Merging upstream version 1.44.3.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

76 files changed, 13653 insertions, 0 deletions

diff --git a/fluent-bit/lib/cfl/.github/actionlint-matcher.json b/fluent-bit/lib/cfl/.github/actionlint-matcher.json
new file mode 100644
index 00000000..7b1ba251
--- /dev/null
+++ b/fluent-bit/lib/cfl/.github/actionlint-matcher.json
@@ -0,0 +1,17 @@
+{
+    "problemMatcher": [
+      {
+        "owner": "actionlint",
+        "pattern": [
+          {
+            "regexp": "^(?:\\x1b\\[\\d+m)?(.+?)(?:\\x1b\\[\\d+m)*:(?:\\x1b\\[\\d+m)*(\\d+)(?:\\x1b\\[\\d+m)*:(?:\\x1b\\[\\d+m)*(\\d+)(?:\\x1b\\[\\d+m)*: (?:\\x1b\\[\\d+m)*(.+?)(?:\\x1b\\[\\d+m)* \\[(.+?)\\]$",
+            "file": 1,
+            "line": 2,
+            "column": 3,
+            "message": 4,
+            "code": 5
+          }
+        ]
+      }
+    ]
+  }
\ No newline at end of file
diff --git a/fluent-bit/lib/cfl/.github/dependabot.yml b/fluent-bit/lib/cfl/.github/dependabot.yml
new file mode 100644
index 00000000..12f9b4a0
--- /dev/null
+++ b/fluent-bit/lib/cfl/.github/dependabot.yml
@@ -0,0 +1,12 @@
+# To get started with Dependabot version updates, you'll need to specify which
+# package ecosystems to update and where the package manifests are located.
+# Please see the documentation for all configuration options:
+# https://help.github.com/github/administering-a-repository/configuration-options-for-dependency-updates
+
+version: 2
+updates:
+  # Maintain dependencies for GitHub Actions
+  - package-ecosystem: "github-actions"
+    directory: "/"
+    schedule:
+      interval: "weekly"
\ No newline at end of file
diff --git a/fluent-bit/lib/cfl/.github/workflows/build.yaml b/fluent-bit/lib/cfl/.github/workflows/build.yaml
new file mode 100644
index 00000000..8f3ad739
--- /dev/null
+++ b/fluent-bit/lib/cfl/.github/workflows/build.yaml
@@ -0,0 +1,168 @@
+name: Build PR(s) and master branch.
+on:
+  push:
+    branches:
+      - master
+  pull_request:
+    branches:
+      - master
+    types: [opened, reopened, synchronize]
+jobs:
+  build-windows:
+    name: Build sources on amd64 for ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [windows-latest, windows-2019]
+    steps:
+      - uses: actions/checkout@v3
+      - name: Build on ${{ matrix.os }} with vs-2019
+        run: |
+          .\scripts\win_build.bat
+      - name: Run unit tests.
+        run: |
+          ctest --rerun-failed --output-on-failure -C Debug --test-dir .\tests\
+
+  build-msys2:
+    name: Build sources on MSYS2 for ${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [windows-latest, windows-2019]
+    steps:
+      - uses: actions/checkout@v3
+      - name: Get dependencies w/ chocolatey
+        uses: crazy-max/ghaction-chocolatey@v2
+        with:
+          args: install -y cmake --installargs 'ADD_CMAKE_TO_PATH=System'
+      - uses: msys2/setup-msys2@v2
+        with:
+          update: true
+          msystem: UCRT64
+          path-type: inherit
+          install: >-
+            mingw-w64-ucrt-x86_64-gcc
+      - name: Build on ${{ matrix.os }} with MSYS2
+        shell: msys2 {0}
+        run: |
+          cmake -G "MinGW Makefiles" -DCFL_TESTS=On .
+          cmake --build .
+      - name: Run unit tests.
+        run: |
+          ctest --rerun-failed --output-on-failure -C Debug --test-dir .\tests\
+
+  build-centos:
+    name: CentOS 7 build to confirm no issues once used downstream
+    runs-on: ubuntu-latest
+    container: centos:7
+    steps:
+      - name: Set up base image dependencies
+        run: |
+          yum -y update
+          yum install -y ca-certificates cmake gcc gcc-c++ make wget
+          wget http://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm
+          rpm -ivh epel-release-latest-7.noarch.rpm
+          yum install -y cmake3
+        shell: bash
+
+      - uses: actions/checkout@v3
+
+      - name: Run compilation
+        run: |
+          cmake3 -DCFL_DEV=on .
+          make
+
+  build-debian:
+    name: Debian Buster build to confirm no issues once used downstream
+    runs-on: ubuntu-latest
+    container: debian:buster
+    steps:
+      - name: Set up base image dependencies
+        run: |
+          apt-get update
+          apt-get install -y build-essential cmake make
+        shell: bash
+
+      - uses: actions/checkout@v3
+
+      - name: Run compilation
+        run: |
+          cmake -DCFL_DEV=on .
+          make
+
+  build-unix-arm64:
+    name: Build sources on arm64 for ${{ matrix.os }} - ${{ matrix.compiler }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest]
+        compiler: [ gcc, clang ]
+    steps:
+      - uses: actions/checkout@v3
+      - name: Build on ${{ matrix.os }} with ${{ matrix.compiler }}
+        uses: uraimo/run-on-arch-action@v2.3.0
+        with:
+          arch: aarch64
+          distro: ubuntu20.04
+          run: |
+            apt-get update && \
+            apt-get install -y --no-install-recommends \
+            build-essential \
+            cmake \
+            file \
+            make 
+            export CC=${{ env.compiler }}
+            cmake -DCFL_TESTS=On .
+            make all
+            CTEST_OUTPUT_ON_FAILURE=1 make test
+        env:
+          CC: ${{ matrix.compiler }}
+
+  build-unix-amd64:
+    name: Build sources on amd64 for ${{ matrix.os }} - ${{ matrix.compiler }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-latest]
+        compiler: [ gcc, clang ]
+    steps:
+      - uses: actions/checkout@v3
+      - name: Build on ${{ matrix.os }} with ${{ matrix.compiler }}
+        run: |
+          echo "CC = $CC, CXX = $CXX"
+          cmake -DCFL_TESTS=On .
+          make all
+          CTEST_OUTPUT_ON_FAILURE=1 make test
+        env:
+          CC: ${{ matrix.compiler }}
+
+  build-analysis-tests:
+    name: Build with various code analysis tools
+    strategy:
+      fail-fast: false
+      matrix:
+        preset:
+          - clang-sanitize-address
+          - clang-sanitize-memory
+          - clang-sanitize-undefined
+          - clang-sanitize-dataflow
+          - clang-sanitize-safe-stack
+          - valgrind
+    permissions:
+      contents: read
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          submodules: true
+
+      - uses: docker://lpenz/ghaction-cmake:0.19
+        with:
+          preset: ${{ matrix.preset }}
+          # dependencies_debian: ''
+          cmakeflags: '-DCFL_TESTS=On -DCFL_DEV=on .'
+          build_command: make all
diff --git a/fluent-bit/lib/cfl/.github/workflows/lint.yaml b/fluent-bit/lib/cfl/.github/workflows/lint.yaml
new file mode 100644
index 00000000..7e3dc05b
--- /dev/null
+++ b/fluent-bit/lib/cfl/.github/workflows/lint.yaml
@@ -0,0 +1,30 @@
+name: Lint PRs
+on:
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  shellcheck:
+    runs-on: ubuntu-latest
+    name: Shellcheck
+    permissions:
+      contents: read
+    steps:
+      - uses: actions/checkout@v3
+      - uses: ludeeus/action-shellcheck@master
+        with:
+          ignore_paths: lib
+
+  actionlint:
+    runs-on: ubuntu-latest
+    name: Actionlint
+    permissions:
+      contents: read
+    steps:
+      - uses: actions/checkout@v3
+      - run: |
+          echo "::add-matcher::.github/actionlint-matcher.json"
+          bash <(curl https://raw.githubusercontent.com/rhysd/actionlint/main/scripts/download-actionlint.bash)
+          ./actionlint -color -shellcheck=
+        shell: bash
+
diff --git a/fluent-bit/lib/cfl/.github/workflows/packages.yaml b/fluent-bit/lib/cfl/.github/workflows/packages.yaml
new file mode 100644
index 00000000..2c5bedbe
--- /dev/null
+++ b/fluent-bit/lib/cfl/.github/workflows/packages.yaml
@@ -0,0 +1,102 @@
+name: Build packages for master or a tagged release
+
+on:
+  push:
+    branches:
+      - master
+    # Any tag starting with 'v'
+    tags:
+      - 'v*'
+  workflow_dispatch:
+
+jobs:
+  build-distro-packages-arm64:
+    runs-on: ubuntu-latest
+    name: build arm64 packages
+    strategy:
+      fail-fast: true
+      matrix:
+        format: [ rpm, deb ]
+    steps:
+      - uses: actions/checkout@v3
+      - uses: uraimo/run-on-arch-action@v2.3.0
+        name: Build the ${{matrix.format}} packages
+        with:
+          arch: aarch64
+          distro: ubuntu20.04
+          run: |
+            apt-get update && \
+            apt-get install -y --no-install-recommends \
+            build-essential \
+            cmake \
+            file \
+            rpm \
+            make 
+            cmake .
+            echo ${{ matrix.format }} | awk '{print toupper($0)}' | xargs -I{} cpack -G {}
+
+      - name: Store the master package artifacts
+        uses: actions/upload-artifact@v3
+        with:
+          name: ${{ matrix.format }}-arm64
+          path: |
+            ./*.${{matrix.format}}
+
+  build-distro-packages-amd64:
+    name: build amd64 packages
+    strategy:
+      fail-fast: true
+      matrix:
+        format: [ rpm, deb ]
+
+    runs-on: [ ubuntu-latest ]
+    steps:
+      - uses: actions/checkout@v3
+      - name: Build the ${{matrix.format}} packages
+        run:  |
+          cmake .
+          echo ${{ matrix.format }} | awk '{print toupper($0)}' | xargs -I{} cpack -G {}
+
+      - name: Store the master package artifacts
+        uses: actions/upload-artifact@v3
+        with:
+          name: ${{ matrix.format }}-amd64
+          path: |
+            ./*.${{matrix.format}}
+
+  release:
+    name: Create release and upload packages
+    needs:
+      - build-distro-packages-amd64
+      - build-distro-packages-arm64
+    runs-on: ubuntu-latest
+    permissions:
+      contents: write
+    steps:
+      - name: Download all artefacts
+        uses: actions/download-artifact@v3
+        with:
+          path: artifacts/
+
+      - name: Display structure of downloaded files
+        run: ls -R
+        working-directory: artifacts
+        shell: bash
+
+      - name: Unstable release on push to master to make it easier to download
+        uses: pyTooling/Actions/releaser@r0
+        continue-on-error: true
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          tag: 'unstable'
+          rm: true
+          files: |
+            artifacts/**/*
+
+      - name: Release on tag
+        uses: softprops/action-gh-release@v1
+        if: startsWith(github.ref, 'refs/tags/')
+        with:
+          generate_release_notes: true
+          files: |
+            artifacts/**/*
diff --git a/fluent-bit/lib/cfl/.gitignore b/fluent-bit/lib/cfl/.gitignore
new file mode 100644
index 00000000..db0ae743
--- /dev/null
+++ b/fluent-bit/lib/cfl/.gitignore
@@ -0,0 +1,6 @@
+build/*
+include/cfl/cfl_info.h
+include/cfl/cfl_version.h
+cfl.sublime*
+tests/cfl_tests_internal.h
+
diff --git a/fluent-bit/lib/cfl/CMakeLists.txt b/fluent-bit/lib/cfl/CMakeLists.txt
new file mode 100644
index 00000000..98ee88f7
--- /dev/null
+++ b/fluent-bit/lib/cfl/CMakeLists.txt
@@ -0,0 +1,270 @@
+cmake_minimum_required(VERSION 3.12)
+project(cfl C)
+set(CMAKE_POSITION_INDEPENDENT_CODE ON)
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+
+# C Floppy Version
+set(CFL_VERSION_MAJOR  0)
+set(CFL_VERSION_MINOR  2)
+set(CFL_VERSION_PATCH  2)
+set(CFL_VERSION_STR    "${CFL_VERSION_MAJOR}.${CFL_VERSION_MINOR}.${CFL_VERSION_PATCH}")
+
+# Configuration options
+option(CFL_DEV                            "Enable development mode"                    No)
+option(CFL_TESTS                          "Enable unit testing"                        No)
+option(CFL_INSTALL_BUNDLED_XXHASH_HEADERS "Enable bundled xxHash headers installation" Yes)
+
+if(CFL_DEV)
+  set(CMAKE_BUILD_TYPE Debug)
+  set(CFL_TESTS        On)
+endif()
+
+# Include helpers
+include(cmake/macros.cmake)
+include(CheckCSourceCompiles)
+include(GNUInstallDirs)
+
+# Define macro to identify Windows system (without Cygwin)
+if(CMAKE_SYSTEM_NAME MATCHES "Windows")
+  set(CFL_SYSTEM_WINDOWS On)
+  add_definitions(-DCFL_SYSTEM_WINDOWS)
+endif()
+
+# Define macro to identify macOS system
+if(CMAKE_SYSTEM_NAME MATCHES "Darwin")
+  set(CFL_SYSTEM_MACOS On)
+  add_definitions(-DCFL_SYSTEM_MACOS)
+endif()
+
+if(NOT MSVC)
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wall")
+endif()
+
+# Define __FILENAME__ consistently across Operating Systems
+if(NOT ${CMAKE_SYSTEM_NAME} MATCHES "Windows")
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -D__FILENAME__='\"$(subst ${CMAKE_SOURCE_DIR}/,,$(abspath $<))\"'")
+else()
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -D__FILENAME__=__FILE__")
+endif()
+
+
+
+# timespec_get() support
+check_c_source_compiles("
+  #include <time.h>
+  int main() {
+     struct tm tm;
+     return timespec_get(&tm, TIME_UTC);
+  }" CFL_HAVE_TIMESPEC_GET)
+if(CFL_HAVE_TIMESPEC_GET)
+  CFL_DEFINITION(CFL_HAVE_TIMESPEC_GET)
+endif()
+
+# gmtime_r() support
+check_c_source_compiles("
+  #include <time.h>
+  int main() {
+     struct tm tm;
+     struct timespec tms;
+     return gmtime_r(&tms.tv_sec, &tm);
+  }" CFL_HAVE_GMTIME_R)
+if(CFL_HAVE_GMTIME_R)
+  CFL_DEFINITION(CFL_HAVE_GMTIME_R)
+endif()
+
+# gmtime_s() support
+check_c_source_compiles("
+  #include <time.h>
+  int main() {
+     struct tm tm;
+     struct timespec tms;
+     return gmtime_s(&tm, &tms.tv_sec);
+  }" CFL_HAVE_GMTIME_S)
+if(CFL_HAVE_GMTIME_S)
+  CFL_DEFINITION(CFL_HAVE_GMTIME_S)
+endif()
+
+# clock_get_time() support for macOS.
+check_c_source_compiles("
+  #include <mach/clock.h>
+  #include <mach/mach.h>
+  int main() {
+      clock_serv_t cclock;
+      mach_timespec_t mts;
+      host_get_clock_service(mach_host_self(), CALENDAR_CLOCK, &cclock);
+      clock_get_time(cclock, &mts);
+      return mach_port_deallocate(mach_task_self(), cclock);
+  }" CFL_HAVE_CLOCK_GET_TIME)
+if(CFL_HAVE_CLOCK_GET_TIME)
+  CFL_DEFINITION(CFL_HAVE_CLOCK_GET_TIME)
+endif()
+
+configure_file(
+  "${PROJECT_SOURCE_DIR}/include/cfl/cfl_info.h.in"
+  "${PROJECT_SOURCE_DIR}/include/cfl/cfl_info.h"
+  )
+
+configure_file(
+  "${PROJECT_SOURCE_DIR}/include/cfl/cfl_version.h.in"
+  "${PROJECT_SOURCE_DIR}/include/cfl/cfl_version.h"
+  )
+
+# Include headers and dependency headers
+include_directories(
+  lib/xxhash/
+  include
+  )
+
+# xxHash
+if(NOT TARGET xxhash)
+  # Do something when target found
+  set(XXHASH_BUILD_ENABLE_INLINE_API OFF)
+  set(XXHASH_BUILD_XXHSUM OFF)
+  set(BUILD_SHARED_LIBS OFF)
+  add_subdirectory(lib/xxhash/cmake_unofficial EXCLUDE_FROM_ALL)
+endif()
+
+# Installation Directories
+# ========================
+if(${CMAKE_SYSTEM_NAME} MATCHES "Windows")
+  set(CFL_INSTALL_BINDIR "bin")
+  set(CFL_INSTALL_LIBDIR "lib")
+  set(CFL_INSTALL_INCLUDEDIR "include")
+else()
+  set(CFL_INSTALL_BINDIR ${CMAKE_INSTALL_FULL_BINDIR})
+  set(CFL_INSTALL_LIBDIR "${CMAKE_INSTALL_LIBDIR}")
+  set(CFL_INSTALL_INCLUDEDIR "${CMAKE_INSTALL_PREFIX}/include")
+endif()
+
+# Output paths
+set(CFL_ROOT "${CMAKE_CURRENT_SOURCE_DIR}")
+set(CFL_BUILD_DIR "${CFL_ROOT}/build")
+
+# CFL sources
+add_subdirectory(include)
+add_subdirectory(src)
+
+# Tests
+if(CFL_TESTS)
+  enable_testing()
+  add_subdirectory(tests)
+endif()
+
+# Installer Generation (Cpack)
+# ============================
+
+set(CPACK_PACKAGE_VERSION ${CFL_VERSION_STR})
+set(CPACK_PACKAGE_NAME "cfl")
+set(CPACK_PACKAGE_RELEASE 1)
+set(CPACK_PACKAGE_CONTACT "Eduardo Silva <eduardo@calyptia.com>")
+set(CPACK_PACKAGE_VENDOR "Calyptia")
+set(CPACK_RESOURCE_FILE_LICENSE "${PROJECT_SOURCE_DIR}/LICENSE")
+set(CPACK_PACKAGING_INSTALL_PREFIX "/")
+
+set(CPACK_PACKAGE_FILE_NAME "${CPACK_PACKAGE_NAME}-${CPACK_PACKAGE_VERSION}-${CPACK_PACKAGE_RELEASE}.${CMAKE_SYSTEM_PROCESSOR}")
+
+if(CFL_SYSTEM_WINDOWS)
+  set(CPACK_GENERATOR "ZIP")
+
+  if(CMAKE_SIZEOF_VOID_P EQUAL 8)
+    set(CPACK_PACKAGE_FILE_NAME "${CPACK_PACKAGE_NAME}-${CPACK_PACKAGE_VERSION}-win64")
+  else()
+    set(CPACK_PACKAGE_FILE_NAME "${CPACK_PACKAGE_NAME}-${CPACK_PACKAGE_VERSION}-win32")
+  endif()
+endif()
+
+
+# Enable components
+set(CPACK_DEB_COMPONENT_INSTALL ON)
+set(CPACK_RPM_COMPONENT_INSTALL ON)
+set(CPACK_productbuild_COMPONENT_INSTALL ON)
+set(CPACK_COMPONENTS_ALL ${CPACK_COMPONENTS_ALL} binary library headers)
+set(CPACK_COMPONENTS_GROUPING "ONE_PER_GROUP")
+
+set(CPACK_COMPONENT_BINARY_GROUP "RUNTIME")
+set(CPACK_COMPONENT_LIBRARY_GROUP "RUNTIME")
+
+# Debian package setup and name sanitizer
+set(CPACK_DEBIAN_PACKAGE_SHLIBDEPS ON)
+
+find_program(DPKG_PROGRAM dpkg DOC "dpkg program of Debian-based systems")
+if(DPKG_PROGRAM)
+  execute_process(
+    COMMAND ${DPKG_PROGRAM} --print-architecture
+    OUTPUT_VARIABLE CPACK_DEBIAN_PACKAGE_ARCHITECTURE
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    )
+
+  set(CPACK_DEBIAN_HEADERS_FILE_NAME "${CPACK_PACKAGE_NAME}_${CPACK_PACKAGE_VERSION}_${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}-headers.deb")
+  set(CPACK_DEBIAN_RUNTIME_PACKAGE_NAME "${CPACK_PACKAGE_NAME}")
+  set(CPACK_DEBIAN_RUNTIME_FILE_NAME "${CPACK_PACKAGE_NAME}_${CPACK_PACKAGE_VERSION}_${CPACK_DEBIAN_PACKAGE_ARCHITECTURE}.deb")
+  set(CPACK_DEBIAN_RUNTIME_PACKAGE_CONTROL_EXTRA
+    ${CMAKE_CURRENT_SOURCE_DIR}/debian/conffiles
+    )
+endif()
+
+# RPM Generation information
+set(CPACK_RPM_PACKAGE_GROUP "System Environment/Daemons")
+set(CPACK_RPM_PACKAGE_LICENSE "Apache v2.0")
+set(CPACK_RPM_PACKAGE_RELEASE ${CPACK_PACKAGE_RELEASE})
+set(CPACK_PACKAGE_DESCRIPTION_FILE "${PROJECT_SOURCE_DIR}/cpack/description")
+set(CPACK_PACKAGE_DESCRIPTION_SUMMARY "C Traces Library")
+set(CPACK_RPM_SPEC_MORE_DEFINE "%define ignore \#")
+set(CPACK_RPM_USER_FILELIST
+  "%ignore /lib"
+  "%ignore /lib64"
+  "%ignore /lib64/pkgconfig"
+  "%ignore /usr/local"
+  "%ignore /usr/local/bin")
+
+set(CPACK_RPM_PACKAGE_AUTOREQ ON)
+set(CPACK_RPM_RUNTIME_PACKAGE_NAME "${CPACK_PACKAGE_NAME}")
+set(CPACK_RPM_HEADERS_FILE_NAME "${CPACK_PACKAGE_NAME}-${CPACK_PACKAGE_VERSION}-${CPACK_PACKAGE_RELEASE}.${CMAKE_SYSTEM_PROCESSOR}-headers.rpm")
+set(CPACK_RPM_RUNTIME_FILE_NAME "${CPACK_PACKAGE_NAME}-${CPACK_PACKAGE_VERSION}-${CPACK_PACKAGE_RELEASE}.${CMAKE_SYSTEM_PROCESSOR}.rpm")
+
+# CPack: DEB
+set(CPACK_DEBIAN_PACKAGE_SHLIBDEPS ON)
+
+# CPack: Windows System
+if(CPACK_GENERATOR MATCHES "ZIP")
+  set(CPACK_MONOLITHIC_INSTALL 1)
+  set(CPACK_PACKAGE_INSTALL_DIRECTORY "cfl")
+endif()
+
+# CPack: macOS w/ productbuild
+if(CFL_SYSTEM_MACOS)
+  # Determine the platform suffix
+  execute_process(
+    COMMAND uname -m
+    RESULT_VARIABLE UNAME_M_RESULT
+    OUTPUT_VARIABLE UNAME_ARCH
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    )
+  if (UNAME_M_RESULT EQUAL 0 AND UNAME_ARCH STREQUAL "arm64")
+    set(CFL_PKG ${CFL_BUILD_DIR}/${CPACK_PACKAGE_NAME}-${CFL_VERSION_STR}-apple)
+  elseif(UNAME_M_RESULT EQUAL 0 AND UNAME_ARCH STREQUAL "x86_64")
+    set(CFL_PKG ${CFL_BUILD_DIR}/${CPACK_PACKAGE_NAME}-${CFL_VERSION_STR}-intel)
+  else()
+    set(CFL_PKG ${CFL_BUILD_DIR}/${CPACK_PACKAGE_NAME}-${CFL_VERSION_STR}-${UNAME_ARCH})
+  endif()
+
+  if (CPACK_GENERATOR MATCHES "productbuild")
+    set(CPACK_SET_DESTDIR "ON")
+    configure_file(cpack/macos/welcome.txt.cmakein ${CFL_BUILD_DIR}/welcome.txt)
+    configure_file(LICENSE ${CFL_BUILD_DIR}/LICENSE.txt)
+    find_program(CONVERTER textutil)
+    if (NOT CONVERTER)
+      message(FATAL_ERROR "textutil not found.")
+    endif()
+    if (CONVERTER)
+      execute_process(COMMAND ${CONVERTER} -convert html "${CMAKE_SOURCE_DIR}/README.md" -output "${CFL_BUILD_DIR}/README.html")
+    endif()
+    set(CPACK_PACKAGE_FILE_NAME "${CFL_PKG}")
+    set(CPACK_RESOURCE_FILE_WELCOME ${CFL_BUILD_DIR}/welcome.txt)
+    set(CPACK_RESOURCE_FILE_LICENSE ${CFL_BUILD_DIR}/LICENSE.txt)
+    set(CPACK_RESOURCE_FILE_README ${CFL_BUILD_DIR}/README.html)
+    set(CPACK_PRODUCTBUILD_IDENTIFIER "com.calyptia.${CPACK_PACKAGE_NAME}")
+  endif()
+endif()
+
+include(CPack)
diff --git a/fluent-bit/lib/cfl/LICENSE b/fluent-bit/lib/cfl/LICENSE
new file mode 100644
index 00000000..f433b1a5
--- /dev/null
+++ b/fluent-bit/lib/cfl/LICENSE
@@ -0,0 +1,177 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
diff --git a/fluent-bit/lib/cfl/README.md b/fluent-bit/lib/cfl/README.md
new file mode 100644
index 00000000..4050bef8
--- /dev/null
+++ b/fluent-bit/lib/cfl/README.md
@@ -0,0 +1,23 @@
+# CFL
+
+CFL is a tiny library that provides interfaces for data structures, originally created to satisfy the needs of Fluent Bit and other libraries used internally like CMetrics and CTraces projects.
+
+note: The name doesn't mean anything specific, you can call it `c:\ floppy` if you want.
+
+## Interfaces
+
+- cfl_sds: string manipulation
+- cfl_list: linked list 
+- cfl_kv: key value pairs by using a linked list (cfl_list)
+- cfl_array: array of elements
+- cfl_variant: interface to manage contexts with vairant types
+- cfl_time: time utilities
+- cfl_hash: 64bit hashing functions
+
+## License
+
+This program is under the terms of the [Apache License v2.0](http://www.apache.org/licenses/LICENSE-2.0).
+
+## Authors
+
+Copyright is assigned to the `CFL Authors`, you can see a list of contributors [here](https://github.com/fluent/cfl/graphs/contributors).
diff --git a/fluent-bit/lib/cfl/appveyor.yml b/fluent-bit/lib/cfl/appveyor.yml
new file mode 100644
index 00000000..1f60cfeb
--- /dev/null
+++ b/fluent-bit/lib/cfl/appveyor.yml
@@ -0,0 +1,34 @@
+version: v1-winbuild-{build}
+
+image: Visual Studio 2019
+
+platform:
+  - Win32
+  - x64
+
+environment:
+  vspath: 'C:\Program Files (x86)\Microsoft Visual Studio\2019\Community'
+  winflexbison: https://github.com/lexxmark/winflexbison/releases/download/v2.5.22/win_flex_bison-2.5.22.zip
+  PATH: '%PATH%;C:\WinFlexBison'
+
+configuration:
+  - Release
+  
+skip_commits:
+  message: /workflows/
+  files:
+    - '.github/**'
+
+install:
+  - ps: Invoke-WebRequest -O winflexbison.zip $env:winflexbison
+  - ps: Expand-Archive winflexbison.zip -Destination /WinFlexBison
+  - ps: Copy-Item -Path /WinFlexBison/win_bison.exe /WinFlexBison/bison.exe
+  - ps: Copy-Item -Path /WinFlexBison/win_flex.exe /WinFlexBison/flex.exe
+
+before_build:
+  - if %PLATFORM%==Win32 call "%vspath%\VC\Auxiliary\Build\vcvars32.bat"
+  - if %PLATFORM%==x64   call "%vspatH%\VC\Auxiliary\Build\vcvars64.bat"
+  
+build_script:
+  - .\scripts\win_build.bat
+  - ctest -C Debug --test-dir .\tests\
diff --git a/fluent-bit/lib/cfl/cmake/macros.cmake b/fluent-bit/lib/cfl/cmake/macros.cmake
new file mode 100644
index 00000000..bbbd09be
--- /dev/null
+++ b/fluent-bit/lib/cfl/cmake/macros.cmake
@@ -0,0 +1,10 @@
+# Macro to set definitions
+macro(CFL_DEFINITION var)
+  add_definitions(-D${var})
+  set(CFL_BUILD_FLAGS "${CFL_BUILD_FLAGS}#ifndef ${var}\n#define ${var}\n#endif\n")
+  set(CFL_INFO_FLAGS "${CFL_INFO_FLAGS} ${var}")
+endmacro()
+
+macro(CFL_OPTION option value)
+  set(${option} ${value} CACHE INTERNAL "" FORCE)
+endmacro()
diff --git a/fluent-bit/lib/cfl/cpack/description b/fluent-bit/lib/cfl/cpack/description
new file mode 100644
index 00000000..9293bf13
--- /dev/null
+++ b/fluent-bit/lib/cfl/cpack/description
@@ -0,0 +1 @@
+Tiny library for data structures, call it C Floppy ;)
diff --git a/fluent-bit/lib/cfl/cpack/macos/welcome.txt.cmakein b/fluent-bit/lib/cfl/cpack/macos/welcome.txt.cmakein
new file mode 100644
index 00000000..0ef28bc2
--- /dev/null
+++ b/fluent-bit/lib/cfl/cpack/macos/welcome.txt.cmakein
@@ -0,0 +1,6 @@
+This will install @CPACK_PACKAGE_NAME@ on your Mac.
+
+--------------------------------------------------
+
+Thank you for trying @CPACK_PACKAGE_NAME@! Have a fantastic day!
+
diff --git a/fluent-bit/lib/cfl/include/CMakeLists.txt b/fluent-bit/lib/cfl/include/CMakeLists.txt
new file mode 100644
index 00000000..bf2e3879
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/CMakeLists.txt
@@ -0,0 +1,17 @@
+file(GLOB cflHeaders "cfl/*.h")
+install(FILES ${cflHeaders}
+    DESTINATION ${CFL_INSTALL_INCLUDEDIR}/cfl
+    COMPONENT headers
+    PERMISSIONS OWNER_READ OWNER_WRITE GROUP_READ WORLD_READ)
+
+# xxHash
+if(CFL_INSTALL_BUNDLED_XXHASH_HEADERS)
+  install(FILES ${CFL_ROOT}/lib/xxhash/xxh3.h
+    DESTINATION ${CFL_INSTALL_INCLUDEDIR}
+    COMPONENT headers
+    PERMISSIONS OWNER_READ OWNER_WRITE GROUP_READ WORLD_READ)
+  install(FILES ${CFL_ROOT}/lib/xxhash/xxhash.h
+    DESTINATION ${CFL_INSTALL_INCLUDEDIR}
+    COMPONENT headers
+    PERMISSIONS OWNER_READ OWNER_WRITE GROUP_READ WORLD_READ)
+endif()
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl.h b/fluent-bit/lib/cfl/include/cfl/cfl.h
new file mode 100644
index 00000000..fee8fcf4
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl.h
@@ -0,0 +1,42 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_H
+#define CFL_H
+
+#define CFL_FALSE   0
+#define CFL_TRUE    !CFL_FALSE
+
+/* headers that are needed in general */
+#include <cfl/cfl_info.h>
+#include <cfl/cfl_version.h>
+#include <cfl/cfl_compat.h>
+#include <cfl/cfl_log.h>
+#include <cfl/cfl_sds.h>
+#include <cfl/cfl_list.h>
+#include <cfl/cfl_hash.h>
+#include <cfl/cfl_array.h>
+#include <cfl/cfl_kv.h>
+#include <cfl/cfl_kvlist.h>
+#include <cfl/cfl_time.h>
+#include <cfl/cfl_variant.h>
+
+int cfl_init();
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_array.h b/fluent-bit/lib/cfl/include/cfl/cfl_array.h
new file mode 100644
index 00000000..e260d24d
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_array.h
@@ -0,0 +1,62 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_ARRAY_H
+#define CFL_ARRAY_H
+
+#include <stdio.h>
+#include <cfl/cfl_variant.h>
+
+struct cfl_array {
+    int                  resizable;
+    struct cfl_variant **entries;
+    size_t               slot_count;
+    size_t               entry_count;
+};
+
+struct cfl_array *cfl_array_create(size_t slot_count);
+void cfl_array_destroy(struct cfl_array *array);
+
+static inline struct cfl_variant *cfl_array_fetch_by_index(struct cfl_array *array,
+                                                           size_t position)
+{
+    if (position >= array->entry_count) {
+        return NULL;
+    }
+
+    return array->entries[position];
+}
+
+int cfl_array_resizable(struct cfl_array *array, int v);
+int cfl_array_remove_by_index(struct cfl_array *array, size_t position);
+int cfl_array_remove_by_reference(struct cfl_array *array, struct cfl_variant *value);
+int cfl_array_append(struct cfl_array *array, struct cfl_variant *value);
+int cfl_array_append_string(struct cfl_array *array, char *value);
+int cfl_array_append_bytes(struct cfl_array *array, char *value, size_t length);
+int cfl_array_append_reference(struct cfl_array *array, void *value);
+int cfl_array_append_bool(struct cfl_array *array, int value);
+int cfl_array_append_int64(struct cfl_array *array, int64_t value);
+int cfl_array_append_double(struct cfl_array *array, double value);
+int cfl_array_append_array(struct cfl_array *array, struct cfl_array *value);
+int cfl_array_append_new_array(struct cfl_array *array, size_t size);
+int cfl_array_append_kvlist(struct cfl_array *array, struct 
+cfl_kvlist *value);
+int cfl_array_print(FILE *fp, struct cfl_array *array);
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_checksum.h b/fluent-bit/lib/cfl/include/cfl/cfl_checksum.h
new file mode 100644
index 00000000..d9690aa8
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_checksum.h
@@ -0,0 +1,27 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_CHECKSUM_H
+#define CFL_CHECKSUM_H
+
+#include <stdint.h>
+
+uint32_t cfl_checksum_crc32c(unsigned char *buffer, size_t length);
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_compat.h b/fluent-bit/lib/cfl/include/cfl/cfl_compat.h
new file mode 100644
index 00000000..36c56233
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_compat.h
@@ -0,0 +1,46 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*
+ * This file contains compatibility functions and macros for various platforms.
+ *
+ * Including this header file should make platforms behave more consistently;
+ * Add more macros if you find any missing features.
+ */
+
+#ifndef CFL_COMPAT_H
+#define CFL_COMPAT_H
+
+#ifdef CFL_SYSTEM_WINDOWS
+
+#ifdef _MSC_VER
+/*
+ * cl.exe that is one of the C++ compilers for Windows prefers
+ * to add an underscore to each POSIX function.
+ * To suppress compiler warnings, we need these trivial macros.
+ * For MSYS2 platform on Windows, we don't need to do.
+ */
+#define timezone _timezone
+#define tzname _tzname
+#define strncasecmp _strnicmp
+#define timegm _mkgmtime
+#endif /* _MSC_VER */
+
+#endif
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_found.h b/fluent-bit/lib/cfl/include/cfl/cfl_found.h
new file mode 100644
index 00000000..d6416be8
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_found.h
@@ -0,0 +1,47 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*
+ * This is a dummy header that can be used by parent projects to check if
+ * CFL headers are found in their path.
+ *
+ * Why ?, because <cfl/cfl.h> includes "cfl_info.h" which is only generated once
+ * CMake runs in CFL project. Likely this check is done before that.
+ *
+ *
+ * In order to use it, the caller might try to use check_c_source_compiles() CMake function
+ * and try to include this header and invoke the inline function defined here, e.g:
+ *
+ *   check_c_source_compiles("
+ *     include <cfl/cfl_found.h>
+ *
+ *     int main() {
+ *        return cfl_found();
+ *     }" CFL_FOUND)
+ */
+
+#ifndef CFL_FOUND_H
+#define CFL_FOUND_H
+
+static inline int cfl_found()
+{
+    return 0;
+}
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_hash.h b/fluent-bit/lib/cfl/include/cfl/cfl_hash.h
new file mode 100644
index 00000000..fb2fc4e6
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_hash.h
@@ -0,0 +1,38 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_HASH_H
+#define CFL_HASH_H
+
+/* NOTE: this is just a wrapper for naming convention */
+
+#include <stdint.h>
+#include "xxh3.h"
+
+#define cfl_hash_64bits_t      XXH64_hash_t
+#define cfl_hash_state_t       XXH3_state_t
+#define cfl_hash_64bits_reset  XXH3_64bits_reset
+#define cfl_hash_64bits_update XXH3_64bits_update
+#define cfl_hash_64bits_digest XXH3_64bits_digest
+#define cfl_hash_64bits        XXH3_64bits
+
+#define cfl_hash_128bits_t     XXH128_hash_t
+#define cfl_hash_128bits       XXH3_128bits
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_info.h.in b/fluent-bit/lib/cfl/include/cfl/cfl_info.h.in
new file mode 100644
index 00000000..e2f0d511
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_info.h.in
@@ -0,0 +1,28 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_INFO_H
+#define CFL_INFO_H
+
+#define CFL_SOURCE_DIR "@CMAKE_SOURCE_DIR@"
+
+/* General flags set by /CMakeLists.txt */
+@CFL_BUILD_FLAGS@
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_kv.h b/fluent-bit/lib/cfl/include/cfl/cfl_kv.h
new file mode 100644
index 00000000..f44c5be6
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_kv.h
@@ -0,0 +1,43 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_KV_H
+#define CFL_KV_H
+
+#include <cfl/cfl_info.h>
+#include <cfl/cfl_sds.h>
+#include <cfl/cfl_list.h>
+
+struct cfl_kv {
+  cfl_sds_t       key;
+  cfl_sds_t       val;
+  struct cfl_list _head;
+};
+
+void cfl_kv_init(struct cfl_list *list);
+struct cfl_kv *cfl_kv_item_create_len(struct cfl_list *list,
+                                      char *k_buf, size_t k_len,
+                                      char *v_buf, size_t v_len);
+struct cfl_kv *cfl_kv_item_create(struct cfl_list *list,
+                                  char *k_buf, char *v_buf);
+void cfl_kv_item_destroy(struct cfl_kv *kv);
+void cfl_kv_release(struct cfl_list *list);
+const char *cfl_kv_get_key_value(const char *key, struct cfl_list *list);
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_kvlist.h b/fluent-bit/lib/cfl/include/cfl/cfl_kvlist.h
new file mode 100644
index 00000000..5f920e58
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_kvlist.h
@@ -0,0 +1,76 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_KVLIST_H
+#define CFL_KVLIST_H
+
+#include <stdio.h>
+#include <cfl/cfl_sds.h>
+#include <cfl/cfl_list.h>
+#include <cfl/cfl_variant.h>
+
+struct cfl_kvpair {
+    cfl_sds_t            key;    /* Key */
+    struct cfl_variant   *val;   /* Value */
+    struct cfl_list      _head;  /* Link to list cfl_kvlist->list */
+};
+
+struct cfl_kvlist {
+    struct cfl_list list;
+};
+
+struct cfl_kvlist *cfl_kvlist_create();
+void cfl_kvlist_destroy(struct cfl_kvlist *list);
+
+int cfl_kvlist_insert_string(struct cfl_kvlist *list,
+                             char *key, char *value);
+
+int cfl_kvlist_insert_bytes(struct cfl_kvlist *list,
+                             char *key, char *value,
+                             size_t value_length);
+
+int cfl_kvlist_insert_reference(struct cfl_kvlist *list,
+                                char *key, void *value);
+
+int cfl_kvlist_insert_bool(struct cfl_kvlist *list,
+                           char *key, int value);
+
+int cfl_kvlist_insert_int64(struct cfl_kvlist *list,
+                            char *key, int64_t value);
+
+int cfl_kvlist_insert_double(struct cfl_kvlist *list,
+                             char *key, double value);
+
+int cfl_kvlist_insert_array(struct cfl_kvlist *list,
+                            char *key, struct cfl_array *value);
+
+int cfl_kvlist_insert_new_array(struct cfl_kvlist *list,
+                                char *key, size_t size);
+
+int cfl_kvlist_insert_kvlist(struct cfl_kvlist *list,
+                             char *key, struct cfl_kvlist *value);
+
+int cfl_kvlist_insert(struct cfl_kvlist *list,
+                      char *key, struct cfl_variant *value);
+
+int cfl_kvlist_count(struct cfl_kvlist *list);
+struct cfl_variant *cfl_kvlist_fetch(struct cfl_kvlist *list, char *key);
+int cfl_kvlist_print(FILE *fp, struct cfl_kvlist *list);
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_list.h b/fluent-bit/lib/cfl/include/cfl/cfl_list.h
new file mode 100644
index 00000000..f2709288
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_list.h
@@ -0,0 +1,236 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*
+ * This is a rebranded copy of the original Monkey HTTP Server linked list
+ * interface (cfl_list).
+ *
+ * - http://monkey-project.com
+ * - https://github.com/monkey/monkey
+ */
+
+#ifndef CFL_LIST_H
+#define CFL_LIST_H
+
+#include <stddef.h>
+#include <stdint.h>
+
+#ifdef _WIN32
+/* Windows */
+#define cfl_container_of(address, type, field) ((type *)(                   \
+                                                        (unsigned char *)(address) - \
+                                                        (intptr_t)(&((type *)0)->field)))
+#else
+/* Rest of the world */
+#ifndef offsetof
+#define offsetof(TYPE, MEMBER) ((size_t) &((TYPE *)0)->MEMBER)
+#endif
+
+#define cfl_container_of(ptr, type, member) ({                  \
+      const typeof( ((type *)0)->member ) *__mptr = (ptr);      \
+      (type *)( (char *)__mptr - offsetof(type,member) );})
+#endif
+
+struct cfl_list {
+    struct cfl_list *prev, *next;
+};
+
+static inline int cfl_list_is_empty(struct cfl_list *head)
+{
+    if (head->next == head) {
+        return 1;
+    }
+
+    return 0;
+}
+
+static inline void cfl_list_init(struct cfl_list *list)
+{
+    list->next = list;
+    list->prev = list;
+}
+
+static inline void __cfl_list_del(struct cfl_list *prev,
+                                  struct cfl_list *next)
+{
+    prev->next = next;
+    next->prev = prev;
+}
+
+static inline void cfl_list_del(struct cfl_list *entry)
+{
+    __cfl_list_del(entry->prev, entry->next);
+
+    entry->prev = NULL;
+    entry->next = NULL;
+}
+
+static inline void __cfl_list_add(struct cfl_list *_new,
+                                  struct cfl_list *prev,
+                                  struct cfl_list *next)
+{
+    next->prev = _new;
+    _new->next = next;
+    _new->prev = prev;
+    prev->next = _new;
+}
+
+static inline void cfl_list_add(struct cfl_list *_new,
+                                struct cfl_list *head)
+{
+    __cfl_list_add(_new, head->prev, head);
+}
+
+static inline void cfl_list_add_after(struct cfl_list *_new,
+                                      struct cfl_list *prev,
+                                      struct cfl_list *head)
+{
+    struct cfl_list *next;
+
+    if (_new == NULL || prev == NULL || head == NULL) {
+        return;
+    }
+
+    next = prev->next;
+    next->prev = prev;
+    _new->next = next;
+    _new->prev = prev;
+    prev->next = _new;
+}
+
+static inline void cfl_list_add_before(struct cfl_list *_new,
+                                       struct cfl_list *next,
+                                       struct cfl_list *head)
+{
+    struct cfl_list *prev;
+
+    if (_new == NULL || next == NULL || head == NULL) {
+        return;
+    }
+
+    prev = next->prev;
+    _new->next = next;
+    _new->prev = prev;
+    prev->next = _new;
+    next->prev = _new;
+}
+
+static inline void cfl_list_append(struct cfl_list *_new,
+                                   struct cfl_list *head)
+{
+    if (cfl_list_is_empty(head)) {
+        __cfl_list_add(_new, head->prev, head);
+    }
+    else {
+        cfl_list_add_after(_new,
+                           head->prev,
+                           head);
+    }
+}
+
+static inline void cfl_list_prepend(struct cfl_list *_new,
+                                    struct cfl_list *head)
+{
+    if (cfl_list_is_empty(head)) {
+        __cfl_list_add(_new, head->prev, head);
+    }
+    else {
+        cfl_list_add_before(_new,
+                           head->next,
+                           head);
+    }
+}
+
+static inline int cfl_list_size(struct cfl_list *head)
+{
+    int ret = 0;
+    struct cfl_list *it;
+
+    for (it = head->next; it != head; it = it->next, ret++);
+
+    return ret;
+}
+
+static inline void cfl_list_entry_init(struct cfl_list *entry)
+{
+    entry->next = NULL;
+    entry->prev = NULL;
+}
+
+static inline int cfl_list_entry_is_orphan(struct cfl_list *entry)
+{
+    if (entry->next != NULL &&
+        entry->prev != NULL) {
+        return CFL_FALSE;
+    }
+
+    return CFL_TRUE;
+}
+
+static inline void cfl_list_cat(struct cfl_list *list, struct cfl_list *head)
+{
+    struct cfl_list *last;
+
+    last = head->prev;
+    last->next = list->next;
+    list->next->prev = last;
+    list->prev->next = head;
+    head->prev = list->prev;
+}
+
+#define cfl_list_foreach(curr, head) for( curr = (head)->next; curr != (head); curr = curr->next )
+#define cfl_list_foreach_safe(curr, n, head) \
+    for (curr = (head)->next, n = curr->next; curr != (head); curr = n, n = curr->next)
+
+
+#define cfl_list_foreach_r(curr, head) for( curr = (head)->prev; curr != (head); curr = curr->prev )
+#define cfl_list_foreach_safe_r(curr, n, head) \
+    for (curr = (head)->prev, n = curr->prev; curr != (head); curr = n, n = curr->prev)
+
+#define cfl_list_entry( ptr, type, member ) cfl_container_of( ptr, type, member )
+
+/*
+ * First node of the list
+ * ----------------------
+ * Be careful with this Macro, its intended to be used when some node is already linked
+ * to the list (ptr). If the list is empty it will return the list address as it points
+ * to it self: list == list->prev == list->next.
+ *
+ * If exists some possiblity that your code handle an empty list, use cfl_list_is_empty()
+ * previously to check if its empty or not.
+ */
+#define cfl_list_entry_first(ptr, type, member) cfl_container_of((ptr)->next, type, member)
+
+/* First node of the list
+ * ---------------------
+ * Be careful with this Macro, its intended to be used when some node is already linked
+ * to the list (ptr). If the list is empty it will return the list address as it points
+ * to it self: list == list->prev == list->next.
+ *
+ * If exists some possiblity that your code handle an empty list, use cfl_list_is_empty()
+ * previously to check if its empty or not.
+ */
+#define cfl_list_entry_last(ptr, type, member) cfl_container_of((ptr)->prev, type, member)
+
+/* Next node */
+#define cfl_list_entry_next(ptr, type, member, head)                     \
+    (ptr)->next == (head) ? cfl_container_of((head)->next, type, member) :  \
+        cfl_container_of((ptr)->next, type, member);
+
+#endif /* !cfl_list_H_ */
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_log.h b/fluent-bit/lib/cfl/include/cfl/cfl_log.h
new file mode 100644
index 00000000..9ceb0015
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_log.h
@@ -0,0 +1,35 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_LOG_H
+#define CFL_LOG_H
+
+#include <errno.h>
+
+int cfl_report_runtime_error_impl(int errnum, char *file, int line);
+
+#ifdef __FILENAME__
+#define cfl_report_runtime_error() cfl_report_runtime_error_impl(errno, __FILENAME__, __LINE__)
+#else
+#define cfl_report_runtime_error() cfl_report_runtime_error_impl(errno, __FILE__, __LINE__)
+#endif
+
+#define cfl_errno() do {} while (0)
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_sds.h b/fluent-bit/lib/cfl/include/cfl/cfl_sds.h
new file mode 100644
index 00000000..35054737
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_sds.h
@@ -0,0 +1,65 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_SDS_H
+#define CFL_SDS_H
+
+/*
+ * This interface is a minimized version of Fluent Bit SDS just for easily
+ * string storage
+ */
+
+#include <stdlib.h>
+#include <string.h>
+#include <inttypes.h>
+
+#define CFL_SDS_HEADER_SIZE (sizeof(uint64_t) + sizeof(uint64_t))
+
+typedef char *cfl_sds_t;
+
+#pragma pack(push, 1)
+struct cfl_sds {
+    uint64_t len;        /* used */
+    uint64_t alloc;      /* excluding the header and null terminator */
+    char buf[];
+};
+#pragma pack(pop)
+
+#define CFL_SDS_HEADER(s)  ((struct cfl_sds *) (s - CFL_SDS_HEADER_SIZE))
+
+static inline void cfl_sds_len_set(cfl_sds_t s, size_t len)
+{
+    CFL_SDS_HEADER(s)->len = len;
+}
+
+size_t cfl_sds_avail(cfl_sds_t s);
+cfl_sds_t sds_alloc(size_t size);
+size_t cfl_sds_alloc(cfl_sds_t s);
+cfl_sds_t cfl_sds_increase(cfl_sds_t s, size_t len);
+size_t cfl_sds_len(cfl_sds_t s);
+cfl_sds_t cfl_sds_create_len(const char *str, int len);
+cfl_sds_t cfl_sds_create(const char *str);
+void cfl_sds_destroy(cfl_sds_t s);
+cfl_sds_t cfl_sds_cat(cfl_sds_t s, const char *str, int len);
+cfl_sds_t cfl_sds_create_size(size_t size);
+void cfl_sds_set_len(cfl_sds_t s, size_t len);
+void cfl_sds_cat_safe(cfl_sds_t *buf, const char *str, int len);
+cfl_sds_t cfl_sds_printf(cfl_sds_t *sds, const char *fmt, ...);
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_time.h b/fluent-bit/lib/cfl/include/cfl/cfl_time.h
new file mode 100644
index 00000000..9c141c1b
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_time.h
@@ -0,0 +1,27 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_TIME_H
+#define CFL_TIME_H
+
+#include <cfl/cfl.h>
+
+uint64_t cfl_time_now();
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_variant.h b/fluent-bit/lib/cfl/include/cfl/cfl_variant.h
new file mode 100644
index 00000000..6249ffdd
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_variant.h
@@ -0,0 +1,65 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_VARIANT_H
+#define CFL_VARIANT_H
+
+#include <stdio.h>
+#include <inttypes.h>
+
+#define CFL_VARIANT_STRING    1
+#define CFL_VARIANT_BOOL      2
+#define CFL_VARIANT_INT       3
+#define CFL_VARIANT_DOUBLE    4
+#define CFL_VARIANT_ARRAY     5
+#define CFL_VARIANT_KVLIST    6
+#define CFL_VARIANT_BYTES     7
+#define CFL_VARIANT_REFERENCE 8
+
+struct cfl_array;
+struct cfl_kvlist;
+
+struct cfl_variant {
+    int type;
+
+    union {
+        cfl_sds_t as_string;
+        cfl_sds_t as_bytes;
+        unsigned int as_bool;
+        int64_t as_int64;
+        double as_double;
+        void *as_reference;
+        struct cfl_array *as_array;
+        struct cfl_kvlist *as_kvlist;
+    } data;
+};
+int cfl_variant_print(FILE *fp, struct cfl_variant *val);
+struct cfl_variant *cfl_variant_create_from_string(char *value);
+struct cfl_variant *cfl_variant_create_from_bytes(char *value, size_t length);
+struct cfl_variant *cfl_variant_create_from_bool(int value);
+struct cfl_variant *cfl_variant_create_from_int64(int64_t value);
+struct cfl_variant *cfl_variant_create_from_double(double value);
+struct cfl_variant *cfl_variant_create_from_array(struct cfl_array *value);
+struct cfl_variant *cfl_variant_create_from_kvlist(struct cfl_kvlist *value);
+struct cfl_variant *cfl_variant_create_from_reference(void *value);
+struct cfl_variant *cfl_variant_create();
+
+void cfl_variant_destroy(struct cfl_variant *instance);
+
+#endif
diff --git a/fluent-bit/lib/cfl/include/cfl/cfl_version.h.in b/fluent-bit/lib/cfl/include/cfl/cfl_version.h.in
new file mode 100644
index 00000000..9df8459a
--- /dev/null
+++ b/fluent-bit/lib/cfl/include/cfl/cfl_version.h.in
@@ -0,0 +1,36 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_VERSION_H
+#define CFL_VERSION_H
+
+/* Helpers to convert/format version string */
+#define STR_HELPER(s)      #s
+#define STR(s)             STR_HELPER(s)
+
+/* CTraces Version */
+#define CFL_VERSION_MAJOR   @CFL_VERSION_MAJOR@
+#define CFL_VERSION_MINOR   @CFL_VERSION_MINOR@
+#define CFL_VERSION_PATCH   @CFL_VERSION_PATCH@
+#define CFL_VERSION         (CFL_VERSION_MAJOR * 10000 \
+                             CFL_VERSION_MINOR * 100   \
+                             CFL_VERSION_PATCH)
+#define CFL_VERSION_STR     "@CFL_VERSION_STR@"
+
+#endif
diff --git a/fluent-bit/lib/cfl/lib/xxhash/.gitattributes b/fluent-bit/lib/cfl/lib/xxhash/.gitattributes
new file mode 100644
index 00000000..fbcf75b5
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/.gitattributes
@@ -0,0 +1,10 @@
+# Set the default behavior
+* text eol=lf
+
+# Explicitly declare source files
+*.c text eol=lf
+*.h text eol=lf
+
+# Denote files that should not be modified.
+*.odt binary
+
diff --git a/fluent-bit/lib/cfl/lib/xxhash/.github/workflows/ci.yml b/fluent-bit/lib/cfl/lib/xxhash/.github/workflows/ci.yml
new file mode 100644
index 00000000..882925f8
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/.github/workflows/ci.yml
@@ -0,0 +1,439 @@
+# Known critical issues:
+# - AVX512 related tests are incomplete.  Because default environment of
+#   GitHub Actions doesn't guarantee to support AVX512.
+#   As of May 2021, they're using Xeon E5-2673 (which doesn't support
+#   AVX512) and Xeon Platinum 8171M (which supports AVX512).
+#   See also https://github.com/actions/runner/issues/1069
+#
+#   In this CI script, it always run `make default` which compiles xxHash
+#   with AVX512 intrinsics.  But if test runner doesn't support AVX512,
+#   it doesn't run `make check` which tests runtime error/consistency.
+#   It means that this test stochastically detects a failure in AVX512
+#   code path.
+#
+# Known issues:
+# - This test script ignores exit code of cppcheck which can see under
+#   Job:Linux x64 misc tests > cppcheck in the GitHub Actions report.
+#   Because xxHash project doesn't 100% follow their recommendation.
+#   Also sometimes it reports false positives.
+#
+# - GitHub Actions doesn't support Visual Studio 2015 and 2013.
+#   https://github.com/actions/virtual-environments/issues/387
+#
+# - Setup procedure for msys2 environment is painfully slow.  It takes
+#   3..5 minutes.
+#
+# Notes:
+# - You can investigate various information at the right pane of GitHub
+#   Actions report page.
+#
+#   | Item                      | Section in the right pane             |
+#   | ------------------------- | ------------------------------------- |
+#   | OS, VM                    | Set up job                            |
+#   | git repo, commit hash     | Run actions/checkout@v2               |
+#   | gcc, tools                | Environment info                      |
+#
+# - To fail earlier, order of tests in the same job are roughly sorted by
+#   elapsed time.
+#
+# Todos:
+# - [ ] Linux: Add native ARM runner.
+# - [ ] Linux: Add native ARM64 runner.
+# - [ ] Linux: Add native PPC64LE runner.
+# - [ ] Linux: Add native S390X runner.
+# - [ ] Windows: Add VS2013.
+# - [ ] Windows: Add VS2015.
+# - [ ] Windows: Add clang for msys2.
+# - [ ] Windows: Add native or emulated ARM runner.
+# - [ ] Windows: Add native or emulated ARM64 runner.
+
+
+# Name of the workflow is also displayed as a SVG badge
+name: xxHash CI tests
+
+on: [push, pull_request]
+
+jobs:
+  xxhash-c-compilers:
+    name: CC=${{ matrix.cc }}, ${{ matrix.os }}
+    strategy:
+      fail-fast: false  # 'false' means Don't stop matrix workflows even if some matrix entry fails.
+      matrix:
+        include: [
+          # You can access the following values via ${{ matrix.??? }}
+          #
+          #   pkgs    : apt-get package names.  It can include multiple package names which are delimited by space.
+          #   cc      : C compiler executable.
+          #   cxx     : C++ compiler executable for `make ctocpptest`.
+          #   avx512  : Set 'true' if compiler supports avx512.  Otherwise, set 'false'.
+          #   os      : GitHub Actions YAML workflow label.  See https://github.com/actions/virtual-environments#available-environments
+
+          # cc
+          { pkgs: '',                                  cc: cc,        cxx: c++,         avx512: 'true',  os: ubuntu-latest, },
+
+          # gcc
+          { pkgs: '',                                  cc: gcc,       cxx: g++,         avx512: 'true',  os: ubuntu-latest, },
+          { pkgs: 'gcc-11  g++-11  lib32gcc-11-dev',   cc: gcc-11,    cxx: g++-11,      avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'gcc-10  g++-10  lib32gcc-10-dev',   cc: gcc-10,    cxx: g++-10,      avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'gcc-9   g++-9   lib32gcc-9-dev',    cc: gcc-9,     cxx: g++-9,       avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'gcc-8   g++-8   lib32gcc-8-dev',    cc: gcc-8,     cxx: g++-8,       avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'gcc-7   g++-7   lib32gcc-7-dev',    cc: gcc-7,     cxx: g++-7,       avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'gcc-6   g++-6   lib32gcc-6-dev',    cc: gcc-6,     cxx: g++-6,       avx512: 'true',  os: ubuntu-18.04,  },
+          { pkgs: 'gcc-5   g++-5   lib32gcc-5-dev',    cc: gcc-5,     cxx: g++-5,       avx512: 'true',  os: ubuntu-18.04,  },
+          { pkgs: 'gcc-4.8 g++-4.8 lib32gcc-4.8-dev ', cc: gcc-4.8,   cxx: g++-4.8,     avx512: 'false', os: ubuntu-18.04,  },
+
+          # clang
+          { pkgs: '',                                  cc: clang,     cxx: clang++,     avx512: 'true',  os: ubuntu-latest, },
+          { pkgs: 'clang-12',                          cc: clang-12,  cxx: clang++-12,  avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'clang-11',                          cc: clang-11,  cxx: clang++-11,  avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'clang-10',                          cc: clang-10,  cxx: clang++-10,  avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'clang-9',                           cc: clang-9,   cxx: clang++-9,   avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'clang-8',                           cc: clang-8,   cxx: clang++-8,   avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'clang-7',                           cc: clang-7,   cxx: clang++-7,   avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'clang-6.0',                         cc: clang-6.0, cxx: clang++-6.0, avx512: 'true',  os: ubuntu-20.04,  },
+          { pkgs: 'clang-5.0',                         cc: clang-5.0, cxx: clang++-5.0, avx512: 'true',  os: ubuntu-18.04,  },
+          { pkgs: 'clang-4.0',                         cc: clang-4.0, cxx: clang++-4.0, avx512: 'true',  os: ubuntu-18.04,  },
+          { pkgs: 'clang-3.9',                         cc: clang-3.9, cxx: clang++-3.9, avx512: 'true',  os: ubuntu-18.04,  },
+        ]
+
+    runs-on: ${{ matrix.os }}
+    env:                        # Set environment variables
+      # We globally set CC and CXX to improve compatibility with .travis.yml
+      CC: ${{ matrix.cc }}
+      CXX: ${{ matrix.cxx }}
+    steps:
+    - uses: actions/checkout@v2 # https://github.com/actions/checkout
+
+    - name: apt-get install
+      run: |
+        sudo apt-get update
+        sudo apt-get install gcc-multilib
+        sudo apt-get install ${{ matrix.pkgs }}
+
+    - name: Environment info
+      run: |
+        echo && type $CC && which $CC && $CC --version
+        echo && type $CXX && which $CXX && $CXX --version
+        echo && type make && make -v
+        echo && cat /proc/cpuinfo || echo /proc/cpuinfo is not present
+
+    - name: C90 + no-long-long compliance
+      if: always()
+      run: |
+        CFLAGS="-std=c90 -pedantic -Wno-long-long -Werror" make clean xxhsum
+
+    - name: C90 + XXH_NO_LONG_LONG
+      if: always()
+      run: |
+        # strict c90, with no long long support; resulting in no XXH64_* symbol
+        make clean c90test
+
+    - name: dispatch
+      if: always()
+      run: |
+        # removing sign conversion warnings due to a bug in gcc-5's definition of some AVX512 intrinsics
+        CFLAGS="-Werror" MOREFLAGS="-Wno-sign-conversion" make clean dispatch
+
+    - name: DISPATCH=1
+      if: always()
+      run: |
+        CFLAGS="-Wall -Wextra -Werror" make DISPATCH=1 clean default
+
+    - name: noxxh3test
+      if: always()
+      run: |
+        # check library can be compiled with XXH_NO_XXH3, resulting in no XXH3_* symbol
+        make clean noxxh3test
+
+    - name: make avx512f
+      if: ${{ matrix.avx512 == 'true' }}
+      run: |
+        CFLAGS="-O1 -mavx512f -Werror" make clean default
+
+    - name: test-all
+      if: always()
+      run: |
+        make clean test-all
+
+
+  ubuntu-consistency:
+    name: Linux x64 check results consistency
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2 # https://github.com/actions/checkout
+
+    - name: Environment info
+      run: |
+        echo && gcc --version
+        echo && make -v
+        echo && cat /proc/cpuinfo || echo /proc/cpuinfo is not present
+
+    - name: Scalar code path
+      run: |
+        CPPFLAGS=-DXXH_VECTOR=XXH_SCALAR make clean check
+
+    - name: SSE2 code path
+      run: |
+        CPPFLAGS=-DXXH_VECTOR=XXH_SSE2 make clean check
+
+    - name: AVX2 code path
+      run: |
+        CPPFLAGS="-mavx2 -DXXH_VECTOR=XXH_AVX2" make clean check
+
+    # As for AVX512, see "Known critical issues" at the top of this file
+    - name: AVX512 code path
+      run: |
+        # Run "make check" if /proc/cpuinfo has flags for avx512.
+        grep -q "^flags.*\bavx512\b" /proc/cpuinfo && CPPFLAGS="-mavx512f -DXXH_VECTOR=XXH_AVX512" make clean check || (echo This test runner does not support AVX512. && $(exit 0))
+
+    - name: reroll code path (#240)
+      run: |
+        CPPFLAGS=-DXXH_REROLL=1 make clean check
+
+    - name: tests/bench
+      run: |
+        make -C tests/bench
+
+
+  ubuntu-misc:
+    name: Linux x64 misc tests
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2 # https://github.com/actions/checkout
+
+    - name: apt-get install
+      run: |
+        sudo apt-get install valgrind cppcheck
+
+    - name: Environment info
+      run: |
+        echo && gcc --version
+        echo && clang --version
+        echo && valgrind --version
+        echo && cppcheck --version
+        echo && make -v
+        echo && cat /proc/cpuinfo || echo /proc/cpuinfo is not present
+
+    - name: cppcheck
+      run: |
+        # This test script ignores exit code of cppcheck.  See knowin issues
+        # at the top of this file.
+        make clean cppcheck || echo There are some cppcheck reports
+
+    - name: test-mem (valgrind)
+      run: |
+        make clean test-mem
+
+    - name: usan
+      run: |
+        make clean usan
+
+    - name: Lint Unicode in root-dir, cli/, tests/, tests/bench/, tests/collisions/.
+      run: |
+        make lint-unicode
+
+
+  ubuntu-cmake-unofficial:
+    name: Linux x64 cmake unofficial build test
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v2 # https://github.com/actions/checkout
+
+    - name: Environment info
+      run: |
+        echo && gcc --version
+        echo && cmake --version
+        echo && make -v
+        echo && cat /proc/cpuinfo || echo /proc/cpuinfo is not present
+
+    - name: cmake
+      run: |
+        cd cmake_unofficial
+        mkdir build
+        cd build
+        cmake ..
+        CFLAGS=-Werror make
+
+
+  # Linux, { ARM, ARM64, PPC64LE, PPC64, S390X }
+  # All tests are using QEMU and gcc cross compiler.
+
+  qemu-consistency:
+    name: QEMU ${{ matrix.name }}
+    runs-on: ubuntu-latest
+    strategy:
+      fail-fast: false  # 'false' means Don't stop matrix workflows even if some matrix failed.
+      matrix:
+        include: [
+          { name: ARM,      xcc_pkg: gcc-arm-linux-gnueabi,     xcc: arm-linux-gnueabi-gcc,     xemu_pkg: qemu-system-arm,    xemu: qemu-arm-static     },
+          { name: ARM64,    xcc_pkg: gcc-aarch64-linux-gnu,     xcc: aarch64-linux-gnu-gcc,     xemu_pkg: qemu-system-arm,    xemu: qemu-aarch64-static },
+          { name: PPC64LE-gcc9,  xcc_pkg: gcc-9-powerpc64le-linux-gnu,  xcc: powerpc64le-linux-gnu-gcc-9,  xemu_pkg: qemu-system-ppc, xemu: qemu-ppc64le-static },
+          { name: PPC64LE-gcc10, xcc_pkg: gcc-10-powerpc64le-linux-gnu, xcc: powerpc64le-linux-gnu-gcc-10, xemu_pkg: qemu-system-ppc, xemu: qemu-ppc64le-static },
+          { name: PPC64-gcc9,    xcc_pkg: gcc-9-powerpc64-linux-gnu,    xcc: powerpc64-linux-gnu-gcc-9,    xemu_pkg: qemu-system-ppc, xemu: qemu-ppc64-static },
+          { name: PPC64-gcc10,   xcc_pkg: gcc-10-powerpc64-linux-gnu,   xcc: powerpc64-linux-gnu-gcc-10,   xemu_pkg: qemu-system-ppc, xemu: qemu-ppc64-static },
+          { name: S390X,    xcc_pkg: gcc-s390x-linux-gnu,       xcc: s390x-linux-gnu-gcc,       xemu_pkg: qemu-system-s390x,  xemu: qemu-s390x-static   },
+          { name: MIPS,     xcc_pkg: gcc-mips-linux-gnu,        xcc: mips-linux-gnu-gcc,        xemu_pkg: qemu-system-mips,   xemu: qemu-mips-static    },
+        ]
+    env:                        # Set environment variables
+      XCC: ${{ matrix.xcc }}
+      XEMU: ${{ matrix.xemu }}
+    steps:
+    - uses: actions/checkout@v2 # https://github.com/actions/checkout
+    - name: apt update & install
+      run: |
+        sudo apt-get update
+        sudo apt-get install gcc-multilib g++-multilib qemu-utils qemu-user-static
+        sudo apt-get install ${{ matrix.xcc_pkg }} ${{ matrix.xemu_pkg }}
+
+    - name: Environment info
+      run: |
+        echo && which $XCC
+        echo && $XCC --version
+        echo && $XCC -v  # Show built-in specs
+        echo && which $XEMU
+        echo && $XEMU --version
+
+    - name: ARM (XXH_VECTOR=[ scalar, NEON ])
+      if: ${{ matrix.name == 'ARM' }}
+      run: |
+        CPPFLAGS="-DXXH_VECTOR=XXH_SCALAR" LDFLAGS="-static" CC=$XCC RUN_ENV=$XEMU make clean check
+        CPPFLAGS="-DXXH_VECTOR=XXH_NEON" CFLAGS="-O3 -march=armv7-a -fPIC -mfloat-abi=softfp -mfpu=neon-vfpv4" LDFLAGS="-static" CC=$XCC RUN_ENV=$XEMU make clean check
+
+    - name: ARM64 (XXH_VECTOR=[ scalar, NEON ])
+      if: ${{ matrix.name == 'ARM64' }}
+      run: |
+        CPPFLAGS="-DXXH_VECTOR=XXH_SCALAR" LDFLAGS="-static" CC=$XCC RUN_ENV=$XEMU make clean check
+        CPPFLAGS="-DXXH_VECTOR=XXH_NEON" LDFLAGS="-static" CC=$XCC RUN_ENV=$XEMU make clean check
+
+    - name: ${{ matrix.name }} (XXH_VECTOR=[ scalar, VSX ])
+      if: ${{ startsWith(matrix.name, 'PPC64') }}
+      run: |
+        CPPFLAGS="-DXXH_VECTOR=XXH_SCALAR" LDFLAGS="-static" CC=$XCC RUN_ENV=$XEMU make clean check
+        CPPFLAGS="-DXXH_VECTOR=XXH_VSX" CFLAGS="-O3 -maltivec -mvsx -mpower8-vector -mcpu=power8" LDFLAGS="-static" CC=$XCC RUN_ENV=$XEMU make clean check
+
+    - name: S390X (XXH_VECTOR=[ scalar, VSX ])
+      if: ${{ matrix.name == 'S390X' }}
+      run: |
+        CPPFLAGS="-DXXH_VECTOR=XXH_SCALAR" LDFLAGS="-static" CC=$XCC RUN_ENV=$XEMU make clean check
+        CPPFLAGS=-DXXH_VECTOR=XXH_VSX CFLAGS="-O3 -march=arch11 -mzvector" LDFLAGS="-static" CC=$XCC RUN_ENV=$XEMU make clean check
+
+    - name: MIPS (XXH_VECTOR=[ scalar ])
+      if: ${{ matrix.name == 'MIPS' }}
+      run: |
+        LDFLAGS="-static" CC=$XCC RUN_ENV=$XEMU make clean check
+
+  # macOS
+
+  macos-latest-general:
+    name: macOS general test
+    runs-on: macos-latest
+    steps:
+    - uses: actions/checkout@v2
+
+    - name: Environment info
+      run: |
+        echo && clang --version
+        echo && sysctl -a | grep machdep.cpu   # cpuinfo
+
+    - name: make
+      run: |
+        CFLAGS="-Werror" make clean default
+
+    - name: make test
+      run: |
+        # test scenario where "stdout" is not the console
+        make clean test MOREFLAGS='-Werror' | tee
+
+
+  # Windows, { VC++2019, VC++2017 } x { x64, Win32, ARM, ARM64 }
+  #
+  # - Default shell for Windows environment is PowerShell Core.
+  #   https://docs.github.com/en/actions/reference/workflow-syntax-for-github-actions#using-a-specific-shell
+  #
+  # - "windows-2019" uses Visual Studio 2019.
+  #   https://github.com/actions/virtual-environments/blob/main/images/win/Windows2019-Readme.md#visual-studio-enterprise-2019
+  #
+  # - "windows-2016" uses Visual Studio 2017.
+  #   https://github.com/actions/virtual-environments/blob/main/images/win/Windows2016-Readme.md#visual-studio-enterprise-2017
+
+  windows-visualc-general:
+    name: ${{ matrix.system.vc }}, ${{ matrix.arch }}
+    runs-on: ${{ matrix.system.os }}   # Runs-on foreach value of strategy.matrix.system.os
+    strategy:
+      fail-fast: false  # 'false' means Don't stop matrix workflows even if some matrix failed.
+      matrix:
+        system: [
+          { os: windows-2019, vc: "VC++ 2019" },
+          { os: windows-2016, vc: "VC++ 2017" },
+        ]
+        arch: [ x64, Win32, ARM, ARM64 ]
+    steps:
+    - uses: actions/checkout@v2
+
+    - name: Build ${{ matrix.system.os }}, ${{ matrix.arch }}
+      run: |
+        cd cmake_unofficial
+        mkdir build
+        cd build
+        cmake .. -DCMAKE_BUILD_TYPE=Release -A ${{ matrix.arch }} -DXXHASH_C_FLAGS="/WX"
+        cmake --build . --config Release
+
+    - name: Test
+      # Run benchmark for testing only if target arch is x64 or Win32.
+      if: ${{ matrix.arch == 'x64' || matrix.arch == 'Win32' }}
+      run: |
+        .\cmake_unofficial\build\Release\xxhsum.exe -bi1
+
+
+  # Windows, { mingw64, mingw32 }
+  #
+  # - Shell for msys2 is sh (msys2).  defaults.run.shell is for this setting.
+  #
+  # https://github.com/msys2/MINGW-packages/blob/master/.github/workflows/main.yml
+  # https://github.com/actions/starter-workflows/issues/95
+
+  windows-msys2-general:
+    name: Windows ${{ matrix.msystem }}
+    runs-on: windows-latest
+    strategy:
+      fail-fast: false  # 'false' means Don't stop matrix workflows even if some matrix failed.
+      matrix:
+        include: [
+          { msystem: mingw64, toolchain: mingw-w64-x86_64-toolchain },
+          { msystem: mingw32, toolchain: mingw-w64-i686-toolchain },
+        ]
+    defaults:
+      run:
+        shell: msys2 {0}
+    steps:
+      - uses: actions/checkout@v2  # https://github.com/actions/checkout
+      - uses: msys2/setup-msys2@v2 # https://github.com/msys2/setup-msys2
+        with:
+          msystem: MSYS
+          install: mingw-w64-i686-make ${{ matrix.toolchain }}
+          update: true
+
+      - name: Update
+        run: |
+          pacman --noconfirm -Suuy
+          pacman --noconfirm -Suu
+
+      - name: mingw64
+        if: ${{ matrix.msystem == 'mingw64' }}
+        run: |
+          PATH=/mingw64/bin:$PATH /mingw32/bin/mingw32-make clean test MOREFLAGS=-Werror
+          PATH=/mingw64/bin:$PATH /mingw32/bin/mingw32-make -C tests/bench
+          # Abort if result of "file ./xxhsum.exe" doesn't contain 'x86-64'.
+          # Expected output is "./xxhsum.exe: PE32+ executable (console) x86-64, for MS Windows"
+          file ./xxhsum.exe | grep -q 'x86-64' || $(exit 1)
+          ./xxhsum.exe --version
+
+      - name: mingw32
+        if: ${{ matrix.msystem == 'mingw32' }}
+        run: |
+          PATH=/mingw32/bin:$PATH /mingw32/bin/mingw32-make.exe clean test MOREFLAGS=-Werror
+          PATH=/mingw32/bin:$PATH /mingw32/bin/mingw32-make.exe -C tests/bench
+          # Abort if result of "file ./xxhsum.exe" doesn't contain '80386'.
+          # Expected output is "./xxhsum.exe: PE32 executable (console) Intel 80386, for MS Windows"
+          file ./xxhsum.exe | grep -q '80386' || $(exit 1)
+          ./xxhsum.exe --version
diff --git a/fluent-bit/lib/cfl/lib/xxhash/.gitignore b/fluent-bit/lib/cfl/lib/xxhash/.gitignore
new file mode 100644
index 00000000..e8645286
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/.gitignore
@@ -0,0 +1,47 @@
+# objects
+*.o
+*.obj
+*.s
+
+# libraries
+libxxhash.*
+!libxxhash.pc.in
+
+# Executables
+*.exe
+xxh32sum
+xxh64sum
+xxh128sum
+xxhsum
+xxhsum32
+xxhsum_privateXXH
+xxhsum_inlinedXXH
+dispatch
+tests/generate_unicode_test
+
+# local conf
+.clang_complete
+
+# Mac OS-X artefacts
+*.dSYM
+.DS_Store
+
+# Wasm / emcc / emscripten artefacts
+*.html
+*.wasm
+*.js
+
+# CMake build directories
+build*/
+
+# project managers artifacts
+.projectile
+
+# analyzer artifacts
+infer-out
+
+# test artifacts
+.test*
+tmp*
+tests/*.unicode
+tests/unicode_test*
diff --git a/fluent-bit/lib/cfl/lib/xxhash/.travis.yml b/fluent-bit/lib/cfl/lib/xxhash/.travis.yml
new file mode 100644
index 00000000..9f9e42ca
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/.travis.yml
@@ -0,0 +1,142 @@
+language: c
+
+# Dump CPU info before start
+before_install:
+  - cat /proc/cpuinfo || echo /proc/cpuinfo is not present
+
+matrix:
+  fast_finish: true
+  include:
+
+    - name: General linux x64 tests
+      arch: amd64
+      addons:
+        apt:
+          packages:
+            - g++-multilib
+            - gcc-multilib
+            - cppcheck
+      script:
+        - make -B test-all
+        - make clean
+        - CFLAGS="-Werror" MOREFLAGS="-Wno-sign-conversion" make dispatch  # removing sign conversion warnings due to a bug in gcc-5's definition of some AVX512 intrinsics
+        - make clean
+        - CFLAGS="-O1 -mavx512f -Werror" make
+        - make clean
+        - CFLAGS="-Wall -Wextra -Werror" make DISPATCH=1
+        - make clean
+        - CFLAGS="-std=c90 -pedantic -Wno-long-long -Werror" make xxhsum  # check C90 + long long compliance
+        - make c90test # strict c90, with no long long support; resulting in no XXH64_* symbol
+        - make noxxh3test # check library can be compiled with XXH_NO_XXH3, resulting in no XXH3_* symbol
+
+
+    - name: Check results consistency on x64
+      arch: amd64
+      script:
+        - CPPFLAGS=-DXXH_VECTOR=XXH_SCALAR make check   # Scalar code path
+        - make clean
+        - CPPFLAGS=-DXXH_VECTOR=XXH_SSE2 make check   # SSE2 code path
+        - make clean
+        - CPPFLAGS="-mavx2 -DXXH_VECTOR=XXH_AVX2" make check   # AVX2 code path
+        - make clean
+        - CPPFLAGS="-mavx512f -DXXH_VECTOR=XXH_AVX512" make check   # AVX512 code path
+        - make clean
+        - CPPFLAGS=-DXXH_REROLL=1 make check   # reroll code path (#240)
+        - make -C tests/bench
+
+    - name: macOS General Test
+      os: osx
+      compiler: clang
+      script:
+        - CFLAGS="-Werror" make   # test library build
+        - make clean
+        - make test MOREFLAGS='-Werror' | tee # test scenario where `stdout` is not the console
+
+    - name: ARM compilation and consistency checks (Qemu)
+      dist: xenial
+      arch: amd64
+      addons:
+        apt:
+          packages:
+            - qemu-system-arm
+            - qemu-user-static
+            - gcc-arm-linux-gnueabi
+            - libc6-dev-armel-cross
+      script:
+        # arm (32-bit)
+        - CC=arm-linux-gnueabi-gcc CPPFLAGS=-DXXH_VECTOR=XXH_SCALAR LDFLAGS=-static RUN_ENV=qemu-arm-static make check   # Scalar code path
+        - make clean
+        # NEON (32-bit)
+        - CC=arm-linux-gnueabi-gcc CPPFLAGS=-DXXH_VECTOR=XXH_NEON CFLAGS="-O3 -march=armv7-a -fPIC -mfloat-abi=softfp -mfpu=neon-vfpv4" LDFLAGS=-static RUN_ENV=qemu-arm-static make check   # NEON code path
+
+    - name: aarch64 compilation and consistency checks
+      dist: xenial
+      arch: arm64
+      script:
+        # aarch64
+        - CPPFLAGS=-DXXH_VECTOR=XXH_SCALAR make check   # Scalar code path
+        # NEON (64-bit)
+        - make clean
+        - CPPFLAGS=-DXXH_VECTOR=XXH_NEON make check   # NEON code path
+        # clang
+        - make clean
+        - CC=clang CPPFLAGS=-DXXH_VECTOR=XXH_SCALAR make check   # Scalar code path
+        # clang + NEON
+        - make clean
+        - CC=clang CPPFLAGS=-DXXH_VECTOR=XXH_NEON make check   # NEON code path
+
+    # We need Bionic here because the QEMU versions shipped in the older repos
+    # do not support POWER8 emulation, and compiling QEMU from source is a pain.
+    - name: PowerPC + PPC64 compilation and consistency checks (Qemu on Bionic)
+      dist: bionic
+      arch: amd64
+      addons:
+        apt:
+          packages:
+            - qemu-system-ppc
+            - qemu-user-static
+            - gcc-powerpc-linux-gnu
+            - gcc-powerpc64-linux-gnu
+            - libc6-dev-powerpc-cross
+            - libc6-dev-ppc64-cross
+      script:
+        - CC=powerpc-linux-gnu-gcc RUN_ENV=qemu-ppc-static LDFLAGS=-static make check   # Scalar code path
+        - make clean
+        - CC=powerpc64-linux-gnu-gcc RUN_ENV=qemu-ppc64-static CPPFLAGS=-DXXH_VECTOR=XXH_SCALAR CFLAGS="-O3" LDFLAGS="-static -m64" make check   # Scalar code path
+        # VSX code
+        - make clean
+        - CC=powerpc64-linux-gnu-gcc RUN_ENV="qemu-ppc64-static -cpu power8" CPPFLAGS=-DXXH_VECTOR=XXH_VSX CFLAGS="-O3 -maltivec -mvsx -mcpu=power8 -mpower8-vector" LDFLAGS="-static -m64" make check   # VSX code path
+        # altivec.h redefinition issue #426
+        - make clean
+        - CC=powerpc64-linux-gnu-gcc CPPFLAGS=-DXXH_VECTOR=XXH_VSX CFLAGS="-maltivec -mvsx -mcpu=power8 -mpower8-vector" make -C tests test_ppc_redefine
+
+    - name: PPC64LE compilation and consistency checks
+      dist: xenial
+      arch: ppc64le
+      script:
+        # Scalar (universal) code path
+        - CPPFLAGS=-DXXH_VECTOR=XXH_SCALAR LDFLAGS=-static make check
+        # VSX code path (64-bit)
+        - make clean
+        - CPPFLAGS=-DXXH_VECTOR=XXH_VSX CFLAGS="-O3 -maltivec -mvsx -mpower8-vector -mcpu=power8" LDFLAGS="-static" make check
+        # altivec.h redefinition issue #426
+        - make clean
+        - CPPFLAGS=-DXXH_VECTOR=XXH_VSX CFLAGS="-maltivec -mvsx -mcpu=power8 -mpower8-vector" make -C tests test_ppc_redefine
+
+    - name: IBM s390x compilation and consistency checks
+      dist: bionic
+      arch: s390x
+      script:
+        # Scalar (universal) code path
+        - CPPFLAGS=-DXXH_VECTOR=XXH_SCALAR LDFLAGS=-static make check
+        # s390x code path (64-bit)
+        - make clean
+        - CPPFLAGS=-DXXH_VECTOR=XXH_VSX CFLAGS="-O3 -march=arch11 -mzvector" LDFLAGS="-static" make check
+
+    - name: cmake build test
+      script:
+        - cd cmake_unofficial
+        - mkdir build
+        - cd build
+        - cmake ..
+        - CFLAGS=-Werror make
diff --git a/fluent-bit/lib/cfl/lib/xxhash/CHANGELOG b/fluent-bit/lib/cfl/lib/xxhash/CHANGELOG
new file mode 100644
index 00000000..ff59d8bb
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/CHANGELOG
@@ -0,0 +1,71 @@
+v0.8.1
+- perf : much improved performance for XXH3 streaming variants, notably on gcc and msvc
+- perf : improved XXH64 speed and latency on small inputs
+- perf : small XXH32 speed and latency improvement on small inputs of random size
+- perf : minor stack usage improvement for XXH32 and XXH64
+- api  : new experimental variants XXH3_*_withSecretandSeed()
+- api  : update XXH3_generateSecret(), can no generate secret of any size (>= XXH3_SECRET_SIZE_MIN)
+- cli  : xxhsum can now generate and check XXH3 checksums, using command `-H3`
+- build: can build xxhash without XXH3, with new build macro XXH_NO_XXH3
+- build: fix xxh_x86dispatch build with MSVC, by @apankrat
+- build: XXH_INLINE_ALL can always be used safely, even after XXH_NAMESPACE or a previous XXH_INLINE_ALL
+- build: improved PPC64LE vector support, by @mpe
+- install: fix pkgconfig, by @ellert
+- install: compatibility with Haiku, by @Begasus
+- doc  : code comments made compatible with doxygen, by @easyaspi314
+- misc : XXH_ACCEPT_NULL_INPUT_POINTER is no longer necessary, all functions can accept NULL input pointers, as long as size == 0
+- misc : complete refactor of CI tests on Github Actions, offering much larger coverage, by @t-mat
+- misc : xxhsum code base split into multiple specialized units, within directory cli/, by @easyaspi314
+
+v0.8.0
+- api : stabilize XXH3
+- cli : xxhsum can parse BSD-style --check lines, by @WayneD
+- cli : `xxhsum -` accepts console input, requested by @jaki
+- cli : xxhsum accepts -- separator, by @jaki
+- cli : fix : print correct default algo for symlinked helpers, by @martinetd
+- install: improved pkgconfig script, allowing custom install locations, requested by @ellert
+
+v0.7.4
+- perf: automatic vector detection and selection at runtime (`xxh_x86dispatch.h`), initiated by @easyaspi314
+- perf: added AVX512 support, by @gzm55
+- api : new: secret generator `XXH_generateSecret()`, suggested by @koraa
+- api : fix: XXH3_state_t is movable, identified by @koraa
+- api : fix: state is correctly aligned in AVX mode (unlike `malloc()`), by @easyaspi314
+- api : fix: streaming generated wrong values in some combination of random ingestion lengths, reported by @WayneD
+- cli : fix unicode print on Windows, by @easyaspi314
+- cli : can `-c` check file generated by sfv
+- build: `make DISPATCH=1` generates `xxhsum` and `libxxhash` with runtime vector detection (x86/x64 only)
+- install: cygwin installation support
+- doc : Cryptol specification of XXH32 and XXH64, by @weaversa
+
+v0.7.3
+- perf: improved speed for large inputs (~+20%)
+- perf: improved latency for small inputs (~10%)
+- perf: s390x Vectorial code, by @easyaspi314
+- cli: improved support for Unicode filenames on Windows, thanks to @easyaspi314 and @t-mat
+- api: `xxhash.h` can now be included in any order, with and without `XXH_STATIC_LINKING_ONLY` and `XXH_INLINE_ALL`
+- build: xxHash's implementation transferred into `xxhash.h`. No more need to have `xxhash.c` in the `/include` directory for `XXH_INLINE_ALL` to work
+- install: created pkg-config file, by @bket
+- install: VCpkg installation instructions, by @LilyWangL
+- doc: Highly improved code documentation, by @easyaspi314
+- misc: New test tool in `/tests/collisions`: brute force collision tester for 64-bit hashes
+
+v0.7.2
+- Fixed collision ratio of `XXH128` for some specific input lengths, reported by @svpv
+- Improved `VSX` and `NEON` variants, by @easyaspi314
+- Improved performance of scalar code path (`XXH_VECTOR=0`), by @easyaspi314
+- `xxhsum`: can generate 128-bit hashes with the `-H2` option (note: for experimental purposes only! `XXH128` is not yet frozen)
+- `xxhsum`: option `-q` removes status notifications
+
+v0.7.1
+- Secret first: the algorithm computation can be altered by providing a "secret", which is any blob of bytes, of size >= `XXH3_SECRET_SIZE_MIN`.
+- `seed` is still available, and acts as a secret generator
+- updated `ARM NEON` variant by @easyaspi314
+- Streaming implementation is available
+- Improve compatibility and performance with Visual Studio, with help from @aras-p
+- Better integration when using `XXH_INLINE_ALL`: do not pollute host namespace, use its own macros, such as `XXH_ASSERT()`, `XXH_ALIGN`, etc.
+- 128-bit variant provides helper functions for comparison of hashes.
+- Better `clang` generation of `rotl` instruction, thanks to @easyaspi314
+- `XXH_REROLL` build macro to reduce binary size, by @easyaspi314
+- Improved `cmake` script, by @Mezozoysky
+- Full benchmark program provided in `/tests/bench`
diff --git a/fluent-bit/lib/cfl/lib/xxhash/Doxyfile b/fluent-bit/lib/cfl/lib/xxhash/Doxyfile
new file mode 100644
index 00000000..634e1e1a
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/Doxyfile
@@ -0,0 +1,58 @@
+# Doxygen config for xxHash
+DOXYFILE_ENCODING      = UTF-8
+
+PROJECT_NAME           = "xxHash"
+PROJECT_NUMBER         = "0.8.0"
+PROJECT_BRIEF          = "Extremely fast non-cryptographic hash function"
+OUTPUT_DIRECTORY       = doxygen
+OUTPUT_LANGUAGE        = English
+
+# We already separate the internal docs.
+INTERNAL_DOCS          = YES
+# Consistency
+SORT_MEMBER_DOCS       = NO
+BRIEF_MEMBER_DESC      = YES
+REPEAT_BRIEF           = YES
+
+# Warnings
+QUIET                  = YES
+# Until we document everything
+WARN_IF_UNDOCUMENTED   = NO
+
+# TODO: Add the other files. It is just xxhash.h for now.
+FILE_PATTERNS          = xxhash.h xxh_x86dispatch.c
+# Note: xxHash's source files are technically ASCII only.
+INPUT_ENCODING         = UTF-8
+TAB_SIZE               = 4
+MARKDOWN_SUPPORT       = YES
+
+# xxHash is a C library
+OPTIMIZE_OUTPUT_FOR_C  = YES
+# So we can document the internals
+EXTRACT_STATIC         = YES
+# Document the macros
+MACRO_EXPANSION        = YES
+EXPAND_ONLY_PREDEF     = YES
+# Predefine some macros to clean up the output.
+PREDEFINED             = "XXH_DOXYGEN=" \
+                         "XXH_PUBLIC_API=" \
+                         "XXH_FORCE_INLINE=static inline" \
+                         "XXH_NO_INLINE=static" \
+                         "XXH_RESTRICT=restrict" \
+                         "XSUM_API=" \
+                         "XXH_STATIC_LINKING_ONLY" \
+                         "XXH_IMPLEMENTATION" \
+                         "XXH_ALIGN(N)=alignas(N)" \
+                         "XXH_ALIGN_MEMBER(align,type)=alignas(align) type"
+
+# We want HTML docs
+GENERATE_HTML          = YES
+HTML_OUTPUT            = html
+HTML_FILE_EXTENSION    = .html
+# Tweak the colors a bit
+HTML_COLORSTYLE_HUE    = 220
+HTML_COLORSTYLE_GAMMA  = 100
+HTML_COLORSTYLE_SAT    = 100
+
+# We don't want LaTeX.
+GENERATE_LATEX         = NO
diff --git a/fluent-bit/lib/cfl/lib/xxhash/LICENSE b/fluent-bit/lib/cfl/lib/xxhash/LICENSE
new file mode 100644
index 00000000..6bc30a1b
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/LICENSE
@@ -0,0 +1,26 @@
+xxHash Library
+Copyright (c) 2012-2020 Yann Collet
+All rights reserved.
+
+BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
+
+Redistribution and use in source and binary forms, with or without modification,
+are permitted provided that the following conditions are met:
+
+* Redistributions of source code must retain the above copyright notice, this
+  list of conditions and the following disclaimer.
+
+* Redistributions in binary form must reproduce the above copyright notice, this
+  list of conditions and the following disclaimer in the documentation and/or
+  other materials provided with the distribution.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR
+ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON
+ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
diff --git a/fluent-bit/lib/cfl/lib/xxhash/README.md b/fluent-bit/lib/cfl/lib/xxhash/README.md
new file mode 100644
index 00000000..2406c8d2
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/README.md
@@ -0,0 +1,236 @@
+
+xxHash - Extremely fast hash algorithm
+======================================
+
+xxHash is an Extremely fast Hash algorithm, running at RAM speed limits.
+It successfully completes the [SMHasher](https://code.google.com/p/smhasher/wiki/SMHasher) test suite
+which evaluates collision, dispersion and randomness qualities of hash functions.
+Code is highly portable, and hashes are identical across all platforms (little / big endian).
+
+|Branch      |Status   |
+|------------|---------|
+|dev         | [![Build Status](https://github.com/Cyan4973/xxHash/actions/workflows/ci.yml/badge.svg?branch=dev)](https://github.com/Cyan4973/xxHash/actions?query=branch%3Adev+) |
+
+
+Benchmarks
+-------------------------
+
+The reference system uses an Intel i7-9700K cpu, and runs Ubuntu x64 20.04.
+The [open source benchmark program] is compiled with `clang` v10.0 using `-O3` flag.
+
+| Hash Name     | Width | Bandwidth (GB/s) | Small Data Velocity | Quality | Comment |
+| ---------     | ----- | ---------------- | ----- | --- | --- |
+| __XXH3__ (SSE2) |  64 | 31.5 GB/s        | 133.1 | 10
+| __XXH128__ (SSE2) | 128 | 29.6 GB/s      | 118.1 | 10
+| _RAM sequential read_ | N/A | 28.0 GB/s  |   N/A | N/A | _for reference_
+| City64        |    64 | 22.0 GB/s        |  76.6 | 10
+| T1ha2         |    64 | 22.0 GB/s        |  99.0 |  9 | Slightly worse [collisions]
+| City128       |   128 | 21.7 GB/s        |  57.7 | 10
+| __XXH64__     |    64 | 19.4 GB/s        |  71.0 | 10
+| SpookyHash    |    64 | 19.3 GB/s        |  53.2 | 10
+| Mum           |    64 | 18.0 GB/s        |  67.0 |  9 | Slightly worse [collisions]
+| __XXH32__     |    32 |  9.7 GB/s        |  71.9 | 10
+| City32        |    32 |  9.1 GB/s        |  66.0 | 10
+| Murmur3       |    32 |  3.9 GB/s        |  56.1 | 10
+| SipHash       |    64 |  3.0 GB/s        |  43.2 | 10
+| FNV64         |    64 |  1.2 GB/s        |  62.7 |  5 | Poor avalanche properties
+| Blake2        |   256 |  1.1 GB/s        |   5.1 | 10 | Cryptographic
+| SHA1          |   160 |  0.8 GB/s        |   5.6 | 10 | Cryptographic but broken
+| MD5           |   128 |  0.6 GB/s        |   7.8 | 10 | Cryptographic but broken
+
+[open source benchmark program]: https://github.com/Cyan4973/xxHash/tree/release/tests/bench
+[collisions]: https://github.com/Cyan4973/xxHash/wiki/Collision-ratio-comparison#collision-study
+
+note 1: Small data velocity is a _rough_ evaluation of algorithm's efficiency on small data. For more detailed analysis, please refer to next paragraph.
+
+note 2: some algorithms feature _faster than RAM_ speed. In which case, they can only reach their full speed when input data is already in CPU cache (L3 or better). Otherwise, they max out on RAM speed limit.
+
+### Small data
+
+Performance on large data is only one part of the picture.
+Hashing is also very useful in constructions like hash tables and bloom filters.
+In these use cases, it's frequent to hash a lot of small data (starting at a few bytes).
+Algorithm's performance can be very different for such scenarios, since parts of the algorithm,
+such as initialization or finalization, become fixed cost.
+The impact of branch mis-prediction also becomes much more present.
+
+XXH3 has been designed for excellent performance on both long and small inputs,
+which can be observed in the following graph:
+
+![XXH3, latency, random size](https://user-images.githubusercontent.com/750081/61976089-aedeab00-af9f-11e9-9239-e5375d6c080f.png)
+
+For a more detailed analysis, visit the wiki :
+https://github.com/Cyan4973/xxHash/wiki/Performance-comparison#benchmarks-concentrating-on-small-data-
+
+Quality
+-------------------------
+
+Speed is not the only property that matters.
+Produced hash values must respect excellent dispersion and randomness properties,
+so that any sub-section of it can be used to maximally spread out a table or index,
+as well as reduce the amount of collisions to the minimal theoretical level, following the [birthday paradox].
+
+`xxHash` has been tested with Austin Appleby's excellent SMHasher test suite,
+and passes all tests, ensuring reasonable quality levels.
+It also passes extended tests from [newer forks of SMHasher], featuring additional scenarios and conditions.
+
+Finally, xxHash provides its own [massive collision tester](https://github.com/Cyan4973/xxHash/tree/dev/tests/collisions),
+able to generate and compare billions of hashes to test the limits of 64-bit hash algorithms.
+On this front too, xxHash features good results, in line with the [birthday paradox].
+A more detailed analysis is documented [in the wiki](https://github.com/Cyan4973/xxHash/wiki/Collision-ratio-comparison).
+
+[birthday paradox]: https://en.wikipedia.org/wiki/Birthday_problem
+[newer forks of SMHasher]: https://github.com/rurban/smhasher
+
+
+### Build modifiers
+
+The following macros can be set at compilation time to modify libxxhash's behavior. They are generally disabled by default.
+
+- `XXH_INLINE_ALL`: Make all functions `inline`, with implementations being directly included within `xxhash.h`.
+                    Inlining functions is beneficial for speed on small keys.
+                    It's _extremely effective_ when key length is expressed as _a compile time constant_,
+                    with performance improvements observed in the +200% range .
+                    See [this article](https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html) for details.
+- `XXH_PRIVATE_API`: same outcome as `XXH_INLINE_ALL`. Still available for legacy support.
+                     The name underlines that `XXH_*` symbols will not be exported.
+- `XXH_NAMESPACE`: Prefixes all symbols with the value of `XXH_NAMESPACE`.
+                   This macro can only use compilable character set.
+                   Useful to evade symbol naming collisions,
+                   in case of multiple inclusions of xxHash's source code.
+                   Client applications still use the regular function names,
+                   as symbols are automatically translated through `xxhash.h`.
+- `XXH_FORCE_MEMORY_ACCESS`: The default method `0` uses a portable `memcpy()` notation.
+                             Method `1` uses a gcc-specific `packed` attribute, which can provide better performance for some targets.
+                             Method `2` forces unaligned reads, which is not standards compliant, but might sometimes be the only way to extract better read performance.
+                             Method `3` uses a byteshift operation, which is best for old compilers which don't inline `memcpy()` or big-endian systems without a byteswap instruction
+- `XXH_FORCE_ALIGN_CHECK`: Use a faster direct read path when input is aligned.
+                           This option can result in dramatic performance improvement when input to hash is aligned on 32 or 64-bit boundaries,
+                           when running on architectures unable to load memory from unaligned addresses, or suffering a performance penalty from it.
+                           It is (slightly) detrimental on platform with good unaligned memory access performance (same instruction for both aligned and unaligned accesses).
+                           This option is automatically disabled on `x86`, `x64` and `aarch64`, and enabled on all other platforms.
+- `XXH_VECTOR` : manually select a vector instruction set (default: auto-selected at compilation time). Available instruction sets are `XXH_SCALAR`, `XXH_SSE2`, `XXH_AVX2`, `XXH_AVX512`, `XXH_NEON` and `XXH_VSX`. Compiler may require additional flags to ensure proper support (for example, `gcc` on linux will require `-mavx2` for AVX2, and `-mavx512f` for AVX512).
+- `XXH_NO_PREFETCH` : disable prefetching. Some platforms or situations may perform better without prefetching. XXH3 only.
+- `XXH_PREFETCH_DIST` : select prefetching distance. For close-to-metal adaptation to specific hardware platforms. XXH3 only.
+- `XXH_NO_INLINE_HINTS`: By default, xxHash uses `__attribute__((always_inline))` and `__forceinline` to improve performance at the cost of code size.
+                         Defining this macro to 1 will mark all internal functions as `static`, allowing the compiler to decide whether to inline a function or not.
+                         This is very useful when optimizing for smallest binary size,
+                         and is automatically defined when compiling with `-O0`, `-Os`, `-Oz`, or `-fno-inline` on GCC and Clang.
+                         This may also increase performance depending on compiler and architecture.
+- `XXH32_ENDJMP`: Switch multi-branch finalization stage of XXH32 by a single jump.
+                  This is generally undesirable for performance, especially when hashing inputs of random sizes.
+                  But depending on exact architecture and compiler, a jump might provide slightly better performance on small inputs. Disabled by default.
+- `XXH_STATIC_LINKING_ONLY`: gives access to internal state declaration, required for static allocation.
+                             Incompatible with dynamic linking, due to risks of ABI changes.
+- `XXH_NO_XXH3` : removes symbols related to `XXH3` (both 64 & 128 bits) from generated binary.
+                  Useful to reduce binary size, notably for applications which do not use `XXH3`.
+- `XXH_NO_LONG_LONG`: removes compilation of algorithms relying on 64-bit types (XXH3 and XXH64). Only XXH32 will be compiled.
+                      Useful for targets (architectures and compilers) without 64-bit support.
+- `XXH_IMPORT`: MSVC specific: should only be defined for dynamic linking, as it prevents linkage errors.
+- `XXH_CPU_LITTLE_ENDIAN`: By default, endianness is determined by a runtime test resolved at compile time.
+                           If, for some reason, the compiler cannot simplify the runtime test, it can cost performance.
+                           It's possible to skip auto-detection and simply state that the architecture is little-endian by setting this macro to 1.
+                           Setting it to 0 states big-endian.
+- `XXH_DEBUGLEVEL` : When set to any value >= 1, enables `assert()` statements.
+                     This (slightly) slows down execution, but may help finding bugs during debugging sessions.
+
+When compiling the Command Line Interface `xxhsum` with `make`, the following environment variables can also be set :
+- `DISPATCH=1` : use `xxh_x86dispatch.c`, to automatically select between `scalar`, `sse2`, `avx2` or `avx512` instruction set at runtime, depending on local host. This option is only valid for `x86`/`x64` systems.
+
+
+### Building xxHash - Using vcpkg
+
+You can download and install xxHash using the [vcpkg](https://github.com/Microsoft/vcpkg) dependency manager:
+
+    git clone https://github.com/Microsoft/vcpkg.git
+    cd vcpkg
+    ./bootstrap-vcpkg.sh
+    ./vcpkg integrate install
+    ./vcpkg install xxhash
+
+The xxHash port in vcpkg is kept up to date by Microsoft team members and community contributors. If the version is out of date, please [create an issue or pull request](https://github.com/Microsoft/vcpkg) on the vcpkg repository.
+
+
+### Example
+
+The simplest example calls xxhash 64-bit variant as a one-shot function
+generating a hash value from a single buffer, and invoked from a C/C++ program:
+
+```C
+#include "xxhash.h"
+
+    (...)
+    XXH64_hash_t hash = XXH64(buffer, size, seed);
+}
+```
+
+Streaming variant is more involved, but makes it possible to provide data incrementally:
+
+```C
+#include "stdlib.h"   /* abort() */
+#include "xxhash.h"
+
+
+XXH64_hash_t calcul_hash_streaming(FileHandler fh)
+{
+    /* create a hash state */
+    XXH64_state_t* const state = XXH64_createState();
+    if (state==NULL) abort();
+
+    size_t const bufferSize = SOME_SIZE;
+    void* const buffer = malloc(bufferSize);
+    if (buffer==NULL) abort();
+
+    /* Initialize state with selected seed */
+    XXH64_hash_t const seed = 0;   /* or any other value */
+    if (XXH64_reset(state, seed) == XXH_ERROR) abort();
+
+    /* Feed the state with input data, any size, any number of times */
+    (...)
+    while ( /* some data left */ ) {
+        size_t const length = get_more_data(buffer, bufferSize, fh);
+        if (XXH64_update(state, buffer, length) == XXH_ERROR) abort();
+        (...)
+    }
+    (...)
+
+    /* Produce the final hash value */
+    XXH64_hash_t const hash = XXH64_digest(state);
+
+    /* State could be re-used; but in this example, it is simply freed  */
+    free(buffer);
+    XXH64_freeState(state);
+
+    return hash;
+}
+```
+
+
+### License
+
+The library files `xxhash.c` and `xxhash.h` are BSD licensed.
+The utility `xxhsum` is GPL licensed.
+
+
+### Other programming languages
+
+Beyond the C reference version,
+xxHash is also available from many different programming languages,
+thanks to great contributors.
+They are [listed here](http://www.xxhash.com/#other-languages).
+
+
+### Packaging status
+
+Many distributions bundle a package manager
+which allows easy xxhash installation as both a `libxxhash` library
+and `xxhsum` command line interface.
+
+[![Packaging status](https://repology.org/badge/vertical-allrepos/xxhash.svg)](https://repology.org/project/xxhash/versions)
+
+
+### Special Thanks
+
+- Takayuki Matsuoka, aka @t-mat, for creating `xxhsum -c` and great support during early xxh releases
+- Mathias Westerdahl, aka @JCash, for introducing the first version of `XXH64`
+- Devin Hussey, aka @easyaspi314, for incredible low-level optimizations on `XXH3` and `XXH128`
diff --git a/fluent-bit/lib/cfl/lib/xxhash/appveyor.yml b/fluent-bit/lib/cfl/lib/xxhash/appveyor.yml
new file mode 100644
index 00000000..7aef900f
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/appveyor.yml
@@ -0,0 +1,115 @@
+#---------------------------------#
+#      general configuration      #
+#---------------------------------#
+version: 1.0.{build}
+max_jobs: 2
+
+#---------------------------------#
+#    environment configuration    #
+#---------------------------------#
+clone_depth: 2
+environment:
+  matrix:
+  - COMPILER: "visual"
+    ARCH: "x64"
+    TEST_XXHSUM: "true"
+  - COMPILER: "visual"
+    ARCH: "Win32"
+    TEST_XXHSUM: "true"
+  - COMPILER: "visual"
+    ARCH: "Win32"
+    APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2013
+    TEST_XXHSUM: "true"
+  - COMPILER: "visual"
+    ARCH: "ARM"
+# Below tests are now disabled due to redundancy.
+# Their equivalent already runs correctly on Github Actions.
+#  - COMPILER: "visual"
+#    ARCH: "x64"
+#    APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017
+#    TEST_XXHSUM: "true"
+#  - COMPILER: "visual"
+#    ARCH: "ARM64"
+#    APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017
+#    # note: ARM64 is not available with Visual Studio 14 2015, which is default for Appveyor
+
+# The following tests were also flacky on Appveyor, for various reasons.
+#  - COMPILER: "gcc"
+#    PLATFORM: "mingw64"
+#  - COMPILER: "gcc"
+#    PLATFORM: "mingw32"
+#  - COMPILER: "gcc"
+#    PLATFORM: "clang"
+
+install:
+  - ECHO Installing %COMPILER% %PLATFORM% %ARCH%
+  - MKDIR bin
+  - if [%COMPILER%]==[gcc] SET PATH_ORIGINAL=%PATH%
+  - if [%COMPILER%]==[gcc] (
+      SET "PATH_MINGW32=c:\MinGW\bin;c:\MinGW\usr\bin" &&
+      SET "PATH_MINGW64=c:\msys64\mingw64\bin;c:\msys64\usr\bin" &&
+      COPY C:\MinGW\bin\mingw32-make.exe C:\MinGW\bin\make.exe &&
+      COPY C:\MinGW\bin\gcc.exe C:\MinGW\bin\cc.exe
+    )
+
+#---------------------------------#
+#       build configuration       #
+#---------------------------------#
+build_script:
+  - if [%PLATFORM%]==[mingw32] SET PATH=%PATH_MINGW32%;%PATH_ORIGINAL%
+  - if [%PLATFORM%]==[mingw64] SET PATH=%PATH_MINGW64%;%PATH_ORIGINAL%
+  - if [%PLATFORM%]==[clang]   SET PATH=%PATH_MINGW64%;%PATH_ORIGINAL%
+  - ECHO ***
+  - ECHO Building %COMPILER% %PLATFORM% %ARCH%
+  - ECHO ***
+
+  - if [%COMPILER%]==[gcc] (
+      if [%PLATFORM%]==[clang] (
+        clang -v
+      ) ELSE (
+        gcc -v
+      )
+    )
+  - if [%COMPILER%]==[gcc] (
+      echo ----- &&
+      make -v &&
+      echo ----- &&
+      if not [%PLATFORM%]==[clang] (
+        if [%PLATFORM%]==[mingw32] ( SET CPPFLAGS=-DPOOL_MT=0 ) &&
+        make -B clean test MOREFLAGS=-Werror
+      ) ELSE (
+        SET CXXFLAGS=--std=c++14 &&
+        make -B clean test CC=clang CXX=clang++ MOREFLAGS="--target=x86_64-w64-mingw32 -Werror -Wno-pass-failed" NO_C90_TEST=true
+      ) &&
+      make -C tests/bench
+    )
+    # note 1: strict c90 tests with clang fail, due to (erroneous) presence on `inline` keyword in some included system file
+    # note 2: multi-threading code doesn't work with mingw32, disabled through POOL_MT=0
+    # note 3: clang requires C++14 to compile sort because its own code contains c++14-only code
+
+  - if [%COMPILER%]==[visual] (
+      cd cmake_unofficial &&
+      cmake . -DCMAKE_BUILD_TYPE=Release -A %ARCH% -DXXHASH_C_FLAGS="/WX" &&
+      cmake --build . --config Release
+    )
+
+#---------------------------------#
+#       tests configuration       #
+#---------------------------------#
+test_script:
+  # note: can only run x86 and x64 binaries on Appveyor
+  # note: if %COMPILER%==gcc, xxhsum was already tested within `make test`
+  - if [%TEST_XXHSUM%]==[true] (
+      ECHO ***  &&
+      ECHO Testing %COMPILER% %PLATFORM% %ARCH%  &&
+      ECHO ***  &&
+      cd Release &&
+      xxhsum.exe -bi1 &&
+      ECHO ------- xxhsum tested -------
+    )
+
+
+#---------------------------------#
+#      artifacts configuration    #
+#---------------------------------#
+# none yet
diff --git a/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/.gitignore b/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/.gitignore
new file mode 100644
index 00000000..93d9fe4f
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/.gitignore
@@ -0,0 +1,12 @@
+# cmake artifacts
+
+CMakeCache.txt
+CMakeFiles
+Makefile
+cmake_install.cmake
+
+
+# make compilation results
+
+*.dylib
+*.a
diff --git a/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/CMakeLists.txt b/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/CMakeLists.txt
new file mode 100644
index 00000000..d5456b01
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/CMakeLists.txt
@@ -0,0 +1,173 @@
+# To the extent possible under law, the author(s) have dedicated all
+# copyright and related and neighboring rights to this software to
+# the public domain worldwide. This software is distributed without
+# any warranty.
+#
+# For details, see <https://creativecommons.org/publicdomain/zero/1.0/>.
+
+cmake_minimum_required (VERSION 2.8.12 FATAL_ERROR)
+
+set(XXHASH_DIR "${CMAKE_CURRENT_SOURCE_DIR}/..")
+
+file(STRINGS "${XXHASH_DIR}/xxhash.h" XXHASH_VERSION_MAJOR REGEX "^#define XXH_VERSION_MAJOR +([0-9]+) *$")
+string(REGEX REPLACE "^#define XXH_VERSION_MAJOR +([0-9]+) *$" "\\1" XXHASH_VERSION_MAJOR "${XXHASH_VERSION_MAJOR}")
+file(STRINGS "${XXHASH_DIR}/xxhash.h" XXHASH_VERSION_MINOR REGEX "^#define XXH_VERSION_MINOR +([0-9]+) *$")
+string(REGEX REPLACE "^#define XXH_VERSION_MINOR +([0-9]+) *$" "\\1" XXHASH_VERSION_MINOR "${XXHASH_VERSION_MINOR}")
+file(STRINGS "${XXHASH_DIR}/xxhash.h" XXHASH_VERSION_RELEASE REGEX "^#define XXH_VERSION_RELEASE +([0-9]+) *$")
+string(REGEX REPLACE "^#define XXH_VERSION_RELEASE +([0-9]+) *$" "\\1" XXHASH_VERSION_RELEASE "${XXHASH_VERSION_RELEASE}")
+set(XXHASH_VERSION_STRING "${XXHASH_VERSION_MAJOR}.${XXHASH_VERSION_MINOR}.${XXHASH_VERSION_RELEASE}")
+set(XXHASH_LIB_VERSION ${XXHASH_VERSION_STRING})
+set(XXHASH_LIB_SOVERSION "${XXHASH_VERSION_MAJOR}")
+mark_as_advanced(XXHASH_VERSION_MAJOR XXHASH_VERSION_MINOR XXHASH_VERSION_RELEASE XXHASH_VERSION_STRING XXHASH_LIB_VERSION XXHASH_LIB_SOVERSION)
+
+if("${CMAKE_VERSION}" VERSION_LESS "3.13")
+    #message(WARNING "CMake ${CMAKE_VERSION} has no CMP0077 policy: options will erase uncached/untyped normal vars!")
+else()
+    cmake_policy (SET CMP0077 NEW)
+endif()
+if("${CMAKE_VERSION}" VERSION_LESS "3.0")
+    project(xxHash C)
+else()
+    cmake_policy (SET CMP0048 NEW)
+    project(xxHash
+        VERSION ${XXHASH_VERSION_STRING}
+        LANGUAGES C)
+endif()
+
+if(NOT CMAKE_BUILD_TYPE AND NOT CMAKE_CONFIGURATION_TYPES)
+  set(CMAKE_BUILD_TYPE "Release" CACHE STRING "Project build type" FORCE)
+  set_property(CACHE CMAKE_BUILD_TYPE
+    PROPERTY STRINGS "Debug" "Release" "RelWithDebInfo" "MinSizeRel")
+endif()
+if(NOT CMAKE_CONFIGURATION_TYPES)
+  message(STATUS "xxHash build type: ${CMAKE_BUILD_TYPE}")
+endif()
+# Enable assert() statements in debug builds
+if("${CMAKE_BUILD_TYPE}" STREQUAL "Debug")
+  add_compile_definitions(XXH_DEBUGLEVEL=1)
+endif()
+
+option(BUILD_SHARED_LIBS "Build shared library" ON)
+option(XXHASH_BUILD_XXHSUM "Build the xxhsum binary" ON)
+
+# If XXHASH is being bundled in another project, we don't want to
+# install anything.  However, we want to let people override this, so
+# we'll use the XXHASH_BUNDLED_MODE variable to let them do that; just
+# set it to OFF in your project before you add_subdirectory(xxhash/cmake_unofficial).
+if(NOT DEFINED XXHASH_BUNDLED_MODE)
+  if("${PROJECT_SOURCE_DIR}" STREQUAL "${CMAKE_SOURCE_DIR}")
+    set(XXHASH_BUNDLED_MODE OFF)
+  else()
+    set(XXHASH_BUNDLED_MODE ON)
+  endif()
+endif()
+set(XXHASH_BUNDLED_MODE ${XXHASH_BUNDLED_MODE} CACHE BOOL "" FORCE)
+mark_as_advanced(XXHASH_BUNDLED_MODE)
+
+# Allow people to choose whether to build shared or static libraries
+# via the BUILD_SHARED_LIBS option unless we are in bundled mode, in
+# which case we always use static libraries.
+include(CMakeDependentOption)
+CMAKE_DEPENDENT_OPTION(BUILD_SHARED_LIBS "Build shared libraries" ON "NOT XXHASH_BUNDLED_MODE" OFF)
+
+# libxxhash
+add_library(xxhash "${XXHASH_DIR}/xxhash.c")
+add_library(${PROJECT_NAME}::xxhash ALIAS xxhash)
+
+target_include_directories(xxhash
+  PUBLIC
+    $<BUILD_INTERFACE:${XXHASH_DIR}>
+    $<INSTALL_INTERFACE:include/>)
+if (BUILD_SHARED_LIBS)
+  target_compile_definitions(xxhash PUBLIC XXH_EXPORT)
+endif ()
+set_target_properties(xxhash PROPERTIES
+  SOVERSION "${XXHASH_LIB_SOVERSION}"
+  VERSION "${XXHASH_VERSION_STRING}")
+
+if(XXHASH_BUILD_XXHSUM)
+  set(XXHSUM_DIR "${XXHASH_DIR}/cli")
+  # xxhsum
+  add_executable(xxhsum "${XXHSUM_DIR}/xxhsum.c"
+                        "${XXHSUM_DIR}/xsum_os_specific.c"
+                        "${XXHSUM_DIR}/xsum_output.c"
+                        "${XXHSUM_DIR}/xsum_sanity_check.c"
+                        "${XXHSUM_DIR}/xsum_bench.c"
+                )
+  add_executable(${PROJECT_NAME}::xxhsum ALIAS xxhsum)
+
+  target_link_libraries(xxhsum PRIVATE xxhash)
+  target_include_directories(xxhsum PRIVATE "${XXHASH_DIR}")
+endif(XXHASH_BUILD_XXHSUM)
+
+# Extra warning flags
+include (CheckCCompilerFlag)
+if (XXHASH_C_FLAGS)
+  set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${XXHASH_C_FLAGS}")
+endif()
+
+if(NOT XXHASH_BUNDLED_MODE)
+  include(GNUInstallDirs)
+
+  install(TARGETS xxhash
+    EXPORT xxHashTargets
+    RUNTIME DESTINATION "${CMAKE_INSTALL_BINDIR}"
+    LIBRARY DESTINATION "${CMAKE_INSTALL_LIBDIR}"
+    ARCHIVE DESTINATION "${CMAKE_INSTALL_LIBDIR}")
+  install(FILES "${XXHASH_DIR}/xxhash.h"
+    DESTINATION "${CMAKE_INSTALL_INCLUDEDIR}")
+  install(FILES "${XXHASH_DIR}/xxh3.h"
+    DESTINATION "${CMAKE_INSTALL_INCLUDEDIR}")
+  if(XXHASH_BUILD_XXHSUM)
+    install(TARGETS xxhsum
+      EXPORT xxHashTargets
+      RUNTIME DESTINATION "${CMAKE_INSTALL_BINDIR}")
+    install(FILES "${XXHASH_DIR}/xxhsum.1"
+      DESTINATION "${CMAKE_INSTALL_MANDIR}/man1")
+  endif(XXHASH_BUILD_XXHSUM)
+
+  include(CMakePackageConfigHelpers)
+
+  set(xxHash_VERSION_CONFIG "${PROJECT_BINARY_DIR}/xxHashConfigVersion.cmake")
+  set(xxHash_PROJECT_CONFIG "${PROJECT_BINARY_DIR}/xxHashConfig.cmake")
+  set(xxHash_TARGETS_CONFIG "${PROJECT_BINARY_DIR}/xxHashTargets.cmake")
+  set(xxHash_CONFIG_INSTALL_DIR "${CMAKE_INSTALL_LIBDIR}/cmake/xxHash/")
+  write_basic_package_version_file(${xxHash_VERSION_CONFIG}
+    VERSION ${XXHASH_VERSION_STRING}
+    COMPATIBILITY AnyNewerVersion)
+  configure_package_config_file(
+    ${PROJECT_SOURCE_DIR}/xxHashConfig.cmake.in
+    ${xxHash_PROJECT_CONFIG}
+    INSTALL_DESTINATION ${xxHash_CONFIG_INSTALL_DIR})
+  if("${CMAKE_VERSION}" VERSION_LESS "3.0")
+      set(XXHASH_EXPORT_SET xxhash)
+      if(XXHASH_BUILD_XXHSUM)
+        set(XXHASH_EXPORT_SET ${XXHASH_EXPORT_SET} xxhsum)
+      endif()
+      export(TARGETS ${XXHASH_EXPORT_SET}
+      FILE ${xxHash_TARGETS_CONFIG}
+      NAMESPACE ${PROJECT_NAME}::)
+  else()
+    export(EXPORT xxHashTargets
+      FILE ${xxHash_TARGETS_CONFIG}
+      NAMESPACE ${PROJECT_NAME}::)
+  endif()
+
+  install(FILES ${xxHash_PROJECT_CONFIG} ${xxHash_VERSION_CONFIG}
+    DESTINATION ${xxHash_CONFIG_INSTALL_DIR})
+  install(EXPORT xxHashTargets
+    DESTINATION ${xxHash_CONFIG_INSTALL_DIR}
+    NAMESPACE ${PROJECT_NAME}::)
+
+  # configure and install pkg-config
+  set(PREFIX ${CMAKE_INSTALL_PREFIX})
+  set(EXECPREFIX "\${prefix}")
+  set(INCLUDEDIR "${CMAKE_INSTALL_INCLUDEDIR}")
+  set(LIBDIR "${CMAKE_INSTALL_LIBDIR}")
+  set(VERSION "${XXHASH_VERSION_STRING}")
+  configure_file(${XXHASH_DIR}/libxxhash.pc.in ${CMAKE_BINARY_DIR}/libxxhash.pc @ONLY)
+
+  install(FILES ${CMAKE_BINARY_DIR}/libxxhash.pc
+    DESTINATION ${CMAKE_INSTALL_LIBDIR}/pkgconfig)
+
+endif(NOT XXHASH_BUNDLED_MODE)
diff --git a/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/README.md b/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/README.md
new file mode 100644
index 00000000..66df7909
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/README.md
@@ -0,0 +1,36 @@
+
+## Usage
+
+### Way 1: import targets
+Build xxHash targets:
+
+    cd </path/to/xxHash/>
+    mkdir build
+    cd build
+    cmake ../cmake_unofficial [options]
+    cmake --build .
+    cmake --build . --target install #optional
+
+Where possible options are:
+- `-DXXHASH_BUILD_ENABLE_INLINE_API=<ON|OFF>`: adds xxhash.c for the `-DXXH_INLINE_ALL` api. ON by default.
+- `-DXXHASH_BUILD_XXHSUM=<ON|OFF>`: build the command line binary. ON by default
+- `-DBUILD_SHARED_LIBS=<ON|OFF>`: build dynamic library. ON by default.
+- `-DCMAKE_INSTALL_PREFIX=<path>`: use custom install prefix path.
+
+Add lines into downstream CMakeLists.txt:
+
+    find_package(xxHash 0.7 CONFIG REQUIRED)
+    ...
+    target_link_libraries(MyTarget PRIVATE xxHash::xxhash)
+
+### Way 2: Add subdirectory
+Add lines into downstream CMakeLists.txt:
+
+    option(BUILD_SHARED_LIBS "Build shared libs" OFF) #optional
+    ...
+    set(XXHASH_BUILD_ENABLE_INLINE_API OFF) #optional
+    set(XXHASH_BUILD_XXHSUM OFF) #optional
+    add_subdirectory(</path/to/xxHash/cmake_unofficial/> </path/to/xxHash/build/> EXCLUDE_FROM_ALL)
+    ...
+    target_link_libraries(MyTarget PRIVATE xxHash::xxhash)
+
diff --git a/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/xxHashConfig.cmake.in b/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/xxHashConfig.cmake.in
new file mode 100644
index 00000000..fd282bee
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/cmake_unofficial/xxHashConfig.cmake.in
@@ -0,0 +1,4 @@
+@PACKAGE_INIT@
+
+include(${CMAKE_CURRENT_LIST_DIR}/xxHashTargets.cmake)
+
diff --git a/fluent-bit/lib/cfl/lib/xxhash/doc/README.md b/fluent-bit/lib/cfl/lib/xxhash/doc/README.md
new file mode 100644
index 00000000..a73ad729
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/doc/README.md
@@ -0,0 +1,9 @@
+xxHash Specification
+=======================
+
+This directory contains material defining the xxHash algorithm.
+It's described in [this specification document](xxhash_spec.md).
+
+The algorithm is also be illustrated by a [simple educational library](https://github.com/easyaspi314/xxhash-clean),
+written by @easyaspi314 and designed for readability
+(as opposed to the reference library which is designed for speed).
diff --git a/fluent-bit/lib/cfl/lib/xxhash/doc/xxhash.cry b/fluent-bit/lib/cfl/lib/xxhash/doc/xxhash.cry
new file mode 100644
index 00000000..984e1c8b
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/doc/xxhash.cry
@@ -0,0 +1,206 @@
+module xxhash where
+
+/**
+ * The 32-bit variant of xxHash. The first argument is the sequence
+ * of L bytes to hash. The second argument is a seed value.
+ */
+XXH32 : {L} (fin L) => [L][8] -> [32] -> [32]
+XXH32 input seed = XXH32_avalanche acc1
+  where (stripes16 # stripes4 # stripes1) = input
+        accR = foldl XXH32_rounds (XXH32_init seed) (split stripes16 : [L/16][16][8])
+        accL = `(L % 2^^32) + if (`L:Integer) < 16
+                              then seed + PRIME32_5
+                              else XXH32_converge accR
+        acc4 = foldl XXH32_digest4 accL (split stripes4 : [(L%16)/4][4][8])
+        acc1 = foldl XXH32_digest1 acc4 (stripes1 : [L%4][8])
+
+/**
+ * The 64-bit variant of xxHash. The first argument is the sequence
+ * of L bytes to hash. The second argument is a seed value.
+ */
+XXH64 : {L} (fin L) => [L][8] -> [64] -> [64]
+XXH64 input seed = XXH64_avalanche acc1
+  where (stripes32 # stripes8 # stripes4 # stripes1) = input
+        accR = foldl XXH64_rounds (XXH64_init seed) (split stripes32 : [L/32][32][8])
+        accL = `(L % 2^^64) + if (`L:Integer) < 32
+                              then seed + PRIME64_5
+                              else XXH64_converge accR
+        acc8 = foldl XXH64_digest8 accL (split stripes8 : [(L%32)/8][8][8])
+        acc4 = foldl XXH64_digest4 acc8 (split stripes4 : [(L%8)/4][4][8])
+        acc1 = foldl XXH64_digest1 acc4 (stripes1 : [L%4][8])
+
+private
+
+  //Utility functions
+
+  /**
+   * Combines a sequence of bytes into a word using the little-endian
+   * convention.
+   */
+  toLE bytes = join (reverse bytes)
+
+  //32-bit xxHash helper functions
+
+  //32-bit prime number constants
+  PRIME32_1 = 0x9E3779B1 : [32]
+  PRIME32_2 = 0x85EBCA77 : [32]
+  PRIME32_3 = 0xC2B2AE3D : [32]
+  PRIME32_4 = 0x27D4EB2F : [32]
+  PRIME32_5 = 0x165667B1 : [32]
+
+  /**
+   * The property shows that the hexadecimal representation of the
+   * PRIME32 constants is the same as the binary representation.
+   */
+  property PRIME32s_as_bits_correct =
+    (PRIME32_1 == 0b10011110001101110111100110110001) /\
+    (PRIME32_2 == 0b10000101111010111100101001110111) /\
+    (PRIME32_3 == 0b11000010101100101010111000111101) /\
+    (PRIME32_4 == 0b00100111110101001110101100101111) /\
+    (PRIME32_5 == 0b00010110010101100110011110110001)
+
+  /**
+   * This function initializes the four internal accumulators of XXH32.
+   */
+  XXH32_init : [32] -> [4][32]
+  XXH32_init seed = [acc1, acc2, acc3, acc4]
+    where acc1 = seed + PRIME32_1 + PRIME32_2
+          acc2 = seed + PRIME32_2
+          acc3 = seed + 0
+          acc4 = seed - PRIME32_1
+
+  /**
+   * This processes a single lane of the main round function of XXH32.
+   */
+  XXH32_round : [32] -> [32] -> [32]
+  XXH32_round accN laneN = ((accN + laneN * PRIME32_2) <<< 13) * PRIME32_1
+
+  /**
+   * This is the main round function of XXH32 and processes a stripe,
+   * i.e. 4 lanes with 4 bytes each.
+   */
+  XXH32_rounds : [4][32] -> [16][8] -> [4][32]
+  XXH32_rounds accs stripe =
+    [ XXH32_round accN (toLE laneN) | accN <- accs | laneN <- split stripe ]
+
+  /**
+   * This function combines the four lane accumulators into a single
+   * 32-bit value.
+   */
+  XXH32_converge : [4][32] -> [32]
+  XXH32_converge [acc1, acc2, acc3, acc4] =
+    (acc1 <<< 1) + (acc2 <<< 7) + (acc3 <<< 12) + (acc4 <<< 18)
+
+  /**
+   * This function digests a four byte lane
+   */
+  XXH32_digest4 : [32] -> [4][8] -> [32]
+  XXH32_digest4 acc lane = ((acc + toLE lane * PRIME32_3) <<< 17) * PRIME32_4
+
+  /**
+   * This function digests a single byte lane
+   */
+  XXH32_digest1 : [32] -> [8] -> [32]
+  XXH32_digest1 acc lane = ((acc + (0 # lane) * PRIME32_5) <<< 11) * PRIME32_1
+
+  /**
+   * This function ensures that all input bits have a chance to impact
+   * any bit in the output digest, resulting in an unbiased
+   * distribution.
+   */
+  XXH32_avalanche : [32] -> [32]
+  XXH32_avalanche acc0 = acc5
+    where acc1 = acc0 ^ (acc0 >> 15)
+          acc2 = acc1 * PRIME32_2
+          acc3 = acc2 ^ (acc2 >> 13)
+          acc4 = acc3 * PRIME32_3
+          acc5 = acc4 ^ (acc4 >> 16)
+
+  //64-bit xxHash helper functions
+
+  //64-bit prime number constants
+  PRIME64_1 = 0x9E3779B185EBCA87 : [64]
+  PRIME64_2 = 0xC2B2AE3D27D4EB4F : [64]
+  PRIME64_3 = 0x165667B19E3779F9 : [64]
+  PRIME64_4 = 0x85EBCA77C2B2AE63 : [64]
+  PRIME64_5 = 0x27D4EB2F165667C5 : [64]
+
+  /**
+   * The property shows that the hexadecimal representation of the
+   * PRIME64 constants is the same as the binary representation.
+   */
+  property PRIME64s_as_bits_correct =
+    (PRIME64_1 == 0b1001111000110111011110011011000110000101111010111100101010000111) /\
+    (PRIME64_2 == 0b1100001010110010101011100011110100100111110101001110101101001111) /\
+    (PRIME64_3 == 0b0001011001010110011001111011000110011110001101110111100111111001) /\
+    (PRIME64_4 == 0b1000010111101011110010100111011111000010101100101010111001100011) /\
+    (PRIME64_5 == 0b0010011111010100111010110010111100010110010101100110011111000101)
+
+  /**
+   * This function initializes the four internal accumulators of XXH64.
+   */
+  XXH64_init : [64] -> [4][64]
+  XXH64_init seed = [acc1, acc2, acc3, acc4]
+    where acc1 = seed + PRIME64_1 + PRIME64_2
+          acc2 = seed + PRIME64_2
+          acc3 = seed + 0
+          acc4 = seed - PRIME64_1
+
+  /**
+   * This processes a single lane of the main round function of XXH64.
+   */
+  XXH64_round : [64] -> [64] -> [64]
+  XXH64_round accN laneN = ((accN + laneN * PRIME64_2) <<< 31) * PRIME64_1
+
+  /**
+   * This is the main round function of XXH64 and processes a stripe,
+   * i.e. 4 lanes with 8 bytes each.
+   */
+  XXH64_rounds : [4][64] -> [32][8] -> [4][64]
+  XXH64_rounds accs stripe =
+    [ XXH64_round accN (toLE laneN) | accN <- accs | laneN <- split stripe ]
+
+  /**
+   * This is a helper function, used to merge the four lane accumulators.
+   */
+  mergeAccumulator : [64] -> [64] -> [64]
+  mergeAccumulator acc accN = (acc ^ XXH64_round 0 accN) * PRIME64_1 + PRIME64_4
+
+  /**
+   * This function combines the four lane accumulators into a single
+   * 64-bit value.
+   */
+  XXH64_converge : [4][64] -> [64]
+  XXH64_converge [acc1, acc2, acc3, acc4] =
+    foldl mergeAccumulator ((acc1 <<< 1) + (acc2 <<< 7) + (acc3 <<< 12) + (acc4 <<< 18)) [acc1, acc2, acc3, acc4]
+
+  /**
+   * This function digests an eight byte lane
+   */
+  XXH64_digest8 : [64] -> [8][8] -> [64]
+  XXH64_digest8 acc lane = ((acc ^ XXH64_round 0 (toLE lane)) <<< 27) * PRIME64_1 + PRIME64_4
+
+  /**
+   * This function digests a four byte lane
+   */
+  XXH64_digest4 : [64] -> [4][8] -> [64]
+  XXH64_digest4 acc lane = ((acc ^ (0 # toLE lane) * PRIME64_1) <<< 23) * PRIME64_2 + PRIME64_3
+
+  /**
+   * This function digests a single byte lane
+   */
+  XXH64_digest1 : [64] -> [8] -> [64]
+  XXH64_digest1 acc lane = ((acc ^ (0 # lane) * PRIME64_5) <<< 11) * PRIME64_1
+
+  /**
+   * This function ensures that all input bits have a chance to impact
+   * any bit in the output digest, resulting in an unbiased
+   * distribution.
+   */
+  XXH64_avalanche : [64] -> [64]
+  XXH64_avalanche acc0 = acc5
+    where acc1 = acc0 ^ (acc0 >> 33)
+          acc2 = acc1 * PRIME64_2
+          acc3 = acc2 ^ (acc2 >> 29)
+          acc4 = acc3 * PRIME64_3
+          acc5 = acc4 ^ (acc4 >> 32)
diff --git a/fluent-bit/lib/cfl/lib/xxhash/doc/xxhash_spec.md b/fluent-bit/lib/cfl/lib/xxhash/doc/xxhash_spec.md
new file mode 100644
index 00000000..80576d6b
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/doc/xxhash_spec.md
@@ -0,0 +1,351 @@
+xxHash fast digest algorithm
+======================
+
+### Notices
+
+Copyright (c) Yann Collet
+
+Permission is granted to copy and distribute this document
+for any purpose and without charge,
+including translations into other languages
+and incorporation into compilations,
+provided that the copyright notice and this notice are preserved,
+and that any substantive changes or deletions from the original
+are clearly marked.
+Distribution of this document is unlimited.
+
+### Version
+
+0.1.1 (10/10/18)
+
+
+Table of Contents
+---------------------
+- [Introduction](#introduction)
+- [XXH32 algorithm description](#xxh32-algorithm-description)
+- [XXH64 algorithm description](#xxh64-algorithm-description)
+- [Performance considerations](#performance-considerations)
+- [Reference Implementation](#reference-implementation)
+
+
+Introduction
+----------------
+
+This document describes the xxHash digest algorithm for both 32-bit and 64-bit variants, named `XXH32` and `XXH64`. The algorithm takes an input a message of arbitrary length and an optional seed value, then produces an output of 32 or 64-bit as "fingerprint" or "digest".
+
+xxHash is primarily designed for speed. It is labeled non-cryptographic, and is not meant to avoid intentional collisions (same digest for 2 different messages), or to prevent producing a message with a predefined digest.
+
+XXH32 is designed to be fast on 32-bit machines.
+XXH64 is designed to be fast on 64-bit machines.
+Both variants produce different output.
+However, a given variant shall produce exactly the same output, irrespective of the cpu / os used. In particular, the result remains identical whatever the endianness and width of the cpu is.
+
+### Operation notations
+
+All operations are performed modulo {32,64} bits. Arithmetic overflows are expected.
+`XXH32` uses 32-bit modular operations. `XXH64` uses 64-bit modular operations.
+
+- `+`: denotes modular addition
+- `*`: denotes modular multiplication
+- `X <<< s`: denotes the value obtained by circularly shifting (rotating) `X` left by `s` bit positions.
+- `X >> s`: denotes the value obtained by shifting `X` right by s bit positions. Upper `s` bits become `0`.
+- `X xor Y`: denotes the bit-wise XOR of `X` and `Y` (same width).
+
+
+XXH32 Algorithm Description
+-------------------------------------
+
+### Overview
+
+We begin by supposing that we have a message of any length `L` as input, and that we wish to find its digest. Here `L` is an arbitrary nonnegative integer; `L` may be zero. The following steps are performed to compute the digest of the message.
+
+The algorithm collect and transform input in _stripes_ of 16 bytes. The transforms are stored inside 4 "accumulators", each one storing an unsigned 32-bit value. Each accumulator can be processed independently in parallel, speeding up processing for cpu with multiple execution units.
+
+The algorithm uses 32-bits addition, multiplication, rotate, shift and xor operations. Many operations require some 32-bits prime number constants, all defined below:
+
+```c
+  static const u32 PRIME32_1 = 0x9E3779B1U;  // 0b10011110001101110111100110110001
+  static const u32 PRIME32_2 = 0x85EBCA77U;  // 0b10000101111010111100101001110111
+  static const u32 PRIME32_3 = 0xC2B2AE3DU;  // 0b11000010101100101010111000111101
+  static const u32 PRIME32_4 = 0x27D4EB2FU;  // 0b00100111110101001110101100101111
+  static const u32 PRIME32_5 = 0x165667B1U;  // 0b00010110010101100110011110110001
+```
+
+These constants are prime numbers, and feature a good mix of bits 1 and 0, neither too regular, nor too dissymmetric. These properties help dispersion capabilities.
+
+### Step 1. Initialize internal accumulators
+
+Each accumulator gets an initial value based on optional `seed` input. Since the `seed` is optional, it can be `0`.
+
+```c
+  u32 acc1 = seed + PRIME32_1 + PRIME32_2;
+  u32 acc2 = seed + PRIME32_2;
+  u32 acc3 = seed + 0;
+  u32 acc4 = seed - PRIME32_1;
+```
+
+#### Special case: input is less than 16 bytes
+
+When the input is too small (< 16 bytes), the algorithm will not process any stripes. Consequently, it will not make use of parallel accumulators.
+
+In this case, a simplified initialization is performed, using a single accumulator:
+
+```c
+  u32 acc  = seed + PRIME32_5;
+```
+
+The algorithm then proceeds directly to step 4.
+
+### Step 2. Process stripes
+
+A stripe is a contiguous segment of 16 bytes.
+It is evenly divided into 4 _lanes_, of 4 bytes each.
+The first lane is used to update accumulator 1, the second lane is used to update accumulator 2, and so on.
+
+Each lane read its associated 32-bit value using __little-endian__ convention.
+
+For each {lane, accumulator}, the update process is called a _round_, and applies the following formula:
+
+```c
+  accN = accN + (laneN * PRIME32_2);
+  accN = accN <<< 13;
+  accN = accN * PRIME32_1;
+```
+
+This shuffles the bits so that any bit from input _lane_ impacts several bits in output _accumulator_. All operations are performed modulo 2^32.
+
+Input is consumed one full stripe at a time. Step 2 is looped as many times as necessary to consume the whole input, except for the last remaining bytes which cannot form a stripe (< 16 bytes).
+When that happens, move to step 3.
+
+### Step 3. Accumulator convergence
+
+All 4 lane accumulators from the previous steps are merged to produce a single remaining accumulator of the same width (32-bit). The associated formula is as follows:
+
+```c
+  acc = (acc1 <<< 1) + (acc2 <<< 7) + (acc3 <<< 12) + (acc4 <<< 18);
+```
+
+### Step 4. Add input length
+
+The input total length is presumed known at this stage. This step is just about adding the length to accumulator, so that it participates to final mixing.
+
+```c
+  acc = acc + (u32)inputLength;
+```
+
+Note that, if input length is so large that it requires more than 32-bits, only the lower 32-bits are added to the accumulator.
+
+### Step 5. Consume remaining input
+
+There may be up to 15 bytes remaining to consume from the input.
+The final stage will digest them according to following pseudo-code:
+
+```c
+  while (remainingLength >= 4) {
+      lane = read_32bit_little_endian(input_ptr);
+      acc = acc + lane * PRIME32_3;
+      acc = (acc <<< 17) * PRIME32_4;
+      input_ptr += 4; remainingLength -= 4;
+  }
+
+  while (remainingLength >= 1) {
+      lane = read_byte(input_ptr);
+      acc = acc + lane * PRIME32_5;
+      acc = (acc <<< 11) * PRIME32_1;
+      input_ptr += 1; remainingLength -= 1;
+  }
+```
+
+This process ensures that all input bytes are present in the final mix.
+
+### Step 6. Final mix (avalanche)
+
+The final mix ensures that all input bits have a chance to impact any bit in the output digest, resulting in an unbiased distribution. This is also called avalanche effect.
+
+```c
+  acc = acc xor (acc >> 15);
+  acc = acc * PRIME32_2;
+  acc = acc xor (acc >> 13);
+  acc = acc * PRIME32_3;
+  acc = acc xor (acc >> 16);
+```
+
+### Step 7. Output
+
+The `XXH32()` function produces an unsigned 32-bit value as output.
+
+For systems which require to store and/or display the result in binary or hexadecimal format, the canonical format is defined to reproduce the same value as the natural decimal format, hence follows __big-endian__ convention (most significant byte first).
+
+
+XXH64 Algorithm Description
+-------------------------------------
+
+### Overview
+
+`XXH64`'s algorithm structure is very similar to `XXH32` one. The major difference is that `XXH64` uses 64-bit arithmetic, speeding up memory transfer for 64-bit compliant systems, but also relying on cpu capability to efficiently perform 64-bit operations.
+
+The algorithm collects and transforms input in _stripes_ of 32 bytes. The transforms are stored inside 4 "accumulators", each one storing an unsigned 64-bit value. Each accumulator can be processed independently in parallel, speeding up processing for cpu with multiple execution units.
+
+The algorithm uses 64-bit addition, multiplication, rotate, shift and xor operations. Many operations require some 64-bit prime number constants, all defined below:
+
+```c
+  static const u64 PRIME64_1 = 0x9E3779B185EBCA87ULL;  // 0b1001111000110111011110011011000110000101111010111100101010000111
+  static const u64 PRIME64_2 = 0xC2B2AE3D27D4EB4FULL;  // 0b1100001010110010101011100011110100100111110101001110101101001111
+  static const u64 PRIME64_3 = 0x165667B19E3779F9ULL;  // 0b0001011001010110011001111011000110011110001101110111100111111001
+  static const u64 PRIME64_4 = 0x85EBCA77C2B2AE63ULL;  // 0b1000010111101011110010100111011111000010101100101010111001100011
+  static const u64 PRIME64_5 = 0x27D4EB2F165667C5ULL;  // 0b0010011111010100111010110010111100010110010101100110011111000101
+```
+
+These constants are prime numbers, and feature a good mix of bits 1 and 0, neither too regular, nor too dissymmetric. These properties help dispersion capabilities.
+
+### Step 1. Initialize internal accumulators
+
+Each accumulator gets an initial value based on optional `seed` input. Since the `seed` is optional, it can be `0`.
+
+```c
+  u64 acc1 = seed + PRIME64_1 + PRIME64_2;
+  u64 acc2 = seed + PRIME64_2;
+  u64 acc3 = seed + 0;
+  u64 acc4 = seed - PRIME64_1;
+```
+
+#### Special case: input is less than 32 bytes
+
+When the input is too small (< 32 bytes), the algorithm will not process any stripes. Consequently, it will not make use of parallel accumulators.
+
+In this case, a simplified initialization is performed, using a single accumulator:
+
+```c
+  u64 acc  = seed + PRIME64_5;
+```
+
+The algorithm then proceeds directly to step 4.
+
+### Step 2. Process stripes
+
+A stripe is a contiguous segment of 32 bytes.
+It is evenly divided into 4 _lanes_, of 8 bytes each.
+The first lane is used to update accumulator 1, the second lane is used to update accumulator 2, and so on.
+
+Each lane read its associated 64-bit value using __little-endian__ convention.
+
+For each {lane, accumulator}, the update process is called a _round_, and applies the following formula:
+
+```c
+  round(accN,laneN):
+  accN = accN + (laneN * PRIME64_2);
+  accN = accN <<< 31;
+  return accN * PRIME64_1;
+```
+
+This shuffles the bits so that any bit from input _lane_ impacts several bits in output _accumulator_. All operations are performed modulo 2^64.
+
+Input is consumed one full stripe at a time. Step 2 is looped as many times as necessary to consume the whole input, except for the last remaining bytes which cannot form a stripe (< 32 bytes).
+When that happens, move to step 3.
+
+### Step 3. Accumulator convergence
+
+All 4 lane accumulators from previous steps are merged to produce a single remaining accumulator of same width (64-bit). The associated formula is as follows.
+
+Note that accumulator convergence is more complex than 32-bit variant, and requires to define another function called _mergeAccumulator()_:
+
+```c
+  mergeAccumulator(acc,accN):
+  acc  = acc xor round(0, accN);
+  acc  = acc * PRIME64_1;
+  return acc + PRIME64_4;
+```
+
+which is then used in the convergence formula:
+
+```c
+  acc = (acc1 <<< 1) + (acc2 <<< 7) + (acc3 <<< 12) + (acc4 <<< 18);
+  acc = mergeAccumulator(acc, acc1);
+  acc = mergeAccumulator(acc, acc2);
+  acc = mergeAccumulator(acc, acc3);
+  acc = mergeAccumulator(acc, acc4);
+```
+
+### Step 4. Add input length
+
+The input total length is presumed known at this stage. This step is just about adding the length to accumulator, so that it participates to final mixing.
+
+```c
+  acc = acc + inputLength;
+```
+
+### Step 5. Consume remaining input
+
+There may be up to 31 bytes remaining to consume from the input.
+The final stage will digest them according to following pseudo-code:
+
+```c
+  while (remainingLength >= 8) {
+      lane = read_64bit_little_endian(input_ptr);
+      acc = acc xor round(0, lane);
+      acc = (acc <<< 27) * PRIME64_1;
+      acc = acc + PRIME64_4;
+      input_ptr += 8; remainingLength -= 8;
+  }
+
+  if (remainingLength >= 4) {
+      lane = read_32bit_little_endian(input_ptr);
+      acc = acc xor (lane * PRIME64_1);
+      acc = (acc <<< 23) * PRIME64_2;
+      acc = acc + PRIME64_3;
+      input_ptr += 4; remainingLength -= 4;
+  }
+
+  while (remainingLength >= 1) {
+      lane = read_byte(input_ptr);
+      acc = acc xor (lane * PRIME64_5);
+      acc = (acc <<< 11) * PRIME64_1;
+      input_ptr += 1; remainingLength -= 1;
+  }
+```
+
+This process ensures that all input bytes are present in the final mix.
+
+### Step 6. Final mix (avalanche)
+
+The final mix ensures that all input bits have a chance to impact any bit in the output digest, resulting in an unbiased distribution. This is also called avalanche effect.
+
+```c
+  acc = acc xor (acc >> 33);
+  acc = acc * PRIME64_2;
+  acc = acc xor (acc >> 29);
+  acc = acc * PRIME64_3;
+  acc = acc xor (acc >> 32);
+```
+
+### Step 7. Output
+
+The `XXH64()` function produces an unsigned 64-bit value as output.
+
+For systems which require to store and/or display the result in binary or hexadecimal format, the canonical format is defined to reproduce the same value as the natural decimal format, hence follows __big-endian__ convention (most significant byte first).
+
+Performance considerations
+----------------------------------
+
+The xxHash algorithms are simple and compact to implement. They provide a system independent "fingerprint" or digest of a message of arbitrary length.
+
+The algorithm allows input to be streamed and processed in multiple steps. In such case, an internal buffer is needed to ensure data is presented to the algorithm in full stripes.
+
+On 64-bit systems, the 64-bit variant `XXH64` is generally faster to compute, so it is a recommended variant, even when only 32-bit are needed.
+
+On 32-bit systems though, positions are reversed: `XXH64` performance is reduced, due to its usage of 64-bit arithmetic. `XXH32` becomes a faster variant.
+
+
+Reference Implementation
+----------------------------------------
+
+A reference library written in C is available at https://www.xxhash.com.
+The web page also links to multiple other implementations written in many different languages.
+It links to the [github project page](https://github.com/Cyan4973/xxHash) where an [issue board](https://github.com/Cyan4973/xxHash/issues) can be used for further public discussions on the topic.
+
+
+Version changes
+--------------------
+v0.7.3: Minor fixes
+v0.1.1: added a note on rationale for selection of constants
+v0.1.0: initial release
diff --git a/fluent-bit/lib/cfl/lib/xxhash/libxxhash.pc.in b/fluent-bit/lib/cfl/lib/xxhash/libxxhash.pc.in
new file mode 100644
index 00000000..28c16448
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/libxxhash.pc.in
@@ -0,0 +1,15 @@
+#   xxHash - Extremely fast hash algorithm
+#   Copyright (C) 2012-2020, Yann Collet, Facebook
+#   BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
+
+prefix=@PREFIX@
+exec_prefix=@EXECPREFIX@
+includedir=${prefix}/@INCLUDEDIR@
+libdir=${exec_prefix}/@LIBDIR@
+
+Name: xxhash
+Description: extremely fast hash algorithm
+URL: http://www.xxhash.com/
+Version: @VERSION@
+Libs: -L${libdir} -lxxhash
+Cflags: -I${includedir}
diff --git a/fluent-bit/lib/cfl/lib/xxhash/xxh3.h b/fluent-bit/lib/cfl/lib/xxhash/xxh3.h
new file mode 100644
index 00000000..f7dc1959
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/xxh3.h
@@ -0,0 +1,55 @@
+/*
+ * xxHash - Extremely Fast Hash algorithm
+ * Development source file for `xxh3`
+ * Copyright (C) 2019-2020 Yann Collet
+ *
+ * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *    * Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *    * Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following disclaimer
+ *      in the documentation and/or other materials provided with the
+ *      distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * You can contact the author at:
+ *   - xxHash homepage: https://www.xxhash.com
+ *   - xxHash source repository: https://github.com/Cyan4973/xxHash
+ */
+
+/*
+ * Note: This file used to host the source code of XXH3_* variants.
+ * during the development period.
+ * The source code is now properly integrated within xxhash.h.
+ *
+ * xxh3.h is no longer useful,
+ * but it is still provided for compatibility with source code
+ * which used to include it directly.
+ *
+ * Programs are now highly discouraged to include xxh3.h.
+ * Include `xxhash.h` instead, which is the officially supported interface.
+ *
+ * In the future, xxh3.h will start to generate warnings, then errors,
+ * then it will be removed from source package and from include directory.
+ */
+
+/* Simulate the same impact as including the old xxh3.h source file */
+
+#define XXH_INLINE_ALL
+#include "xxhash.h"
diff --git a/fluent-bit/lib/cfl/lib/xxhash/xxh_x86dispatch.c b/fluent-bit/lib/cfl/lib/xxhash/xxh_x86dispatch.c
new file mode 100644
index 00000000..399bad90
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/xxh_x86dispatch.c
@@ -0,0 +1,770 @@
+/*
+ * xxHash - Extremely Fast Hash algorithm
+ * Copyright (C) 2020 Yann Collet
+ *
+ * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *    * Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *    * Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following disclaimer
+ *      in the documentation and/or other materials provided with the
+ *      distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * You can contact the author at:
+ *   - xxHash homepage: https://www.xxhash.com
+ *   - xxHash source repository: https://github.com/Cyan4973/xxHash
+ */
+
+
+/*!
+ * @file xxh_x86dispatch.c
+ *
+ * Automatic dispatcher code for the @ref xxh3_family on x86-based targets.
+ *
+ * Optional add-on.
+ *
+ * **Compile this file with the default flags for your target.** Do not compile
+ * with flags like `-mavx*`, `-march=native`, or `/arch:AVX*`, there will be
+ * an error. See @ref XXH_X86DISPATCH_ALLOW_AVX for details.
+ *
+ * @defgroup dispatch x86 Dispatcher
+ * @{
+ */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+#if !(defined(__x86_64__) || defined(__i386__) || defined(_M_IX86) || defined(_M_X64))
+#  error "Dispatching is currently only supported on x86 and x86_64."
+#endif
+
+/*!
+ * @def XXH_X86DISPATCH_ALLOW_AVX
+ * @brief Disables the AVX sanity check.
+ *
+ * Don't compile xxh_x86dispatch.c with options like `-mavx*`, `-march=native`,
+ * or `/arch:AVX*`. It is intended to be compiled for the minimum target, and
+ * it selectively enables SSE2, AVX2, and AVX512 when it is needed.
+ *
+ * Using this option _globally_ allows this feature, and therefore makes it
+ * undefined behavior to execute on any CPU without said feature.
+ *
+ * Even if the source code isn't directly using AVX intrinsics in a function,
+ * the compiler can still generate AVX code from autovectorization and by
+ * "upgrading" SSE2 intrinsics to use the VEX prefixes (a.k.a. AVX128).
+ *
+ * Use the same flags that you use to compile the rest of the program; this
+ * file will safely generate SSE2, AVX2, and AVX512 without these flags.
+ *
+ * Define XXH_X86DISPATCH_ALLOW_AVX to ignore this check, and feel free to open
+ * an issue if there is a target in the future where AVX is a default feature.
+ */
+#ifdef XXH_DOXYGEN
+#  define XXH_X86DISPATCH_ALLOW_AVX
+#endif
+
+#if defined(__AVX__) && !defined(XXH_X86DISPATCH_ALLOW_AVX)
+#  error "Do not compile xxh_x86dispatch.c with AVX enabled! See the comment above."
+#endif
+
+#ifdef __has_include
+#  define XXH_HAS_INCLUDE(header) __has_include(header)
+#else
+#  define XXH_HAS_INCLUDE(header) 0
+#endif
+
+/*!
+ * @def XXH_DISPATCH_SCALAR
+ * @brief Enables/dispatching the scalar code path.
+ *
+ * If this is defined to 0, SSE2 support is assumed. This reduces code size
+ * when the scalar path is not needed.
+ *
+ * This is automatically defined to 0 when...
+ *   - SSE2 support is enabled in the compiler
+ *   - Targeting x86_64
+ *   - Targeting Android x86
+ *   - Targeting macOS
+ */
+#ifndef XXH_DISPATCH_SCALAR
+#  if defined(__SSE2__) || (defined(_M_IX86_FP) && _M_IX86_FP >= 2) /* SSE2 on by default */ \
+     || defined(__x86_64__) || defined(_M_X64) /* x86_64 */ \
+     || defined(__ANDROID__) || defined(__APPLEv__) /* Android or macOS */
+#     define XXH_DISPATCH_SCALAR 0 /* disable */
+#  else
+#     define XXH_DISPATCH_SCALAR 1
+#  endif
+#endif
+/*!
+ * @def XXH_DISPATCH_AVX2
+ * @brief Enables/disables dispatching for AVX2.
+ *
+ * This is automatically detected if it is not defined.
+ *  - GCC 4.7 and later are known to support AVX2, but >4.9 is required for
+ *    to get the AVX2 intrinsics and typedefs without -mavx -mavx2.
+ *  - Visual Studio 2013 Update 2 and later are known to support AVX2.
+ *  - The GCC/Clang internal header `<avx2intrin.h>` is detected. While this is
+ *    not allowed to be included directly, it still appears in the builtin
+ *    include path and is detectable with `__has_include`.
+ *
+ * @see XXH_AVX2
+ */
+#ifndef XXH_DISPATCH_AVX2
+#  if (defined(__GNUC__) && (__GNUC__ > 4)) /* GCC 5.0+ */ \
+   || (defined(_MSC_VER) && _MSC_VER >= 1900) /* VS 2015+ */ \
+   || (defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 180030501) /* VS 2013 Update 2 */ \
+   || XXH_HAS_INCLUDE(<avx2intrin.h>) /* GCC/Clang internal header */
+#    define XXH_DISPATCH_AVX2 1   /* enable dispatch towards AVX2 */
+#  else
+#    define XXH_DISPATCH_AVX2 0
+#  endif
+#endif /* XXH_DISPATCH_AVX2 */
+
+/*!
+ * @def XXH_DISPATCH_AVX512
+ * @brief Enables/disables dispatching for AVX512.
+ *
+ * Automatically detected if one of the following conditions is met:
+ *  - GCC 4.9 and later are known to support AVX512.
+ *  - Visual Studio 2017  and later are known to support AVX2.
+ *  - The GCC/Clang internal header `<avx512fintrin.h>` is detected. While this
+ *    is not allowed to be included directly, it still appears in the builtin
+ *    include path and is detectable with `__has_include`.
+ *
+ * @see XXH_AVX512
+ */
+#ifndef XXH_DISPATCH_AVX512
+#  if (defined(__GNUC__) \
+       && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 9))) /* GCC 4.9+ */ \
+   || (defined(_MSC_VER) && _MSC_VER >= 1910) /* VS 2017+ */ \
+   || XXH_HAS_INCLUDE(<avx512fintrin.h>) /* GCC/Clang internal header */
+#    define XXH_DISPATCH_AVX512 1   /* enable dispatch towards AVX512 */
+#  else
+#    define XXH_DISPATCH_AVX512 0
+#  endif
+#endif /* XXH_DISPATCH_AVX512 */
+
+/*!
+ * @def XXH_TARGET_SSE2
+ * @brief Allows a function to be compiled with SSE2 intrinsics.
+ *
+ * Uses `__attribute__((__target__("sse2")))` on GCC to allow SSE2 to be used
+ * even with `-mno-sse2`.
+ *
+ * @def XXH_TARGET_AVX2
+ * @brief Like @ref XXH_TARGET_SSE2, but for AVX2.
+ *
+ * @def XXH_TARGET_AVX512
+ * @brief Like @ref XXH_TARGET_SSE2, but for AVX512.
+ */
+#if defined(__GNUC__)
+#  include <emmintrin.h> /* SSE2 */
+#  if XXH_DISPATCH_AVX2 || XXH_DISPATCH_AVX512
+#    include <immintrin.h> /* AVX2, AVX512F */
+#  endif
+#  define XXH_TARGET_SSE2 __attribute__((__target__("sse2")))
+#  define XXH_TARGET_AVX2 __attribute__((__target__("avx2")))
+#  define XXH_TARGET_AVX512 __attribute__((__target__("avx512f")))
+#elif defined(_MSC_VER)
+#  include <intrin.h>
+#  define XXH_TARGET_SSE2
+#  define XXH_TARGET_AVX2
+#  define XXH_TARGET_AVX512
+#else
+#  error "Dispatching is currently not supported for your compiler."
+#endif
+
+#ifdef XXH_DISPATCH_DEBUG
+/* debug logging */
+#  include <stdio.h>
+#  define XXH_debugPrint(str) { fprintf(stderr, "DEBUG: xxHash dispatch: %s \n", str); fflush(NULL); }
+#else
+#  define XXH_debugPrint(str) ((void)0)
+#  undef NDEBUG /* avoid redefinition */
+#  define NDEBUG
+#endif
+#include <assert.h>
+
+#define XXH_INLINE_ALL
+#define XXH_X86DISPATCH
+#include "xxhash.h"
+
+/*
+ * Support both AT&T and Intel dialects
+ *
+ * GCC doesn't convert AT&T syntax to Intel syntax, and will error out if
+ * compiled with -masm=intel. Instead, it supports dialect switching with
+ * curly braces: { AT&T syntax | Intel syntax }
+ *
+ * Clang's integrated assembler automatically converts AT&T syntax to Intel if
+ * needed, making the dialect switching useless (it isn't even supported).
+ *
+ * Note: Comments are written in the inline assembly itself.
+ */
+#ifdef __clang__
+#  define XXH_I_ATT(intel, att) att "\n\t"
+#else
+#  define XXH_I_ATT(intel, att) "{" att "|" intel "}\n\t"
+#endif
+
+/*!
+ * @internal
+ * @brief Runs CPUID.
+ *
+ * @param eax , ecx The parameters to pass to CPUID, %eax and %ecx respectively.
+ * @param abcd The array to store the result in, `{ eax, ebx, ecx, edx }`
+ */
+static void XXH_cpuid(xxh_u32 eax, xxh_u32 ecx, xxh_u32* abcd)
+{
+#if defined(_MSC_VER)
+    __cpuidex(abcd, eax, ecx);
+#else
+    xxh_u32 ebx, edx;
+# if defined(__i386__) && defined(__PIC__)
+    __asm__(
+        "# Call CPUID\n\t"
+        "#\n\t"
+        "# On 32-bit x86 with PIC enabled, we are not allowed to overwrite\n\t"
+        "# EBX, so we use EDI instead.\n\t"
+        XXH_I_ATT("mov     edi, ebx",   "movl    %%ebx, %%edi")
+        XXH_I_ATT("cpuid",              "cpuid"               )
+        XXH_I_ATT("xchg    edi, ebx",   "xchgl   %%ebx, %%edi")
+        : "=D" (ebx),
+# else
+    __asm__(
+        "# Call CPUID\n\t"
+        XXH_I_ATT("cpuid",              "cpuid")
+        : "=b" (ebx),
+# endif
+              "+a" (eax), "+c" (ecx), "=d" (edx));
+    abcd[0] = eax;
+    abcd[1] = ebx;
+    abcd[2] = ecx;
+    abcd[3] = edx;
+#endif
+}
+
+/*
+ * Modified version of Intel's guide
+ * https://software.intel.com/en-us/articles/how-to-detect-new-instruction-support-in-the-4th-generation-intel-core-processor-family
+ */
+
+#if XXH_DISPATCH_AVX2 || XXH_DISPATCH_AVX512
+/*!
+ * @internal
+ * @brief Runs `XGETBV`.
+ *
+ * While the CPU may support AVX2, the operating system might not properly save
+ * the full YMM/ZMM registers.
+ *
+ * xgetbv is used for detecting this: Any compliant operating system will define
+ * a set of flags in the xcr0 register indicating how it saves the AVX registers.
+ *
+ * You can manually disable this flag on Windows by running, as admin:
+ *
+ *   bcdedit.exe /set xsavedisable 1
+ *
+ * and rebooting. Run the same command with 0 to re-enable it.
+ */
+static xxh_u64 XXH_xgetbv(void)
+{
+#if defined(_MSC_VER)
+    return _xgetbv(0);  /* min VS2010 SP1 compiler is required */
+#else
+    xxh_u32 xcr0_lo, xcr0_hi;
+    __asm__(
+        "# Call XGETBV\n\t"
+        "#\n\t"
+        "# Older assemblers (e.g. macOS's ancient GAS version) don't support\n\t"
+        "# the XGETBV opcode, so we encode it by hand instead.\n\t"
+        "# See <https://github.com/asmjit/asmjit/issues/78> for details.\n\t"
+        ".byte   0x0f, 0x01, 0xd0\n\t"
+       : "=a" (xcr0_lo), "=d" (xcr0_hi) : "c" (0));
+    return xcr0_lo | ((xxh_u64)xcr0_hi << 32);
+#endif
+}
+#endif
+
+#define XXH_SSE2_CPUID_MASK (1 << 26)
+#define XXH_OSXSAVE_CPUID_MASK ((1 << 26) | (1 << 27))
+#define XXH_AVX2_CPUID_MASK (1 << 5)
+#define XXH_AVX2_XGETBV_MASK ((1 << 2) | (1 << 1))
+#define XXH_AVX512F_CPUID_MASK (1 << 16)
+#define XXH_AVX512F_XGETBV_MASK ((7 << 5) | (1 << 2) | (1 << 1))
+
+/*!
+ * @internal
+ * @brief Returns the best XXH3 implementation.
+ *
+ * Runs various CPUID/XGETBV tests to try and determine the best implementation.
+ *
+ * @ret The best @ref XXH_VECTOR implementation.
+ * @see XXH_VECTOR_TYPES
+ */
+static int XXH_featureTest(void)
+{
+    xxh_u32 abcd[4];
+    xxh_u32 max_leaves;
+    int best = XXH_SCALAR;
+#if XXH_DISPATCH_AVX2 || XXH_DISPATCH_AVX512
+    xxh_u64 xgetbv_val;
+#endif
+#if defined(__GNUC__) && defined(__i386__)
+    xxh_u32 cpuid_supported;
+    __asm__(
+        "# For the sake of ruthless backwards compatibility, check if CPUID\n\t"
+        "# is supported in the EFLAGS on i386.\n\t"
+        "# This is not necessary on x86_64 - CPUID is mandatory.\n\t"
+        "#   The ID flag (bit 21) in the EFLAGS register indicates support\n\t"
+        "#   for the CPUID instruction. If a software procedure can set and\n\t"
+        "#   clear this flag, the processor executing the procedure supports\n\t"
+        "#   the CPUID instruction.\n\t"
+        "#   <https://c9x.me/x86/html/file_module_x86_id_45.html>\n\t"
+        "#\n\t"
+        "# Routine is from <https://wiki.osdev.org/CPUID>.\n\t"
+
+        "# Save EFLAGS\n\t"
+        XXH_I_ATT("pushfd",                           "pushfl"                    )
+        "# Store EFLAGS\n\t"
+        XXH_I_ATT("pushfd",                           "pushfl"                    )
+        "# Invert the ID bit in stored EFLAGS\n\t"
+        XXH_I_ATT("xor     dword ptr[esp], 0x200000", "xorl    $0x200000, (%%esp)")
+        "# Load stored EFLAGS (with ID bit inverted)\n\t"
+        XXH_I_ATT("popfd",                            "popfl"                     )
+        "# Store EFLAGS again (ID bit may or not be inverted)\n\t"
+        XXH_I_ATT("pushfd",                           "pushfl"                    )
+        "# eax = modified EFLAGS (ID bit may or may not be inverted)\n\t"
+        XXH_I_ATT("pop     eax",                      "popl    %%eax"             )
+        "# eax = whichever bits were changed\n\t"
+        XXH_I_ATT("xor     eax, dword ptr[esp]",      "xorl    (%%esp), %%eax"    )
+        "# Restore original EFLAGS\n\t"
+        XXH_I_ATT("popfd",                            "popfl"                     )
+        "# eax = zero if ID bit can't be changed, else non-zero\n\t"
+        XXH_I_ATT("and     eax, 0x200000",            "andl    $0x200000, %%eax"  )
+        : "=a" (cpuid_supported) :: "cc");
+
+    if (XXH_unlikely(!cpuid_supported)) {
+        XXH_debugPrint("CPUID support is not detected!");
+        return best;
+    }
+
+#endif
+    /* Check how many CPUID pages we have */
+    XXH_cpuid(0, 0, abcd);
+    max_leaves = abcd[0];
+
+    /* Shouldn't happen on hardware, but happens on some QEMU configs. */
+    if (XXH_unlikely(max_leaves == 0)) {
+        XXH_debugPrint("Max CPUID leaves == 0!");
+        return best;
+    }
+
+    /* Check for SSE2, OSXSAVE and xgetbv */
+    XXH_cpuid(1, 0, abcd);
+
+    /*
+     * Test for SSE2. The check is redundant on x86_64, but it doesn't hurt.
+     */
+    if (XXH_unlikely((abcd[3] & XXH_SSE2_CPUID_MASK) != XXH_SSE2_CPUID_MASK))
+        return best;
+
+    XXH_debugPrint("SSE2 support detected.");
+
+    best = XXH_SSE2;
+#if XXH_DISPATCH_AVX2 || XXH_DISPATCH_AVX512
+    /* Make sure we have enough leaves */
+    if (XXH_unlikely(max_leaves < 7))
+        return best;
+
+    /* Test for OSXSAVE and XGETBV */
+    if ((abcd[2] & XXH_OSXSAVE_CPUID_MASK) != XXH_OSXSAVE_CPUID_MASK)
+        return best;
+
+    /* CPUID check for AVX features */
+    XXH_cpuid(7, 0, abcd);
+
+    xgetbv_val = XXH_xgetbv();
+#if XXH_DISPATCH_AVX2
+    /* Validate that AVX2 is supported by the CPU */
+    if ((abcd[1] & XXH_AVX2_CPUID_MASK) != XXH_AVX2_CPUID_MASK)
+        return best;
+
+    /* Validate that the OS supports YMM registers */
+    if ((xgetbv_val & XXH_AVX2_XGETBV_MASK) != XXH_AVX2_XGETBV_MASK) {
+        XXH_debugPrint("AVX2 supported by the CPU, but not the OS.");
+        return best;
+    }
+
+    /* AVX2 supported */
+    XXH_debugPrint("AVX2 support detected.");
+    best = XXH_AVX2;
+#endif
+#if XXH_DISPATCH_AVX512
+    /* Check if AVX512F is supported by the CPU */
+    if ((abcd[1] & XXH_AVX512F_CPUID_MASK) != XXH_AVX512F_CPUID_MASK) {
+        XXH_debugPrint("AVX512F not supported by CPU");
+        return best;
+    }
+
+    /* Validate that the OS supports ZMM registers */
+    if ((xgetbv_val & XXH_AVX512F_XGETBV_MASK) != XXH_AVX512F_XGETBV_MASK) {
+        XXH_debugPrint("AVX512F supported by the CPU, but not the OS.");
+        return best;
+    }
+
+    /* AVX512F supported */
+    XXH_debugPrint("AVX512F support detected.");
+    best = XXH_AVX512;
+#endif
+#endif
+    return best;
+}
+
+
+/* ===   Vector implementations   === */
+
+/*!
+ * @internal
+ * @brief Defines the various dispatch functions.
+ *
+ * TODO: Consolidate?
+ *
+ * @param suffix The suffix for the functions, e.g. sse2 or scalar
+ * @param target XXH_TARGET_* or empty.
+ */
+#define XXH_DEFINE_DISPATCH_FUNCS(suffix, target)                             \
+                                                                              \
+/* ===   XXH3, default variants   === */                                      \
+                                                                              \
+XXH_NO_INLINE target XXH64_hash_t                                             \
+XXHL64_default_##suffix(const void* XXH_RESTRICT input, size_t len)           \
+{                                                                             \
+    return XXH3_hashLong_64b_internal(                                        \
+               input, len, XXH3_kSecret, sizeof(XXH3_kSecret),                \
+               XXH3_accumulate_512_##suffix, XXH3_scrambleAcc_##suffix        \
+    );                                                                        \
+}                                                                             \
+                                                                              \
+/* ===   XXH3, Seeded variants   === */                                       \
+                                                                              \
+XXH_NO_INLINE target XXH64_hash_t                                             \
+XXHL64_seed_##suffix(const void* XXH_RESTRICT input, size_t len,              \
+                     XXH64_hash_t seed)                                       \
+{                                                                             \
+    return XXH3_hashLong_64b_withSeed_internal(                               \
+                    input, len, seed, XXH3_accumulate_512_##suffix,           \
+                    XXH3_scrambleAcc_##suffix, XXH3_initCustomSecret_##suffix \
+    );                                                                        \
+}                                                                             \
+                                                                              \
+/* ===   XXH3, Secret variants   === */                                       \
+                                                                              \
+XXH_NO_INLINE target XXH64_hash_t                                             \
+XXHL64_secret_##suffix(const void* XXH_RESTRICT input, size_t len,            \
+                       const void* secret, size_t secretLen)                  \
+{                                                                             \
+    return XXH3_hashLong_64b_internal(                                        \
+                    input, len, secret, secretLen,                            \
+                    XXH3_accumulate_512_##suffix, XXH3_scrambleAcc_##suffix   \
+    );                                                                        \
+}                                                                             \
+                                                                              \
+/* ===   XXH3 update variants   === */                                        \
+                                                                              \
+XXH_NO_INLINE target XXH_errorcode                                            \
+XXH3_update_##suffix(XXH3_state_t* state, const void* input, size_t len)      \
+{                                                                             \
+    return XXH3_update(state, (const xxh_u8*)input, len,                      \
+                    XXH3_accumulate_512_##suffix, XXH3_scrambleAcc_##suffix); \
+}                                                                             \
+                                                                              \
+/* ===   XXH128 default variants   === */                                     \
+                                                                              \
+XXH_NO_INLINE target XXH128_hash_t                                            \
+XXHL128_default_##suffix(const void* XXH_RESTRICT input, size_t len)          \
+{                                                                             \
+    return XXH3_hashLong_128b_internal(                                       \
+                    input, len, XXH3_kSecret, sizeof(XXH3_kSecret),           \
+                    XXH3_accumulate_512_##suffix, XXH3_scrambleAcc_##suffix   \
+    );                                                                        \
+}                                                                             \
+                                                                              \
+/* ===   XXH128 Secret variants   === */                                      \
+                                                                              \
+XXH_NO_INLINE target XXH128_hash_t                                            \
+XXHL128_secret_##suffix(const void* XXH_RESTRICT input, size_t len,           \
+                        const void* XXH_RESTRICT secret, size_t secretLen)    \
+{                                                                             \
+    return XXH3_hashLong_128b_internal(                                       \
+                    input, len, (const xxh_u8*)secret, secretLen,             \
+                    XXH3_accumulate_512_##suffix, XXH3_scrambleAcc_##suffix); \
+}                                                                             \
+                                                                              \
+/* ===   XXH128 Seeded variants   === */                                      \
+                                                                              \
+XXH_NO_INLINE target XXH128_hash_t                                            \
+XXHL128_seed_##suffix(const void* XXH_RESTRICT input, size_t len,             \
+                      XXH64_hash_t seed)                                      \
+{                                                                             \
+    return XXH3_hashLong_128b_withSeed_internal(input, len, seed,             \
+                    XXH3_accumulate_512_##suffix, XXH3_scrambleAcc_##suffix,  \
+                    XXH3_initCustomSecret_##suffix);                          \
+}
+
+/* End XXH_DEFINE_DISPATCH_FUNCS */
+
+#if XXH_DISPATCH_SCALAR
+XXH_DEFINE_DISPATCH_FUNCS(scalar, /* nothing */)
+#endif
+XXH_DEFINE_DISPATCH_FUNCS(sse2, XXH_TARGET_SSE2)
+#if XXH_DISPATCH_AVX2
+XXH_DEFINE_DISPATCH_FUNCS(avx2, XXH_TARGET_AVX2)
+#endif
+#if XXH_DISPATCH_AVX512
+XXH_DEFINE_DISPATCH_FUNCS(avx512, XXH_TARGET_AVX512)
+#endif
+#undef XXH_DEFINE_DISPATCH_FUNCS
+
+/* ====    Dispatchers    ==== */
+
+typedef XXH64_hash_t (*XXH3_dispatchx86_hashLong64_default)(const void* XXH_RESTRICT, size_t);
+
+typedef XXH64_hash_t (*XXH3_dispatchx86_hashLong64_withSeed)(const void* XXH_RESTRICT, size_t, XXH64_hash_t);
+
+typedef XXH64_hash_t (*XXH3_dispatchx86_hashLong64_withSecret)(const void* XXH_RESTRICT, size_t, const void* XXH_RESTRICT, size_t);
+
+typedef XXH_errorcode (*XXH3_dispatchx86_update)(XXH3_state_t*, const void*, size_t);
+
+typedef struct {
+    XXH3_dispatchx86_hashLong64_default    hashLong64_default;
+    XXH3_dispatchx86_hashLong64_withSeed   hashLong64_seed;
+    XXH3_dispatchx86_hashLong64_withSecret hashLong64_secret;
+    XXH3_dispatchx86_update                update;
+} XXH_dispatchFunctions_s;
+
+#define XXH_NB_DISPATCHES 4
+
+/*!
+ * @internal
+ * @brief Table of dispatchers for @ref XXH3_64bits().
+ *
+ * @pre The indices must match @ref XXH_VECTOR_TYPE.
+ */
+static const XXH_dispatchFunctions_s XXH_kDispatch[XXH_NB_DISPATCHES] = {
+#if XXH_DISPATCH_SCALAR
+    /* Scalar */ { XXHL64_default_scalar, XXHL64_seed_scalar, XXHL64_secret_scalar, XXH3_update_scalar },
+#else
+    /* Scalar */ { NULL, NULL, NULL, NULL },
+#endif
+    /* SSE2   */ { XXHL64_default_sse2,   XXHL64_seed_sse2,   XXHL64_secret_sse2,   XXH3_update_sse2 },
+#if XXH_DISPATCH_AVX2
+    /* AVX2   */ { XXHL64_default_avx2,   XXHL64_seed_avx2,   XXHL64_secret_avx2,   XXH3_update_avx2 },
+#else
+    /* AVX2   */ { NULL, NULL, NULL, NULL },
+#endif
+#if XXH_DISPATCH_AVX512
+    /* AVX512 */ { XXHL64_default_avx512, XXHL64_seed_avx512, XXHL64_secret_avx512, XXH3_update_avx512 }
+#else
+    /* AVX512 */ { NULL, NULL, NULL, NULL }
+#endif
+};
+/*!
+ * @internal
+ * @brief The selected dispatch table for @ref XXH3_64bits().
+ */
+static XXH_dispatchFunctions_s XXH_g_dispatch = { NULL, NULL, NULL, NULL };
+
+
+typedef XXH128_hash_t (*XXH3_dispatchx86_hashLong128_default)(const void* XXH_RESTRICT, size_t);
+
+typedef XXH128_hash_t (*XXH3_dispatchx86_hashLong128_withSeed)(const void* XXH_RESTRICT, size_t, XXH64_hash_t);
+
+typedef XXH128_hash_t (*XXH3_dispatchx86_hashLong128_withSecret)(const void* XXH_RESTRICT, size_t, const void* XXH_RESTRICT, size_t);
+
+typedef struct {
+    XXH3_dispatchx86_hashLong128_default    hashLong128_default;
+    XXH3_dispatchx86_hashLong128_withSeed   hashLong128_seed;
+    XXH3_dispatchx86_hashLong128_withSecret hashLong128_secret;
+    XXH3_dispatchx86_update                 update;
+} XXH_dispatch128Functions_s;
+
+
+/*!
+ * @internal
+ * @brief Table of dispatchers for @ref XXH3_128bits().
+ *
+ * @pre The indices must match @ref XXH_VECTOR_TYPE.
+ */
+static const XXH_dispatch128Functions_s XXH_kDispatch128[XXH_NB_DISPATCHES] = {
+#if XXH_DISPATCH_SCALAR
+    /* Scalar */ { XXHL128_default_scalar, XXHL128_seed_scalar, XXHL128_secret_scalar, XXH3_update_scalar },
+#else
+    /* Scalar */ { NULL, NULL, NULL, NULL },
+#endif
+    /* SSE2   */ { XXHL128_default_sse2,   XXHL128_seed_sse2,   XXHL128_secret_sse2,   XXH3_update_sse2 },
+#if XXH_DISPATCH_AVX2
+    /* AVX2   */ { XXHL128_default_avx2,   XXHL128_seed_avx2,   XXHL128_secret_avx2,   XXH3_update_avx2 },
+#else
+    /* AVX2   */ { NULL, NULL, NULL, NULL },
+#endif
+#if XXH_DISPATCH_AVX512
+    /* AVX512 */ { XXHL128_default_avx512, XXHL128_seed_avx512, XXHL128_secret_avx512, XXH3_update_avx512 }
+#else
+    /* AVX512 */ { NULL, NULL, NULL, NULL }
+#endif
+};
+
+/*!
+ * @internal
+ * @brief The selected dispatch table for @ref XXH3_64bits().
+ */
+static XXH_dispatch128Functions_s XXH_g_dispatch128 = { NULL, NULL, NULL, NULL };
+
+/*!
+ * @internal
+ * @brief Runs a CPUID check and sets the correct dispatch tables.
+ */
+static void XXH_setDispatch(void)
+{
+    int vecID = XXH_featureTest();
+    XXH_STATIC_ASSERT(XXH_AVX512 == XXH_NB_DISPATCHES-1);
+    assert(XXH_SCALAR <= vecID && vecID <= XXH_AVX512);
+#if !XXH_DISPATCH_SCALAR
+    assert(vecID != XXH_SCALAR);
+#endif
+#if !XXH_DISPATCH_AVX512
+    assert(vecID != XXH_AVX512);
+#endif
+#if !XXH_DISPATCH_AVX2
+    assert(vecID != XXH_AVX2);
+#endif
+    XXH_g_dispatch = XXH_kDispatch[vecID];
+    XXH_g_dispatch128 = XXH_kDispatch128[vecID];
+}
+
+
+/* ====    XXH3 public functions    ==== */
+
+static XXH64_hash_t
+XXH3_hashLong_64b_defaultSecret_selection(const void* input, size_t len,
+                                          XXH64_hash_t seed64, const xxh_u8* secret, size_t secretLen)
+{
+    (void)seed64; (void)secret; (void)secretLen;
+    if (XXH_g_dispatch.hashLong64_default == NULL) XXH_setDispatch();
+    return XXH_g_dispatch.hashLong64_default(input, len);
+}
+
+XXH64_hash_t XXH3_64bits_dispatch(const void* input, size_t len)
+{
+    return XXH3_64bits_internal(input, len, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_defaultSecret_selection);
+}
+
+static XXH64_hash_t
+XXH3_hashLong_64b_withSeed_selection(const void* input, size_t len,
+                                     XXH64_hash_t seed64, const xxh_u8* secret, size_t secretLen)
+{
+    (void)secret; (void)secretLen;
+    if (XXH_g_dispatch.hashLong64_seed == NULL) XXH_setDispatch();
+    return XXH_g_dispatch.hashLong64_seed(input, len, seed64);
+}
+
+XXH64_hash_t XXH3_64bits_withSeed_dispatch(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed_selection);
+}
+
+static XXH64_hash_t
+XXH3_hashLong_64b_withSecret_selection(const void* input, size_t len,
+                                       XXH64_hash_t seed64, const xxh_u8* secret, size_t secretLen)
+{
+    (void)seed64;
+    if (XXH_g_dispatch.hashLong64_secret == NULL) XXH_setDispatch();
+    return XXH_g_dispatch.hashLong64_secret(input, len, secret, secretLen);
+}
+
+XXH64_hash_t XXH3_64bits_withSecret_dispatch(const void* input, size_t len, const void* secret, size_t secretLen)
+{
+    return XXH3_64bits_internal(input, len, 0, secret, secretLen, XXH3_hashLong_64b_withSecret_selection);
+}
+
+XXH_errorcode
+XXH3_64bits_update_dispatch(XXH3_state_t* state, const void* input, size_t len)
+{
+    if (XXH_g_dispatch.update == NULL) XXH_setDispatch();
+    return XXH_g_dispatch.update(state, (const xxh_u8*)input, len);
+}
+
+
+/* ====    XXH128 public functions    ==== */
+
+static XXH128_hash_t
+XXH3_hashLong_128b_defaultSecret_selection(const void* input, size_t len,
+                                           XXH64_hash_t seed64, const void* secret, size_t secretLen)
+{
+    (void)seed64; (void)secret; (void)secretLen;
+    if (XXH_g_dispatch128.hashLong128_default == NULL) XXH_setDispatch();
+    return XXH_g_dispatch128.hashLong128_default(input, len);
+}
+
+XXH128_hash_t XXH3_128bits_dispatch(const void* input, size_t len)
+{
+    return XXH3_128bits_internal(input, len, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_128b_defaultSecret_selection);
+}
+
+static XXH128_hash_t
+XXH3_hashLong_128b_withSeed_selection(const void* input, size_t len,
+                                     XXH64_hash_t seed64, const void* secret, size_t secretLen)
+{
+    (void)secret; (void)secretLen;
+    if (XXH_g_dispatch128.hashLong128_seed == NULL) XXH_setDispatch();
+    return XXH_g_dispatch128.hashLong128_seed(input, len, seed64);
+}
+
+XXH128_hash_t XXH3_128bits_withSeed_dispatch(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_128b_withSeed_selection);
+}
+
+static XXH128_hash_t
+XXH3_hashLong_128b_withSecret_selection(const void* input, size_t len,
+                                        XXH64_hash_t seed64, const void* secret, size_t secretLen)
+{
+    (void)seed64;
+    if (XXH_g_dispatch128.hashLong128_secret == NULL) XXH_setDispatch();
+    return XXH_g_dispatch128.hashLong128_secret(input, len, secret, secretLen);
+}
+
+XXH128_hash_t XXH3_128bits_withSecret_dispatch(const void* input, size_t len, const void* secret, size_t secretLen)
+{
+    return XXH3_128bits_internal(input, len, 0, secret, secretLen, XXH3_hashLong_128b_withSecret_selection);
+}
+
+XXH_errorcode
+XXH3_128bits_update_dispatch(XXH3_state_t* state, const void* input, size_t len)
+{
+    if (XXH_g_dispatch128.update == NULL) XXH_setDispatch();
+    return XXH_g_dispatch128.update(state, (const xxh_u8*)input, len);
+}
+
+#if defined (__cplusplus)
+}
+#endif
+/*! @} */
diff --git a/fluent-bit/lib/cfl/lib/xxhash/xxh_x86dispatch.h b/fluent-bit/lib/cfl/lib/xxhash/xxh_x86dispatch.h
new file mode 100644
index 00000000..6bc17bcb
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/xxh_x86dispatch.h
@@ -0,0 +1,86 @@
+/*
+ * xxHash - XXH3 Dispatcher for x86-based targets
+ * Copyright (C) 2020 Yann Collet
+ *
+ * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *    * Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *    * Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following disclaimer
+ *      in the documentation and/or other materials provided with the
+ *      distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * You can contact the author at:
+ *   - xxHash homepage: https://www.xxhash.com
+ *   - xxHash source repository: https://github.com/Cyan4973/xxHash
+ */
+
+#ifndef XXH_X86DISPATCH_H_13563687684
+#define XXH_X86DISPATCH_H_13563687684
+
+#include "xxhash.h"  /* XXH64_hash_t, XXH3_state_t */
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+XXH_PUBLIC_API XXH64_hash_t  XXH3_64bits_dispatch(const void* input, size_t len);
+XXH_PUBLIC_API XXH64_hash_t  XXH3_64bits_withSeed_dispatch(const void* input, size_t len, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH64_hash_t  XXH3_64bits_withSecret_dispatch(const void* input, size_t len, const void* secret, size_t secretLen);
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update_dispatch(XXH3_state_t* state, const void* input, size_t len);
+
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_dispatch(const void* input, size_t len);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed_dispatch(const void* input, size_t len, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret_dispatch(const void* input, size_t len, const void* secret, size_t secretLen);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update_dispatch(XXH3_state_t* state, const void* input, size_t len);
+
+#if defined (__cplusplus)
+}
+#endif
+
+
+/* automatic replacement of XXH3 functions.
+ * can be disabled by setting XXH_DISPATCH_DISABLE_REPLACE */
+#ifndef XXH_DISPATCH_DISABLE_REPLACE
+
+# undef  XXH3_64bits
+# define XXH3_64bits XXH3_64bits_dispatch
+# undef  XXH3_64bits_withSeed
+# define XXH3_64bits_withSeed XXH3_64bits_withSeed_dispatch
+# undef  XXH3_64bits_withSecret
+# define XXH3_64bits_withSecret XXH3_64bits_withSecret_dispatch
+# undef  XXH3_64bits_update
+# define XXH3_64bits_update XXH3_64bits_update_dispatch
+
+# undef  XXH128
+# define XXH128 XXH3_128bits_withSeed_dispatch
+# define XXH3_128bits XXH3_128bits_dispatch
+# undef  XXH3_128bits
+# define XXH3_128bits XXH3_128bits_dispatch
+# undef  XXH3_128bits_withSeed
+# define XXH3_128bits_withSeed XXH3_128bits_withSeed_dispatch
+# undef  XXH3_128bits_withSecret
+# define XXH3_128bits_withSecret XXH3_128bits_withSecret_dispatch
+# undef  XXH3_128bits_update
+# define XXH3_128bits_update XXH3_128bits_update_dispatch
+
+#endif /* XXH_DISPATCH_DISABLE_REPLACE */
+
+#endif /* XXH_X86DISPATCH_H_13563687684 */
diff --git a/fluent-bit/lib/cfl/lib/xxhash/xxhash.c b/fluent-bit/lib/cfl/lib/xxhash/xxhash.c
new file mode 100644
index 00000000..0fae88c5
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/xxhash.c
@@ -0,0 +1,43 @@
+/*
+ * xxHash - Extremely Fast Hash algorithm
+ * Copyright (C) 2012-2020 Yann Collet
+ *
+ * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *    * Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *    * Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following disclaimer
+ *      in the documentation and/or other materials provided with the
+ *      distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * You can contact the author at:
+ *   - xxHash homepage: https://www.xxhash.com
+ *   - xxHash source repository: https://github.com/Cyan4973/xxHash
+ */
+
+
+/*
+ * xxhash.c instantiates functions defined in xxhash.h
+ */
+
+#define XXH_STATIC_LINKING_ONLY   /* access advanced declarations */
+#define XXH_IMPLEMENTATION   /* access definitions */
+
+#include "xxhash.h"
diff --git a/fluent-bit/lib/cfl/lib/xxhash/xxhash.h b/fluent-bit/lib/cfl/lib/xxhash/xxhash.h
new file mode 100644
index 00000000..511c4d12
--- /dev/null
+++ b/fluent-bit/lib/cfl/lib/xxhash/xxhash.h
@@ -0,0 +1,5579 @@
+/*
+ * xxHash - Extremely Fast Hash algorithm
+ * Header File
+ * Copyright (C) 2012-2020 Yann Collet
+ *
+ * BSD 2-Clause License (https://www.opensource.org/licenses/bsd-license.php)
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *
+ *    * Redistributions of source code must retain the above copyright
+ *      notice, this list of conditions and the following disclaimer.
+ *    * Redistributions in binary form must reproduce the above
+ *      copyright notice, this list of conditions and the following disclaimer
+ *      in the documentation and/or other materials provided with the
+ *      distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * You can contact the author at:
+ *   - xxHash homepage: https://www.xxhash.com
+ *   - xxHash source repository: https://github.com/Cyan4973/xxHash
+ */
+/*!
+ * @mainpage xxHash
+ *
+ * @file xxhash.h
+ * xxHash prototypes and implementation
+ */
+/* TODO: update */
+/* Notice extracted from xxHash homepage:
+
+xxHash is an extremely fast hash algorithm, running at RAM speed limits.
+It also successfully passes all tests from the SMHasher suite.
+
+Comparison (single thread, Windows Seven 32 bits, using SMHasher on a Core 2 Duo @3GHz)
+
+Name            Speed       Q.Score   Author
+xxHash          5.4 GB/s     10
+CrapWow         3.2 GB/s      2       Andrew
+MurmurHash 3a   2.7 GB/s     10       Austin Appleby
+SpookyHash      2.0 GB/s     10       Bob Jenkins
+SBox            1.4 GB/s      9       Bret Mulvey
+Lookup3         1.2 GB/s      9       Bob Jenkins
+SuperFastHash   1.2 GB/s      1       Paul Hsieh
+CityHash64      1.05 GB/s    10       Pike & Alakuijala
+FNV             0.55 GB/s     5       Fowler, Noll, Vo
+CRC32           0.43 GB/s     9
+MD5-32          0.33 GB/s    10       Ronald L. Rivest
+SHA1-32         0.28 GB/s    10
+
+Q.Score is a measure of quality of the hash function.
+It depends on successfully passing SMHasher test set.
+10 is a perfect score.
+
+Note: SMHasher's CRC32 implementation is not the fastest one.
+Other speed-oriented implementations can be faster,
+especially in combination with PCLMUL instruction:
+https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html?showComment=1552696407071#c3490092340461170735
+
+A 64-bit version, named XXH64, is available since r35.
+It offers much better speed, but for 64-bit applications only.
+Name     Speed on 64 bits    Speed on 32 bits
+XXH64       13.8 GB/s            1.9 GB/s
+XXH32        6.8 GB/s            6.0 GB/s
+*/
+
+#if defined (__cplusplus)
+extern "C" {
+#endif
+
+/* ****************************
+ *  INLINE mode
+ ******************************/
+/*!
+ * XXH_INLINE_ALL (and XXH_PRIVATE_API)
+ * Use these build macros to inline xxhash into the target unit.
+ * Inlining improves performance on small inputs, especially when the length is
+ * expressed as a compile-time constant:
+ *
+ *      https://fastcompression.blogspot.com/2018/03/xxhash-for-small-keys-impressive-power.html
+ *
+ * It also keeps xxHash symbols private to the unit, so they are not exported.
+ *
+ * Usage:
+ *     #define XXH_INLINE_ALL
+ *     #include "xxhash.h"
+ *
+ * Do not compile and link xxhash.o as a separate object, as it is not useful.
+ */
+#if (defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)) \
+    && !defined(XXH_INLINE_ALL_31684351384)
+   /* this section should be traversed only once */
+#  define XXH_INLINE_ALL_31684351384
+   /* give access to the advanced API, required to compile implementations */
+#  undef XXH_STATIC_LINKING_ONLY   /* avoid macro redef */
+#  define XXH_STATIC_LINKING_ONLY
+   /* make all functions private */
+#  undef XXH_PUBLIC_API
+#  if defined(__GNUC__)
+#    define XXH_PUBLIC_API static __inline __attribute__((unused))
+#  elif defined (__cplusplus) || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */)
+#    define XXH_PUBLIC_API static inline
+#  elif defined(_MSC_VER)
+#    define XXH_PUBLIC_API static __inline
+#  else
+     /* note: this version may generate warnings for unused static functions */
+#    define XXH_PUBLIC_API static
+#  endif
+
+   /*
+    * This part deals with the special case where a unit wants to inline xxHash,
+    * but "xxhash.h" has previously been included without XXH_INLINE_ALL,
+    * such as part of some previously included *.h header file.
+    * Without further action, the new include would just be ignored,
+    * and functions would effectively _not_ be inlined (silent failure).
+    * The following macros solve this situation by prefixing all inlined names,
+    * avoiding naming collision with previous inclusions.
+    */
+   /* Before that, we unconditionally #undef all symbols,
+    * in case they were already defined with XXH_NAMESPACE.
+    * They will then be redefined for XXH_INLINE_ALL
+    */
+#  undef XXH_versionNumber
+    /* XXH32 */
+#  undef XXH32
+#  undef XXH32_createState
+#  undef XXH32_freeState
+#  undef XXH32_reset
+#  undef XXH32_update
+#  undef XXH32_digest
+#  undef XXH32_copyState
+#  undef XXH32_canonicalFromHash
+#  undef XXH32_hashFromCanonical
+    /* XXH64 */
+#  undef XXH64
+#  undef XXH64_createState
+#  undef XXH64_freeState
+#  undef XXH64_reset
+#  undef XXH64_update
+#  undef XXH64_digest
+#  undef XXH64_copyState
+#  undef XXH64_canonicalFromHash
+#  undef XXH64_hashFromCanonical
+    /* XXH3_64bits */
+#  undef XXH3_64bits
+#  undef XXH3_64bits_withSecret
+#  undef XXH3_64bits_withSeed
+#  undef XXH3_64bits_withSecretandSeed
+#  undef XXH3_createState
+#  undef XXH3_freeState
+#  undef XXH3_copyState
+#  undef XXH3_64bits_reset
+#  undef XXH3_64bits_reset_withSeed
+#  undef XXH3_64bits_reset_withSecret
+#  undef XXH3_64bits_update
+#  undef XXH3_64bits_digest
+#  undef XXH3_generateSecret
+    /* XXH3_128bits */
+#  undef XXH128
+#  undef XXH3_128bits
+#  undef XXH3_128bits_withSeed
+#  undef XXH3_128bits_withSecret
+#  undef XXH3_128bits_reset
+#  undef XXH3_128bits_reset_withSeed
+#  undef XXH3_128bits_reset_withSecret
+#  undef XXH3_128bits_reset_withSecretandSeed
+#  undef XXH3_128bits_update
+#  undef XXH3_128bits_digest
+#  undef XXH128_isEqual
+#  undef XXH128_cmp
+#  undef XXH128_canonicalFromHash
+#  undef XXH128_hashFromCanonical
+    /* Finally, free the namespace itself */
+#  undef XXH_NAMESPACE
+
+    /* employ the namespace for XXH_INLINE_ALL */
+#  define XXH_NAMESPACE XXH_INLINE_
+   /*
+    * Some identifiers (enums, type names) are not symbols,
+    * but they must nonetheless be renamed to avoid redeclaration.
+    * Alternative solution: do not redeclare them.
+    * However, this requires some #ifdefs, and has a more dispersed impact.
+    * Meanwhile, renaming can be achieved in a single place.
+    */
+#  define XXH_IPREF(Id)   XXH_NAMESPACE ## Id
+#  define XXH_OK XXH_IPREF(XXH_OK)
+#  define XXH_ERROR XXH_IPREF(XXH_ERROR)
+#  define XXH_errorcode XXH_IPREF(XXH_errorcode)
+#  define XXH32_canonical_t  XXH_IPREF(XXH32_canonical_t)
+#  define XXH64_canonical_t  XXH_IPREF(XXH64_canonical_t)
+#  define XXH128_canonical_t XXH_IPREF(XXH128_canonical_t)
+#  define XXH32_state_s XXH_IPREF(XXH32_state_s)
+#  define XXH32_state_t XXH_IPREF(XXH32_state_t)
+#  define XXH64_state_s XXH_IPREF(XXH64_state_s)
+#  define XXH64_state_t XXH_IPREF(XXH64_state_t)
+#  define XXH3_state_s  XXH_IPREF(XXH3_state_s)
+#  define XXH3_state_t  XXH_IPREF(XXH3_state_t)
+#  define XXH128_hash_t XXH_IPREF(XXH128_hash_t)
+   /* Ensure the header is parsed again, even if it was previously included */
+#  undef XXHASH_H_5627135585666179
+#  undef XXHASH_H_STATIC_13879238742
+#endif /* XXH_INLINE_ALL || XXH_PRIVATE_API */
+
+
+
+/* ****************************************************************
+ *  Stable API
+ *****************************************************************/
+#ifndef XXHASH_H_5627135585666179
+#define XXHASH_H_5627135585666179 1
+
+
+/*!
+ * @defgroup public Public API
+ * Contains details on the public xxHash functions.
+ * @{
+ */
+/* specific declaration modes for Windows */
+#if !defined(XXH_INLINE_ALL) && !defined(XXH_PRIVATE_API)
+#  if defined(WIN32) && defined(_MSC_VER) && (defined(XXH_IMPORT) || defined(XXH_EXPORT))
+#    ifdef XXH_EXPORT
+#      define XXH_PUBLIC_API __declspec(dllexport)
+#    elif XXH_IMPORT
+#      define XXH_PUBLIC_API __declspec(dllimport)
+#    endif
+#  else
+#    define XXH_PUBLIC_API   /* do nothing */
+#  endif
+#endif
+
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Emulate a namespace by transparently prefixing all symbols.
+ *
+ * If you want to include _and expose_ xxHash functions from within your own
+ * library, but also want to avoid symbol collisions with other libraries which
+ * may also include xxHash, you can use XXH_NAMESPACE to automatically prefix
+ * any public symbol from xxhash library with the value of XXH_NAMESPACE
+ * (therefore, avoid empty or numeric values).
+ *
+ * Note that no change is required within the calling program as long as it
+ * includes `xxhash.h`: Regular symbol names will be automatically translated
+ * by this header.
+ */
+#  define XXH_NAMESPACE /* YOUR NAME HERE */
+#  undef XXH_NAMESPACE
+#endif
+
+#ifdef XXH_NAMESPACE
+#  define XXH_CAT(A,B) A##B
+#  define XXH_NAME2(A,B) XXH_CAT(A,B)
+#  define XXH_versionNumber XXH_NAME2(XXH_NAMESPACE, XXH_versionNumber)
+/* XXH32 */
+#  define XXH32 XXH_NAME2(XXH_NAMESPACE, XXH32)
+#  define XXH32_createState XXH_NAME2(XXH_NAMESPACE, XXH32_createState)
+#  define XXH32_freeState XXH_NAME2(XXH_NAMESPACE, XXH32_freeState)
+#  define XXH32_reset XXH_NAME2(XXH_NAMESPACE, XXH32_reset)
+#  define XXH32_update XXH_NAME2(XXH_NAMESPACE, XXH32_update)
+#  define XXH32_digest XXH_NAME2(XXH_NAMESPACE, XXH32_digest)
+#  define XXH32_copyState XXH_NAME2(XXH_NAMESPACE, XXH32_copyState)
+#  define XXH32_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH32_canonicalFromHash)
+#  define XXH32_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH32_hashFromCanonical)
+/* XXH64 */
+#  define XXH64 XXH_NAME2(XXH_NAMESPACE, XXH64)
+#  define XXH64_createState XXH_NAME2(XXH_NAMESPACE, XXH64_createState)
+#  define XXH64_freeState XXH_NAME2(XXH_NAMESPACE, XXH64_freeState)
+#  define XXH64_reset XXH_NAME2(XXH_NAMESPACE, XXH64_reset)
+#  define XXH64_update XXH_NAME2(XXH_NAMESPACE, XXH64_update)
+#  define XXH64_digest XXH_NAME2(XXH_NAMESPACE, XXH64_digest)
+#  define XXH64_copyState XXH_NAME2(XXH_NAMESPACE, XXH64_copyState)
+#  define XXH64_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH64_canonicalFromHash)
+#  define XXH64_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH64_hashFromCanonical)
+/* XXH3_64bits */
+#  define XXH3_64bits XXH_NAME2(XXH_NAMESPACE, XXH3_64bits)
+#  define XXH3_64bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecret)
+#  define XXH3_64bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSeed)
+#  define XXH3_64bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_withSecretandSeed)
+#  define XXH3_createState XXH_NAME2(XXH_NAMESPACE, XXH3_createState)
+#  define XXH3_freeState XXH_NAME2(XXH_NAMESPACE, XXH3_freeState)
+#  define XXH3_copyState XXH_NAME2(XXH_NAMESPACE, XXH3_copyState)
+#  define XXH3_64bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset)
+#  define XXH3_64bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSeed)
+#  define XXH3_64bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecret)
+#  define XXH3_64bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_reset_withSecretandSeed)
+#  define XXH3_64bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_update)
+#  define XXH3_64bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_64bits_digest)
+#  define XXH3_generateSecret XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret)
+#  define XXH3_generateSecret_fromSeed XXH_NAME2(XXH_NAMESPACE, XXH3_generateSecret_fromSeed)
+/* XXH3_128bits */
+#  define XXH128 XXH_NAME2(XXH_NAMESPACE, XXH128)
+#  define XXH3_128bits XXH_NAME2(XXH_NAMESPACE, XXH3_128bits)
+#  define XXH3_128bits_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSeed)
+#  define XXH3_128bits_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecret)
+#  define XXH3_128bits_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_withSecretandSeed)
+#  define XXH3_128bits_reset XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset)
+#  define XXH3_128bits_reset_withSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSeed)
+#  define XXH3_128bits_reset_withSecret XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecret)
+#  define XXH3_128bits_reset_withSecretandSeed XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_reset_withSecretandSeed)
+#  define XXH3_128bits_update XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_update)
+#  define XXH3_128bits_digest XXH_NAME2(XXH_NAMESPACE, XXH3_128bits_digest)
+#  define XXH128_isEqual XXH_NAME2(XXH_NAMESPACE, XXH128_isEqual)
+#  define XXH128_cmp     XXH_NAME2(XXH_NAMESPACE, XXH128_cmp)
+#  define XXH128_canonicalFromHash XXH_NAME2(XXH_NAMESPACE, XXH128_canonicalFromHash)
+#  define XXH128_hashFromCanonical XXH_NAME2(XXH_NAMESPACE, XXH128_hashFromCanonical)
+#endif
+
+
+/* *************************************
+*  Version
+***************************************/
+#define XXH_VERSION_MAJOR    0
+#define XXH_VERSION_MINOR    8
+#define XXH_VERSION_RELEASE  1
+#define XXH_VERSION_NUMBER  (XXH_VERSION_MAJOR *100*100 + XXH_VERSION_MINOR *100 + XXH_VERSION_RELEASE)
+
+/*!
+ * @brief Obtains the xxHash version.
+ *
+ * This is mostly useful when xxHash is compiled as a shared library,
+ * since the returned value comes from the library, as opposed to header file.
+ *
+ * @return `XXH_VERSION_NUMBER` of the invoked library.
+ */
+XXH_PUBLIC_API unsigned XXH_versionNumber (void);
+
+
+/* ****************************
+*  Common basic types
+******************************/
+#include <stddef.h>   /* size_t */
+typedef enum { XXH_OK=0, XXH_ERROR } XXH_errorcode;
+
+
+/*-**********************************************************************
+*  32-bit hash
+************************************************************************/
+#if defined(XXH_DOXYGEN) /* Don't show <stdint.h> include */
+/*!
+ * @brief An unsigned 32-bit integer.
+ *
+ * Not necessarily defined to `uint32_t` but functionally equivalent.
+ */
+typedef uint32_t XXH32_hash_t;
+
+#elif !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#   include <stdint.h>
+    typedef uint32_t XXH32_hash_t;
+
+#else
+#   include <limits.h>
+#   if UINT_MAX == 0xFFFFFFFFUL
+      typedef unsigned int XXH32_hash_t;
+#   else
+#     if ULONG_MAX == 0xFFFFFFFFUL
+        typedef unsigned long XXH32_hash_t;
+#     else
+#       error "unsupported platform: need a 32-bit type"
+#     endif
+#   endif
+#endif
+
+/*!
+ * @}
+ *
+ * @defgroup xxh32_family XXH32 family
+ * @ingroup public
+ * Contains functions used in the classic 32-bit xxHash algorithm.
+ *
+ * @note
+ *   XXH32 is useful for older platforms, with no or poor 64-bit performance.
+ *   Note that @ref xxh3_family provides competitive speed
+ *   for both 32-bit and 64-bit systems, and offers true 64/128 bit hash results.
+ *
+ * @see @ref xxh64_family, @ref xxh3_family : Other xxHash families
+ * @see @ref xxh32_impl for implementation details
+ * @{
+ */
+
+/*!
+ * @brief Calculates the 32-bit hash of @p input using xxHash32.
+ *
+ * Speed on Core 2 Duo @ 3 GHz (single thread, SMHasher benchmark): 5.4 GB/s
+ *
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ * @param seed The 32-bit seed to alter the hash's output predictably.
+ *
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return The calculated 32-bit hash value.
+ *
+ * @see
+ *    XXH64(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
+ *    Direct equivalents for the other variants of xxHash.
+ * @see
+ *    XXH32_createState(), XXH32_update(), XXH32_digest(): Streaming version.
+ */
+XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t length, XXH32_hash_t seed);
+
+/*!
+ * Streaming functions generate the xxHash value from an incremental input.
+ * This method is slower than single-call functions, due to state management.
+ * For small inputs, prefer `XXH32()` and `XXH64()`, which are better optimized.
+ *
+ * An XXH state must first be allocated using `XXH*_createState()`.
+ *
+ * Start a new hash by initializing the state with a seed using `XXH*_reset()`.
+ *
+ * Then, feed the hash state by calling `XXH*_update()` as many times as necessary.
+ *
+ * The function returns an error code, with 0 meaning OK, and any other value
+ * meaning there is an error.
+ *
+ * Finally, a hash value can be produced anytime, by using `XXH*_digest()`.
+ * This function returns the nn-bits hash as an int or long long.
+ *
+ * It's still possible to continue inserting input into the hash state after a
+ * digest, and generate new hash values later on by invoking `XXH*_digest()`.
+ *
+ * When done, release the state using `XXH*_freeState()`.
+ *
+ * Example code for incrementally hashing a file:
+ * @code{.c}
+ *    #include <stdio.h>
+ *    #include <xxhash.h>
+ *    #define BUFFER_SIZE 256
+ *
+ *    // Note: XXH64 and XXH3 use the same interface.
+ *    XXH32_hash_t
+ *    hashFile(FILE* stream)
+ *    {
+ *        XXH32_state_t* state;
+ *        unsigned char buf[BUFFER_SIZE];
+ *        size_t amt;
+ *        XXH32_hash_t hash;
+ *
+ *        state = XXH32_createState();       // Create a state
+ *        assert(state != NULL);             // Error check here
+ *        XXH32_reset(state, 0xbaad5eed);    // Reset state with our seed
+ *        while ((amt = fread(buf, 1, sizeof(buf), stream)) != 0) {
+ *            XXH32_update(state, buf, amt); // Hash the file in chunks
+ *        }
+ *        hash = XXH32_digest(state);        // Finalize the hash
+ *        XXH32_freeState(state);            // Clean up
+ *        return hash;
+ *    }
+ * @endcode
+ */
+
+/*!
+ * @typedef struct XXH32_state_s XXH32_state_t
+ * @brief The opaque state struct for the XXH32 streaming API.
+ *
+ * @see XXH32_state_s for details.
+ */
+typedef struct XXH32_state_s XXH32_state_t;
+
+/*!
+ * @brief Allocates an @ref XXH32_state_t.
+ *
+ * Must be freed with XXH32_freeState().
+ * @return An allocated XXH32_state_t on success, `NULL` on failure.
+ */
+XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void);
+/*!
+ * @brief Frees an @ref XXH32_state_t.
+ *
+ * Must be allocated with XXH32_createState().
+ * @param statePtr A pointer to an @ref XXH32_state_t allocated with @ref XXH32_createState().
+ * @return XXH_OK.
+ */
+XXH_PUBLIC_API XXH_errorcode  XXH32_freeState(XXH32_state_t* statePtr);
+/*!
+ * @brief Copies one @ref XXH32_state_t to another.
+ *
+ * @param dst_state The state to copy to.
+ * @param src_state The state to copy from.
+ * @pre
+ *   @p dst_state and @p src_state must not be `NULL` and must not overlap.
+ */
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dst_state, const XXH32_state_t* src_state);
+
+/*!
+ * @brief Resets an @ref XXH32_state_t to begin a new hash.
+ *
+ * This function resets and seeds a state. Call it before @ref XXH32_update().
+ *
+ * @param statePtr The state struct to reset.
+ * @param seed The 32-bit seed to alter the hash result predictably.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH32_reset  (XXH32_state_t* statePtr, XXH32_hash_t seed);
+
+/*!
+ * @brief Consumes a block of @p input to an @ref XXH32_state_t.
+ *
+ * Call this to incrementally consume blocks of data.
+ *
+ * @param statePtr The state struct to update.
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ *
+ * @pre
+ *   @p statePtr must not be `NULL`.
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return @ref XXH_OK on success, @ref XXH_ERROR on failure.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH32_update (XXH32_state_t* statePtr, const void* input, size_t length);
+
+/*!
+ * @brief Returns the calculated hash value from an @ref XXH32_state_t.
+ *
+ * @note
+ *   Calling XXH32_digest() will not affect @p statePtr, so you can update,
+ *   digest, and update again.
+ *
+ * @param statePtr The state struct to calculate the hash from.
+ *
+ * @pre
+ *  @p statePtr must not be `NULL`.
+ *
+ * @return The calculated xxHash32 value from that state.
+ */
+XXH_PUBLIC_API XXH32_hash_t  XXH32_digest (const XXH32_state_t* statePtr);
+
+/*******   Canonical representation   *******/
+
+/*
+ * The default return values from XXH functions are unsigned 32 and 64 bit
+ * integers.
+ * This the simplest and fastest format for further post-processing.
+ *
+ * However, this leaves open the question of what is the order on the byte level,
+ * since little and big endian conventions will store the same number differently.
+ *
+ * The canonical representation settles this issue by mandating big-endian
+ * convention, the same convention as human-readable numbers (large digits first).
+ *
+ * When writing hash values to storage, sending them over a network, or printing
+ * them, it's highly recommended to use the canonical representation to ensure
+ * portability across a wider range of systems, present and future.
+ *
+ * The following functions allow transformation of hash values to and from
+ * canonical format.
+ */
+
+/*!
+ * @brief Canonical (big endian) representation of @ref XXH32_hash_t.
+ */
+typedef struct {
+    unsigned char digest[4]; /*!< Hash bytes, big endian */
+} XXH32_canonical_t;
+
+/*!
+ * @brief Converts an @ref XXH32_hash_t to a big endian @ref XXH32_canonical_t.
+ *
+ * @param dst The @ref XXH32_canonical_t pointer to be stored to.
+ * @param hash The @ref XXH32_hash_t to be converted.
+ *
+ * @pre
+ *   @p dst must not be `NULL`.
+ */
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash);
+
+/*!
+ * @brief Converts an @ref XXH32_canonical_t to a native @ref XXH32_hash_t.
+ *
+ * @param src The @ref XXH32_canonical_t to convert.
+ *
+ * @pre
+ *   @p src must not be `NULL`.
+ *
+ * @return The converted hash.
+ */
+XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src);
+
+
+#ifdef __has_attribute
+# define XXH_HAS_ATTRIBUTE(x) __has_attribute(x)
+#else
+# define XXH_HAS_ATTRIBUTE(x) 0
+#endif
+
+/* C-language Attributes are added in C23. */
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ > 201710L) && defined(__has_c_attribute)
+# define XXH_HAS_C_ATTRIBUTE(x) __has_c_attribute(x)
+#else
+# define XXH_HAS_C_ATTRIBUTE(x) 0
+#endif
+
+#if defined(__cplusplus) && defined(__has_cpp_attribute)
+# define XXH_HAS_CPP_ATTRIBUTE(x) __has_cpp_attribute(x)
+#else
+# define XXH_HAS_CPP_ATTRIBUTE(x) 0
+#endif
+
+/*
+Define XXH_FALLTHROUGH macro for annotating switch case with the 'fallthrough' attribute
+introduced in CPP17 and C23.
+CPP17 : https://en.cppreference.com/w/cpp/language/attributes/fallthrough
+C23   : https://en.cppreference.com/w/c/language/attributes/fallthrough
+*/
+#if XXH_HAS_C_ATTRIBUTE(x)
+# define XXH_FALLTHROUGH [[fallthrough]]
+#elif XXH_HAS_CPP_ATTRIBUTE(x)
+# define XXH_FALLTHROUGH [[fallthrough]]
+#elif XXH_HAS_ATTRIBUTE(__fallthrough__)
+# define XXH_FALLTHROUGH __attribute__ ((fallthrough))
+#else
+# define XXH_FALLTHROUGH
+#endif
+
+/*!
+ * @}
+ * @ingroup public
+ * @{
+ */
+
+#ifndef XXH_NO_LONG_LONG
+/*-**********************************************************************
+*  64-bit hash
+************************************************************************/
+#if defined(XXH_DOXYGEN) /* don't include <stdint.h> */
+/*!
+ * @brief An unsigned 64-bit integer.
+ *
+ * Not necessarily defined to `uint64_t` but functionally equivalent.
+ */
+typedef uint64_t XXH64_hash_t;
+#elif !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+#  include <stdint.h>
+   typedef uint64_t XXH64_hash_t;
+#else
+#  include <limits.h>
+#  if defined(__LP64__) && ULONG_MAX == 0xFFFFFFFFFFFFFFFFULL
+     /* LP64 ABI says uint64_t is unsigned long */
+     typedef unsigned long XXH64_hash_t;
+#  else
+     /* the following type must have a width of 64-bit */
+     typedef unsigned long long XXH64_hash_t;
+#  endif
+#endif
+
+/*!
+ * @}
+ *
+ * @defgroup xxh64_family XXH64 family
+ * @ingroup public
+ * @{
+ * Contains functions used in the classic 64-bit xxHash algorithm.
+ *
+ * @note
+ *   XXH3 provides competitive speed for both 32-bit and 64-bit systems,
+ *   and offers true 64/128 bit hash results.
+ *   It provides better speed for systems with vector processing capabilities.
+ */
+
+
+/*!
+ * @brief Calculates the 64-bit hash of @p input using xxHash64.
+ *
+ * This function usually runs faster on 64-bit systems, but slower on 32-bit
+ * systems (see benchmark).
+ *
+ * @param input The block of data to be hashed, at least @p length bytes in size.
+ * @param length The length of @p input, in bytes.
+ * @param seed The 64-bit seed to alter the hash's output predictably.
+ *
+ * @pre
+ *   The memory between @p input and @p input + @p length must be valid,
+ *   readable, contiguous memory. However, if @p length is `0`, @p input may be
+ *   `NULL`. In C++, this also must be *TriviallyCopyable*.
+ *
+ * @return The calculated 64-bit hash.
+ *
+ * @see
+ *    XXH32(), XXH3_64bits_withSeed(), XXH3_128bits_withSeed(), XXH128():
+ *    Direct equivalents for the other variants of xxHash.
+ * @see
+ *    XXH64_createState(), XXH64_update(), XXH64_digest(): Streaming version.
+ */
+XXH_PUBLIC_API XXH64_hash_t XXH64(const void* input, size_t length, XXH64_hash_t seed);
+
+/*******   Streaming   *******/
+/*!
+ * @brief The opaque state struct for the XXH64 streaming API.
+ *
+ * @see XXH64_state_s for details.
+ */
+typedef struct XXH64_state_s XXH64_state_t;   /* incomplete type */
+XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void);
+XXH_PUBLIC_API XXH_errorcode  XXH64_freeState(XXH64_state_t* statePtr);
+XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dst_state, const XXH64_state_t* src_state);
+
+XXH_PUBLIC_API XXH_errorcode XXH64_reset  (XXH64_state_t* statePtr, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_errorcode XXH64_update (XXH64_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH64_hash_t  XXH64_digest (const XXH64_state_t* statePtr);
+
+/*******   Canonical representation   *******/
+typedef struct { unsigned char digest[sizeof(XXH64_hash_t)]; } XXH64_canonical_t;
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash);
+XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src);
+
+/*!
+ * @}
+ * ************************************************************************
+ * @defgroup xxh3_family XXH3 family
+ * @ingroup public
+ * @{
+ *
+ * XXH3 is a more recent hash algorithm featuring:
+ *  - Improved speed for both small and large inputs
+ *  - True 64-bit and 128-bit outputs
+ *  - SIMD acceleration
+ *  - Improved 32-bit viability
+ *
+ * Speed analysis methodology is explained here:
+ *
+ *    https://fastcompression.blogspot.com/2019/03/presenting-xxh3.html
+ *
+ * Compared to XXH64, expect XXH3 to run approximately
+ * ~2x faster on large inputs and >3x faster on small ones,
+ * exact differences vary depending on platform.
+ *
+ * XXH3's speed benefits greatly from SIMD and 64-bit arithmetic,
+ * but does not require it.
+ * Any 32-bit and 64-bit targets that can run XXH32 smoothly
+ * can run XXH3 at competitive speeds, even without vector support.
+ * Further details are explained in the implementation.
+ *
+ * Optimized implementations are provided for AVX512, AVX2, SSE2, NEON, POWER8,
+ * ZVector and scalar targets. This can be controlled via the XXH_VECTOR macro.
+ *
+ * XXH3 implementation is portable:
+ * it has a generic C90 formulation that can be compiled on any platform,
+ * all implementations generage exactly the same hash value on all platforms.
+ * Starting from v0.8.0, it's also labelled "stable", meaning that
+ * any future version will also generate the same hash value.
+ *
+ * XXH3 offers 2 variants, _64bits and _128bits.
+ *
+ * When only 64 bits are needed, prefer invoking the _64bits variant, as it
+ * reduces the amount of mixing, resulting in faster speed on small inputs.
+ * It's also generally simpler to manipulate a scalar return type than a struct.
+ *
+ * The API supports one-shot hashing, streaming mode, and custom secrets.
+ */
+
+/*-**********************************************************************
+*  XXH3 64-bit variant
+************************************************************************/
+
+/* XXH3_64bits():
+ * default 64-bit variant, using default secret and default seed of 0.
+ * It's the fastest variant. */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* data, size_t len);
+
+/*
+ * XXH3_64bits_withSeed():
+ * This variant generates a custom secret on the fly
+ * based on default secret altered using the `seed` value.
+ * While this operation is decently fast, note that it's not completely free.
+ * Note: seed==0 produces the same results as XXH3_64bits().
+ */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSeed(const void* data, size_t len, XXH64_hash_t seed);
+
+/*!
+ * The bare minimum size for a custom secret.
+ *
+ * @see
+ *  XXH3_64bits_withSecret(), XXH3_64bits_reset_withSecret(),
+ *  XXH3_128bits_withSecret(), XXH3_128bits_reset_withSecret().
+ */
+#define XXH3_SECRET_SIZE_MIN 136
+
+/*
+ * XXH3_64bits_withSecret():
+ * It's possible to provide any blob of bytes as a "secret" to generate the hash.
+ * This makes it more difficult for an external actor to prepare an intentional collision.
+ * The main condition is that secretSize *must* be large enough (>= XXH3_SECRET_SIZE_MIN).
+ * However, the quality of the secret impacts the dispersion of the hash algorithm.
+ * Therefore, the secret _must_ look like a bunch of random bytes.
+ * Avoid "trivial" or structured data such as repeated sequences or a text document.
+ * Whenever in doubt about the "randomness" of the blob of bytes,
+ * consider employing "XXH3_generateSecret()" instead (see below).
+ * It will generate a proper high entropy secret derived from the blob of bytes.
+ * Another advantage of using XXH3_generateSecret() is that
+ * it guarantees that all bits within the initial blob of bytes
+ * will impact every bit of the output.
+ * This is not necessarily the case when using the blob of bytes directly
+ * because, when hashing _small_ inputs, only a portion of the secret is employed.
+ */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize);
+
+
+/*******   Streaming   *******/
+/*
+ * Streaming requires state maintenance.
+ * This operation costs memory and CPU.
+ * As a consequence, streaming is slower than one-shot hashing.
+ * For better performance, prefer one-shot functions whenever applicable.
+ */
+
+/*!
+ * @brief The state struct for the XXH3 streaming API.
+ *
+ * @see XXH3_state_s for details.
+ */
+typedef struct XXH3_state_s XXH3_state_t;
+XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void);
+XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr);
+XXH_PUBLIC_API void XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state);
+
+/*
+ * XXH3_64bits_reset():
+ * Initialize with default parameters.
+ * digest will be equivalent to `XXH3_64bits()`.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset(XXH3_state_t* statePtr);
+/*
+ * XXH3_64bits_reset_withSeed():
+ * Generate a custom secret from `seed`, and store it into `statePtr`.
+ * digest will be equivalent to `XXH3_64bits_withSeed()`.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed);
+/*
+ * XXH3_64bits_reset_withSecret():
+ * `secret` is referenced, it _must outlive_ the hash streaming session.
+ * Similar to one-shot API, `secretSize` must be >= `XXH3_SECRET_SIZE_MIN`,
+ * and the quality of produced hash values depends on secret's entropy
+ * (secret's content should look like a bunch of random bytes).
+ * When in doubt about the randomness of a candidate `secret`,
+ * consider employing `XXH3_generateSecret()` instead (see below).
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize);
+
+XXH_PUBLIC_API XXH_errorcode XXH3_64bits_update (XXH3_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH64_hash_t  XXH3_64bits_digest (const XXH3_state_t* statePtr);
+
+/* note : canonical representation of XXH3 is the same as XXH64
+ * since they both produce XXH64_hash_t values */
+
+
+/*-**********************************************************************
+*  XXH3 128-bit variant
+************************************************************************/
+
+/*!
+ * @brief The return value from 128-bit hashes.
+ *
+ * Stored in little endian order, although the fields themselves are in native
+ * endianness.
+ */
+typedef struct {
+    XXH64_hash_t low64;   /*!< `value & 0xFFFFFFFFFFFFFFFF` */
+    XXH64_hash_t high64;  /*!< `value >> 64` */
+} XXH128_hash_t;
+
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* data, size_t len);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSeed(const void* data, size_t len, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_withSecret(const void* data, size_t len, const void* secret, size_t secretSize);
+
+/*******   Streaming   *******/
+/*
+ * Streaming requires state maintenance.
+ * This operation costs memory and CPU.
+ * As a consequence, streaming is slower than one-shot hashing.
+ * For better performance, prefer one-shot functions whenever applicable.
+ *
+ * XXH3_128bits uses the same XXH3_state_t as XXH3_64bits().
+ * Use already declared XXH3_createState() and XXH3_freeState().
+ *
+ * All reset and streaming functions have same meaning as their 64-bit counterpart.
+ */
+
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset(XXH3_state_t* statePtr);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed);
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize);
+
+XXH_PUBLIC_API XXH_errorcode XXH3_128bits_update (XXH3_state_t* statePtr, const void* input, size_t length);
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* statePtr);
+
+/* Following helper functions make it possible to compare XXH128_hast_t values.
+ * Since XXH128_hash_t is a structure, this capability is not offered by the language.
+ * Note: For better performance, these functions can be inlined using XXH_INLINE_ALL */
+
+/*!
+ * XXH128_isEqual():
+ * Return: 1 if `h1` and `h2` are equal, 0 if they are not.
+ */
+XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2);
+
+/*!
+ * XXH128_cmp():
+ *
+ * This comparator is compatible with stdlib's `qsort()`/`bsearch()`.
+ *
+ * return: >0 if *h128_1  > *h128_2
+ *         =0 if *h128_1 == *h128_2
+ *         <0 if *h128_1  < *h128_2
+ */
+XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2);
+
+
+/*******   Canonical representation   *******/
+typedef struct { unsigned char digest[sizeof(XXH128_hash_t)]; } XXH128_canonical_t;
+XXH_PUBLIC_API void XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash);
+XXH_PUBLIC_API XXH128_hash_t XXH128_hashFromCanonical(const XXH128_canonical_t* src);
+
+
+#endif  /* XXH_NO_LONG_LONG */
+
+/*!
+ * @}
+ */
+#endif /* XXHASH_H_5627135585666179 */
+
+
+
+#if defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742)
+#define XXHASH_H_STATIC_13879238742
+/* ****************************************************************************
+ * This section contains declarations which are not guaranteed to remain stable.
+ * They may change in future versions, becoming incompatible with a different
+ * version of the library.
+ * These declarations should only be used with static linking.
+ * Never use them in association with dynamic linking!
+ ***************************************************************************** */
+
+/*
+ * These definitions are only present to allow static allocation
+ * of XXH states, on stack or in a struct, for example.
+ * Never **ever** access their members directly.
+ */
+
+/*!
+ * @internal
+ * @brief Structure for XXH32 streaming API.
+ *
+ * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
+ * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is
+ * an opaque type. This allows fields to safely be changed.
+ *
+ * Typedef'd to @ref XXH32_state_t.
+ * Do not access the members of this struct directly.
+ * @see XXH64_state_s, XXH3_state_s
+ */
+struct XXH32_state_s {
+   XXH32_hash_t total_len_32; /*!< Total length hashed, modulo 2^32 */
+   XXH32_hash_t large_len;    /*!< Whether the hash is >= 16 (handles @ref total_len_32 overflow) */
+   XXH32_hash_t v[4];         /*!< Accumulator lanes */
+   XXH32_hash_t mem32[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[16]. */
+   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem32 */
+   XXH32_hash_t reserved;     /*!< Reserved field. Do not read or write to it, it may be removed. */
+};   /* typedef'd to XXH32_state_t */
+
+
+#ifndef XXH_NO_LONG_LONG  /* defined when there is no 64-bit support */
+
+/*!
+ * @internal
+ * @brief Structure for XXH64 streaming API.
+ *
+ * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
+ * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined. Otherwise it is
+ * an opaque type. This allows fields to safely be changed.
+ *
+ * Typedef'd to @ref XXH64_state_t.
+ * Do not access the members of this struct directly.
+ * @see XXH32_state_s, XXH3_state_s
+ */
+struct XXH64_state_s {
+   XXH64_hash_t total_len;    /*!< Total length hashed. This is always 64-bit. */
+   XXH64_hash_t v[4];         /*!< Accumulator lanes */
+   XXH64_hash_t mem64[4];     /*!< Internal buffer for partial reads. Treated as unsigned char[32]. */
+   XXH32_hash_t memsize;      /*!< Amount of data in @ref mem64 */
+   XXH32_hash_t reserved32;   /*!< Reserved field, needed for padding anyways*/
+   XXH64_hash_t reserved64;   /*!< Reserved field. Do not read or write to it, it may be removed. */
+};   /* typedef'd to XXH64_state_t */
+
+#if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L) /* >= C11 */
+#  include <stdalign.h>
+#  define XXH_ALIGN(n)      alignas(n)
+#elif defined(__cplusplus) && (__cplusplus >= 201103L) /* >= C++11 */
+/* In C++ alignas() is a keyword */
+#  define XXH_ALIGN(n)      alignas(n)
+#elif defined(__GNUC__)
+#  define XXH_ALIGN(n)      __attribute__ ((aligned(n)))
+#elif defined(_MSC_VER)
+#  define XXH_ALIGN(n)      __declspec(align(n))
+#else
+#  define XXH_ALIGN(n)   /* disabled */
+#endif
+
+/* Old GCC versions only accept the attribute after the type in structures. */
+#if !(defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L))   /* C11+ */ \
+    && ! (defined(__cplusplus) && (__cplusplus >= 201103L)) /* >= C++11 */ \
+    && defined(__GNUC__)
+#   define XXH_ALIGN_MEMBER(align, type) type XXH_ALIGN(align)
+#else
+#   define XXH_ALIGN_MEMBER(align, type) XXH_ALIGN(align) type
+#endif
+
+/*!
+ * @brief The size of the internal XXH3 buffer.
+ *
+ * This is the optimal update size for incremental hashing.
+ *
+ * @see XXH3_64b_update(), XXH3_128b_update().
+ */
+#define XXH3_INTERNALBUFFER_SIZE 256
+
+/*!
+ * @brief Default size of the secret buffer (and @ref XXH3_kSecret).
+ *
+ * This is the size used in @ref XXH3_kSecret and the seeded functions.
+ *
+ * Not to be confused with @ref XXH3_SECRET_SIZE_MIN.
+ */
+#define XXH3_SECRET_DEFAULT_SIZE 192
+
+/*!
+ * @internal
+ * @brief Structure for XXH3 streaming API.
+ *
+ * @note This is only defined when @ref XXH_STATIC_LINKING_ONLY,
+ * @ref XXH_INLINE_ALL, or @ref XXH_IMPLEMENTATION is defined.
+ * Otherwise it is an opaque type.
+ * Never use this definition in combination with dynamic library.
+ * This allows fields to safely be changed in the future.
+ *
+ * @note ** This structure has a strict alignment requirement of 64 bytes!! **
+ * Do not allocate this with `malloc()` or `new`,
+ * it will not be sufficiently aligned.
+ * Use @ref XXH3_createState() and @ref XXH3_freeState(), or stack allocation.
+ *
+ * Typedef'd to @ref XXH3_state_t.
+ * Do never access the members of this struct directly.
+ *
+ * @see XXH3_INITSTATE() for stack initialization.
+ * @see XXH3_createState(), XXH3_freeState().
+ * @see XXH32_state_s, XXH64_state_s
+ */
+struct XXH3_state_s {
+   XXH_ALIGN_MEMBER(64, XXH64_hash_t acc[8]);
+       /*!< The 8 accumulators. Similar to `vN` in @ref XXH32_state_s::v1 and @ref XXH64_state_s */
+   XXH_ALIGN_MEMBER(64, unsigned char customSecret[XXH3_SECRET_DEFAULT_SIZE]);
+       /*!< Used to store a custom secret generated from a seed. */
+   XXH_ALIGN_MEMBER(64, unsigned char buffer[XXH3_INTERNALBUFFER_SIZE]);
+       /*!< The internal buffer. @see XXH32_state_s::mem32 */
+   XXH32_hash_t bufferedSize;
+       /*!< The amount of memory in @ref buffer, @see XXH32_state_s::memsize */
+   XXH32_hash_t useSeed;
+       /*!< Reserved field. Needed for padding on 64-bit. */
+   size_t nbStripesSoFar;
+       /*!< Number or stripes processed. */
+   XXH64_hash_t totalLen;
+       /*!< Total length hashed. 64-bit even on 32-bit targets. */
+   size_t nbStripesPerBlock;
+       /*!< Number of stripes per block. */
+   size_t secretLimit;
+       /*!< Size of @ref customSecret or @ref extSecret */
+   XXH64_hash_t seed;
+       /*!< Seed for _withSeed variants. Must be zero otherwise, @see XXH3_INITSTATE() */
+   XXH64_hash_t reserved64;
+       /*!< Reserved field. */
+   const unsigned char* extSecret;
+       /*!< Reference to an external secret for the _withSecret variants, NULL
+        *   for other variants. */
+   /* note: there may be some padding at the end due to alignment on 64 bytes */
+}; /* typedef'd to XXH3_state_t */
+
+#undef XXH_ALIGN_MEMBER
+
+/*!
+ * @brief Initializes a stack-allocated `XXH3_state_s`.
+ *
+ * When the @ref XXH3_state_t structure is merely emplaced on stack,
+ * it should be initialized with XXH3_INITSTATE() or a memset()
+ * in case its first reset uses XXH3_NNbits_reset_withSeed().
+ * This init can be omitted if the first reset uses default or _withSecret mode.
+ * This operation isn't necessary when the state is created with XXH3_createState().
+ * Note that this doesn't prepare the state for a streaming operation,
+ * it's still necessary to use XXH3_NNbits_reset*() afterwards.
+ */
+#define XXH3_INITSTATE(XXH3_state_ptr)   { (XXH3_state_ptr)->seed = 0; }
+
+
+/* XXH128() :
+ * simple alias to pre-selected XXH3_128bits variant
+ */
+XXH_PUBLIC_API XXH128_hash_t XXH128(const void* data, size_t len, XXH64_hash_t seed);
+
+
+/* ===   Experimental API   === */
+/* Symbols defined below must be considered tied to a specific library version. */
+
+/*
+ * XXH3_generateSecret():
+ *
+ * Derive a high-entropy secret from any user-defined content, named customSeed.
+ * The generated secret can be used in combination with `*_withSecret()` functions.
+ * The `_withSecret()` variants are useful to provide a higher level of protection than 64-bit seed,
+ * as it becomes much more difficult for an external actor to guess how to impact the calculation logic.
+ *
+ * The function accepts as input a custom seed of any length and any content,
+ * and derives from it a high-entropy secret of length @secretSize
+ * into an already allocated buffer @secretBuffer.
+ * @secretSize must be >= XXH3_SECRET_SIZE_MIN
+ *
+ * The generated secret can then be used with any `*_withSecret()` variant.
+ * Functions `XXH3_128bits_withSecret()`, `XXH3_64bits_withSecret()`,
+ * `XXH3_128bits_reset_withSecret()` and `XXH3_64bits_reset_withSecret()`
+ * are part of this list. They all accept a `secret` parameter
+ * which must be large enough for implementation reasons (>= XXH3_SECRET_SIZE_MIN)
+ * _and_ feature very high entropy (consist of random-looking bytes).
+ * These conditions can be a high bar to meet, so
+ * XXH3_generateSecret() can be employed to ensure proper quality.
+ *
+ * customSeed can be anything. It can have any size, even small ones,
+ * and its content can be anything, even "poor entropy" sources such as a bunch of zeroes.
+ * The resulting `secret` will nonetheless provide all required qualities.
+ *
+ * When customSeedSize > 0, supplying NULL as customSeed is undefined behavior.
+ */
+XXH_PUBLIC_API XXH_errorcode XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSeed, size_t customSeedSize);
+
+
+/*
+ * XXH3_generateSecret_fromSeed():
+ *
+ * Generate the same secret as the _withSeed() variants.
+ *
+ * The resulting secret has a length of XXH3_SECRET_DEFAULT_SIZE (necessarily).
+ * @secretBuffer must be already allocated, of size at least XXH3_SECRET_DEFAULT_SIZE bytes.
+ *
+ * The generated secret can be used in combination with
+ *`*_withSecret()` and `_withSecretandSeed()` variants.
+ * This generator is notably useful in combination with `_withSecretandSeed()`,
+ * as a way to emulate a faster `_withSeed()` variant.
+ */
+XXH_PUBLIC_API void XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed);
+
+/*
+ * *_withSecretandSeed() :
+ * These variants generate hash values using either
+ * @seed for "short" keys (< XXH3_MIDSIZE_MAX = 240 bytes)
+ * or @secret for "large" keys (>= XXH3_MIDSIZE_MAX).
+ *
+ * This generally benefits speed, compared to `_withSeed()` or `_withSecret()`.
+ * `_withSeed()` has to generate the secret on the fly for "large" keys.
+ * It's fast, but can be perceptible for "not so large" keys (< 1 KB).
+ * `_withSecret()` has to generate the masks on the fly for "small" keys,
+ * which requires more instructions than _withSeed() variants.
+ * Therefore, _withSecretandSeed variant combines the best of both worlds.
+ *
+ * When @secret has been generated by XXH3_generateSecret_fromSeed(),
+ * this variant produces *exactly* the same results as `_withSeed()` variant,
+ * hence offering only a pure speed benefit on "large" input,
+ * by skipping the need to regenerate the secret for every large input.
+ *
+ * Another usage scenario is to hash the secret to a 64-bit hash value,
+ * for example with XXH3_64bits(), which then becomes the seed,
+ * and then employ both the seed and the secret in _withSecretandSeed().
+ * On top of speed, an added benefit is that each bit in the secret
+ * has a 50% chance to swap each bit in the output,
+ * via its impact to the seed.
+ * This is not guaranteed when using the secret directly in "small data" scenarios,
+ * because only portions of the secret are employed for small data.
+ */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecretandSeed(const void* data, size_t len,
+                              const void* secret, size_t secretSize,
+                              XXH64_hash_t seed);
+
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecretandSeed(const void* data, size_t len,
+                               const void* secret, size_t secretSize,
+                               XXH64_hash_t seed64);
+
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
+                                    const void* secret, size_t secretSize,
+                                    XXH64_hash_t seed64);
+
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr,
+                                     const void* secret, size_t secretSize,
+                                     XXH64_hash_t seed64);
+
+
+#endif  /* XXH_NO_LONG_LONG */
+#if defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API)
+#  define XXH_IMPLEMENTATION
+#endif
+
+#endif  /* defined(XXH_STATIC_LINKING_ONLY) && !defined(XXHASH_H_STATIC_13879238742) */
+
+
+/* ======================================================================== */
+/* ======================================================================== */
+/* ======================================================================== */
+
+
+/*-**********************************************************************
+ * xxHash implementation
+ *-**********************************************************************
+ * xxHash's implementation used to be hosted inside xxhash.c.
+ *
+ * However, inlining requires implementation to be visible to the compiler,
+ * hence be included alongside the header.
+ * Previously, implementation was hosted inside xxhash.c,
+ * which was then #included when inlining was activated.
+ * This construction created issues with a few build and install systems,
+ * as it required xxhash.c to be stored in /include directory.
+ *
+ * xxHash implementation is now directly integrated within xxhash.h.
+ * As a consequence, xxhash.c is no longer needed in /include.
+ *
+ * xxhash.c is still available and is still useful.
+ * In a "normal" setup, when xxhash is not inlined,
+ * xxhash.h only exposes the prototypes and public symbols,
+ * while xxhash.c can be built into an object file xxhash.o
+ * which can then be linked into the final binary.
+ ************************************************************************/
+
+#if ( defined(XXH_INLINE_ALL) || defined(XXH_PRIVATE_API) \
+   || defined(XXH_IMPLEMENTATION) ) && !defined(XXH_IMPLEM_13a8737387)
+#  define XXH_IMPLEM_13a8737387
+
+/* *************************************
+*  Tuning parameters
+***************************************/
+
+/*!
+ * @defgroup tuning Tuning parameters
+ * @{
+ *
+ * Various macros to control xxHash's behavior.
+ */
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Define this to disable 64-bit code.
+ *
+ * Useful if only using the @ref xxh32_family and you have a strict C90 compiler.
+ */
+#  define XXH_NO_LONG_LONG
+#  undef XXH_NO_LONG_LONG /* don't actually */
+/*!
+ * @brief Controls how unaligned memory is accessed.
+ *
+ * By default, access to unaligned memory is controlled by `memcpy()`, which is
+ * safe and portable.
+ *
+ * Unfortunately, on some target/compiler combinations, the generated assembly
+ * is sub-optimal.
+ *
+ * The below switch allow selection of a different access method
+ * in the search for improved performance.
+ *
+ * @par Possible options:
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=0` (default): `memcpy`
+ *   @par
+ *     Use `memcpy()`. Safe and portable. Note that most modern compilers will
+ *     eliminate the function call and treat it as an unaligned access.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=1`: `__attribute__((packed))`
+ *   @par
+ *     Depends on compiler extensions and is therefore not portable.
+ *     This method is safe _if_ your compiler supports it,
+ *     and *generally* as fast or faster than `memcpy`.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=2`: Direct cast
+ *  @par
+ *     Casts directly and dereferences. This method doesn't depend on the
+ *     compiler, but it violates the C standard as it directly dereferences an
+ *     unaligned pointer. It can generate buggy code on targets which do not
+ *     support unaligned memory accesses, but in some circumstances, it's the
+ *     only known way to get the most performance.
+ *
+ *  - `XXH_FORCE_MEMORY_ACCESS=3`: Byteshift
+ *  @par
+ *     Also portable. This can generate the best code on old compilers which don't
+ *     inline small `memcpy()` calls, and it might also be faster on big-endian
+ *     systems which lack a native byteswap instruction. However, some compilers
+ *     will emit literal byteshifts even if the target supports unaligned access.
+ *  .
+ *
+ * @warning
+ *   Methods 1 and 2 rely on implementation-defined behavior. Use these with
+ *   care, as what works on one compiler/platform/optimization level may cause
+ *   another to read garbage data or even crash.
+ *
+ * See http://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html for details.
+ *
+ * Prefer these methods in priority order (0 > 3 > 1 > 2)
+ */
+#  define XXH_FORCE_MEMORY_ACCESS 0
+
+/*!
+ * @def XXH_FORCE_ALIGN_CHECK
+ * @brief If defined to non-zero, adds a special path for aligned inputs (XXH32()
+ * and XXH64() only).
+ *
+ * This is an important performance trick for architectures without decent
+ * unaligned memory access performance.
+ *
+ * It checks for input alignment, and when conditions are met, uses a "fast
+ * path" employing direct 32-bit/64-bit reads, resulting in _dramatically
+ * faster_ read speed.
+ *
+ * The check costs one initial branch per hash, which is generally negligible,
+ * but not zero.
+ *
+ * Moreover, it's not useful to generate an additional code path if memory
+ * access uses the same instruction for both aligned and unaligned
+ * addresses (e.g. x86 and aarch64).
+ *
+ * In these cases, the alignment check can be removed by setting this macro to 0.
+ * Then the code will always use unaligned memory access.
+ * Align check is automatically disabled on x86, x64 & arm64,
+ * which are platforms known to offer good unaligned memory accesses performance.
+ *
+ * This option does not affect XXH3 (only XXH32 and XXH64).
+ */
+#  define XXH_FORCE_ALIGN_CHECK 0
+
+/*!
+ * @def XXH_NO_INLINE_HINTS
+ * @brief When non-zero, sets all functions to `static`.
+ *
+ * By default, xxHash tries to force the compiler to inline almost all internal
+ * functions.
+ *
+ * This can usually improve performance due to reduced jumping and improved
+ * constant folding, but significantly increases the size of the binary which
+ * might not be favorable.
+ *
+ * Additionally, sometimes the forced inlining can be detrimental to performance,
+ * depending on the architecture.
+ *
+ * XXH_NO_INLINE_HINTS marks all internal functions as static, giving the
+ * compiler full control on whether to inline or not.
+ *
+ * When not optimizing (-O0), optimizing for size (-Os, -Oz), or using
+ * -fno-inline with GCC or Clang, this will automatically be defined.
+ */
+#  define XXH_NO_INLINE_HINTS 0
+
+/*!
+ * @def XXH32_ENDJMP
+ * @brief Whether to use a jump for `XXH32_finalize`.
+ *
+ * For performance, `XXH32_finalize` uses multiple branches in the finalizer.
+ * This is generally preferable for performance,
+ * but depending on exact architecture, a jmp may be preferable.
+ *
+ * This setting is only possibly making a difference for very small inputs.
+ */
+#  define XXH32_ENDJMP 0
+
+/*!
+ * @internal
+ * @brief Redefines old internal names.
+ *
+ * For compatibility with code that uses xxHash's internals before the names
+ * were changed to improve namespacing. There is no other reason to use this.
+ */
+#  define XXH_OLD_NAMES
+#  undef XXH_OLD_NAMES /* don't actually use, it is ugly. */
+#endif /* XXH_DOXYGEN */
+/*!
+ * @}
+ */
+
+#ifndef XXH_FORCE_MEMORY_ACCESS   /* can be defined externally, on command line for example */
+   /* prefer __packed__ structures (method 1) for gcc on armv7+ and mips */
+#  if !defined(__clang__) && \
+( \
+    (defined(__INTEL_COMPILER) && !defined(_WIN32)) || \
+    ( \
+        defined(__GNUC__) && ( \
+            (defined(__ARM_ARCH) && __ARM_ARCH >= 7) || \
+            ( \
+                defined(__mips__) && \
+                (__mips <= 5 || __mips_isa_rev < 6) && \
+                (!defined(__mips16) || defined(__mips_mips16e2)) \
+            ) \
+        ) \
+    ) \
+)
+#    define XXH_FORCE_MEMORY_ACCESS 1
+#  endif
+#endif
+
+#ifndef XXH_FORCE_ALIGN_CHECK  /* can be defined externally */
+#  if defined(__i386)  || defined(__x86_64__) || defined(__aarch64__) \
+   || defined(_M_IX86) || defined(_M_X64)     || defined(_M_ARM64) /* visual */
+#    define XXH_FORCE_ALIGN_CHECK 0
+#  else
+#    define XXH_FORCE_ALIGN_CHECK 1
+#  endif
+#endif
+
+#ifndef XXH_NO_INLINE_HINTS
+#  if defined(__OPTIMIZE_SIZE__) /* -Os, -Oz */ \
+   || defined(__NO_INLINE__)     /* -O0, -fno-inline */
+#    define XXH_NO_INLINE_HINTS 1
+#  else
+#    define XXH_NO_INLINE_HINTS 0
+#  endif
+#endif
+
+#ifndef XXH32_ENDJMP
+/* generally preferable for performance */
+#  define XXH32_ENDJMP 0
+#endif
+
+/*!
+ * @defgroup impl Implementation
+ * @{
+ */
+
+
+/* *************************************
+*  Includes & Memory related functions
+***************************************/
+/*
+ * Modify the local functions below should you wish to use
+ * different memory routines for malloc() and free()
+ */
+#include <stdlib.h>
+
+/*!
+ * @internal
+ * @brief Modify this function to use a different routine than malloc().
+ */
+static void* XXH_malloc(size_t s) { return malloc(s); }
+
+/*!
+ * @internal
+ * @brief Modify this function to use a different routine than free().
+ */
+static void XXH_free(void* p) { free(p); }
+
+#include <string.h>
+
+/*!
+ * @internal
+ * @brief Modify this function to use a different routine than memcpy().
+ */
+static void* XXH_memcpy(void* dest, const void* src, size_t size)
+{
+    return memcpy(dest,src,size);
+}
+
+#include <limits.h>   /* ULLONG_MAX */
+
+
+/* *************************************
+*  Compiler Specific Options
+***************************************/
+#ifdef _MSC_VER /* Visual Studio warning fix */
+#  pragma warning(disable : 4127) /* disable: C4127: conditional expression is constant */
+#endif
+
+#if XXH_NO_INLINE_HINTS  /* disable inlining hints */
+#  if defined(__GNUC__) || defined(__clang__)
+#    define XXH_FORCE_INLINE static __attribute__((unused))
+#  else
+#    define XXH_FORCE_INLINE static
+#  endif
+#  define XXH_NO_INLINE static
+/* enable inlining hints */
+#elif defined(__GNUC__) || defined(__clang__)
+#  define XXH_FORCE_INLINE static __inline__ __attribute__((always_inline, unused))
+#  define XXH_NO_INLINE static __attribute__((noinline))
+#elif defined(_MSC_VER)  /* Visual Studio */
+#  define XXH_FORCE_INLINE static __forceinline
+#  define XXH_NO_INLINE static __declspec(noinline)
+#elif defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L))   /* C99 */
+#  define XXH_FORCE_INLINE static inline
+#  define XXH_NO_INLINE static
+#else
+#  define XXH_FORCE_INLINE static
+#  define XXH_NO_INLINE static
+#endif
+
+
+
+/* *************************************
+*  Debug
+***************************************/
+/*!
+ * @ingroup tuning
+ * @def XXH_DEBUGLEVEL
+ * @brief Sets the debugging level.
+ *
+ * XXH_DEBUGLEVEL is expected to be defined externally, typically via the
+ * compiler's command line options. The value must be a number.
+ */
+#ifndef XXH_DEBUGLEVEL
+#  ifdef DEBUGLEVEL /* backwards compat */
+#    define XXH_DEBUGLEVEL DEBUGLEVEL
+#  else
+#    define XXH_DEBUGLEVEL 0
+#  endif
+#endif
+
+#if (XXH_DEBUGLEVEL>=1)
+#  include <assert.h>   /* note: can still be disabled with NDEBUG */
+#  define XXH_ASSERT(c)   assert(c)
+#else
+#  define XXH_ASSERT(c)   ((void)0)
+#endif
+
+/* note: use after variable declarations */
+#ifndef XXH_STATIC_ASSERT
+#  if defined(__STDC_VERSION__) && (__STDC_VERSION__ >= 201112L)    /* C11 */
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { _Static_assert((c),m); } while(0)
+#  elif defined(__cplusplus) && (__cplusplus >= 201103L)            /* C++11 */
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { static_assert((c),m); } while(0)
+#  else
+#    define XXH_STATIC_ASSERT_WITH_MESSAGE(c,m) do { struct xxh_sa { char x[(c) ? 1 : -1]; }; } while(0)
+#  endif
+#  define XXH_STATIC_ASSERT(c) XXH_STATIC_ASSERT_WITH_MESSAGE((c),#c)
+#endif
+
+/*!
+ * @internal
+ * @def XXH_COMPILER_GUARD(var)
+ * @brief Used to prevent unwanted optimizations for @p var.
+ *
+ * It uses an empty GCC inline assembly statement with a register constraint
+ * which forces @p var into a general purpose register (eg eax, ebx, ecx
+ * on x86) and marks it as modified.
+ *
+ * This is used in a few places to avoid unwanted autovectorization (e.g.
+ * XXH32_round()). All vectorization we want is explicit via intrinsics,
+ * and _usually_ isn't wanted elsewhere.
+ *
+ * We also use it to prevent unwanted constant folding for AArch64 in
+ * XXH3_initCustomSecret_scalar().
+ */
+#if defined(__GNUC__) || defined(__clang__)
+#  define XXH_COMPILER_GUARD(var) __asm__ __volatile__("" : "+r" (var))
+#else
+#  define XXH_COMPILER_GUARD(var) ((void)0)
+#endif
+
+/* *************************************
+*  Basic Types
+***************************************/
+#if !defined (__VMS) \
+ && (defined (__cplusplus) \
+ || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+# include <stdint.h>
+  typedef uint8_t xxh_u8;
+#else
+  typedef unsigned char xxh_u8;
+#endif
+typedef XXH32_hash_t xxh_u32;
+
+#ifdef XXH_OLD_NAMES
+#  define BYTE xxh_u8
+#  define U8   xxh_u8
+#  define U32  xxh_u32
+#endif
+
+/* ***   Memory access   *** */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_read32(const void* ptr)
+ * @brief Reads an unaligned 32-bit integer from @p ptr in native endianness.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit native endian integer from the bytes at @p ptr.
+ */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_readLE32(const void* ptr)
+ * @brief Reads an unaligned 32-bit little endian integer from @p ptr.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit little endian integer from the bytes at @p ptr.
+ */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_readBE32(const void* ptr)
+ * @brief Reads an unaligned 32-bit big endian integer from @p ptr.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ *
+ * @param ptr The pointer to read from.
+ * @return The 32-bit big endian integer from the bytes at @p ptr.
+ */
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_readLE32_align(const void* ptr, XXH_alignment align)
+ * @brief Like @ref XXH_readLE32(), but has an option for aligned reads.
+ *
+ * Affected by @ref XXH_FORCE_MEMORY_ACCESS.
+ * Note that when @ref XXH_FORCE_ALIGN_CHECK == 0, the @p align parameter is
+ * always @ref XXH_alignment::XXH_unaligned.
+ *
+ * @param ptr The pointer to read from.
+ * @param align Whether @p ptr is aligned.
+ * @pre
+ *   If @p align == @ref XXH_alignment::XXH_aligned, @p ptr must be 4 byte
+ *   aligned.
+ * @return The 32-bit little endian integer from the bytes at @p ptr.
+ */
+
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+/*
+ * Manual byteshift. Best for old compilers which don't inline memcpy.
+ * We actually directly use XXH_readLE32 and XXH_readBE32.
+ */
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
+
+/*
+ * Force direct memory access. Only works on CPU which support unaligned memory
+ * access in hardware.
+ */
+static xxh_u32 XXH_read32(const void* memPtr) { return *(const xxh_u32*) memPtr; }
+
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
+
+/*
+ * __pack instructions are safer but compiler specific, hence potentially
+ * problematic for some compilers.
+ *
+ * Currently only defined for GCC and ICC.
+ */
+#ifdef XXH_OLD_NAMES
+typedef union { xxh_u32 u32; } __attribute__((packed)) unalign;
+#endif
+static xxh_u32 XXH_read32(const void* ptr)
+{
+    typedef union { xxh_u32 u32; } __attribute__((packed)) xxh_unalign;
+    return ((const xxh_unalign*)ptr)->u32;
+}
+
+#else
+
+/*
+ * Portable and safe solution. Generally efficient.
+ * see: http://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ */
+static xxh_u32 XXH_read32(const void* memPtr)
+{
+    xxh_u32 val;
+    XXH_memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
+
+#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+
+/* ***   Endianness   *** */
+
+/*!
+ * @ingroup tuning
+ * @def XXH_CPU_LITTLE_ENDIAN
+ * @brief Whether the target is little endian.
+ *
+ * Defined to 1 if the target is little endian, or 0 if it is big endian.
+ * It can be defined externally, for example on the compiler command line.
+ *
+ * If it is not defined,
+ * a runtime check (which is usually constant folded) is used instead.
+ *
+ * @note
+ *   This is not necessarily defined to an integer constant.
+ *
+ * @see XXH_isLittleEndian() for the runtime check.
+ */
+#ifndef XXH_CPU_LITTLE_ENDIAN
+/*
+ * Try to detect endianness automatically, to avoid the nonstandard behavior
+ * in `XXH_isLittleEndian()`
+ */
+#  if defined(_WIN32) /* Windows is always little endian */ \
+     || defined(__LITTLE_ENDIAN__) \
+     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
+#    define XXH_CPU_LITTLE_ENDIAN 1
+#  elif defined(__BIG_ENDIAN__) \
+     || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+#    define XXH_CPU_LITTLE_ENDIAN 0
+#  else
+/*!
+ * @internal
+ * @brief Runtime check for @ref XXH_CPU_LITTLE_ENDIAN.
+ *
+ * Most compilers will constant fold this.
+ */
+static int XXH_isLittleEndian(void)
+{
+    /*
+     * Portable and well-defined behavior.
+     * Don't use static: it is detrimental to performance.
+     */
+    const union { xxh_u32 u; xxh_u8 c[4]; } one = { 1 };
+    return one.c[0];
+}
+#   define XXH_CPU_LITTLE_ENDIAN   XXH_isLittleEndian()
+#  endif
+#endif
+
+
+
+
+/* ****************************************
+*  Compiler-specific Functions and Macros
+******************************************/
+#define XXH_GCC_VERSION (__GNUC__ * 100 + __GNUC_MINOR__)
+
+#ifdef __has_builtin
+#  define XXH_HAS_BUILTIN(x) __has_builtin(x)
+#else
+#  define XXH_HAS_BUILTIN(x) 0
+#endif
+
+/*!
+ * @internal
+ * @def XXH_rotl32(x,r)
+ * @brief 32-bit rotate left.
+ *
+ * @param x The 32-bit integer to be rotated.
+ * @param r The number of bits to rotate.
+ * @pre
+ *   @p r > 0 && @p r < 32
+ * @note
+ *   @p x and @p r may be evaluated multiple times.
+ * @return The rotated result.
+ */
+#if !defined(NO_CLANG_BUILTIN) && XXH_HAS_BUILTIN(__builtin_rotateleft32) \
+                               && XXH_HAS_BUILTIN(__builtin_rotateleft64)
+#  define XXH_rotl32 __builtin_rotateleft32
+#  define XXH_rotl64 __builtin_rotateleft64
+/* Note: although _rotl exists for minGW (GCC under windows), performance seems poor */
+#elif defined(_MSC_VER)
+#  define XXH_rotl32(x,r) _rotl(x,r)
+#  define XXH_rotl64(x,r) _rotl64(x,r)
+#else
+#  define XXH_rotl32(x,r) (((x) << (r)) | ((x) >> (32 - (r))))
+#  define XXH_rotl64(x,r) (((x) << (r)) | ((x) >> (64 - (r))))
+#endif
+
+/*!
+ * @internal
+ * @fn xxh_u32 XXH_swap32(xxh_u32 x)
+ * @brief A 32-bit byteswap.
+ *
+ * @param x The 32-bit integer to byteswap.
+ * @return @p x, byteswapped.
+ */
+#if defined(_MSC_VER)     /* Visual Studio */
+#  define XXH_swap32 _byteswap_ulong
+#elif XXH_GCC_VERSION >= 403
+#  define XXH_swap32 __builtin_bswap32
+#else
+static xxh_u32 XXH_swap32 (xxh_u32 x)
+{
+    return  ((x << 24) & 0xff000000 ) |
+            ((x <<  8) & 0x00ff0000 ) |
+            ((x >>  8) & 0x0000ff00 ) |
+            ((x >> 24) & 0x000000ff );
+}
+#endif
+
+
+/* ***************************
+*  Memory reads
+*****************************/
+
+/*!
+ * @internal
+ * @brief Enum to indicate whether a pointer is aligned.
+ */
+typedef enum {
+    XXH_aligned,  /*!< Aligned */
+    XXH_unaligned /*!< Possibly unaligned */
+} XXH_alignment;
+
+/*
+ * XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load.
+ *
+ * This is ideal for older compilers which don't inline memcpy.
+ */
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[0]
+         | ((xxh_u32)bytePtr[1] << 8)
+         | ((xxh_u32)bytePtr[2] << 16)
+         | ((xxh_u32)bytePtr[3] << 24);
+}
+
+XXH_FORCE_INLINE xxh_u32 XXH_readBE32(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[3]
+         | ((xxh_u32)bytePtr[2] << 8)
+         | ((xxh_u32)bytePtr[1] << 16)
+         | ((xxh_u32)bytePtr[0] << 24);
+}
+
+#else
+XXH_FORCE_INLINE xxh_u32 XXH_readLE32(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_read32(ptr) : XXH_swap32(XXH_read32(ptr));
+}
+
+static xxh_u32 XXH_readBE32(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap32(XXH_read32(ptr)) : XXH_read32(ptr);
+}
+#endif
+
+XXH_FORCE_INLINE xxh_u32
+XXH_readLE32_align(const void* ptr, XXH_alignment align)
+{
+    if (align==XXH_unaligned) {
+        return XXH_readLE32(ptr);
+    } else {
+        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u32*)ptr : XXH_swap32(*(const xxh_u32*)ptr);
+    }
+}
+
+
+/* *************************************
+*  Misc
+***************************************/
+/*! @ingroup public */
+XXH_PUBLIC_API unsigned XXH_versionNumber (void) { return XXH_VERSION_NUMBER; }
+
+
+/* *******************************************************************
+*  32-bit hash functions
+*********************************************************************/
+/*!
+ * @}
+ * @defgroup xxh32_impl XXH32 implementation
+ * @ingroup impl
+ * @{
+ */
+ /* #define instead of static const, to be used as initializers */
+#define XXH_PRIME32_1  0x9E3779B1U  /*!< 0b10011110001101110111100110110001 */
+#define XXH_PRIME32_2  0x85EBCA77U  /*!< 0b10000101111010111100101001110111 */
+#define XXH_PRIME32_3  0xC2B2AE3DU  /*!< 0b11000010101100101010111000111101 */
+#define XXH_PRIME32_4  0x27D4EB2FU  /*!< 0b00100111110101001110101100101111 */
+#define XXH_PRIME32_5  0x165667B1U  /*!< 0b00010110010101100110011110110001 */
+
+#ifdef XXH_OLD_NAMES
+#  define PRIME32_1 XXH_PRIME32_1
+#  define PRIME32_2 XXH_PRIME32_2
+#  define PRIME32_3 XXH_PRIME32_3
+#  define PRIME32_4 XXH_PRIME32_4
+#  define PRIME32_5 XXH_PRIME32_5
+#endif
+
+/*!
+ * @internal
+ * @brief Normal stripe processing routine.
+ *
+ * This shuffles the bits so that any bit from @p input impacts several bits in
+ * @p acc.
+ *
+ * @param acc The accumulator lane.
+ * @param input The stripe of input to mix.
+ * @return The mixed accumulator lane.
+ */
+static xxh_u32 XXH32_round(xxh_u32 acc, xxh_u32 input)
+{
+    acc += input * XXH_PRIME32_2;
+    acc  = XXH_rotl32(acc, 13);
+    acc *= XXH_PRIME32_1;
+#if (defined(__SSE4_1__) || defined(__aarch64__)) && !defined(XXH_ENABLE_AUTOVECTORIZE)
+    /*
+     * UGLY HACK:
+     * A compiler fence is the only thing that prevents GCC and Clang from
+     * autovectorizing the XXH32 loop (pragmas and attributes don't work for some
+     * reason) without globally disabling SSE4.1.
+     *
+     * The reason we want to avoid vectorization is because despite working on
+     * 4 integers at a time, there are multiple factors slowing XXH32 down on
+     * SSE4:
+     * - There's a ridiculous amount of lag from pmulld (10 cycles of latency on
+     *   newer chips!) making it slightly slower to multiply four integers at
+     *   once compared to four integers independently. Even when pmulld was
+     *   fastest, Sandy/Ivy Bridge, it is still not worth it to go into SSE
+     *   just to multiply unless doing a long operation.
+     *
+     * - Four instructions are required to rotate,
+     *      movqda tmp,  v // not required with VEX encoding
+     *      pslld  tmp, 13 // tmp <<= 13
+     *      psrld  v,   19 // x >>= 19
+     *      por    v,  tmp // x |= tmp
+     *   compared to one for scalar:
+     *      roll   v, 13    // reliably fast across the board
+     *      shldl  v, v, 13 // Sandy Bridge and later prefer this for some reason
+     *
+     * - Instruction level parallelism is actually more beneficial here because
+     *   the SIMD actually serializes this operation: While v1 is rotating, v2
+     *   can load data, while v3 can multiply. SSE forces them to operate
+     *   together.
+     *
+     * This is also enabled on AArch64, as Clang autovectorizes it incorrectly
+     * and it is pointless writing a NEON implementation that is basically the
+     * same speed as scalar for XXH32.
+     */
+    XXH_COMPILER_GUARD(acc);
+#endif
+    return acc;
+}
+
+/*!
+ * @internal
+ * @brief Mixes all bits to finalize the hash.
+ *
+ * The final mix ensures that all input bits have a chance to impact any bit in
+ * the output digest, resulting in an unbiased distribution.
+ *
+ * @param h32 The hash to avalanche.
+ * @return The avalanched hash.
+ */
+static xxh_u32 XXH32_avalanche(xxh_u32 h32)
+{
+    h32 ^= h32 >> 15;
+    h32 *= XXH_PRIME32_2;
+    h32 ^= h32 >> 13;
+    h32 *= XXH_PRIME32_3;
+    h32 ^= h32 >> 16;
+    return(h32);
+}
+
+#define XXH_get32bits(p) XXH_readLE32_align(p, align)
+
+/*!
+ * @internal
+ * @brief Processes the last 0-15 bytes of @p ptr.
+ *
+ * There may be up to 15 bytes remaining to consume from the input.
+ * This final stage will digest them to ensure that all input bytes are present
+ * in the final mix.
+ *
+ * @param h32 The hash to finalize.
+ * @param ptr The pointer to the remaining input.
+ * @param len The remaining length, modulo 16.
+ * @param align Whether @p ptr is aligned.
+ * @return The finalized hash.
+ */
+static xxh_u32
+XXH32_finalize(xxh_u32 h32, const xxh_u8* ptr, size_t len, XXH_alignment align)
+{
+#define XXH_PROCESS1 do {                           \
+    h32 += (*ptr++) * XXH_PRIME32_5;                \
+    h32 = XXH_rotl32(h32, 11) * XXH_PRIME32_1;      \
+} while (0)
+
+#define XXH_PROCESS4 do {                           \
+    h32 += XXH_get32bits(ptr) * XXH_PRIME32_3;      \
+    ptr += 4;                                   \
+    h32  = XXH_rotl32(h32, 17) * XXH_PRIME32_4;     \
+} while (0)
+
+    if (ptr==NULL) XXH_ASSERT(len == 0);
+
+    /* Compact rerolled version; generally faster */
+    if (!XXH32_ENDJMP) {
+        len &= 15;
+        while (len >= 4) {
+            XXH_PROCESS4;
+            len -= 4;
+        }
+        while (len > 0) {
+            XXH_PROCESS1;
+            --len;
+        }
+        return XXH32_avalanche(h32);
+    } else {
+         switch(len&15) /* or switch(bEnd - p) */ {
+           case 12:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 8:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 4:       XXH_PROCESS4;
+                         return XXH32_avalanche(h32);
+
+           case 13:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 9:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 5:       XXH_PROCESS4;
+                         XXH_PROCESS1;
+                         return XXH32_avalanche(h32);
+
+           case 14:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 10:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 6:       XXH_PROCESS4;
+                         XXH_PROCESS1;
+                         XXH_PROCESS1;
+                         return XXH32_avalanche(h32);
+
+           case 15:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 11:      XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 7:       XXH_PROCESS4;
+                         XXH_FALLTHROUGH;
+           case 3:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;
+           case 2:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;
+           case 1:       XXH_PROCESS1;
+                         XXH_FALLTHROUGH;
+           case 0:       return XXH32_avalanche(h32);
+        }
+        XXH_ASSERT(0);
+        return h32;   /* reaching this point is deemed impossible */
+    }
+}
+
+#ifdef XXH_OLD_NAMES
+#  define PROCESS1 XXH_PROCESS1
+#  define PROCESS4 XXH_PROCESS4
+#else
+#  undef XXH_PROCESS1
+#  undef XXH_PROCESS4
+#endif
+
+/*!
+ * @internal
+ * @brief The implementation for @ref XXH32().
+ *
+ * @param input , len , seed Directly passed from @ref XXH32().
+ * @param align Whether @p input is aligned.
+ * @return The calculated hash.
+ */
+XXH_FORCE_INLINE xxh_u32
+XXH32_endian_align(const xxh_u8* input, size_t len, xxh_u32 seed, XXH_alignment align)
+{
+    xxh_u32 h32;
+
+    if (input==NULL) XXH_ASSERT(len == 0);
+
+    if (len>=16) {
+        const xxh_u8* const bEnd = input + len;
+        const xxh_u8* const limit = bEnd - 15;
+        xxh_u32 v1 = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+        xxh_u32 v2 = seed + XXH_PRIME32_2;
+        xxh_u32 v3 = seed + 0;
+        xxh_u32 v4 = seed - XXH_PRIME32_1;
+
+        do {
+            v1 = XXH32_round(v1, XXH_get32bits(input)); input += 4;
+            v2 = XXH32_round(v2, XXH_get32bits(input)); input += 4;
+            v3 = XXH32_round(v3, XXH_get32bits(input)); input += 4;
+            v4 = XXH32_round(v4, XXH_get32bits(input)); input += 4;
+        } while (input < limit);
+
+        h32 = XXH_rotl32(v1, 1)  + XXH_rotl32(v2, 7)
+            + XXH_rotl32(v3, 12) + XXH_rotl32(v4, 18);
+    } else {
+        h32  = seed + XXH_PRIME32_5;
+    }
+
+    h32 += (xxh_u32)len;
+
+    return XXH32_finalize(h32, input, len&15, align);
+}
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32 (const void* input, size_t len, XXH32_hash_t seed)
+{
+#if 0
+    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
+    XXH32_state_t state;
+    XXH32_reset(&state, seed);
+    XXH32_update(&state, (const xxh_u8*)input, len);
+    return XXH32_digest(&state);
+#else
+    if (XXH_FORCE_ALIGN_CHECK) {
+        if ((((size_t)input) & 3) == 0) {   /* Input is 4-bytes aligned, leverage the speed benefit */
+            return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
+    }   }
+
+    return XXH32_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
+#endif
+}
+
+
+
+/*******   Hash streaming   *******/
+/*!
+ * @ingroup xxh32_family
+ */
+XXH_PUBLIC_API XXH32_state_t* XXH32_createState(void)
+{
+    return (XXH32_state_t*)XXH_malloc(sizeof(XXH32_state_t));
+}
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH_errorcode XXH32_freeState(XXH32_state_t* statePtr)
+{
+    XXH_free(statePtr);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API void XXH32_copyState(XXH32_state_t* dstState, const XXH32_state_t* srcState)
+{
+    XXH_memcpy(dstState, srcState, sizeof(*dstState));
+}
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH_errorcode XXH32_reset(XXH32_state_t* statePtr, XXH32_hash_t seed)
+{
+    XXH32_state_t state;   /* using a local state to memcpy() in order to avoid strict-aliasing warnings */
+    memset(&state, 0, sizeof(state));
+    state.v[0] = seed + XXH_PRIME32_1 + XXH_PRIME32_2;
+    state.v[1] = seed + XXH_PRIME32_2;
+    state.v[2] = seed + 0;
+    state.v[3] = seed - XXH_PRIME32_1;
+    /* do not write into reserved, planned to be removed in a future version */
+    XXH_memcpy(statePtr, &state, sizeof(state) - sizeof(state.reserved));
+    return XXH_OK;
+}
+
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH32_update(XXH32_state_t* state, const void* input, size_t len)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
+    }
+
+    {   const xxh_u8* p = (const xxh_u8*)input;
+        const xxh_u8* const bEnd = p + len;
+
+        state->total_len_32 += (XXH32_hash_t)len;
+        state->large_len |= (XXH32_hash_t)((len>=16) | (state->total_len_32>=16));
+
+        if (state->memsize + len < 16)  {   /* fill in tmp buffer */
+            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, len);
+            state->memsize += (XXH32_hash_t)len;
+            return XXH_OK;
+        }
+
+        if (state->memsize) {   /* some data left from previous update */
+            XXH_memcpy((xxh_u8*)(state->mem32) + state->memsize, input, 16-state->memsize);
+            {   const xxh_u32* p32 = state->mem32;
+                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p32)); p32++;
+                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p32)); p32++;
+                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p32)); p32++;
+                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p32));
+            }
+            p += 16-state->memsize;
+            state->memsize = 0;
+        }
+
+        if (p <= bEnd-16) {
+            const xxh_u8* const limit = bEnd - 16;
+
+            do {
+                state->v[0] = XXH32_round(state->v[0], XXH_readLE32(p)); p+=4;
+                state->v[1] = XXH32_round(state->v[1], XXH_readLE32(p)); p+=4;
+                state->v[2] = XXH32_round(state->v[2], XXH_readLE32(p)); p+=4;
+                state->v[3] = XXH32_round(state->v[3], XXH_readLE32(p)); p+=4;
+            } while (p<=limit);
+
+        }
+
+        if (p < bEnd) {
+            XXH_memcpy(state->mem32, p, (size_t)(bEnd-p));
+            state->memsize = (unsigned)(bEnd-p);
+        }
+    }
+
+    return XXH_OK;
+}
+
+
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32_digest(const XXH32_state_t* state)
+{
+    xxh_u32 h32;
+
+    if (state->large_len) {
+        h32 = XXH_rotl32(state->v[0], 1)
+            + XXH_rotl32(state->v[1], 7)
+            + XXH_rotl32(state->v[2], 12)
+            + XXH_rotl32(state->v[3], 18);
+    } else {
+        h32 = state->v[2] /* == seed */ + XXH_PRIME32_5;
+    }
+
+    h32 += state->total_len_32;
+
+    return XXH32_finalize(h32, (const xxh_u8*)state->mem32, state->memsize, XXH_aligned);
+}
+
+
+/*******   Canonical representation   *******/
+
+/*!
+ * @ingroup xxh32_family
+ * The default return values from XXH functions are unsigned 32 and 64 bit
+ * integers.
+ *
+ * The canonical representation uses big endian convention, the same convention
+ * as human-readable numbers (large digits first).
+ *
+ * This way, hash values can be written into a file or buffer, remaining
+ * comparable across different systems.
+ *
+ * The following functions allow transformation of hash values to and from their
+ * canonical format.
+ */
+XXH_PUBLIC_API void XXH32_canonicalFromHash(XXH32_canonical_t* dst, XXH32_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH32_canonical_t) == sizeof(XXH32_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap32(hash);
+    XXH_memcpy(dst, &hash, sizeof(*dst));
+}
+/*! @ingroup xxh32_family */
+XXH_PUBLIC_API XXH32_hash_t XXH32_hashFromCanonical(const XXH32_canonical_t* src)
+{
+    return XXH_readBE32(src);
+}
+
+
+#ifndef XXH_NO_LONG_LONG
+
+/* *******************************************************************
+*  64-bit hash functions
+*********************************************************************/
+/*!
+ * @}
+ * @ingroup impl
+ * @{
+ */
+/*******   Memory access   *******/
+
+typedef XXH64_hash_t xxh_u64;
+
+#ifdef XXH_OLD_NAMES
+#  define U64 xxh_u64
+#endif
+
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+/*
+ * Manual byteshift. Best for old compilers which don't inline memcpy.
+ * We actually directly use XXH_readLE64 and XXH_readBE64.
+ */
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==2))
+
+/* Force direct memory access. Only works on CPU which support unaligned memory access in hardware */
+static xxh_u64 XXH_read64(const void* memPtr)
+{
+    return *(const xxh_u64*) memPtr;
+}
+
+#elif (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==1))
+
+/*
+ * __pack instructions are safer, but compiler specific, hence potentially
+ * problematic for some compilers.
+ *
+ * Currently only defined for GCC and ICC.
+ */
+#ifdef XXH_OLD_NAMES
+typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) unalign64;
+#endif
+static xxh_u64 XXH_read64(const void* ptr)
+{
+    typedef union { xxh_u32 u32; xxh_u64 u64; } __attribute__((packed)) xxh_unalign64;
+    return ((const xxh_unalign64*)ptr)->u64;
+}
+
+#else
+
+/*
+ * Portable and safe solution. Generally efficient.
+ * see: http://fastcompression.blogspot.com/2015/08/accessing-unaligned-memory.html
+ */
+static xxh_u64 XXH_read64(const void* memPtr)
+{
+    xxh_u64 val;
+    XXH_memcpy(&val, memPtr, sizeof(val));
+    return val;
+}
+
+#endif   /* XXH_FORCE_DIRECT_MEMORY_ACCESS */
+
+#if defined(_MSC_VER)     /* Visual Studio */
+#  define XXH_swap64 _byteswap_uint64
+#elif XXH_GCC_VERSION >= 403
+#  define XXH_swap64 __builtin_bswap64
+#else
+static xxh_u64 XXH_swap64(xxh_u64 x)
+{
+    return  ((x << 56) & 0xff00000000000000ULL) |
+            ((x << 40) & 0x00ff000000000000ULL) |
+            ((x << 24) & 0x0000ff0000000000ULL) |
+            ((x << 8)  & 0x000000ff00000000ULL) |
+            ((x >> 8)  & 0x00000000ff000000ULL) |
+            ((x >> 24) & 0x0000000000ff0000ULL) |
+            ((x >> 40) & 0x000000000000ff00ULL) |
+            ((x >> 56) & 0x00000000000000ffULL);
+}
+#endif
+
+
+/* XXH_FORCE_MEMORY_ACCESS==3 is an endian-independent byteshift load. */
+#if (defined(XXH_FORCE_MEMORY_ACCESS) && (XXH_FORCE_MEMORY_ACCESS==3))
+
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[0]
+         | ((xxh_u64)bytePtr[1] << 8)
+         | ((xxh_u64)bytePtr[2] << 16)
+         | ((xxh_u64)bytePtr[3] << 24)
+         | ((xxh_u64)bytePtr[4] << 32)
+         | ((xxh_u64)bytePtr[5] << 40)
+         | ((xxh_u64)bytePtr[6] << 48)
+         | ((xxh_u64)bytePtr[7] << 56);
+}
+
+XXH_FORCE_INLINE xxh_u64 XXH_readBE64(const void* memPtr)
+{
+    const xxh_u8* bytePtr = (const xxh_u8 *)memPtr;
+    return bytePtr[7]
+         | ((xxh_u64)bytePtr[6] << 8)
+         | ((xxh_u64)bytePtr[5] << 16)
+         | ((xxh_u64)bytePtr[4] << 24)
+         | ((xxh_u64)bytePtr[3] << 32)
+         | ((xxh_u64)bytePtr[2] << 40)
+         | ((xxh_u64)bytePtr[1] << 48)
+         | ((xxh_u64)bytePtr[0] << 56);
+}
+
+#else
+XXH_FORCE_INLINE xxh_u64 XXH_readLE64(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_read64(ptr) : XXH_swap64(XXH_read64(ptr));
+}
+
+static xxh_u64 XXH_readBE64(const void* ptr)
+{
+    return XXH_CPU_LITTLE_ENDIAN ? XXH_swap64(XXH_read64(ptr)) : XXH_read64(ptr);
+}
+#endif
+
+XXH_FORCE_INLINE xxh_u64
+XXH_readLE64_align(const void* ptr, XXH_alignment align)
+{
+    if (align==XXH_unaligned)
+        return XXH_readLE64(ptr);
+    else
+        return XXH_CPU_LITTLE_ENDIAN ? *(const xxh_u64*)ptr : XXH_swap64(*(const xxh_u64*)ptr);
+}
+
+
+/*******   xxh64   *******/
+/*!
+ * @}
+ * @defgroup xxh64_impl XXH64 implementation
+ * @ingroup impl
+ * @{
+ */
+/* #define rather that static const, to be used as initializers */
+#define XXH_PRIME64_1  0x9E3779B185EBCA87ULL  /*!< 0b1001111000110111011110011011000110000101111010111100101010000111 */
+#define XXH_PRIME64_2  0xC2B2AE3D27D4EB4FULL  /*!< 0b1100001010110010101011100011110100100111110101001110101101001111 */
+#define XXH_PRIME64_3  0x165667B19E3779F9ULL  /*!< 0b0001011001010110011001111011000110011110001101110111100111111001 */
+#define XXH_PRIME64_4  0x85EBCA77C2B2AE63ULL  /*!< 0b1000010111101011110010100111011111000010101100101010111001100011 */
+#define XXH_PRIME64_5  0x27D4EB2F165667C5ULL  /*!< 0b0010011111010100111010110010111100010110010101100110011111000101 */
+
+#ifdef XXH_OLD_NAMES
+#  define PRIME64_1 XXH_PRIME64_1
+#  define PRIME64_2 XXH_PRIME64_2
+#  define PRIME64_3 XXH_PRIME64_3
+#  define PRIME64_4 XXH_PRIME64_4
+#  define PRIME64_5 XXH_PRIME64_5
+#endif
+
+static xxh_u64 XXH64_round(xxh_u64 acc, xxh_u64 input)
+{
+    acc += input * XXH_PRIME64_2;
+    acc  = XXH_rotl64(acc, 31);
+    acc *= XXH_PRIME64_1;
+    return acc;
+}
+
+static xxh_u64 XXH64_mergeRound(xxh_u64 acc, xxh_u64 val)
+{
+    val  = XXH64_round(0, val);
+    acc ^= val;
+    acc  = acc * XXH_PRIME64_1 + XXH_PRIME64_4;
+    return acc;
+}
+
+static xxh_u64 XXH64_avalanche(xxh_u64 h64)
+{
+    h64 ^= h64 >> 33;
+    h64 *= XXH_PRIME64_2;
+    h64 ^= h64 >> 29;
+    h64 *= XXH_PRIME64_3;
+    h64 ^= h64 >> 32;
+    return h64;
+}
+
+
+#define XXH_get64bits(p) XXH_readLE64_align(p, align)
+
+static xxh_u64
+XXH64_finalize(xxh_u64 h64, const xxh_u8* ptr, size_t len, XXH_alignment align)
+{
+    if (ptr==NULL) XXH_ASSERT(len == 0);
+    len &= 31;
+    while (len >= 8) {
+        xxh_u64 const k1 = XXH64_round(0, XXH_get64bits(ptr));
+        ptr += 8;
+        h64 ^= k1;
+        h64  = XXH_rotl64(h64,27) * XXH_PRIME64_1 + XXH_PRIME64_4;
+        len -= 8;
+    }
+    if (len >= 4) {
+        h64 ^= (xxh_u64)(XXH_get32bits(ptr)) * XXH_PRIME64_1;
+        ptr += 4;
+        h64 = XXH_rotl64(h64, 23) * XXH_PRIME64_2 + XXH_PRIME64_3;
+        len -= 4;
+    }
+    while (len > 0) {
+        h64 ^= (*ptr++) * XXH_PRIME64_5;
+        h64 = XXH_rotl64(h64, 11) * XXH_PRIME64_1;
+        --len;
+    }
+    return  XXH64_avalanche(h64);
+}
+
+#ifdef XXH_OLD_NAMES
+#  define PROCESS1_64 XXH_PROCESS1_64
+#  define PROCESS4_64 XXH_PROCESS4_64
+#  define PROCESS8_64 XXH_PROCESS8_64
+#else
+#  undef XXH_PROCESS1_64
+#  undef XXH_PROCESS4_64
+#  undef XXH_PROCESS8_64
+#endif
+
+XXH_FORCE_INLINE xxh_u64
+XXH64_endian_align(const xxh_u8* input, size_t len, xxh_u64 seed, XXH_alignment align)
+{
+    xxh_u64 h64;
+    if (input==NULL) XXH_ASSERT(len == 0);
+
+    if (len>=32) {
+        const xxh_u8* const bEnd = input + len;
+        const xxh_u8* const limit = bEnd - 31;
+        xxh_u64 v1 = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+        xxh_u64 v2 = seed + XXH_PRIME64_2;
+        xxh_u64 v3 = seed + 0;
+        xxh_u64 v4 = seed - XXH_PRIME64_1;
+
+        do {
+            v1 = XXH64_round(v1, XXH_get64bits(input)); input+=8;
+            v2 = XXH64_round(v2, XXH_get64bits(input)); input+=8;
+            v3 = XXH64_round(v3, XXH_get64bits(input)); input+=8;
+            v4 = XXH64_round(v4, XXH_get64bits(input)); input+=8;
+        } while (input<limit);
+
+        h64 = XXH_rotl64(v1, 1) + XXH_rotl64(v2, 7) + XXH_rotl64(v3, 12) + XXH_rotl64(v4, 18);
+        h64 = XXH64_mergeRound(h64, v1);
+        h64 = XXH64_mergeRound(h64, v2);
+        h64 = XXH64_mergeRound(h64, v3);
+        h64 = XXH64_mergeRound(h64, v4);
+
+    } else {
+        h64  = seed + XXH_PRIME64_5;
+    }
+
+    h64 += (xxh_u64) len;
+
+    return XXH64_finalize(h64, input, len, align);
+}
+
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64 (const void* input, size_t len, XXH64_hash_t seed)
+{
+#if 0
+    /* Simple version, good for code maintenance, but unfortunately slow for small inputs */
+    XXH64_state_t state;
+    XXH64_reset(&state, seed);
+    XXH64_update(&state, (const xxh_u8*)input, len);
+    return XXH64_digest(&state);
+#else
+    if (XXH_FORCE_ALIGN_CHECK) {
+        if ((((size_t)input) & 7)==0) {  /* Input is aligned, let's leverage the speed advantage */
+            return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_aligned);
+    }   }
+
+    return XXH64_endian_align((const xxh_u8*)input, len, seed, XXH_unaligned);
+
+#endif
+}
+
+/*******   Hash Streaming   *******/
+
+/*! @ingroup xxh64_family*/
+XXH_PUBLIC_API XXH64_state_t* XXH64_createState(void)
+{
+    return (XXH64_state_t*)XXH_malloc(sizeof(XXH64_state_t));
+}
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH_errorcode XXH64_freeState(XXH64_state_t* statePtr)
+{
+    XXH_free(statePtr);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API void XXH64_copyState(XXH64_state_t* dstState, const XXH64_state_t* srcState)
+{
+    XXH_memcpy(dstState, srcState, sizeof(*dstState));
+}
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH_errorcode XXH64_reset(XXH64_state_t* statePtr, XXH64_hash_t seed)
+{
+    XXH64_state_t state;   /* use a local state to memcpy() in order to avoid strict-aliasing warnings */
+    memset(&state, 0, sizeof(state));
+    state.v[0] = seed + XXH_PRIME64_1 + XXH_PRIME64_2;
+    state.v[1] = seed + XXH_PRIME64_2;
+    state.v[2] = seed + 0;
+    state.v[3] = seed - XXH_PRIME64_1;
+     /* do not write into reserved64, might be removed in a future version */
+    XXH_memcpy(statePtr, &state, sizeof(state) - sizeof(state.reserved64));
+    return XXH_OK;
+}
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH64_update (XXH64_state_t* state, const void* input, size_t len)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
+    }
+
+    {   const xxh_u8* p = (const xxh_u8*)input;
+        const xxh_u8* const bEnd = p + len;
+
+        state->total_len += len;
+
+        if (state->memsize + len < 32) {  /* fill in tmp buffer */
+            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, len);
+            state->memsize += (xxh_u32)len;
+            return XXH_OK;
+        }
+
+        if (state->memsize) {   /* tmp buffer is full */
+            XXH_memcpy(((xxh_u8*)state->mem64) + state->memsize, input, 32-state->memsize);
+            state->v[0] = XXH64_round(state->v[0], XXH_readLE64(state->mem64+0));
+            state->v[1] = XXH64_round(state->v[1], XXH_readLE64(state->mem64+1));
+            state->v[2] = XXH64_round(state->v[2], XXH_readLE64(state->mem64+2));
+            state->v[3] = XXH64_round(state->v[3], XXH_readLE64(state->mem64+3));
+            p += 32 - state->memsize;
+            state->memsize = 0;
+        }
+
+        if (p+32 <= bEnd) {
+            const xxh_u8* const limit = bEnd - 32;
+
+            do {
+                state->v[0] = XXH64_round(state->v[0], XXH_readLE64(p)); p+=8;
+                state->v[1] = XXH64_round(state->v[1], XXH_readLE64(p)); p+=8;
+                state->v[2] = XXH64_round(state->v[2], XXH_readLE64(p)); p+=8;
+                state->v[3] = XXH64_round(state->v[3], XXH_readLE64(p)); p+=8;
+            } while (p<=limit);
+
+        }
+
+        if (p < bEnd) {
+            XXH_memcpy(state->mem64, p, (size_t)(bEnd-p));
+            state->memsize = (unsigned)(bEnd-p);
+        }
+    }
+
+    return XXH_OK;
+}
+
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64_digest(const XXH64_state_t* state)
+{
+    xxh_u64 h64;
+
+    if (state->total_len >= 32) {
+        h64 = XXH_rotl64(state->v[0], 1) + XXH_rotl64(state->v[1], 7) + XXH_rotl64(state->v[2], 12) + XXH_rotl64(state->v[3], 18);
+        h64 = XXH64_mergeRound(h64, state->v[0]);
+        h64 = XXH64_mergeRound(h64, state->v[1]);
+        h64 = XXH64_mergeRound(h64, state->v[2]);
+        h64 = XXH64_mergeRound(h64, state->v[3]);
+    } else {
+        h64  = state->v[2] /*seed*/ + XXH_PRIME64_5;
+    }
+
+    h64 += (xxh_u64) state->total_len;
+
+    return XXH64_finalize(h64, (const xxh_u8*)state->mem64, (size_t)state->total_len, XXH_aligned);
+}
+
+
+/******* Canonical representation   *******/
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API void XXH64_canonicalFromHash(XXH64_canonical_t* dst, XXH64_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH64_canonical_t) == sizeof(XXH64_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) hash = XXH_swap64(hash);
+    XXH_memcpy(dst, &hash, sizeof(*dst));
+}
+
+/*! @ingroup xxh64_family */
+XXH_PUBLIC_API XXH64_hash_t XXH64_hashFromCanonical(const XXH64_canonical_t* src)
+{
+    return XXH_readBE64(src);
+}
+
+#ifndef XXH_NO_XXH3
+
+/* *********************************************************************
+*  XXH3
+*  New generation hash designed for speed on small keys and vectorization
+************************************************************************ */
+/*!
+ * @}
+ * @defgroup xxh3_impl XXH3 implementation
+ * @ingroup impl
+ * @{
+ */
+
+/* ===   Compiler specifics   === */
+
+#if ((defined(sun) || defined(__sun)) && __cplusplus) /* Solaris includes __STDC_VERSION__ with C++. Tested with GCC 5.5 */
+#  define XXH_RESTRICT /* disable */
+#elif defined (__STDC_VERSION__) && __STDC_VERSION__ >= 199901L   /* >= C99 */
+#  define XXH_RESTRICT   restrict
+#else
+/* Note: it might be useful to define __restrict or __restrict__ for some C++ compilers */
+#  define XXH_RESTRICT   /* disable */
+#endif
+
+#if (defined(__GNUC__) && (__GNUC__ >= 3))  \
+  || (defined(__INTEL_COMPILER) && (__INTEL_COMPILER >= 800)) \
+  || defined(__clang__)
+#    define XXH_likely(x) __builtin_expect(x, 1)
+#    define XXH_unlikely(x) __builtin_expect(x, 0)
+#else
+#    define XXH_likely(x) (x)
+#    define XXH_unlikely(x) (x)
+#endif
+
+#if defined(__GNUC__)
+#  if defined(__AVX2__)
+#    include <immintrin.h>
+#  elif defined(__SSE2__)
+#    include <emmintrin.h>
+#  elif defined(__ARM_NEON__) || defined(__ARM_NEON)
+#    define inline __inline__  /* circumvent a clang bug */
+#    include <arm_neon.h>
+#    undef inline
+#  endif
+#elif defined(_MSC_VER)
+#  include <intrin.h>
+#endif
+
+/*
+ * One goal of XXH3 is to make it fast on both 32-bit and 64-bit, while
+ * remaining a true 64-bit/128-bit hash function.
+ *
+ * This is done by prioritizing a subset of 64-bit operations that can be
+ * emulated without too many steps on the average 32-bit machine.
+ *
+ * For example, these two lines seem similar, and run equally fast on 64-bit:
+ *
+ *   xxh_u64 x;
+ *   x ^= (x >> 47); // good
+ *   x ^= (x >> 13); // bad
+ *
+ * However, to a 32-bit machine, there is a major difference.
+ *
+ * x ^= (x >> 47) looks like this:
+ *
+ *   x.lo ^= (x.hi >> (47 - 32));
+ *
+ * while x ^= (x >> 13) looks like this:
+ *
+ *   // note: funnel shifts are not usually cheap.
+ *   x.lo ^= (x.lo >> 13) | (x.hi << (32 - 13));
+ *   x.hi ^= (x.hi >> 13);
+ *
+ * The first one is significantly faster than the second, simply because the
+ * shift is larger than 32. This means:
+ *  - All the bits we need are in the upper 32 bits, so we can ignore the lower
+ *    32 bits in the shift.
+ *  - The shift result will always fit in the lower 32 bits, and therefore,
+ *    we can ignore the upper 32 bits in the xor.
+ *
+ * Thanks to this optimization, XXH3 only requires these features to be efficient:
+ *
+ *  - Usable unaligned access
+ *  - A 32-bit or 64-bit ALU
+ *      - If 32-bit, a decent ADC instruction
+ *  - A 32 or 64-bit multiply with a 64-bit result
+ *  - For the 128-bit variant, a decent byteswap helps short inputs.
+ *
+ * The first two are already required by XXH32, and almost all 32-bit and 64-bit
+ * platforms which can run XXH32 can run XXH3 efficiently.
+ *
+ * Thumb-1, the classic 16-bit only subset of ARM's instruction set, is one
+ * notable exception.
+ *
+ * First of all, Thumb-1 lacks support for the UMULL instruction which
+ * performs the important long multiply. This means numerous __aeabi_lmul
+ * calls.
+ *
+ * Second of all, the 8 functional registers are just not enough.
+ * Setup for __aeabi_lmul, byteshift loads, pointers, and all arithmetic need
+ * Lo registers, and this shuffling results in thousands more MOVs than A32.
+ *
+ * A32 and T32 don't have this limitation. They can access all 14 registers,
+ * do a 32->64 multiply with UMULL, and the flexible operand allowing free
+ * shifts is helpful, too.
+ *
+ * Therefore, we do a quick sanity check.
+ *
+ * If compiling Thumb-1 for a target which supports ARM instructions, we will
+ * emit a warning, as it is not a "sane" platform to compile for.
+ *
+ * Usually, if this happens, it is because of an accident and you probably need
+ * to specify -march, as you likely meant to compile for a newer architecture.
+ *
+ * Credit: large sections of the vectorial and asm source code paths
+ *         have been contributed by @easyaspi314
+ */
+#if defined(__thumb__) && !defined(__thumb2__) && defined(__ARM_ARCH_ISA_ARM)
+#   warning "XXH3 is highly inefficient without ARM or Thumb-2."
+#endif
+
+/* ==========================================
+ * Vectorization detection
+ * ========================================== */
+
+#ifdef XXH_DOXYGEN
+/*!
+ * @ingroup tuning
+ * @brief Overrides the vectorization implementation chosen for XXH3.
+ *
+ * Can be defined to 0 to disable SIMD or any of the values mentioned in
+ * @ref XXH_VECTOR_TYPE.
+ *
+ * If this is not defined, it uses predefined macros to determine the best
+ * implementation.
+ */
+#  define XXH_VECTOR XXH_SCALAR
+/*!
+ * @ingroup tuning
+ * @brief Possible values for @ref XXH_VECTOR.
+ *
+ * Note that these are actually implemented as macros.
+ *
+ * If this is not defined, it is detected automatically.
+ * @ref XXH_X86DISPATCH overrides this.
+ */
+enum XXH_VECTOR_TYPE /* fake enum */ {
+    XXH_SCALAR = 0,  /*!< Portable scalar version */
+    XXH_SSE2   = 1,  /*!<
+                      * SSE2 for Pentium 4, Opteron, all x86_64.
+                      *
+                      * @note SSE2 is also guaranteed on Windows 10, macOS, and
+                      * Android x86.
+                      */
+    XXH_AVX2   = 2,  /*!< AVX2 for Haswell and Bulldozer */
+    XXH_AVX512 = 3,  /*!< AVX512 for Skylake and Icelake */
+    XXH_NEON   = 4,  /*!< NEON for most ARMv7-A and all AArch64 */
+    XXH_VSX    = 5,  /*!< VSX and ZVector for POWER8/z13 (64-bit) */
+};
+/*!
+ * @ingroup tuning
+ * @brief Selects the minimum alignment for XXH3's accumulators.
+ *
+ * When using SIMD, this should match the alignment reqired for said vector
+ * type, so, for example, 32 for AVX2.
+ *
+ * Default: Auto detected.
+ */
+#  define XXH_ACC_ALIGN 8
+#endif
+
+/* Actual definition */
+#ifndef XXH_DOXYGEN
+#  define XXH_SCALAR 0
+#  define XXH_SSE2   1
+#  define XXH_AVX2   2
+#  define XXH_AVX512 3
+#  define XXH_NEON   4
+#  define XXH_VSX    5
+#endif
+
+#ifndef XXH_VECTOR    /* can be defined on command line */
+#  if defined(__AVX512F__)
+#    define XXH_VECTOR XXH_AVX512
+#  elif defined(__AVX2__)
+#    define XXH_VECTOR XXH_AVX2
+#  elif defined(__SSE2__) || defined(_M_AMD64) || defined(_M_X64) || (defined(_M_IX86_FP) && (_M_IX86_FP == 2))
+#    define XXH_VECTOR XXH_SSE2
+#  elif ( \
+        defined(__ARM_NEON__) || defined(__ARM_NEON) /* gcc */ \
+     || defined(_M_ARM64) || defined(_M_ARM_ARMV7VE) /* msvc */ \
+   ) && ( \
+        defined(_WIN32) || defined(__LITTLE_ENDIAN__) /* little endian only */ \
+    || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) \
+   )
+#    define XXH_VECTOR XXH_NEON
+#  elif (defined(__PPC64__) && defined(__POWER8_VECTOR__)) \
+     || (defined(__s390x__) && defined(__VEC__)) \
+     && defined(__GNUC__) /* TODO: IBM XL */
+#    define XXH_VECTOR XXH_VSX
+#  else
+#    define XXH_VECTOR XXH_SCALAR
+#  endif
+#endif
+
+/*
+ * Controls the alignment of the accumulator,
+ * for compatibility with aligned vector loads, which are usually faster.
+ */
+#ifndef XXH_ACC_ALIGN
+#  if defined(XXH_X86DISPATCH)
+#     define XXH_ACC_ALIGN 64  /* for compatibility with avx512 */
+#  elif XXH_VECTOR == XXH_SCALAR  /* scalar */
+#     define XXH_ACC_ALIGN 8
+#  elif XXH_VECTOR == XXH_SSE2  /* sse2 */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_AVX2  /* avx2 */
+#     define XXH_ACC_ALIGN 32
+#  elif XXH_VECTOR == XXH_NEON  /* neon */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_VSX   /* vsx */
+#     define XXH_ACC_ALIGN 16
+#  elif XXH_VECTOR == XXH_AVX512  /* avx512 */
+#     define XXH_ACC_ALIGN 64
+#  endif
+#endif
+
+#if defined(XXH_X86DISPATCH) || XXH_VECTOR == XXH_SSE2 \
+    || XXH_VECTOR == XXH_AVX2 || XXH_VECTOR == XXH_AVX512
+#  define XXH_SEC_ALIGN XXH_ACC_ALIGN
+#else
+#  define XXH_SEC_ALIGN 8
+#endif
+
+/*
+ * UGLY HACK:
+ * GCC usually generates the best code with -O3 for xxHash.
+ *
+ * However, when targeting AVX2, it is overzealous in its unrolling resulting
+ * in code roughly 3/4 the speed of Clang.
+ *
+ * There are other issues, such as GCC splitting _mm256_loadu_si256 into
+ * _mm_loadu_si128 + _mm256_inserti128_si256. This is an optimization which
+ * only applies to Sandy and Ivy Bridge... which don't even support AVX2.
+ *
+ * That is why when compiling the AVX2 version, it is recommended to use either
+ *   -O2 -mavx2 -march=haswell
+ * or
+ *   -O2 -mavx2 -mno-avx256-split-unaligned-load
+ * for decent performance, or to use Clang instead.
+ *
+ * Fortunately, we can control the first one with a pragma that forces GCC into
+ * -O2, but the other one we can't control without "failed to inline always
+ * inline function due to target mismatch" warnings.
+ */
+#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
+  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) /* respect -O0 and -Os */
+#  pragma GCC push_options
+#  pragma GCC optimize("-O2")
+#endif
+
+
+#if XXH_VECTOR == XXH_NEON
+/*
+ * NEON's setup for vmlal_u32 is a little more complicated than it is on
+ * SSE2, AVX2, and VSX.
+ *
+ * While PMULUDQ and VMULEUW both perform a mask, VMLAL.U32 performs an upcast.
+ *
+ * To do the same operation, the 128-bit 'Q' register needs to be split into
+ * two 64-bit 'D' registers, performing this operation::
+ *
+ *   [                a                 |                 b                ]
+ *            |              '---------. .--------'                |
+ *            |                         x                          |
+ *            |              .---------' '--------.                |
+ *   [ a & 0xFFFFFFFF | b & 0xFFFFFFFF ],[    a >> 32     |     b >> 32    ]
+ *
+ * Due to significant changes in aarch64, the fastest method for aarch64 is
+ * completely different than the fastest method for ARMv7-A.
+ *
+ * ARMv7-A treats D registers as unions overlaying Q registers, so modifying
+ * D11 will modify the high half of Q5. This is similar to how modifying AH
+ * will only affect bits 8-15 of AX on x86.
+ *
+ * VZIP takes two registers, and puts even lanes in one register and odd lanes
+ * in the other.
+ *
+ * On ARMv7-A, this strangely modifies both parameters in place instead of
+ * taking the usual 3-operand form.
+ *
+ * Therefore, if we want to do this, we can simply use a D-form VZIP.32 on the
+ * lower and upper halves of the Q register to end up with the high and low
+ * halves where we want - all in one instruction.
+ *
+ *   vzip.32   d10, d11       @ d10 = { d10[0], d11[0] }; d11 = { d10[1], d11[1] }
+ *
+ * Unfortunately we need inline assembly for this: Instructions modifying two
+ * registers at once is not possible in GCC or Clang's IR, and they have to
+ * create a copy.
+ *
+ * aarch64 requires a different approach.
+ *
+ * In order to make it easier to write a decent compiler for aarch64, many
+ * quirks were removed, such as conditional execution.
+ *
+ * NEON was also affected by this.
+ *
+ * aarch64 cannot access the high bits of a Q-form register, and writes to a
+ * D-form register zero the high bits, similar to how writes to W-form scalar
+ * registers (or DWORD registers on x86_64) work.
+ *
+ * The formerly free vget_high intrinsics now require a vext (with a few
+ * exceptions)
+ *
+ * Additionally, VZIP was replaced by ZIP1 and ZIP2, which are the equivalent
+ * of PUNPCKL* and PUNPCKH* in SSE, respectively, in order to only modify one
+ * operand.
+ *
+ * The equivalent of the VZIP.32 on the lower and upper halves would be this
+ * mess:
+ *
+ *   ext     v2.4s, v0.4s, v0.4s, #2 // v2 = { v0[2], v0[3], v0[0], v0[1] }
+ *   zip1    v1.2s, v0.2s, v2.2s     // v1 = { v0[0], v2[0] }
+ *   zip2    v0.2s, v0.2s, v1.2s     // v0 = { v0[1], v2[1] }
+ *
+ * Instead, we use a literal downcast, vmovn_u64 (XTN), and vshrn_n_u64 (SHRN):
+ *
+ *   shrn    v1.2s, v0.2d, #32  // v1 = (uint32x2_t)(v0 >> 32);
+ *   xtn     v0.2s, v0.2d       // v0 = (uint32x2_t)(v0 & 0xFFFFFFFF);
+ *
+ * This is available on ARMv7-A, but is less efficient than a single VZIP.32.
+ */
+
+/*!
+ * Function-like macro:
+ * void XXH_SPLIT_IN_PLACE(uint64x2_t &in, uint32x2_t &outLo, uint32x2_t &outHi)
+ * {
+ *     outLo = (uint32x2_t)(in & 0xFFFFFFFF);
+ *     outHi = (uint32x2_t)(in >> 32);
+ *     in = UNDEFINED;
+ * }
+ */
+# if !defined(XXH_NO_VZIP_HACK) /* define to disable */ \
+   && defined(__GNUC__) \
+   && !defined(__aarch64__) && !defined(__arm64__) && !defined(_M_ARM64)
+#  define XXH_SPLIT_IN_PLACE(in, outLo, outHi)                                              \
+    do {                                                                                    \
+      /* Undocumented GCC/Clang operand modifier: %e0 = lower D half, %f0 = upper D half */ \
+      /* https://github.com/gcc-mirror/gcc/blob/38cf91e5/gcc/config/arm/arm.c#L22486 */     \
+      /* https://github.com/llvm-mirror/llvm/blob/2c4ca683/lib/Target/ARM/ARMAsmPrinter.cpp#L399 */ \
+      __asm__("vzip.32  %e0, %f0" : "+w" (in));                                             \
+      (outLo) = vget_low_u32 (vreinterpretq_u32_u64(in));                                   \
+      (outHi) = vget_high_u32(vreinterpretq_u32_u64(in));                                   \
+   } while (0)
+# else
+#  define XXH_SPLIT_IN_PLACE(in, outLo, outHi)                                            \
+    do {                                                                                  \
+      (outLo) = vmovn_u64    (in);                                                        \
+      (outHi) = vshrn_n_u64  ((in), 32);                                                  \
+    } while (0)
+# endif
+#endif  /* XXH_VECTOR == XXH_NEON */
+
+/*
+ * VSX and Z Vector helpers.
+ *
+ * This is very messy, and any pull requests to clean this up are welcome.
+ *
+ * There are a lot of problems with supporting VSX and s390x, due to
+ * inconsistent intrinsics, spotty coverage, and multiple endiannesses.
+ */
+#if XXH_VECTOR == XXH_VSX
+#  if defined(__s390x__)
+#    include <s390intrin.h>
+#  else
+/* gcc's altivec.h can have the unwanted consequence to unconditionally
+ * #define bool, vector, and pixel keywords,
+ * with bad consequences for programs already using these keywords for other purposes.
+ * The paragraph defining these macros is skipped when __APPLE_ALTIVEC__ is defined.
+ * __APPLE_ALTIVEC__ is _generally_ defined automatically by the compiler,
+ * but it seems that, in some cases, it isn't.
+ * Force the build macro to be defined, so that keywords are not altered.
+ */
+#    if defined(__GNUC__) && !defined(__APPLE_ALTIVEC__)
+#      define __APPLE_ALTIVEC__
+#    endif
+#    include <altivec.h>
+#  endif
+
+typedef __vector unsigned long long xxh_u64x2;
+typedef __vector unsigned char xxh_u8x16;
+typedef __vector unsigned xxh_u32x4;
+
+# ifndef XXH_VSX_BE
+#  if defined(__BIG_ENDIAN__) \
+  || (defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
+#    define XXH_VSX_BE 1
+#  elif defined(__VEC_ELEMENT_REG_ORDER__) && __VEC_ELEMENT_REG_ORDER__ == __ORDER_BIG_ENDIAN__
+#    warning "-maltivec=be is not recommended. Please use native endianness."
+#    define XXH_VSX_BE 1
+#  else
+#    define XXH_VSX_BE 0
+#  endif
+# endif /* !defined(XXH_VSX_BE) */
+
+# if XXH_VSX_BE
+#  if defined(__POWER9_VECTOR__) || (defined(__clang__) && defined(__s390x__))
+#    define XXH_vec_revb vec_revb
+#  else
+/*!
+ * A polyfill for POWER9's vec_revb().
+ */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_revb(xxh_u64x2 val)
+{
+    xxh_u8x16 const vByteSwap = { 0x07, 0x06, 0x05, 0x04, 0x03, 0x02, 0x01, 0x00,
+                                  0x0F, 0x0E, 0x0D, 0x0C, 0x0B, 0x0A, 0x09, 0x08 };
+    return vec_perm(val, val, vByteSwap);
+}
+#  endif
+# endif /* XXH_VSX_BE */
+
+/*!
+ * Performs an unaligned vector load and byte swaps it on big endian.
+ */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_loadu(const void *ptr)
+{
+    xxh_u64x2 ret;
+    XXH_memcpy(&ret, ptr, sizeof(xxh_u64x2));
+# if XXH_VSX_BE
+    ret = XXH_vec_revb(ret);
+# endif
+    return ret;
+}
+
+/*
+ * vec_mulo and vec_mule are very problematic intrinsics on PowerPC
+ *
+ * These intrinsics weren't added until GCC 8, despite existing for a while,
+ * and they are endian dependent. Also, their meaning swap depending on version.
+ * */
+# if defined(__s390x__)
+ /* s390x is always big endian, no issue on this platform */
+#  define XXH_vec_mulo vec_mulo
+#  define XXH_vec_mule vec_mule
+# elif defined(__clang__) && XXH_HAS_BUILTIN(__builtin_altivec_vmuleuw)
+/* Clang has a better way to control this, we can just use the builtin which doesn't swap. */
+#  define XXH_vec_mulo __builtin_altivec_vmulouw
+#  define XXH_vec_mule __builtin_altivec_vmuleuw
+# else
+/* gcc needs inline assembly */
+/* Adapted from https://github.com/google/highwayhash/blob/master/highwayhash/hh_vsx.h. */
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mulo(xxh_u32x4 a, xxh_u32x4 b)
+{
+    xxh_u64x2 result;
+    __asm__("vmulouw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
+    return result;
+}
+XXH_FORCE_INLINE xxh_u64x2 XXH_vec_mule(xxh_u32x4 a, xxh_u32x4 b)
+{
+    xxh_u64x2 result;
+    __asm__("vmuleuw %0, %1, %2" : "=v" (result) : "v" (a), "v" (b));
+    return result;
+}
+# endif /* XXH_vec_mulo, XXH_vec_mule */
+#endif /* XXH_VECTOR == XXH_VSX */
+
+
+/* prefetch
+ * can be disabled, by declaring XXH_NO_PREFETCH build macro */
+#if defined(XXH_NO_PREFETCH)
+#  define XXH_PREFETCH(ptr)  (void)(ptr)  /* disabled */
+#else
+#  if defined(_MSC_VER) && (defined(_M_X64) || defined(_M_IX86))  /* _mm_prefetch() not defined outside of x86/x64 */
+#    include <mmintrin.h>   /* https://msdn.microsoft.com/fr-fr/library/84szxsww(v=vs.90).aspx */
+#    define XXH_PREFETCH(ptr)  _mm_prefetch((const char*)(ptr), _MM_HINT_T0)
+#  elif defined(__GNUC__) && ( (__GNUC__ >= 4) || ( (__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) ) )
+#    define XXH_PREFETCH(ptr)  __builtin_prefetch((ptr), 0 /* rw==read */, 3 /* locality */)
+#  else
+#    define XXH_PREFETCH(ptr) (void)(ptr)  /* disabled */
+#  endif
+#endif  /* XXH_NO_PREFETCH */
+
+
+/* ==========================================
+ * XXH3 default settings
+ * ========================================== */
+
+#define XXH_SECRET_DEFAULT_SIZE 192   /* minimum XXH3_SECRET_SIZE_MIN */
+
+#if (XXH_SECRET_DEFAULT_SIZE < XXH3_SECRET_SIZE_MIN)
+#  error "default keyset is not large enough"
+#endif
+
+/*! Pseudorandom secret taken directly from FARSH. */
+XXH_ALIGN(64) static const xxh_u8 XXH3_kSecret[XXH_SECRET_DEFAULT_SIZE] = {
+    0xb8, 0xfe, 0x6c, 0x39, 0x23, 0xa4, 0x4b, 0xbe, 0x7c, 0x01, 0x81, 0x2c, 0xf7, 0x21, 0xad, 0x1c,
+    0xde, 0xd4, 0x6d, 0xe9, 0x83, 0x90, 0x97, 0xdb, 0x72, 0x40, 0xa4, 0xa4, 0xb7, 0xb3, 0x67, 0x1f,
+    0xcb, 0x79, 0xe6, 0x4e, 0xcc, 0xc0, 0xe5, 0x78, 0x82, 0x5a, 0xd0, 0x7d, 0xcc, 0xff, 0x72, 0x21,
+    0xb8, 0x08, 0x46, 0x74, 0xf7, 0x43, 0x24, 0x8e, 0xe0, 0x35, 0x90, 0xe6, 0x81, 0x3a, 0x26, 0x4c,
+    0x3c, 0x28, 0x52, 0xbb, 0x91, 0xc3, 0x00, 0xcb, 0x88, 0xd0, 0x65, 0x8b, 0x1b, 0x53, 0x2e, 0xa3,
+    0x71, 0x64, 0x48, 0x97, 0xa2, 0x0d, 0xf9, 0x4e, 0x38, 0x19, 0xef, 0x46, 0xa9, 0xde, 0xac, 0xd8,
+    0xa8, 0xfa, 0x76, 0x3f, 0xe3, 0x9c, 0x34, 0x3f, 0xf9, 0xdc, 0xbb, 0xc7, 0xc7, 0x0b, 0x4f, 0x1d,
+    0x8a, 0x51, 0xe0, 0x4b, 0xcd, 0xb4, 0x59, 0x31, 0xc8, 0x9f, 0x7e, 0xc9, 0xd9, 0x78, 0x73, 0x64,
+    0xea, 0xc5, 0xac, 0x83, 0x34, 0xd3, 0xeb, 0xc3, 0xc5, 0x81, 0xa0, 0xff, 0xfa, 0x13, 0x63, 0xeb,
+    0x17, 0x0d, 0xdd, 0x51, 0xb7, 0xf0, 0xda, 0x49, 0xd3, 0x16, 0x55, 0x26, 0x29, 0xd4, 0x68, 0x9e,
+    0x2b, 0x16, 0xbe, 0x58, 0x7d, 0x47, 0xa1, 0xfc, 0x8f, 0xf8, 0xb8, 0xd1, 0x7a, 0xd0, 0x31, 0xce,
+    0x45, 0xcb, 0x3a, 0x8f, 0x95, 0x16, 0x04, 0x28, 0xaf, 0xd7, 0xfb, 0xca, 0xbb, 0x4b, 0x40, 0x7e,
+};
+
+
+#ifdef XXH_OLD_NAMES
+#  define kSecret XXH3_kSecret
+#endif
+
+#ifdef XXH_DOXYGEN
+/*!
+ * @brief Calculates a 32-bit to 64-bit long multiply.
+ *
+ * Implemented as a macro.
+ *
+ * Wraps `__emulu` on MSVC x86 because it tends to call `__allmul` when it doesn't
+ * need to (but it shouldn't need to anyways, it is about 7 instructions to do
+ * a 64x64 multiply...). Since we know that this will _always_ emit `MULL`, we
+ * use that instead of the normal method.
+ *
+ * If you are compiling for platforms like Thumb-1 and don't have a better option,
+ * you may also want to write your own long multiply routine here.
+ *
+ * @param x, y Numbers to be multiplied
+ * @return 64-bit product of the low 32 bits of @p x and @p y.
+ */
+XXH_FORCE_INLINE xxh_u64
+XXH_mult32to64(xxh_u64 x, xxh_u64 y)
+{
+   return (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF);
+}
+#elif defined(_MSC_VER) && defined(_M_IX86)
+#    include <intrin.h>
+#    define XXH_mult32to64(x, y) __emulu((unsigned)(x), (unsigned)(y))
+#else
+/*
+ * Downcast + upcast is usually better than masking on older compilers like
+ * GCC 4.2 (especially 32-bit ones), all without affecting newer compilers.
+ *
+ * The other method, (x & 0xFFFFFFFF) * (y & 0xFFFFFFFF), will AND both operands
+ * and perform a full 64x64 multiply -- entirely redundant on 32-bit.
+ */
+#    define XXH_mult32to64(x, y) ((xxh_u64)(xxh_u32)(x) * (xxh_u64)(xxh_u32)(y))
+#endif
+
+/*!
+ * @brief Calculates a 64->128-bit long multiply.
+ *
+ * Uses `__uint128_t` and `_umul128` if available, otherwise uses a scalar
+ * version.
+ *
+ * @param lhs , rhs The 64-bit integers to be multiplied
+ * @return The 128-bit result represented in an @ref XXH128_hash_t.
+ */
+static XXH128_hash_t
+XXH_mult64to128(xxh_u64 lhs, xxh_u64 rhs)
+{
+    /*
+     * GCC/Clang __uint128_t method.
+     *
+     * On most 64-bit targets, GCC and Clang define a __uint128_t type.
+     * This is usually the best way as it usually uses a native long 64-bit
+     * multiply, such as MULQ on x86_64 or MUL + UMULH on aarch64.
+     *
+     * Usually.
+     *
+     * Despite being a 32-bit platform, Clang (and emscripten) define this type
+     * despite not having the arithmetic for it. This results in a laggy
+     * compiler builtin call which calculates a full 128-bit multiply.
+     * In that case it is best to use the portable one.
+     * https://github.com/Cyan4973/xxHash/issues/211#issuecomment-515575677
+     */
+#if defined(__GNUC__) && !defined(__wasm__) \
+    && defined(__SIZEOF_INT128__) \
+    || (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128)
+
+    __uint128_t const product = (__uint128_t)lhs * (__uint128_t)rhs;
+    XXH128_hash_t r128;
+    r128.low64  = (xxh_u64)(product);
+    r128.high64 = (xxh_u64)(product >> 64);
+    return r128;
+
+    /*
+     * MSVC for x64's _umul128 method.
+     *
+     * xxh_u64 _umul128(xxh_u64 Multiplier, xxh_u64 Multiplicand, xxh_u64 *HighProduct);
+     *
+     * This compiles to single operand MUL on x64.
+     */
+#elif defined(_M_X64) || defined(_M_IA64)
+
+#ifndef _MSC_VER
+#   pragma intrinsic(_umul128)
+#endif
+    xxh_u64 product_high;
+    xxh_u64 const product_low = _umul128(lhs, rhs, &product_high);
+    XXH128_hash_t r128;
+    r128.low64  = product_low;
+    r128.high64 = product_high;
+    return r128;
+
+    /*
+     * MSVC for ARM64's __umulh method.
+     *
+     * This compiles to the same MUL + UMULH as GCC/Clang's __uint128_t method.
+     */
+#elif defined(_M_ARM64)
+
+#ifndef _MSC_VER
+#   pragma intrinsic(__umulh)
+#endif
+    XXH128_hash_t r128;
+    r128.low64  = lhs * rhs;
+    r128.high64 = __umulh(lhs, rhs);
+    return r128;
+
+#else
+    /*
+     * Portable scalar method. Optimized for 32-bit and 64-bit ALUs.
+     *
+     * This is a fast and simple grade school multiply, which is shown below
+     * with base 10 arithmetic instead of base 0x100000000.
+     *
+     *           9 3 // D2 lhs = 93
+     *         x 7 5 // D2 rhs = 75
+     *     ----------
+     *           1 5 // D2 lo_lo = (93 % 10) * (75 % 10) = 15
+     *         4 5 | // D2 hi_lo = (93 / 10) * (75 % 10) = 45
+     *         2 1 | // D2 lo_hi = (93 % 10) * (75 / 10) = 21
+     *     + 6 3 | | // D2 hi_hi = (93 / 10) * (75 / 10) = 63
+     *     ---------
+     *         2 7 | // D2 cross = (15 / 10) + (45 % 10) + 21 = 27
+     *     + 6 7 | | // D2 upper = (27 / 10) + (45 / 10) + 63 = 67
+     *     ---------
+     *       6 9 7 5 // D4 res = (27 * 10) + (15 % 10) + (67 * 100) = 6975
+     *
+     * The reasons for adding the products like this are:
+     *  1. It avoids manual carry tracking. Just like how
+     *     (9 * 9) + 9 + 9 = 99, the same applies with this for UINT64_MAX.
+     *     This avoids a lot of complexity.
+     *
+     *  2. It hints for, and on Clang, compiles to, the powerful UMAAL
+     *     instruction available in ARM's Digital Signal Processing extension
+     *     in 32-bit ARMv6 and later, which is shown below:
+     *
+     *         void UMAAL(xxh_u32 *RdLo, xxh_u32 *RdHi, xxh_u32 Rn, xxh_u32 Rm)
+     *         {
+     *             xxh_u64 product = (xxh_u64)*RdLo * (xxh_u64)*RdHi + Rn + Rm;
+     *             *RdLo = (xxh_u32)(product & 0xFFFFFFFF);
+     *             *RdHi = (xxh_u32)(product >> 32);
+     *         }
+     *
+     *     This instruction was designed for efficient long multiplication, and
+     *     allows this to be calculated in only 4 instructions at speeds
+     *     comparable to some 64-bit ALUs.
+     *
+     *  3. It isn't terrible on other platforms. Usually this will be a couple
+     *     of 32-bit ADD/ADCs.
+     */
+
+    /* First calculate all of the cross products. */
+    xxh_u64 const lo_lo = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs & 0xFFFFFFFF);
+    xxh_u64 const hi_lo = XXH_mult32to64(lhs >> 32,        rhs & 0xFFFFFFFF);
+    xxh_u64 const lo_hi = XXH_mult32to64(lhs & 0xFFFFFFFF, rhs >> 32);
+    xxh_u64 const hi_hi = XXH_mult32to64(lhs >> 32,        rhs >> 32);
+
+    /* Now add the products together. These will never overflow. */
+    xxh_u64 const cross = (lo_lo >> 32) + (hi_lo & 0xFFFFFFFF) + lo_hi;
+    xxh_u64 const upper = (hi_lo >> 32) + (cross >> 32)        + hi_hi;
+    xxh_u64 const lower = (cross << 32) | (lo_lo & 0xFFFFFFFF);
+
+    XXH128_hash_t r128;
+    r128.low64  = lower;
+    r128.high64 = upper;
+    return r128;
+#endif
+}
+
+/*!
+ * @brief Calculates a 64-bit to 128-bit multiply, then XOR folds it.
+ *
+ * The reason for the separate function is to prevent passing too many structs
+ * around by value. This will hopefully inline the multiply, but we don't force it.
+ *
+ * @param lhs , rhs The 64-bit integers to multiply
+ * @return The low 64 bits of the product XOR'd by the high 64 bits.
+ * @see XXH_mult64to128()
+ */
+static xxh_u64
+XXH3_mul128_fold64(xxh_u64 lhs, xxh_u64 rhs)
+{
+    XXH128_hash_t product = XXH_mult64to128(lhs, rhs);
+    return product.low64 ^ product.high64;
+}
+
+/*! Seems to produce slightly better code on GCC for some reason. */
+XXH_FORCE_INLINE xxh_u64 XXH_xorshift64(xxh_u64 v64, int shift)
+{
+    XXH_ASSERT(0 <= shift && shift < 64);
+    return v64 ^ (v64 >> shift);
+}
+
+/*
+ * This is a fast avalanche stage,
+ * suitable when input bits are already partially mixed
+ */
+static XXH64_hash_t XXH3_avalanche(xxh_u64 h64)
+{
+    h64 = XXH_xorshift64(h64, 37);
+    h64 *= 0x165667919E3779F9ULL;
+    h64 = XXH_xorshift64(h64, 32);
+    return h64;
+}
+
+/*
+ * This is a stronger avalanche,
+ * inspired by Pelle Evensen's rrmxmx
+ * preferable when input has not been previously mixed
+ */
+static XXH64_hash_t XXH3_rrmxmx(xxh_u64 h64, xxh_u64 len)
+{
+    /* this mix is inspired by Pelle Evensen's rrmxmx */
+    h64 ^= XXH_rotl64(h64, 49) ^ XXH_rotl64(h64, 24);
+    h64 *= 0x9FB21C651E98DF25ULL;
+    h64 ^= (h64 >> 35) + len ;
+    h64 *= 0x9FB21C651E98DF25ULL;
+    return XXH_xorshift64(h64, 28);
+}
+
+
+/* ==========================================
+ * Short keys
+ * ==========================================
+ * One of the shortcomings of XXH32 and XXH64 was that their performance was
+ * sub-optimal on short lengths. It used an iterative algorithm which strongly
+ * favored lengths that were a multiple of 4 or 8.
+ *
+ * Instead of iterating over individual inputs, we use a set of single shot
+ * functions which piece together a range of lengths and operate in constant time.
+ *
+ * Additionally, the number of multiplies has been significantly reduced. This
+ * reduces latency, especially when emulating 64-bit multiplies on 32-bit.
+ *
+ * Depending on the platform, this may or may not be faster than XXH32, but it
+ * is almost guaranteed to be faster than XXH64.
+ */
+
+/*
+ * At very short lengths, there isn't enough input to fully hide secrets, or use
+ * the entire secret.
+ *
+ * There is also only a limited amount of mixing we can do before significantly
+ * impacting performance.
+ *
+ * Therefore, we use different sections of the secret and always mix two secret
+ * samples with an XOR. This should have no effect on performance on the
+ * seedless or withSeed variants because everything _should_ be constant folded
+ * by modern compilers.
+ *
+ * The XOR mixing hides individual parts of the secret and increases entropy.
+ *
+ * This adds an extra layer of strength for custom secrets.
+ */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_1to3_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(1 <= len && len <= 3);
+    XXH_ASSERT(secret != NULL);
+    /*
+     * len = 1: combined = { input[0], 0x01, input[0], input[0] }
+     * len = 2: combined = { input[1], 0x02, input[0], input[1] }
+     * len = 3: combined = { input[2], 0x03, input[0], input[1] }
+     */
+    {   xxh_u8  const c1 = input[0];
+        xxh_u8  const c2 = input[len >> 1];
+        xxh_u8  const c3 = input[len - 1];
+        xxh_u32 const combined = ((xxh_u32)c1 << 16) | ((xxh_u32)c2  << 24)
+                               | ((xxh_u32)c3 <<  0) | ((xxh_u32)len << 8);
+        xxh_u64 const bitflip = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
+        xxh_u64 const keyed = (xxh_u64)combined ^ bitflip;
+        return XXH64_avalanche(keyed);
+    }
+}
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_4to8_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(4 <= len && len <= 8);
+    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
+    {   xxh_u32 const input1 = XXH_readLE32(input);
+        xxh_u32 const input2 = XXH_readLE32(input + len - 4);
+        xxh_u64 const bitflip = (XXH_readLE64(secret+8) ^ XXH_readLE64(secret+16)) - seed;
+        xxh_u64 const input64 = input2 + (((xxh_u64)input1) << 32);
+        xxh_u64 const keyed = input64 ^ bitflip;
+        return XXH3_rrmxmx(keyed, len);
+    }
+}
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_9to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(9 <= len && len <= 16);
+    {   xxh_u64 const bitflip1 = (XXH_readLE64(secret+24) ^ XXH_readLE64(secret+32)) + seed;
+        xxh_u64 const bitflip2 = (XXH_readLE64(secret+40) ^ XXH_readLE64(secret+48)) - seed;
+        xxh_u64 const input_lo = XXH_readLE64(input)           ^ bitflip1;
+        xxh_u64 const input_hi = XXH_readLE64(input + len - 8) ^ bitflip2;
+        xxh_u64 const acc = len
+                          + XXH_swap64(input_lo) + input_hi
+                          + XXH3_mul128_fold64(input_lo, input_hi);
+        return XXH3_avalanche(acc);
+    }
+}
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_0to16_64b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(len <= 16);
+    {   if (XXH_likely(len >  8)) return XXH3_len_9to16_64b(input, len, secret, seed);
+        if (XXH_likely(len >= 4)) return XXH3_len_4to8_64b(input, len, secret, seed);
+        if (len) return XXH3_len_1to3_64b(input, len, secret, seed);
+        return XXH64_avalanche(seed ^ (XXH_readLE64(secret+56) ^ XXH_readLE64(secret+64)));
+    }
+}
+
+/*
+ * DISCLAIMER: There are known *seed-dependent* multicollisions here due to
+ * multiplication by zero, affecting hashes of lengths 17 to 240.
+ *
+ * However, they are very unlikely.
+ *
+ * Keep this in mind when using the unseeded XXH3_64bits() variant: As with all
+ * unseeded non-cryptographic hashes, it does not attempt to defend itself
+ * against specially crafted inputs, only random inputs.
+ *
+ * Compared to classic UMAC where a 1 in 2^31 chance of 4 consecutive bytes
+ * cancelling out the secret is taken an arbitrary number of times (addressed
+ * in XXH3_accumulate_512), this collision is very unlikely with random inputs
+ * and/or proper seeding:
+ *
+ * This only has a 1 in 2^63 chance of 8 consecutive bytes cancelling out, in a
+ * function that is only called up to 16 times per hash with up to 240 bytes of
+ * input.
+ *
+ * This is not too bad for a non-cryptographic hash function, especially with
+ * only 64 bit outputs.
+ *
+ * The 128-bit variant (which trades some speed for strength) is NOT affected
+ * by this, although it is always a good idea to use a proper seed if you care
+ * about strength.
+ */
+XXH_FORCE_INLINE xxh_u64 XXH3_mix16B(const xxh_u8* XXH_RESTRICT input,
+                                     const xxh_u8* XXH_RESTRICT secret, xxh_u64 seed64)
+{
+#if defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__i386__) && defined(__SSE2__)  /* x86 + SSE2 */ \
+  && !defined(XXH_ENABLE_AUTOVECTORIZE)      /* Define to disable like XXH32 hack */
+    /*
+     * UGLY HACK:
+     * GCC for x86 tends to autovectorize the 128-bit multiply, resulting in
+     * slower code.
+     *
+     * By forcing seed64 into a register, we disrupt the cost model and
+     * cause it to scalarize. See `XXH32_round()`
+     *
+     * FIXME: Clang's output is still _much_ faster -- On an AMD Ryzen 3600,
+     * XXH3_64bits @ len=240 runs at 4.6 GB/s with Clang 9, but 3.3 GB/s on
+     * GCC 9.2, despite both emitting scalar code.
+     *
+     * GCC generates much better scalar code than Clang for the rest of XXH3,
+     * which is why finding a more optimal codepath is an interest.
+     */
+    XXH_COMPILER_GUARD(seed64);
+#endif
+    {   xxh_u64 const input_lo = XXH_readLE64(input);
+        xxh_u64 const input_hi = XXH_readLE64(input+8);
+        return XXH3_mul128_fold64(
+            input_lo ^ (XXH_readLE64(secret)   + seed64),
+            input_hi ^ (XXH_readLE64(secret+8) - seed64)
+        );
+    }
+}
+
+/* For mid range keys, XXH3 uses a Mum-hash variant. */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_len_17to128_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                     const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                     XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(16 < len && len <= 128);
+
+    {   xxh_u64 acc = len * XXH_PRIME64_1;
+        if (len > 32) {
+            if (len > 64) {
+                if (len > 96) {
+                    acc += XXH3_mix16B(input+48, secret+96, seed);
+                    acc += XXH3_mix16B(input+len-64, secret+112, seed);
+                }
+                acc += XXH3_mix16B(input+32, secret+64, seed);
+                acc += XXH3_mix16B(input+len-48, secret+80, seed);
+            }
+            acc += XXH3_mix16B(input+16, secret+32, seed);
+            acc += XXH3_mix16B(input+len-32, secret+48, seed);
+        }
+        acc += XXH3_mix16B(input+0, secret+0, seed);
+        acc += XXH3_mix16B(input+len-16, secret+16, seed);
+
+        return XXH3_avalanche(acc);
+    }
+}
+
+#define XXH3_MIDSIZE_MAX 240
+
+XXH_NO_INLINE XXH64_hash_t
+XXH3_len_129to240_64b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                      XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+
+    #define XXH3_MIDSIZE_STARTOFFSET 3
+    #define XXH3_MIDSIZE_LASTOFFSET  17
+
+    {   xxh_u64 acc = len * XXH_PRIME64_1;
+        int const nbRounds = (int)len / 16;
+        int i;
+        for (i=0; i<8; i++) {
+            acc += XXH3_mix16B(input+(16*i), secret+(16*i), seed);
+        }
+        acc = XXH3_avalanche(acc);
+        XXH_ASSERT(nbRounds >= 8);
+#if defined(__clang__)                                /* Clang */ \
+    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */ \
+    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
+        /*
+         * UGLY HACK:
+         * Clang for ARMv7-A tries to vectorize this loop, similar to GCC x86.
+         * In everywhere else, it uses scalar code.
+         *
+         * For 64->128-bit multiplies, even if the NEON was 100% optimal, it
+         * would still be slower than UMAAL (see XXH_mult64to128).
+         *
+         * Unfortunately, Clang doesn't handle the long multiplies properly and
+         * converts them to the nonexistent "vmulq_u64" intrinsic, which is then
+         * scalarized into an ugly mess of VMOV.32 instructions.
+         *
+         * This mess is difficult to avoid without turning autovectorization
+         * off completely, but they are usually relatively minor and/or not
+         * worth it to fix.
+         *
+         * This loop is the easiest to fix, as unlike XXH32, this pragma
+         * _actually works_ because it is a loop vectorization instead of an
+         * SLP vectorization.
+         */
+        #pragma clang loop vectorize(disable)
+#endif
+        for (i=8 ; i < nbRounds; i++) {
+            acc += XXH3_mix16B(input+(16*i), secret+(16*(i-8)) + XXH3_MIDSIZE_STARTOFFSET, seed);
+        }
+        /* last bytes */
+        acc += XXH3_mix16B(input + len - 16, secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET, seed);
+        return XXH3_avalanche(acc);
+    }
+}
+
+
+/* =======     Long Keys     ======= */
+
+#define XXH_STRIPE_LEN 64
+#define XXH_SECRET_CONSUME_RATE 8   /* nb of secret bytes consumed at each accumulation */
+#define XXH_ACC_NB (XXH_STRIPE_LEN / sizeof(xxh_u64))
+
+#ifdef XXH_OLD_NAMES
+#  define STRIPE_LEN XXH_STRIPE_LEN
+#  define ACC_NB XXH_ACC_NB
+#endif
+
+XXH_FORCE_INLINE void XXH_writeLE64(void* dst, xxh_u64 v64)
+{
+    if (!XXH_CPU_LITTLE_ENDIAN) v64 = XXH_swap64(v64);
+    XXH_memcpy(dst, &v64, sizeof(v64));
+}
+
+/* Several intrinsic functions below are supposed to accept __int64 as argument,
+ * as documented in https://software.intel.com/sites/landingpage/IntrinsicsGuide/ .
+ * However, several environments do not define __int64 type,
+ * requiring a workaround.
+ */
+#if !defined (__VMS) \
+  && (defined (__cplusplus) \
+  || (defined (__STDC_VERSION__) && (__STDC_VERSION__ >= 199901L) /* C99 */) )
+    typedef int64_t xxh_i64;
+#else
+    /* the following type must have a width of 64-bit */
+    typedef long long xxh_i64;
+#endif
+
+/*
+ * XXH3_accumulate_512 is the tightest loop for long inputs, and it is the most optimized.
+ *
+ * It is a hardened version of UMAC, based off of FARSH's implementation.
+ *
+ * This was chosen because it adapts quite well to 32-bit, 64-bit, and SIMD
+ * implementations, and it is ridiculously fast.
+ *
+ * We harden it by mixing the original input to the accumulators as well as the product.
+ *
+ * This means that in the (relatively likely) case of a multiply by zero, the
+ * original input is preserved.
+ *
+ * On 128-bit inputs, we swap 64-bit pairs when we add the input to improve
+ * cross-pollination, as otherwise the upper and lower halves would be
+ * essentially independent.
+ *
+ * This doesn't matter on 64-bit hashes since they all get merged together in
+ * the end, so we skip the extra step.
+ *
+ * Both XXH3_64bits and XXH3_128bits use this subroutine.
+ */
+
+#if (XXH_VECTOR == XXH_AVX512) \
+     || (defined(XXH_DISPATCH_AVX512) && XXH_DISPATCH_AVX512 != 0)
+
+#ifndef XXH_TARGET_AVX512
+# define XXH_TARGET_AVX512  /* disable attribute target */
+#endif
+
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_accumulate_512_avx512(void* XXH_RESTRICT acc,
+                     const void* XXH_RESTRICT input,
+                     const void* XXH_RESTRICT secret)
+{
+    __m512i* const xacc = (__m512i *) acc;
+    XXH_ASSERT((((size_t)acc) & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
+
+    {
+        /* data_vec    = input[0]; */
+        __m512i const data_vec    = _mm512_loadu_si512   (input);
+        /* key_vec     = secret[0]; */
+        __m512i const key_vec     = _mm512_loadu_si512   (secret);
+        /* data_key    = data_vec ^ key_vec; */
+        __m512i const data_key    = _mm512_xor_si512     (data_vec, key_vec);
+        /* data_key_lo = data_key >> 32; */
+        __m512i const data_key_lo = _mm512_shuffle_epi32 (data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
+        /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+        __m512i const product     = _mm512_mul_epu32     (data_key, data_key_lo);
+        /* xacc[0] += swap(data_vec); */
+        __m512i const data_swap = _mm512_shuffle_epi32(data_vec, (_MM_PERM_ENUM)_MM_SHUFFLE(1, 0, 3, 2));
+        __m512i const sum       = _mm512_add_epi64(*xacc, data_swap);
+        /* xacc[0] += product; */
+        *xacc = _mm512_add_epi64(product, sum);
+    }
+}
+
+/*
+ * XXH3_scrambleAcc: Scrambles the accumulators to improve mixing.
+ *
+ * Multiplication isn't perfect, as explained by Google in HighwayHash:
+ *
+ *  // Multiplication mixes/scrambles bytes 0-7 of the 64-bit result to
+ *  // varying degrees. In descending order of goodness, bytes
+ *  // 3 4 2 5 1 6 0 7 have quality 228 224 164 160 100 96 36 32.
+ *  // As expected, the upper and lower bytes are much worse.
+ *
+ * Source: https://github.com/google/highwayhash/blob/0aaf66b/highwayhash/hh_avx2.h#L291
+ *
+ * Since our algorithm uses a pseudorandom secret to add some variance into the
+ * mix, we don't need to (or want to) mix as often or as much as HighwayHash does.
+ *
+ * This isn't as tight as XXH3_accumulate, but still written in SIMD to avoid
+ * extraction.
+ *
+ * Both XXH3_64bits and XXH3_128bits use this subroutine.
+ */
+
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_scrambleAcc_avx512(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_STRIPE_LEN == sizeof(__m512i));
+    {   __m512i* const xacc = (__m512i*) acc;
+        const __m512i prime32 = _mm512_set1_epi32((int)XXH_PRIME32_1);
+
+        /* xacc[0] ^= (xacc[0] >> 47) */
+        __m512i const acc_vec     = *xacc;
+        __m512i const shifted     = _mm512_srli_epi64    (acc_vec, 47);
+        __m512i const data_vec    = _mm512_xor_si512     (acc_vec, shifted);
+        /* xacc[0] ^= secret; */
+        __m512i const key_vec     = _mm512_loadu_si512   (secret);
+        __m512i const data_key    = _mm512_xor_si512     (data_vec, key_vec);
+
+        /* xacc[0] *= XXH_PRIME32_1; */
+        __m512i const data_key_hi = _mm512_shuffle_epi32 (data_key, (_MM_PERM_ENUM)_MM_SHUFFLE(0, 3, 0, 1));
+        __m512i const prod_lo     = _mm512_mul_epu32     (data_key, prime32);
+        __m512i const prod_hi     = _mm512_mul_epu32     (data_key_hi, prime32);
+        *xacc = _mm512_add_epi64(prod_lo, _mm512_slli_epi64(prod_hi, 32));
+    }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_AVX512 void
+XXH3_initCustomSecret_avx512(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 63) == 0);
+    XXH_STATIC_ASSERT(XXH_SEC_ALIGN == 64);
+    XXH_ASSERT(((size_t)customSecret & 63) == 0);
+    (void)(&XXH_writeLE64);
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m512i);
+        __m512i const seed = _mm512_mask_set1_epi64(_mm512_set1_epi64((xxh_i64)seed64), 0xAA, (xxh_i64)(0U - seed64));
+
+        const __m512i* const src  = (const __m512i*) ((const void*) XXH3_kSecret);
+              __m512i* const dest = (      __m512i*) customSecret;
+        int i;
+        XXH_ASSERT(((size_t)src & 63) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dest & 63) == 0);
+        for (i=0; i < nbRounds; ++i) {
+            /* GCC has a bug, _mm512_stream_load_si512 accepts 'void*', not 'void const*',
+             * this will warn "discards 'const' qualifier". */
+            union {
+                const __m512i* cp;
+                void* p;
+            } remote_const_void;
+            remote_const_void.cp = src + i;
+            dest[i] = _mm512_add_epi64(_mm512_stream_load_si512(remote_const_void.p), seed);
+    }   }
+}
+
+#endif
+
+#if (XXH_VECTOR == XXH_AVX2) \
+    || (defined(XXH_DISPATCH_AVX2) && XXH_DISPATCH_AVX2 != 0)
+
+#ifndef XXH_TARGET_AVX2
+# define XXH_TARGET_AVX2  /* disable attribute target */
+#endif
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void
+XXH3_accumulate_512_avx2( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 31) == 0);
+    {   __m256i* const xacc    =       (__m256i *) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires  a const __m256i * pointer for some reason. */
+        const         __m256i* const xinput  = (const __m256i *) input;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
+        const         __m256i* const xsecret = (const __m256i *) secret;
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
+            /* data_vec    = xinput[i]; */
+            __m256i const data_vec    = _mm256_loadu_si256    (xinput+i);
+            /* key_vec     = xsecret[i]; */
+            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
+            /* data_key    = data_vec ^ key_vec; */
+            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
+            /* data_key_lo = data_key >> 32; */
+            __m256i const data_key_lo = _mm256_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+            __m256i const product     = _mm256_mul_epu32     (data_key, data_key_lo);
+            /* xacc[i] += swap(data_vec); */
+            __m256i const data_swap = _mm256_shuffle_epi32(data_vec, _MM_SHUFFLE(1, 0, 3, 2));
+            __m256i const sum       = _mm256_add_epi64(xacc[i], data_swap);
+            /* xacc[i] += product; */
+            xacc[i] = _mm256_add_epi64(product, sum);
+    }   }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void
+XXH3_scrambleAcc_avx2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 31) == 0);
+    {   __m256i* const xacc = (__m256i*) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm256_loadu_si256 requires a const __m256i * pointer for some reason. */
+        const         __m256i* const xsecret = (const __m256i *) secret;
+        const __m256i prime32 = _mm256_set1_epi32((int)XXH_PRIME32_1);
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m256i); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47) */
+            __m256i const acc_vec     = xacc[i];
+            __m256i const shifted     = _mm256_srli_epi64    (acc_vec, 47);
+            __m256i const data_vec    = _mm256_xor_si256     (acc_vec, shifted);
+            /* xacc[i] ^= xsecret; */
+            __m256i const key_vec     = _mm256_loadu_si256   (xsecret+i);
+            __m256i const data_key    = _mm256_xor_si256     (data_vec, key_vec);
+
+            /* xacc[i] *= XXH_PRIME32_1; */
+            __m256i const data_key_hi = _mm256_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            __m256i const prod_lo     = _mm256_mul_epu32     (data_key, prime32);
+            __m256i const prod_hi     = _mm256_mul_epu32     (data_key_hi, prime32);
+            xacc[i] = _mm256_add_epi64(prod_lo, _mm256_slli_epi64(prod_hi, 32));
+        }
+    }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_AVX2 void XXH3_initCustomSecret_avx2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 31) == 0);
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE / sizeof(__m256i)) == 6);
+    XXH_STATIC_ASSERT(XXH_SEC_ALIGN <= 64);
+    (void)(&XXH_writeLE64);
+    XXH_PREFETCH(customSecret);
+    {   __m256i const seed = _mm256_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64, (xxh_i64)(0U - seed64), (xxh_i64)seed64);
+
+        const __m256i* const src  = (const __m256i*) ((const void*) XXH3_kSecret);
+              __m256i*       dest = (      __m256i*) customSecret;
+
+#       if defined(__GNUC__) || defined(__clang__)
+        /*
+         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
+         *   - do not extract the secret from sse registers in the internal loop
+         *   - use less common registers, and avoid pushing these reg into stack
+         */
+        XXH_COMPILER_GUARD(dest);
+#       endif
+        XXH_ASSERT(((size_t)src & 31) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dest & 31) == 0);
+
+        /* GCC -O2 need unroll loop manually */
+        dest[0] = _mm256_add_epi64(_mm256_stream_load_si256(src+0), seed);
+        dest[1] = _mm256_add_epi64(_mm256_stream_load_si256(src+1), seed);
+        dest[2] = _mm256_add_epi64(_mm256_stream_load_si256(src+2), seed);
+        dest[3] = _mm256_add_epi64(_mm256_stream_load_si256(src+3), seed);
+        dest[4] = _mm256_add_epi64(_mm256_stream_load_si256(src+4), seed);
+        dest[5] = _mm256_add_epi64(_mm256_stream_load_si256(src+5), seed);
+    }
+}
+
+#endif
+
+/* x86dispatch always generates SSE2 */
+#if (XXH_VECTOR == XXH_SSE2) || defined(XXH_X86DISPATCH)
+
+#ifndef XXH_TARGET_SSE2
+# define XXH_TARGET_SSE2  /* disable attribute target */
+#endif
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void
+XXH3_accumulate_512_sse2( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    /* SSE2 is just a half-scale version of the AVX2 version. */
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    {   __m128i* const xacc    =       (__m128i *) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xinput  = (const __m128i *) input;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xsecret = (const __m128i *) secret;
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
+            /* data_vec    = xinput[i]; */
+            __m128i const data_vec    = _mm_loadu_si128   (xinput+i);
+            /* key_vec     = xsecret[i]; */
+            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
+            /* data_key    = data_vec ^ key_vec; */
+            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
+            /* data_key_lo = data_key >> 32; */
+            __m128i const data_key_lo = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            /* product     = (data_key & 0xffffffff) * (data_key_lo & 0xffffffff); */
+            __m128i const product     = _mm_mul_epu32     (data_key, data_key_lo);
+            /* xacc[i] += swap(data_vec); */
+            __m128i const data_swap = _mm_shuffle_epi32(data_vec, _MM_SHUFFLE(1,0,3,2));
+            __m128i const sum       = _mm_add_epi64(xacc[i], data_swap);
+            /* xacc[i] += product; */
+            xacc[i] = _mm_add_epi64(product, sum);
+    }   }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void
+XXH3_scrambleAcc_sse2(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    {   __m128i* const xacc = (__m128i*) acc;
+        /* Unaligned. This is mainly for pointer arithmetic, and because
+         * _mm_loadu_si128 requires a const __m128i * pointer for some reason. */
+        const         __m128i* const xsecret = (const __m128i *) secret;
+        const __m128i prime32 = _mm_set1_epi32((int)XXH_PRIME32_1);
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(__m128i); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47) */
+            __m128i const acc_vec     = xacc[i];
+            __m128i const shifted     = _mm_srli_epi64    (acc_vec, 47);
+            __m128i const data_vec    = _mm_xor_si128     (acc_vec, shifted);
+            /* xacc[i] ^= xsecret[i]; */
+            __m128i const key_vec     = _mm_loadu_si128   (xsecret+i);
+            __m128i const data_key    = _mm_xor_si128     (data_vec, key_vec);
+
+            /* xacc[i] *= XXH_PRIME32_1; */
+            __m128i const data_key_hi = _mm_shuffle_epi32 (data_key, _MM_SHUFFLE(0, 3, 0, 1));
+            __m128i const prod_lo     = _mm_mul_epu32     (data_key, prime32);
+            __m128i const prod_hi     = _mm_mul_epu32     (data_key_hi, prime32);
+            xacc[i] = _mm_add_epi64(prod_lo, _mm_slli_epi64(prod_hi, 32));
+        }
+    }
+}
+
+XXH_FORCE_INLINE XXH_TARGET_SSE2 void XXH3_initCustomSecret_sse2(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+    (void)(&XXH_writeLE64);
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / sizeof(__m128i);
+
+#       if defined(_MSC_VER) && defined(_M_IX86) && _MSC_VER < 1900
+        /* MSVC 32bit mode does not support _mm_set_epi64x before 2015 */
+        XXH_ALIGN(16) const xxh_i64 seed64x2[2] = { (xxh_i64)seed64, (xxh_i64)(0U - seed64) };
+        __m128i const seed = _mm_load_si128((__m128i const*)seed64x2);
+#       else
+        __m128i const seed = _mm_set_epi64x((xxh_i64)(0U - seed64), (xxh_i64)seed64);
+#       endif
+        int i;
+
+        const void* const src16 = XXH3_kSecret;
+        __m128i* dst16 = (__m128i*) customSecret;
+#       if defined(__GNUC__) || defined(__clang__)
+        /*
+         * On GCC & Clang, marking 'dest' as modified will cause the compiler:
+         *   - do not extract the secret from sse registers in the internal loop
+         *   - use less common registers, and avoid pushing these reg into stack
+         */
+        XXH_COMPILER_GUARD(dst16);
+#       endif
+        XXH_ASSERT(((size_t)src16 & 15) == 0); /* control alignment */
+        XXH_ASSERT(((size_t)dst16 & 15) == 0);
+
+        for (i=0; i < nbRounds; ++i) {
+            dst16[i] = _mm_add_epi64(_mm_load_si128((const __m128i *)src16+i), seed);
+    }   }
+}
+
+#endif
+
+#if (XXH_VECTOR == XXH_NEON)
+
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_neon( void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+    {
+        uint64x2_t* const xacc = (uint64x2_t *) acc;
+        /* We don't use a uint32x4_t pointer because it causes bus errors on ARMv7. */
+        uint8_t const* const xinput = (const uint8_t *) input;
+        uint8_t const* const xsecret  = (const uint8_t *) secret;
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN / sizeof(uint64x2_t); i++) {
+            /* data_vec = xinput[i]; */
+            uint8x16_t data_vec    = vld1q_u8(xinput  + (i * 16));
+            /* key_vec  = xsecret[i];  */
+            uint8x16_t key_vec     = vld1q_u8(xsecret + (i * 16));
+            uint64x2_t data_key;
+            uint32x2_t data_key_lo, data_key_hi;
+            /* xacc[i] += swap(data_vec); */
+            uint64x2_t const data64  = vreinterpretq_u64_u8(data_vec);
+            uint64x2_t const swapped = vextq_u64(data64, data64, 1);
+            xacc[i] = vaddq_u64 (xacc[i], swapped);
+            /* data_key = data_vec ^ key_vec; */
+            data_key = vreinterpretq_u64_u8(veorq_u8(data_vec, key_vec));
+            /* data_key_lo = (uint32x2_t) (data_key & 0xFFFFFFFF);
+             * data_key_hi = (uint32x2_t) (data_key >> 32);
+             * data_key = UNDEFINED; */
+            XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
+            /* xacc[i] += (uint64x2_t) data_key_lo * (uint64x2_t) data_key_hi; */
+            xacc[i] = vmlal_u32 (xacc[i], data_key_lo, data_key_hi);
+
+        }
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_neon(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+
+    {   uint64x2_t* xacc       = (uint64x2_t*) acc;
+        uint8_t const* xsecret = (uint8_t const*) secret;
+        uint32x2_t prime       = vdup_n_u32 (XXH_PRIME32_1);
+
+        size_t i;
+        for (i=0; i < XXH_STRIPE_LEN/sizeof(uint64x2_t); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47); */
+            uint64x2_t acc_vec  = xacc[i];
+            uint64x2_t shifted  = vshrq_n_u64 (acc_vec, 47);
+            uint64x2_t data_vec = veorq_u64   (acc_vec, shifted);
+
+            /* xacc[i] ^= xsecret[i]; */
+            uint8x16_t key_vec  = vld1q_u8    (xsecret + (i * 16));
+            uint64x2_t data_key = veorq_u64   (data_vec, vreinterpretq_u64_u8(key_vec));
+
+            /* xacc[i] *= XXH_PRIME32_1 */
+            uint32x2_t data_key_lo, data_key_hi;
+            /* data_key_lo = (uint32x2_t) (xacc[i] & 0xFFFFFFFF);
+             * data_key_hi = (uint32x2_t) (xacc[i] >> 32);
+             * xacc[i] = UNDEFINED; */
+            XXH_SPLIT_IN_PLACE(data_key, data_key_lo, data_key_hi);
+            {   /*
+                 * prod_hi = (data_key >> 32) * XXH_PRIME32_1;
+                 *
+                 * Avoid vmul_u32 + vshll_n_u32 since Clang 6 and 7 will
+                 * incorrectly "optimize" this:
+                 *   tmp     = vmul_u32(vmovn_u64(a), vmovn_u64(b));
+                 *   shifted = vshll_n_u32(tmp, 32);
+                 * to this:
+                 *   tmp     = "vmulq_u64"(a, b); // no such thing!
+                 *   shifted = vshlq_n_u64(tmp, 32);
+                 *
+                 * However, unlike SSE, Clang lacks a 64-bit multiply routine
+                 * for NEON, and it scalarizes two 64-bit multiplies instead.
+                 *
+                 * vmull_u32 has the same timing as vmul_u32, and it avoids
+                 * this bug completely.
+                 * See https://bugs.llvm.org/show_bug.cgi?id=39967
+                 */
+                uint64x2_t prod_hi = vmull_u32 (data_key_hi, prime);
+                /* xacc[i] = prod_hi << 32; */
+                xacc[i] = vshlq_n_u64(prod_hi, 32);
+                /* xacc[i] += (prod_hi & 0xFFFFFFFF) * XXH_PRIME32_1; */
+                xacc[i] = vmlal_u32(xacc[i], data_key_lo, prime);
+            }
+    }   }
+}
+
+#endif
+
+#if (XXH_VECTOR == XXH_VSX)
+
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_vsx(  void* XXH_RESTRICT acc,
+                    const void* XXH_RESTRICT input,
+                    const void* XXH_RESTRICT secret)
+{
+    /* presumed aligned */
+    unsigned long long* const xacc = (unsigned long long*) acc;
+    xxh_u64x2 const* const xinput   = (xxh_u64x2 const*) input;   /* no alignment restriction */
+    xxh_u64x2 const* const xsecret  = (xxh_u64x2 const*) secret;    /* no alignment restriction */
+    xxh_u64x2 const v32 = { 32, 32 };
+    size_t i;
+    for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
+        /* data_vec = xinput[i]; */
+        xxh_u64x2 const data_vec = XXH_vec_loadu(xinput + i);
+        /* key_vec = xsecret[i]; */
+        xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + i);
+        xxh_u64x2 const data_key = data_vec ^ key_vec;
+        /* shuffled = (data_key << 32) | (data_key >> 32); */
+        xxh_u32x4 const shuffled = (xxh_u32x4)vec_rl(data_key, v32);
+        /* product = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)shuffled & 0xFFFFFFFF); */
+        xxh_u64x2 const product  = XXH_vec_mulo((xxh_u32x4)data_key, shuffled);
+        /* acc_vec = xacc[i]; */
+        xxh_u64x2 acc_vec        = vec_xl(0, xacc + 2 * i);
+        acc_vec += product;
+
+        /* swap high and low halves */
+#ifdef __s390x__
+        acc_vec += vec_permi(data_vec, data_vec, 2);
+#else
+        acc_vec += vec_xxpermdi(data_vec, data_vec, 2);
+#endif
+        /* xacc[i] = acc_vec; */
+        vec_xst(acc_vec, 0, xacc + 2 * i);
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_vsx(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    XXH_ASSERT((((size_t)acc) & 15) == 0);
+
+    {         xxh_u64x2* const xacc    =       (xxh_u64x2*) acc;
+        const xxh_u64x2* const xsecret = (const xxh_u64x2*) secret;
+        /* constants */
+        xxh_u64x2 const v32  = { 32, 32 };
+        xxh_u64x2 const v47 = { 47, 47 };
+        xxh_u32x4 const prime = { XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1, XXH_PRIME32_1 };
+        size_t i;
+        for (i = 0; i < XXH_STRIPE_LEN / sizeof(xxh_u64x2); i++) {
+            /* xacc[i] ^= (xacc[i] >> 47); */
+            xxh_u64x2 const acc_vec  = xacc[i];
+            xxh_u64x2 const data_vec = acc_vec ^ (acc_vec >> v47);
+
+            /* xacc[i] ^= xsecret[i]; */
+            xxh_u64x2 const key_vec  = XXH_vec_loadu(xsecret + i);
+            xxh_u64x2 const data_key = data_vec ^ key_vec;
+
+            /* xacc[i] *= XXH_PRIME32_1 */
+            /* prod_lo = ((xxh_u64x2)data_key & 0xFFFFFFFF) * ((xxh_u64x2)prime & 0xFFFFFFFF);  */
+            xxh_u64x2 const prod_even  = XXH_vec_mule((xxh_u32x4)data_key, prime);
+            /* prod_hi = ((xxh_u64x2)data_key >> 32) * ((xxh_u64x2)prime >> 32);  */
+            xxh_u64x2 const prod_odd  = XXH_vec_mulo((xxh_u32x4)data_key, prime);
+            xacc[i] = prod_odd + (prod_even << v32);
+    }   }
+}
+
+#endif
+
+/* scalar variants - universal */
+
+XXH_FORCE_INLINE void
+XXH3_accumulate_512_scalar(void* XXH_RESTRICT acc,
+                     const void* XXH_RESTRICT input,
+                     const void* XXH_RESTRICT secret)
+{
+    xxh_u64* const xacc = (xxh_u64*) acc; /* presumed aligned */
+    const xxh_u8* const xinput  = (const xxh_u8*) input;  /* no alignment restriction */
+    const xxh_u8* const xsecret = (const xxh_u8*) secret;   /* no alignment restriction */
+    size_t i;
+    XXH_ASSERT(((size_t)acc & (XXH_ACC_ALIGN-1)) == 0);
+    for (i=0; i < XXH_ACC_NB; i++) {
+        xxh_u64 const data_val = XXH_readLE64(xinput + 8*i);
+        xxh_u64 const data_key = data_val ^ XXH_readLE64(xsecret + i*8);
+        xacc[i ^ 1] += data_val; /* swap adjacent lanes */
+        xacc[i] += XXH_mult32to64(data_key & 0xFFFFFFFF, data_key >> 32);
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_scrambleAcc_scalar(void* XXH_RESTRICT acc, const void* XXH_RESTRICT secret)
+{
+    xxh_u64* const xacc = (xxh_u64*) acc;   /* presumed aligned */
+    const xxh_u8* const xsecret = (const xxh_u8*) secret;   /* no alignment restriction */
+    size_t i;
+    XXH_ASSERT((((size_t)acc) & (XXH_ACC_ALIGN-1)) == 0);
+    for (i=0; i < XXH_ACC_NB; i++) {
+        xxh_u64 const key64 = XXH_readLE64(xsecret + 8*i);
+        xxh_u64 acc64 = xacc[i];
+        acc64 = XXH_xorshift64(acc64, 47);
+        acc64 ^= key64;
+        acc64 *= XXH_PRIME32_1;
+        xacc[i] = acc64;
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_initCustomSecret_scalar(void* XXH_RESTRICT customSecret, xxh_u64 seed64)
+{
+    /*
+     * We need a separate pointer for the hack below,
+     * which requires a non-const pointer.
+     * Any decent compiler will optimize this out otherwise.
+     */
+    const xxh_u8* kSecretPtr = XXH3_kSecret;
+    XXH_STATIC_ASSERT((XXH_SECRET_DEFAULT_SIZE & 15) == 0);
+
+#if defined(__clang__) && defined(__aarch64__)
+    /*
+     * UGLY HACK:
+     * Clang generates a bunch of MOV/MOVK pairs for aarch64, and they are
+     * placed sequentially, in order, at the top of the unrolled loop.
+     *
+     * While MOVK is great for generating constants (2 cycles for a 64-bit
+     * constant compared to 4 cycles for LDR), long MOVK chains stall the
+     * integer pipelines:
+     *   I   L   S
+     * MOVK
+     * MOVK
+     * MOVK
+     * MOVK
+     * ADD
+     * SUB      STR
+     *          STR
+     * By forcing loads from memory (as the asm line causes Clang to assume
+     * that XXH3_kSecretPtr has been changed), the pipelines are used more
+     * efficiently:
+     *   I   L   S
+     *      LDR
+     *  ADD LDR
+     *  SUB     STR
+     *          STR
+     * XXH3_64bits_withSeed, len == 256, Snapdragon 835
+     *   without hack: 2654.4 MB/s
+     *   with hack:    3202.9 MB/s
+     */
+    XXH_COMPILER_GUARD(kSecretPtr);
+#endif
+    /*
+     * Note: in debug mode, this overrides the asm optimization
+     * and Clang will emit MOVK chains again.
+     */
+    XXH_ASSERT(kSecretPtr == XXH3_kSecret);
+
+    {   int const nbRounds = XXH_SECRET_DEFAULT_SIZE / 16;
+        int i;
+        for (i=0; i < nbRounds; i++) {
+            /*
+             * The asm hack causes Clang to assume that kSecretPtr aliases with
+             * customSecret, and on aarch64, this prevented LDP from merging two
+             * loads together for free. Putting the loads together before the stores
+             * properly generates LDP.
+             */
+            xxh_u64 lo = XXH_readLE64(kSecretPtr + 16*i)     + seed64;
+            xxh_u64 hi = XXH_readLE64(kSecretPtr + 16*i + 8) - seed64;
+            XXH_writeLE64((xxh_u8*)customSecret + 16*i,     lo);
+            XXH_writeLE64((xxh_u8*)customSecret + 16*i + 8, hi);
+    }   }
+}
+
+
+typedef void (*XXH3_f_accumulate_512)(void* XXH_RESTRICT, const void*, const void*);
+typedef void (*XXH3_f_scrambleAcc)(void* XXH_RESTRICT, const void*);
+typedef void (*XXH3_f_initCustomSecret)(void* XXH_RESTRICT, xxh_u64);
+
+
+#if (XXH_VECTOR == XXH_AVX512)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_avx512
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx512
+#define XXH3_initCustomSecret XXH3_initCustomSecret_avx512
+
+#elif (XXH_VECTOR == XXH_AVX2)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_avx2
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_avx2
+#define XXH3_initCustomSecret XXH3_initCustomSecret_avx2
+
+#elif (XXH_VECTOR == XXH_SSE2)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_sse2
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_sse2
+#define XXH3_initCustomSecret XXH3_initCustomSecret_sse2
+
+#elif (XXH_VECTOR == XXH_NEON)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_neon
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_neon
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+
+#elif (XXH_VECTOR == XXH_VSX)
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_vsx
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_vsx
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+
+#else /* scalar */
+
+#define XXH3_accumulate_512 XXH3_accumulate_512_scalar
+#define XXH3_scrambleAcc    XXH3_scrambleAcc_scalar
+#define XXH3_initCustomSecret XXH3_initCustomSecret_scalar
+
+#endif
+
+
+
+#ifndef XXH_PREFETCH_DIST
+#  ifdef __clang__
+#    define XXH_PREFETCH_DIST 320
+#  else
+#    if (XXH_VECTOR == XXH_AVX512)
+#      define XXH_PREFETCH_DIST 512
+#    else
+#      define XXH_PREFETCH_DIST 384
+#    endif
+#  endif  /* __clang__ */
+#endif  /* XXH_PREFETCH_DIST */
+
+/*
+ * XXH3_accumulate()
+ * Loops over XXH3_accumulate_512().
+ * Assumption: nbStripes will not overflow the secret size
+ */
+XXH_FORCE_INLINE void
+XXH3_accumulate(     xxh_u64* XXH_RESTRICT acc,
+                const xxh_u8* XXH_RESTRICT input,
+                const xxh_u8* XXH_RESTRICT secret,
+                      size_t nbStripes,
+                      XXH3_f_accumulate_512 f_acc512)
+{
+    size_t n;
+    for (n = 0; n < nbStripes; n++ ) {
+        const xxh_u8* const in = input + n*XXH_STRIPE_LEN;
+        XXH_PREFETCH(in + XXH_PREFETCH_DIST);
+        f_acc512(acc,
+                 in,
+                 secret + n*XXH_SECRET_CONSUME_RATE);
+    }
+}
+
+XXH_FORCE_INLINE void
+XXH3_hashLong_internal_loop(xxh_u64* XXH_RESTRICT acc,
+                      const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                            XXH3_f_accumulate_512 f_acc512,
+                            XXH3_f_scrambleAcc f_scramble)
+{
+    size_t const nbStripesPerBlock = (secretSize - XXH_STRIPE_LEN) / XXH_SECRET_CONSUME_RATE;
+    size_t const block_len = XXH_STRIPE_LEN * nbStripesPerBlock;
+    size_t const nb_blocks = (len - 1) / block_len;
+
+    size_t n;
+
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+
+    for (n = 0; n < nb_blocks; n++) {
+        XXH3_accumulate(acc, input + n*block_len, secret, nbStripesPerBlock, f_acc512);
+        f_scramble(acc, secret + secretSize - XXH_STRIPE_LEN);
+    }
+
+    /* last partial block */
+    XXH_ASSERT(len > XXH_STRIPE_LEN);
+    {   size_t const nbStripes = ((len - 1) - (block_len * nb_blocks)) / XXH_STRIPE_LEN;
+        XXH_ASSERT(nbStripes <= (secretSize / XXH_SECRET_CONSUME_RATE));
+        XXH3_accumulate(acc, input + nb_blocks*block_len, secret, nbStripes, f_acc512);
+
+        /* last stripe */
+        {   const xxh_u8* const p = input + len - XXH_STRIPE_LEN;
+#define XXH_SECRET_LASTACC_START 7  /* not aligned on 8, last secret is different from acc & scrambler */
+            f_acc512(acc, p, secret + secretSize - XXH_STRIPE_LEN - XXH_SECRET_LASTACC_START);
+    }   }
+}
+
+XXH_FORCE_INLINE xxh_u64
+XXH3_mix2Accs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret)
+{
+    return XXH3_mul128_fold64(
+               acc[0] ^ XXH_readLE64(secret),
+               acc[1] ^ XXH_readLE64(secret+8) );
+}
+
+static XXH64_hash_t
+XXH3_mergeAccs(const xxh_u64* XXH_RESTRICT acc, const xxh_u8* XXH_RESTRICT secret, xxh_u64 start)
+{
+    xxh_u64 result64 = start;
+    size_t i = 0;
+
+    for (i = 0; i < 4; i++) {
+        result64 += XXH3_mix2Accs(acc+2*i, secret + 16*i);
+#if defined(__clang__)                                /* Clang */ \
+    && (defined(__arm__) || defined(__thumb__))       /* ARMv7 */ \
+    && (defined(__ARM_NEON) || defined(__ARM_NEON__)) /* NEON */  \
+    && !defined(XXH_ENABLE_AUTOVECTORIZE)             /* Define to disable */
+        /*
+         * UGLY HACK:
+         * Prevent autovectorization on Clang ARMv7-a. Exact same problem as
+         * the one in XXH3_len_129to240_64b. Speeds up shorter keys > 240b.
+         * XXH3_64bits, len == 256, Snapdragon 835:
+         *   without hack: 2063.7 MB/s
+         *   with hack:    2560.7 MB/s
+         */
+        XXH_COMPILER_GUARD(result64);
+#endif
+    }
+
+    return XXH3_avalanche(result64);
+}
+
+#define XXH3_INIT_ACC { XXH_PRIME32_3, XXH_PRIME64_1, XXH_PRIME64_2, XXH_PRIME64_3, \
+                        XXH_PRIME64_4, XXH_PRIME32_2, XXH_PRIME64_5, XXH_PRIME32_1 }
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_internal(const void* XXH_RESTRICT input, size_t len,
+                           const void* XXH_RESTRICT secret, size_t secretSize,
+                           XXH3_f_accumulate_512 f_acc512,
+                           XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
+
+    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, (const xxh_u8*)secret, secretSize, f_acc512, f_scramble);
+
+    /* converge into final hash */
+    XXH_STATIC_ASSERT(sizeof(acc) == 64);
+    /* do not align on 8, so that the secret is different from the accumulator */
+#define XXH_SECRET_MERGEACCS_START 11
+    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+    return XXH3_mergeAccs(acc, (const xxh_u8*)secret + XXH_SECRET_MERGEACCS_START, (xxh_u64)len * XXH_PRIME64_1);
+}
+
+/*
+ * It's important for performance to transmit secret's size (when it's static)
+ * so that the compiler can properly optimize the vectorized loop.
+ * This makes a big performance difference for "medium" keys (<1 KB) when using AVX instruction set.
+ */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSecret(const void* XXH_RESTRICT input, size_t len,
+                             XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64;
+    return XXH3_hashLong_64b_internal(input, len, secret, secretLen, XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*
+ * It's preferable for performance that XXH3_hashLong is not inlined,
+ * as it results in a smaller function for small data, easier to the instruction cache.
+ * Note that inside this no_inline function, we do inline the internal loop,
+ * and provide a statically defined secret size to allow optimization of vector loop.
+ */
+XXH_NO_INLINE XXH64_hash_t
+XXH3_hashLong_64b_default(const void* XXH_RESTRICT input, size_t len,
+                          XXH64_hash_t seed64, const xxh_u8* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64; (void)secret; (void)secretLen;
+    return XXH3_hashLong_64b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*
+ * XXH3_hashLong_64b_withSeed():
+ * Generate a custom key based on alteration of default XXH3_kSecret with the seed,
+ * and then use this key for long mode hashing.
+ *
+ * This operation is decently fast but nonetheless costs a little bit of time.
+ * Try to avoid it whenever possible (typically when seed==0).
+ *
+ * It's important for performance that XXH3_hashLong is not inlined. Not sure
+ * why (uop cache maybe?), but the difference is large and easily measurable.
+ */
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSeed_internal(const void* input, size_t len,
+                                    XXH64_hash_t seed,
+                                    XXH3_f_accumulate_512 f_acc512,
+                                    XXH3_f_scrambleAcc f_scramble,
+                                    XXH3_f_initCustomSecret f_initSec)
+{
+    if (seed == 0)
+        return XXH3_hashLong_64b_internal(input, len,
+                                          XXH3_kSecret, sizeof(XXH3_kSecret),
+                                          f_acc512, f_scramble);
+    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+        f_initSec(secret, seed);
+        return XXH3_hashLong_64b_internal(input, len, secret, sizeof(secret),
+                                          f_acc512, f_scramble);
+    }
+}
+
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH64_hash_t
+XXH3_hashLong_64b_withSeed(const void* input, size_t len,
+                           XXH64_hash_t seed, const xxh_u8* secret, size_t secretLen)
+{
+    (void)secret; (void)secretLen;
+    return XXH3_hashLong_64b_withSeed_internal(input, len, seed,
+                XXH3_accumulate_512, XXH3_scrambleAcc, XXH3_initCustomSecret);
+}
+
+
+typedef XXH64_hash_t (*XXH3_hashLong64_f)(const void* XXH_RESTRICT, size_t,
+                                          XXH64_hash_t, const xxh_u8* XXH_RESTRICT, size_t);
+
+XXH_FORCE_INLINE XXH64_hash_t
+XXH3_64bits_internal(const void* XXH_RESTRICT input, size_t len,
+                     XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
+                     XXH3_hashLong64_f f_hashLong)
+{
+    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
+    /*
+     * If an action is to be taken if `secretLen` condition is not respected,
+     * it should be done here.
+     * For now, it's a contract pre-condition.
+     * Adding a check and a branch here would cost performance at every hash.
+     * Also, note that function signature doesn't offer room to return an error.
+     */
+    if (len <= 16)
+        return XXH3_len_0to16_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
+    if (len <= 128)
+        return XXH3_len_17to128_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_len_129to240_64b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    return f_hashLong(input, len, seed64, (const xxh_u8*)secret, secretLen);
+}
+
+
+/* ===   Public entry point   === */
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits(const void* input, size_t len)
+{
+    return XXH3_64bits_internal(input, len, 0, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_default);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecret(const void* input, size_t len, const void* secret, size_t secretSize)
+{
+    return XXH3_64bits_internal(input, len, 0, secret, secretSize, XXH3_hashLong_64b_withSecret);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSeed(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), XXH3_hashLong_64b_withSeed);
+}
+
+XXH_PUBLIC_API XXH64_hash_t
+XXH3_64bits_withSecretandSeed(const void* input, size_t len, const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_64bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
+    return XXH3_hashLong_64b_withSecret(input, len, seed, (const xxh_u8*)secret, secretSize);
+}
+
+
+/* ===   XXH3 streaming   === */
+
+/*
+ * Malloc's a pointer that is always aligned to align.
+ *
+ * This must be freed with `XXH_alignedFree()`.
+ *
+ * malloc typically guarantees 16 byte alignment on 64-bit systems and 8 byte
+ * alignment on 32-bit. This isn't enough for the 32 byte aligned loads in AVX2
+ * or on 32-bit, the 16 byte aligned loads in SSE2 and NEON.
+ *
+ * This underalignment previously caused a rather obvious crash which went
+ * completely unnoticed due to XXH3_createState() not actually being tested.
+ * Credit to RedSpah for noticing this bug.
+ *
+ * The alignment is done manually: Functions like posix_memalign or _mm_malloc
+ * are avoided: To maintain portability, we would have to write a fallback
+ * like this anyways, and besides, testing for the existence of library
+ * functions without relying on external build tools is impossible.
+ *
+ * The method is simple: Overallocate, manually align, and store the offset
+ * to the original behind the returned pointer.
+ *
+ * Align must be a power of 2 and 8 <= align <= 128.
+ */
+static void* XXH_alignedMalloc(size_t s, size_t align)
+{
+    XXH_ASSERT(align <= 128 && align >= 8); /* range check */
+    XXH_ASSERT((align & (align-1)) == 0);   /* power of 2 */
+    XXH_ASSERT(s != 0 && s < (s + align));  /* empty/overflow */
+    {   /* Overallocate to make room for manual realignment and an offset byte */
+        xxh_u8* base = (xxh_u8*)XXH_malloc(s + align);
+        if (base != NULL) {
+            /*
+             * Get the offset needed to align this pointer.
+             *
+             * Even if the returned pointer is aligned, there will always be
+             * at least one byte to store the offset to the original pointer.
+             */
+            size_t offset = align - ((size_t)base & (align - 1)); /* base % align */
+            /* Add the offset for the now-aligned pointer */
+            xxh_u8* ptr = base + offset;
+
+            XXH_ASSERT((size_t)ptr % align == 0);
+
+            /* Store the offset immediately before the returned pointer. */
+            ptr[-1] = (xxh_u8)offset;
+            return ptr;
+        }
+        return NULL;
+    }
+}
+/*
+ * Frees an aligned pointer allocated by XXH_alignedMalloc(). Don't pass
+ * normal malloc'd pointers, XXH_alignedMalloc has a specific data layout.
+ */
+static void XXH_alignedFree(void* p)
+{
+    if (p != NULL) {
+        xxh_u8* ptr = (xxh_u8*)p;
+        /* Get the offset byte we added in XXH_malloc. */
+        xxh_u8 offset = ptr[-1];
+        /* Free the original malloc'd pointer */
+        xxh_u8* base = ptr - offset;
+        XXH_free(base);
+    }
+}
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH3_state_t* XXH3_createState(void)
+{
+    XXH3_state_t* const state = (XXH3_state_t*)XXH_alignedMalloc(sizeof(XXH3_state_t), 64);
+    if (state==NULL) return NULL;
+    XXH3_INITSTATE(state);
+    return state;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode XXH3_freeState(XXH3_state_t* statePtr)
+{
+    XXH_alignedFree(statePtr);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API void
+XXH3_copyState(XXH3_state_t* dst_state, const XXH3_state_t* src_state)
+{
+    XXH_memcpy(dst_state, src_state, sizeof(*dst_state));
+}
+
+static void
+XXH3_reset_internal(XXH3_state_t* statePtr,
+                    XXH64_hash_t seed,
+                    const void* secret, size_t secretSize)
+{
+    size_t const initStart = offsetof(XXH3_state_t, bufferedSize);
+    size_t const initLength = offsetof(XXH3_state_t, nbStripesPerBlock) - initStart;
+    XXH_ASSERT(offsetof(XXH3_state_t, nbStripesPerBlock) > initStart);
+    XXH_ASSERT(statePtr != NULL);
+    /* set members from bufferedSize to nbStripesPerBlock (excluded) to 0 */
+    memset((char*)statePtr + initStart, 0, initLength);
+    statePtr->acc[0] = XXH_PRIME32_3;
+    statePtr->acc[1] = XXH_PRIME64_1;
+    statePtr->acc[2] = XXH_PRIME64_2;
+    statePtr->acc[3] = XXH_PRIME64_3;
+    statePtr->acc[4] = XXH_PRIME64_4;
+    statePtr->acc[5] = XXH_PRIME32_2;
+    statePtr->acc[6] = XXH_PRIME64_5;
+    statePtr->acc[7] = XXH_PRIME32_1;
+    statePtr->seed = seed;
+    statePtr->useSeed = (seed != 0);
+    statePtr->extSecret = (const unsigned char*)secret;
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+    statePtr->secretLimit = secretSize - XXH_STRIPE_LEN;
+    statePtr->nbStripesPerBlock = statePtr->secretLimit / XXH_SECRET_CONSUME_RATE;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset(XXH3_state_t* statePtr)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, 0, XXH3_kSecret, XXH_SECRET_DEFAULT_SIZE);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, 0, secret, secretSize);
+    if (secret == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    if (seed==0) return XXH3_64bits_reset(statePtr);
+    if ((seed != statePtr->seed) || (statePtr->extSecret != NULL))
+        XXH3_initCustomSecret(statePtr->customSecret, seed);
+    XXH3_reset_internal(statePtr, seed, NULL, XXH_SECRET_DEFAULT_SIZE);
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret, size_t secretSize, XXH64_hash_t seed64)
+{
+    if (statePtr == NULL) return XXH_ERROR;
+    if (secret == NULL) return XXH_ERROR;
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+    XXH3_reset_internal(statePtr, seed64, secret, secretSize);
+    statePtr->useSeed = 1; /* always, even if seed64==0 */
+    return XXH_OK;
+}
+
+/* Note : when XXH3_consumeStripes() is invoked,
+ * there must be a guarantee that at least one more byte must be consumed from input
+ * so that the function can blindly consume all stripes using the "normal" secret segment */
+XXH_FORCE_INLINE void
+XXH3_consumeStripes(xxh_u64* XXH_RESTRICT acc,
+                    size_t* XXH_RESTRICT nbStripesSoFarPtr, size_t nbStripesPerBlock,
+                    const xxh_u8* XXH_RESTRICT input, size_t nbStripes,
+                    const xxh_u8* XXH_RESTRICT secret, size_t secretLimit,
+                    XXH3_f_accumulate_512 f_acc512,
+                    XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ASSERT(nbStripes <= nbStripesPerBlock);  /* can handle max 1 scramble per invocation */
+    XXH_ASSERT(*nbStripesSoFarPtr < nbStripesPerBlock);
+    if (nbStripesPerBlock - *nbStripesSoFarPtr <= nbStripes) {
+        /* need a scrambling operation */
+        size_t const nbStripesToEndofBlock = nbStripesPerBlock - *nbStripesSoFarPtr;
+        size_t const nbStripesAfterBlock = nbStripes - nbStripesToEndofBlock;
+        XXH3_accumulate(acc, input, secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE, nbStripesToEndofBlock, f_acc512);
+        f_scramble(acc, secret + secretLimit);
+        XXH3_accumulate(acc, input + nbStripesToEndofBlock * XXH_STRIPE_LEN, secret, nbStripesAfterBlock, f_acc512);
+        *nbStripesSoFarPtr = nbStripesAfterBlock;
+    } else {
+        XXH3_accumulate(acc, input, secret + nbStripesSoFarPtr[0] * XXH_SECRET_CONSUME_RATE, nbStripes, f_acc512);
+        *nbStripesSoFarPtr += nbStripes;
+    }
+}
+
+#ifndef XXH3_STREAM_USE_STACK
+# ifndef __clang__ /* clang doesn't need additional stack space */
+#   define XXH3_STREAM_USE_STACK 1
+# endif
+#endif
+/*
+ * Both XXH3_64bits_update and XXH3_128bits_update use this routine.
+ */
+XXH_FORCE_INLINE XXH_errorcode
+XXH3_update(XXH3_state_t* XXH_RESTRICT const state,
+            const xxh_u8* XXH_RESTRICT input, size_t len,
+            XXH3_f_accumulate_512 f_acc512,
+            XXH3_f_scrambleAcc f_scramble)
+{
+    if (input==NULL) {
+        XXH_ASSERT(len == 0);
+        return XXH_OK;
+    }
+
+    XXH_ASSERT(state != NULL);
+    {   const xxh_u8* const bEnd = input + len;
+        const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
+        /* For some reason, gcc and MSVC seem to suffer greatly
+         * when operating accumulators directly into state.
+         * Operating into stack space seems to enable proper optimization.
+         * clang, on the other hand, doesn't seem to need this trick */
+        XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[8]; memcpy(acc, state->acc, sizeof(acc));
+#else
+        xxh_u64* XXH_RESTRICT const acc = state->acc;
+#endif
+        state->totalLen += len;
+        XXH_ASSERT(state->bufferedSize <= XXH3_INTERNALBUFFER_SIZE);
+
+        /* small input : just fill in tmp buffer */
+        if (state->bufferedSize + len <= XXH3_INTERNALBUFFER_SIZE) {
+            XXH_memcpy(state->buffer + state->bufferedSize, input, len);
+            state->bufferedSize += (XXH32_hash_t)len;
+            return XXH_OK;
+        }
+
+        /* total input is now > XXH3_INTERNALBUFFER_SIZE */
+        #define XXH3_INTERNALBUFFER_STRIPES (XXH3_INTERNALBUFFER_SIZE / XXH_STRIPE_LEN)
+        XXH_STATIC_ASSERT(XXH3_INTERNALBUFFER_SIZE % XXH_STRIPE_LEN == 0);   /* clean multiple */
+
+        /*
+         * Internal buffer is partially filled (always, except at beginning)
+         * Complete it, then consume it.
+         */
+        if (state->bufferedSize) {
+            size_t const loadSize = XXH3_INTERNALBUFFER_SIZE - state->bufferedSize;
+            XXH_memcpy(state->buffer + state->bufferedSize, input, loadSize);
+            input += loadSize;
+            XXH3_consumeStripes(acc,
+                               &state->nbStripesSoFar, state->nbStripesPerBlock,
+                                state->buffer, XXH3_INTERNALBUFFER_STRIPES,
+                                secret, state->secretLimit,
+                                f_acc512, f_scramble);
+            state->bufferedSize = 0;
+        }
+        XXH_ASSERT(input < bEnd);
+
+        /* large input to consume : ingest per full block */
+        if ((size_t)(bEnd - input) > state->nbStripesPerBlock * XXH_STRIPE_LEN) {
+            size_t nbStripes = (size_t)(bEnd - 1 - input) / XXH_STRIPE_LEN;
+            XXH_ASSERT(state->nbStripesPerBlock >= state->nbStripesSoFar);
+            /* join to current block's end */
+            {   size_t const nbStripesToEnd = state->nbStripesPerBlock - state->nbStripesSoFar;
+                XXH_ASSERT(nbStripes <= nbStripes);
+                XXH3_accumulate(acc, input, secret + state->nbStripesSoFar * XXH_SECRET_CONSUME_RATE, nbStripesToEnd, f_acc512);
+                f_scramble(acc, secret + state->secretLimit);
+                state->nbStripesSoFar = 0;
+                input += nbStripesToEnd * XXH_STRIPE_LEN;
+                nbStripes -= nbStripesToEnd;
+            }
+            /* consume per entire blocks */
+            while(nbStripes >= state->nbStripesPerBlock) {
+                XXH3_accumulate(acc, input, secret, state->nbStripesPerBlock, f_acc512);
+                f_scramble(acc, secret + state->secretLimit);
+                input += state->nbStripesPerBlock * XXH_STRIPE_LEN;
+                nbStripes -= state->nbStripesPerBlock;
+            }
+            /* consume last partial block */
+            XXH3_accumulate(acc, input, secret, nbStripes, f_acc512);
+            input += nbStripes * XXH_STRIPE_LEN;
+            XXH_ASSERT(input < bEnd);  /* at least some bytes left */
+            state->nbStripesSoFar = nbStripes;
+            /* buffer predecessor of last partial stripe */
+            XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
+            XXH_ASSERT(bEnd - input <= XXH_STRIPE_LEN);
+        } else {
+            /* content to consume <= block size */
+            /* Consume input by a multiple of internal buffer size */
+            if (bEnd - input > XXH3_INTERNALBUFFER_SIZE) {
+                const xxh_u8* const limit = bEnd - XXH3_INTERNALBUFFER_SIZE;
+                do {
+                    XXH3_consumeStripes(acc,
+                                       &state->nbStripesSoFar, state->nbStripesPerBlock,
+                                        input, XXH3_INTERNALBUFFER_STRIPES,
+                                        secret, state->secretLimit,
+                                        f_acc512, f_scramble);
+                    input += XXH3_INTERNALBUFFER_SIZE;
+                } while (input<limit);
+                /* buffer predecessor of last partial stripe */
+                XXH_memcpy(state->buffer + sizeof(state->buffer) - XXH_STRIPE_LEN, input - XXH_STRIPE_LEN, XXH_STRIPE_LEN);
+            }
+        }
+
+        /* Some remaining input (always) : buffer it */
+        XXH_ASSERT(input < bEnd);
+        XXH_ASSERT(bEnd - input <= XXH3_INTERNALBUFFER_SIZE);
+        XXH_ASSERT(state->bufferedSize == 0);
+        XXH_memcpy(state->buffer, input, (size_t)(bEnd-input));
+        state->bufferedSize = (XXH32_hash_t)(bEnd-input);
+#if defined(XXH3_STREAM_USE_STACK) && XXH3_STREAM_USE_STACK >= 1
+        /* save stack accumulators into state */
+        memcpy(state->acc, acc, sizeof(acc));
+#endif
+    }
+
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_64bits_update(XXH3_state_t* state, const void* input, size_t len)
+{
+    return XXH3_update(state, (const xxh_u8*)input, len,
+                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+
+XXH_FORCE_INLINE void
+XXH3_digest_long (XXH64_hash_t* acc,
+                  const XXH3_state_t* state,
+                  const unsigned char* secret)
+{
+    /*
+     * Digest on a local copy. This way, the state remains unaltered, and it can
+     * continue ingesting more input afterwards.
+     */
+    XXH_memcpy(acc, state->acc, sizeof(state->acc));
+    if (state->bufferedSize >= XXH_STRIPE_LEN) {
+        size_t const nbStripes = (state->bufferedSize - 1) / XXH_STRIPE_LEN;
+        size_t nbStripesSoFar = state->nbStripesSoFar;
+        XXH3_consumeStripes(acc,
+                           &nbStripesSoFar, state->nbStripesPerBlock,
+                            state->buffer, nbStripes,
+                            secret, state->secretLimit,
+                            XXH3_accumulate_512, XXH3_scrambleAcc);
+        /* last stripe */
+        XXH3_accumulate_512(acc,
+                            state->buffer + state->bufferedSize - XXH_STRIPE_LEN,
+                            secret + state->secretLimit - XXH_SECRET_LASTACC_START);
+    } else {  /* bufferedSize < XXH_STRIPE_LEN */
+        xxh_u8 lastStripe[XXH_STRIPE_LEN];
+        size_t const catchupSize = XXH_STRIPE_LEN - state->bufferedSize;
+        XXH_ASSERT(state->bufferedSize > 0);  /* there is always some input buffered */
+        XXH_memcpy(lastStripe, state->buffer + sizeof(state->buffer) - catchupSize, catchupSize);
+        XXH_memcpy(lastStripe + catchupSize, state->buffer, state->bufferedSize);
+        XXH3_accumulate_512(acc,
+                            lastStripe,
+                            secret + state->secretLimit - XXH_SECRET_LASTACC_START);
+    }
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH64_hash_t XXH3_64bits_digest (const XXH3_state_t* state)
+{
+    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+    if (state->totalLen > XXH3_MIDSIZE_MAX) {
+        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
+        XXH3_digest_long(acc, state, secret);
+        return XXH3_mergeAccs(acc,
+                              secret + XXH_SECRET_MERGEACCS_START,
+                              (xxh_u64)state->totalLen * XXH_PRIME64_1);
+    }
+    /* totalLen <= XXH3_MIDSIZE_MAX: digesting a short input */
+    if (state->useSeed)
+        return XXH3_64bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
+    return XXH3_64bits_withSecret(state->buffer, (size_t)(state->totalLen),
+                                  secret, state->secretLimit + XXH_STRIPE_LEN);
+}
+
+
+
+/* ==========================================
+ * XXH3 128 bits (a.k.a XXH128)
+ * ==========================================
+ * XXH3's 128-bit variant has better mixing and strength than the 64-bit variant,
+ * even without counting the significantly larger output size.
+ *
+ * For example, extra steps are taken to avoid the seed-dependent collisions
+ * in 17-240 byte inputs (See XXH3_mix16B and XXH128_mix32B).
+ *
+ * This strength naturally comes at the cost of some speed, especially on short
+ * lengths. Note that longer hashes are about as fast as the 64-bit version
+ * due to it using only a slight modification of the 64-bit loop.
+ *
+ * XXH128 is also more oriented towards 64-bit machines. It is still extremely
+ * fast for a _128-bit_ hash on 32-bit (it usually clears XXH64).
+ */
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_1to3_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    /* A doubled version of 1to3_64b with different constants. */
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(1 <= len && len <= 3);
+    XXH_ASSERT(secret != NULL);
+    /*
+     * len = 1: combinedl = { input[0], 0x01, input[0], input[0] }
+     * len = 2: combinedl = { input[1], 0x02, input[0], input[1] }
+     * len = 3: combinedl = { input[2], 0x03, input[0], input[1] }
+     */
+    {   xxh_u8 const c1 = input[0];
+        xxh_u8 const c2 = input[len >> 1];
+        xxh_u8 const c3 = input[len - 1];
+        xxh_u32 const combinedl = ((xxh_u32)c1 <<16) | ((xxh_u32)c2 << 24)
+                                | ((xxh_u32)c3 << 0) | ((xxh_u32)len << 8);
+        xxh_u32 const combinedh = XXH_rotl32(XXH_swap32(combinedl), 13);
+        xxh_u64 const bitflipl = (XXH_readLE32(secret) ^ XXH_readLE32(secret+4)) + seed;
+        xxh_u64 const bitfliph = (XXH_readLE32(secret+8) ^ XXH_readLE32(secret+12)) - seed;
+        xxh_u64 const keyed_lo = (xxh_u64)combinedl ^ bitflipl;
+        xxh_u64 const keyed_hi = (xxh_u64)combinedh ^ bitfliph;
+        XXH128_hash_t h128;
+        h128.low64  = XXH64_avalanche(keyed_lo);
+        h128.high64 = XXH64_avalanche(keyed_hi);
+        return h128;
+    }
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_4to8_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(4 <= len && len <= 8);
+    seed ^= (xxh_u64)XXH_swap32((xxh_u32)seed) << 32;
+    {   xxh_u32 const input_lo = XXH_readLE32(input);
+        xxh_u32 const input_hi = XXH_readLE32(input + len - 4);
+        xxh_u64 const input_64 = input_lo + ((xxh_u64)input_hi << 32);
+        xxh_u64 const bitflip = (XXH_readLE64(secret+16) ^ XXH_readLE64(secret+24)) + seed;
+        xxh_u64 const keyed = input_64 ^ bitflip;
+
+        /* Shift len to the left to ensure it is even, this avoids even multiplies. */
+        XXH128_hash_t m128 = XXH_mult64to128(keyed, XXH_PRIME64_1 + (len << 2));
+
+        m128.high64 += (m128.low64 << 1);
+        m128.low64  ^= (m128.high64 >> 3);
+
+        m128.low64   = XXH_xorshift64(m128.low64, 35);
+        m128.low64  *= 0x9FB21C651E98DF25ULL;
+        m128.low64   = XXH_xorshift64(m128.low64, 28);
+        m128.high64  = XXH3_avalanche(m128.high64);
+        return m128;
+    }
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_9to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(input != NULL);
+    XXH_ASSERT(secret != NULL);
+    XXH_ASSERT(9 <= len && len <= 16);
+    {   xxh_u64 const bitflipl = (XXH_readLE64(secret+32) ^ XXH_readLE64(secret+40)) - seed;
+        xxh_u64 const bitfliph = (XXH_readLE64(secret+48) ^ XXH_readLE64(secret+56)) + seed;
+        xxh_u64 const input_lo = XXH_readLE64(input);
+        xxh_u64       input_hi = XXH_readLE64(input + len - 8);
+        XXH128_hash_t m128 = XXH_mult64to128(input_lo ^ input_hi ^ bitflipl, XXH_PRIME64_1);
+        /*
+         * Put len in the middle of m128 to ensure that the length gets mixed to
+         * both the low and high bits in the 128x64 multiply below.
+         */
+        m128.low64 += (xxh_u64)(len - 1) << 54;
+        input_hi   ^= bitfliph;
+        /*
+         * Add the high 32 bits of input_hi to the high 32 bits of m128, then
+         * add the long product of the low 32 bits of input_hi and XXH_PRIME32_2 to
+         * the high 64 bits of m128.
+         *
+         * The best approach to this operation is different on 32-bit and 64-bit.
+         */
+        if (sizeof(void *) < sizeof(xxh_u64)) { /* 32-bit */
+            /*
+             * 32-bit optimized version, which is more readable.
+             *
+             * On 32-bit, it removes an ADC and delays a dependency between the two
+             * halves of m128.high64, but it generates an extra mask on 64-bit.
+             */
+            m128.high64 += (input_hi & 0xFFFFFFFF00000000ULL) + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2);
+        } else {
+            /*
+             * 64-bit optimized (albeit more confusing) version.
+             *
+             * Uses some properties of addition and multiplication to remove the mask:
+             *
+             * Let:
+             *    a = input_hi.lo = (input_hi & 0x00000000FFFFFFFF)
+             *    b = input_hi.hi = (input_hi & 0xFFFFFFFF00000000)
+             *    c = XXH_PRIME32_2
+             *
+             *    a + (b * c)
+             * Inverse Property: x + y - x == y
+             *    a + (b * (1 + c - 1))
+             * Distributive Property: x * (y + z) == (x * y) + (x * z)
+             *    a + (b * 1) + (b * (c - 1))
+             * Identity Property: x * 1 == x
+             *    a + b + (b * (c - 1))
+             *
+             * Substitute a, b, and c:
+             *    input_hi.hi + input_hi.lo + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
+             *
+             * Since input_hi.hi + input_hi.lo == input_hi, we get this:
+             *    input_hi + ((xxh_u64)input_hi.lo * (XXH_PRIME32_2 - 1))
+             */
+            m128.high64 += input_hi + XXH_mult32to64((xxh_u32)input_hi, XXH_PRIME32_2 - 1);
+        }
+        /* m128 ^= XXH_swap64(m128 >> 64); */
+        m128.low64  ^= XXH_swap64(m128.high64);
+
+        {   /* 128x64 multiply: h128 = m128 * XXH_PRIME64_2; */
+            XXH128_hash_t h128 = XXH_mult64to128(m128.low64, XXH_PRIME64_2);
+            h128.high64 += m128.high64 * XXH_PRIME64_2;
+
+            h128.low64   = XXH3_avalanche(h128.low64);
+            h128.high64  = XXH3_avalanche(h128.high64);
+            return h128;
+    }   }
+}
+
+/*
+ * Assumption: `secret` size is >= XXH3_SECRET_SIZE_MIN
+ */
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_0to16_128b(const xxh_u8* input, size_t len, const xxh_u8* secret, XXH64_hash_t seed)
+{
+    XXH_ASSERT(len <= 16);
+    {   if (len > 8) return XXH3_len_9to16_128b(input, len, secret, seed);
+        if (len >= 4) return XXH3_len_4to8_128b(input, len, secret, seed);
+        if (len) return XXH3_len_1to3_128b(input, len, secret, seed);
+        {   XXH128_hash_t h128;
+            xxh_u64 const bitflipl = XXH_readLE64(secret+64) ^ XXH_readLE64(secret+72);
+            xxh_u64 const bitfliph = XXH_readLE64(secret+80) ^ XXH_readLE64(secret+88);
+            h128.low64 = XXH64_avalanche(seed ^ bitflipl);
+            h128.high64 = XXH64_avalanche( seed ^ bitfliph);
+            return h128;
+    }   }
+}
+
+/*
+ * A bit slower than XXH3_mix16B, but handles multiply by zero better.
+ */
+XXH_FORCE_INLINE XXH128_hash_t
+XXH128_mix32B(XXH128_hash_t acc, const xxh_u8* input_1, const xxh_u8* input_2,
+              const xxh_u8* secret, XXH64_hash_t seed)
+{
+    acc.low64  += XXH3_mix16B (input_1, secret+0, seed);
+    acc.low64  ^= XXH_readLE64(input_2) + XXH_readLE64(input_2 + 8);
+    acc.high64 += XXH3_mix16B (input_2, secret+16, seed);
+    acc.high64 ^= XXH_readLE64(input_1) + XXH_readLE64(input_1 + 8);
+    return acc;
+}
+
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_len_17to128_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                      const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                      XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(16 < len && len <= 128);
+
+    {   XXH128_hash_t acc;
+        acc.low64 = len * XXH_PRIME64_1;
+        acc.high64 = 0;
+        if (len > 32) {
+            if (len > 64) {
+                if (len > 96) {
+                    acc = XXH128_mix32B(acc, input+48, input+len-64, secret+96, seed);
+                }
+                acc = XXH128_mix32B(acc, input+32, input+len-48, secret+64, seed);
+            }
+            acc = XXH128_mix32B(acc, input+16, input+len-32, secret+32, seed);
+        }
+        acc = XXH128_mix32B(acc, input, input+len-16, secret, seed);
+        {   XXH128_hash_t h128;
+            h128.low64  = acc.low64 + acc.high64;
+            h128.high64 = (acc.low64    * XXH_PRIME64_1)
+                        + (acc.high64   * XXH_PRIME64_4)
+                        + ((len - seed) * XXH_PRIME64_2);
+            h128.low64  = XXH3_avalanche(h128.low64);
+            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
+            return h128;
+        }
+    }
+}
+
+XXH_NO_INLINE XXH128_hash_t
+XXH3_len_129to240_128b(const xxh_u8* XXH_RESTRICT input, size_t len,
+                       const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                       XXH64_hash_t seed)
+{
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN); (void)secretSize;
+    XXH_ASSERT(128 < len && len <= XXH3_MIDSIZE_MAX);
+
+    {   XXH128_hash_t acc;
+        int const nbRounds = (int)len / 32;
+        int i;
+        acc.low64 = len * XXH_PRIME64_1;
+        acc.high64 = 0;
+        for (i=0; i<4; i++) {
+            acc = XXH128_mix32B(acc,
+                                input  + (32 * i),
+                                input  + (32 * i) + 16,
+                                secret + (32 * i),
+                                seed);
+        }
+        acc.low64 = XXH3_avalanche(acc.low64);
+        acc.high64 = XXH3_avalanche(acc.high64);
+        XXH_ASSERT(nbRounds >= 4);
+        for (i=4 ; i < nbRounds; i++) {
+            acc = XXH128_mix32B(acc,
+                                input + (32 * i),
+                                input + (32 * i) + 16,
+                                secret + XXH3_MIDSIZE_STARTOFFSET + (32 * (i - 4)),
+                                seed);
+        }
+        /* last bytes */
+        acc = XXH128_mix32B(acc,
+                            input + len - 16,
+                            input + len - 32,
+                            secret + XXH3_SECRET_SIZE_MIN - XXH3_MIDSIZE_LASTOFFSET - 16,
+                            0ULL - seed);
+
+        {   XXH128_hash_t h128;
+            h128.low64  = acc.low64 + acc.high64;
+            h128.high64 = (acc.low64    * XXH_PRIME64_1)
+                        + (acc.high64   * XXH_PRIME64_4)
+                        + ((len - seed) * XXH_PRIME64_2);
+            h128.low64  = XXH3_avalanche(h128.low64);
+            h128.high64 = (XXH64_hash_t)0 - XXH3_avalanche(h128.high64);
+            return h128;
+        }
+    }
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_internal(const void* XXH_RESTRICT input, size_t len,
+                            const xxh_u8* XXH_RESTRICT secret, size_t secretSize,
+                            XXH3_f_accumulate_512 f_acc512,
+                            XXH3_f_scrambleAcc f_scramble)
+{
+    XXH_ALIGN(XXH_ACC_ALIGN) xxh_u64 acc[XXH_ACC_NB] = XXH3_INIT_ACC;
+
+    XXH3_hashLong_internal_loop(acc, (const xxh_u8*)input, len, secret, secretSize, f_acc512, f_scramble);
+
+    /* converge into final hash */
+    XXH_STATIC_ASSERT(sizeof(acc) == 64);
+    XXH_ASSERT(secretSize >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+    {   XXH128_hash_t h128;
+        h128.low64  = XXH3_mergeAccs(acc,
+                                     secret + XXH_SECRET_MERGEACCS_START,
+                                     (xxh_u64)len * XXH_PRIME64_1);
+        h128.high64 = XXH3_mergeAccs(acc,
+                                     secret + secretSize
+                                            - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
+                                     ~((xxh_u64)len * XXH_PRIME64_2));
+        return h128;
+    }
+}
+
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH128_hash_t
+XXH3_hashLong_128b_default(const void* XXH_RESTRICT input, size_t len,
+                           XXH64_hash_t seed64,
+                           const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64; (void)secret; (void)secretLen;
+    return XXH3_hashLong_128b_internal(input, len, XXH3_kSecret, sizeof(XXH3_kSecret),
+                                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*
+ * It's important for performance to pass @secretLen (when it's static)
+ * to the compiler, so that it can properly optimize the vectorized loop.
+ */
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSecret(const void* XXH_RESTRICT input, size_t len,
+                              XXH64_hash_t seed64,
+                              const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)seed64;
+    return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, secretLen,
+                                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSeed_internal(const void* XXH_RESTRICT input, size_t len,
+                                XXH64_hash_t seed64,
+                                XXH3_f_accumulate_512 f_acc512,
+                                XXH3_f_scrambleAcc f_scramble,
+                                XXH3_f_initCustomSecret f_initSec)
+{
+    if (seed64 == 0)
+        return XXH3_hashLong_128b_internal(input, len,
+                                           XXH3_kSecret, sizeof(XXH3_kSecret),
+                                           f_acc512, f_scramble);
+    {   XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+        f_initSec(secret, seed64);
+        return XXH3_hashLong_128b_internal(input, len, (const xxh_u8*)secret, sizeof(secret),
+                                           f_acc512, f_scramble);
+    }
+}
+
+/*
+ * It's important for performance that XXH3_hashLong is not inlined.
+ */
+XXH_NO_INLINE XXH128_hash_t
+XXH3_hashLong_128b_withSeed(const void* input, size_t len,
+                            XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen)
+{
+    (void)secret; (void)secretLen;
+    return XXH3_hashLong_128b_withSeed_internal(input, len, seed64,
+                XXH3_accumulate_512, XXH3_scrambleAcc, XXH3_initCustomSecret);
+}
+
+typedef XXH128_hash_t (*XXH3_hashLong128_f)(const void* XXH_RESTRICT, size_t,
+                                            XXH64_hash_t, const void* XXH_RESTRICT, size_t);
+
+XXH_FORCE_INLINE XXH128_hash_t
+XXH3_128bits_internal(const void* input, size_t len,
+                      XXH64_hash_t seed64, const void* XXH_RESTRICT secret, size_t secretLen,
+                      XXH3_hashLong128_f f_hl128)
+{
+    XXH_ASSERT(secretLen >= XXH3_SECRET_SIZE_MIN);
+    /*
+     * If an action is to be taken if `secret` conditions are not respected,
+     * it should be done here.
+     * For now, it's a contract pre-condition.
+     * Adding a check and a branch here would cost performance at every hash.
+     */
+    if (len <= 16)
+        return XXH3_len_0to16_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, seed64);
+    if (len <= 128)
+        return XXH3_len_17to128_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_len_129to240_128b((const xxh_u8*)input, len, (const xxh_u8*)secret, secretLen, seed64);
+    return f_hl128(input, len, seed64, secret, secretLen);
+}
+
+
+/* ===   Public XXH128 API   === */
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits(const void* input, size_t len)
+{
+    return XXH3_128bits_internal(input, len, 0,
+                                 XXH3_kSecret, sizeof(XXH3_kSecret),
+                                 XXH3_hashLong_128b_default);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecret(const void* input, size_t len, const void* secret, size_t secretSize)
+{
+    return XXH3_128bits_internal(input, len, 0,
+                                 (const xxh_u8*)secret, secretSize,
+                                 XXH3_hashLong_128b_withSecret);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSeed(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_128bits_internal(input, len, seed,
+                                 XXH3_kSecret, sizeof(XXH3_kSecret),
+                                 XXH3_hashLong_128b_withSeed);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH3_128bits_withSecretandSeed(const void* input, size_t len, const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    if (len <= XXH3_MIDSIZE_MAX)
+        return XXH3_128bits_internal(input, len, seed, XXH3_kSecret, sizeof(XXH3_kSecret), NULL);
+    return XXH3_hashLong_128b_withSecret(input, len, seed, secret, secretSize);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH128(const void* input, size_t len, XXH64_hash_t seed)
+{
+    return XXH3_128bits_withSeed(input, len, seed);
+}
+
+
+/* ===   XXH3 128-bit streaming   === */
+
+/*
+ * All initialization and update functions are identical to 64-bit streaming variant.
+ * The only difference is the finalization routine.
+ */
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset(XXH3_state_t* statePtr)
+{
+    return XXH3_64bits_reset(statePtr);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecret(XXH3_state_t* statePtr, const void* secret, size_t secretSize)
+{
+    return XXH3_64bits_reset_withSecret(statePtr, secret, secretSize);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSeed(XXH3_state_t* statePtr, XXH64_hash_t seed)
+{
+    return XXH3_64bits_reset_withSeed(statePtr, seed);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_reset_withSecretandSeed(XXH3_state_t* statePtr, const void* secret, size_t secretSize, XXH64_hash_t seed)
+{
+    return XXH3_64bits_reset_withSecretandSeed(statePtr, secret, secretSize, seed);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_128bits_update(XXH3_state_t* state, const void* input, size_t len)
+{
+    return XXH3_update(state, (const xxh_u8*)input, len,
+                       XXH3_accumulate_512, XXH3_scrambleAcc);
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t XXH3_128bits_digest (const XXH3_state_t* state)
+{
+    const unsigned char* const secret = (state->extSecret == NULL) ? state->customSecret : state->extSecret;
+    if (state->totalLen > XXH3_MIDSIZE_MAX) {
+        XXH_ALIGN(XXH_ACC_ALIGN) XXH64_hash_t acc[XXH_ACC_NB];
+        XXH3_digest_long(acc, state, secret);
+        XXH_ASSERT(state->secretLimit + XXH_STRIPE_LEN >= sizeof(acc) + XXH_SECRET_MERGEACCS_START);
+        {   XXH128_hash_t h128;
+            h128.low64  = XXH3_mergeAccs(acc,
+                                         secret + XXH_SECRET_MERGEACCS_START,
+                                         (xxh_u64)state->totalLen * XXH_PRIME64_1);
+            h128.high64 = XXH3_mergeAccs(acc,
+                                         secret + state->secretLimit + XXH_STRIPE_LEN
+                                                - sizeof(acc) - XXH_SECRET_MERGEACCS_START,
+                                         ~((xxh_u64)state->totalLen * XXH_PRIME64_2));
+            return h128;
+        }
+    }
+    /* len <= XXH3_MIDSIZE_MAX : short code */
+    if (state->seed)
+        return XXH3_128bits_withSeed(state->buffer, (size_t)state->totalLen, state->seed);
+    return XXH3_128bits_withSecret(state->buffer, (size_t)(state->totalLen),
+                                   secret, state->secretLimit + XXH_STRIPE_LEN);
+}
+
+/* 128-bit utility functions */
+
+#include <string.h>   /* memcmp, memcpy */
+
+/* return : 1 is equal, 0 if different */
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API int XXH128_isEqual(XXH128_hash_t h1, XXH128_hash_t h2)
+{
+    /* note : XXH128_hash_t is compact, it has no padding byte */
+    return !(memcmp(&h1, &h2, sizeof(h1)));
+}
+
+/* This prototype is compatible with stdlib's qsort().
+ * return : >0 if *h128_1  > *h128_2
+ *          <0 if *h128_1  < *h128_2
+ *          =0 if *h128_1 == *h128_2  */
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API int XXH128_cmp(const void* h128_1, const void* h128_2)
+{
+    XXH128_hash_t const h1 = *(const XXH128_hash_t*)h128_1;
+    XXH128_hash_t const h2 = *(const XXH128_hash_t*)h128_2;
+    int const hcmp = (h1.high64 > h2.high64) - (h2.high64 > h1.high64);
+    /* note : bets that, in most cases, hash values are different */
+    if (hcmp) return hcmp;
+    return (h1.low64 > h2.low64) - (h2.low64 > h1.low64);
+}
+
+
+/*======   Canonical representation   ======*/
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API void
+XXH128_canonicalFromHash(XXH128_canonical_t* dst, XXH128_hash_t hash)
+{
+    XXH_STATIC_ASSERT(sizeof(XXH128_canonical_t) == sizeof(XXH128_hash_t));
+    if (XXH_CPU_LITTLE_ENDIAN) {
+        hash.high64 = XXH_swap64(hash.high64);
+        hash.low64  = XXH_swap64(hash.low64);
+    }
+    XXH_memcpy(dst, &hash.high64, sizeof(hash.high64));
+    XXH_memcpy((char*)dst + sizeof(hash.high64), &hash.low64, sizeof(hash.low64));
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH128_hash_t
+XXH128_hashFromCanonical(const XXH128_canonical_t* src)
+{
+    XXH128_hash_t h;
+    h.high64 = XXH_readBE64(src);
+    h.low64  = XXH_readBE64(src->digest + 8);
+    return h;
+}
+
+
+
+/* ==========================================
+ * Secret generators
+ * ==========================================
+ */
+#define XXH_MIN(x, y) (((x) > (y)) ? (y) : (x))
+
+static void XXH3_combine16(void* dst, XXH128_hash_t h128)
+{
+    XXH_writeLE64( dst, XXH_readLE64(dst) ^ h128.low64 );
+    XXH_writeLE64( (char*)dst+8, XXH_readLE64((char*)dst+8) ^ h128.high64 );
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API XXH_errorcode
+XXH3_generateSecret(void* secretBuffer, size_t secretSize, const void* customSeed, size_t customSeedSize)
+{
+    XXH_ASSERT(secretBuffer != NULL);
+    if (secretBuffer == NULL) return XXH_ERROR;
+    XXH_ASSERT(secretSize >= XXH3_SECRET_SIZE_MIN);
+    if (secretSize < XXH3_SECRET_SIZE_MIN) return XXH_ERROR;
+    if (customSeedSize == 0) {
+        customSeed = XXH3_kSecret;
+        customSeedSize = XXH_SECRET_DEFAULT_SIZE;
+    }
+    XXH_ASSERT(customSeed != NULL);
+    if (customSeed == NULL) return XXH_ERROR;
+
+    /* Fill secretBuffer with a copy of customSeed - repeat as needed */
+    {   size_t pos = 0;
+        while (pos < secretSize) {
+            size_t const toCopy = XXH_MIN((secretSize - pos), customSeedSize);
+            memcpy((char*)secretBuffer + pos, customSeed, toCopy);
+            pos += toCopy;
+    }   }
+
+    {   size_t const nbSeg16 = secretSize / 16;
+        size_t n;
+        XXH128_canonical_t scrambler;
+        XXH128_canonicalFromHash(&scrambler, XXH128(customSeed, customSeedSize, 0));
+        for (n=0; n<nbSeg16; n++) {
+            XXH128_hash_t const h128 = XXH128(&scrambler, sizeof(scrambler), n);
+            XXH3_combine16((char*)secretBuffer + n*16, h128);
+        }
+        /* last segment */
+        XXH3_combine16((char*)secretBuffer + secretSize - 16, XXH128_hashFromCanonical(&scrambler));
+    }
+    return XXH_OK;
+}
+
+/*! @ingroup xxh3_family */
+XXH_PUBLIC_API void
+XXH3_generateSecret_fromSeed(void* secretBuffer, XXH64_hash_t seed)
+{
+    XXH_ALIGN(XXH_SEC_ALIGN) xxh_u8 secret[XXH_SECRET_DEFAULT_SIZE];
+    XXH3_initCustomSecret(secret, seed);
+    XXH_ASSERT(secretBuffer != NULL);
+    memcpy(secretBuffer, secret, XXH_SECRET_DEFAULT_SIZE);
+}
+
+
+
+/* Pop our optimization override from above */
+#if XXH_VECTOR == XXH_AVX2 /* AVX2 */ \
+  && defined(__GNUC__) && !defined(__clang__) /* GCC, not Clang */ \
+  && defined(__OPTIMIZE__) && !defined(__OPTIMIZE_SIZE__) /* respect -O0 and -Os */
+#  pragma GCC pop_options
+#endif
+
+#endif  /* XXH_NO_LONG_LONG */
+
+#endif  /* XXH_NO_XXH3 */
+
+/*!
+ * @}
+ */
+#endif  /* XXH_IMPLEMENTATION */
+
+
+#if defined (__cplusplus)
+}
+#endif
diff --git a/fluent-bit/lib/cfl/scripts/win_build.bat b/fluent-bit/lib/cfl/scripts/win_build.bat
new file mode 100755
index 00000000..a30ab170
--- /dev/null
+++ b/fluent-bit/lib/cfl/scripts/win_build.bat
@@ -0,0 +1,7 @@
+setlocal
+call "C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\Common7\Tools\VsDevCmd.bat"
+path "C:\Program Files (x86)\MSBuild\16.0\Bin;C:\Program Files (x86)\Microsoft Visual Studio\2019\Enterprise\MSBuild\Current\Bin";%path%
+cd build
+cmake -G "NMake Makefiles"  -DCFL_TESTS=On ..\
+cmake --build .
+endlocal
diff --git a/fluent-bit/lib/cfl/src/CMakeLists.txt b/fluent-bit/lib/cfl/src/CMakeLists.txt
new file mode 100644
index 00000000..53f2a620
--- /dev/null
+++ b/fluent-bit/lib/cfl/src/CMakeLists.txt
@@ -0,0 +1,29 @@
+set(src
+  cfl.c
+  cfl_log.c
+  cfl_sds.c
+  cfl_time.c
+  cfl_kv.c
+  cfl_kvlist.c
+  cfl_array.c
+  cfl_variant.c
+  cfl_checksum.c
+  )
+
+# Static Library
+add_library(cfl-static STATIC ${src})
+target_link_libraries(cfl-static xxhash)
+
+# Install Library
+if(MSVC)
+  # Rename the output for Windows environment to avoid naming issues
+  set_target_properties(cfl-static PROPERTIES OUTPUT_NAME libcfl)
+else()
+  set_target_properties(cfl-static PROPERTIES OUTPUT_NAME cfl)
+endif(MSVC)
+
+install(TARGETS cfl-static
+  RUNTIME DESTINATION ${CFL_INSTALL_BINDIR}
+  LIBRARY DESTINATION ${CFL_INSTALL_LIBDIR}
+  ARCHIVE DESTINATION ${CFL_INSTALL_LIBDIR}
+  COMPONENT library)
diff --git a/fluent-bit/lib/cfl/src/cfl.c b/fluent-bit/lib/cfl/src/cfl.c
new file mode 100644
index 00000000..7afb1109
--- /dev/null
+++ b/fluent-bit/lib/cfl/src/cfl.c
@@ -0,0 +1,23 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+int cfl_init()
+{
+    return 0;
+}
diff --git a/fluent-bit/lib/cfl/src/cfl_array.c b/fluent-bit/lib/cfl/src/cfl_array.c
new file mode 100644
index 00000000..e64fbf52
--- /dev/null
+++ b/fluent-bit/lib/cfl/src/cfl_array.c
@@ -0,0 +1,376 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <cfl/cfl.h>
+#include <cfl/cfl_array.h>
+#include <cfl/cfl_variant.h>
+
+struct cfl_array *cfl_array_create(size_t slot_count)
+{
+    struct cfl_array *array;
+
+    array = malloc(sizeof(struct cfl_array));
+    if (array == NULL) {
+        cfl_errno();
+        return NULL;
+    }
+
+    /* by default arrays are not resizable */
+    array->resizable = CFL_FALSE;
+
+    /* allocate fixed number of entries */
+    array->entries = calloc(slot_count, sizeof(void *));
+    if (array->entries == NULL) {
+        cfl_errno();
+        free(array);
+        return NULL;
+    }
+
+    array->entry_count = 0;
+    array->slot_count = slot_count;
+
+    return array;
+}
+
+void cfl_array_destroy(struct cfl_array *array)
+{
+    size_t index;
+
+    if (!array) {
+        return;
+    }
+
+    if (array->entries != NULL) {
+        for (index = 0 ; index < array->entry_count ; index++) {
+            if(array->entries[index] != NULL) {
+                cfl_variant_destroy(array->entries[index]);
+            }
+        }
+
+        free(array->entries);
+    }
+    free(array);
+}
+
+int cfl_array_resizable(struct cfl_array *array, int v)
+{
+    if (v != CFL_TRUE && v != CFL_FALSE) {
+        return -1;
+    }
+
+    array->resizable = v;
+    return 0;
+}
+
+int cfl_array_remove_by_index(struct cfl_array *array,
+                              size_t position)
+{
+    if (position >= array->entry_count) {
+        return -1;
+    }
+
+    cfl_variant_destroy(array->entries[position]);
+
+    if (position != array->entry_count - 1) {
+        memcpy(&array->entries[position],
+               &array->entries[position + 1],
+               sizeof(void *) * (array->entry_count - (position + 1)));
+    }
+    else {
+        array->entries[position] = NULL;
+    }
+    array->entry_count--;
+
+    return 0;
+}
+
+int cfl_array_remove_by_reference(struct cfl_array *array,
+                                  struct cfl_variant *value)
+{
+    size_t index;
+
+    for (index = 0 ; index < array->entry_count ; index++) {
+        if (array->entries[index] == value) {
+            return cfl_array_remove_by_index(array, index);
+        }
+    }
+
+    return 0;
+}
+
+int cfl_array_append(struct cfl_array *array,
+                     struct cfl_variant *value)
+{
+    void *tmp;
+    size_t new_slot_count;
+    size_t new_size;
+
+    if (array->entry_count >= array->slot_count) {
+        /*
+         * if there is no more space but the caller allowed to resize
+         * the array, just double the size. Yeah, this is scary and should
+         * be used only when the caller 'knows this is safe to do' because
+         * it controls the input data.
+         */
+        if (array->resizable) {
+            /* set new number of slots and total size */
+            new_slot_count = (array->slot_count * 2);
+            new_size = (new_slot_count * sizeof(void *));
+
+            tmp = realloc(array->entries, new_size);
+            if (!tmp) {
+                cfl_report_runtime_error();
+                return -1;
+            }
+            array->slot_count = new_slot_count;
+            array->entries = tmp;
+        }
+        else {
+            return -1;
+        }
+    }
+    array->entries[array->entry_count++] = value;
+
+    return 0;
+}
+
+int cfl_array_append_string(struct cfl_array *array,
+                            char *value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_string(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_array_append(array, value_instance);
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_array_append_bytes(struct cfl_array *array,
+                           char *value,
+                           size_t length)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_bytes(value, length);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_array_append(array, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_array_append_reference(struct cfl_array *array, void *value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_reference(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_array_append(array, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_array_append_bool(struct cfl_array *array, int value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_bool(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_array_append(array, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_array_append_int64(struct cfl_array *array, int64_t value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_int64(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_array_append(array, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+        return -2;
+    }
+
+    return 0;
+}
+
+
+int cfl_array_append_double(struct cfl_array *array, double value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_double(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_array_append(array, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+
+int cfl_array_append_array(struct cfl_array *array, struct cfl_array *value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_array(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_array_append(array, value_instance);
+    if (result) {
+        cfl_variant_destroy(value_instance);
+        return -2;
+    }
+
+    return 0;
+}
+
+
+int cfl_array_append_new_array(struct cfl_array *array, size_t size)
+{
+    int               result;
+    struct cfl_array *value;
+
+    value = cfl_array_create(size);
+
+    if (value == NULL) {
+        return -1;
+    }
+
+    result = cfl_array_append_array(array, value);
+
+    if (result) {
+        cfl_array_destroy(value);
+    }
+
+    return result;
+}
+
+int cfl_array_append_kvlist(struct cfl_array *array, struct 
+cfl_kvlist *value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_kvlist(value);
+    if (value_instance == NULL) {
+        return -1;
+    }
+    result = cfl_array_append(array, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+
+int cfl_array_print(FILE *fp, struct cfl_array *array)
+{
+    size_t size;
+    size_t i;
+    int ret;
+
+    if (fp == NULL || array == NULL) {
+        return -1;
+    }
+
+    size = array->entry_count;
+    if (size == 0) {
+        fputs("[]", fp);
+        return 0;
+    }
+
+    fputs("[", fp);
+    for (i=0; i<size-1; i++) {
+        ret = cfl_variant_print(fp, array->entries[i]);
+        fputs(",", fp);
+    }
+    ret = cfl_variant_print(fp, array->entries[size-1]);
+    fputs("]", fp);
+
+    return ret;
+}
diff --git a/fluent-bit/lib/cfl/src/cfl_checksum.c b/fluent-bit/lib/cfl/src/cfl_checksum.c
new file mode 100644
index 00000000..9b0297ea
--- /dev/null
+++ b/fluent-bit/lib/cfl/src/cfl_checksum.c
@@ -0,0 +1,111 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <stdint.h>
+#include <stdio.h>
+
+static const uint32_t crc32c_table[256] = {
+    0x00000000, 0xF26B8303, 0xE13B70F7, 0x1350F3F4,
+    0xC79A971F, 0x35F1141C, 0x26A1E7E8, 0xD4CA64EB,
+    0x8AD958CF, 0x78B2DBCC, 0x6BE22838, 0x9989AB3B,
+    0x4D43CFD0, 0xBF284CD3, 0xAC78BF27, 0x5E133C24,
+    0x105EC76F, 0xE235446C, 0xF165B798, 0x030E349B,
+    0xD7C45070, 0x25AFD373, 0x36FF2087, 0xC494A384,
+    0x9A879FA0, 0x68EC1CA3, 0x7BBCEF57, 0x89D76C54,
+    0x5D1D08BF, 0xAF768BBC, 0xBC267848, 0x4E4DFB4B,
+    0x20BD8EDE, 0xD2D60DDD, 0xC186FE29, 0x33ED7D2A,
+    0xE72719C1, 0x154C9AC2, 0x061C6936, 0xF477EA35,
+    0xAA64D611, 0x580F5512, 0x4B5FA6E6, 0xB93425E5,
+    0x6DFE410E, 0x9F95C20D, 0x8CC531F9, 0x7EAEB2FA,
+    0x30E349B1, 0xC288CAB2, 0xD1D83946, 0x23B3BA45,
+    0xF779DEAE, 0x05125DAD, 0x1642AE59, 0xE4292D5A,
+    0xBA3A117E, 0x4851927D, 0x5B016189, 0xA96AE28A,
+    0x7DA08661, 0x8FCB0562, 0x9C9BF696, 0x6EF07595,
+    0x417B1DBC, 0xB3109EBF, 0xA0406D4B, 0x522BEE48,
+    0x86E18AA3, 0x748A09A0, 0x67DAFA54, 0x95B17957,
+    0xCBA24573, 0x39C9C670, 0x2A993584, 0xD8F2B687,
+    0x0C38D26C, 0xFE53516F, 0xED03A29B, 0x1F682198,
+    0x5125DAD3, 0xA34E59D0, 0xB01EAA24, 0x42752927,
+    0x96BF4DCC, 0x64D4CECF, 0x77843D3B, 0x85EFBE38,
+    0xDBFC821C, 0x2997011F, 0x3AC7F2EB, 0xC8AC71E8,
+    0x1C661503, 0xEE0D9600, 0xFD5D65F4, 0x0F36E6F7,
+    0x61C69362, 0x93AD1061, 0x80FDE395, 0x72966096,
+    0xA65C047D, 0x5437877E, 0x4767748A, 0xB50CF789,
+    0xEB1FCBAD, 0x197448AE, 0x0A24BB5A, 0xF84F3859,
+    0x2C855CB2, 0xDEEEDFB1, 0xCDBE2C45, 0x3FD5AF46,
+    0x7198540D, 0x83F3D70E, 0x90A324FA, 0x62C8A7F9,
+    0xB602C312, 0x44694011, 0x5739B3E5, 0xA55230E6,
+    0xFB410CC2, 0x092A8FC1, 0x1A7A7C35, 0xE811FF36,
+    0x3CDB9BDD, 0xCEB018DE, 0xDDE0EB2A, 0x2F8B6829,
+    0x82F63B78, 0x709DB87B, 0x63CD4B8F, 0x91A6C88C,
+    0x456CAC67, 0xB7072F64, 0xA457DC90, 0x563C5F93,
+    0x082F63B7, 0xFA44E0B4, 0xE9141340, 0x1B7F9043,
+    0xCFB5F4A8, 0x3DDE77AB, 0x2E8E845F, 0xDCE5075C,
+    0x92A8FC17, 0x60C37F14, 0x73938CE0, 0x81F80FE3,
+    0x55326B08, 0xA759E80B, 0xB4091BFF, 0x466298FC,
+    0x1871A4D8, 0xEA1A27DB, 0xF94AD42F, 0x0B21572C,
+    0xDFEB33C7, 0x2D80B0C4, 0x3ED04330, 0xCCBBC033,
+    0xA24BB5A6, 0x502036A5, 0x4370C551, 0xB11B4652,
+    0x65D122B9, 0x97BAA1BA, 0x84EA524E, 0x7681D14D,
+    0x2892ED69, 0xDAF96E6A, 0xC9A99D9E, 0x3BC21E9D,
+    0xEF087A76, 0x1D63F975, 0x0E330A81, 0xFC588982,
+    0xB21572C9, 0x407EF1CA, 0x532E023E, 0xA145813D,
+    0x758FE5D6, 0x87E466D5, 0x94B49521, 0x66DF1622,
+    0x38CC2A06, 0xCAA7A905, 0xD9F75AF1, 0x2B9CD9F2,
+    0xFF56BD19, 0x0D3D3E1A, 0x1E6DCDEE, 0xEC064EED,
+    0xC38D26C4, 0x31E6A5C7, 0x22B65633, 0xD0DDD530,
+    0x0417B1DB, 0xF67C32D8, 0xE52CC12C, 0x1747422F,
+    0x49547E0B, 0xBB3FFD08, 0xA86F0EFC, 0x5A048DFF,
+    0x8ECEE914, 0x7CA56A17, 0x6FF599E3, 0x9D9E1AE0,
+    0xD3D3E1AB, 0x21B862A8, 0x32E8915C, 0xC083125F,
+    0x144976B4, 0xE622F5B7, 0xF5720643, 0x07198540,
+    0x590AB964, 0xAB613A67, 0xB831C993, 0x4A5A4A90,
+    0x9E902E7B, 0x6CFBAD78, 0x7FAB5E8C, 0x8DC0DD8F,
+    0xE330A81A, 0x115B2B19, 0x020BD8ED, 0xF0605BEE,
+    0x24AA3F05, 0xD6C1BC06, 0xC5914FF2, 0x37FACCF1,
+    0x69E9F0D5, 0x9B8273D6, 0x88D28022, 0x7AB90321,
+    0xAE7367CA, 0x5C18E4C9, 0x4F48173D, 0xBD23943E,
+    0xF36E6F75, 0x0105EC76, 0x12551F82, 0xE03E9C81,
+    0x34F4F86A, 0xC69F7B69, 0xD5CF889D, 0x27A40B9E,
+    0x79B737BA, 0x8BDCB4B9, 0x988C474D, 0x6AE7C44E,
+    0xBE2DA0A5, 0x4C4623A6, 0x5F16D052, 0xAD7D5351
+};
+
+uint32_t cfl_checksum_crc32c(unsigned char *buffer, size_t length)
+{
+    uint32_t checksum_table_index;
+    uint32_t checksum;
+    size_t   index;
+
+    checksum = 0xFFFFFFFF;
+
+    /* Keeping in mind that compilers are smart enough to optimize these
+     * expensive divisions / modulo with bitwise operations it makes no sense
+     * to make this code cryptic.
+     */
+
+    for (index = 0 ; index < length ; index++) {
+        checksum_table_index  = checksum ^ buffer[index];
+        checksum_table_index %= 256;
+
+        checksum = crc32c_table[checksum_table_index] ^ (checksum / 256);
+    }
+
+    return checksum ^ 0xFFFFFFFF;
+}
diff --git a/fluent-bit/lib/cfl/src/cfl_kv.c b/fluent-bit/lib/cfl/src/cfl_kv.c
new file mode 100644
index 00000000..23d3c4d9
--- /dev/null
+++ b/fluent-bit/lib/cfl/src/cfl_kv.c
@@ -0,0 +1,147 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <cfl/cfl.h>
+#include <cfl/cfl_kv.h>
+#include <cfl/cfl_log.h>
+
+
+void cfl_kv_init(struct cfl_list *list)
+{
+    cfl_list_init(list);
+}
+
+struct cfl_kv *cfl_kv_item_create_len(struct cfl_list *list,
+                                      char *k_buf, size_t k_len,
+                                      char *v_buf, size_t v_len)
+{
+    struct cfl_kv *kv;
+
+    kv = calloc(1, sizeof(struct cfl_kv));
+
+    if (kv == NULL) {
+        cfl_report_runtime_error();
+
+        return NULL;
+    }
+
+    kv->key = cfl_sds_create_len(k_buf, k_len);
+
+    if (kv->key == NULL) {
+        free(kv);
+
+        return NULL;
+    }
+
+    if (v_len > 0) {
+        kv->val = cfl_sds_create_len(v_buf, v_len);
+
+        if (kv->val == NULL) {
+            cfl_sds_destroy(kv->key);
+            free(kv);
+
+            return NULL;
+        }
+    }
+
+    cfl_list_add(&kv->_head, list);
+
+    return kv;
+}
+
+struct cfl_kv *cfl_kv_item_create(struct cfl_list *list,
+                                  char *k_buf, char *v_buf)
+{
+    int k_len;
+    int v_len;
+
+    if (k_buf == NULL) {
+        return NULL;
+    }
+
+    k_len = strlen(k_buf);
+
+    if (v_buf != NULL) {
+        v_len = strlen(v_buf);
+    }
+    else {
+        v_len = 0;
+    }
+
+    return cfl_kv_item_create_len(list, k_buf, k_len, v_buf, v_len);
+}
+
+void cfl_kv_item_destroy(struct cfl_kv *kv)
+{
+    if (kv->key != NULL) {
+        cfl_sds_destroy(kv->key);
+    }
+
+    if (kv->val != NULL) {
+        cfl_sds_destroy(kv->val);
+    }
+
+    cfl_list_del(&kv->_head);
+
+    free(kv);
+}
+
+void cfl_kv_release(struct cfl_list *list)
+{
+    struct cfl_list *head;
+    struct cfl_list *tmp;
+    struct cfl_kv   *kv;
+
+    cfl_list_foreach_safe(head, tmp, list) {
+        kv = cfl_list_entry(head, struct cfl_kv, _head);
+
+        cfl_kv_item_destroy(kv);
+    }
+}
+
+const char *cfl_kv_get_key_value(const char *key, struct cfl_list *list)
+{
+    struct cfl_list *head;
+    int              len;
+    struct cfl_kv   *kv;
+
+    if (key == NULL) {
+        return NULL;
+    }
+
+    len = strlen(key);
+
+    if (len == 0) {
+        return NULL;
+    }
+
+    cfl_list_foreach(head, list) {
+        kv = cfl_list_entry(head, struct cfl_kv, _head);
+
+        if (cfl_sds_len(kv->key) != len) {
+            continue;
+        }
+
+        if (strncasecmp(kv->key, key, len) == 0) {
+            return kv->val;
+        }
+    }
+
+    return NULL;
+}
diff --git a/fluent-bit/lib/cfl/src/cfl_kvlist.c b/fluent-bit/lib/cfl/src/cfl_kvlist.c
new file mode 100644
index 00000000..56139b87
--- /dev/null
+++ b/fluent-bit/lib/cfl/src/cfl_kvlist.c
@@ -0,0 +1,390 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <cfl/cfl.h>
+#include <cfl/cfl_kvlist.h>
+#include <cfl/cfl_array.h>
+#include <cfl/cfl_variant.h>
+
+struct cfl_kvlist *cfl_kvlist_create()
+{
+    struct cfl_kvlist *list;
+
+    list = malloc(sizeof(struct cfl_kvlist));
+    if (list == NULL) {
+        cfl_report_runtime_error();
+        return NULL;
+    }
+
+    cfl_list_init(&list->list);
+    return list;
+}
+
+void cfl_kvlist_destroy(struct cfl_kvlist *list)
+{
+    struct cfl_list *tmp;
+    struct cfl_list *head;
+    struct cfl_kvpair *pair;
+
+    cfl_list_foreach_safe(head, tmp, &list->list) {
+        pair = cfl_list_entry(head, struct cfl_kvpair, _head);
+
+        if (pair->key) {
+            cfl_sds_destroy(pair->key);
+        }
+
+        if (pair->val) {
+            cfl_variant_destroy(pair->val);
+        }
+        cfl_list_del(&pair->_head);
+        free(pair);
+    }
+
+    free(list);
+}
+
+/*
+int cfl_kvlist_insert(struct cfl_kvlist *list,
+                      char *key, void *value,
+                      size_t value_length,
+                      int value_type)
+{
+    struct cfl_kvpair *pair;
+
+    pair = malloc(sizeof(struct cfl_kvpair));
+
+    if (pair == NULL) {
+        cfl_errno();
+
+        return -1;
+    }
+
+    pair->key = cfl_sds_create(key);
+
+    if (pair->key == NULL) {
+        free(pair);
+
+        return -2;
+    }
+
+    pair->val = cfl_variant_create(value, value_length, value_type);
+
+    if (pair->val == NULL) {
+        cfl_sds_destroy(pair->key);
+        free(pair);
+
+        return -3;
+    }
+
+    cfl_list_add(&pair->_head, &list->list);
+
+    return 0;
+}
+*/
+
+int cfl_kvlist_insert_string(struct cfl_kvlist *list,
+                             char *key, char *value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_string(value);
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_kvlist_insert(list, key, value_instance);
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_kvlist_insert_bytes(struct cfl_kvlist *list,
+                             char *key, char *value,
+                             size_t length)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_bytes(value, length);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_kvlist_insert(list, key, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_kvlist_insert_reference(struct cfl_kvlist *list,
+                                char *key, void *value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_reference(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_kvlist_insert(list, key, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_kvlist_insert_bool(struct cfl_kvlist *list,
+                           char *key, int value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_bool(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_kvlist_insert(list, key, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_kvlist_insert_int64(struct cfl_kvlist *list,
+                            char *key, int64_t value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_int64(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_kvlist_insert(list, key, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_kvlist_insert_double(struct cfl_kvlist *list,
+                             char *key, double value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_double(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_kvlist_insert(list, key, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_kvlist_insert_array(struct cfl_kvlist *list,
+                            char *key, struct cfl_array *value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_array(value);
+
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_kvlist_insert(list, key, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_kvlist_insert_new_array(struct cfl_kvlist *list,
+                                char *key, size_t size)
+{
+    int               result;
+    struct cfl_array *value;
+
+    value = cfl_array_create(size);
+
+    if (value == NULL) {
+        return -1;
+    }
+
+    result = cfl_kvlist_insert_array(list, key, value);
+
+    if (result) {
+        cfl_array_destroy(value);
+    }
+
+    return result;
+}
+
+int cfl_kvlist_insert_kvlist(struct cfl_kvlist *list,
+                             char *key, struct cfl_kvlist *value)
+{
+    struct cfl_variant *value_instance;
+    int                 result;
+
+    value_instance = cfl_variant_create_from_kvlist(value);
+    if (value_instance == NULL) {
+        return -1;
+    }
+
+    result = cfl_kvlist_insert(list, key, value_instance);
+
+    if (result) {
+        cfl_variant_destroy(value_instance);
+
+        return -2;
+    }
+
+    return 0;
+}
+
+int cfl_kvlist_insert(struct cfl_kvlist *list,
+                      char *key, struct cfl_variant *value)
+{
+    struct cfl_kvpair *pair;
+
+    pair = malloc(sizeof(struct cfl_kvpair));
+    if (pair == NULL) {
+        cfl_report_runtime_error();
+        return -1;
+    }
+
+    pair->key = cfl_sds_create(key);
+
+    if (pair->key == NULL) {
+        free(pair);
+
+        return -2;
+    }
+
+    pair->val = value;
+
+    cfl_list_add(&pair->_head, &list->list);
+    return 0;
+}
+
+int cfl_kvlist_count(struct cfl_kvlist *list)
+{
+    return cfl_list_size(&list->list);
+}
+
+struct cfl_variant *cfl_kvlist_fetch(struct cfl_kvlist *list, char *key)
+{
+    int len;
+    struct cfl_list *head;
+    struct cfl_kvpair *pair;
+
+    len = strlen(key);
+
+    cfl_list_foreach(head, &list->list) {
+        pair = cfl_list_entry(head, struct cfl_kvpair, _head);
+
+        if (cfl_sds_len(pair->key) != len) {
+            continue;
+        }
+
+        if (strcmp(pair->key, key) == 0) {
+            return pair->val;
+        }
+    }
+
+    return NULL;
+}
+
+int cfl_kvlist_print(FILE *fp, struct cfl_kvlist *list)
+{
+    size_t size;
+    size_t i;
+    int ret = -1;
+
+    struct cfl_list *head = NULL;
+    struct cfl_kvpair *pair = NULL;
+
+    if (fp == NULL || list == NULL) {
+        return -1;
+    }
+
+    size = (size_t)cfl_kvlist_count(list);
+    i = 0;
+    fputs("{", fp);
+    cfl_list_foreach(head, &list->list) {
+        pair = cfl_list_entry(head, struct cfl_kvpair, _head);
+        if (pair == NULL || pair->key == NULL || pair->val == NULL) {
+            continue;
+        }
+
+        fprintf(fp, "\"%s\":", pair->key);
+        ret = cfl_variant_print(fp, pair->val);
+
+        i++;
+        if (i != size) {
+            fputs(",", fp);
+        }
+    }
+    fputs("}", fp);
+
+    return ret;
+}
diff --git a/fluent-bit/lib/cfl/src/cfl_log.c b/fluent-bit/lib/cfl/src/cfl_log.c
new file mode 100644
index 00000000..5918a71c
--- /dev/null
+++ b/fluent-bit/lib/cfl/src/cfl_log.c
@@ -0,0 +1,40 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <cfl/cfl_log.h>
+
+#ifdef _WIN32
+    #define strerror_r(errnum, buf, buf_size) strerror_s(buf, buf_size, errnum)
+#endif
+
+int cfl_report_runtime_error_impl(int errnum, char *file, int line)
+{
+    char buf[256];
+
+    strerror_r(errnum, buf, sizeof(buf) - 1);
+
+    fprintf(stderr, "[%s:%i errno=%i] %s\n",
+            file, line, errnum, buf);
+
+    return 0;
+}
diff --git a/fluent-bit/lib/cfl/src/cfl_sds.c b/fluent-bit/lib/cfl/src/cfl_sds.c
new file mode 100644
index 00000000..eff66fe3
--- /dev/null
+++ b/fluent-bit/lib/cfl/src/cfl_sds.c
@@ -0,0 +1,236 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+/*
+ * This interface is a minimized version of Fluent Bit SDS just for easily
+ * string storage
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdarg.h>
+#include <inttypes.h>
+
+#include <cfl/cfl_sds.h>
+
+size_t cfl_sds_avail(cfl_sds_t s)
+{
+    struct cfl_sds *h;
+
+    h = CFL_SDS_HEADER(s);
+    return (size_t) (h->alloc - h->len);
+}
+
+cfl_sds_t sds_alloc(size_t size)
+{
+    void *buf;
+    cfl_sds_t s;
+    struct cfl_sds *head;
+
+    buf = malloc(CFL_SDS_HEADER_SIZE + size + 1);
+    if (!buf) {
+        return NULL;
+    }
+
+    head = buf;
+    head->len = 0;
+    head->alloc = size;
+
+    s = head->buf;
+    *s = '\0';
+
+    return s;
+}
+
+size_t cfl_sds_alloc(cfl_sds_t s)
+{
+    return (size_t) CFL_SDS_HEADER(s)->alloc;
+}
+
+cfl_sds_t cfl_sds_increase(cfl_sds_t s, size_t len)
+{
+    size_t new_size;
+    struct cfl_sds *head;
+    cfl_sds_t out;
+    void *tmp;
+
+    out = s;
+    new_size = (CFL_SDS_HEADER_SIZE + cfl_sds_alloc(s) + len + 1);
+    head = CFL_SDS_HEADER(s);
+    tmp = realloc(head, new_size);
+    if (!tmp) {
+        return NULL;
+    }
+    head = (struct cfl_sds *) tmp;
+    head->alloc += len;
+    out = head->buf;
+
+    return out;
+}
+
+size_t cfl_sds_len(cfl_sds_t s)
+{
+    return (size_t) CFL_SDS_HEADER(s)->len;
+}
+
+cfl_sds_t cfl_sds_create_len(const char *str, int len)
+{
+    cfl_sds_t s;
+    struct cfl_sds *head;
+
+    s = sds_alloc(len);
+    if (!s) {
+        return NULL;
+    }
+
+    if (str) {
+        memcpy(s, str, len);
+        s[len] = '\0';
+
+        head = CFL_SDS_HEADER(s);
+        head->len = len;
+    }
+    return s;
+}
+
+cfl_sds_t cfl_sds_create(const char *str)
+{
+    size_t len;
+
+    if (!str) {
+        len = 0;
+    }
+    else {
+        len = strlen(str);
+    }
+
+    return cfl_sds_create_len(str, len);
+}
+
+void cfl_sds_destroy(cfl_sds_t s)
+{
+    struct cfl_sds *head;
+
+    if (!s) {
+        return;
+    }
+
+    head = CFL_SDS_HEADER(s);
+    free(head);
+}
+
+cfl_sds_t cfl_sds_cat(cfl_sds_t s, const char *str, int len)
+{
+    size_t avail;
+    struct cfl_sds *head;
+    cfl_sds_t tmp = NULL;
+
+    avail = cfl_sds_avail(s);
+    if (avail < len) {
+        tmp = cfl_sds_increase(s, len);
+        if (!tmp) {
+            return NULL;
+        }
+        s = tmp;
+    }
+    memcpy((char *) (s + cfl_sds_len(s)), str, len);
+
+    head = CFL_SDS_HEADER(s);
+    head->len += len;
+    s[head->len] = '\0';
+
+    return s;
+}
+
+cfl_sds_t cfl_sds_create_size(size_t size)
+{
+    return sds_alloc(size);
+}
+
+void cfl_sds_set_len(cfl_sds_t s, size_t len)
+{
+    struct cfl_sds *head;
+
+    head = CFL_SDS_HEADER(s);
+    head->len = len;
+}
+
+void cfl_sds_cat_safe(cfl_sds_t *buf, const char *str, int len)
+{
+    cfl_sds_t tmp;
+
+    tmp = cfl_sds_cat(*buf, str, len);
+    if (!tmp) {
+        return;
+    }
+    *buf = tmp;
+}
+
+cfl_sds_t cfl_sds_printf(cfl_sds_t *sds, const char *fmt, ...)
+{
+    va_list ap;
+    int len = strlen(fmt)*2;
+    int size;
+    cfl_sds_t tmp = NULL;
+    cfl_sds_t s;
+    struct cfl_sds *head;
+
+    if (len < 64) len = 64;
+
+    s = *sds;
+    if (cfl_sds_avail(s)< len) {
+        tmp = cfl_sds_increase(s, len);
+        if (!tmp) {
+            return NULL;
+        }
+        *sds = s = tmp;
+    }
+
+    va_start(ap, fmt);
+    size = vsnprintf((char *) (s + cfl_sds_len(s)), cfl_sds_avail(s), fmt, ap);
+    if (size < 0) {
+        va_end(ap);
+        return NULL;
+    }
+    va_end(ap);
+
+    if (size >= cfl_sds_avail(s)) {
+        tmp = cfl_sds_increase(s, size - cfl_sds_avail(s) + 1);
+        if (!tmp) {
+            return NULL;
+        }
+        *sds = s = tmp;
+
+        va_start(ap, fmt);
+        size = vsnprintf((char *) (s + cfl_sds_len(s)), cfl_sds_avail(s), fmt, ap);
+        if (size > cfl_sds_avail(s)) {
+            va_end(ap);
+            return NULL;
+        }
+        va_end(ap);
+    }
+
+    head = CFL_SDS_HEADER(s);
+    head->len += size;
+    s[head->len] = '\0';
+
+    return s;
+}
+
diff --git a/fluent-bit/lib/cfl/src/cfl_time.c b/fluent-bit/lib/cfl/src/cfl_time.c
new file mode 100644
index 00000000..262cf744
--- /dev/null
+++ b/fluent-bit/lib/cfl/src/cfl_time.c
@@ -0,0 +1,62 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <cfl/cfl.h>
+
+/* MacOS */
+#ifdef CFL_HAVE_CLOCK_GET_TIME
+#include <mach/clock.h>
+#include <mach/mach.h>
+#endif
+
+#include <inttypes.h>
+#include <time.h>
+
+uint64_t cfl_time_now()
+{
+    struct timespec tm = {0};
+
+    /*
+     * macOS note: newer versions of macOS provides a timespec_get() function, but for
+     * some reason it does not provide full nanoseconds resolution, last 3 digits are
+     * always zeros.
+     *
+     * if CFL_HAVE_CLOCK_GET_TIME and CFL_HAVE_TIMESPEC_GET are defined, let's
+     * prioritize CFL_HAVE_CLOCK_GET_TIME.
+     */
+
+#if defined CFL_HAVE_CLOCK_GET_TIME
+    /* MacOS */
+    clock_serv_t cclock;
+    mach_timespec_t mts;
+    host_get_clock_service(mach_host_self(), CALENDAR_CLOCK, &cclock);
+    clock_get_time(cclock, &mts);
+    tm.tv_sec = mts.tv_sec;
+    tm.tv_nsec = mts.tv_nsec;
+    mach_port_deallocate(mach_task_self(), cclock);
+#elif defined CFL_HAVE_TIMESPEC_GET
+    /* C11 supported */
+    timespec_get(&tm, TIME_UTC);
+
+#else /* __STDC_VERSION__ */
+    clock_gettime(CLOCK_REALTIME, &tm);
+#endif
+
+    return (((uint64_t) tm.tv_sec * 1000000000L) + tm.tv_nsec);
+}
diff --git a/fluent-bit/lib/cfl/src/cfl_variant.c b/fluent-bit/lib/cfl/src/cfl_variant.c
new file mode 100644
index 00000000..6181066e
--- /dev/null
+++ b/fluent-bit/lib/cfl/src/cfl_variant.c
@@ -0,0 +1,234 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <cfl/cfl.h>
+#include <cfl/cfl_variant.h>
+#include <cfl/cfl_array.h>
+#include <cfl/cfl_kvlist.h>
+
+#if defined(__MINGW32__) || defined(__MINGW64__)
+#define HEXDUMPFORMAT "%#x"
+#else
+#define HEXDUMPFORMAT "%p"
+#endif
+
+int cfl_variant_print(FILE *fp, struct cfl_variant *val)
+{
+    int ret = -1;
+    size_t size;
+    size_t i;
+
+    if (fp == NULL || val == NULL) {
+        return -1;
+    }
+
+    switch (val->type) {
+    case CFL_VARIANT_STRING:
+        ret = fprintf(fp, "\"%s\"", val->data.as_string);
+        break;
+    case CFL_VARIANT_BOOL:
+        if (val->data.as_bool) {
+            ret = fputs("true",fp);
+        }
+        else {
+            ret = fputs("false", fp);
+        }
+        break;
+    case CFL_VARIANT_INT:
+        ret = fprintf(fp, "%" PRId64, val->data.as_int64);
+        break;
+    case CFL_VARIANT_DOUBLE:
+        ret = fprintf(fp, "%lf", val->data.as_double);
+        break;
+    case CFL_VARIANT_BYTES:
+        size = cfl_sds_len(val->data.as_bytes);
+        for (i=0; i<size; i++) {
+            ret = fprintf(fp, "%02x", (unsigned char)val->data.as_bytes[i]);
+        }
+        break;
+
+    case CFL_VARIANT_REFERENCE:
+        ret = fprintf(fp, HEXDUMPFORMAT, val->data.as_reference);
+        break;
+    case CFL_VARIANT_ARRAY:
+        ret = cfl_array_print(fp, val->data.as_array);
+        break;
+
+    case CFL_VARIANT_KVLIST:
+        ret = cfl_kvlist_print(fp, val->data.as_kvlist);
+        break;
+
+    default:
+        ret = fputs("!Unknown Type", fp);
+    }
+    return ret;
+}
+
+struct cfl_variant *cfl_variant_create_from_string(char *value)
+{
+    struct cfl_variant *instance;
+
+    instance = cfl_variant_create();
+
+    if (instance != NULL) {
+        instance->data.as_string = cfl_sds_create(value);
+        if (instance->data.as_string == NULL) {
+            free(instance);
+            instance = NULL;
+        }
+        else {
+            instance->type = CFL_VARIANT_STRING;
+        }
+    }
+
+    return instance;
+}
+
+struct cfl_variant *cfl_variant_create_from_bytes(char *value, size_t length)
+{
+    struct cfl_variant *instance;
+
+    instance = cfl_variant_create();
+
+    if (instance != NULL) {
+        instance->data.as_bytes = cfl_sds_create_len(value, length);
+        if (instance->data.as_bytes == NULL) {
+            free(instance);
+            instance = NULL;
+        }
+        else {
+            instance->type = CFL_VARIANT_BYTES;
+        }
+    }
+
+    return instance;
+}
+
+struct cfl_variant *cfl_variant_create_from_bool(int value)
+{
+    struct cfl_variant *instance;
+
+    instance = cfl_variant_create();
+    if (instance != NULL) {
+        instance->data.as_bool = value;
+        instance->type = CFL_VARIANT_BOOL;
+    }
+
+    return instance;
+}
+
+struct cfl_variant *cfl_variant_create_from_int64(int64_t value)
+{
+    struct cfl_variant *instance;
+
+    instance = cfl_variant_create();
+    if (instance != NULL) {
+        instance->data.as_int64 = value;
+        instance->type = CFL_VARIANT_INT;
+    }
+
+    return instance;
+}
+
+struct cfl_variant *cfl_variant_create_from_double(double value)
+{
+    struct cfl_variant *instance;
+
+    instance = cfl_variant_create();
+    if (instance != NULL) {
+        instance->data.as_double = value;
+        instance->type = CFL_VARIANT_DOUBLE;
+    }
+
+    return instance;
+}
+
+struct cfl_variant *cfl_variant_create_from_array(struct cfl_array *value)
+{
+    struct cfl_variant *instance;
+
+    instance = cfl_variant_create();
+    if (instance != NULL) {
+        instance->data.as_array = value;
+        instance->type = CFL_VARIANT_ARRAY;
+    }
+
+    return instance;
+}
+
+struct cfl_variant *cfl_variant_create_from_kvlist(struct cfl_kvlist *value)
+{
+    struct cfl_variant *instance;
+
+    instance = cfl_variant_create();
+    if (instance != NULL) {
+        instance->data.as_kvlist = value;
+        instance->type = CFL_VARIANT_KVLIST;
+    }
+
+    return instance;
+}
+
+struct cfl_variant *cfl_variant_create_from_reference(void *value)
+{
+    struct cfl_variant *instance;
+
+    instance = cfl_variant_create();
+    if (instance != NULL) {
+        instance->data.as_reference = value;
+        instance->type = CFL_VARIANT_REFERENCE;
+    }
+
+    return instance;
+}
+
+struct cfl_variant *cfl_variant_create()
+{
+    struct cfl_variant *instance;
+
+    instance = calloc(1, sizeof(struct cfl_variant));
+    if (instance == NULL) {
+        cfl_errno();
+        return NULL;
+    }
+
+    return instance;
+}
+
+void cfl_variant_destroy(struct cfl_variant *instance)
+{
+    if (!instance) {
+        return;
+    }
+
+    if (instance->type == CFL_VARIANT_STRING ||
+        instance->type == CFL_VARIANT_BYTES) {
+        if (instance->data.as_string != NULL) {
+            cfl_sds_destroy(instance->data.as_string);
+        }
+    }
+    else if (instance->type == CFL_VARIANT_ARRAY) {
+        cfl_array_destroy(instance->data.as_array);
+    }
+    else if (instance->type == CFL_VARIANT_KVLIST) {
+        cfl_kvlist_destroy(instance->data.as_kvlist);
+    }
+
+    free(instance);
+}
diff --git a/fluent-bit/lib/cfl/tests/CMakeLists.txt b/fluent-bit/lib/cfl/tests/CMakeLists.txt
new file mode 100644
index 00000000..0fac328c
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/CMakeLists.txt
@@ -0,0 +1,31 @@
+include_directories(lib/acutest)
+
+set(UNIT_TESTS_FILES
+  kv.c
+  sds.c
+  hash.c
+  list.c
+  variant.c
+  )
+
+configure_file(
+  "${CMAKE_CURRENT_SOURCE_DIR}/cfl_tests_internal.h.in"
+  "${CMAKE_CURRENT_SOURCE_DIR}/cfl_tests_internal.h"
+  )
+
+# Prepare list of unit tests
+foreach(source_file ${UNIT_TESTS_FILES})
+  get_filename_component(source_file_we ${source_file} NAME_WE)
+  set(source_file_we cfl-test-${source_file_we})
+  add_executable(
+    ${source_file_we}
+    ${source_file}
+    )
+  target_link_libraries(${source_file_we} cfl-static)
+
+  if (CFL_SANITIZE_ADDRESS)
+    add_sanitizers(${source_file_we})
+  endif()
+
+  add_test(${source_file_we} ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/${source_file_we})
+endforeach()
diff --git a/fluent-bit/lib/cfl/tests/cfl_tests_internal.h.in b/fluent-bit/lib/cfl/tests/cfl_tests_internal.h.in
new file mode 100644
index 00000000..b14ef012
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/cfl_tests_internal.h.in
@@ -0,0 +1,25 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#ifndef CFL_TEST_INTERNAL_H
+#define CFL_TEST_INTERNAL_H
+
+#include "lib/acutest/include/acutest.h"
+
+#endif
diff --git a/fluent-bit/lib/cfl/tests/hash.c b/fluent-bit/lib/cfl/tests/hash.c
new file mode 100644
index 00000000..afc7311f
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/hash.c
@@ -0,0 +1,57 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <cfl/cfl_hash.h>
+
+#include "cfl_tests_internal.h"
+
+static void checks()
+{
+    int len;
+    uint64_t cfl_hash;
+    uint64_t xxh_hash;
+    XXH3_state_t xxh_state;
+    cfl_hash_state_t cfl_state;
+    char *str = "this is a test";
+
+    len = strlen(str);
+
+    /* a dummy test, we just need to make sure the wrapper is working */
+    cfl_hash = cfl_hash_64bits(str, len);
+    xxh_hash = XXH3_64bits(str, len);
+    TEST_CHECK(cfl_hash == xxh_hash);
+
+    /* state and updates */
+    cfl_hash_64bits_reset(&cfl_state);
+    cfl_hash_64bits_update(&cfl_state, str, len);
+    cfl_hash_64bits_update(&cfl_state, str, len);
+    cfl_hash = cfl_hash_64bits_digest(&cfl_state);
+
+    XXH3_64bits_reset(&xxh_state);
+    XXH3_64bits_update(&xxh_state, str, len);
+    XXH3_64bits_update(&xxh_state, str, len);
+    xxh_hash = XXH3_64bits_digest(&xxh_state);
+
+    TEST_CHECK(cfl_hash == xxh_hash);
+}
+
+TEST_LIST = {
+    {"checks",  checks},
+    { 0 }
+};
diff --git a/fluent-bit/lib/cfl/tests/kv.c b/fluent-bit/lib/cfl/tests/kv.c
new file mode 100644
index 00000000..f49ade9d
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/kv.c
@@ -0,0 +1,82 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <cfl/cfl.h>
+
+#include "cfl_tests_internal.h"
+
+
+static void regular_operation()
+{
+    struct cfl_list *entry_iterator;
+    size_t           entry_index;
+    struct cfl_list  entry_list;
+    struct cfl_kv   *entry;
+
+    cfl_kv_init(&entry_list);
+
+    entry = cfl_kv_item_create(&entry_list, "first entry", "dummy value");
+    TEST_CHECK(entry != NULL);
+
+    entry = cfl_kv_item_create(&entry_list, "second entry", NULL);
+    TEST_CHECK(entry != NULL);
+
+    entry = cfl_kv_item_create(&entry_list, NULL, NULL);
+    TEST_CHECK(entry == NULL);
+
+    entry = cfl_kv_item_create_len(&entry_list, "third entry", 0, "dummy value", 12);
+    TEST_CHECK(entry != NULL);
+
+    entry = cfl_kv_item_create_len(&entry_list, "fourth entry", 13, "dummy value", 0);
+    TEST_CHECK(entry != NULL);
+
+    entry_index = 0;
+
+    cfl_list_foreach(entry_iterator, &entry_list) {
+        entry = cfl_list_entry(entry_iterator, struct cfl_kv, _head);
+
+        switch (entry_index) {
+            case 0:
+                TEST_CHECK(strcmp(entry->key, "first entry") == 0);
+                TEST_CHECK(strcmp(entry->val, "dummy value") == 0);
+                break;
+            case 1:
+                TEST_CHECK(strcmp(entry->key, "second entry") == 0);
+                TEST_CHECK(entry->val == NULL);
+                break;
+            case 2:
+                TEST_CHECK(strlen(entry->key) == 0);
+                TEST_CHECK(strcmp(entry->val, "dummy value") == 0);
+                break;
+            case 3:
+                TEST_CHECK(strcmp(entry->key, "fourth entry") == 0);
+                TEST_CHECK(entry->val == NULL);
+                break;
+        }
+
+        entry_index++;
+    }
+
+    cfl_kv_release(&entry_list);
+}
+
+TEST_LIST = {
+    {"regular_operation",  regular_operation},
+    { 0 }
+};
diff --git a/fluent-bit/lib/cfl/tests/lib/acutest/.travis.yml b/fluent-bit/lib/cfl/tests/lib/acutest/.travis.yml
new file mode 100644
index 00000000..92082198
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/lib/acutest/.travis.yml
@@ -0,0 +1,34 @@
+# YAML definition for travis-ci.com continuous integration.
+# See https://docs.travis-ci.com/user/languages/c
+
+# Container-based infrastructure (Linux)
+# * https://docs.travis-ci.com/user/migrating-from-legacy/#How-can-I-use-container-based-infrastructure%3F
+sudo:
+- false
+
+language: c++
+
+compiler:
+    - gcc
+    - clang
+
+env:
+    - CFLAGS="-std=c99 -pedantic" CONFIG=Debug          # C99
+    - CFLAGS="-std=c99 -pedantic" CONFIG=Release
+    - CFLAGS="-std=c11 -pedantic" CONFIG=Debug          # C11
+    - CFLAGS="-std=c11 -pedantic" CONFIG=Release
+    - CXXFLAGS="-std=c++11 -pedantic" CONFIG=Debug      # C++11
+    - CXXFLAGS="-std=c++11 -pedantic" CONFIG=Release
+    # Too old compilers on travis-ci.org for these:
+    #- CXXFLAGS="-std=c++14 -pedantic" CONFIG=Debug      # C++14
+    #- CXXFLAGS="-std=c++14 -pedantic" CONFIG=Release
+    #- CXXFLAGS="-std=c++17 -pedantic" CONFIG=Debug      # C++17
+    #- CXXFLAGS="-std=c++17 -pedantic" CONFIG=Release
+
+before_script:
+    - mkdir build
+    - cd build
+    - cmake -DCMAKE_BUILD_TYPE=$CONFIG -G 'Unix Makefiles' ..
+
+script:
+    - make VERBOSE=1
diff --git a/fluent-bit/lib/cfl/tests/lib/acutest/CMakeLists.txt b/fluent-bit/lib/cfl/tests/lib/acutest/CMakeLists.txt
new file mode 100644
index 00000000..4a741a42
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/lib/acutest/CMakeLists.txt
@@ -0,0 +1,44 @@
+
+cmake_minimum_required(VERSION 2.8)
+project(Acutest C CXX)
+
+
+set(CMAKE_CONFIGURATION_TYPES Debug Release RelWithDebInfo MinSizeRel)
+if("${CMAKE_BUILD_TYPE}" STREQUAL "")
+    set(CMAKE_BUILD_TYPE $ENV{CMAKE_BUILD_TYPE})
+
+    if("${CMAKE_BUILD_TYPE}" STREQUAL "")
+        set(CMAKE_BUILD_TYPE "Release")
+    endif()
+endif()
+
+
+set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
+
+
+if ("${CMAKE_C_COMPILER_ID}" STREQUAL "Clang")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wall -Wextra")
+elseif ("${CMAKE_C_COMPILER_ID}" STREQUAL "GNU")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wall -Wextra")
+elseif ("${CMAKE_C_COMPILER_ID}" STREQUAL "MSVC")
+    # Specify proper C runtime library:
+    set(CMAKE_C_FLAGS_DEBUG "${CMAKE_C_FLAGS_DEBUG} /MTd")
+    set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} /MT")
+    set(CMAKE_C_FLAGS_RELWITHDEBINFO "${CMAKE_C_FLAGS_RELEASE} /MT")
+    set(CMAKE_C_FLAGS_MINSIZEREL "${CMAKE_C_FLAGS_RELEASE} /MT")
+endif()
+
+if ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Wextra")
+elseif ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "GNU")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Wextra")
+elseif ("${CMAKE_CXX_COMPILER_ID}" STREQUAL "MSVC")
+    # Specify proper C runtime library:
+    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /MTd")
+    set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} /MT")
+    set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELEASE} /MT")
+    set(CMAKE_CXX_FLAGS_MINSIZEREL "${CMAKE_CXX_FLAGS_RELEASE} /MT")
+endif()
+
+
+add_subdirectory(examples)
diff --git a/fluent-bit/lib/cfl/tests/lib/acutest/GIT b/fluent-bit/lib/cfl/tests/lib/acutest/GIT
new file mode 100644
index 00000000..379ff148
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/lib/acutest/GIT
@@ -0,0 +1 @@
+03885e23ea27535251b91abf681257d3d4241eae (git master)
\ No newline at end of file
diff --git a/fluent-bit/lib/cfl/tests/lib/acutest/LICENSE.md b/fluent-bit/lib/cfl/tests/lib/acutest/LICENSE.md
new file mode 100644
index 00000000..97309201
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/lib/acutest/LICENSE.md
@@ -0,0 +1,22 @@
+
+# The MIT License (MIT)
+
+Copyright © 2013-2017 Martin Mitáš
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the “Software”),
+to deal in the Software without restriction, including without limitation
+the rights to use, copy, modify, merge, publish, distribute, sublicense,
+and/or sell copies of the Software, and to permit persons to whom the
+Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included
+in all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS
+OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+IN THE SOFTWARE.
diff --git a/fluent-bit/lib/cfl/tests/lib/acutest/README.md b/fluent-bit/lib/cfl/tests/lib/acutest/README.md
new file mode 100644
index 00000000..e7f75b65
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/lib/acutest/README.md
@@ -0,0 +1,4 @@
+Taken from https://github.com/mity/acutest
+
+MIT License
+
diff --git a/fluent-bit/lib/cfl/tests/lib/acutest/appveyor.yml b/fluent-bit/lib/cfl/tests/lib/acutest/appveyor.yml
new file mode 100644
index 00000000..6e85c7a0
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/lib/acutest/appveyor.yml
@@ -0,0 +1,26 @@
+# YAML definition for Appveyor.com continuous integration.
+# See http://www.appveyor.com/docs/appveyor-yml
+
+version: '{branch}-{build}'
+
+before_build:
+  - 'cmake --version'
+  - 'if "%PLATFORM%"=="x64" cmake -G "Visual Studio 12 Win64"'
+  - 'if not "%PLATFORM%"=="x64" cmake -G "Visual Studio 12"'
+
+build:
+  project: Acutest.sln
+  verbosity: minimal
+
+skip_tags: true
+
+os:
+  - Windows Server 2012 R2
+
+configuration:
+  - Debug
+  - Release
+
+platform:
+  - x64    # 64-bit build
+  - win32  # 32-bit build
diff --git a/fluent-bit/lib/cfl/tests/lib/acutest/include/acutest.h b/fluent-bit/lib/cfl/tests/lib/acutest/include/acutest.h
new file mode 100644
index 00000000..cd5ffb2c
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/lib/acutest/include/acutest.h
@@ -0,0 +1,864 @@
+/*
+ * Acutest -- Another C/C++ Unit Test facility
+ * <http://github.com/mity/acutest>
+ *
+ * Copyright (c) 2013-2017 Martin Mitas
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#ifndef ACUTEST_H__
+#define ACUTEST_H__
+
+
+/************************
+ *** Public interface ***
+ ************************/
+
+/* By default, "acutest.h" provides the main program entry point (function
+ * main()). However, if the test suite is composed of multiple source files
+ * which include "acutest.h", then this causes a problem of multiple main()
+ * definitions. To avoid this problem, #define macro TEST_NO_MAIN in all
+ * compilation units but one.
+ */
+
+/* Macro to specify list of unit tests in the suite.
+ * The unit test implementation MUST provide list of unit tests it implements
+ * with this macro:
+ *
+ *   TEST_LIST = {
+ *       { "test1_name", test1_func_ptr },
+ *       { "test2_name", test2_func_ptr },
+ *       ...
+ *       { 0 }
+ *   };
+ *
+ * The list specifies names of each test (must be unique) and pointer to
+ * a function implementing it. The function does not take any arguments
+ * and has no return values, i.e. every test function has to be compatible
+ * with this prototype:
+ *
+ *   void test_func(void);
+ */
+#define TEST_LIST              const struct test__ test_list__[]
+
+
+/* Macros for testing whether an unit test succeeds or fails. These macros
+ * can be used arbitrarily in functions implementing the unit tests.
+ *
+ * If any condition fails throughout execution of a test, the test fails.
+ *
+ * TEST_CHECK takes only one argument (the condition), TEST_CHECK_ allows
+ * also to specify an error message to print out if the condition fails.
+ * (It expects printf-like format string and its parameters). The macros
+ * return non-zero (condition passes) or 0 (condition fails).
+ *
+ * That can be useful when more conditions should be checked only if some
+ * preceding condition passes, as illustrated in this code snippet:
+ *
+ *   SomeStruct* ptr = allocate_some_struct();
+ *   if(TEST_CHECK(ptr != NULL)) {
+ *       TEST_CHECK(ptr->member1 < 100);
+ *       TEST_CHECK(ptr->member2 > 200);
+ *   }
+ */
+#define TEST_CHECK_(cond,...)  test_check__((cond), __FILE__, __LINE__, __VA_ARGS__)
+#define TEST_CHECK(cond)       test_check__((cond), __FILE__, __LINE__, "%s", #cond)
+
+
+/* printf-like macro for outputting an extra information about a failure.
+ *
+ * Note it does not output anything if there was not (yet) failed condition
+ * in the current test. Intended use is to output some computed output
+ * versus the expected value, e.g. like this:
+ *
+ *   if(!TEST_CHECK(produced == expected)) {
+ *       TEST_MSG("Expected: %d", expected);
+ *       TEST_MSG("Produced: %d", produced);
+ *   }
+ *
+ * The macro can deal with multi-line output fairly well. It also automatically
+ * adds a final new-line if there is none present.
+ */
+#define TEST_MSG(...)          test_message__(__VA_ARGS__)
+
+/* Maximal output per TEST_MSG call. Longer messages are cut.
+ * You may define another limit prior including "acutest.h"
+ */
+#ifndef TEST_MSG_MAXSIZE
+    #define TEST_MSG_MAXSIZE   1024
+#endif
+
+
+/**********************
+ *** Implementation ***
+ **********************/
+
+/* The unit test files should not rely on anything below. */
+
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#if defined(unix) || defined(__unix__) || defined(__unix) || defined(__APPLE__)
+    #define ACUTEST_UNIX__      1
+    #include <errno.h>
+    #include <unistd.h>
+    #include <sys/types.h>
+    #include <sys/wait.h>
+    #include <signal.h>
+#endif
+
+#if defined(__gnu_linux__)
+    #define ACUTEST_LINUX__     1
+    #include <fcntl.h>
+    #include <sys/stat.h>
+#endif
+
+#if defined(_WIN32) || defined(__WIN32__) || defined(__WINDOWS__)
+    #define ACUTEST_WIN__       1
+    #include <windows.h>
+    #include <io.h>
+#endif
+
+#ifdef __cplusplus
+    #include <exception>
+#endif
+
+
+/* Note our global private identifiers end with '__' to mitigate risk of clash
+ * with the unit tests implementation. */
+
+
+#ifdef __cplusplus
+    extern "C" {
+#endif
+
+
+struct test__ {
+    const char* name;
+    void (*func)(void);
+};
+
+extern const struct test__ test_list__[];
+
+int test_check__(int cond, const char* file, int line, const char* fmt, ...);
+void test_message__(const char* fmt, ...);
+
+
+#ifndef TEST_NO_MAIN
+
+static char* test_argv0__ = NULL;
+static size_t test_list_size__ = 0;
+static const struct test__** tests__ = NULL;
+static char* test_flags__ = NULL;
+static size_t test_count__ = 0;
+static int test_no_exec__ = -1;
+static int test_no_summary__ = 0;
+static int test_skip_mode__ = 0;
+
+static int test_stat_failed_units__ = 0;
+static int test_stat_run_units__ = 0;
+
+static const struct test__* test_current_unit__ = NULL;
+static int test_current_already_logged__ = 0;
+static int test_verbose_level__ = 2;
+static int test_current_failures__ = 0;
+static int test_colorize__ = 0;
+
+#define TEST_COLOR_DEFAULT__            0
+#define TEST_COLOR_GREEN__              1
+#define TEST_COLOR_RED__                2
+#define TEST_COLOR_DEFAULT_INTENSIVE__  3
+#define TEST_COLOR_GREEN_INTENSIVE__    4
+#define TEST_COLOR_RED_INTENSIVE__      5
+
+static int
+test_print_in_color__(int color, const char* fmt, ...)
+{
+    va_list args;
+    char buffer[256];
+    int n;
+
+    va_start(args, fmt);
+    vsnprintf(buffer, sizeof(buffer), fmt, args);
+    va_end(args);
+    buffer[sizeof(buffer)-1] = '\0';
+
+    if(!test_colorize__) {
+        return printf("%s", buffer);
+    }
+
+#if defined ACUTEST_UNIX__
+    {
+        const char* col_str;
+        switch(color) {
+            case TEST_COLOR_GREEN__:             col_str = "\033[0;32m"; break;
+            case TEST_COLOR_RED__:               col_str = "\033[0;31m"; break;
+            case TEST_COLOR_GREEN_INTENSIVE__:   col_str = "\033[1;32m"; break;
+            case TEST_COLOR_RED_INTENSIVE__:     col_str = "\033[1;31m"; break;
+            case TEST_COLOR_DEFAULT_INTENSIVE__: col_str = "\033[1m"; break;
+            default:                                col_str = "\033[0m"; break;
+        }
+        printf("%s", col_str);
+        n = printf("%s", buffer);
+        printf("\033[0m");
+        return n;
+    }
+#elif defined ACUTEST_WIN__
+    {
+        HANDLE h;
+        CONSOLE_SCREEN_BUFFER_INFO info;
+        WORD attr;
+
+        h = GetStdHandle(STD_OUTPUT_HANDLE);
+        GetConsoleScreenBufferInfo(h, &info);
+
+        switch(color) {
+            case TEST_COLOR_GREEN__:             attr = FOREGROUND_GREEN; break;
+            case TEST_COLOR_RED__:               attr = FOREGROUND_RED; break;
+            case TEST_COLOR_GREEN_INTENSIVE__:   attr = FOREGROUND_GREEN | FOREGROUND_INTENSITY; break;
+            case TEST_COLOR_RED_INTENSIVE__:     attr = FOREGROUND_RED | FOREGROUND_INTENSITY; break;
+            case TEST_COLOR_DEFAULT_INTENSIVE__: attr = FOREGROUND_BLUE | FOREGROUND_GREEN | FOREGROUND_RED | FOREGROUND_INTENSITY; break;
+            default:                                attr = 0; break;
+        }
+        if(attr != 0)
+            SetConsoleTextAttribute(h, attr);
+        n = printf("%s", buffer);
+        SetConsoleTextAttribute(h, info.wAttributes);
+        return n;
+    }
+#else
+    n = printf("%s", buffer);
+    return n;
+#endif
+}
+
+int
+test_check__(int cond, const char* file, int line, const char* fmt, ...)
+{
+    const char *result_str;
+    int result_color;
+    int verbose_level;
+
+    if(cond) {
+        result_str = "ok";
+        result_color = TEST_COLOR_GREEN__;
+        verbose_level = 3;
+    } else {
+        if(!test_current_already_logged__  &&  test_current_unit__ != NULL) {
+            printf("[ ");
+            test_print_in_color__(TEST_COLOR_RED_INTENSIVE__, "FAILED");
+            printf(" ]\n");
+        }
+        result_str = "failed";
+        result_color = TEST_COLOR_RED__;
+        verbose_level = 2;
+        test_current_failures__++;
+        test_current_already_logged__++;
+    }
+
+    if(test_verbose_level__ >= verbose_level) {
+        va_list args;
+
+        printf("  ");
+
+        if(file != NULL) {
+            if(test_verbose_level__ < 3) {
+#ifdef ACUTEST_WIN__
+                const char* lastsep1 = strrchr(file, '\\');
+                const char* lastsep2 = strrchr(file, '/');
+                if(lastsep1 == NULL)
+                    lastsep1 = file-1;
+                if(lastsep2 == NULL)
+                    lastsep2 = file-1;
+                file = (lastsep1 > lastsep2 ? lastsep1 : lastsep2) + 1;
+#else
+                const char* lastsep = strrchr(file, '/');
+                if(lastsep != NULL)
+                    file = lastsep+1;
+#endif
+            }
+            printf("%s:%d: Check ", file, line);
+        }
+
+        va_start(args, fmt);
+        vprintf(fmt, args);
+        va_end(args);
+
+        printf("... ");
+        test_print_in_color__(result_color, result_str);
+        printf("\n");
+        test_current_already_logged__++;
+    }
+
+    return (cond != 0);
+}
+
+void
+test_message__(const char* fmt, ...)
+{
+    char buffer[TEST_MSG_MAXSIZE];
+    char* line_beg;
+    char* line_end;
+    va_list args;
+
+    if(test_verbose_level__ < 2)
+        return;
+
+    /* We allow extra message only when something is already wrong in the
+     * current test. */
+    if(!test_current_already_logged__  ||  test_current_unit__ == NULL)
+        return;
+
+    va_start(args, fmt);
+    vsnprintf(buffer, TEST_MSG_MAXSIZE, fmt, args);
+    va_end(args);
+    buffer[TEST_MSG_MAXSIZE-1] = '\0';
+
+    line_beg = buffer;
+    while(1) {
+        line_end = strchr(line_beg, '\n');
+        if(line_end == NULL)
+            break;
+        printf("    %.*s\n", (int)(line_end - line_beg), line_beg);
+        line_beg = line_end + 1;
+    }
+    if(line_beg[0] != '\0')
+        printf("    %s\n", line_beg);
+}
+
+static void
+test_list_names__(void)
+{
+    const struct test__* test;
+
+    printf("Unit tests:\n");
+    for(test = &test_list__[0]; test->func != NULL; test++)
+        printf("  %s\n", test->name);
+}
+
+static void
+test_remember__(int i)
+{
+    if(test_flags__[i])
+        return;
+    else
+        test_flags__[i] = 1;
+
+    tests__[test_count__] = &test_list__[i];
+    test_count__++;
+}
+
+static int
+test_name_contains_word__(const char* name, const char* pattern)
+{
+    static const char word_delim[] = " \t-_.";
+    const char* substr;
+    size_t pattern_len;
+    int starts_on_word_boundary;
+    int ends_on_word_boundary;
+
+    pattern_len = strlen(pattern);
+
+    substr = strstr(name, pattern);
+    while(substr != NULL) {
+        starts_on_word_boundary = (substr == name || strchr(word_delim, substr[-1]) != NULL);
+        ends_on_word_boundary = (substr[pattern_len] == '\0' || strchr(word_delim, substr[pattern_len]) != NULL);
+
+        if(starts_on_word_boundary && ends_on_word_boundary)
+            return 1;
+
+        substr = strstr(substr+1, pattern);
+    }
+
+    return 0;
+}
+
+static int
+test_lookup__(const char* pattern)
+{
+    int i;
+    int n = 0;
+
+    /* Try exact match. */
+    for(i = 0; i < (int) test_list_size__; i++) {
+        if(strcmp(test_list__[i].name, pattern) == 0) {
+            test_remember__(i);
+            n++;
+            break;
+        }
+    }
+    if(n > 0)
+        return n;
+
+    /* Try word match. */
+    for(i = 0; i < (int) test_list_size__; i++) {
+        if(test_name_contains_word__(test_list__[i].name, pattern)) {
+            test_remember__(i);
+            n++;
+        }
+    }
+    if(n > 0)
+        return n;
+
+    /* Try relaxed match. */
+    for(i = 0; i < (int) test_list_size__; i++) {
+        if(strstr(test_list__[i].name, pattern) != NULL) {
+            test_remember__(i);
+            n++;
+        }
+    }
+
+    return n;
+}
+
+/* Call directly the given test unit function. */
+static int
+test_do_run__(const struct test__* test)
+{
+    test_current_unit__ = test;
+    test_current_failures__ = 0;
+    test_current_already_logged__ = 0;
+
+    if(test_verbose_level__ >= 3) {
+        test_print_in_color__(TEST_COLOR_DEFAULT_INTENSIVE__, "Test %s:\n", test->name);
+        test_current_already_logged__++;
+    } else if(test_verbose_level__ >= 1) {
+        int n;
+        char spaces[48];
+
+        n = test_print_in_color__(TEST_COLOR_DEFAULT_INTENSIVE__, "Test %s... ", test->name);
+        memset(spaces, ' ', sizeof(spaces));
+        if(n < (int) sizeof(spaces))
+            printf("%.*s", (int) sizeof(spaces) - n, spaces);
+    } else {
+        test_current_already_logged__ = 1;
+    }
+
+#ifdef __cplusplus
+    try {
+#endif
+
+        /* This is good to do for case the test unit e.g. crashes. */
+        fflush(stdout);
+        fflush(stderr);
+
+        test->func();
+
+#ifdef __cplusplus
+    } catch(std::exception& e) {
+        const char* what = e.what();
+        if(what != NULL)
+            test_check__(0, NULL, 0, "Threw std::exception: %s", what);
+        else
+            test_check__(0, NULL, 0, "Threw std::exception");
+    } catch(...) {
+        test_check__(0, NULL, 0, "Threw an exception");
+    }
+#endif
+
+    if(test_verbose_level__ >= 3) {
+        switch(test_current_failures__) {
+            case 0:  test_print_in_color__(TEST_COLOR_GREEN_INTENSIVE__, "  All conditions have passed.\n\n"); break;
+            case 1:  test_print_in_color__(TEST_COLOR_RED_INTENSIVE__, "  One condition has FAILED.\n\n"); break;
+            default: test_print_in_color__(TEST_COLOR_RED_INTENSIVE__, "  %d conditions have FAILED.\n\n", test_current_failures__); break;
+        }
+    } else if(test_verbose_level__ >= 1 && test_current_failures__ == 0) {
+        printf("[   ");
+        test_print_in_color__(TEST_COLOR_GREEN_INTENSIVE__, "OK");
+        printf("   ]\n");
+    }
+
+    test_current_unit__ = NULL;
+    return (test_current_failures__ == 0) ? 0 : -1;
+}
+
+#if defined(ACUTEST_UNIX__) || defined(ACUTEST_WIN__)
+/* Called if anything goes bad in Acutest, or if the unit test ends in other
+ * way then by normal returning from its function (e.g. exception or some
+ * abnormal child process termination). */
+static void
+test_error__(const char* fmt, ...)
+{
+    va_list args;
+
+    if(test_verbose_level__ == 0)
+        return;
+
+    if(test_verbose_level__ <= 2  &&  !test_current_already_logged__  &&  test_current_unit__ != NULL) {
+        printf("[ ");
+        test_print_in_color__(TEST_COLOR_RED_INTENSIVE__, "FAILED");
+        printf(" ]\n");
+    }
+
+    if(test_verbose_level__ >= 2) {
+        test_print_in_color__(TEST_COLOR_RED_INTENSIVE__, "  Error: ");
+        va_start(args, fmt);
+        vprintf(fmt, args);
+        va_end(args);
+        printf("\n");
+    }
+}
+#endif
+
+/* Trigger the unit test. If possible (and not suppressed) it starts a child
+ * process who calls test_do_run__(), otherwise it calls test_do_run__()
+ * directly. */
+static void
+test_run__(const struct test__* test)
+{
+    int failed = 1;
+
+    test_current_unit__ = test;
+    test_current_already_logged__ = 0;
+
+    if(!test_no_exec__) {
+
+#if defined(ACUTEST_UNIX__)
+
+        pid_t pid;
+        int exit_code;
+
+        pid = fork();
+        if(pid == (pid_t)-1) {
+            test_error__("Cannot fork. %s [%d]", strerror(errno), errno);
+            failed = 1;
+        } else if(pid == 0) {
+            /* Child: Do the test. */
+            failed = (test_do_run__(test) != 0);
+            exit(failed ? 1 : 0);
+        } else {
+            /* Parent: Wait until child terminates and analyze its exit code. */
+            waitpid(pid, &exit_code, 0);
+            if(WIFEXITED(exit_code)) {
+                switch(WEXITSTATUS(exit_code)) {
+                    case 0:   failed = 0; break;   /* test has passed. */
+                    case 1:   /* noop */ break;    /* "normal" failure. */
+                    default:  test_error__("Unexpected exit code [%d]", WEXITSTATUS(exit_code));
+                }
+            } else if(WIFSIGNALED(exit_code)) {
+                char tmp[32];
+                const char* signame;
+                switch(WTERMSIG(exit_code)) {
+                    case SIGINT:  signame = "SIGINT"; break;
+                    case SIGHUP:  signame = "SIGHUP"; break;
+                    case SIGQUIT: signame = "SIGQUIT"; break;
+                    case SIGABRT: signame = "SIGABRT"; break;
+                    case SIGKILL: signame = "SIGKILL"; break;
+                    case SIGSEGV: signame = "SIGSEGV"; break;
+                    case SIGILL:  signame = "SIGILL"; break;
+                    case SIGTERM: signame = "SIGTERM"; break;
+                    default:      sprintf(tmp, "signal %d", WTERMSIG(exit_code)); signame = tmp; break;
+                }
+                test_error__("Test interrupted by %s", signame);
+            } else {
+                test_error__("Test ended in an unexpected way [%d]", exit_code);
+            }
+        }
+
+#elif defined(ACUTEST_WIN__)
+
+        char buffer[512] = {0};
+        STARTUPINFOA startupInfo;
+        PROCESS_INFORMATION processInfo;
+        DWORD exitCode;
+
+        /* Windows has no fork(). So we propagate all info into the child
+         * through a command line arguments. */
+        _snprintf(buffer, sizeof(buffer)-1,
+                 "%s --no-exec --no-summary --verbose=%d --color=%s -- \"%s\"",
+                 test_argv0__, test_verbose_level__,
+                 test_colorize__ ? "always" : "never", test->name);
+        memset(&startupInfo, 0, sizeof(startupInfo));
+        startupInfo.cb = sizeof(STARTUPINFO);
+        if(CreateProcessA(NULL, buffer, NULL, NULL, FALSE, 0, NULL, NULL, &startupInfo, &processInfo)) {
+            WaitForSingleObject(processInfo.hProcess, INFINITE);
+            GetExitCodeProcess(processInfo.hProcess, &exitCode);
+            CloseHandle(processInfo.hThread);
+            CloseHandle(processInfo.hProcess);
+            failed = (exitCode != 0);
+        } else {
+            test_error__("Cannot create unit test subprocess [%ld].", GetLastError());
+            failed = 1;
+        }
+
+#else
+
+        /* A platform where we don't know how to run child process. */
+        failed = (test_do_run__(test) != 0);
+
+#endif
+
+    } else {
+        /* Child processes suppressed through --no-exec. */
+        failed = (test_do_run__(test) != 0);
+    }
+
+    test_current_unit__ = NULL;
+
+    test_stat_run_units__++;
+    if(failed)
+        test_stat_failed_units__++;
+}
+
+#if defined(ACUTEST_WIN__)
+/* Callback for SEH events. */
+static LONG CALLBACK
+test_exception_filter__(EXCEPTION_POINTERS *ptrs)
+{
+    test_error__("Unhandled SEH exception %08lx at %p.",
+                 ptrs->ExceptionRecord->ExceptionCode,
+                 ptrs->ExceptionRecord->ExceptionAddress);
+    fflush(stdout);
+    fflush(stderr);
+    return EXCEPTION_EXECUTE_HANDLER;
+}
+#endif
+
+
+static void
+test_help__(void)
+{
+    printf("Usage: %s [options] [test...]\n", test_argv0__);
+    printf("Run the specified unit tests; or if the option '--skip' is used, run all\n");
+    printf("tests in the suite but those listed.  By default, if no tests are specified\n");
+    printf("on the command line, all unit tests in the suite are run.\n");
+    printf("\n");
+    printf("Options:\n");
+    printf("  -s, --skip            Execute all unit tests but the listed ones\n");
+    printf("      --exec=WHEN       If supported, execute unit tests as child processes\n");
+    printf("                          (WHEN is one of 'auto', 'always', 'never')\n");
+    printf("  -E, --no-exec         Same as --exec=never\n");
+    printf("      --no-summary      Suppress printing of test results summary\n");
+    printf("  -l, --list            List unit tests in the suite and exit\n");
+    printf("  -v, --verbose         Enable more verbose output\n");
+    printf("      --verbose=LEVEL   Set verbose level to LEVEL:\n");
+    printf("                          0 ... Be silent\n");
+    printf("                          1 ... Output one line per test (and summary)\n");
+    printf("                          2 ... As 1 and failed conditions (this is default)\n");
+    printf("                          3 ... As 1 and all conditions (and extended summary)\n");
+    printf("      --color=WHEN      Enable colorized output\n");
+    printf("                          (WHEN is one of 'auto', 'always', 'never')\n");
+    printf("  -h, --help            Display this help and exit\n");
+
+    if(test_list_size__ < 16) {
+        printf("\n");
+        test_list_names__();
+    }
+}
+
+#ifdef ACUTEST_LINUX__
+static int
+test_is_tracer_present__(void)
+{
+    char buf[256+32+1];
+    int tracer_present = 0;
+    int fd;
+    ssize_t n_read;
+
+    fd = open("/proc/self/status", O_RDONLY);
+    if(fd == -1)
+        return 0;
+
+    n_read = read(fd, buf, sizeof(buf)-1);
+    while(n_read > 0) {
+        static const char pattern[] = "TracerPid:";
+        const char* field;
+
+        buf[n_read] = '\0';
+        field = strstr(buf, pattern);
+        if(field != NULL  &&  field < buf + sizeof(buf) - 32) {
+            pid_t tracer_pid = (pid_t) atoi(field + sizeof(pattern) - 1);
+            tracer_present = (tracer_pid != 0);
+            break;
+        }
+
+        if(n_read == sizeof(buf)-1) {
+            memmove(buf, buf + sizeof(buf)-1 - 32, 32);
+            n_read = read(fd, buf+32, sizeof(buf)-1-32);
+            if(n_read > 0)
+                n_read += 32;
+        }
+    }
+
+    close(fd);
+    return tracer_present;
+}
+#endif
+
+int
+main(int argc, char** argv)
+{
+    int i;
+    int seen_double_dash = 0;
+
+    test_argv0__ = argv[0];
+
+#if defined ACUTEST_UNIX__
+    test_colorize__ = isatty(STDOUT_FILENO);
+#elif defined ACUTEST_WIN__
+ #if defined __BORLANDC__
+    test_colorize__ = isatty(_fileno(stdout));
+ #else
+    test_colorize__ = _isatty(_fileno(stdout));
+ #endif
+#else
+    test_colorize__ = 0;
+#endif
+
+    /* Count all test units */
+    test_list_size__ = 0;
+    for(i = 0; test_list__[i].func != NULL; i++)
+        test_list_size__++;
+
+    tests__ = (const struct test__**) malloc(sizeof(const struct test__*) * test_list_size__);
+    test_flags__ = (char*) malloc(sizeof(char) * test_list_size__);
+    if(tests__ == NULL || test_flags__ == NULL) {
+        fprintf(stderr, "Out of memory.\n");
+        exit(2);
+    }
+    memset((void*) test_flags__, 0, sizeof(char) * test_list_size__);
+
+    /* Parse options */
+    for(i = 1; i < argc; i++) {
+        if(seen_double_dash || argv[i][0] != '-') {
+            if(test_lookup__(argv[i]) == 0) {
+                fprintf(stderr, "%s: Unrecognized unit test '%s'\n", argv[0], argv[i]);
+                fprintf(stderr, "Try '%s --list' for list of unit tests.\n", argv[0]);
+                exit(2);
+            }
+        } else if(strcmp(argv[i], "--") == 0) {
+            seen_double_dash = 1;
+        } else if(strcmp(argv[i], "--help") == 0 || strcmp(argv[i], "-h") == 0) {
+            test_help__();
+            exit(0);
+        } else if(strcmp(argv[i], "--verbose") == 0 || strcmp(argv[i], "-v") == 0) {
+            test_verbose_level__++;
+        } else if(strncmp(argv[i], "--verbose=", 10) == 0) {
+            test_verbose_level__ = atoi(argv[i] + 10);
+        } else if(strcmp(argv[i], "--color=auto") == 0) {
+            /* noop (set from above) */
+        } else if(strcmp(argv[i], "--color=always") == 0 || strcmp(argv[i], "--color") == 0) {
+            test_colorize__ = 1;
+        } else if(strcmp(argv[i], "--color=never") == 0 || strcmp(argv[i], "--no-color") == 0) {
+            test_colorize__ = 0;
+        } else if(strcmp(argv[i], "--skip") == 0 || strcmp(argv[i], "-s") == 0) {
+            test_skip_mode__ = 1;
+        } else if(strcmp(argv[i], "--exec=auto") == 0) {
+            /* noop (set from above) */
+        } else if(strcmp(argv[i], "--exec=always") == 0 || strcmp(argv[i], "--exec") == 0) {
+            test_no_exec__ = 0;
+        } else if(strcmp(argv[i], "--exec=never") == 0 || strcmp(argv[i], "--no-exec") == 0 || strcmp(argv[i], "-E") == 0) {
+            test_no_exec__ = 1;
+        } else if(strcmp(argv[i], "--no-summary") == 0) {
+            test_no_summary__ = 1;
+        } else if(strcmp(argv[i], "--list") == 0 || strcmp(argv[i], "-l") == 0) {
+            test_list_names__();
+            exit(0);
+        } else {
+            fprintf(stderr, "%s: Unrecognized option '%s'\n", argv[0], argv[i]);
+            fprintf(stderr, "Try '%s --help' for more information.\n", argv[0]);
+            exit(2);
+        }
+    }
+
+#if defined(ACUTEST_WIN__)
+    SetUnhandledExceptionFilter(test_exception_filter__);
+#endif
+
+    /* By default, we want to run all tests. */
+    if(test_count__ == 0) {
+        for(i = 0; test_list__[i].func != NULL; i++)
+            tests__[i] = &test_list__[i];
+        test_count__ = test_list_size__;
+    }
+
+    /* Guess whether we want to run unit tests as child processes. */
+    if(test_no_exec__ < 0) {
+        test_no_exec__ = 0;
+
+        if(test_count__ <= 1) {
+            test_no_exec__ = 1;
+        } else {
+#ifdef ACUTEST_WIN__
+            if(IsDebuggerPresent())
+                test_no_exec__ = 1;
+#endif
+#ifdef ACUTEST_LINUX__
+            if(test_is_tracer_present__())
+                test_no_exec__ = 1;
+#endif
+        }
+    }
+
+    /* Run the tests */
+    if(!test_skip_mode__) {
+        /* Run the listed tests. */
+        for(i = 0; i < (int) test_count__; i++)
+            test_run__(tests__[i]);
+    } else {
+        /* Run all tests except those listed. */
+        for(i = 0; test_list__[i].func != NULL; i++) {
+            if(!test_flags__[i])
+                test_run__(&test_list__[i]);
+        }
+    }
+
+    /* Write a summary */
+    if(!test_no_summary__ && test_verbose_level__ >= 1) {
+        if(test_verbose_level__ >= 3) {
+            test_print_in_color__(TEST_COLOR_DEFAULT_INTENSIVE__, "Summary:\n");
+
+            printf("  Count of all unit tests:     %4d\n", (int) test_list_size__);
+            printf("  Count of run unit tests:     %4d\n", test_stat_run_units__);
+            printf("  Count of failed unit tests:  %4d\n", test_stat_failed_units__);
+            printf("  Count of skipped unit tests: %4d\n", (int) test_list_size__ - test_stat_run_units__);
+            printf("  ");
+        }
+
+        if(test_stat_failed_units__ == 0) {
+            test_print_in_color__(TEST_COLOR_GREEN_INTENSIVE__, "SUCCESS:");
+            printf(" All unit tests have passed.\n");
+        } else {
+            test_print_in_color__(TEST_COLOR_RED_INTENSIVE__, "FAILED:");
+            printf(" %d of %d unit tests have failed.\n",
+                    test_stat_failed_units__, test_stat_run_units__);
+        }
+
+        if(test_verbose_level__ >= 3)
+            printf("\n");
+    }
+
+    free((void*) tests__);
+    free((void*) test_flags__);
+
+    return (test_stat_failed_units__ == 0) ? 0 : 1;
+}
+
+
+#endif  /* #ifndef TEST_NO_MAIN */
+
+#ifdef __cplusplus
+    }  /* extern "C" */
+#endif
+
+
+#endif  /* #ifndef ACUTEST_H__ */
diff --git a/fluent-bit/lib/cfl/tests/list.c b/fluent-bit/lib/cfl/tests/list.c
new file mode 100644
index 00000000..0c12f888
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/list.c
@@ -0,0 +1,50 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <cfl/cfl.h>
+
+#include "cfl_tests_internal.h"
+
+struct test {
+	struct cfl_list _head;
+};
+
+static void checks()
+{
+	struct test *t;
+	struct cfl_list list;
+
+	cfl_list_init(&list);
+	TEST_CHECK(cfl_list_is_empty(&list));
+
+	t = malloc(sizeof(struct test));
+	cfl_list_add(&t->_head, &list);
+	TEST_CHECK(!cfl_list_is_empty(&list));
+
+	cfl_list_del(&t->_head);
+	TEST_CHECK(cfl_list_is_empty(&list));
+
+	free(t);
+}
+
+TEST_LIST = {
+    {"checks",  checks},
+    { 0 }
+};
+
diff --git a/fluent-bit/lib/cfl/tests/sds.c b/fluent-bit/lib/cfl/tests/sds.c
new file mode 100644
index 00000000..899725e1
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/sds.c
@@ -0,0 +1,59 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <cfl/cfl.h>
+#include "cfl_tests_internal.h"
+
+static void test_sds_usage()
+{
+    cfl_sds_t s;
+
+    s = cfl_sds_create("test");
+    TEST_CHECK(s != NULL);
+    TEST_CHECK(cfl_sds_len(s) == 4);
+    TEST_CHECK(cfl_sds_alloc(s) == 4);
+    TEST_CHECK(strcmp("test", s) == 0);
+
+    s = cfl_sds_cat(s, ",cat message", 12);
+    TEST_CHECK(strcmp("test,cat message", s) == 0);
+
+    cfl_sds_destroy(s);
+}
+
+static void test_sds_printf()
+{
+    int len;
+    cfl_sds_t s;
+    cfl_sds_t tmp;
+    char *str = "0123456789ABCDEFGHIJQLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvewxyz";
+
+    s = cfl_sds_create_size(10);
+    tmp = cfl_sds_printf(&s, "%s=%s", str, str);
+
+    len = (strlen(str) * 2) + 1;
+    TEST_CHECK(tmp == s);
+    TEST_CHECK(cfl_sds_len(s) == len);
+    cfl_sds_destroy(s);
+}
+
+TEST_LIST = {
+    { "sds_usage" , test_sds_usage},
+    { "sds_printf", test_sds_printf},
+    { 0 }
+};
diff --git a/fluent-bit/lib/cfl/tests/variant.c b/fluent-bit/lib/cfl/tests/variant.c
new file mode 100644
index 00000000..83dcb33a
--- /dev/null
+++ b/fluent-bit/lib/cfl/tests/variant.c
@@ -0,0 +1,470 @@
+/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 4 -*- */
+
+/*  CFL
+ *  ===
+ *  Copyright (C) 2022 The CFL Authors
+ *
+ *  Licensed under the Apache License, Version 2.0 (the "License");
+ *  you may not use this file except in compliance with the License.
+ *  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ *  Unless required by applicable law or agreed to in writing, software
+ *  distributed under the License is distributed on an "AS IS" BASIS,
+ *  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ *  See the License for the specific language governing permissions and
+ *  limitations under the License.
+ */
+
+#include <stdio.h>
+#include <string.h>
+#include <cfl/cfl.h>
+#include <cfl/cfl_variant.h>
+
+#include "cfl_tests_internal.h"
+
+static int compare(FILE *fp, char *expect, int ignore_len)
+{
+    int ret;
+    size_t len = strlen(expect);
+    size_t ret_fp;
+    char buf[1024] = {0};
+
+    ret = fseek(fp, 0, SEEK_SET);
+    if (!TEST_CHECK(ret == 0)) {
+        TEST_MSG("fseek failed");
+        return ret;
+    }
+
+    ret_fp = fread(&buf[0], 1, sizeof(buf), fp);
+    if (ret_fp == 0) {
+        if (!TEST_CHECK(feof(fp) == 0)) {
+            TEST_MSG("fread error.");
+            return -1;
+        }
+    }
+    if (ignore_len) {
+        if (!TEST_CHECK(strstr(&buf[0], expect) != NULL)) {
+            TEST_MSG("compare error. got=%s expect=%s", &buf[0], expect);
+            return -1;
+        }
+
+    } else {
+        if (!TEST_CHECK(strlen(buf) == len)) {
+            TEST_MSG("length error. len=%d got=%s expect=%s", strlen(buf), &buf[0], expect);
+            return -1;
+        }
+        if (!TEST_CHECK(strncmp(expect, &buf[0], len) == 0)) {
+            TEST_MSG("compare error. got=%s expect=%s", &buf[0], expect);
+            return -1;
+        }
+    }
+    return 0;
+}
+
+static void test_variant_print_bool()
+{
+    int ret;
+    int i;
+    int inputs[] = {CFL_TRUE, CFL_FALSE};
+    char *expects[] = {"true", "false"};
+
+    FILE *fp = NULL;
+    struct cfl_variant *val = NULL;
+
+    for (i=0; i<sizeof(inputs)/sizeof(int); i++) {
+        fp = tmpfile();
+        if (!TEST_CHECK(fp != NULL)) {
+            TEST_MSG("%d: fp is NULL", i);
+            continue;
+        }
+
+        val = cfl_variant_create_from_bool(inputs[i]);
+        if (!TEST_CHECK(val != NULL)) {
+            TEST_MSG("%d: cfl_variant_create_from_bool failed", i);
+            fclose(fp);
+            continue;
+        }
+
+        ret = cfl_variant_print(fp, val);
+        /* Check whether EOF or not. Not checking for positive
+         * number here. */
+        if (!TEST_CHECK(ret != EOF)) {
+            TEST_MSG("%d:cfl_variant_print failed", i);
+            cfl_variant_destroy(val);
+            fclose(fp);
+            continue;
+        }
+        ret = compare(fp, expects[i], 0);
+        if (!TEST_CHECK(ret == 0)) {
+            TEST_MSG("%d:compare failed", i);
+        }
+        cfl_variant_destroy(val);
+        fclose(fp);
+    }
+}
+
+static void test_variant_print_int64()
+{
+    int ret;
+    int i;
+    int inputs[] = {1, 0, -123};
+    char *expects[] = {"1", "0", "-123"};
+
+    FILE *fp = NULL;
+    struct cfl_variant *val = NULL;
+
+    for (i=0; i<sizeof(inputs)/sizeof(int); i++) {
+        fp = tmpfile();
+        if (!TEST_CHECK(fp != NULL)) {
+            TEST_MSG("%d: fp is NULL", i);
+            continue;
+        }
+        val = cfl_variant_create_from_int64(inputs[i]);
+        if (!TEST_CHECK(val != NULL)) {
+            TEST_MSG("%d: cfl_variant_create_from_int64 failed", i);
+            fclose(fp);
+            continue;
+        }
+
+        ret = cfl_variant_print(fp, val);
+        if (!TEST_CHECK(ret > 0)) {
+            TEST_MSG("%d:cfl_variant_print failed", i);
+            cfl_variant_destroy(val);
+            fclose(fp);
+            continue;
+        }
+        ret = compare(fp, expects[i], 0);
+        if (!TEST_CHECK(ret == 0)) {
+            TEST_MSG("%d:compare failed", i);
+        }
+        cfl_variant_destroy(val);
+        fclose(fp);
+    }
+}
+
+static void test_variant_print_array()
+{
+    int ret;
+    int i;
+    int64_t inputs[] = {1, 0, -123};
+    char *expect = {"[1,0,-123]"};
+
+    FILE *fp = NULL;
+    struct cfl_array *array = NULL;
+    struct cfl_variant *val = NULL;
+
+    fp = tmpfile();
+    if (!TEST_CHECK(fp != NULL)) {
+        TEST_MSG("fp is NULL");
+        exit(1);
+    }
+
+    array = cfl_array_create(sizeof(inputs)/sizeof(int64_t));
+    if (!TEST_CHECK(array != NULL)) {
+        TEST_MSG("cfl_array_create failed");
+        fclose(fp);
+        exit(1);
+    }
+
+
+    for (i=0; i<sizeof(inputs)/sizeof(int64_t); i++) {
+        ret = cfl_array_append_int64(array, inputs[i]);
+        if (!TEST_CHECK(ret == 0)) {
+            TEST_MSG("%d: cfl_array_append_int64 failed", i);
+            fclose(fp);
+            cfl_array_destroy(array);
+            exit(1);
+        }
+    }
+
+
+    val = cfl_variant_create_from_array(array);
+    if (!TEST_CHECK(val != NULL)) {
+        TEST_MSG("cfl_variant_create_from_array failed");
+        cfl_array_destroy(array);
+        fclose(fp);
+        exit(1);
+    }
+    ret = cfl_variant_print(fp, val);
+    if (!TEST_CHECK(ret > 0)) {
+        TEST_MSG("cfl_variant_print failed");
+        cfl_variant_destroy(val);
+        fclose(fp);
+        exit(1);
+    }
+    ret = compare(fp, expect, 0);
+    if (!TEST_CHECK(ret == 0)) {
+        TEST_MSG("compare failed");
+    }
+    cfl_variant_destroy(val);
+    fclose(fp);
+}
+
+static void test_variant_print_kvlist()
+{
+    int ret;
+    int i;
+    char *key_inputs[] = {"key", "key2", "aaa"};
+    int64_t val_inputs[] = {1, 0, -123};
+    char *expect = {"{\"key\":1,\"key2\":0,\"aaa\":-123}"};
+
+    FILE *fp = NULL;
+    struct cfl_kvlist *list = NULL;
+    struct cfl_variant *val = NULL;
+
+    if (!TEST_CHECK(sizeof(key_inputs)/sizeof(char*) == sizeof(val_inputs)/sizeof(int64_t))) {
+        TEST_MSG("key val array size mismatch. key_len=%d val_len=%d", 
+                 sizeof(key_inputs)/sizeof(char*),
+                 sizeof(val_inputs)/sizeof(int64_t));
+        exit(1);
+    }
+
+    fp = tmpfile();
+    if (!TEST_CHECK(fp != NULL)) {
+        TEST_MSG("fp is NULL");
+        exit(1);
+    }
+
+    list = cfl_kvlist_create();
+    if (!TEST_CHECK(list != NULL)) {
+        TEST_MSG("cfl_kvlist_create failed");
+        fclose(fp);
+        exit(1);
+    }
+
+    for (i=0; i<sizeof(key_inputs)/sizeof(char*); i++) {
+        ret = cfl_kvlist_insert_int64(list, key_inputs[i], val_inputs[i]);
+        if (!TEST_CHECK(ret == 0)) {
+            TEST_MSG("%d: cfl_kvlist_insert_int64 failed", i);
+            fclose(fp);
+            cfl_kvlist_destroy(list);
+            exit(1);
+        }
+    }
+
+    val = cfl_variant_create_from_kvlist(list);
+    if (!TEST_CHECK(val != NULL)) {
+        TEST_MSG("cfl_variant_create_from_kvlist failed");
+        cfl_kvlist_destroy(list);
+        fclose(fp);
+        exit(1);
+    }
+    ret = cfl_variant_print(fp, val);
+    if (!TEST_CHECK(ret > 0)) {
+        TEST_MSG("cfl_variant_print failed");
+        cfl_variant_destroy(val);
+        fclose(fp);
+        exit(1);
+    }
+    ret = compare(fp, expect, 0);
+    if (!TEST_CHECK(ret == 0)) {
+        TEST_MSG("compare failed");
+    }
+    cfl_variant_destroy(val);
+    fclose(fp);
+}
+
+static void test_variant_print_double()
+{
+    int ret;
+    int i;
+    double inputs[] = {1.0, -12.3};
+    char *expects[] = {"1.0", "-12.3"};
+
+    FILE *fp = NULL;
+    struct cfl_variant *val = NULL;
+
+    for (i=0; i<sizeof(inputs)/sizeof(double); i++) {
+        fp = tmpfile();
+        if (!TEST_CHECK(fp != NULL)) {
+            TEST_MSG("%d: fp is NULL", i);
+            continue;
+        }
+        val = cfl_variant_create_from_double(inputs[i]);
+        if (!TEST_CHECK(val != NULL)) {
+            TEST_MSG("%d: cfl_variant_create_from_double failed", i);
+            fclose(fp);
+            continue;
+        }
+
+        ret = cfl_variant_print(fp, val);
+        if (!TEST_CHECK(ret > 0)) {
+            TEST_MSG("%d:cfl_variant_print failed", i);
+            cfl_variant_destroy(val);
+            fclose(fp);
+            continue;
+        }
+        ret = compare(fp, expects[i], 1);
+        if (!TEST_CHECK(ret == 0)) {
+            TEST_MSG("%d:compare failed", i);
+        }
+        cfl_variant_destroy(val);
+        fclose(fp);
+    }
+}
+
+static void test_variant_print_string()
+{
+    int ret;
+    int i;
+    char *inputs[] = {"hoge", "aaa"};
+    char *expects[] = {"\"hoge\"", "\"aaa\""};
+
+    FILE *fp = NULL;
+    struct cfl_variant *val = NULL;
+
+    for (i=0; i<sizeof(inputs)/sizeof(char*); i++) {
+        fp = tmpfile();
+        if (!TEST_CHECK(fp != NULL)) {
+            TEST_MSG("%d: fp is NULL", i);
+            continue;
+        }
+
+        val = cfl_variant_create_from_string(inputs[i]);
+        if (!TEST_CHECK(val != NULL)) {
+            TEST_MSG("%d: cfl_variant_create_from_string failed", i);
+            fclose(fp);
+            continue;
+        }
+
+        ret = cfl_variant_print(fp, val);
+        if (!TEST_CHECK(ret > 0)) {
+            TEST_MSG("%d:cfl_variant_print failed", i);
+            cfl_variant_destroy(val);
+            fclose(fp);
+            continue;
+        }
+        ret = compare(fp, expects[i], 0);
+        if (!TEST_CHECK(ret == 0)) {
+            TEST_MSG("%d:compare failed", i);
+        }
+        cfl_variant_destroy(val);
+        fclose(fp);
+    }
+}
+
+static void test_variant_print_bytes()
+{
+    int ret;
+    char input[] = {0x1f, 0xaa, 0x0a, 0xff};
+    char *expect = "1faa0aff";
+
+    FILE *fp = NULL;
+    struct cfl_variant *val = NULL;
+
+    fp = tmpfile();
+    if (!TEST_CHECK(fp != NULL)) {
+        TEST_MSG("fp is NULL");
+        exit(1);
+    }
+
+    val = cfl_variant_create_from_bytes(input, 4);
+    if (!TEST_CHECK(val != NULL)) {
+        TEST_MSG("cfl_variant_create_from_bytes failed");
+        fclose(fp);
+        exit(1);
+    }
+
+    ret = cfl_variant_print(fp, val);
+    if (!TEST_CHECK(ret > 0)) {
+        TEST_MSG("cfl_variant_print failed");
+        cfl_variant_destroy(val);
+        fclose(fp);
+        exit(1);
+    }
+    ret = compare(fp, expect, 0);
+    if (!TEST_CHECK(ret == 0)) {
+        TEST_MSG("compare failed");
+    }
+    cfl_variant_destroy(val);
+    fclose(fp);
+}
+
+
+static void test_variant_print_reference()
+{
+    int ret;
+    int *input = (int*)0x12345678;
+    char expect[] = "0x12345678";
+
+    FILE *fp = NULL;
+    struct cfl_variant *val = NULL;
+
+    fp = tmpfile();
+    if (!TEST_CHECK(fp != NULL)) {
+        TEST_MSG("fp is NULL");
+        exit(1);
+    }
+
+    val = cfl_variant_create_from_reference(input);
+    if (!TEST_CHECK(val != NULL)) {
+        TEST_MSG("cfl_variant_create_from_reference failed");
+        fclose(fp);
+        exit(1);
+    }
+
+    ret = cfl_variant_print(fp, val);
+    if (!TEST_CHECK(ret > 0)) {
+        TEST_MSG("cfl_variant_print failed");
+        fclose(fp);
+        cfl_variant_destroy(val);
+        exit(1);
+    }
+    ret = compare(fp, &expect[0], 0);
+    if (!TEST_CHECK(ret == 0)) {
+        TEST_MSG("compare failed");
+    }
+    cfl_variant_destroy(val);
+    fclose(fp);
+}
+
+static void test_variant_print_unknown()
+{
+    int ret;
+    char expect[] = "Unknown";
+
+    FILE *fp = NULL;
+    struct cfl_variant *val = NULL;
+
+    fp = tmpfile();
+    if (!TEST_CHECK(fp != NULL)) {
+        TEST_MSG("fp is NULL");
+        exit(1);
+    }
+
+    val = cfl_variant_create();
+    if (!TEST_CHECK(val != NULL)) {
+        TEST_MSG("cfl_variant_create failed");
+        fclose(fp);
+        exit(1);
+    }
+
+    ret = cfl_variant_print(fp, val);
+    if (!TEST_CHECK(ret != EOF)) {
+        TEST_MSG("cfl_variant_print failed");
+        fclose(fp);
+        cfl_variant_destroy(val);
+        exit(1);
+    }
+    ret = compare(fp, &expect[0], 1);
+    if (!TEST_CHECK(ret == 0)) {
+        TEST_MSG("compare failed");
+    }
+    cfl_variant_destroy(val);
+    fclose(fp);
+}
+
+TEST_LIST = {
+    {"variant_print_bool", test_variant_print_bool},
+    {"variant_print_int64", test_variant_print_int64},
+    {"variant_print_double", test_variant_print_double},
+    {"variant_print_string", test_variant_print_string},
+    {"variant_print_bytes", test_variant_print_bytes},
+    {"variant_print_array", test_variant_print_array},
+    {"variant_print_kvlist", test_variant_print_kvlist},
+    {"variant_print_reference", test_variant_print_reference},
+    {"variant_print_unknown", test_variant_print_unknown},
+    { 0 }
+};