB), though each
+individual document must be no larger than 4 GB.

 Here is an example:

@@ -1438,8 +1473,8 @@ The simdjson library is fully compliant with  the [RFC 8259](https://www.tbray.o
 - A single string or a single number is considered to be a valid JSON document.
 - We fully validate the numbers according to the JSON specification. For example,  the string `01` is not valid JSON document since the specification states that *leading zeros are not allowed*.
 - The specification allows implementations to set limits on the range and precision of numbers accepted.  We support 64-bit floating-point numbers as well as integer values.
-  - We parse integers and floating-point numbers afloating-point number followed by an integer.s separate types which allows us to support all signed (two's complement) 64-bit integersfloating-point number followed by an integer., like a Java `long` or a C/C++ `long long` and all 64-bit unsigned integers. When we cannot represent exactly an integer as a signed or unsigned 64-bit value, we reject the JSON document.
-  - We support the full range of 64-bit floating-point numbers (binary64). The values range from `std::numeric_limits<double>::lowest()`  to `std::numefloating-point number followed by an integer.ric_limits<double>::max()`, so from -1.7976e308 all the way to 1.7975e308. Extreme values (less or equal to -1e308, greater or equal to 1e308) are rejected: we refuse to parse the input document. Numbers are parsed with a perfect accuracy (ULP 0): the nearest floating-point value is chosen, rounding to even when needed. If you serialized your floating-point numbers with 17floating-point value  followed by an integer. significant digits in a standard compliant manner, the simdjson library is guaranfloating-point number followed by an integer.teed to recover the same numbers, exactly.
+  - We parse integers and floating-point numbers as separate types which allows us to support all signed (two's complement) 64-bit integers, like a Java `long` or a C/C++ `long long` and all 64-bit unsigned integers. When we cannot represent exactly an integer as a signed or unsigned 64-bit value, we reject the JSON document.
+  - We support the full range of 64-bit floating-point numbers (binary64). The values range from `std::numeric_limits<double>::lowest()`  to `std::numeric_limits<double>::max()`, so from -1.7976e308 all the way to 1.7975e308. Extreme values (less or equal to -1e308, greater or equal to 1e308) are rejected: we refuse to parse the input document. Numbers are parsed with a perfect accuracy (ULP 0): the nearest floating-point value is chosen, rounding to even when needed. If you serialized your floating-point numbers with 17 significant digits in a standard compliant manner, the simdjson library is guaranteed to recover the same numbers, exactly.
 - The specification states that JSON text exchanged between systems that are not part of a closed ecosystem MUST be encoded using UTF-8. The simdjson library does full UTF-8 validation as part of the parsing. The specification states that implementations MUST NOT add a byte order mark: the simdjson library rejects documents starting with a  byte order mark.
 - The simdjson library validates string content for unescaped characters. Unescaped line breaks and tabs in strings are not allowed.
 - The simdjson library accepts objects with repeated keys: all of the name/value pairs, including duplicates, are reported. We do not enforce key uniqueness.
@@ -3,8 +3,8 @@ The Document-Object-Model (DOM) front-end

 An overview of what you need to know to use simdjson, with examples.

-* [DOM vs On Demand](#dom-vs-ondemand)
-* [The Basics: Loading and Parsing JSON Documents](#the-basics-loading-and-parsing-json-documents)
+* [DOM vs On Demand](#dom-vs-on-demand)
+* [The Basics: Loading and Parsing JSON Documents](#the-basics-loading-and-parsing-json-documents-using-the-dom-front-end)
 * [Using the Parsed JSON](#using-the-parsed-json)
 * [C++17 Support](#c17-support)
 * [JSON Pointer](#json-pointer)
@@ -81,7 +81,7 @@ Once you have an element, you can navigate it with idiomatic C++ iterators, oper
  know the type of the value, you can cast it right there, too! `for (double value : array) { ... }`
 * **Object Iteration:** You can iterate through an object's fields, too: `for (auto [key, value] : object)`
 * **Array Index:** To get at an array value by index, use the at() method: `array.at(0)` gets the
-  first element.
+  first element. The at() method has linear-time complexity so it should not be used to iterate over the values of an array.
  > Note that array[0] does not compile, because implementing [] gives the impression indexing is a
  > O(1) operation, which it is not presently in simdjson. Instead, you should iterate over the elements
  > using a for-loop, as in our examples.
@@ -249,7 +249,11 @@ auto error = parser.parse(json).get(doc);
 if (error) { cerr << error << endl; exit(1); }
 ```

-When you use the code this way, it is your responsibility to check for error before using the
+When there is no error, the error code simdjson::SUCCESS is returned: it evaluates as false as a Boolean.
+We have several error codes to indicate errors, they all evaluate to true as a Boolean: your software should not generally not depend on exact
+error codes. We may change the error codes in future releases and the exact error codes could vary depending on your system.
+
+When you use the code without exceptions, it is your responsibility to check for error before using the
 result: if there is an error, the result value will not be valid and using it will caused undefined
 behavior.

@@ -318,6 +322,9 @@ int main(void) {
 }
 ```

+The `at()` method has linear-time complexity: it should not be used to iterate
+over the content of an array.
+
 ### Error Handling Example

 This is how the example in "Using the Parsed JSON" could be written using only error code checking:
@@ -50,9 +50,9 @@ Inspecting the Detected Implementation
 You can check what implementation is running with `active_implementation`:

 ```c++
-cout << "simdjson v" << STRINGIFY(SIMDJSON_VERSION) << endl;
-cout << "Detected the best implementation for your machine: " << simdjson::active_implementation->name();
-cout << "(" << simdjson::active_implementation->description() << ")" << endl;
+cout << "simdjson v" << SIMDJSON_STRINGIFY(SIMDJSON_VERSION) << endl;
+cout << "Detected the best implementation for your machine: " << simdjson::get_active_implementation()->name();
+cout << "(" << simdjson::get_active_implementation()->description() << ")" << endl;
 ```

 Implementation detection will happen in this case when you first call `name()`.
@@ -63,7 +63,7 @@ Querying Available Implementations
 You can list all available implementations, regardless of which one was selected:

 ```c++
-for (auto implementation : simdjson::available_implementations) {
+for (auto implementation : simdjson::get_available_implementations()) {
  cout << implementation->name() << ": " << implementation->description() << endl;
 }
 ```
@@ -71,10 +71,10 @@ for (auto implementation : simdjson::available_implementations) {
 And look them up by name:

 ```c++
-cout << simdjson::available_implementations["fallback"]->description() << endl;
+cout << simdjson::get_available_implementations()["fallback"]->description() << endl;
 ```
 Though the fallback implementation should always be available, others might be missing. When
-an implementation is not available, the bracket call `simdjson::available_implementations[name]`
+an implementation is not available, the bracket call `simdjson::get_available_implementations()[name]`
 will return the null pointer.

 The available implementations have been compiled but may not necessarily be run safely on your system
@@ -90,18 +90,18 @@ can select the CPU architecture yourself:

 ```c++
 // Use the fallback implementation, even though my machine is fast enough for anything
-simdjson::active_implementation = simdjson::available_implementations["fallback"];
+simdjson::get_active_implementation() = simdjson::get_available_implementations()["fallback"];
 ```

 You are responsible for ensuring that the requirements of the selected implementation match your current system.
-Furthermore, you should check that the implementation is available before setting it to `simdjson::active_implementation`
+Furthermore, you should check that the implementation is available before setting it to `simdjson::get_active_implementation()`
 by comparing it with the null pointer.

 ```c++
-auto my_implementation = simdjson::available_implementations["haswell"];
+auto my_implementation = simdjson::get_available_implementations()["haswell"];
 if(! my_implementation) { exit(1); }
 if(! my_implementation->supported_by_runtime_system()) { exit(1); }
-simdjson::active_implementation = my_implementation;
+simdjson::get_active_implementation() = my_implementation;
 ```

 Checking that an Implementation can Run on your System
@@ -110,12 +110,12 @@ Checking that an Implementation can Run on your System
 You should call `supported_by_runtime_system()` to compare the processor's features with the need of the implementation.

 ```c++
-for (auto implementation : simdjson::available_implementations) {
+for (auto implementation : simdjson::get_available_implementations()) {
  if(implementation->supported_by_runtime_system()) {
    cout << implementation->name() << ": " << implementation->description() << endl;
  }
 }
 ```

-The call to `supported_by_runtime_system()` maybe relatively expensive. Do not call  `supported_by_runtime_system()` each
-time you parse a JSON input (for example). It is meant to be called a handful of times at most in the life of a program.
+The call to `supported_by_runtime_system()` may be relatively expensive. Do not call  `supported_by_runtime_system()` each
+time you parse a JSON input (for example). It is meant to be called a handful of times at most in the life of a program.
@@ -1,7 +1,9 @@
 iterate_many
 ==========

-An interface providing features to work with files or streams containing multiple small JSON documents. Given an input such as
+When serializing large databases, it is often better to write out many independent JSON
+documents, instead of one large monolithic document containing many records. The simdjson
+library provides high-speed access to files or streams containing multiple small JSON documents separated by ASCII white-space characters. Given an input such as
 ```JSON
 {"text":"a"}
 {"text":"b"}
@@ -114,7 +116,9 @@ Whitespace Characters:
 - **Linefeed**
 - **Carriage return**
 - **Horizontal tab**
- **Nothing**
+
+If your documents are all objects or arrays, then you may even have nothing between them.
+E.g., `[1,2]{"32":1}` is recognized as two documents.

 Some official formats **(non-exhaustive list)**:
 - [Newline-Delimited JSON (NDJSON)](http://ndjson.org/)
@@ -709,42 +709,6 @@ in production systems:
    }
 ```

-
-### Long-Running Processes and Memory Capacity
-
-The On Demand approach also automatically expands its memory capacity when larger documents are parsed. However, for longer processes where very large files are processed (such as server loops), this capacity is not resized down. Similarly to the DOM-based approach (see [here](https://github.com/simdjson/simdjson/blob/master/doc/dom.md#server-loops-long-running-processes-and-memory-capacity)]), On Demand also lets you adjust the maximal capacity that the parser can process:
-
-* You can set an upper bound (*max_capacity*) when construction the parser:
-```C++
-    ondemand::parser parser(1000*1000);  // Never grows past documents > 1 MB
-    auto doc = parser.iterate(json);
-    for (web_request request : listen()) {
-      padded_string json;
-      padded_string json = padded_string::load(request.body);
-      auto error = parser.iterate(json);
-      // If the document was above our limit, emit 413 = payload too large
-      if (error == CAPACITY) { request.respond(413); continue; }
-      // ...
-    }
-```
-
-The capacity will grow as the parser encounters larger documents up to 1 MB.
-
-* You can also allocate a *fixed capacity* that will never grow:
-```C++
-    ondemand::parser parser(1000*1000);
-    parser.allocate(1000*1000)  // Fix the capacity to 1 MB
-    auto doc = parser.iterate(json);
-    for (web_request request : listen()) {
-      padded_string json;
-      padded_string json = padded_string::load(request.body);
-      auto error = parser.iterate(json);
-      // If the document was above our limit, emit 413 = payload too large
-      if (error == CAPACITY) { request.respond(413); continue; }
-      // ...
-    }
-```
-You can also manually set the maximal capacity using the method `set_max_capacity()`.
 ### Benefits of the On Demand Approach

 We expect that the On Demand approach has many of the performance benefits of the schema-based approach, while providing a flexibility that is similar to that of the DOM-based approach.
@@ -764,7 +728,6 @@ There are currently additional technical limitations which we expect to resolve

 * The simdjson library offers runtime dispatching which allows you to compile one binary and have it run at full speed on different processors, taking advantage of the specific features of the processor. The On Demand API has limited runtime dispatch support. Under x64 systems, to fully benefit from the On Demand API, we recommend that you compile your code for a specific processor. E.g., if your processor supports AVX2 instructions, you should compile your binary executable with AVX2 instruction support (by using your compiler's commands). If you are sufficiently technically proficient, you can implement runtime dispatching within your application, by compiling your On Demand code for different processors.
 * There is an initial phase which scans the entire document quickly, irrespective of the size of the document. We plan to break this phase into distinct steps for large files in a future release as we have done with other components of our API (e.g., `parse_many`).
-* The On Demand API does not support JSON Pointer. This capability is currently limited to our core API.

 ### Applicability of the On Demand Approach

@@ -56,6 +56,43 @@ or simply
 ```


+Server Loops: Long-Running Processes and Memory Capacity
+---------------------------------
+
+The On Demand approach also automatically expands its memory capacity when larger documents are parsed. However, for longer processes where very large files are processed (such as server loops), this capacity is not resized down. On Demand also lets you adjust the maximal capacity that the parser can process:
+
+* You can set an upper bound (*max_capacity*) when construction the parser:
+```C++
+    ondemand::parser parser(1000*1000);  // Never grows past documents > 1 MB
+    auto doc = parser.iterate(json);
+    for (web_request request : listen()) {
+      padded_string json;
+      padded_string json = padded_string::load(request.body);
+      auto error = parser.iterate(json);
+      // If the document was above our limit, emit 413 = payload too large
+      if (error == CAPACITY) { request.respond(413); continue; }
+      // ...
+    }
+```
+
+The capacity will grow as the parser encounters larger documents up to 1 MB.
+
+* You can also allocate a *fixed capacity* that will never grow:
+```C++
+    ondemand::parser parser(1000*1000);
+    parser.allocate(1000*1000)  // Fix the capacity to 1 MB
+    auto doc = parser.iterate(json);
+    for (web_request request : listen()) {
+      padded_string json;
+      padded_string json = padded_string::load(request.body);
+      auto error = parser.iterate(json);
+      // If the document was above our limit, emit 413 = payload too large
+      if (error == CAPACITY) { request.respond(413); continue; }
+      // ...
+    }
+```
+You can also manually set the maximal capacity using the method `set_max_capacity()`.
+
 Large files and huge page support
 ---------------------------------

@@ -123,6 +160,8 @@ You should not expect the simdjson library to cause *downclocking* of your recen
 - [Whenever 512-bit AVX-512 instructions are used](https://lemire.me/blog/2018/09/07/avx-512-when-and-how-to-use-these-new-instructions/).
 - Whenever heavy 256-bit or wider instructions are used. Heavy instructions are those involving floating point operations or integer multiplications (since these execute on the floating point unit).

-The simdjson library does not currently support AVX-512 instructions and it does not make use of heavy 256-bit instructions. We do use vectorized multiplications, but only using 128-bit registers. Thus there should be no downclocking due to simdjson on recent processors.
+The simdjson library does not make use of heavy 256-bit instructions. We do use vectorized multiplications, but only using 128-bit registers. Thus there should be no downclocking due to simdjson on recent processors, except when AVX-512 is allowed and
+detected. However, we only allow AVX-512 on recent processors (Ice Lake/Tiger Lake or better) where [little to no frequency throttling is expected](https://travisdowns.github.io/blog/2020/08/19/icl-avx512-freq.html). If you can still concerned, you can easily disable AVX-512 with the CMake option `SIMDJSON_AVX512_ALLOWED` set to `OFF` (e.g., `cmake -D SIMDJSON_AVX512_ALLOWED=OFF -B build && cmake --build build`) or by setting
+the macro `SIMDJSON_AVX512_ALLOWED` to `0` in C++ prior to importing the headers.

 You may still be worried about which SIMD instruction set is used by simdjson.  Thankfully,  [you can always determine and change which architecture-specific implementation is used](implementation-selection.md) by simdjson. Thus even if your CPU supports AVX2, you do not need to use AVX2. You are in control.
@@ -93,7 +93,7 @@ extern "C" int LLVMFuzzerTestOneInput(const uint8_t *Data, size_t Size) {
    std::size_t nerrors=0;
    for(std::size_t i=0; i<Nimplementations; ++i) {
        auto& e=implementations[i];
-        simdjson::active_implementation=e.impl;
+        simdjson::get_active_implementation()=e.impl;
        e.error=e.parser.parse(Data,Size).get(e.element);
        if(e.error) {
            ++nerrors;
@@ -8,24 +8,26 @@

 extern "C" int LLVMFuzzerTestOneInput(const uint8_t *Data, size_t Size) {
  FuzzData fd(Data, Size);
-  const auto batch_size=static_cast<size_t>(fd.getInt<0,1000>());
-  const auto json=simdjson::padded_string{fd.remainder_as_stringview()};
+  const auto batch_size = static_cast<size_t>(fd.getInt<0,1000>());
+  const auto json = simdjson::padded_string{fd.remainder_as_stringview()};
  simdjson::dom::parser parser;
-#if SIMDJSON_EXCEPTIONS
-  try {
-#endif
-    simdjson::dom::document_stream docs;
-    if(parser.parse_many(json,batch_size).get(docs)) {
-      return 0;
-    }
-
-  size_t bool_count=0;
+  simdjson::dom::document_stream docs;
+  if(parser.parse_many(json,batch_size).get(docs)) { return 0; }
+  size_t bool_count1 = 0;
+  size_t total_count1 = 0;
  for (auto doc : docs) {
-    bool_count+=doc.is_bool();
+    total_count1++;
+    bool_count1 += doc.is_bool();
  }
-#if SIMDJSON_EXCEPTIONS
-  } catch(...) {
+  // Restart, if we made it this far, the document *must* be accessible.
+  if(parser.parse_many(json,batch_size).get(docs)) { return EXIT_FAILURE; }
+  size_t bool_count2 = 0;
+  size_t total_count2 = 0;
+  for (auto doc : docs) {
+    total_count2++;
+    bool_count2 += doc.is_bool();
  }
-#endif
+  // They should agree!!!
+  if((total_count2 != total_count1) || (bool_count2 != bool_count1)) { return EXIT_FAILURE; }
  return 0;
 }
@@ -17,7 +17,7 @@
 std::vector<const simdjson::implementation*>
 get_runtime_supported_implementations() {
    std::vector<const simdjson::implementation*> ret;
-    for(auto& e: simdjson::available_implementations) {
+    for(auto& e: simdjson::get_available_implementations()) {
        if(e->supported_by_runtime_system()) {
            ret.emplace_back(e);
        }
@@ -34,6 +34,7 @@
    }
 */

+#include "simdjson/simdjson_version.h"
 #include "simdjson/dom.h"
 #include "simdjson/builtin.h"

@@ -8,7 +8,7 @@ namespace {
 // We sometimes call trailing_zero on inputs that are zero,
 // but the algorithms do not end up using the returned value.
 // Sadly, sanitizers are not smart enough to figure it out.
-NO_SANITIZE_UNDEFINED
+SIMDJSON_NO_SANITIZE_UNDEFINED
 simdjson_really_inline int trailing_zeroes(uint64_t input_num) {
 #ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
  unsigned long ret;
@@ -73,7 +73,7 @@ simdjson_really_inline uint64_t reverse_bits(uint64_t input_num) {
 * greating or equal to 63 in which case we trigger undefined behavior, but the output
 * of such undefined behavior is never used.
 **/
-NO_SANITIZE_UNDEFINED
+SIMDJSON_NO_SANITIZE_UNDEFINED
 simdjson_really_inline uint64_t zero_leading_bit(uint64_t rev_bits, int leading_zeroes) {
  return rev_bits ^ (uint64_t(0x8000000000000000) >> leading_zeroes);
 }
@@ -132,7 +132,7 @@ simdjson_really_inline int8x16_t make_int8x16_t(int8_t x1,  int8_t x2,  int8_t x
    simdjson_really_inline simd8<T>& operator&=(const simd8<T> other) { auto this_cast = static_cast<simd8<T>*>(this); *this_cast = *this_cast & other; return *this_cast; }
    simdjson_really_inline simd8<T>& operator^=(const simd8<T> other) { auto this_cast = static_cast<simd8<T>*>(this); *this_cast = *this_cast ^ other; return *this_cast; }

-    simdjson_really_inline Mask operator==(const simd8<T> other) const { return vceqq_u8(*this, other); }
+    friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return vceqq_u8(lhs, rhs); }

    template<int N=1>
    simdjson_really_inline simd8<T> prev(const simd8<T> prev_chunk) const {
@@ -5,7 +5,9 @@

 // Determine the best builtin implementation
 #ifndef SIMDJSON_BUILTIN_IMPLEMENTATION
-#if SIMDJSON_CAN_ALWAYS_RUN_HASWELL
+#if SIMDJSON_CAN_ALWAYS_RUN_ICELAKE
+#define SIMDJSON_BUILTIN_IMPLEMENTATION icelake
+#elif SIMDJSON_CAN_ALWAYS_RUN_HASWELL
 #define SIMDJSON_BUILTIN_IMPLEMENTATION haswell
 #elif SIMDJSON_CAN_ALWAYS_RUN_WESTMERE
 #define SIMDJSON_BUILTIN_IMPLEMENTATION westmere
@@ -176,7 +176,7 @@ inline error_code parser::allocate(size_t capacity, size_t max_depth) noexcept {
  if (implementation) {
    err = implementation->allocate(capacity, max_depth);
  } else {
-    err = simdjson::active_implementation->create_dom_parser_implementation(capacity, max_depth, implementation);
+    err = simdjson::get_active_implementation()->create_dom_parser_implementation(capacity, max_depth, implementation);
  }
  if (err) { return err; }
  return SUCCESS;
@@ -157,7 +157,10 @@ simdjson_really_inline void mini_formatter::string(std::string_view unescaped) {
  size_t i = 0;
  // Fast path for the case where we have no control character, no ", and no backslash.
  // This should include most keys.
-  constexpr static bool needs_escaping[] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  //
+  // We would like to use 'bool' but some compilers take offense to bitwise operation
+  // with bool types.
+  constexpr static char needs_escaping[] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0,
@@ -169,6 +172,8 @@ simdjson_really_inline void mini_formatter::string(std::string_view unescaped) {
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
  for(;i + 8 <= unescaped.length(); i += 8) {
    // Poor's man vectorization. This could get much faster if we used SIMD.
+    //
+    // It is not the case that replacing '|' with '||' would be neutral performance-wise.
    if(needs_escaping[uint8_t(unescaped[i])] | needs_escaping[uint8_t(unescaped[i+1])]
      | needs_escaping[uint8_t(unescaped[i+2])] | needs_escaping[uint8_t(unescaped[i+3])]
      | needs_escaping[uint8_t(unescaped[i+4])] | needs_escaping[uint8_t(unescaped[i+5])]
@@ -7,7 +7,12 @@
 namespace simdjson {

 /**
- * All possible errors returned by simdjson.
+ * All possible errors returned by simdjson. These error codes are subject to change
+ * and not all simdjson kernel returns the same error code given the same input: it is not
+ * well defined which error a given input should produce.
+ *
+ * Only SUCCESS evaluates to false as a Boolean. All other error codes will evaluate
+ * to true as a Boolean.
 */
 enum error_code {
  SUCCESS = 0,                ///< No error
@@ -361,7 +361,7 @@ error_code slow_float_parsing(simdjson_unused const uint8_t * src, W writer) {
 }

 template<typename I>
-NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
+SIMDJSON_NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
 simdjson_really_inline bool parse_digit(const uint8_t c, I &i) {
  const uint8_t digit = static_cast<uint8_t>(c - '0');
  if (digit > 9) {
@@ -609,7 +609,7 @@ simdjson_really_inline error_code parse_number(const uint8_t *const src, W &writ
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    }  else if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INVALID_NUMBER(src); }
  }
@@ -732,7 +732,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -782,7 +782,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -830,9 +830,11 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned_
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
-    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
+    // Note: we use src[1] and not src[0] because src[0] is the quote character in this
+    // instance.
+    if (src[1] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }

  return i;
@@ -104,7 +104,7 @@ public:

  /**
   * Get the value at the given index. This function has linear-time complexity.
-   * This function should only be called once as the array iterator is not reset between each call.
+   * This function should only be called once on an array instance since the array iterator is not reset between each call.
   *
   * @return The value at the given index, or:
   *         - INDEX_OUT_OF_BOUNDS if the array index is larger than an array length
@@ -527,7 +527,7 @@ simdjson_really_inline simdjson_result<value> document_reference::find_field_uno
 simdjson_really_inline simdjson_result<value> document_reference::find_field_unordered(const char *key) & noexcept { return doc->find_field_unordered(key); }
 simdjson_really_inline simdjson_result<json_type> document_reference::type() noexcept { return doc->type(); }
 simdjson_really_inline simdjson_result<bool> document_reference::is_scalar() noexcept { return doc->is_scalar(); }
-simdjson_really_inline simdjson_result<const char *> document_reference::current_location() noexcept { return doc->current_location(); };
+simdjson_really_inline simdjson_result<const char *> document_reference::current_location() noexcept { return doc->current_location(); }
 simdjson_really_inline bool document_reference::is_negative() noexcept { return doc->is_negative(); }
 simdjson_really_inline simdjson_result<bool> document_reference::is_integer() noexcept { return doc->is_integer(); }
 simdjson_really_inline simdjson_result<number_type> document_reference::get_number_type() noexcept { return doc->get_number_type(); }
@@ -715,4 +715,4 @@ simdjson_really_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value>
 }


-} // namespace simdjson
+} // namespace simdjson
@@ -94,6 +94,8 @@ public:
   *
   * The string is guaranteed to be valid UTF-8.
   *
+   * Important: Calling get_string() twice on the same document is an error.
+   *
   * @returns An UTF-8 string. The string is stored in the parser and will be invalidated the next
   *          time it parses a document or when it is destroyed.
   * @returns INCORRECT_TYPE if the JSON value is not a string.
@@ -267,7 +269,7 @@ public:
  simdjson_really_inline simdjson_result<size_t> count_fields() & noexcept;
  /**
   * Get the value at the given index in the array. This function has linear-time complexity.
-   * This function should only be called once as the array iterator is not reset between each call.
+   * This function should only be called once on an array instance since the array iterator is not reset between each call.
   *
   * @return The value at the given index, or:
   *         - INDEX_OUT_OF_BOUNDS if the array index is larger than an array length
@@ -303,6 +305,18 @@ public:
   * **Raw Keys:** The lookup will be done against the *raw* key, and will not unescape keys.
   * e.g. `object["a"]` will match `{ "a": 1 }`, but will *not* match `{ "\u0061": 1 }`.
   *
+   *
+   * You must consume the fields on an object one at a time. A request for a new key
+   * invalidates previous field values: it makes them unsafe. E.g., the array
+   * given by content["bids"].get_array() should not be accessed after you have called
+   * content["asks"].get_array(). You can detect such mistakes by first compiling and running
+   * the code in Debug mode (or with the macro `SIMDJSON_DEVELOPMENT_CHECKS` set to 1): an
+   * OUT_OF_ORDER_ITERATION error is generated.
+   *
+   * You are expected to access keys only once. You should access the value corresponding to
+   * a key a single time. Doing object["mykey"].to_string()and then again object["mykey"].to_string()
+   * is an error.
+   *
   * @param key The key to look up.
   * @returns The value of the field, or NO_SUCH_FIELD if the field is not in the object.
   */
@@ -326,6 +340,17 @@ public:
   * Use find_field() if you are sure fields will be in order (or are willing to treat it as if the
   * field wasn't there when they aren't).
   *
+   * You must consume the fields on an object one at a time. A request for a new key
+   * invalidates previous field values: it makes them unsafe. E.g., the array
+   * given by content["bids"].get_array() should not be accessed after you have called
+   * content["asks"].get_array(). You can detect such mistakes by first compiling and running
+   * the code in Debug mode (or with the macro `SIMDJSON_DEVELOPMENT_CHECKS` set to 1): an
+   * OUT_OF_ORDER_ITERATION error is generated.
+   *
+   * You are expected to access keys only once. You should access the value corresponding to a key
+   * a single time. Doing object["mykey"].to_string() and then again object["mykey"].to_string()
+   * is an error.
+   *
   * @param key The key to look up.
   * @returns The value of the field, or NO_SUCH_FIELD if the field is not in the object.
   */
@@ -399,7 +424,7 @@ public:
   * type.
   *
   * number.get_number_type() is number_type::signed_integer if we have
-   * a integer in [-9223372036854775808,9223372036854775808)
+   * an integer in [-9223372036854775808,9223372036854775808)
   * You can recover the value by calling number.get_int64() and you
   * have that number.is_int64() is true.
   *
@@ -498,7 +523,7 @@ public:
  simdjson_really_inline simdjson_result<value> at_pointer(std::string_view json_pointer) noexcept;
  /**
   * Consumes the document and returns a string_view instance corresponding to the
-   * document as represented in JSON. It points inside the original byte array containg
+   * document as represented in JSON. It points inside the original byte array containing
   * the JSON document.
   */
  simdjson_really_inline simdjson_result<std::string_view> raw_json() noexcept;
@@ -541,6 +566,7 @@ public:
  simdjson_really_inline document_reference() noexcept;
  simdjson_really_inline document_reference(document &d) noexcept;
  simdjson_really_inline document_reference(const document_reference &other) noexcept = default;
+  simdjson_really_inline document_reference& operator=(const document_reference &other) noexcept = default;
  simdjson_really_inline void rewind() noexcept;
  simdjson_really_inline simdjson_result<array> get_array() & noexcept;
  simdjson_really_inline simdjson_result<object> get_object() & noexcept;
@@ -53,17 +53,6 @@ inline void json_iterator::rewind() noexcept {
 SIMDJSON_PUSH_DISABLE_WARNINGS
 SIMDJSON_DISABLE_STRICT_OVERFLOW_WARNING
 simdjson_warn_unused simdjson_really_inline error_code json_iterator::skip_child(depth_t parent_depth) noexcept {
-  /***
-   * WARNING:
-   * Inside an object, a string value is a depth of +1 compared to the object. Yet a key
-   * is at the same depth as the object.
-   * But json_iterator cannot easily tell whether we are pointing at a key or a string value.
-   * Instead, it assumes that if you are pointing at a string, then it is a value, not a key.
-   * To be clear...
-   * the following code assumes that we are *not* pointing at a key. If we are then a bug
-   * will follow. Unfortunately, it is not possible for the json_iterator its to make this
-   * check.
-   */
  if (depth() <= parent_depth) { return SUCCESS; }
  switch (*return_current_and_advance()) {
    // TODO consider whether matching braces is a requirement: if non-matching braces indicates
@@ -91,19 +80,18 @@ simdjson_warn_unused simdjson_really_inline error_code json_iterator::skip_child
      if (at_end()) { return report_error(INCOMPLETE_ARRAY_OR_OBJECT, "Missing [ or { at start"); }
 #endif // SIMDJSON_CHECK_EOF
      break;
-    /*case '"':
+    case '"':
      if(*peek() == ':') {
-        // we are at a key!!! This is
-        // only possible if someone searched
-        // for a key in an object and the key
-        // was not found but our code then
-        // decided the consume the separating
-        // comma before returning.
+        // We are at a key!!!
+        // This might happen if you just started an object and you skip it immediately.
+        // Performance note: it would be nice to get rid of this check as it is somewhat
+        // expensive.
+        // https://github.com/simdjson/simdjson/issues/1742
        logger::log_value(*this, "key");
-        advance(); // eat up the ':'
+        return_current_and_advance(); // eat up the ':'
        break; // important!!!
      }
-      simdjson_fallthrough;*/
+      simdjson_fallthrough;
    // Anything else must be a scalar value
    default:
      // For the first scalar, we will have incremented depth already, so we decrement it here.
@@ -111,7 +111,6 @@ protected:
    uint64_t unsigned_integer;
  } payload{0};
  number_type type{number_type::signed_integer};
-  friend class value_iterator;
 };

 /**
@@ -37,6 +37,17 @@ public:
   * **Raw Keys:** The lookup will be done against the *raw* key, and will not unescape keys.
   * e.g. `object["a"]` will match `{ "a": 1 }`, but will *not* match `{ "\u0061": 1 }`.
   *
+   * You must consume the fields on an object one at a time. A request for a new key
+   * invalidates previous field values: it makes them unsafe. E.g., the array
+   * given by content["bids"].get_array() should not be accessed after you have called
+   * content["asks"].get_array(). You can detect such mistakes by first compiling and running
+   * the code in Debug mode (or with the macro `SIMDJSON_DEVELOPMENT_CHECKS` set to 1): an
+   * OUT_OF_ORDER_ITERATION error is generated.
+   *
+   * You are expected to access keys only once. You should access the value corresponding to a
+   * key a single time. Doing object["mykey"].to_string() and then again object["mykey"].to_string()
+   * is an error.
+   *
   * @param key The key to look up.
   * @returns The value of the field, or NO_SUCH_FIELD if the field is not in the object.
   */
@@ -63,6 +74,16 @@ public:
   * If you have multiple fields with a matching key ({"x": 1,  "x": 1}) be mindful
   * that only one field is returned.
   *
+   * You must consume the fields on an object one at a time. A request for a new key
+   * invalidates previous field values: it makes them unsafe. E.g., the array
+   * given by content["bids"].get_array() should not be accessed after you have called
+   * content["asks"].get_array(). You can detect such mistakes by first compiling and running
+   * the code in Debug mode (or with the macro `SIMDJSON_DEVELOPMENT_CHECKS` set to 1): an
+   * OUT_OF_ORDER_ITERATION error is generated.
+   *
+   * You are expected to access keys only once. You should access the value corresponding to a key
+   * a single time. Doing object["mykey"].to_string() and then again object["mykey"].to_string() is an error.
+   *
   * @param key The key to look up.
   * @returns The value of the field, or NO_SUCH_FIELD if the field is not in the object.
   */
@@ -21,7 +21,7 @@ simdjson_warn_unused simdjson_really_inline error_code parser::allocate(size_t n
    SIMDJSON_TRY( implementation->set_capacity(new_capacity) );
    SIMDJSON_TRY( implementation->set_max_depth(new_max_depth) );
  } else {
-    SIMDJSON_TRY( simdjson::active_implementation->create_dom_parser_implementation(new_capacity, new_max_depth, implementation) );
+    SIMDJSON_TRY( simdjson::get_active_implementation()->create_dom_parser_implementation(new_capacity, new_max_depth, implementation) );
  }
  _capacity = new_capacity;
  _max_depth = new_max_depth;
@@ -54,6 +54,11 @@ public:
   *   ondemand::parser parser;
   *   document doc = parser.iterate(json);
   *
+   * It is expected that the content is a valid UTF-8 file, containing a valid JSON document.
+   * Otherwise the iterate method may return an error. In particular, the whole input should be
+   * valid: we do not attempt to tolerate incorrect content either before or after a JSON
+   * document.
+   *
   * ### IMPORTANT: Validate what you use
   *
   * Calling iterate on an invalid JSON document may not immediately trigger an error. The call to
@@ -166,16 +171,18 @@ public:
   * ### Format
   *
   * The buffer must contain a series of one or more JSON documents, concatenated into a single
-   * buffer, separated by whitespace. It effectively parses until it has a fully valid document,
+   * buffer, separated by ASCII whitespace. It effectively parses until it has a fully valid document,
   * then starts parsing the next document at that point. (It does this with more parallelism and
   * lookahead than you might think, though.)
   *
   * documents that consist of an object or array may omit the whitespace between them, concatenating
-   * with no separator. documents that consist of a single primitive (i.e. documents that are not
-   * arrays or objects) MUST be separated with whitespace.
+   * with no separator. Documents that consist of a single primitive (i.e. documents that are not
+   * arrays or objects) MUST be separated with ASCII whitespace.
+   *
+   * The characters inside a JSON document, and between JSON documents, must be valid Unicode (UTF-8).
   *
   * The documents must not exceed batch_size bytes (by default 1MB) or they will fail to parse.
-   * Setting batch_size to excessively large or excesively small values may impact negatively the
+   * Setting batch_size to excessively large or excessively small values may impact negatively the
   * performance.
   *
   * ### REQUIRED: Buffer Padding
@@ -99,6 +99,7 @@ inline simdjson_result<std::string_view> to_json_string(simdjson_result<SIMDJSON
 }
 } // namespace simdjson

+namespace simdjson { namespace SIMDJSON_IMPLEMENTATION { namespace ondemand {

 #if SIMDJSON_EXCEPTIONS
 inline std::ostream& operator<<(std::ostream& out, simdjson::SIMDJSON_IMPLEMENTATION::ondemand::value x) {
@@ -216,3 +217,4 @@ inline std::ostream& operator<<(std::ostream& out, simdjson::SIMDJSON_IMPLEMENTA
  }
 }
 #endif
+}}} // namespace simdjson::SIMDJSON_IMPLEMENTATION::ondemand
@@ -29,6 +29,14 @@ inline simdjson_result<std::string_view> to_json_string(simdjson_result<SIMDJSON
 inline simdjson_result<std::string_view> to_json_string(simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::array> x);
 } // namespace simdjson

+/**
+ * We want to support argument-dependent lookup (ADL).
+ * Hence we should define operator<< in the namespace
+ * where the argument (here value, object, etc.) resides.
+ * Credit: @madhur4127
+ * See https://github.com/simdjson/simdjson/issues/1768
+ */
+namespace simdjson { namespace SIMDJSON_IMPLEMENTATION { namespace ondemand {

 /**
 * Print JSON to an output stream.
@@ -78,3 +86,4 @@ inline std::ostream& operator<<(std::ostream& out, simdjson::SIMDJSON_IMPLEMENTA
 #if SIMDJSON_EXCEPTIONS
 inline std::ostream& operator<<(std::ostream& out, simdjson::simdjson_result<simdjson::SIMDJSON_IMPLEMENTATION::ondemand::object> x);
 #endif
+}}} // namespace simdjson::SIMDJSON_IMPLEMENTATION::ondemand
@@ -62,6 +62,7 @@ template<> simdjson_really_inline simdjson_result<array> value::get() noexcept {
 template<> simdjson_really_inline simdjson_result<object> value::get() noexcept { return get_object(); }
 template<> simdjson_really_inline simdjson_result<raw_json_string> value::get() noexcept { return get_raw_json_string(); }
 template<> simdjson_really_inline simdjson_result<std::string_view> value::get() noexcept { return get_string(); }
+template<> simdjson_really_inline simdjson_result<number> value::get() noexcept { return get_number(); }
 template<> simdjson_really_inline simdjson_result<double> value::get() noexcept { return get_double(); }
 template<> simdjson_really_inline simdjson_result<uint64_t> value::get() noexcept { return get_uint64(); }
 template<> simdjson_really_inline simdjson_result<int64_t> value::get() noexcept { return get_int64(); }
@@ -114,6 +115,13 @@ simdjson_really_inline simdjson_result<size_t> value::count_elements() & noexcep
  iter.move_at_start();
  return answer;
 }
+simdjson_really_inline simdjson_result<size_t> value::count_fields() & noexcept {
+  simdjson_result<size_t> answer;
+  auto a = get_object();
+  answer = a.count_fields();
+  iter.move_at_start();
+  return answer;
+}
 simdjson_really_inline simdjson_result<value> value::at(size_t index) noexcept {
  auto a = get_array();
  return a.at(index);
@@ -169,6 +177,10 @@ simdjson_really_inline std::string_view value::raw_json_token() noexcept {
  return std::string_view(reinterpret_cast<const char*>(iter.peek_start()), iter.peek_start_length());
 }

+simdjson_really_inline simdjson_result<const char *> value::current_location() noexcept {
+  return iter.json_iter().current_location();
+}
+
 simdjson_really_inline simdjson_result<value> value::at_pointer(std::string_view json_pointer) noexcept {
  json_type t;
  SIMDJSON_TRY(type().get(t));
@@ -207,6 +219,10 @@ simdjson_really_inline simdjson_result<size_t> simdjson_result<SIMDJSON_IMPLEMEN
  if (error()) { return error(); }
  return first.count_elements();
 }
+simdjson_really_inline simdjson_result<size_t> simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value>::count_fields() & noexcept {
+  if (error()) { return error(); }
+  return first.count_fields();
+}
 simdjson_really_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value>::at(size_t index) noexcept {
  if (error()) { return error(); }
  return first.at(index);
@@ -379,6 +395,11 @@ simdjson_really_inline simdjson_result<std::string_view> simdjson_result<SIMDJSO
  return first.raw_json_token();
 }

+simdjson_really_inline simdjson_result<const char *> simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value>::current_location() noexcept {
+  if (error()) { return error(); }
+  return first.current_location();
+}
+
 simdjson_really_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value>::at_pointer(std::string_view json_pointer) noexcept {
  if (error()) { return error(); }
  return first.at_pointer(json_pointer);
@@ -121,6 +121,9 @@ public:
   *
   * Equivalent to get<std::string_view>().
   *
+   * Important: a value should be consumed once. Calling get_string() twice on the same value
+   * is an error.
+   *
   * @returns An UTF-8 string. The string is stored in the parser and will be invalidated the next
   *          time it parses a document or when it is destroyed.
   * @returns INCORRECT_TYPE if the JSON value is not a string.
@@ -244,9 +247,24 @@ public:
   * safe to continue.
   */
  simdjson_really_inline simdjson_result<size_t> count_elements() & noexcept;
+  /**
+   * This method scans the object and counts the number of key-value pairs.
+   * The count_fields method should always be called before you have begun
+   * iterating through the object: it is expected that you are pointing at
+   * the beginning of the object.
+   * The runtime complexity is linear in the size of the object. After
+   * calling this function, if successful, the object is 'rewinded' at its
+   * beginning as if it had never been accessed. If the JSON is malformed (e.g.,
+   * there is a missing comma), then an error is returned and it is no longer
+   * safe to continue.
+   *
+   * To check that an object is empty, it is more performant to use
+   * the is_empty() method on the object instance.
+   */
+  simdjson_really_inline simdjson_result<size_t> count_fields() & noexcept;
  /**
   * Get the value at the given index in the array. This function has linear-time complexity.
-   * This function should only be called once as the array iterator is not reset between each call.
+   * This function should only be called once on an array instance since the array iterator is not reset between each call.
   *
   * @return The value at the given index, or:
   *         - INDEX_OUT_OF_BOUNDS if the array index is larger than an array length
@@ -375,7 +393,7 @@ public:
   * type.
   *
   * number.get_number_type() is number_type::signed_integer if we have
-   * a integer in [-9223372036854775808,9223372036854775808)
+   * an integer in [-9223372036854775808,9223372036854775808)
   * You can recover the value by calling number.get_int64() and you
   * have that number.is_int64() is true.
   *
@@ -425,6 +443,11 @@ public:
   */
  simdjson_really_inline std::string_view raw_json_token() noexcept;

+  /**
+   * Returns the current location in the document if in bounds.
+   */
+  simdjson_really_inline simdjson_result<const char *> current_location() noexcept;
+
  /**
   * Get the value associated with the given JSON pointer.  We use the RFC 6901
   * https://tools.ietf.org/html/rfc6901 standard.
@@ -552,6 +575,7 @@ public:
  simdjson_really_inline operator bool() noexcept(false);
 #endif
  simdjson_really_inline simdjson_result<size_t> count_elements() & noexcept;
+  simdjson_really_inline simdjson_result<size_t> count_fields() & noexcept;
  simdjson_really_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> at(size_t index) noexcept;
  simdjson_really_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::array_iterator> begin() & noexcept;
  simdjson_really_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::array_iterator> end() & noexcept;
@@ -624,6 +648,9 @@ public:
  /** @copydoc simdjson_really_inline std::string_view value::raw_json_token() const noexcept */
  simdjson_really_inline simdjson_result<std::string_view> raw_json_token() noexcept;

+  /** @copydoc simdjson_really_inline simdjson_result<const char *> current_location() noexcept */
+  simdjson_really_inline simdjson_result<const char *> current_location() noexcept;
+
  simdjson_really_inline simdjson_result<SIMDJSON_IMPLEMENTATION::ondemand::value> at_pointer(std::string_view json_pointer) noexcept;
 };

@@ -8,7 +8,7 @@ namespace {
 // We sometimes call trailing_zero on inputs that are zero,
 // but the algorithms do not end up using the returned value.
 // Sadly, sanitizers are not smart enough to figure it out.
-NO_SANITIZE_UNDEFINED
+SIMDJSON_NO_SANITIZE_UNDEFINED
 simdjson_really_inline int trailing_zeroes(uint64_t input_num) {
 #ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
  return (int)_tzcnt_u64(input_num);
@@ -41,11 +41,7 @@
 // has it as a macro.
 #ifndef _blsr_u64
 // we roll our own
-SIMDJSON_TARGET_HASWELL
-static simdjson_really_inline uint64_t _blsr_u64(uint64_t n) {
-  return (n - 1) & n;
-}
-SIMDJSON_UNTARGET_HASWELL
+#define _blsr_u64(n) ((n - 1) & n)
 #endif //  _blsr_u64
 #endif // SIMDJSON_CLANG_VISUAL_STUDIO

@@ -45,7 +45,7 @@ namespace simd {
    simdjson_really_inline base8() : base<simd8<T>>() {}
    simdjson_really_inline base8(const __m256i _value) : base<simd8<T>>(_value) {}

-    simdjson_really_inline Mask operator==(const simd8<T> other) const { return _mm256_cmpeq_epi8(*this, other); }
+    friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return _mm256_cmpeq_epi8(lhs, rhs); }

    static const int SIZE = sizeof(base<T>::value);

@@ -0,0 +1,47 @@
+#ifndef SIMDJSON_ICELAKE_H
+#define SIMDJSON_ICELAKE_H
+
+#include "simdjson/implementation-base.h"
+
+#if SIMDJSON_IMPLEMENTATION_ICELAKE
+
+#if SIMDJSON_CAN_ALWAYS_RUN_ICELAKE
+#define SIMDJSON_TARGET_ICELAKE
+#define SIMDJSON_UNTARGET_ICELAKE
+#else
+#define SIMDJSON_TARGET_ICELAKE SIMDJSON_TARGET_REGION("avx512f,avx512dq,avx512cd,avx512bw,avx512vbmi,avx512vbmi2,avx512vl,avx2,bmi,pclmul,lzcnt")
+#define SIMDJSON_UNTARGET_ICELAKE SIMDJSON_UNTARGET_REGION
+#endif
+
+namespace simdjson {
+/**
+ * Implementation for Icelake (Intel AVX512).
+ */
+namespace icelake {
+} // namespace icelake
+} // namespace simdjson
+
+//
+// These two need to be included outside SIMDJSON_TARGET_ICELAKE
+//
+#include "simdjson/icelake/implementation.h"
+#include "simdjson/icelake/intrinsics.h"
+
+//
+// The rest need to be inside the region
+//
+#include "simdjson/icelake/begin.h"
+
+// Declarations
+#include "simdjson/generic/dom_parser_implementation.h"
+#include "simdjson/icelake/bitmanipulation.h"
+#include "simdjson/icelake/bitmask.h"
+#include "simdjson/icelake/simd.h"
+#include "simdjson/generic/jsoncharutils.h"
+#include "simdjson/generic/atomparsing.h"
+#include "simdjson/icelake/stringparsing.h"
+#include "simdjson/icelake/numberparsing.h"
+#include "simdjson/icelake/end.h"
+
+#endif // SIMDJSON_IMPLEMENTATION_ICELAKE
+#endif // SIMDJSON_ICELAKE_H
@@ -0,0 +1,2 @@
+#define SIMDJSON_IMPLEMENTATION icelake
+SIMDJSON_TARGET_ICELAKE
@@ -0,0 +1,61 @@
+#ifndef SIMDJSON_ICELAKE_BITMANIPULATION_H
+#define SIMDJSON_ICELAKE_BITMANIPULATION_H
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace {
+
+// We sometimes call trailing_zero on inputs that are zero,
+// but the algorithms do not end up using the returned value.
+// Sadly, sanitizers are not smart enough to figure it out.
+SIMDJSON_NO_SANITIZE_UNDEFINED
+simdjson_really_inline int trailing_zeroes(uint64_t input_num) {
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+  return (int)_tzcnt_u64(input_num);
+#else // SIMDJSON_REGULAR_VISUAL_STUDIO
+  ////////
+  // You might expect the next line to be equivalent to
+  // return (int)_tzcnt_u64(input_num);
+  // but the generated code differs and might be less efficient?
+  ////////
+  return __builtin_ctzll(input_num);
+#endif // SIMDJSON_REGULAR_VISUAL_STUDIO
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_really_inline uint64_t clear_lowest_bit(uint64_t input_num) {
+  return _blsr_u64(input_num);
+}
+
+/* result might be undefined when input_num is zero */
+simdjson_really_inline int leading_zeroes(uint64_t input_num) {
+  return int(_lzcnt_u64(input_num));
+}
+
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+simdjson_really_inline unsigned __int64 count_ones(uint64_t input_num) {
+  // note: we do not support legacy 32-bit Windows
+  return __popcnt64(input_num);// Visual Studio wants two underscores
+}
+#else
+simdjson_really_inline long long int count_ones(uint64_t input_num) {
+  return _popcnt64(input_num);
+}
+#endif
+
+simdjson_really_inline bool add_overflow(uint64_t value1, uint64_t value2,
+                                uint64_t *result) {
+#ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
+  return _addcarry_u64(0, value1, value2,
+                       reinterpret_cast<unsigned __int64 *>(result));
+#else
+  return __builtin_uaddll_overflow(value1, value2,
+                                   reinterpret_cast<unsigned long long *>(result));
+#endif
+}
+
+} // unnamed namespace
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_ICELAKE_BITMANIPULATION_H
@@ -0,0 +1,25 @@
+#ifndef SIMDJSON_ICELAKE_BITMASK_H
+#define SIMDJSON_ICELAKE_BITMASK_H
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace {
+
+//
+// Perform a "cumulative bitwise xor," flipping bits each time a 1 is encountered.
+//
+// For example, prefix_xor(00100100) == 00011100
+//
+simdjson_really_inline uint64_t prefix_xor(const uint64_t bitmask) {
+  // There should be no such thing with a processor supporting avx2
+  // but not clmul.
+  __m128i all_ones = _mm_set1_epi8('\xFF');
+  __m128i result = _mm_clmulepi64_si128(_mm_set_epi64x(0ULL, bitmask), all_ones, 0);
+  return _mm_cvtsi128_si64(result);
+}
+
+} // unnamed namespace
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_ICELAKE_BITMASK_H
@@ -0,0 +1,2 @@
+SIMDJSON_UNTARGET_ICELAKE
+#undef SIMDJSON_IMPLEMENTATION
@@ -0,0 +1,31 @@
+#ifndef SIMDJSON_ICELAKE_IMPLEMENTATION_H
+#define SIMDJSON_ICELAKE_IMPLEMENTATION_H
+
+#include "simdjson/implementation.h"
+
+// The constructor may be executed on any host, so we take care not to use SIMDJSON_TARGET_ICELAKE
+namespace simdjson {
+namespace icelake {
+
+using namespace simdjson;
+
+class implementation final : public simdjson::implementation {
+public:
+  simdjson_really_inline implementation() : simdjson::implementation(
+      "icelake",
+      "Intel/AMD AVX512",
+      internal::instruction_set::AVX2 | internal::instruction_set::PCLMULQDQ | internal::instruction_set::BMI1 | internal::instruction_set::BMI2 | internal::instruction_set::AVX512F | internal::instruction_set::AVX512DQ | internal::instruction_set::AVX512CD | internal::instruction_set::AVX512BW | internal::instruction_set::AVX512VL | internal::instruction_set::AVX512VBMI2
+  ) {}
+  simdjson_warn_unused error_code create_dom_parser_implementation(
+    size_t capacity,
+    size_t max_length,
+    std::unique_ptr<internal::dom_parser_implementation>& dst
+  ) const noexcept final;
+  simdjson_warn_unused error_code minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept final;
+  simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) const noexcept final;
+};
+
+} // namespace icelake
+} // namespace simdjson
+
+#endif // SIMDJSON_ICELAKE_IMPLEMENTATION_H
@@ -0,0 +1,56 @@
+#ifndef SIMDJSON_ICELAKE_INTRINSICS_H
+#define SIMDJSON_ICELAKE_INTRINSICS_H
+
+#include "simdjson/base.h"
+
+#ifdef SIMDJSON_VISUAL_STUDIO
+// under clang within visual studio, this will include <x86intrin.h>
+#include <intrin.h>  // visual studio or clang
+#else
+#include <x86intrin.h> // elsewhere
+#endif // SIMDJSON_VISUAL_STUDIO
+
+#ifdef SIMDJSON_CLANG_VISUAL_STUDIO
+/**
+ * You are not supposed, normally, to include these
+ * headers directly. Instead you should either include intrin.h
+ * or x86intrin.h. However, when compiling with clang
+ * under Windows (i.e., when _MSC_VER is set), these headers
+ * only get included *if* the corresponding features are detected
+ * from macros:
+ * e.g., if __AVX2__ is set... in turn,  we normally set these
+ * macros by compiling against the corresponding architecture
+ * (e.g., arch:AVX2, -mavx2, etc.) which compiles the whole
+ * software with these advanced instructions. In simdjson, we
+ * want to compile the whole program for a generic target,
+ * and only target our specific kernels. As a workaround,
+ * we directly include the needed headers. These headers would
+ * normally guard against such usage, but we carefully included
+ * <x86intrin.h>  (or <intrin.h>) before, so the headers
+ * are fooled.
+ */
+#include <bmiintrin.h>   // for _blsr_u64
+#include <lzcntintrin.h> // for  __lzcnt64
+#include <immintrin.h>   // for most things (AVX2, AVX512, _popcnt64)
+#include <smmintrin.h>
+#include <tmmintrin.h>
+#include <avxintrin.h>
+#include <avx2intrin.h>
+#include <wmmintrin.h>   // for  _mm_clmulepi64_si128
+// Important: we need the AVX-512 headers:
+#include <avx512fintrin.h>
+#include <avx512dqintrin.h>
+#include <avx512cdintrin.h>
+#include <avx512bwintrin.h>
+#include <avx512vlintrin.h>
+#include <avx512vbmiintrin.h>
+#include <avx512vbmi2intrin.h>
+// unfortunately, we may not get _blsr_u64, but, thankfully, clang
+// has it as a macro.
+#ifndef _blsr_u64
+// we roll our own
+#define _blsr_u64(n) ((n - 1) & n)
+#endif //  _blsr_u64
+#endif // SIMDJSON_CLANG_VISUAL_STUDIO
+
+#endif // SIMDJSON_ICELAKE_INTRINSICS_H
@@ -0,0 +1,34 @@
+#ifndef SIMDJSON_ICELAKE_NUMBERPARSING_H
+#define SIMDJSON_ICELAKE_NUMBERPARSING_H
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace {
+
+static simdjson_really_inline uint32_t parse_eight_digits_unrolled(const uint8_t *chars) {
+  // this actually computes *16* values so we are being wasteful.
+  const __m128i ascii0 = _mm_set1_epi8('0');
+  const __m128i mul_1_10 =
+      _mm_setr_epi8(10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1, 10, 1);
+  const __m128i mul_1_100 = _mm_setr_epi16(100, 1, 100, 1, 100, 1, 100, 1);
+  const __m128i mul_1_10000 =
+      _mm_setr_epi16(10000, 1, 10000, 1, 10000, 1, 10000, 1);
+  const __m128i input = _mm_sub_epi8(
+      _mm_loadu_si128(reinterpret_cast<const __m128i *>(chars)), ascii0);
+  const __m128i t1 = _mm_maddubs_epi16(input, mul_1_10);
+  const __m128i t2 = _mm_madd_epi16(t1, mul_1_100);
+  const __m128i t3 = _mm_packus_epi32(t2, t2);
+  const __m128i t4 = _mm_madd_epi16(t3, mul_1_10000);
+  return _mm_cvtsi128_si32(
+      t4); // only captures the sum of the first 8 digits, drop the rest
+}
+
+} // unnamed namespace
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#define SIMDJSON_SWAR_NUMBER_PARSING 1
+
+#include "simdjson/generic/numberparsing.h"
+
+#endif // SIMDJSON_ICELAKE_NUMBERPARSING_H
@@ -0,0 +1,341 @@
+#ifndef SIMDJSON_ICELAKE_SIMD_H
+#define SIMDJSON_ICELAKE_SIMD_H
+
+#include "simdjson/internal/simdprune_tables.h"
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace {
+namespace simd {
+
+  // Forward-declared so they can be used by splat and friends.
+  template<typename Child>
+  struct base {
+    __m512i value;
+
+    // Zero constructor
+    simdjson_really_inline base() : value{__m512i()} {}
+
+    // Conversion from SIMD register
+    simdjson_really_inline base(const __m512i _value) : value(_value) {}
+
+    // Conversion to SIMD register
+    simdjson_really_inline operator const __m512i&() const { return this->value; }
+    simdjson_really_inline operator __m512i&() { return this->value; }
+
+    // Bit operations
+    simdjson_really_inline Child operator|(const Child other) const { return _mm512_or_si512(*this, other); }
+    simdjson_really_inline Child operator&(const Child other) const { return _mm512_and_si512(*this, other); }
+    simdjson_really_inline Child operator^(const Child other) const { return _mm512_xor_si512(*this, other); }
+    simdjson_really_inline Child bit_andnot(const Child other) const { return _mm512_andnot_si512(other, *this); }
+    simdjson_really_inline Child& operator|=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast | other; return *this_cast; }
+    simdjson_really_inline Child& operator&=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast & other; return *this_cast; }
+    simdjson_really_inline Child& operator^=(const Child other) { auto this_cast = static_cast<Child*>(this); *this_cast = *this_cast ^ other; return *this_cast; }
+  };
+
+  // Forward-declared so they can be used by splat and friends.
+  template<typename T>
+  struct simd8;
+
+  template<typename T, typename Mask=simd8<bool>>
+  struct base8: base<simd8<T>> {
+    typedef uint32_t bitmask_t;
+    typedef uint64_t bitmask2_t;
+
+    simdjson_really_inline base8() : base<simd8<T>>() {}
+    simdjson_really_inline base8(const __m512i _value) : base<simd8<T>>(_value) {}
+
+    simdjson_really_inline uint64_t operator==(const simd8<T> other) const { return _mm512_cmpeq_epi8_mask(*this, other); }
+
+    static const int SIZE = sizeof(base<T>::value);
+
+    template<int N=1>
+    simdjson_really_inline simd8<T> prev(const simd8<T> prev_chunk) const {
+      return _mm512_alignr_epi8(*this, _mm512_permutex2var_epi64(prev_chunk, _mm512_set_epi64(13, 12, 11, 10, 9, 8, 7, 6), *this), 16 - N);
+    }
+  };
+
+  // SIMD byte mask type (returned by things like eq and gt)
+  template<>
+  struct simd8<bool>: base8<bool> {
+    static simdjson_really_inline simd8<bool> splat(bool _value) { return _mm512_set1_epi8(uint8_t(-(!!_value))); }
+
+    simdjson_really_inline simd8<bool>() : base8() {}
+    simdjson_really_inline simd8<bool>(const __m512i _value) : base8<bool>(_value) {}
+    // Splat constructor
+    simdjson_really_inline simd8<bool>(bool _value) : base8<bool>(splat(_value)) {}
+    simdjson_really_inline bool any() const { return !!_mm512_test_epi8_mask (*this, *this); }
+    simdjson_really_inline simd8<bool> operator~() const { return *this ^ true; }
+  };
+
+  template<typename T>
+  struct base8_numeric: base8<T> {
+    static simdjson_really_inline simd8<T> splat(T _value) { return _mm512_set1_epi8(_value); }
+    static simdjson_really_inline simd8<T> zero() { return _mm512_setzero_si512(); }
+    static simdjson_really_inline simd8<T> load(const T values[64]) {
+      return _mm512_loadu_si512(reinterpret_cast<const __m512i *>(values));
+    }
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    static simdjson_really_inline simd8<T> repeat_16(
+      T v0,  T v1,  T v2,  T v3,  T v4,  T v5,  T v6,  T v7,
+      T v8,  T v9,  T v10, T v11, T v12, T v13, T v14, T v15
+    ) {
+      return simd8<T>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    simdjson_really_inline base8_numeric() : base8<T>() {}
+    simdjson_really_inline base8_numeric(const __m512i _value) : base8<T>(_value) {}
+
+    // Store to array
+    simdjson_really_inline void store(T dst[64]) const { return _mm512_storeu_si512(reinterpret_cast<__m512i *>(dst), *this); }
+
+    // Addition/subtraction are the same for signed and unsigned
+    simdjson_really_inline simd8<T> operator+(const simd8<T> other) const { return _mm512_add_epi8(*this, other); }
+    simdjson_really_inline simd8<T> operator-(const simd8<T> other) const { return _mm512_sub_epi8(*this, other); }
+    simdjson_really_inline simd8<T>& operator+=(const simd8<T> other) { *this = *this + other; return *static_cast<simd8<T>*>(this); }
+    simdjson_really_inline simd8<T>& operator-=(const simd8<T> other) { *this = *this - other; return *static_cast<simd8<T>*>(this); }
+
+    // Override to distinguish from bool version
+    simdjson_really_inline simd8<T> operator~() const { return *this ^ 0xFFu; }
+
+    // Perform a lookup assuming the value is between 0 and 16 (undefined behavior for out of range values)
+    template<typename L>
+    simdjson_really_inline simd8<L> lookup_16(simd8<L> lookup_table) const {
+      return _mm512_shuffle_epi8(lookup_table, *this);
+    }
+
+    // Copies to 'output" all bytes corresponding to a 0 in the mask (interpreted as a bitset).
+    // Passing a 0 value for mask would be equivalent to writing out every byte to output.
+    // Only the first 32 - count_ones(mask) bytes of the result are significant but 32 bytes
+    // get written.
+    // Design consideration: it seems like a function with the
+    // signature simd8<L> compress(uint32_t mask) would be
+    // sensible, but the AVX ISA makes this kind of approach difficult.
+    template<typename L>
+    simdjson_really_inline void compress(uint64_t mask, L * output) const {
+      _mm512_mask_compressstoreu_epi8 (output,~mask,*this);
+    }
+
+    template<typename L>
+    simdjson_really_inline simd8<L> lookup_16(
+        L replace0,  L replace1,  L replace2,  L replace3,
+        L replace4,  L replace5,  L replace6,  L replace7,
+        L replace8,  L replace9,  L replace10, L replace11,
+        L replace12, L replace13, L replace14, L replace15) const {
+      return lookup_16(simd8<L>::repeat_16(
+        replace0,  replace1,  replace2,  replace3,
+        replace4,  replace5,  replace6,  replace7,
+        replace8,  replace9,  replace10, replace11,
+        replace12, replace13, replace14, replace15
+      ));
+    }
+  };
+
+  // Signed bytes
+  template<>
+  struct simd8<int8_t> : base8_numeric<int8_t> {
+    simdjson_really_inline simd8() : base8_numeric<int8_t>() {}
+    simdjson_really_inline simd8(const __m512i _value) : base8_numeric<int8_t>(_value) {}
+    // Splat constructor
+    simdjson_really_inline simd8(int8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_really_inline simd8(const int8_t values[64]) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_really_inline simd8(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15,
+      int8_t v16, int8_t v17, int8_t v18, int8_t v19, int8_t v20, int8_t v21, int8_t v22, int8_t v23,
+      int8_t v24, int8_t v25, int8_t v26, int8_t v27, int8_t v28, int8_t v29, int8_t v30, int8_t v31,
+      int8_t v32, int8_t v33, int8_t v34, int8_t v35, int8_t v36, int8_t v37, int8_t v38, int8_t v39,
+      int8_t v40, int8_t v41, int8_t v42, int8_t v43, int8_t v44, int8_t v45, int8_t v46, int8_t v47,
+      int8_t v48, int8_t v49, int8_t v50, int8_t v51, int8_t v52, int8_t v53, int8_t v54, int8_t v55,
+      int8_t v56, int8_t v57, int8_t v58, int8_t v59, int8_t v60, int8_t v61, int8_t v62, int8_t v63
+    ) : simd8(_mm512_set_epi8(
+      v63, v62, v61, v60, v59, v58, v57, v56,
+      v55, v54, v53, v52, v51, v50, v49, v48,
+      v47, v46, v45, v44, v43, v42, v41, v40,
+      v39, v38, v37, v36, v35, v34, v33, v32,
+      v31, v30, v29, v28, v27, v26, v25, v24,
+      v23, v22, v21, v20, v19, v18, v17, v16,
+      v15, v14, v13, v12, v11, v10,  v9,  v8,
+       v7,  v6,  v5,  v4,  v3,  v2,  v1,  v0
+    )) {}
+
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_really_inline static simd8<int8_t> repeat_16(
+      int8_t v0,  int8_t v1,  int8_t v2,  int8_t v3,  int8_t v4,  int8_t v5,  int8_t v6,  int8_t v7,
+      int8_t v8,  int8_t v9,  int8_t v10, int8_t v11, int8_t v12, int8_t v13, int8_t v14, int8_t v15
+    ) {
+      return simd8<int8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Order-sensitive comparisons
+    simdjson_really_inline simd8<int8_t> max_val(const simd8<int8_t> other) const { return _mm512_max_epi8(*this, other); }
+    simdjson_really_inline simd8<int8_t> min_val(const simd8<int8_t> other) const { return _mm512_min_epi8(*this, other); }
+
+    simdjson_really_inline simd8<bool> operator>(const simd8<int8_t> other) const { return _mm512_maskz_abs_epi8(_mm512_cmpgt_epi8_mask(*this, other),_mm512_set1_epi8(uint8_t(0x80))); }
+    simdjson_really_inline simd8<bool> operator<(const simd8<int8_t> other) const { return _mm512_maskz_abs_epi8(_mm512_cmpgt_epi8_mask(other, *this),_mm512_set1_epi8(uint8_t(0x80))); }
+  };
+
+  // Unsigned bytes
+  template<>
+  struct simd8<uint8_t>: base8_numeric<uint8_t> {
+    simdjson_really_inline simd8() : base8_numeric<uint8_t>() {}
+    simdjson_really_inline simd8(const __m512i _value) : base8_numeric<uint8_t>(_value) {}
+    // Splat constructor
+    simdjson_really_inline simd8(uint8_t _value) : simd8(splat(_value)) {}
+    // Array constructor
+    simdjson_really_inline simd8(const uint8_t values[64]) : simd8(load(values)) {}
+    // Member-by-member initialization
+    simdjson_really_inline simd8(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15,
+      uint8_t v16, uint8_t v17, uint8_t v18, uint8_t v19, uint8_t v20, uint8_t v21, uint8_t v22, uint8_t v23,
+      uint8_t v24, uint8_t v25, uint8_t v26, uint8_t v27, uint8_t v28, uint8_t v29, uint8_t v30, uint8_t v31,
+      uint8_t v32, uint8_t v33, uint8_t v34, uint8_t v35, uint8_t v36, uint8_t v37, uint8_t v38, uint8_t v39,
+      uint8_t v40, uint8_t v41, uint8_t v42, uint8_t v43, uint8_t v44, uint8_t v45, uint8_t v46, uint8_t v47,
+      uint8_t v48, uint8_t v49, uint8_t v50, uint8_t v51, uint8_t v52, uint8_t v53, uint8_t v54, uint8_t v55,
+      uint8_t v56, uint8_t v57, uint8_t v58, uint8_t v59, uint8_t v60, uint8_t v61, uint8_t v62, uint8_t v63
+    ) : simd8(_mm512_set_epi8(
+      v63, v62, v61, v60, v59, v58, v57, v56,
+      v55, v54, v53, v52, v51, v50, v49, v48,
+      v47, v46, v45, v44, v43, v42, v41, v40,
+      v39, v38, v37, v36, v35, v34, v33, v32,
+      v31, v30, v29, v28, v27, v26, v25, v24,
+      v23, v22, v21, v20, v19, v18, v17, v16,
+      v15, v14, v13, v12, v11, v10,  v9,  v8,
+       v7,  v6,  v5,  v4,  v3,  v2,  v1,  v0
+    )) {}
+
+    // Repeat 16 values as many times as necessary (usually for lookup tables)
+    simdjson_really_inline static simd8<uint8_t> repeat_16(
+      uint8_t v0,  uint8_t v1,  uint8_t v2,  uint8_t v3,  uint8_t v4,  uint8_t v5,  uint8_t v6,  uint8_t v7,
+      uint8_t v8,  uint8_t v9,  uint8_t v10, uint8_t v11, uint8_t v12, uint8_t v13, uint8_t v14, uint8_t v15
+    ) {
+      return simd8<uint8_t>(
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15,
+        v0, v1, v2, v3, v4, v5, v6, v7,
+        v8, v9, v10,v11,v12,v13,v14,v15
+      );
+    }
+
+    // Saturated math
+    simdjson_really_inline simd8<uint8_t> saturating_add(const simd8<uint8_t> other) const { return _mm512_adds_epu8(*this, other); }
+    simdjson_really_inline simd8<uint8_t> saturating_sub(const simd8<uint8_t> other) const { return _mm512_subs_epu8(*this, other); }
+
+    // Order-specific operations
+    simdjson_really_inline simd8<uint8_t> max_val(const simd8<uint8_t> other) const { return _mm512_max_epu8(*this, other); }
+    simdjson_really_inline simd8<uint8_t> min_val(const simd8<uint8_t> other) const { return _mm512_min_epu8(other, *this); }
+    // Same as >, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_really_inline simd8<uint8_t> gt_bits(const simd8<uint8_t> other) const { return this->saturating_sub(other); }
+    // Same as <, but only guarantees true is nonzero (< guarantees true = -1)
+    simdjson_really_inline simd8<uint8_t> lt_bits(const simd8<uint8_t> other) const { return other.saturating_sub(*this); }
+    simdjson_really_inline uint64_t operator<=(const simd8<uint8_t> other) const { return other.max_val(*this) == other; }
+    simdjson_really_inline uint64_t operator>=(const simd8<uint8_t> other) const { return other.min_val(*this) == other; }
+    simdjson_really_inline simd8<bool> operator>(const simd8<uint8_t> other) const { return this->gt_bits(other).any_bits_set(); }
+    simdjson_really_inline simd8<bool> operator<(const simd8<uint8_t> other) const { return this->lt_bits(other).any_bits_set(); }
+
+    // Bit-specific operations
+    simdjson_really_inline simd8<bool> bits_not_set() const { return _mm512_mask_blend_epi8(*this == uint8_t(0), _mm512_set1_epi8(0), _mm512_set1_epi8(-1)); }
+    simdjson_really_inline simd8<bool> bits_not_set(simd8<uint8_t> bits) const { return (*this & bits).bits_not_set(); }
+    simdjson_really_inline simd8<bool> any_bits_set() const { return ~this->bits_not_set(); }
+    simdjson_really_inline simd8<bool> any_bits_set(simd8<uint8_t> bits) const { return ~this->bits_not_set(bits); }
+
+    simdjson_really_inline bool is_ascii() const { return _mm512_movepi8_mask(*this) == 0; }
+    simdjson_really_inline bool bits_not_set_anywhere() const {
+      return !_mm512_test_epi8_mask(*this, *this);
+    }
+    simdjson_really_inline bool any_bits_set_anywhere() const { return !bits_not_set_anywhere(); }
+    simdjson_really_inline bool bits_not_set_anywhere(simd8<uint8_t> bits) const { return !_mm512_test_epi8_mask(*this, bits); }
+    simdjson_really_inline bool any_bits_set_anywhere(simd8<uint8_t> bits) const { return !bits_not_set_anywhere(bits); }
+    template<int N>
+    simdjson_really_inline simd8<uint8_t> shr() const { return simd8<uint8_t>(_mm512_srli_epi16(*this, N)) & uint8_t(0xFFu >> N); }
+    template<int N>
+    simdjson_really_inline simd8<uint8_t> shl() const { return simd8<uint8_t>(_mm512_slli_epi16(*this, N)) & uint8_t(0xFFu << N); }
+    // Get one of the bits and make a bitmask out of it.
+    // e.g. value.get_bit<7>() gets the high bit
+    template<int N>
+    simdjson_really_inline uint64_t get_bit() const { return _mm512_movepi8_mask(_mm512_slli_epi16(*this, 7-N)); }
+  };
+
+  template<typename T>
+  struct simd8x64 {
+    static constexpr int NUM_CHUNKS = 64 / sizeof(simd8<T>);
+    static_assert(NUM_CHUNKS == 1, "Icelake kernel should use one register per 64-byte block.");
+    const simd8<T> chunks[NUM_CHUNKS];
+
+    simd8x64(const simd8x64<T>& o) = delete; // no copy allowed
+    simd8x64<T>& operator=(const simd8<T>& other) = delete; // no assignment allowed
+    simd8x64() = delete; // no default constructor allowed
+
+    simdjson_really_inline simd8x64(const simd8<T> chunk0, const simd8<T> chunk1) : chunks{chunk0, chunk1} {}
+    simdjson_really_inline simd8x64(const simd8<T> chunk0) : chunks{chunk0} {}
+    simdjson_really_inline simd8x64(const T ptr[64]) : chunks{simd8<T>::load(ptr)} {}
+
+    simdjson_really_inline uint64_t compress(uint64_t mask, T * output) const {
+      this->chunks[0].compress(mask, output);
+      return 64 - count_ones(mask);
+    }
+
+    simdjson_really_inline void store(T ptr[64]) const {
+      this->chunks[0].store(ptr+sizeof(simd8<T>)*0);
+    }
+
+    simdjson_really_inline simd8<T> reduce_or() const {
+      return this->chunks[0];
+    }
+
+    simdjson_really_inline simd8x64<T> bit_or(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return simd8x64<T>(
+        this->chunks[0] | mask
+      );
+    }
+
+    simdjson_really_inline uint64_t eq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return this->chunks[0] == mask;
+    }
+
+    simdjson_really_inline uint64_t eq(const simd8x64<uint8_t> &other) const {
+      return this->chunks[0] == other.chunks[0];
+    }
+
+    simdjson_really_inline uint64_t lteq(const T m) const {
+      const simd8<T> mask = simd8<T>::splat(m);
+      return this->chunks[0] <= mask;
+    }
+    simdjson_really_inline operator __m512i() const { return __m512i(this->chunks[0]); }
+
+  }; // struct simd8x64<T>
+
+} // namespace simd
+
+} // unnamed namespace
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#endif // SIMDJSON_ICELAKE_SIMD_H
@@ -0,0 +1,48 @@
+#ifndef SIMDJSON_ICELAKE_STRINGPARSING_H
+#define SIMDJSON_ICELAKE_STRINGPARSING_H
+
+#include "simdjson/base.h"
+#include "simdjson/icelake/simd.h"
+#include "simdjson/icelake/bitmanipulation.h"
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace {
+
+using namespace simd;
+
+// Holds backslashes and quotes locations.
+struct backslash_and_quote {
+public:
+  static constexpr uint32_t BYTES_PROCESSED = 32;
+  simdjson_really_inline static backslash_and_quote copy_and_find(const uint8_t *src, uint8_t *dst);
+
+  simdjson_really_inline bool has_quote_first() { return ((bs_bits - 1) & quote_bits) != 0; }
+  simdjson_really_inline bool has_backslash() { return ((quote_bits - 1) & bs_bits) != 0; }
+  simdjson_really_inline int quote_index() { return trailing_zeroes(quote_bits); }
+  simdjson_really_inline int backslash_index() { return trailing_zeroes(bs_bits); }
+
+  uint64_t bs_bits;
+  uint64_t quote_bits;
+}; // struct backslash_and_quote
+
+simdjson_really_inline backslash_and_quote backslash_and_quote::copy_and_find(const uint8_t *src, uint8_t *dst) {
+  // this can read up to 15 bytes beyond the buffer size, but we require
+  // SIMDJSON_PADDING of padding
+  static_assert(SIMDJSON_PADDING >= (BYTES_PROCESSED - 1), "backslash and quote finder must process fewer than SIMDJSON_PADDING bytes");
+  simd8<uint8_t> v(src);
+  // store to dest unconditionally - we can overwrite the bits we don't like later
+  v.store(dst);
+  return {
+      static_cast<uint64_t>(v == '\\'), // bs_bits
+      static_cast<uint64_t>(v == '"'), // quote_bits
+  };
+}
+
+} // unnamed namespace
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#include "simdjson/generic/stringparsing.h"
+
+#endif // SIMDJSON_ICELAKE_STRINGPARSING_H
@@ -56,7 +56,7 @@ public:
  /**
   * The name of this implementation.
   *
-   *     const implementation *impl = simdjson::active_implementation;
+   *     const implementation *impl = simdjson::get_active_implementation();
   *     cout << "simdjson is optimized for " << impl->name() << "(" << impl->description() << ")" << endl;
   *
   * @return the name of the implementation, e.g. "haswell", "westmere", "arm64"
@@ -66,7 +66,7 @@ public:
  /**
   * The description of this implementation.
   *
-   *     const implementation *impl = simdjson::active_implementation;
+   *     const implementation *impl = simdjson::get_active_implementation();
   *     cout << "simdjson is optimized for " << impl->name() << "(" << impl->description() << ")" << endl;
   *
   * @return the name of the implementation, e.g. "haswell", "westmere", "arm64"
@@ -95,7 +95,7 @@ public:
  /**
   * @private For internal implementation use
   *
-   *     const implementation *impl = simdjson::active_implementation;
+   *     const implementation *impl = simdjson::get_active_implementation();
   *     cout << "simdjson is optimized for " << impl->name() << "(" << impl->description() << ")" << endl;
   *
   * @param capacity The largest document that will be passed to the parser.
@@ -189,10 +189,10 @@ public:
   *
   * Case sensitive.
   *
-   *     const implementation *impl = simdjson::available_implementations["westmere"];
+   *     const implementation *impl = simdjson::get_available_implementations()["westmere"];
   *     if (!impl) { exit(1); }
   *     if (!imp->supported_by_runtime_system()) { exit(1); }
-   *     simdjson::active_implementation = impl;
+   *     simdjson::get_active_implementation() = impl;
   *
   * @param name the implementation to find, e.g. "westmere", "haswell", "arm64"
   * @return the implementation, or nullptr if the parse failed.
@@ -210,7 +210,7 @@ public:
   * This is used to initialize the implementation on startup.
   *
   *     const implementation *impl = simdjson::available_implementation::detect_best_supported();
-   *     simdjson::active_implementation = impl;
+   *     simdjson::get_active_implementation() = impl;
   *
   * @return the most advanced supported implementation for the current host, or an
   *         implementation that returns UNSUPPORTED_ARCHITECTURE if there is no supported
@@ -242,14 +242,14 @@ private:
 /**
 * The list of available implementations compiled into simdjson.
 */
-extern SIMDJSON_DLLIMPORTEXPORT const internal::available_implementation_list available_implementations;
+extern SIMDJSON_DLLIMPORTEXPORT const internal::available_implementation_list& get_available_implementations();

 /**
  * The active implementation.
  *
  * Automatically initialized on first use to the most advanced implementation supported by this hardware.
  */
-extern SIMDJSON_DLLIMPORTEXPORT internal::atomic_ptr<const implementation> active_implementation;
+extern SIMDJSON_DLLIMPORTEXPORT internal::atomic_ptr<const implementation>& get_active_implementation();

 } // namespace simdjson

@@ -13,16 +13,42 @@
 #endif
 #define SIMDJSON_CAN_ALWAYS_RUN_ARM64 SIMDJSON_IMPLEMENTATION_ARM64 && SIMDJSON_IS_ARM64

+#ifdef __has_include
+// How do we detect that a compiler supports vbmi2?
+// For sure if the following header is found, we are ok?
+#if __has_include(<avx512vbmi2intrin.h>)
+#define SIMDJSON_COMPILER_SUPPORTS_VBMI2 1
+#endif
+#endif
+
+// Default Icelake to on if this is x86-64. Even if we're not compiled for it, it could be selected
+// at runtime.
+#ifndef SIMDJSON_IMPLEMENTATION_ICELAKE
+#define SIMDJSON_IMPLEMENTATION_ICELAKE ((SIMDJSON_IS_X86_64) && (SIMDJSON_AVX512_ALLOWED) && (SIMDJSON_COMPILER_SUPPORTS_VBMI2))
+#endif
+
+#ifdef _MSC_VER
+// To see why  (__BMI__) && (__PCLMUL__) && (__LZCNT__) are not part of this next line, see
+// https://github.com/simdjson/simdjson/issues/1247
+#define SIMDJSON_CAN_ALWAYS_RUN_ICELAKE ((SIMDJSON_IMPLEMENTATION_ICELAKE) && (__AVX2__) && (__AVX512F__) && (__AVX512DQ__) && (__AVX512CD__) && (__AVX512BW__) && (__AVX512VL__) && (__AVX512VBMI2__))
+#else
+#define SIMDJSON_CAN_ALWAYS_RUN_ICELAKE ((SIMDJSON_IMPLEMENTATION_ICELAKE) && (__AVX2__) && (__BMI__) && (__PCLMUL__) && (__LZCNT__) && (__AVX512F__) && (__AVX512DQ__) && (__AVX512CD__) && (__AVX512BW__) && (__AVX512VL__) && (__AVX512VBMI2__))
+#endif
+
 // Default Haswell to on if this is x86-64. Even if we're not compiled for it, it could be selected
 // at runtime.
 #ifndef SIMDJSON_IMPLEMENTATION_HASWELL
-#define SIMDJSON_IMPLEMENTATION_HASWELL (SIMDJSON_IS_X86_64)
+#define SIMDJSON_IMPLEMENTATION_HASWELL SIMDJSON_IS_X86_64
 #endif
+#ifdef _MSC_VER
 // To see why  (__BMI__) && (__PCLMUL__) && (__LZCNT__) are not part of this next line, see
 // https://github.com/simdjson/simdjson/issues/1247
 #define SIMDJSON_CAN_ALWAYS_RUN_HASWELL ((SIMDJSON_IMPLEMENTATION_HASWELL) && (SIMDJSON_IS_X86_64) && (__AVX2__))
+#else
+#define SIMDJSON_CAN_ALWAYS_RUN_HASWELL ((SIMDJSON_IMPLEMENTATION_HASWELL) && (SIMDJSON_IS_X86_64) && (__AVX2__) && (__BMI__) && (__PCLMUL__) && (__LZCNT__))
+#endif

-// Default Westmere to on if this is x86-64, unless we'll always select Haswell.
+// Default Westmere to on if this is x86-64. Note that the macro SIMDJSON_REQUIRES_HASWELL appears unused.
 #ifndef SIMDJSON_IMPLEMENTATION_WESTMERE
 #define SIMDJSON_IMPLEMENTATION_WESTMERE (SIMDJSON_IS_X86_64 && !SIMDJSON_REQUIRES_HASWELL)
 #endif
@@ -45,6 +71,7 @@ SIMDJSON_DISABLE_UNDESIRED_WARNINGS
 // Implementations
 #include "simdjson/arm64.h"
 #include "simdjson/fallback.h"
+#include "simdjson/icelake.h"
 #include "simdjson/haswell.h"
 #include "simdjson/ppc64.h"
 #include "simdjson/westmere.h"
@@ -66,7 +66,16 @@ enum instruction_set {
  PCLMULQDQ = 0x10,
  BMI1 = 0x20,
  BMI2 = 0x40,
-  ALTIVEC = 0x80
+  ALTIVEC = 0x80,
+  AVX512F = 0x100,
+  AVX512DQ = 0x200,
+  AVX512IFMA = 0x400,
+  AVX512PF = 0x800,
+  AVX512ER = 0x1000,
+  AVX512CD = 0x2000,
+  AVX512BW = 0x4000,
+  AVX512VL = 0x8000,
+  AVX512VBMI2 = 0x10000
 };

 #if defined(__PPC64__)
@@ -96,11 +105,20 @@ static inline uint32_t detect_supported_architectures() {

 namespace {
 // Can be found on Intel ISA Reference for CPUID
-constexpr uint32_t cpuid_avx2_bit = 1 << 5;      ///< @private Bit 5 of EBX for EAX=0x7
-constexpr uint32_t cpuid_bmi1_bit = 1 << 3;      ///< @private bit 3 of EBX for EAX=0x7
-constexpr uint32_t cpuid_bmi2_bit = 1 << 8;      ///< @private bit 8 of EBX for EAX=0x7
-constexpr uint32_t cpuid_sse42_bit = 1 << 20;    ///< @private bit 20 of ECX for EAX=0x1
-constexpr uint32_t cpuid_pclmulqdq_bit = 1 << 1; ///< @private bit  1 of ECX for EAX=0x1
+constexpr uint32_t cpuid_avx2_bit = 1 << 5;         ///< @private Bit 5 of EBX for EAX=0x7
+constexpr uint32_t cpuid_bmi1_bit = 1 << 3;         ///< @private bit 3 of EBX for EAX=0x7
+constexpr uint32_t cpuid_bmi2_bit = 1 << 8;         ///< @private bit 8 of EBX for EAX=0x7
+constexpr uint32_t cpuid_avx512f_bit = 1 << 16;     ///< @private bit 16 of EBX for EAX=0x7
+constexpr uint32_t cpuid_avx512dq_bit = 1 << 17;    ///< @private bit 17 of EBX for EAX=0x7
+constexpr uint32_t cpuid_avx512ifma_bit = 1 << 21;  ///< @private bit 21 of EBX for EAX=0x7
+constexpr uint32_t cpuid_avx512pf_bit = 1 << 26;    ///< @private bit 26 of EBX for EAX=0x7
+constexpr uint32_t cpuid_avx512er_bit = 1 << 27;    ///< @private bit 27 of EBX for EAX=0x7
+constexpr uint32_t cpuid_avx512cd_bit = 1 << 28;    ///< @private bit 28 of EBX for EAX=0x7
+constexpr uint32_t cpuid_avx512bw_bit = 1 << 30;    ///< @private bit 30 of EBX for EAX=0x7
+constexpr uint32_t cpuid_avx512vl_bit = 1 << 31;    ///< @private bit 31 of EBX for EAX=0x7
+constexpr uint32_t cpuid_avx512vbmi2_bit = 1 << 6;  ///< @private bit 6 of ECX for EAX=0x7
+constexpr uint32_t cpuid_sse42_bit = 1 << 20;       ///< @private bit 20 of ECX for EAX=0x1
+constexpr uint32_t cpuid_pclmulqdq_bit = 1 << 1;    ///< @private bit  1 of ECX for EAX=0x1
 }


@@ -146,6 +164,42 @@ static inline uint32_t detect_supported_architectures() {
    host_isa |= instruction_set::BMI2;
  }

+  if (ebx & cpuid_avx512f_bit) {
+    host_isa |= instruction_set::AVX512F;
+  }
+
+  if (ebx & cpuid_avx512dq_bit) {
+    host_isa |= instruction_set::AVX512DQ;
+  }
+
+  if (ebx & cpuid_avx512ifma_bit) {
+    host_isa |= instruction_set::AVX512IFMA;
+  }
+
+  if (ebx & cpuid_avx512pf_bit) {
+    host_isa |= instruction_set::AVX512PF;
+  }
+
+  if (ebx & cpuid_avx512er_bit) {
+    host_isa |= instruction_set::AVX512ER;
+  }
+
+  if (ebx & cpuid_avx512cd_bit) {
+    host_isa |= instruction_set::AVX512CD;
+  }
+
+  if (ebx & cpuid_avx512bw_bit) {
+    host_isa |= instruction_set::AVX512BW;
+  }
+
+  if (ebx & cpuid_avx512vl_bit) {
+    host_isa |= instruction_set::AVX512VL;
+  }
+
+  if (ecx & cpuid_avx512vbmi2_bit) {
+    host_isa |= instruction_set::AVX512VBMI2;
+  }
+
  // EBX for EAX=0x1
  eax = 0x1;
  cpuid(&eax, &ebx, &ecx, &edx);
@@ -51,7 +51,7 @@ inline padded_string::padded_string(size_t length) noexcept
 }
 inline padded_string::padded_string(const char *data, size_t length) noexcept
    : viable_size(length), data_ptr(internal::allocate_padded_buffer(length)) {
-  if ((data != nullptr) and (data_ptr != nullptr)) {
+  if ((data != nullptr) && (data_ptr != nullptr)) {
    std::memcpy(data_ptr, data, length);
  }
 }
@@ -71,10 +71,10 @@ use a 64-bit target such as x64, 64-bit ARM or 64-bit PPC.")
 #endif // SIMDJSON_IS_32BITS

 // this is almost standard?
-#undef STRINGIFY_IMPLEMENTATION_
-#undef STRINGIFY
-#define STRINGIFY_IMPLEMENTATION_(a) #a
-#define STRINGIFY(a) STRINGIFY_IMPLEMENTATION_(a)
+#undef SIMDJSON_STRINGIFY_IMPLEMENTATION_
+#undef SIMDJSON_STRINGIFY
+#define SIMDJSON_STRINGIFY_IMPLEMENTATION_(a) #a
+#define SIMDJSON_STRINGIFY(a) SIMDJSON_STRINGIFY_IMPLEMENTATION_(a)

 // Our fast kernels require 64-bit systems.
 //
@@ -98,13 +98,13 @@ use a 64-bit target such as x64, 64-bit ARM or 64-bit PPC.")
 // til 8.0 so SIMDJSON_TARGET_REGION and SIMDJSON_UNTARGET_REGION must be *outside* of a
 // namespace.
 #define SIMDJSON_TARGET_REGION(T)                                                       \
-  _Pragma(STRINGIFY(                                                           \
+  _Pragma(SIMDJSON_STRINGIFY(                                                           \
      clang attribute push(__attribute__((target(T))), apply_to = function)))
 #define SIMDJSON_UNTARGET_REGION _Pragma("clang attribute pop")
 #elif defined(__GNUC__)
 // GCC is easier
 #define SIMDJSON_TARGET_REGION(T)                                                       \
-  _Pragma("GCC push_options") _Pragma(STRINGIFY(GCC target(T)))
+  _Pragma("GCC push_options") _Pragma(SIMDJSON_STRINGIFY(GCC target(T)))
 #define SIMDJSON_UNTARGET_REGION _Pragma("GCC pop_options")
 #endif // clang then gcc

@@ -138,11 +138,11 @@ use a 64-bit target such as x64, 64-bit ARM or 64-bit PPC.")


 #if defined(__clang__)
-#define NO_SANITIZE_UNDEFINED __attribute__((no_sanitize("undefined")))
+#define SIMDJSON_NO_SANITIZE_UNDEFINED __attribute__((no_sanitize("undefined")))
 #elif defined(__GNUC__)
-#define NO_SANITIZE_UNDEFINED __attribute__((no_sanitize_undefined))
+#define SIMDJSON_NO_SANITIZE_UNDEFINED __attribute__((no_sanitize_undefined))
 #else
-#define NO_SANITIZE_UNDEFINED
+#define SIMDJSON_NO_SANITIZE_UNDEFINED
 #endif

 #ifdef SIMDJSON_VISUAL_STUDIO
@@ -8,7 +8,7 @@ namespace {
 // We sometimes call trailing_zero on inputs that are zero,
 // but the algorithms do not end up using the returned value.
 // Sadly, sanitizers are not smart enough to figure it out.
-NO_SANITIZE_UNDEFINED
+SIMDJSON_NO_SANITIZE_UNDEFINED
 simdjson_really_inline int trailing_zeroes(uint64_t input_num) {
 #ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
  unsigned long ret;
@@ -1,7 +1,11 @@
 #ifndef SIMDJSON_PPC64_NUMBERPARSING_H
 #define SIMDJSON_PPC64_NUMBERPARSING_H

+#if defined(__linux__)
 #include <byteswap.h>
+#elif defined(__FreeBSD__)
+#include <sys/endian.h>
+#endif

 namespace simdjson {
 namespace SIMDJSON_IMPLEMENTATION {
@@ -14,7 +18,11 @@ parse_eight_digits_unrolled(const uint8_t *chars) {
  uint64_t val;
  std::memcpy(&val, chars, sizeof(uint64_t));
 #ifdef __BIG_ENDIAN__
+#if defined(__linux__)
  val = bswap_64(val);
+#elif defined(__FreeBSD__)
+  val = bswap64(val);
+#endif
 #endif
  val = (val & 0x0F0F0F0F0F0F0F0F) * 2561 >> 8;
  val = (val & 0x00FF00FF00FF00FF) * 6553601 >> 16;
@@ -69,8 +69,8 @@ struct base8 : base<simd8<T>> {
  simdjson_really_inline base8() : base<simd8<T>>() {}
  simdjson_really_inline base8(const __m128i _value) : base<simd8<T>>(_value) {}

-  simdjson_really_inline Mask operator==(const simd8<T> other) const {
-    return (__m128i)vec_cmpeq(this->value, (__m128i)other);
+  friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) {
+    return (__m128i)vec_cmpeq(lhs.value, (__m128i)rhs);
  }

  static const int SIZE = sizeof(base<simd8<T>>::value);
@@ -4,7 +4,7 @@
 #define SIMDJSON_SIMDJSON_VERSION_H

 /** The version of simdjson being used (major.minor.revision) */
-#define SIMDJSON_VERSION 1.0.0
+#define SIMDJSON_VERSION 1.0.2

 namespace simdjson {
 enum {
@@ -19,7 +19,7 @@ enum {
  /**
   * The revision (major.minor.REVISION) of simdjson being used.
   */
-  SIMDJSON_VERSION_REVISION = 0
+  SIMDJSON_VERSION_REVISION = 2
 };
 } // namespace simdjson

@@ -8,7 +8,7 @@ namespace {
 // We sometimes call trailing_zero on inputs that are zero,
 // but the algorithms do not end up using the returned value.
 // Sadly, sanitizers are not smart enough to figure it out.
-NO_SANITIZE_UNDEFINED
+SIMDJSON_NO_SANITIZE_UNDEFINED
 simdjson_really_inline int trailing_zeroes(uint64_t input_num) {
 #ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
  unsigned long ret;
@@ -44,7 +44,7 @@ namespace simd {
    simdjson_really_inline base8() : base<simd8<T>>() {}
    simdjson_really_inline base8(const __m128i _value) : base<simd8<T>>(_value) {}

-    simdjson_really_inline Mask operator==(const simd8<T> other) const { return _mm_cmpeq_epi8(*this, other); }
+    friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return _mm_cmpeq_epi8(lhs, rhs); }

    static const int SIZE = sizeof(base<simd8<T>>::value);

@@ -7,7 +7,7 @@ int main(int argc, char *argv[]) {
    std::cerr << "Please specify at least one file name and" << std::endl;
    std::cerr << "up to two files." << std::endl;
    std::cerr << "The first file should be a JSON document." << std::endl;
-    std::cerr << "The secod file should container many JSON documents."
+    std::cerr << "The second file should container many JSON documents."
              << std::endl;
    std::cerr << "Try the test files: jsonexamples/twitter.json "
                 "jsonexamples/amazon_cellphones.ndjson"
@@ -39,7 +39,7 @@ int main(int argc, char *argv[]) {
    std::cout << error << std::endl;
    return EXIT_FAILURE;
  }
-  std::cout << "document has the following type at the root: " << type
+  std::cout << "if valid, the document has the following type at the root: " << type
            << std::endl;

  if (argc == 2) {
@@ -73,8 +73,9 @@ int main(int argc, char *argv[]) {
    std::cout << error << std::endl;
    return EXIT_FAILURE;
  } else {
-    std::cout << "iterate_many valid" << std::endl;
-    std::cout << "found " << counter << " documents" << std::endl;
+    std::cout << "I found " << counter << " potential JSON documents." << std::endl;
  }
+  std::cout << "For more information on how simdjson works, please refer to our documentation." << std::endl;
+  std::cout << "https://github.com/simdjson/simdjson/blob/master/doc/basics.md" << std::endl;
  return EXIT_SUCCESS;
 }
@@ -1,4 +1,4 @@
-/* auto-generated on 2021-09-07 14:34:40 -0400. Do not edit! */
+/* auto-generated on 2022-01-31 11:38:54 -0500. Do not edit! */
 /* begin file src/simdjson.cpp */
 #include "simdjson.h"

@@ -2620,19 +2620,34 @@ namespace internal {
 // without requiring a static initializer.

 #if SIMDJSON_IMPLEMENTATION_HASWELL
-const haswell::implementation haswell_singleton{};
+static const haswell::implementation* get_haswell_singleton() {
+  static const haswell::implementation haswell_singleton{};
+  return &haswell_singleton;
+}
 #endif
 #if SIMDJSON_IMPLEMENTATION_WESTMERE
-const westmere::implementation westmere_singleton{};
+static const westmere::implementation* get_westmere_singleton() {
+  static const westmere::implementation westmere_singleton{};
+  return &westmere_singleton;
+}
 #endif // SIMDJSON_IMPLEMENTATION_WESTMERE
 #if SIMDJSON_IMPLEMENTATION_ARM64
-const arm64::implementation arm64_singleton{};
+static const arm64::implementation* get_arm64_singleton() {
+  static const arm64::implementation arm64_singleton{};
+  return &arm64_singleton;
+}
 #endif // SIMDJSON_IMPLEMENTATION_ARM64
 #if SIMDJSON_IMPLEMENTATION_PPC64
-const ppc64::implementation ppc64_singleton{};
+static const ppc64::implementation* get_ppc64_singleton() {
+  static const ppc64::implementation ppc64_singleton{};
+  return &ppc64_singleton;
+}
 #endif // SIMDJSON_IMPLEMENTATION_PPC64
 #if SIMDJSON_IMPLEMENTATION_FALLBACK
-const fallback::implementation fallback_singleton{};
+static const fallback::implementation* get_fallback_singleton() {
+  static const fallback::implementation fallback_singleton{};
+  return &fallback_singleton;
+}
 #endif // SIMDJSON_IMPLEMENTATION_FALLBACK

 /**
@@ -2661,25 +2676,26 @@ private:
  const implementation *set_best() const noexcept;
 };

-const detect_best_supported_implementation_on_first_use detect_best_supported_implementation_on_first_use_singleton;
-
-const std::initializer_list<const implementation *> available_implementation_pointers {
+static const std::initializer_list<const implementation *>& get_available_implementation_pointers() {
+  static const std::initializer_list<const implementation *> available_implementation_pointers {
 #if SIMDJSON_IMPLEMENTATION_HASWELL
-  &haswell_singleton,
+    get_haswell_singleton(),
 #endif
 #if SIMDJSON_IMPLEMENTATION_WESTMERE
-  &westmere_singleton,
+    get_westmere_singleton(),
 #endif
 #if SIMDJSON_IMPLEMENTATION_ARM64
-  &arm64_singleton,
+    get_arm64_singleton(),
 #endif
 #if SIMDJSON_IMPLEMENTATION_PPC64
-  &ppc64_singleton,
+    get_ppc64_singleton(),
 #endif
 #if SIMDJSON_IMPLEMENTATION_FALLBACK
-  &fallback_singleton,
+    get_fallback_singleton(),
 #endif
-}; // available_implementation_pointers
+  }; // available_implementation_pointers
+  return available_implementation_pointers;
+}

 // So we can return UNSUPPORTED_ARCHITECTURE from the parser when there is no support
 class unsupported_implementation final : public implementation {
@@ -2707,25 +2723,28 @@ public:
  unsupported_implementation() : implementation("unsupported", "Unsupported CPU (no detected SIMD instructions)", 0) {}
 };

-const unsupported_implementation unsupported_singleton{};
+const unsupported_implementation* get_unsupported_singleton() {
+    static const unsupported_implementation unsupported_singleton{};
+    return &unsupported_singleton;
+}

 size_t available_implementation_list::size() const noexcept {
-  return internal::available_implementation_pointers.size();
+  return internal::get_available_implementation_pointers().size();
 }
 const implementation * const *available_implementation_list::begin() const noexcept {
-  return internal::available_implementation_pointers.begin();
+  return internal::get_available_implementation_pointers().begin();
 }
 const implementation * const *available_implementation_list::end() const noexcept {
-  return internal::available_implementation_pointers.end();
+  return internal::get_available_implementation_pointers().end();
 }
 const implementation *available_implementation_list::detect_best_supported() const noexcept {
  // They are prelisted in priority order, so we just go down the list
  uint32_t supported_instruction_sets = internal::detect_supported_architectures();
-  for (const implementation *impl : internal::available_implementation_pointers) {
+  for (const implementation *impl : internal::get_available_implementation_pointers()) {
    uint32_t required_instruction_sets = impl->required_instruction_sets();
    if ((supported_instruction_sets & required_instruction_sets) == required_instruction_sets) { return impl; }
  }
-  return &unsupported_singleton; // this should never happen?
+  return get_unsupported_singleton(); // this should never happen?
 }

 const implementation *detect_best_supported_implementation_on_first_use::set_best() const noexcept {
@@ -2735,31 +2754,39 @@ const implementation *detect_best_supported_implementation_on_first_use::set_bes
  SIMDJSON_POP_DISABLE_WARNINGS

  if (force_implementation_name) {
-    auto force_implementation = available_implementations[force_implementation_name];
+    auto force_implementation = get_available_implementations()[force_implementation_name];
    if (force_implementation) {
-      return active_implementation = force_implementation;
+      return get_active_implementation() = force_implementation;
    } else {
      // Note: abort() and stderr usage within the library is forbidden.
-      return active_implementation = &unsupported_singleton;
+      return get_active_implementation() = get_unsupported_singleton();
    }
  }
-  return active_implementation = available_implementations.detect_best_supported();
+  return get_active_implementation() = get_available_implementations().detect_best_supported();
 }

 } // namespace internal

-SIMDJSON_DLLIMPORTEXPORT const internal::available_implementation_list available_implementations{};
-SIMDJSON_DLLIMPORTEXPORT internal::atomic_ptr<const implementation> active_implementation{&internal::detect_best_supported_implementation_on_first_use_singleton};
+SIMDJSON_DLLIMPORTEXPORT const internal::available_implementation_list& get_available_implementations() {
+  static const internal::available_implementation_list available_implementations{};
+  return available_implementations;
+}
+
+SIMDJSON_DLLIMPORTEXPORT internal::atomic_ptr<const implementation>& get_active_implementation() {
+    static const internal::detect_best_supported_implementation_on_first_use detect_best_supported_implementation_on_first_use_singleton;
+    static internal::atomic_ptr<const implementation> active_implementation{&detect_best_supported_implementation_on_first_use_singleton};
+    return active_implementation;
+}

 simdjson_warn_unused error_code minify(const char *buf, size_t len, char *dst, size_t &dst_len) noexcept {
-  return active_implementation->minify(reinterpret_cast<const uint8_t *>(buf), len, reinterpret_cast<uint8_t *>(dst), dst_len);
+  return get_active_implementation()->minify(reinterpret_cast<const uint8_t *>(buf), len, reinterpret_cast<uint8_t *>(dst), dst_len);
 }
 simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) noexcept {
-  return active_implementation->validate_utf8(buf, len);
+  return get_active_implementation()->validate_utf8(buf, len);
 }

 const implementation * builtin_implementation() {
-  static const implementation * builtin_impl = available_implementations[STRINGIFY(SIMDJSON_BUILTIN_IMPLEMENTATION)];
+  static const implementation * builtin_impl = get_available_implementations()[SIMDJSON_STRINGIFY(SIMDJSON_BUILTIN_IMPLEMENTATION)];
  assert(builtin_impl);
  return builtin_impl;
 }
@@ -3473,11 +3500,11 @@ simdjson_really_inline uint64_t follows(const uint64_t match, uint64_t &overflow

 simdjson_really_inline json_block json_scanner::next(const simd::simd8x64<uint8_t>& in) {
  json_string_block strings = string_scanner.next(in);
-  // identifies the white-space and the structurat characters
+  // identifies the white-space and the structural characters
  json_character_block characters = json_character_block::classify(in);
  // The term "scalar" refers to anything except structural characters and white space
  // (so letters, numbers, quotes).
-  // We want  follows_scalar to mark anything that follows a non-quote scalar (so letters and numbers).
+  // We want follows_scalar to mark anything that follows a non-quote scalar (so letters and numbers).
  //
  // A terminal quote should either be followed by a structural character (comma, brace, bracket, colon)
  // or nothing. However, we still want ' "a string"true ' to mark the 't' of 'true' as a potential
@@ -4246,25 +4273,25 @@ public:
  /**
   * Log that a value has been found.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_value(const char *type) const noexcept;
  /**
   * Log the start of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_start_value(const char *type) const noexcept;
  /**
   * Log the end of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_end_value(const char *type) const noexcept;
  /**
   * Log an error.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_error(const char *error) const noexcept;

@@ -5574,25 +5601,25 @@ public:
  /**
   * Log that a value has been found.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_value(const char *type) const noexcept;
  /**
   * Log the start of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_start_value(const char *type) const noexcept;
  /**
   * Log the end of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_end_value(const char *type) const noexcept;
  /**
   * Log an error.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_error(const char *error) const noexcept;

@@ -6937,11 +6964,11 @@ simdjson_really_inline uint64_t follows(const uint64_t match, uint64_t &overflow

 simdjson_really_inline json_block json_scanner::next(const simd::simd8x64<uint8_t>& in) {
  json_string_block strings = string_scanner.next(in);
-  // identifies the white-space and the structurat characters
+  // identifies the white-space and the structural characters
  json_character_block characters = json_character_block::classify(in);
  // The term "scalar" refers to anything except structural characters and white space
  // (so letters, numbers, quotes).
-  // We want  follows_scalar to mark anything that follows a non-quote scalar (so letters and numbers).
+  // We want follows_scalar to mark anything that follows a non-quote scalar (so letters and numbers).
  //
  // A terminal quote should either be followed by a structural character (comma, brace, bracket, colon)
  // or nothing. However, we still want ' "a string"true ' to mark the 't' of 'true' as a potential
@@ -7709,25 +7736,25 @@ public:
  /**
   * Log that a value has been found.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_value(const char *type) const noexcept;
  /**
   * Log the start of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_start_value(const char *type) const noexcept;
  /**
   * Log the end of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_end_value(const char *type) const noexcept;
  /**
   * Log an error.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_error(const char *error) const noexcept;

@@ -9063,11 +9090,11 @@ simdjson_really_inline uint64_t follows(const uint64_t match, uint64_t &overflow

 simdjson_really_inline json_block json_scanner::next(const simd::simd8x64<uint8_t>& in) {
  json_string_block strings = string_scanner.next(in);
-  // identifies the white-space and the structurat characters
+  // identifies the white-space and the structural characters
  json_character_block characters = json_character_block::classify(in);
  // The term "scalar" refers to anything except structural characters and white space
  // (so letters, numbers, quotes).
-  // We want  follows_scalar to mark anything that follows a non-quote scalar (so letters and numbers).
+  // We want follows_scalar to mark anything that follows a non-quote scalar (so letters and numbers).
  //
  // A terminal quote should either be followed by a structural character (comma, brace, bracket, colon)
  // or nothing. However, we still want ' "a string"true ' to mark the 't' of 'true' as a potential
@@ -9836,25 +9863,25 @@ public:
  /**
   * Log that a value has been found.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_value(const char *type) const noexcept;
  /**
   * Log the start of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_start_value(const char *type) const noexcept;
  /**
   * Log the end of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_end_value(const char *type) const noexcept;
  /**
   * Log an error.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_error(const char *error) const noexcept;

@@ -11225,11 +11252,11 @@ simdjson_really_inline uint64_t follows(const uint64_t match, uint64_t &overflow

 simdjson_really_inline json_block json_scanner::next(const simd::simd8x64<uint8_t>& in) {
  json_string_block strings = string_scanner.next(in);
-  // identifies the white-space and the structurat characters
+  // identifies the white-space and the structural characters
  json_character_block characters = json_character_block::classify(in);
  // The term "scalar" refers to anything except structural characters and white space
  // (so letters, numbers, quotes).
-  // We want  follows_scalar to mark anything that follows a non-quote scalar (so letters and numbers).
+  // We want follows_scalar to mark anything that follows a non-quote scalar (so letters and numbers).
  //
  // A terminal quote should either be followed by a structural character (comma, brace, bracket, colon)
  // or nothing. However, we still want ' "a string"true ' to mark the 't' of 'true' as a potential
@@ -11997,25 +12024,25 @@ public:
  /**
   * Log that a value has been found.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_value(const char *type) const noexcept;
  /**
   * Log the start of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_start_value(const char *type) const noexcept;
  /**
   * Log the end of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_end_value(const char *type) const noexcept;
  /**
   * Log an error.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_error(const char *error) const noexcept;

@@ -1,4 +1,4 @@
-/* auto-generated on 2021-09-07 14:34:40 -0400. Do not edit! */
+/* auto-generated on 2022-01-31 11:38:54 -0500. Do not edit! */
 /* begin file include/simdjson.h */
 #ifndef SIMDJSON_H
 #define SIMDJSON_H
@@ -36,6 +36,34 @@
    }
 */

+/* begin file include/simdjson/simdjson_version.h */
+// /include/simdjson/simdjson_version.h automatically generated by release.py,
+// do not change by hand
+#ifndef SIMDJSON_SIMDJSON_VERSION_H
+#define SIMDJSON_SIMDJSON_VERSION_H
+
+/** The version of simdjson being used (major.minor.revision) */
+#define SIMDJSON_VERSION 1.0.2
+
+namespace simdjson {
+enum {
+  /**
+   * The major version (MAJOR.minor.revision) of simdjson being used.
+   */
+  SIMDJSON_VERSION_MAJOR = 1,
+  /**
+   * The minor version (major.MINOR.revision) of simdjson being used.
+   */
+  SIMDJSON_VERSION_MINOR = 0,
+  /**
+   * The revision (major.minor.REVISION) of simdjson being used.
+   */
+  SIMDJSON_VERSION_REVISION = 2
+};
+} // namespace simdjson
+
+#endif // SIMDJSON_SIMDJSON_VERSION_H
+/* end file include/simdjson/simdjson_version.h */
 /* begin file include/simdjson/dom.h */
 #ifndef SIMDJSON_DOM_H
 #define SIMDJSON_DOM_H
@@ -160,10 +188,10 @@ use a 64-bit target such as x64, 64-bit ARM or 64-bit PPC.")
 #endif // SIMDJSON_IS_32BITS

 // this is almost standard?
-#undef STRINGIFY_IMPLEMENTATION_
-#undef STRINGIFY
-#define STRINGIFY_IMPLEMENTATION_(a) #a
-#define STRINGIFY(a) STRINGIFY_IMPLEMENTATION_(a)
+#undef SIMDJSON_STRINGIFY_IMPLEMENTATION_
+#undef SIMDJSON_STRINGIFY
+#define SIMDJSON_STRINGIFY_IMPLEMENTATION_(a) #a
+#define SIMDJSON_STRINGIFY(a) SIMDJSON_STRINGIFY_IMPLEMENTATION_(a)

 // Our fast kernels require 64-bit systems.
 //
@@ -187,13 +215,13 @@ use a 64-bit target such as x64, 64-bit ARM or 64-bit PPC.")
 // til 8.0 so SIMDJSON_TARGET_REGION and SIMDJSON_UNTARGET_REGION must be *outside* of a
 // namespace.
 #define SIMDJSON_TARGET_REGION(T)                                                       \
-  _Pragma(STRINGIFY(                                                           \
+  _Pragma(SIMDJSON_STRINGIFY(                                                           \
      clang attribute push(__attribute__((target(T))), apply_to = function)))
 #define SIMDJSON_UNTARGET_REGION _Pragma("clang attribute pop")
 #elif defined(__GNUC__)
 // GCC is easier
 #define SIMDJSON_TARGET_REGION(T)                                                       \
-  _Pragma("GCC push_options") _Pragma(STRINGIFY(GCC target(T)))
+  _Pragma("GCC push_options") _Pragma(SIMDJSON_STRINGIFY(GCC target(T)))
 #define SIMDJSON_UNTARGET_REGION _Pragma("GCC pop_options")
 #endif // clang then gcc

@@ -227,11 +255,11 @@ use a 64-bit target such as x64, 64-bit ARM or 64-bit PPC.")


 #if defined(__clang__)
-#define NO_SANITIZE_UNDEFINED __attribute__((no_sanitize("undefined")))
+#define SIMDJSON_NO_SANITIZE_UNDEFINED __attribute__((no_sanitize("undefined")))
 #elif defined(__GNUC__)
-#define NO_SANITIZE_UNDEFINED __attribute__((no_sanitize_undefined))
+#define SIMDJSON_NO_SANITIZE_UNDEFINED __attribute__((no_sanitize_undefined))
 #else
-#define NO_SANITIZE_UNDEFINED
+#define SIMDJSON_NO_SANITIZE_UNDEFINED
 #endif

 #ifdef SIMDJSON_VISUAL_STUDIO
@@ -2182,34 +2210,6 @@ SIMDJSON_PUSH_DISABLE_WARNINGS
 SIMDJSON_DISABLE_UNDESIRED_WARNINGS

 // Public API
-/* begin file include/simdjson/simdjson_version.h */
-// /include/simdjson/simdjson_version.h automatically generated by release.py,
-// do not change by hand
-#ifndef SIMDJSON_SIMDJSON_VERSION_H
-#define SIMDJSON_SIMDJSON_VERSION_H
-
-/** The version of simdjson being used (major.minor.revision) */
-#define SIMDJSON_VERSION 1.0.0
-
-namespace simdjson {
-enum {
-  /**
-   * The major version (MAJOR.minor.revision) of simdjson being used.
-   */
-  SIMDJSON_VERSION_MAJOR = 1,
-  /**
-   * The minor version (major.MINOR.revision) of simdjson being used.
-   */
-  SIMDJSON_VERSION_MINOR = 0,
-  /**
-   * The revision (major.minor.REVISION) of simdjson being used.
-   */
-  SIMDJSON_VERSION_REVISION = 0
-};
-} // namespace simdjson
-
-#endif // SIMDJSON_SIMDJSON_VERSION_H
-/* end file include/simdjson/simdjson_version.h */
 /* begin file include/simdjson/error.h */
 #ifndef SIMDJSON_ERROR_H
 #define SIMDJSON_ERROR_H
@@ -2219,7 +2219,12 @@ enum {
 namespace simdjson {

 /**
- * All possible errors returned by simdjson.
+ * All possible errors returned by simdjson. These error codes are subject to change
+ * and not all simdjson kernel returns the same error code given the same input: it is not
+ * well defined which error a given input should produce.
+ *
+ * Only SUCCESS evaluates to false as a Boolean. All other error codes will evaluate
+ * to true as a Boolean.
 */
 enum error_code {
  SUCCESS = 0,                ///< No error
@@ -3247,7 +3252,7 @@ public:
  /**
   * The name of this implementation.
   *
-   *     const implementation *impl = simdjson::active_implementation;
+   *     const implementation *impl = simdjson::get_active_implementation();
   *     cout << "simdjson is optimized for " << impl->name() << "(" << impl->description() << ")" << endl;
   *
   * @return the name of the implementation, e.g. "haswell", "westmere", "arm64"
@@ -3257,7 +3262,7 @@ public:
  /**
   * The description of this implementation.
   *
-   *     const implementation *impl = simdjson::active_implementation;
+   *     const implementation *impl = simdjson::get_active_implementation();
   *     cout << "simdjson is optimized for " << impl->name() << "(" << impl->description() << ")" << endl;
   *
   * @return the name of the implementation, e.g. "haswell", "westmere", "arm64"
@@ -3286,7 +3291,7 @@ public:
  /**
   * @private For internal implementation use
   *
-   *     const implementation *impl = simdjson::active_implementation;
+   *     const implementation *impl = simdjson::get_active_implementation();
   *     cout << "simdjson is optimized for " << impl->name() << "(" << impl->description() << ")" << endl;
   *
   * @param capacity The largest document that will be passed to the parser.
@@ -3380,10 +3385,10 @@ public:
   *
   * Case sensitive.
   *
-   *     const implementation *impl = simdjson::available_implementations["westmere"];
+   *     const implementation *impl = simdjson::get_available_implementations()["westmere"];
   *     if (!impl) { exit(1); }
   *     if (!imp->supported_by_runtime_system()) { exit(1); }
-   *     simdjson::active_implementation = impl;
+   *     simdjson::get_active_implementation() = impl;
   *
   * @param name the implementation to find, e.g. "westmere", "haswell", "arm64"
   * @return the implementation, or nullptr if the parse failed.
@@ -3401,7 +3406,7 @@ public:
   * This is used to initialize the implementation on startup.
   *
   *     const implementation *impl = simdjson::available_implementation::detect_best_supported();
-   *     simdjson::active_implementation = impl;
+   *     simdjson::get_active_implementation() = impl;
   *
   * @return the most advanced supported implementation for the current host, or an
   *         implementation that returns UNSUPPORTED_ARCHITECTURE if there is no supported
@@ -3433,14 +3438,14 @@ private:
 /**
 * The list of available implementations compiled into simdjson.
 */
-extern SIMDJSON_DLLIMPORTEXPORT const internal::available_implementation_list available_implementations;
+extern SIMDJSON_DLLIMPORTEXPORT const internal::available_implementation_list& get_available_implementations();

 /**
  * The active implementation.
  *
  * Automatically initialized on first use to the most advanced implementation supported by this hardware.
  */
-extern SIMDJSON_DLLIMPORTEXPORT internal::atomic_ptr<const implementation> active_implementation;
+extern SIMDJSON_DLLIMPORTEXPORT internal::atomic_ptr<const implementation>& get_active_implementation();

 } // namespace simdjson

@@ -3686,7 +3691,7 @@ inline padded_string::padded_string(size_t length) noexcept
 }
 inline padded_string::padded_string(const char *data, size_t length) noexcept
    : viable_size(length), data_ptr(internal::allocate_padded_buffer(length)) {
-  if ((data != nullptr) and (data_ptr != nullptr)) {
+  if ((data != nullptr) && (data_ptr != nullptr)) {
    std::memcpy(data_ptr, data, length);
  }
 }
@@ -8672,7 +8677,7 @@ inline error_code parser::allocate(size_t capacity, size_t max_depth) noexcept {
  if (implementation) {
    err = implementation->allocate(capacity, max_depth);
  } else {
-    err = simdjson::active_implementation->create_dom_parser_implementation(capacity, max_depth, implementation);
+    err = simdjson::get_active_implementation()->create_dom_parser_implementation(capacity, max_depth, implementation);
  }
  if (err) { return err; }
  return SUCCESS;
@@ -8995,7 +9000,10 @@ simdjson_really_inline void mini_formatter::string(std::string_view unescaped) {
  size_t i = 0;
  // Fast path for the case where we have no control character, no ", and no backslash.
  // This should include most keys.
-  constexpr static bool needs_escaping[] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+  //
+  // We would like to use 'bool' but some compilers take offense to bitwise operation
+  // with bool types.
+  constexpr static char needs_escaping[] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0,
@@ -9007,6 +9015,8 @@ simdjson_really_inline void mini_formatter::string(std::string_view unescaped) {
    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
  for(;i + 8 <= unescaped.length(); i += 8) {
    // Poor's man vectorization. This could get much faster if we used SIMD.
+    //
+    // It is not the case that replacing '|' with '||' would be neutral performance-wise.
    if(needs_escaping[uint8_t(unescaped[i])] | needs_escaping[uint8_t(unescaped[i+1])]
      | needs_escaping[uint8_t(unescaped[i+2])] | needs_escaping[uint8_t(unescaped[i+3])]
      | needs_escaping[uint8_t(unescaped[i+4])] | needs_escaping[uint8_t(unescaped[i+5])]
@@ -9415,9 +9425,13 @@ extern SIMDJSON_DLLIMPORTEXPORT const uint64_t thintable_epi8[256];
 #ifndef SIMDJSON_IMPLEMENTATION_HASWELL
 #define SIMDJSON_IMPLEMENTATION_HASWELL (SIMDJSON_IS_X86_64)
 #endif
+#ifdef _MSC_VER
 // To see why  (__BMI__) && (__PCLMUL__) && (__LZCNT__) are not part of this next line, see
 // https://github.com/simdjson/simdjson/issues/1247
 #define SIMDJSON_CAN_ALWAYS_RUN_HASWELL ((SIMDJSON_IMPLEMENTATION_HASWELL) && (SIMDJSON_IS_X86_64) && (__AVX2__))
+#else
+#define SIMDJSON_CAN_ALWAYS_RUN_HASWELL ((SIMDJSON_IMPLEMENTATION_HASWELL) && (SIMDJSON_IS_X86_64) && (__AVX2__) && (__BMI__) && (__PCLMUL__) && (__LZCNT__))
+#endif

 // Default Westmere to on if this is x86-64, unless we'll always select Haswell.
 #ifndef SIMDJSON_IMPLEMENTATION_WESTMERE
@@ -9593,7 +9607,7 @@ namespace {
 // We sometimes call trailing_zero on inputs that are zero,
 // but the algorithms do not end up using the returned value.
 // Sadly, sanitizers are not smart enough to figure it out.
-NO_SANITIZE_UNDEFINED
+SIMDJSON_NO_SANITIZE_UNDEFINED
 simdjson_really_inline int trailing_zeroes(uint64_t input_num) {
 #ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
  unsigned long ret;
@@ -9658,7 +9672,7 @@ simdjson_really_inline uint64_t reverse_bits(uint64_t input_num) {
 * greating or equal to 63 in which case we trigger undefined behavior, but the output
 * of such undefined behavior is never used.
 **/
-NO_SANITIZE_UNDEFINED
+SIMDJSON_NO_SANITIZE_UNDEFINED
 simdjson_really_inline uint64_t zero_leading_bit(uint64_t rev_bits, int leading_zeroes) {
  return rev_bits ^ (uint64_t(0x8000000000000000) >> leading_zeroes);
 }
@@ -9855,7 +9869,7 @@ simdjson_really_inline int8x16_t make_int8x16_t(int8_t x1,  int8_t x2,  int8_t x
    simdjson_really_inline simd8<T>& operator&=(const simd8<T> other) { auto this_cast = static_cast<simd8<T>*>(this); *this_cast = *this_cast & other; return *this_cast; }
    simdjson_really_inline simd8<T>& operator^=(const simd8<T> other) { auto this_cast = static_cast<simd8<T>*>(this); *this_cast = *this_cast ^ other; return *this_cast; }

-    simdjson_really_inline Mask operator==(const simd8<T> other) const { return vceqq_u8(*this, other); }
+    friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return vceqq_u8(lhs, rhs); }

    template<int N=1>
    simdjson_really_inline simd8<T> prev(const simd8<T> prev_chunk) const {
@@ -11016,7 +11030,7 @@ error_code slow_float_parsing(simdjson_unused const uint8_t * src, W writer) {
 }

 template<typename I>
-NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
+SIMDJSON_NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
 simdjson_really_inline bool parse_digit(const uint8_t c, I &i) {
  const uint8_t digit = static_cast<uint8_t>(c - '0');
  if (digit > 9) {
@@ -11264,7 +11278,7 @@ simdjson_really_inline error_code parse_number(const uint8_t *const src, W &writ
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    }  else if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INVALID_NUMBER(src); }
  }
@@ -11387,7 +11401,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -11437,7 +11451,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -11485,9 +11499,11 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned_
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
-    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
+    // Note: we use src[1] and not src[0] because src[0] is the quote character in this
+    // instance.
+    if (src[1] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }

  return i;
@@ -12847,7 +12863,7 @@ error_code slow_float_parsing(simdjson_unused const uint8_t * src, W writer) {
 }

 template<typename I>
-NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
+SIMDJSON_NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
 simdjson_really_inline bool parse_digit(const uint8_t c, I &i) {
  const uint8_t digit = static_cast<uint8_t>(c - '0');
  if (digit > 9) {
@@ -13095,7 +13111,7 @@ simdjson_really_inline error_code parse_number(const uint8_t *const src, W &writ
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    }  else if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INVALID_NUMBER(src); }
  }
@@ -13218,7 +13234,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -13268,7 +13284,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -13316,9 +13332,11 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned_
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
-    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
+    // Note: we use src[1] and not src[0] because src[0] is the quote character in this
+    // instance.
+    if (src[1] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }

  return i;
@@ -13952,7 +13970,7 @@ namespace {
 // We sometimes call trailing_zero on inputs that are zero,
 // but the algorithms do not end up using the returned value.
 // Sadly, sanitizers are not smart enough to figure it out.
-NO_SANITIZE_UNDEFINED
+SIMDJSON_NO_SANITIZE_UNDEFINED
 simdjson_really_inline int trailing_zeroes(uint64_t input_num) {
 #ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
  return (int)_tzcnt_u64(input_num);
@@ -14078,7 +14096,7 @@ namespace simd {
    simdjson_really_inline base8() : base<simd8<T>>() {}
    simdjson_really_inline base8(const __m256i _value) : base<simd8<T>>(_value) {}

-    simdjson_really_inline Mask operator==(const simd8<T> other) const { return _mm256_cmpeq_epi8(*this, other); }
+    friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return _mm256_cmpeq_epi8(lhs, rhs); }

    static const int SIZE = sizeof(base<T>::value);

@@ -15163,7 +15181,7 @@ error_code slow_float_parsing(simdjson_unused const uint8_t * src, W writer) {
 }

 template<typename I>
-NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
+SIMDJSON_NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
 simdjson_really_inline bool parse_digit(const uint8_t c, I &i) {
  const uint8_t digit = static_cast<uint8_t>(c - '0');
  if (digit > 9) {
@@ -15411,7 +15429,7 @@ simdjson_really_inline error_code parse_number(const uint8_t *const src, W &writ
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    }  else if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INVALID_NUMBER(src); }
  }
@@ -15534,7 +15552,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -15584,7 +15602,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -15632,9 +15650,11 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned_
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
-    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
+    // Note: we use src[1] and not src[0] because src[0] is the quote character in this
+    // instance.
+    if (src[1] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }

  return i;
@@ -16223,7 +16243,7 @@ namespace {
 // We sometimes call trailing_zero on inputs that are zero,
 // but the algorithms do not end up using the returned value.
 // Sadly, sanitizers are not smart enough to figure it out.
-NO_SANITIZE_UNDEFINED
+SIMDJSON_NO_SANITIZE_UNDEFINED
 simdjson_really_inline int trailing_zeroes(uint64_t input_num) {
 #ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
  unsigned long ret;
@@ -16397,8 +16417,8 @@ struct base8 : base<simd8<T>> {
  simdjson_really_inline base8() : base<simd8<T>>() {}
  simdjson_really_inline base8(const __m128i _value) : base<simd8<T>>(_value) {}

-  simdjson_really_inline Mask operator==(const simd8<T> other) const {
-    return (__m128i)vec_cmpeq(this->value, (__m128i)other);
+  friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) {
+    return (__m128i)vec_cmpeq(lhs.value, (__m128i)rhs);
  }

  static const int SIZE = sizeof(base<simd8<T>>::value);
@@ -17189,7 +17209,11 @@ simdjson_unused simdjson_warn_unused simdjson_really_inline error_code parse_str
 #ifndef SIMDJSON_PPC64_NUMBERPARSING_H
 #define SIMDJSON_PPC64_NUMBERPARSING_H

+#if defined(__linux__)
 #include <byteswap.h>
+#elif defined(__FreeBSD__)
+#include <sys/endian.h>
+#endif

 namespace simdjson {
 namespace ppc64 {
@@ -17202,7 +17226,11 @@ parse_eight_digits_unrolled(const uint8_t *chars) {
  uint64_t val;
  std::memcpy(&val, chars, sizeof(uint64_t));
 #ifdef __BIG_ENDIAN__
+#if defined(__linux__)
  val = bswap_64(val);
+#elif defined(__FreeBSD__)
+  val = bswap64(val);
+#endif
 #endif
  val = (val & 0x0F0F0F0F0F0F0F0F) * 2561 >> 8;
  val = (val & 0x00FF00FF00FF00FF) * 6553601 >> 16;
@@ -17578,7 +17606,7 @@ error_code slow_float_parsing(simdjson_unused const uint8_t * src, W writer) {
 }

 template<typename I>
-NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
+SIMDJSON_NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
 simdjson_really_inline bool parse_digit(const uint8_t c, I &i) {
  const uint8_t digit = static_cast<uint8_t>(c - '0');
  if (digit > 9) {
@@ -17826,7 +17854,7 @@ simdjson_really_inline error_code parse_number(const uint8_t *const src, W &writ
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    }  else if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INVALID_NUMBER(src); }
  }
@@ -17949,7 +17977,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -17999,7 +18027,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -18047,9 +18075,11 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned_
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
-    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
+    // Note: we use src[1] and not src[0] because src[0] is the quote character in this
+    // instance.
+    if (src[1] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }

  return i;
@@ -18659,7 +18689,7 @@ namespace {
 // We sometimes call trailing_zero on inputs that are zero,
 // but the algorithms do not end up using the returned value.
 // Sadly, sanitizers are not smart enough to figure it out.
-NO_SANITIZE_UNDEFINED
+SIMDJSON_NO_SANITIZE_UNDEFINED
 simdjson_really_inline int trailing_zeroes(uint64_t input_num) {
 #ifdef SIMDJSON_REGULAR_VISUAL_STUDIO
  unsigned long ret;
@@ -18793,7 +18823,7 @@ namespace simd {
    simdjson_really_inline base8() : base<simd8<T>>() {}
    simdjson_really_inline base8(const __m128i _value) : base<simd8<T>>(_value) {}

-    simdjson_really_inline Mask operator==(const simd8<T> other) const { return _mm_cmpeq_epi8(*this, other); }
+    friend simdjson_really_inline Mask operator==(const simd8<T> lhs, const simd8<T> rhs) { return _mm_cmpeq_epi8(lhs, rhs); }

    static const int SIZE = sizeof(base<simd8<T>>::value);

@@ -19851,7 +19881,7 @@ error_code slow_float_parsing(simdjson_unused const uint8_t * src, W writer) {
 }

 template<typename I>
-NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
+SIMDJSON_NO_SANITIZE_UNDEFINED // We deliberately allow overflow here and check later
 simdjson_really_inline bool parse_digit(const uint8_t c, I &i) {
  const uint8_t digit = static_cast<uint8_t>(c - '0');
  if (digit > 9) {
@@ -20099,7 +20129,7 @@ simdjson_really_inline error_code parse_number(const uint8_t *const src, W &writ
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    }  else if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INVALID_NUMBER(src); }
  }
@@ -20222,7 +20252,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -20272,7 +20302,7 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned(
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }
@@ -20320,9 +20350,11 @@ simdjson_unused simdjson_really_inline simdjson_result<uint64_t> parse_unsigned_
    // - That is smaller than the smallest possible 20-digit number the user could write:
    //   10,000,000,000,000,000,000.
    // - Therefore, if the number is positive and lower than that, it's overflow.
-    // - The value we are looking at is less than or equal to 9,223,372,036,854,775,808 (INT64_MAX).
+    // - The value we are looking at is less than or equal to INT64_MAX.
    //
-    if (src[0] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
+    // Note: we use src[1] and not src[0] because src[0] is the quote character in this
+    // instance.
+    if (src[1] != uint8_t('1') || i <= uint64_t(INT64_MAX)) { return INCORRECT_TYPE; }
  }

  return i;
@@ -21033,7 +21065,6 @@ protected:
    uint64_t unsigned_integer;
  } payload{0};
  number_type type{number_type::signed_integer};
-  friend class value_iterator;
 };

 /**
@@ -22497,7 +22528,7 @@ public:

  /**
   * Get the value at the given index. This function has linear-time complexity.
-   * This function should only be called once as the array iterator is not reset between each call.
+   * This function should only be called once on an array instance since the array iterator is not reset between each call.
   *
   * @return The value at the given index, or:
   *         - INDEX_OUT_OF_BOUNDS if the array index is larger than an array length
@@ -22679,6 +22710,8 @@ public:
   *
   * The string is guaranteed to be valid UTF-8.
   *
+   * Important: Calling get_string() twice on the same document is an error.
+   *
   * @returns An UTF-8 string. The string is stored in the parser and will be invalidated the next
   *          time it parses a document or when it is destroyed.
   * @returns INCORRECT_TYPE if the JSON value is not a string.
@@ -22852,7 +22885,7 @@ public:
  simdjson_really_inline simdjson_result<size_t> count_fields() & noexcept;
  /**
   * Get the value at the given index in the array. This function has linear-time complexity.
-   * This function should only be called once as the array iterator is not reset between each call.
+   * This function should only be called once on an array instance since the array iterator is not reset between each call.
   *
   * @return The value at the given index, or:
   *         - INDEX_OUT_OF_BOUNDS if the array index is larger than an array length
@@ -22888,6 +22921,18 @@ public:
   * **Raw Keys:** The lookup will be done against the *raw* key, and will not unescape keys.
   * e.g. `object["a"]` will match `{ "a": 1 }`, but will *not* match `{ "\u0061": 1 }`.
   *
+   *
+   * You must consume the fields on an object one at a time. A request for a new key
+   * invalidates previous field values: it makes them unsafe. E.g., the array
+   * given by content["bids"].get_array() should not be accessed after you have called
+   * content["asks"].get_array(). You can detect such mistakes by first compiling and running
+   * the code in Debug mode (or with the macro `SIMDJSON_DEVELOPMENT_CHECKS` set to 1): an
+   * OUT_OF_ORDER_ITERATION error is generated.
+   *
+   * You are expected to access keys only once. You should access the value corresponding to
+   * a key a single time. Doing object["mykey"].to_string()and then again object["mykey"].to_string()
+   * is an error.
+   *
   * @param key The key to look up.
   * @returns The value of the field, or NO_SUCH_FIELD if the field is not in the object.
   */
@@ -22911,6 +22956,17 @@ public:
   * Use find_field() if you are sure fields will be in order (or are willing to treat it as if the
   * field wasn't there when they aren't).
   *
+   * You must consume the fields on an object one at a time. A request for a new key
+   * invalidates previous field values: it makes them unsafe. E.g., the array
+   * given by content["bids"].get_array() should not be accessed after you have called
+   * content["asks"].get_array(). You can detect such mistakes by first compiling and running
+   * the code in Debug mode (or with the macro `SIMDJSON_DEVELOPMENT_CHECKS` set to 1): an
+   * OUT_OF_ORDER_ITERATION error is generated.
+   *
+   * You are expected to access keys only once. You should access the value corresponding to a key
+   * a single time. Doing object["mykey"].to_string() and then again object["mykey"].to_string()
+   * is an error.
+   *
   * @param key The key to look up.
   * @returns The value of the field, or NO_SUCH_FIELD if the field is not in the object.
   */
@@ -22984,7 +23040,7 @@ public:
   * type.
   *
   * number.get_number_type() is number_type::signed_integer if we have
-   * a integer in [-9223372036854775808,9223372036854775808)
+   * an integer in [-9223372036854775808,9223372036854775808)
   * You can recover the value by calling number.get_int64() and you
   * have that number.is_int64() is true.
   *
@@ -23083,7 +23139,7 @@ public:
  simdjson_really_inline simdjson_result<value> at_pointer(std::string_view json_pointer) noexcept;
  /**
   * Consumes the document and returns a string_view instance corresponding to the
-   * document as represented in JSON. It points inside the original byte array containg
+   * document as represented in JSON. It points inside the original byte array containing
   * the JSON document.
   */
  simdjson_really_inline simdjson_result<std::string_view> raw_json() noexcept;
@@ -23126,6 +23182,7 @@ public:
  simdjson_really_inline document_reference() noexcept;
  simdjson_really_inline document_reference(document &d) noexcept;
  simdjson_really_inline document_reference(const document_reference &other) noexcept = default;
+  simdjson_really_inline document_reference& operator=(const document_reference &other) noexcept = default;
  simdjson_really_inline void rewind() noexcept;
  simdjson_really_inline simdjson_result<array> get_array() & noexcept;
  simdjson_really_inline simdjson_result<object> get_object() & noexcept;
@@ -23430,6 +23487,9 @@ public:
   *
   * Equivalent to get<std::string_view>().
   *
+   * Important: a value should be consumed once. Calling get_string() twice on the same value
+   * is an error.
+   *
   * @returns An UTF-8 string. The string is stored in the parser and will be invalidated the next
   *          time it parses a document or when it is destroyed.
   * @returns INCORRECT_TYPE if the JSON value is not a string.
@@ -23553,9 +23613,24 @@ public:
   * safe to continue.
   */
  simdjson_really_inline simdjson_result<size_t> count_elements() & noexcept;
+  /**
+   * This method scans the object and counts the number of key-value pairs.
+   * The count_fields method should always be called before you have begun
+   * iterating through the object: it is expected that you are pointing at
+   * the beginning of the object.
+   * The runtime complexity is linear in the size of the object. After
+   * calling this function, if successful, the object is 'rewinded' at its
+   * beginning as if it had never been accessed. If the JSON is malformed (e.g.,
+   * there is a missing comma), then an error is returned and it is no longer
+   * safe to continue.
+   *
+   * To check that an object is empty, it is more performant to use
+   * the is_empty() method on the object instance.
+   */
+  simdjson_really_inline simdjson_result<size_t> count_fields() & noexcept;
  /**
   * Get the value at the given index in the array. This function has linear-time complexity.
-   * This function should only be called once as the array iterator is not reset between each call.
+   * This function should only be called once on an array instance since the array iterator is not reset between each call.
   *
   * @return The value at the given index, or:
   *         - INDEX_OUT_OF_BOUNDS if the array index is larger than an array length
@@ -23684,7 +23759,7 @@ public:
   * type.
   *
   * number.get_number_type() is number_type::signed_integer if we have
-   * a integer in [-9223372036854775808,9223372036854775808)
+   * an integer in [-9223372036854775808,9223372036854775808)
   * You can recover the value by calling number.get_int64() and you
   * have that number.is_int64() is true.
   *
@@ -23734,6 +23809,11 @@ public:
   */
  simdjson_really_inline std::string_view raw_json_token() noexcept;

+  /**
+   * Returns the current location in the document if in bounds.
+   */
+  simdjson_really_inline simdjson_result<const char *> current_location() noexcept;
+
  /**
   * Get the value associated with the given JSON pointer.  We use the RFC 6901
   * https://tools.ietf.org/html/rfc6901 standard.
@@ -23861,6 +23941,7 @@ public:
  simdjson_really_inline operator bool() noexcept(false);
 #endif
  simdjson_really_inline simdjson_result<size_t> count_elements() & noexcept;
+  simdjson_really_inline simdjson_result<size_t> count_fields() & noexcept;
  simdjson_really_inline simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::value> at(size_t index) noexcept;
  simdjson_really_inline simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::array_iterator> begin() & noexcept;
  simdjson_really_inline simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::array_iterator> end() & noexcept;
@@ -23933,6 +24014,9 @@ public:
  /** @copydoc simdjson_really_inline std::string_view value::raw_json_token() const noexcept */
  simdjson_really_inline simdjson_result<std::string_view> raw_json_token() noexcept;

+  /** @copydoc simdjson_really_inline simdjson_result<const char *> current_location() noexcept */
+  simdjson_really_inline simdjson_result<const char *> current_location() noexcept;
+
  simdjson_really_inline simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::value> at_pointer(std::string_view json_pointer) noexcept;
 };

@@ -24050,6 +24134,17 @@ public:
   * **Raw Keys:** The lookup will be done against the *raw* key, and will not unescape keys.
   * e.g. `object["a"]` will match `{ "a": 1 }`, but will *not* match `{ "\u0061": 1 }`.
   *
+   * You must consume the fields on an object one at a time. A request for a new key
+   * invalidates previous field values: it makes them unsafe. E.g., the array
+   * given by content["bids"].get_array() should not be accessed after you have called
+   * content["asks"].get_array(). You can detect such mistakes by first compiling and running
+   * the code in Debug mode (or with the macro `SIMDJSON_DEVELOPMENT_CHECKS` set to 1): an
+   * OUT_OF_ORDER_ITERATION error is generated.
+   *
+   * You are expected to access keys only once. You should access the value corresponding to a
+   * key a single time. Doing object["mykey"].to_string() and then again object["mykey"].to_string()
+   * is an error.
+   *
   * @param key The key to look up.
   * @returns The value of the field, or NO_SUCH_FIELD if the field is not in the object.
   */
@@ -24076,6 +24171,16 @@ public:
   * If you have multiple fields with a matching key ({"x": 1,  "x": 1}) be mindful
   * that only one field is returned.
   *
+   * You must consume the fields on an object one at a time. A request for a new key
+   * invalidates previous field values: it makes them unsafe. E.g., the array
+   * given by content["bids"].get_array() should not be accessed after you have called
+   * content["asks"].get_array(). You can detect such mistakes by first compiling and running
+   * the code in Debug mode (or with the macro `SIMDJSON_DEVELOPMENT_CHECKS` set to 1): an
+   * OUT_OF_ORDER_ITERATION error is generated.
+   *
+   * You are expected to access keys only once. You should access the value corresponding to a key
+   * a single time. Doing object["mykey"].to_string() and then again object["mykey"].to_string() is an error.
+   *
   * @param key The key to look up.
   * @returns The value of the field, or NO_SUCH_FIELD if the field is not in the object.
   */
@@ -24271,6 +24376,11 @@ public:
   *   ondemand::parser parser;
   *   document doc = parser.iterate(json);
   *
+   * It is expected that the content is a valid UTF-8 file, containing a valid JSON document.
+   * Otherwise the iterate method may return an error. In particular, the whole input should be
+   * valid: we do not attempt to tolerate incorrect content either before or after a JSON
+   * document.
+   *
   * ### IMPORTANT: Validate what you use
   *
   * Calling iterate on an invalid JSON document may not immediately trigger an error. The call to
@@ -24383,16 +24493,18 @@ public:
   * ### Format
   *
   * The buffer must contain a series of one or more JSON documents, concatenated into a single
-   * buffer, separated by whitespace. It effectively parses until it has a fully valid document,
+   * buffer, separated by ASCII whitespace. It effectively parses until it has a fully valid document,
   * then starts parsing the next document at that point. (It does this with more parallelism and
   * lookahead than you might think, though.)
   *
   * documents that consist of an object or array may omit the whitespace between them, concatenating
-   * with no separator. documents that consist of a single primitive (i.e. documents that are not
-   * arrays or objects) MUST be separated with whitespace.
+   * with no separator. Documents that consist of a single primitive (i.e. documents that are not
+   * arrays or objects) MUST be separated with ASCII whitespace.
+   *
+   * The characters inside a JSON document, and between JSON documents, must be valid Unicode (UTF-8).
   *
   * The documents must not exceed batch_size bytes (by default 1MB) or they will fail to parse.
-   * Setting batch_size to excessively large or excesively small values may impact negatively the
+   * Setting batch_size to excessively large or excessively small values may impact negatively the
   * performance.
   *
   * ### REQUIRED: Buffer Padding
@@ -24854,6 +24966,14 @@ inline simdjson_result<std::string_view> to_json_string(simdjson_result<SIMDJSON
 inline simdjson_result<std::string_view> to_json_string(simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::array> x);
 } // namespace simdjson

+/**
+ * We want to support argument-dependent lookup (ADL).
+ * Hence we should define operator<< in the namespace
+ * where the argument (here value, object, etc.) resides.
+ * Credit: @madhur4127
+ * See https://github.com/simdjson/simdjson/issues/1768
+ */
+namespace simdjson { namespace SIMDJSON_BUILTIN_IMPLEMENTATION { namespace ondemand {

 /**
 * Print JSON to an output stream.
@@ -24903,6 +25023,7 @@ inline std::ostream& operator<<(std::ostream& out, simdjson::SIMDJSON_BUILTIN_IM
 #if SIMDJSON_EXCEPTIONS
 inline std::ostream& operator<<(std::ostream& out, simdjson::simdjson_result<simdjson::SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::object> x);
 #endif
+}}} // namespace simdjson::SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand
 /* end file include/simdjson/generic/ondemand/serialization.h */
 /* end file include/simdjson/generic/ondemand.h */

@@ -25608,17 +25729,6 @@ inline void json_iterator::rewind() noexcept {
 SIMDJSON_PUSH_DISABLE_WARNINGS
 SIMDJSON_DISABLE_STRICT_OVERFLOW_WARNING
 simdjson_warn_unused simdjson_really_inline error_code json_iterator::skip_child(depth_t parent_depth) noexcept {
-  /***
-   * WARNING:
-   * Inside an object, a string value is a depth of +1 compared to the object. Yet a key
-   * is at the same depth as the object.
-   * But json_iterator cannot easily tell whether we are pointing at a key or a string value.
-   * Instead, it assumes that if you are pointing at a string, then it is a value, not a key.
-   * To be clear...
-   * the following code assumes that we are *not* pointing at a key. If we are then a bug
-   * will follow. Unfortunately, it is not possible for the json_iterator its to make this
-   * check.
-   */
  if (depth() <= parent_depth) { return SUCCESS; }
  switch (*return_current_and_advance()) {
    // TODO consider whether matching braces is a requirement: if non-matching braces indicates
@@ -25646,19 +25756,18 @@ simdjson_warn_unused simdjson_really_inline error_code json_iterator::skip_child
      if (at_end()) { return report_error(INCOMPLETE_ARRAY_OR_OBJECT, "Missing [ or { at start"); }
 #endif // SIMDJSON_CHECK_EOF
      break;
-    /*case '"':
+    case '"':
      if(*peek() == ':') {
-        // we are at a key!!! This is
-        // only possible if someone searched
-        // for a key in an object and the key
-        // was not found but our code then
-        // decided the consume the separating
-        // comma before returning.
+        // We are at a key!!!
+        // This might happen if you just started an object and you skip it immediately.
+        // Performance note: it would be nice to get rid of this check as it is somewhat
+        // expensive.
+        // https://github.com/simdjson/simdjson/issues/1742
        logger::log_value(*this, "key");
-        advance(); // eat up the ':'
+        return_current_and_advance(); // eat up the ':'
        break; // important!!!
      }
-      simdjson_fallthrough;*/
+      simdjson_fallthrough;
    // Anything else must be a scalar value
    default:
      // For the first scalar, we will have incremented depth already, so we decrement it here.
@@ -27789,7 +27898,7 @@ simdjson_really_inline simdjson_result<value> document_reference::find_field_uno
 simdjson_really_inline simdjson_result<value> document_reference::find_field_unordered(const char *key) & noexcept { return doc->find_field_unordered(key); }
 simdjson_really_inline simdjson_result<json_type> document_reference::type() noexcept { return doc->type(); }
 simdjson_really_inline simdjson_result<bool> document_reference::is_scalar() noexcept { return doc->is_scalar(); }
-simdjson_really_inline simdjson_result<const char *> document_reference::current_location() noexcept { return doc->current_location(); };
+simdjson_really_inline simdjson_result<const char *> document_reference::current_location() noexcept { return doc->current_location(); }
 simdjson_really_inline bool document_reference::is_negative() noexcept { return doc->is_negative(); }
 simdjson_really_inline simdjson_result<bool> document_reference::is_integer() noexcept { return doc->is_integer(); }
 simdjson_really_inline simdjson_result<number_type> document_reference::get_number_type() noexcept { return doc->get_number_type(); }
@@ -28044,6 +28153,7 @@ template<> simdjson_really_inline simdjson_result<array> value::get() noexcept {
 template<> simdjson_really_inline simdjson_result<object> value::get() noexcept { return get_object(); }
 template<> simdjson_really_inline simdjson_result<raw_json_string> value::get() noexcept { return get_raw_json_string(); }
 template<> simdjson_really_inline simdjson_result<std::string_view> value::get() noexcept { return get_string(); }
+template<> simdjson_really_inline simdjson_result<number> value::get() noexcept { return get_number(); }
 template<> simdjson_really_inline simdjson_result<double> value::get() noexcept { return get_double(); }
 template<> simdjson_really_inline simdjson_result<uint64_t> value::get() noexcept { return get_uint64(); }
 template<> simdjson_really_inline simdjson_result<int64_t> value::get() noexcept { return get_int64(); }
@@ -28096,6 +28206,13 @@ simdjson_really_inline simdjson_result<size_t> value::count_elements() & noexcep
  iter.move_at_start();
  return answer;
 }
+simdjson_really_inline simdjson_result<size_t> value::count_fields() & noexcept {
+  simdjson_result<size_t> answer;
+  auto a = get_object();
+  answer = a.count_fields();
+  iter.move_at_start();
+  return answer;
+}
 simdjson_really_inline simdjson_result<value> value::at(size_t index) noexcept {
  auto a = get_array();
  return a.at(index);
@@ -28151,6 +28268,10 @@ simdjson_really_inline std::string_view value::raw_json_token() noexcept {
  return std::string_view(reinterpret_cast<const char*>(iter.peek_start()), iter.peek_start_length());
 }

+simdjson_really_inline simdjson_result<const char *> value::current_location() noexcept {
+  return iter.json_iter().current_location();
+}
+
 simdjson_really_inline simdjson_result<value> value::at_pointer(std::string_view json_pointer) noexcept {
  json_type t;
  SIMDJSON_TRY(type().get(t));
@@ -28189,6 +28310,10 @@ simdjson_really_inline simdjson_result<size_t> simdjson_result<SIMDJSON_BUILTIN_
  if (error()) { return error(); }
  return first.count_elements();
 }
+simdjson_really_inline simdjson_result<size_t> simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::value>::count_fields() & noexcept {
+  if (error()) { return error(); }
+  return first.count_fields();
+}
 simdjson_really_inline simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::value> simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::value>::at(size_t index) noexcept {
  if (error()) { return error(); }
  return first.at(index);
@@ -28361,6 +28486,11 @@ simdjson_really_inline simdjson_result<std::string_view> simdjson_result<SIMDJSO
  return first.raw_json_token();
 }

+simdjson_really_inline simdjson_result<const char *> simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::value>::current_location() noexcept {
+  if (error()) { return error(); }
+  return first.current_location();
+}
+
 simdjson_really_inline simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::value> simdjson_result<SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::value>::at_pointer(std::string_view json_pointer) noexcept {
  if (error()) { return error(); }
  return first.at_pointer(json_pointer);
@@ -28703,7 +28833,7 @@ simdjson_warn_unused simdjson_really_inline error_code parser::allocate(size_t n
    SIMDJSON_TRY( implementation->set_capacity(new_capacity) );
    SIMDJSON_TRY( implementation->set_max_depth(new_max_depth) );
  } else {
-    SIMDJSON_TRY( simdjson::active_implementation->create_dom_parser_implementation(new_capacity, new_max_depth, implementation) );
+    SIMDJSON_TRY( simdjson::get_active_implementation()->create_dom_parser_implementation(new_capacity, new_max_depth, implementation) );
  }
  _capacity = new_capacity;
  _max_depth = new_max_depth;
@@ -29323,6 +29453,7 @@ inline simdjson_result<std::string_view> to_json_string(simdjson_result<SIMDJSON
 }
 } // namespace simdjson

+namespace simdjson { namespace SIMDJSON_BUILTIN_IMPLEMENTATION { namespace ondemand {

 #if SIMDJSON_EXCEPTIONS
 inline std::ostream& operator<<(std::ostream& out, simdjson::SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand::value x) {
@@ -29440,6 +29571,7 @@ inline std::ostream& operator<<(std::ostream& out, simdjson::SIMDJSON_BUILTIN_IM
  }
 }
 #endif
+}}} // namespace simdjson::SIMDJSON_BUILTIN_IMPLEMENTATION::ondemand
 /* end file include/simdjson/generic/ondemand/serialization-inl.h */
 /* end file include/simdjson/generic/ondemand-inl.h */

@@ -124,11 +124,11 @@ simdjson_really_inline uint64_t follows(const uint64_t match, uint64_t &overflow

 simdjson_really_inline json_block json_scanner::next(const simd::simd8x64<uint8_t>& in) {
  json_string_block strings = string_scanner.next(in);
-  // identifies the white-space and the structurat characters
+  // identifies the white-space and the structural characters
  json_character_block characters = json_character_block::classify(in);
  // The term "scalar" refers to anything except structural characters and white space
  // (so letters, numbers, quotes).
-  // We want  follows_scalar to mark anything that follows a non-quote scalar (so letters and numbers).
+  // We want follows_scalar to mark anything that follows a non-quote scalar (so letters and numbers).
  //
  // A terminal quote should either be followed by a structural character (comma, brace, bracket, colon)
  // or nothing. However, we still want ' "a string"true ' to mark the 't' of 'true' as a potential
@@ -25,6 +25,12 @@ public:
  // base_ptr[base] incrementing base as we go
  // will potentially store extra values beyond end of valid bits, so base_ptr
  // needs to be large enough to handle this
+  //
+  // If the kernel sets SIMDJSON_CUSTOM_BIT_INDEXER, then it will provide its own
+  // version of the code.
+#ifdef SIMDJSON_CUSTOM_BIT_INDEXER
+  simdjson_really_inline void write(uint32_t idx, uint64_t bits);
+#else
  simdjson_really_inline void write(uint32_t idx, uint64_t bits) {
    // In some instances, the next branch is expensive because it is mispredicted.
    // Unfortunately, in other cases,
@@ -117,6 +123,8 @@ public:
    this->tail += cnt;
 #endif
  }
+#endif // SIMDJSON_CUSTOM_BIT_INDEXER
+
 };

 class json_structural_indexer {
@@ -111,12 +111,25 @@ using namespace simd;
  simdjson_really_inline simd8<uint8_t> is_incomplete(const simd8<uint8_t> input) {
    // If the previous input's last 3 bytes match this, they're too short (they ended at EOF):
    // ... 1111____ 111_____ 11______
+#if SIMDJSON_IMPLEMENTATION_ICELAKE
+    static const uint8_t max_array[64] = {
+      255, 255, 255, 255, 255, 255, 255, 255,
+      255, 255, 255, 255, 255, 255, 255, 255,
+      255, 255, 255, 255, 255, 255, 255, 255,
+      255, 255, 255, 255, 255, 255, 255, 255,
+      255, 255, 255, 255, 255, 255, 255, 255,
+      255, 255, 255, 255, 255, 255, 255, 255,
+      255, 255, 255, 255, 255, 255, 255, 255,
+      255, 255, 255, 255, 255, 0b11110000u-1, 0b11100000u-1, 0b11000000u-1
+    };
+#else
    static const uint8_t max_array[32] = {
      255, 255, 255, 255, 255, 255, 255, 255,
      255, 255, 255, 255, 255, 255, 255, 255,
      255, 255, 255, 255, 255, 255, 255, 255,
      255, 255, 255, 255, 255, 0b11110000u-1, 0b11100000u-1, 0b11000000u-1
    };
+#endif
    const simd8<uint8_t> max_value(&max_array[sizeof(max_array)-sizeof(simd8<uint8_t>)]);
    return input.gt_bits(max_value);
  }
@@ -154,9 +167,13 @@ using namespace simd;
        this->error |= this->prev_incomplete;
      } else {
        // you might think that a for-loop would work, but under Visual Studio, it is not good enough.
-        static_assert((simd8x64<uint8_t>::NUM_CHUNKS == 2) || (simd8x64<uint8_t>::NUM_CHUNKS == 4),
-            "We support either two or four chunks per 64-byte block.");
-        if(simd8x64<uint8_t>::NUM_CHUNKS == 2) {
+        static_assert((simd8x64<uint8_t>::NUM_CHUNKS == 1)
+                ||(simd8x64<uint8_t>::NUM_CHUNKS == 2)
+                || (simd8x64<uint8_t>::NUM_CHUNKS == 4),
+                "We support one, two or four chunks per 64-byte block.");
+        if(simd8x64<uint8_t>::NUM_CHUNKS == 1) {
+          this->check_utf8_bytes(input.chunks[0], this->prev_input_block);
+        } if(simd8x64<uint8_t>::NUM_CHUNKS == 2) {
          this->check_utf8_bytes(input.chunks[0], this->prev_input_block);
          this->check_utf8_bytes(input.chunks[1], input.chunks[0]);
        } else if(simd8x64<uint8_t>::NUM_CHUNKS == 4) {
@@ -80,25 +80,25 @@ public:
  /**
   * Log that a value has been found.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_value(const char *type) const noexcept;
  /**
   * Log the start of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_start_value(const char *type) const noexcept;
  /**
   * Log the end of a multipart value.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_end_value(const char *type) const noexcept;
  /**
   * Log an error.
   *
-   * Set ENABLE_LOGGING=true in logger.h to see logging.
+   * Set LOG_ENABLED=true in logger.h to see logging.
   */
  simdjson_really_inline void log_error(const char *error) const noexcept;

@@ -0,0 +1,206 @@
+#include "simdjson/icelake/begin.h"
+
+//
+// Stage 1
+//
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace {
+
+using namespace simd;
+
+struct json_character_block {
+  static simdjson_really_inline json_character_block classify(const simd::simd8x64<uint8_t>& in);
+  //  ASCII white-space ('\r','\n','\t',' ')
+  simdjson_really_inline uint64_t whitespace() const noexcept;
+  // non-quote structural characters (comma, colon, braces, brackets)
+  simdjson_really_inline uint64_t op() const noexcept;
+  // neither a structural character nor a white-space, so letters, numbers and quotes
+  simdjson_really_inline uint64_t scalar() const noexcept;
+
+  uint64_t _whitespace; // ASCII white-space ('\r','\n','\t',' ')
+  uint64_t _op; // structural characters (comma, colon, braces, brackets but not quotes)
+};
+
+simdjson_really_inline uint64_t json_character_block::whitespace() const noexcept { return _whitespace; }
+simdjson_really_inline uint64_t json_character_block::op() const noexcept { return _op; }
+simdjson_really_inline uint64_t json_character_block::scalar() const noexcept { return ~(op() | whitespace()); }
+
+// This identifies structural characters (comma, colon, braces, brackets),
+// and ASCII white-space ('\r','\n','\t',' ').
+simdjson_really_inline json_character_block json_character_block::classify(const simd::simd8x64<uint8_t>& in) {
+  // These lookups rely on the fact that anything < 127 will match the lower 4 bits, which is why
+  // we can't use the generic lookup_16.
+  const auto whitespace_table = simd8<uint8_t>::repeat_16(' ', 100, 100, 100, 17, 100, 113, 2, 100, '\t', '\n', 112, 100, '\r', 100, 100);
+
+  // The 6 operators (:,[]{}) have these values:
+  //
+  // , 2C
+  // : 3A
+  // [ 5B
+  // { 7B
+  // ] 5D
+  // } 7D
+  //
+  // If you use | 0x20 to turn [ and ] into { and }, the lower 4 bits of each character is unique.
+  // We exploit this, using a simd 4-bit lookup to tell us which character match against, and then
+  // match it (against | 0x20).
+  //
+  // To prevent recognizing other characters, everything else gets compared with 0, which cannot
+  // match due to the | 0x20.
+  //
+  // NOTE: Due to the | 0x20, this ALSO treats <FF> and <SUB> (control characters 0C and 1A) like ,
+  // and :. This gets caught in stage 2, which checks the actual character to ensure the right
+  // operators are in the right places.
+  const auto op_table = simd8<uint8_t>::repeat_16(
+    0, 0, 0, 0,
+    0, 0, 0, 0,
+    0, 0, ':', '{', // : = 3A, [ = 5B, { = 7B
+    ',', '}', 0, 0  // , = 2C, ] = 5D, } = 7D
+  );
+
+  // We compute whitespace and op separately. If later code only uses one or the
+  // other, given the fact that all functions are aggressively inlined, we can
+  // hope that useless computations will be omitted. This is namely case when
+  // minifying (we only need whitespace).
+
+  const uint64_t whitespace = in.eq({
+    _mm512_shuffle_epi8(whitespace_table, in.chunks[0])
+  });
+  // Turn [ and ] into { and }
+  const simd8x64<uint8_t> curlified{
+    in.chunks[0] | 0x20
+  };
+  const uint64_t op = curlified.eq({
+    _mm512_shuffle_epi8(op_table, in.chunks[0])
+  });
+
+  return { whitespace, op };
+}
+
+simdjson_really_inline bool is_ascii(const simd8x64<uint8_t>& input) {
+  return input.reduce_or().is_ascii();
+}
+
+simdjson_unused simdjson_really_inline simd8<bool> must_be_continuation(const simd8<uint8_t> prev1, const simd8<uint8_t> prev2, const simd8<uint8_t> prev3) {
+  simd8<uint8_t> is_second_byte = prev1.saturating_sub(0b11000000u-1); // Only 11______ will be > 0
+  simd8<uint8_t> is_third_byte  = prev2.saturating_sub(0b11100000u-1); // Only 111_____ will be > 0
+  simd8<uint8_t> is_fourth_byte = prev3.saturating_sub(0b11110000u-1); // Only 1111____ will be > 0
+  // Caller requires a bool (all 1's). All values resulting from the subtraction will be <= 64, so signed comparison is fine.
+  return simd8<int8_t>(is_second_byte | is_third_byte | is_fourth_byte) > int8_t(0);
+}
+
+simdjson_really_inline simd8<bool> must_be_2_3_continuation(const simd8<uint8_t> prev2, const simd8<uint8_t> prev3) {
+  simd8<uint8_t> is_third_byte  = prev2.saturating_sub(0b11100000u-1); // Only 111_____ will be > 0
+  simd8<uint8_t> is_fourth_byte = prev3.saturating_sub(0b11110000u-1); // Only 1111____ will be > 0
+  // Caller requires a bool (all 1's). All values resulting from the subtraction will be <= 64, so signed comparison is fine.
+  return simd8<int8_t>(is_third_byte | is_fourth_byte) > int8_t(0);
+}
+
+} // unnamed namespace
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#include "icelake/utf8_lookup4_algorithm.h"
+// defining SIMDJSON_CUSTOM_BIT_INDEXER allows us to provide our own bit_indexer::write
+#define SIMDJSON_CUSTOM_BIT_INDEXER
+#include "generic/stage1/json_structural_indexer.h"
+// We must not forget to undefine it now:
+#undef SIMDJSON_CUSTOM_BIT_INDEXER
+
+/**
+ * We provide a custom version of bit_indexer::write using
+ * naked intrinsics.
+ * TODO: make this code more elegant.
+ */
+namespace simdjson { namespace SIMDJSON_IMPLEMENTATION { namespace { namespace stage1 {
+simdjson_really_inline void bit_indexer::write(uint32_t idx, uint64_t bits) {
+    // In some instances, the next branch is expensive because it is mispredicted.
+    // Unfortunately, in other cases,
+    // it helps tremendously.
+    if (bits == 0) { return; }
+
+    const __m512i indexes = _mm512_maskz_compress_epi8(bits, _mm512_set_epi32(
+      0x3f3e3d3c, 0x3b3a3938, 0x37363534, 0x33323130,
+      0x2f2e2d2c, 0x2b2a2928, 0x27262524, 0x23222120,
+      0x1f1e1d1c, 0x1b1a1918, 0x17161514, 0x13121110,
+      0x0f0e0d0c, 0x0b0a0908, 0x07060504, 0x03020100
+    ));
+    const __m512i start_index = _mm512_set1_epi32(idx);
+
+    const auto count = count_ones(bits);
+    __m512i t0 = _mm512_cvtepu8_epi32(_mm512_castsi512_si128(indexes));
+    _mm512_storeu_si512(this->tail, _mm512_add_epi32(t0, start_index));
+
+    if(count > 16) {
+      const __m512i t1 = _mm512_cvtepu8_epi32(_mm512_extracti32x4_epi32(indexes, 1));
+      _mm512_storeu_si512(this->tail + 16, _mm512_add_epi32(t1, start_index));
+      if(count > 32) {
+        const __m512i t2 = _mm512_cvtepu8_epi32(_mm512_extracti32x4_epi32(indexes, 2));
+        _mm512_storeu_si512(this->tail + 32, _mm512_add_epi32(t2, start_index));
+        if(count > 48) {
+          const __m512i t3 = _mm512_cvtepu8_epi32(_mm512_extracti32x4_epi32(indexes, 3));
+          _mm512_storeu_si512(this->tail + 48, _mm512_add_epi32(t3, start_index));
+        }
+      }
+    }
+    this->tail += count;
+}
+}}}}
+
+#include "generic/stage1/utf8_validator.h"
+
+//
+// Stage 2
+//
+#include "generic/stage2/tape_builder.h"
+
+//
+// Implementation-specific overrides
+//
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace {
+namespace stage1 {
+
+simdjson_really_inline uint64_t json_string_scanner::find_escaped(uint64_t backslash) {
+  if (!backslash) { uint64_t escaped = prev_escaped; prev_escaped = 0; return escaped; }
+  return find_escaped_branchless(backslash);
+}
+
+} // namespace stage1
+} // unnamed namespace
+
+simdjson_warn_unused error_code implementation::minify(const uint8_t *buf, size_t len, uint8_t *dst, size_t &dst_len) const noexcept {
+  return icelake::stage1::json_minifier::minify<128>(buf, len, dst, dst_len);
+}
+
+simdjson_warn_unused error_code dom_parser_implementation::stage1(const uint8_t *_buf, size_t _len, stage1_mode streaming) noexcept {
+  this->buf = _buf;
+  this->len = _len;
+  return icelake::stage1::json_structural_indexer::index<128>(_buf, _len, *this, streaming);
+}
+
+simdjson_warn_unused bool implementation::validate_utf8(const char *buf, size_t len) const noexcept {
+  return icelake::stage1::generic_validate_utf8(buf,len);
+}
+
+simdjson_warn_unused error_code dom_parser_implementation::stage2(dom::document &_doc) noexcept {
+  return stage2::tape_builder::parse_document<false>(*this, _doc);
+}
+
+simdjson_warn_unused error_code dom_parser_implementation::stage2_next(dom::document &_doc) noexcept {
+  return stage2::tape_builder::parse_document<true>(*this, _doc);
+}
+
+simdjson_warn_unused error_code dom_parser_implementation::parse(const uint8_t *_buf, size_t _len, dom::document &_doc) noexcept {
+  auto error = stage1(_buf, _len, stage1_mode::regular);
+  if (error) { return error; }
+  return stage2(_doc);
+}
+
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#include "simdjson/icelake/end.h"
@@ -0,0 +1,24 @@
+#include "simdjson/icelake/begin.h"
+
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+
+simdjson_warn_unused error_code implementation::create_dom_parser_implementation(
+  size_t capacity,
+  size_t max_depth,
+  std::unique_ptr<internal::dom_parser_implementation>& dst
+) const noexcept {
+  dst.reset( new (std::nothrow) dom_parser_implementation() );
+  if (!dst) { return MEMALLOC; }
+  if (auto err = dst->set_capacity(capacity))
+    return err;
+  if (auto err = dst->set_max_depth(max_depth))
+    return err;
+  return SUCCESS;
+}
+
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
+
+#include "simdjson/icelake/end.h"
+
@@ -0,0 +1,141 @@
+namespace simdjson {
+namespace SIMDJSON_IMPLEMENTATION {
+namespace {
+namespace utf8_validation {
+
+simdjson_really_inline __m512i check_special_cases(__m512i input, const __m512i prev1) {
+  __m512i mask1 = _mm512_setr_epi64(
+        0x0202020202020202,
+        0x4915012180808080,
+        0x0202020202020202,
+        0x4915012180808080,
+        0x0202020202020202,
+        0x4915012180808080,
+        0x0202020202020202,
+        0x4915012180808080);
+
+    const __m512i v_0f = _mm512_set1_epi8(0x0f);
+    __m512i index1 = _mm512_and_si512(_mm512_srli_epi16(prev1, 4), v_0f);
+
+    __m512i byte_1_high = _mm512_shuffle_epi8(mask1, index1);
+    __m512i mask2 = _mm512_setr_epi64(
+        0xcbcbcb8b8383a3e7,
+        0xcbcbdbcbcbcbcbcb,
+        0xcbcbcb8b8383a3e7,
+        0xcbcbdbcbcbcbcbcb,
+        0xcbcbcb8b8383a3e7,
+        0xcbcbdbcbcbcbcbcb,
+        0xcbcbcb8b8383a3e7,
+        0xcbcbdbcbcbcbcbcb);
+     __m512i index2 = _mm512_and_si512(prev1, v_0f);
+
+    __m512i byte_1_low = _mm512_shuffle_epi8(mask2, index2);
+    __m512i mask3 = _mm512_setr_epi64(
+        0x101010101010101,
+        0x1010101babaaee6,
+        0x101010101010101,
+        0x1010101babaaee6,
+        0x101010101010101,
+        0x1010101babaaee6,
+        0x101010101010101,
+        0x1010101babaaee6
+    );
+    __m512i index3 = _mm512_and_si512(_mm512_srli_epi16(input, 4), v_0f);
+    __m512i byte_2_high = _mm512_shuffle_epi8(mask3, index3);
+    return _mm512_ternarylogic_epi64(byte_1_high, byte_1_low, byte_2_high, 128);
+  }
+
+  simdjson_really_inline __m512i check_multibyte_lengths(const __m512i prev2,
+      const __m512i prev3, const __m512i sc) {
+
+    __m512i is_third_byte  = _mm512_subs_epu8(prev2, _mm512_set1_epi8(0b11100000u-1)); // Only 111_____ will be > 0
+    __m512i is_fourth_byte  = _mm512_subs_epu8(prev3, _mm512_set1_epi8(0b11110000u-1)); // Only 1111____ will be > 0
+    __m512i is_third_or_fourth_byte = _mm512_or_si512(is_third_byte, is_fourth_byte);
+    const __m512i v_7f = _mm512_set1_epi8(char(0x7f));
+    is_third_or_fourth_byte = _mm512_adds_epu8(v_7f, is_third_or_fourth_byte);
+    // We want to compute (is_third_or_fourth_byte AND v80) XOR sc.
+    const __m512i v_80 = _mm512_set1_epi8(char(0x80));
+    return _mm512_ternarylogic_epi32(is_third_or_fourth_byte, v_80, sc, 0b1101010);
+    // We could also do it the long way:
+    //
+    //__m512i is_third_or_fourth_byte_mask = _mm512_and_si512(is_third_or_fourth_byte, v_80);
+    //return _mm512_xor_si512(is_third_or_fourth_byte_mask, sc);
+  }
+  //
+  // Return nonzero if there are incomplete multibyte characters at the end of the block:
+  // e.g. if there is a 4-byte character, but it's 3 bytes from the end.
+  //
+  simdjson_really_inline __m512i is_incomplete(const __m512i input) {
+    // If the previous input's last 3 bytes match this, they're too short (they ended at EOF):
+    // ... 1111____ 111_____ 11______
+    const __m512i max_value = _mm512_setr_epi64(
+        0xffffffffffffffff,
+        0xffffffffffffffff,
+        0xffffffffffffffff,
+        0xffffffffffffffff,
+        0xffffffffffffffff,
+        0xffffffffffffffff,
+        0xffffffffffffffff,
+        0xbfdfefffffffffff);
+    return _mm512_subs_epu8(input, max_value);
+  }
+
+  struct utf8_checker {
+    // If this is nonzero, there has been a UTF-8 error.
+    __m512i error{};
+
+    // The last input we received
+    __m512i prev_input_block{};
+    // Whether the last input we received was incomplete (used for ASCII fast path)
+    __m512i prev_incomplete{};
+
+    //
+    // Check whether the current bytes are valid UTF-8.
+    //
+    simdjson_really_inline void check_utf8_bytes(const __m512i input, const __m512i prev_input) {
+      // Flip prev1...prev3 so we can easily determine if they are 2+, 3+ or 4+ lead bytes
+      // (2, 3, 4-byte leads become large positive numbers instead of small negative numbers)
+      const __m512i movemask = _mm512_set_epi64(13, 12, 11, 10, 9, 8, 7, 6);
+      const __m512i rotated = _mm512_permutex2var_epi64(prev_input, movemask, input);
+      __m512i prev1 = _mm512_alignr_epi8(input, rotated, 16-1);
+      __m512i prev2 = _mm512_alignr_epi8(input, rotated, 16-2);
+      __m512i prev3 = _mm512_alignr_epi8(input, rotated, 16-3);
+      __m512i sc = check_special_cases(input, prev1);
+      this->error = _mm512_or_si512(check_multibyte_lengths(prev2, prev3, sc), this->error);
+    }
+
+    // The only problem that can happen at EOF is that a multibyte character is too short
+    // or a byte value too large in the last bytes: check_special_cases only checks for bytes
+    // too large in the first of two bytes.
+    simdjson_really_inline void check_eof() {
+      // If the previous block had incomplete UTF-8 characters at the end, an ASCII block can't
+      // possibly finish them.
+      this->error = _mm512_or_si512(this->error, this->prev_incomplete);
+    }
+
+    // returns true if ASCII.
+    simdjson_really_inline bool check_next_input(const __m512i input) {
+      const __m512i v_80 = _mm512_set1_epi8(char(0x80));
+      const __mmask64 ascii = _mm512_test_epi8_mask(input, v_80);
+      if(ascii == 0) {
+        this->error = _mm512_or_si512(this->error, this->prev_incomplete);
+        return true;
+      } else {
+        this->check_utf8_bytes(input, this->prev_input_block);
+        this->prev_incomplete = is_incomplete(input);
+        this->prev_input_block = input;
+        return false;
+      }
+    }
+    // do not forget to call check_eof!
+    simdjson_really_inline error_code errors() {
+      return (_mm512_test_epi8_mask(this->error, this->error) != 0) ? error_code::UTF8_ERROR : error_code::SUCCESS;
+    }
+  }; // struct utf8_checker
+} // namespace utf8_validation
+
+using utf8_validation::utf8_checker;
+
+} // unnamed namespace
+} // namespace SIMDJSON_IMPLEMENTATION
+} // namespace simdjson
@@ -14,20 +14,41 @@ namespace internal {
 // Static array of known implementations. We're hoping these get baked into the executable
 // without requiring a static initializer.

+#if SIMDJSON_IMPLEMENTATION_ICELAKE
+static const icelake::implementation* get_icelake_singleton() {
+  static const icelake::implementation icelake_singleton{};
+  return &icelake_singleton;
+}
+#endif
 #if SIMDJSON_IMPLEMENTATION_HASWELL
-const haswell::implementation haswell_singleton{};
+static const haswell::implementation* get_haswell_singleton() {
+  static const haswell::implementation haswell_singleton{};
+  return &haswell_singleton;
+}
 #endif
 #if SIMDJSON_IMPLEMENTATION_WESTMERE
-const westmere::implementation westmere_singleton{};
+static const westmere::implementation* get_westmere_singleton() {
+  static const westmere::implementation westmere_singleton{};
+  return &westmere_singleton;
+}
 #endif // SIMDJSON_IMPLEMENTATION_WESTMERE
 #if SIMDJSON_IMPLEMENTATION_ARM64
-const arm64::implementation arm64_singleton{};
+static const arm64::implementation* get_arm64_singleton() {
+  static const arm64::implementation arm64_singleton{};
+  return &arm64_singleton;
+}
 #endif // SIMDJSON_IMPLEMENTATION_ARM64
 #if SIMDJSON_IMPLEMENTATION_PPC64
-const ppc64::implementation ppc64_singleton{};
+static const ppc64::implementation* get_ppc64_singleton() {
+  static const ppc64::implementation ppc64_singleton{};
+  return &ppc64_singleton;
+}
 #endif // SIMDJSON_IMPLEMENTATION_PPC64
 #if SIMDJSON_IMPLEMENTATION_FALLBACK
-const fallback::implementation fallback_singleton{};
+static const fallback::implementation* get_fallback_singleton() {
+  static const fallback::implementation fallback_singleton{};
+  return &fallback_singleton;
+}
 #endif // SIMDJSON_IMPLEMENTATION_FALLBACK

 /**
@@ -56,25 +77,29 @@ private:
  const implementation *set_best() const noexcept;
 };

-const detect_best_supported_implementation_on_first_use detect_best_supported_implementation_on_first_use_singleton;
-
-const std::initializer_list<const implementation *> available_implementation_pointers {
+static const std::initializer_list<const implementation *>& get_available_implementation_pointers() {
+  static const std::initializer_list<const implementation *> available_implementation_pointers {
+#if SIMDJSON_IMPLEMENTATION_ICELAKE
+    get_icelake_singleton(),
+#endif
 #if SIMDJSON_IMPLEMENTATION_HASWELL
-  &haswell_singleton,
+    get_haswell_singleton(),
 #endif
 #if SIMDJSON_IMPLEMENTATION_WESTMERE
-  &westmere_singleton,
+    get_westmere_singleton(),
 #endif
 #if SIMDJSON_IMPLEMENTATION_ARM64
-  &arm64_singleton,
+    get_arm64_singleton(),
 #endif
 #if SIMDJSON_IMPLEMENTATION_PPC64
-  &ppc64_singleton,
+    get_ppc64_singleton(),
 #endif
 #if SIMDJSON_IMPLEMENTATION_FALLBACK
-  &fallback_singleton,
+    get_fallback_singleton(),
 #endif
-}; // available_implementation_pointers
+  }; // available_implementation_pointers
+  return available_implementation_pointers;
+}

 // So we can return UNSUPPORTED_ARCHITECTURE from the parser when there is no support
 class unsupported_implementation final : public implementation {
@@ -102,25 +127,28 @@ public:
  unsupported_implementation() : implementation("unsupported", "Unsupported CPU (no detected SIMD instructions)", 0) {}
 };

-const unsupported_implementation unsupported_singleton{};
+const unsupported_implementation* get_unsupported_singleton() {
+    static const unsupported_implementation unsupported_singleton{};
+    return &unsupported_singleton;
+}

 size_t available_implementation_list::size() const noexcept {
-  return internal::available_implementation_pointers.size();
+  return internal::get_available_implementation_pointers().size();
 }
 const implementation * const *available_implementation_list::begin() const noexcept {
-  return internal::available_implementation_pointers.begin();
+  return internal::get_available_implementation_pointers().begin();
 }
 const implementation * const *available_implementation_list::end() const noexcept {
-  return internal::available_implementation_pointers.end();
+  return internal::get_available_implementation_pointers().end();
 }
 const implementation *available_implementation_list::detect_best_supported() const noexcept {
  // They are prelisted in priority order, so we just go down the list
  uint32_t supported_instruction_sets = internal::detect_supported_architectures();
-  for (const implementation *impl : internal::available_implementation_pointers) {
+  for (const implementation *impl : internal::get_available_implementation_pointers()) {
    uint32_t required_instruction_sets = impl->required_instruction_sets();
    if ((supported_instruction_sets & required_instruction_sets) == required_instruction_sets) { return impl; }
  }
-  return &unsupported_singleton; // this should never happen?
+  return get_unsupported_singleton(); // this should never happen?
 }

 const implementation *detect_best_supported_implementation_on_first_use::set_best() const noexcept {
@@ -130,31 +158,39 @@ const implementation *detect_best_supported_implementation_on_first_use::set_bes
  SIMDJSON_POP_DISABLE_WARNINGS

  if (force_implementation_name) {
-    auto force_implementation = available_implementations[force_implementation_name];
+    auto force_implementation = get_available_implementations()[force_implementation_name];
    if (force_implementation) {
-      return active_implementation = force_implementation;
+      return get_active_implementation() = force_implementation;
    } else {
      // Note: abort() and stderr usage within the library is forbidden.
-      return active_implementation = &unsupported_singleton;
+      return get_active_implementation() = get_unsupported_singleton();
    }
  }
-  return active_implementation = available_implementations.detect_best_supported();
+  return get_active_implementation() = get_available_implementations().detect_best_supported();
 }

 } // namespace internal

-SIMDJSON_DLLIMPORTEXPORT const internal::available_implementation_list available_implementations{};
-SIMDJSON_DLLIMPORTEXPORT internal::atomic_ptr<const implementation> active_implementation{&internal::detect_best_supported_implementation_on_first_use_singleton};
+SIMDJSON_DLLIMPORTEXPORT const internal::available_implementation_list& get_available_implementations() {
+  static const internal::available_implementation_list available_implementations{};
+  return available_implementations;
+}
+
+SIMDJSON_DLLIMPORTEXPORT internal::atomic_ptr<const implementation>& get_active_implementation() {
+    static const internal::detect_best_supported_implementation_on_first_use detect_best_supported_implementation_on_first_use_singleton;
+    static internal::atomic_ptr<const implementation> active_implementation{&detect_best_supported_implementation_on_first_use_singleton};
+    return active_implementation;
+}

 simdjson_warn_unused error_code minify(const char *buf, size_t len, char *dst, size_t &dst_len) noexcept {
-  return active_implementation->minify(reinterpret_cast<const uint8_t *>(buf), len, reinterpret_cast<uint8_t *>(dst), dst_len);
+  return get_active_implementation()->minify(reinterpret_cast<const uint8_t *>(buf), len, reinterpret_cast<uint8_t *>(dst), dst_len);
 }
 simdjson_warn_unused bool validate_utf8(const char *buf, size_t len) noexcept {
-  return active_implementation->validate_utf8(buf, len);
+  return get_active_implementation()->validate_utf8(buf, len);
 }

 const implementation * builtin_implementation() {
-  static const implementation * builtin_impl = available_implementations[STRINGIFY(SIMDJSON_BUILTIN_IMPLEMENTATION)];
+  static const implementation * builtin_impl = get_available_implementations()[SIMDJSON_STRINGIFY(SIMDJSON_BUILTIN_IMPLEMENTATION)];
  assert(builtin_impl);
  return builtin_impl;
 }
@@ -1,4 +1,4 @@
-#if SIMDJSON_IMPLEMENTATION_ARM64 || SIMDJSON_IMPLEMENTATION_HASWELL || SIMDJSON_IMPLEMENTATION_WESTMERE || SIMDJSON_IMPLEMENTATION_PPC64
+#if SIMDJSON_IMPLEMENTATION_ARM64 || SIMDJSON_IMPLEMENTATION_ICELAKE || SIMDJSON_IMPLEMENTATION_HASWELL || SIMDJSON_IMPLEMENTATION_WESTMERE || SIMDJSON_IMPLEMENTATION_PPC64

 #include <cstdint>

@@ -128,4 +128,4 @@ SIMDJSON_DLLIMPORTEXPORT  const uint64_t thintable_epi8[256] = {
 } // namespace internal
 } // namespace simdjson

-#endif //  SIMDJSON_IMPLEMENTATION_ARM64 || SIMDJSON_IMPLEMENTATION_HASWELL || SIMDJSON_IMPLEMENTATION_WESTMERE || SIMDJSON_IMPLEMENTATION_PPC64
+#endif //  SIMDJSON_IMPLEMENTATION_ARM64 || SIMDJSON_IMPLEMENTATION_ICELAKE || SIMDJSON_IMPLEMENTATION_HASWELL || SIMDJSON_IMPLEMENTATION_WESTMERE || SIMDJSON_IMPLEMENTATION_PPC64
@@ -19,6 +19,10 @@ SIMDJSON_DISABLE_UNDESIRED_WARNINGS
 #include "fallback/implementation.cpp"
 #include "fallback/dom_parser_implementation.cpp"
 #endif
+#if SIMDJSON_IMPLEMENTATION_ICELAKE
+#include "icelake/implementation.cpp"
+#include "icelake/dom_parser_implementation.cpp"
+#endif
 #if SIMDJSON_IMPLEMENTATION_HASWELL
 #include "haswell/implementation.cpp"
 #include "haswell/dom_parser_implementation.cpp"
@@ -3,7 +3,7 @@
 #include <cstring>

 int main(int argc, const char *argv[]) {
-  std::cout << "simdjson v" << STRINGIFY(SIMDJSON_VERSION) << " is running the " << simdjson::active_implementation->name() << " implementation." << std::endl;
+  std::cout << "simdjson v" << SIMDJSON_STRINGIFY(SIMDJSON_VERSION) << " is running the " << simdjson::get_active_implementation()->name() << " implementation." << std::endl;
  const char *expected_implementation = nullptr;
  if (argc > 1) {
    expected_implementation = argv[1];
@@ -15,7 +15,7 @@ int main(int argc, const char *argv[]) {
    }
    std::cout << "No expected implementation argument, but SIMDJSON_FORCE_IMPLEMENTATION is set to " << expected_implementation << ", so we'll check for that." << std::endl;
  }
-  if (strcmp(expected_implementation, simdjson::active_implementation->name().c_str())) {
+  if (strcmp(expected_implementation, simdjson::get_active_implementation()->name().c_str())) {
    std::cerr << "Wrong implementation! Expected " << expected_implementation << "." << std::endl;
    return EXIT_FAILURE;
  }
@@ -32,7 +32,7 @@
 #endif

 const size_t AMAZON_CELLPHONES_NDJSON_DOC_COUNT = 793;
-#define SIMDJSON_SHOW_DEFINE(x) printf("%s=%s\n", #x, STRINGIFY(x))
+#define SIMDJSON_SHOW_DEFINE(x) printf("%s=%s\n", #x, SIMDJSON_STRINGIFY(x))

 namespace number_tests {

@@ -1617,12 +1617,12 @@ namespace validate_tests {
  bool test_random() {
    std::cout << "Running " << __func__ << std::endl;
    std::vector<uint8_t> source(64,' ');
-    const simdjson::implementation *impl_fallback = simdjson::available_implementations["fallback"];
+    const simdjson::implementation *impl_fallback = simdjson::get_available_implementations()["fallback"];
    if(!impl_fallback) { return true; }
    for(size_t i = 0; i < 10000; i++) {
      std::vector<uint8_t>& s(source);
      s[i%64] ^= uint8_t(1235 * i);
-      const bool active_ok = simdjson::active_implementation->validate_utf8((const char*)s.data(), s.size());
+      const bool active_ok = simdjson::get_active_implementation()->validate_utf8((const char*)s.data(), s.size());
      const bool fallback_ok = impl_fallback->validate_utf8((const char*)s.data(), s.size());
      if(active_ok != fallback_ok) { return false; }
      s[i%64] ^= uint8_t(1235 * i);
@@ -2137,7 +2137,7 @@ int main(int argc, char *argv[]) {
  while ((c = getopt(argc, argv, "a:")) != -1) {
    switch (c) {
    case 'a': {
-      const simdjson::implementation *impl = simdjson::available_implementations[optarg];
+      const simdjson::implementation *impl = simdjson::get_available_implementations()[optarg];
      if (!impl) {
        fprintf(stderr, "Unsupported architecture value -a %s\n", optarg);
        return EXIT_FAILURE;
@@ -2146,7 +2146,7 @@ int main(int argc, char *argv[]) {
        fprintf(stderr, "The selected implementation does not match your current CPU: -a %s\n", optarg);
        return EXIT_FAILURE;
      }
-      simdjson::active_implementation = impl;
+      simdjson::get_active_implementation() = impl;
      break;
    }
    default:
@@ -2156,12 +2156,12 @@ int main(int argc, char *argv[]) {
  }
  // this is put here deliberately to check that the documentation is correct (README),
  // should this fail to compile, you should update the documentation:
-  if (simdjson::active_implementation->name() == "unsupported") {
+  if (simdjson::get_active_implementation()->name() == "unsupported") {
    printf("unsupported CPU\n");
  }
  // We want to know what we are testing.
-  std::cout << "Running tests against this implementation: " << simdjson::active_implementation->name();
-  std::cout << " (" << simdjson::active_implementation->description() << ")" << std::endl;
+  std::cout << "Running tests against this implementation: " << simdjson::get_active_implementation()->name();
+  std::cout << " (" << simdjson::get_active_implementation()->description() << ")" << std::endl;
  std::cout << "------------------------------------------------------------" << std::endl;

  std::cout << "Running basic tests." << std::endl;
@@ -858,7 +858,24 @@ namespace document_stream_tests {
    for (auto doc : docs) {
      bool_count += doc.is_bool();
    }
-    return true;
+    return (bool_count == 0);
+  }
+
+  bool fuzzaccess() {
+    std::cout << "Running " << __func__ << std::endl;
+    // Issue 38801 in oss-fuzz
+    auto json = "\xff         \n~~\n{}"_padded;
+    simdjson::dom::parser parser;
+    simdjson::dom::document_stream docs;
+    ASSERT_SUCCESS(parser.parse_many(json).get(docs));
+    size_t bool_count = 0;
+    size_t total_count = 0;
+
+    for (auto doc : docs) {
+      total_count++;
+      bool_count += doc.is_bool();
+    }
+    return (bool_count == 0) && (total_count == 1);
  }

  bool baby_fuzzer() {
@@ -892,7 +909,8 @@ namespace document_stream_tests {
  }

  bool run() {
-    return baby_fuzzer() &&
+    return fuzzaccess() &&
+           baby_fuzzer() &&
           issue1649() &&
           adversarial_single_document_array() &&
           adversarial_single_document() &&
@@ -934,7 +952,7 @@ int main(int argc, char *argv[]) {
  while ((c = getopt(argc, argv, "a:")) != -1) {
    switch (c) {
    case 'a': {
-      const simdjson::implementation *impl = simdjson::available_implementations[optarg];
+      const simdjson::implementation *impl = simdjson::get_available_implementations()[optarg];
      if (!impl) {
        fprintf(stderr, "Unsupported architecture value -a %s\n", optarg);
        return EXIT_FAILURE;
@@ -943,7 +961,7 @@ int main(int argc, char *argv[]) {
        fprintf(stderr, "The selected implementation does not match your current CPU: -a %s\n", optarg);
        return EXIT_FAILURE;
      }
-      simdjson::active_implementation = impl;
+      simdjson::get_active_implementation() = impl;
      break;
    }
    default:
@@ -954,12 +972,12 @@ int main(int argc, char *argv[]) {

  // this is put here deliberately to check that the documentation is correct (README),
  // should this fail to compile, you should update the documentation:
-  if (simdjson::active_implementation->name() == "unsupported") {
+  if (simdjson::get_active_implementation()->name() == "unsupported") {
    printf("unsupported CPU\n");
  }
  // We want to know what we are testing.
-  std::cout << "Running tests against this implementation: " << simdjson::active_implementation->name();
-  std::cout << " (" << simdjson::active_implementation->description() << ")" << std::endl;
+  std::cout << "Running tests against this implementation: " << simdjson::get_active_implementation()->name();
+  std::cout << " (" << simdjson::get_active_implementation()->description() << ")" << std::endl;
  std::cout << "------------------------------------------------------------" << std::endl;

  std::cout << "Running document_stream tests." << std::endl;
@@ -197,7 +197,7 @@ int main(int argc, char *argv[]) {
  while ((c = getopt(argc, argv, "a:")) != -1) {
    switch (c) {
    case 'a': {
-      const simdjson::implementation *impl = simdjson::available_implementations[optarg];
+      const simdjson::implementation *impl = simdjson::get_available_implementations()[optarg];
      if (!impl) {
        fprintf(stderr, "Unsupported architecture value -a %s\n", optarg);
        return EXIT_FAILURE;
@@ -206,7 +206,7 @@ int main(int argc, char *argv[]) {
        fprintf(stderr, "The selected implementation does not match your current CPU: -a %s\n", optarg);
        return EXIT_FAILURE;
      }
-      simdjson::active_implementation = impl;
+      simdjson::get_active_implementation() = impl;
      break;
    }
    default:
@@ -217,12 +217,12 @@ int main(int argc, char *argv[]) {

  // this is put here deliberately to check that the documentation is correct (README),
  // should this fail to compile, you should update the documentation:
-  if (simdjson::active_implementation->name() == "unsupported") {
+  if (simdjson::get_active_implementation()->name() == "unsupported") {
    printf("unsupported CPU\n");
  }
  // We want to know what we are testing.
-  std::cout << "Running tests against this implementation: " << simdjson::active_implementation->name();
-  std::cout << " (" << simdjson::active_implementation->description() << ")" << std::endl;
+  std::cout << "Running tests against this implementation: " << simdjson::get_active_implementation()->name();
+  std::cout << " (" << simdjson::get_active_implementation()->description() << ")" << std::endl;
  std::cout << "------------------------------------------------------------" << std::endl;

  std::cout << "Running document tests." << std::endl;
@@ -194,7 +194,7 @@ int main(int argc, char *argv[]) {
  while ((c = getopt(argc, argv, "a:")) != -1) {
    switch (c) {
    case 'a': {
-      const simdjson::implementation *impl = simdjson::available_implementations[optarg];
+      const simdjson::implementation *impl = simdjson::get_available_implementations()[optarg];
      if (!impl) {
        fprintf(stderr, "Unsupported architecture value -a %s\n", optarg);
        return EXIT_FAILURE;
@@ -203,7 +203,7 @@ int main(int argc, char *argv[]) {
        fprintf(stderr, "The selected implementation does not match your current CPU: -a %s\n", optarg);
        return EXIT_FAILURE;
      }
-      simdjson::active_implementation = impl;
+      simdjson::get_active_implementation() = impl;
      break;
    }
    default:
@@ -214,12 +214,12 @@ int main(int argc, char *argv[]) {

  // this is put here deliberately to check that the documentation is correct (README),
  // should this fail to compile, you should update the documentation:
-  if (simdjson::active_implementation->name() == "unsupported") {
+  if (simdjson::get_active_implementation()->name() == "unsupported") {
    printf("unsupported CPU\n");
  }
  // We want to know what we are testing.
-  std::cout << "Running tests against this implementation: " << simdjson::active_implementation->name();
-  std::cout << " (" << simdjson::active_implementation->description() << ")" << std::endl;
+  std::cout << "Running tests against this implementation: " << simdjson::get_active_implementation()->name();
+  std::cout << " (" << simdjson::get_active_implementation()->description() << ")" << std::endl;
  std::cout << "------------------------------------------------------------" << std::endl;
  std::cout << "Running error tests." << std::endl;
  if (!(true
@@ -117,7 +117,7 @@ int main(int argc, char *argv[]) {
  while ((c = getopt(argc, argv, "a:")) != -1) {
    switch (c) {
    case 'a': {
-      const simdjson::implementation *impl = simdjson::available_implementations[optarg];
+      const simdjson::implementation *impl = simdjson::get_available_implementations()[optarg];
      if (!impl) {
        fprintf(stderr, "Unsupported architecture value -a %s\n", optarg);
        return EXIT_FAILURE;
@@ -126,7 +126,7 @@ int main(int argc, char *argv[]) {
        fprintf(stderr, "The selected implementation does not match your current CPU: -a %s\n", optarg);
        return EXIT_FAILURE;
      }
-      simdjson::active_implementation = impl;
+      simdjson::get_active_implementation() = impl;
      break;
    }
    default:
@@ -115,7 +115,7 @@ int main(int argc, char *argv[]) {
  while ((c = getopt(argc, argv, "a:")) != -1) {
    switch (c) {
    case 'a': {
-      const simdjson::implementation *impl = simdjson::available_implementations[optarg];
+      const simdjson::implementation *impl = simdjson::get_available_implementations()[optarg];
      if (!impl) {
        fprintf(stderr, "Unsupported architecture value -a %s\n", optarg);
        return EXIT_FAILURE;
@@ -124,7 +124,7 @@ int main(int argc, char *argv[]) {
        fprintf(stderr, "The selected implementation does not match your current CPU: -a %s\n", optarg);
        return EXIT_FAILURE;
      }
-      simdjson::active_implementation = impl;
+      simdjson::get_active_implementation() = impl;
      break;
    }
    default:
@@ -179,7 +179,6 @@ static bool has_extension(const char *filename, const char *extension) {

 bool validate(const char *dirname) {
  parse_error = 0;
-  size_t total_count = 0;
  const char *extension = ".json";
  size_t dirlen = std::strlen(dirname);
  struct dirent **entry_list;
@@ -215,7 +214,6 @@ bool validate(const char *dirname) {
      float_count = 0;
      int_count = 0;
      invalid_count = 0;
-      total_count += float_count + int_count + invalid_count;
      simdjson::dom::parser parser;
      auto err = parser.parse(p).error();
      bool isok = (err == simdjson::error_code::SUCCESS);
@@ -158,7 +158,7 @@ int main(int argc, char *argv[]) {
  while ((c = getopt(argc, argv, "a:")) != -1) {
    switch (c) {
    case 'a': {
-      const simdjson::implementation *impl = simdjson::available_implementations[optarg];
+      const simdjson::implementation *impl = simdjson::get_available_implementations()[optarg];
      if (!impl) {
        fprintf(stderr, "Unsupported architecture value -a %s\n", optarg);
        return EXIT_FAILURE;
@@ -167,7 +167,7 @@ int main(int argc, char *argv[]) {
        fprintf(stderr, "The selected implementation does not match your current CPU: -a %s\n", optarg);
        return EXIT_FAILURE;
      }
-      simdjson::active_implementation = impl;
+      simdjson::get_active_implementation() = impl;
      break;
    }
    default:
@@ -178,12 +178,12 @@ int main(int argc, char *argv[]) {

  // this is put here deliberately to check that the documentation is correct (README),
  // should this fail to compile, you should update the documentation:
-  if (simdjson::active_implementation->name() == "unsupported") {
+  if (simdjson::get_active_implementation()->name() == "unsupported") {
    printf("unsupported CPU\n");
  }
  // We want to know what we are testing.
-  std::cout << "Running tests against this implementation: " << simdjson::active_implementation->name();
-  std::cout << " (" << simdjson::active_implementation->description() << ")" << std::endl;
+  std::cout << "Running tests against this implementation: " << simdjson::get_active_implementation()->name();
+  std::cout << " (" << simdjson::get_active_implementation()->description() << ")" << std::endl;
  std::cout << "------------------------------------------------------------" << std::endl;
  if(optind >= argc) {
        std::cerr << "Usage: " << argv[0] << " <directorywithjsonfiles>"
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Daniel Lemire	84b01e6346	This is a version of the avx-512 kernel which uses ternary instructions and lower-level constructions. It is not faster.	2022-05-24 17:16:06 -04:00
Daniel Lemire	90e1027bc6	Fix for ClangCL (Microsoft)	2022-05-13 20:36:50 -04:00
Daniel Lemire	41b56b2f2a	Updating the old CI tests. Making AVX-512 work by default.	2022-05-13 18:18:15 -04:00
Daniel Lemire	e1eb5d9d30	Let us follow the convention and set the macro to 1	2022-05-11 18:24:02 -04:00
Daniel Lemire	3b8b7e89e4	Further guard the use of AVX-512. We require a compiler which we know supports VBMI2.	2022-05-11 18:17:25 -04:00
Daniel Lemire	b54039bb25	Solve issue where the type of count_ones can lead to spurious warnings.	2022-05-11 17:34:43 -04:00
Daniel Lemire	75178b65e0	1. It makes AVX-512 enabled by default again on tiger-lake and ice-lake processors or better. The expectation is that frequency throttling in these cases is less of a concern for most users. To be clear, older Intel processors with AVX-512 instructions (such as cascade lake or skylake-x) will not be able to run AVX-512 instructions. 2. We adopt a much faster bitset decoding technique described in the following blog post: https://lemire.me/blog/2022/05/10/faster-bitset-decoding-using-intel-avx-512/ Credit to : Kim Walisch and Jatin Bhateja 3. The compress method which is used by _mm512_mask_compressstoreu_epi8 should be a massive performance gain but also a drastic reduction in code complexity. 4. I try to catch up on crediting valued contributors.	2022-05-11 17:26:16 -04:00
wanweiqiangintel	7453fbe878	update the implementation of operator (#1817 )	2022-05-11 14:15:37 -04:00
Eric Zhang	9d767a9f59	Some modification of basic AVX-512 implementation (#1816 ) * Change the implementation name from cascadelake to icelake. * Do VBMI2 check, so as to run AVX512 only on the newer Intel processors. Co-authored-by: root <root@ocsbesrhlrepo01.amr.corp.intel.com>	2022-05-11 09:02:22 -04:00
Daniel Lemire	b015f8959b	Adding some optimizations for AVX-512 and making it optional (this could be reversed).	2022-05-10 11:40:00 -04:00
root	8d06129667	Enabling runtime dispatch for AVX-512 kernel. (This could get reverted.)	2022-05-09 22:30:28 +00:00
Daniel Lemire	a4f3a08c65	Reworded.	2022-05-09 13:10:06 -04:00
mellonyou	6fe66d0fa7	Add cascadelake implementation, which use AVX512 Intrinsics to optimize performance(#1811 ) * Add cascadelake implementation, which use AVX512 Intrinsics to optimization the performance. * update doc/performance.md.	2022-05-09 13:02:46 -04:00
Andrea Pappacoda	1aa1d537fe	docs(basics): ondemand::element -> ::document (#1810 ) simdjson::ondemand::element doesn't exist	2022-05-05 14:55:22 -04:00
Richard Ebeling	c122462278	[skip ci] Fix accidental pastes in doc/basics.md (#1809 ) Co-authored-by: Richard Ebeling <He3lixxx@users.noreply.github.com>	2022-05-02 08:13:55 -04:00
Daniel Lemire	b6b338bea9	This simplifies the build somewhat by always including developer-options.cmake (#1806 )	2022-04-27 12:51:04 -04:00
Dirk Stolle	f72636b5b0	fix a typo in documentation (#1807 )	2022-04-26 18:01:12 -04:00
Daniel Lemire	f6a2bac11b	what happens if we remove this check for CI? (#1804 )	2022-04-23 14:09:22 -04:00
Daniel Lemire	e4ddc5446b	Fixing typo.	2022-04-10 08:16:46 -04:00
PavelP	e0e0aa9e7b	Update simdjson.h to include simdjson_version.h (#1794 ) After this change amalgamated simdjson.h has version info at the beginning of the file.	2022-03-19 08:00:24 -04:00
Daniel Lemire	9f0a2e0e86	Update to vs17 (required) (#1795 ) * Update to vs17 (required) * Moving to windows-latest. * Dropping old vs	2022-03-19 08:00:06 -04:00
Andrei Gritsiuk	203c03e90b	resolves #1784 - GCC 11.1.0 and CLang 12.0.0 complain on 'and' operator in padded_string-inl.h (#1785 ) Co-authored-by: Andrei <C:\Users\Andrei\AppData\Roaming\The Bat!>	2022-01-31 11:38:54 -05:00
Ivan Volnov	6698eb96b9	Fix C++20 build warnings (#1782 ) (#1787 ) * Fix C++20 build warnings (#1782) * Build fix for ppc64 (#1782)	2022-01-20 14:16:27 -05:00
pkubaj	30f7de387c	Fix build on FreeBSD/powerpc64* (#1786 ) FreeBSD uses sys/endian.h and bswap64().	2022-01-18 11:21:47 -05:00
Piotr Doan	e0788507fe	Missing value::get() template specialization for number (#1779 )	2022-01-11 12:07:20 -05:00
Daniel Lemire	e4740c87fb	Update README.md	2022-01-11 12:05:16 -05:00
Daniel Lemire	5beef701e7	Update CONTRIBUTORS	2022-01-03 15:55:47 -05:00
Daniel Lemire	c6f9c93c33	Use static variables to enforce initialization order. (#1773 ) (#1777 ) Fixes https://github.com/simdjson/simdjson/issues/1771 Co-authored-by: Hao Chen <chenhao.yalier@gmail.com>	2022-01-03 15:54:57 -05:00
Daniel Lemire	3a93e45dd1	Removing scoop support.	2021-12-30 12:10:11 -05:00
Daniel Lemire	6db8ceb46c	Verify and fix 1768. (#1769 ) * Verify and fix 1768. * Breaking down the namespaces for C++11 compilers.	2021-12-21 15:12:38 -05:00
Daniel Lemire	23651f0d41	Updating a macro.	2021-12-21 09:54:02 -05:00
N. Kolotov	7e1893db42	Added SIMDJSON_ prefix to STRINGIFY and to NO_SANITIZE_UNDEFINED macros. (#1766 )	2021-12-21 09:53:06 -05:00
Daniel Lemire	b4b1927b0f	Making the documentation more explicit. (#1765 )	2021-12-21 09:51:20 -05:00
Daniel Lemire	e275589832	Update implementation-selection.md	2021-12-07 13:15:05 -05:00
epoll-reactor	69ee84973a	Fix some typos in source files. (#1761 ) Co-authored-by: epoll-reactor <>	2021-12-06 16:56:15 -05:00
epoll-reactor	76bed68ec3	Fix broken link in basics documentation. (#1760 )	2021-12-04 12:21:08 -05:00
epoll-reactor	003d970cd2	Fix -Wredundant-decls warning in json_type.h. (#1759 ) Co-authored-by: threadpoolexecutor <davidkor77@gmail.com>	2021-12-02 13:49:48 -05:00
DavidKorczynski	4aaf3f030c	Add CIFuzz integration (#1757 ) Signed-off-by: David Korczynski <david@adalogics.com>	2021-11-30 10:14:26 -05:00
Daniel Lemire	2763f7e719	Fixing issue 1755 (#1756 )	2021-11-24 09:28:28 -05:00
strager	39035c39be	Fix typos in comments (#1754 )	2021-11-19 09:44:10 -05:00
strager	ac179ff40f	Fix broken links in documentation (#1753 )	2021-11-12 08:47:22 -05:00
Tyler Kennedy	b41bece32c	Fix a couple of broken links in the basics.md TOC (#1752 )	2021-11-07 19:58:32 -05:00
Daniel Lemire	9cb0c5f92d	Patching. (#1750 )	2021-11-05 19:46:57 -04:00
Daniel Lemire	1dec7b04fe	Remark.	2021-11-05 13:32:45 -04:00
strager	c6eb6f98b8	Fix typos in documentation (#1747 ) The variable is called LOG_ENABLED, not ENABLE_LOGGING. Fix the documentation in json_iterator.h.	2021-11-03 10:13:13 -04:00
Daniel Lemire	1b01969bda	Adding tests for issue 1745. (#1746 ) * Adding tests for issue 1745. * Tweaking the documentation so that it is clearer.	2021-11-03 10:12:18 -04:00
Daniel Lemire	17f3148ac7	Version 1.0.2.	2021-10-27 19:29:42 -04:00
Daniel Lemire	35b4a48e99	Fixing issue 1742 (#1743 ) * Fix for issue 1742. * Some additional tests.	2021-10-27 19:25:23 -04:00
mir4cle	c0d18452fc	Add an option to get current location from value (#1738 ) Co-authored-by: Igor Logvanev <igor.logvanev@aimtech.team>	2021-10-24 16:55:49 -04:00
Daniel Lemire	e5c9a310cf	Preparing release 1.0.1.	2021-10-20 12:18:04 -04:00
Daniel Lemire	6d308a08c5	Fixing issue 1736 (#1737 ) * Fixing issue 1736 * Updating google benchmark. * Minor trimming. * Using the variable (to silence a warning). * Adding assignment operator.	2021-10-20 12:15:35 -04:00
Daniel Lemire	9e477ddb00	Let us try to improve the documentation so that it is clearer. (#1733 ) * Let us try to improve the documentation so that it is clearer. * Minor tweaks.	2021-10-19 13:09:41 -04:00
Daniel Lemire	b7c4d1eeef	Adding test for issue 1729. (#1730 ) * Adding test for issue 1729. * Adding comment. * Trying to move to 11.7. * Tweaking. * More tweaking. * Adding additional test. * Missing "<<". * Minor update. * Removing legacy systems.	2021-10-13 09:30:37 -04:00
Daniel Lemire	4a2f10ebcf	Let us trim out these legacy systems from CI.	2021-10-12 18:24:03 -04:00
Daniel Lemire	41cb909604	Minor change to handle paranoid compilers. (#1732 )	2021-10-12 17:23:45 -04:00
Daniel Lemire	91908ade4d	Additional documentation following issue 1723 (#1724 ) * Some extra documentation regarding issue 1723. * Adding comments. * Minor fix. * [no ci] more documentation	2021-10-09 11:41:20 -04:00
Daniel Lemire	9a32c48098	Cleaning up the single-header demo. (#1731 ) * Cleaning up the single-header demo. * Updating xcode	2021-10-09 11:34:21 -04:00
Daniel Lemire	d996ffc494	Minor typo. (#1721 ) * Minor typo. * Minor fixes. * Patching...	2021-09-25 11:34:44 -04:00
Daniel Lemire	d56fdc3d04	[no ci] deleting an irrelevant comment.	2021-09-22 18:06:59 -04:00
Daniel Lemire	1dce4fed6e	[no ci] moving documentation bit to the right location. (#1719 )	2021-09-20 09:33:21 -04:00
Daniel Lemire	cae5e5342f	Additional ndjson tests. (#1717 ) * Additional ndjson tests. * Switching the data source. * Fixing.	2021-09-18 16:29:10 -04:00
Madhur Chauhan	4e609aa955	fix redundant `;` in document-inl.h (#1715 )	2021-09-17 14:02:03 -04:00
Paul Dreik	d28e5534d9	ignore unused variable (#1714 )	2021-09-12 17:37:58 -04:00