Add files using upload-large-folder tool

third_party/tracy/.github/FUNDING.yml

@@ -0,0 +1 @@
+github: wolfpld

third_party/tracy/.github/workflows/build.yml

@@ -0,0 +1,78 @@
+name: build
+
+on:
+  push:
+    branches: [ master ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  build:
+    strategy:
+      matrix:
+        os: [ windows-latest, macos-latest ]
+
+    runs-on: ${{ matrix.os }}
+    continue-on-error: true
+    steps:
+    - uses: actions/checkout@v4
+    - if: startsWith(matrix.os, 'windows')
+      uses: microsoft/setup-msbuild@v2
+    - if: startsWith(matrix.os, 'windows')
+      uses: actions/setup-python@v2
+      with:
+        python-version: '3.x'
+    - if: startsWith(matrix.os, 'windows')
+      run: pip install meson ninja
+    - if: startsWith(matrix.os, 'macos')
+      name: Install macos dependencies
+      run: brew install pkg-config glfw meson
+    - name: Profiler GUI
+      run: |
+        cmake -B profiler/build -S profiler -DCMAKE_BUILD_TYPE=Release
+        cmake --build profiler/build --parallel --config Release
+    - name: Update utility
+      run: |
+        cmake -B update/build -S update -DCMAKE_BUILD_TYPE=Release
+        cmake --build update/build --parallel --config Release
+    - name: Capture utility
+      run: |
+        cmake -B capture/build -S capture -DCMAKE_BUILD_TYPE=Release
+        cmake --build capture/build --parallel --config Release
+    - name: Csvexport utility
+      run: |
+        cmake -B csvexport/build -S csvexport -DCMAKE_BUILD_TYPE=Release
+        cmake --build csvexport/build --parallel --config Release
+    - name: Import utilities
+      run: |
+        cmake -B import/build -S import -DCMAKE_BUILD_TYPE=Release
+        cmake --build import/build --parallel --config Release
+    - if: ${{ !startsWith(matrix.os, 'windows') }}
+      name: Library
+      run: meson setup -Dprefix=$GITHUB_WORKSPACE/bin/lib build && meson compile -C build && meson install -C build
+    - if: ${{ !startsWith(matrix.os, 'windows') }}
+      name: Find Artifacts
+      id: find_artifacts
+      run: |
+        mkdir -p bin
+        cp profiler/build/tracy-profiler bin
+        cp update/build/tracy-update bin
+        cp capture/build/tracy-capture bin
+        cp csvexport/build/tracy-csvexport bin
+        cp import/build/tracy-import-chrome bin
+        cp import/build/tracy-import-fuchsia bin
+    - if: startsWith(matrix.os, 'windows')
+      name: Find Artifacts
+      id: find_artifacts_windows
+      run: |
+        mkdir bin
+        copy profiler\build\Release\tracy-profiler.exe bin
+        copy update\build\Release\tracy-update.exe bin
+        copy capture\build\Release\tracy-capture.exe bin
+        copy csvexport\build\Release\tracy-csvexport.exe bin
+        copy import\build\Release\tracy-import-chrome.exe bin
+        copy import\build\Release\tracy-import-fuchsia.exe bin
+    - uses: actions/upload-artifact@v4
+      with:
+        name: ${{ matrix.os }}
+        path: bin

third_party/tracy/.github/workflows/latex.yml

@@ -0,0 +1,27 @@
+name: Manual
+
+on:
+  push:
+    branches: [ master ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v4
+    - name: Fix stupidity
+      run: |
+        cp LICENSE LICENSE.
+    - name: Compile LaTeX
+      uses: xu-cheng/latex-action@v3
+      with:
+        working_directory: manual
+        root_file: tracy.tex
+    - uses: actions/upload-artifact@v4
+      with:
+        name: manual
+        path: manual/tracy.pdf

third_party/tracy/.github/workflows/linux.yml

@@ -0,0 +1,75 @@
+name: linux
+
+on:
+  push:
+    branches: [ master ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    container: archlinux:base-devel
+    steps:
+    - name: Install dependencies
+      run: pacman -Syu --noconfirm && pacman -S --noconfirm --needed freetype2 tbb debuginfod wayland dbus libxkbcommon libglvnd meson cmake git wayland-protocols nodejs
+    - uses: actions/checkout@v4
+    - name: Profiler GUI
+      run: |
+        cmake -B profiler/build -S profiler -DCMAKE_BUILD_TYPE=Release
+        cmake --build profiler/build --parallel
+    - name: Update utility
+      run: |
+        cmake -B update/build -S update -DCMAKE_BUILD_TYPE=Release
+        cmake --build update/build --parallel
+    - name: Capture utility
+      run: |
+        cmake -B capture/build -S capture -DCMAKE_BUILD_TYPE=Release
+        cmake --build capture/build --parallel
+    - name: Csvexport utility
+      run: |
+        cmake -B csvexport/build -S csvexport -DCMAKE_BUILD_TYPE=Release
+        cmake --build csvexport/build --parallel
+    - name: Import utilities
+      run: |
+        cmake -B import/build -S import -DCMAKE_BUILD_TYPE=Release
+        cmake --build import/build --parallel
+    - name: Library
+      run: meson setup -Dprefix=$GITHUB_WORKSPACE/bin/lib build && meson compile -C build && meson install -C build
+    - name: Test application
+      run: |
+        # test compilation with different flags
+        # we clean the build folder to reset cached variables between runs
+        cmake -B test/build -S test -DCMAKE_BUILD_TYPE=Release
+        cmake --build test/build --parallel
+        rm -rf test/build
+
+        # same with TRACY_ON_DEMAND
+        cmake -B test/build -S test -DCMAKE_BUILD_TYPE=Release -DTRACY_ON_DEMAND=ON .
+        cmake --build test/build --parallel
+        rm -rf test/build
+
+        # same with TRACY_DELAYED_INIT TRACY_MANUAL_LIFETIME
+        cmake -B test/build -S test -DCMAKE_BUILD_TYPE=Release -DTRACY_DELAYED_INIT=ON -DTRACY_MANUAL_LIFETIME=ON .
+        cmake --build test/build --parallel
+        rm -rf test/build
+
+        # same with TRACY_DEMANGLE
+        cmake -B test/build -S test -DCMAKE_BUILD_TYPE=Release -DTRACY_DEMANGLE=ON .
+        cmake --build test/build --parallel
+        rm -rf test/build
+    - name: Find Artifacts
+      id: find_artifacts
+      run: |
+        mkdir -p bin
+        cp profiler/build/tracy-profiler bin
+        cp update/build/tracy-update bin
+        cp capture/build/tracy-capture bin
+        cp csvexport/build/tracy-csvexport bin
+        cp import/build/tracy-import-chrome bin
+        cp import/build/tracy-import-fuchsia bin
+        strip bin/tracy-*
+    - uses: actions/upload-artifact@v4
+      with:
+        name: arch-linux
+        path: bin

third_party/tracy/cmake/CPM.cmake

@@ -0,0 +1,1269 @@
+# CPM.cmake - CMake's missing package manager
+# ===========================================
+# See https://github.com/cpm-cmake/CPM.cmake for usage and update instructions.
+#
+# MIT License
+# -----------
+#[[
+  Copyright (c) 2019-2023 Lars Melchior and contributors
+
+  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.
+]]
+
+cmake_minimum_required(VERSION 3.14 FATAL_ERROR)
+
+# Initialize logging prefix
+if(NOT CPM_INDENT)
+  set(CPM_INDENT
+      "CPM:"
+      CACHE INTERNAL ""
+  )
+endif()
+
+if(NOT COMMAND cpm_message)
+  function(cpm_message)
+    message(${ARGV})
+  endfunction()
+endif()
+
+set(CURRENT_CPM_VERSION 0.40.2)
+
+get_filename_component(CPM_CURRENT_DIRECTORY "${CMAKE_CURRENT_LIST_DIR}" REALPATH)
+if(CPM_DIRECTORY)
+  if(NOT CPM_DIRECTORY STREQUAL CPM_CURRENT_DIRECTORY)
+    if(CPM_VERSION VERSION_LESS CURRENT_CPM_VERSION)
+      message(
+        AUTHOR_WARNING
+          "${CPM_INDENT} \
+A dependency is using a more recent CPM version (${CURRENT_CPM_VERSION}) than the current project (${CPM_VERSION}). \
+It is recommended to upgrade CPM to the most recent version. \
+See https://github.com/cpm-cmake/CPM.cmake for more information."
+      )
+    endif()
+    if(${CMAKE_VERSION} VERSION_LESS "3.17.0")
+      include(FetchContent)
+    endif()
+    return()
+  endif()
+
+  get_property(
+    CPM_INITIALIZED GLOBAL ""
+    PROPERTY CPM_INITIALIZED
+    SET
+  )
+  if(CPM_INITIALIZED)
+    return()
+  endif()
+endif()
+
+if(CURRENT_CPM_VERSION MATCHES "development-version")
+  message(
+    WARNING "${CPM_INDENT} Your project is using an unstable development version of CPM.cmake. \
+Please update to a recent release if possible. \
+See https://github.com/cpm-cmake/CPM.cmake for details."
+  )
+endif()
+
+set_property(GLOBAL PROPERTY CPM_INITIALIZED true)
+
+macro(cpm_set_policies)
+  # the policy allows us to change options without caching
+  cmake_policy(SET CMP0077 NEW)
+  set(CMAKE_POLICY_DEFAULT_CMP0077 NEW)
+
+  # the policy allows us to change set(CACHE) without caching
+  if(POLICY CMP0126)
+    cmake_policy(SET CMP0126 NEW)
+    set(CMAKE_POLICY_DEFAULT_CMP0126 NEW)
+  endif()
+
+  # The policy uses the download time for timestamp, instead of the timestamp in the archive. This
+  # allows for proper rebuilds when a projects url changes
+  if(POLICY CMP0135)
+    cmake_policy(SET CMP0135 NEW)
+    set(CMAKE_POLICY_DEFAULT_CMP0135 NEW)
+  endif()
+
+  # treat relative git repository paths as being relative to the parent project's remote
+  if(POLICY CMP0150)
+    cmake_policy(SET CMP0150 NEW)
+    set(CMAKE_POLICY_DEFAULT_CMP0150 NEW)
+  endif()
+endmacro()
+cpm_set_policies()
+
+option(CPM_USE_LOCAL_PACKAGES "Always try to use `find_package` to get dependencies"
+       $ENV{CPM_USE_LOCAL_PACKAGES}
+)
+option(CPM_LOCAL_PACKAGES_ONLY "Only use `find_package` to get dependencies"
+       $ENV{CPM_LOCAL_PACKAGES_ONLY}
+)
+option(CPM_DOWNLOAD_ALL "Always download dependencies from source" $ENV{CPM_DOWNLOAD_ALL})
+option(CPM_DONT_UPDATE_MODULE_PATH "Don't update the module path to allow using find_package"
+       $ENV{CPM_DONT_UPDATE_MODULE_PATH}
+)
+option(CPM_DONT_CREATE_PACKAGE_LOCK "Don't create a package lock file in the binary path"
+       $ENV{CPM_DONT_CREATE_PACKAGE_LOCK}
+)
+option(CPM_INCLUDE_ALL_IN_PACKAGE_LOCK
+       "Add all packages added through CPM.cmake to the package lock"
+       $ENV{CPM_INCLUDE_ALL_IN_PACKAGE_LOCK}
+)
+option(CPM_USE_NAMED_CACHE_DIRECTORIES
+       "Use additional directory of package name in cache on the most nested level."
+       $ENV{CPM_USE_NAMED_CACHE_DIRECTORIES}
+)
+
+set(CPM_VERSION
+    ${CURRENT_CPM_VERSION}
+    CACHE INTERNAL ""
+)
+set(CPM_DIRECTORY
+    ${CPM_CURRENT_DIRECTORY}
+    CACHE INTERNAL ""
+)
+set(CPM_FILE
+    ${CMAKE_CURRENT_LIST_FILE}
+    CACHE INTERNAL ""
+)
+set(CPM_PACKAGES
+    ""
+    CACHE INTERNAL ""
+)
+set(CPM_DRY_RUN
+    OFF
+    CACHE INTERNAL "Don't download or configure dependencies (for testing)"
+)
+
+if(DEFINED ENV{CPM_SOURCE_CACHE})
+  set(CPM_SOURCE_CACHE_DEFAULT $ENV{CPM_SOURCE_CACHE})
+else()
+  set(CPM_SOURCE_CACHE_DEFAULT OFF)
+endif()
+
+set(CPM_SOURCE_CACHE
+    ${CPM_SOURCE_CACHE_DEFAULT}
+    CACHE PATH "Directory to download CPM dependencies"
+)
+
+if(NOT CPM_DONT_UPDATE_MODULE_PATH)
+  set(CPM_MODULE_PATH
+      "${CMAKE_BINARY_DIR}/CPM_modules"
+      CACHE INTERNAL ""
+  )
+  # remove old modules
+  file(REMOVE_RECURSE ${CPM_MODULE_PATH})
+  file(MAKE_DIRECTORY ${CPM_MODULE_PATH})
+  # locally added CPM modules should override global packages
+  set(CMAKE_MODULE_PATH "${CPM_MODULE_PATH};${CMAKE_MODULE_PATH}")
+endif()
+
+if(NOT CPM_DONT_CREATE_PACKAGE_LOCK)
+  set(CPM_PACKAGE_LOCK_FILE
+      "${CMAKE_BINARY_DIR}/cpm-package-lock.cmake"
+      CACHE INTERNAL ""
+  )
+  file(WRITE ${CPM_PACKAGE_LOCK_FILE}
+       "# CPM Package Lock\n# This file should be committed to version control\n\n"
+  )
+endif()
+
+include(FetchContent)
+
+# Try to infer package name from git repository uri (path or url)
+function(cpm_package_name_from_git_uri URI RESULT)
+  if("${URI}" MATCHES "([^/:]+)/?.git/?$")
+    set(${RESULT}
+        ${CMAKE_MATCH_1}
+        PARENT_SCOPE
+    )
+  else()
+    unset(${RESULT} PARENT_SCOPE)
+  endif()
+endfunction()
+
+# Try to infer package name and version from a url
+function(cpm_package_name_and_ver_from_url url outName outVer)
+  if(url MATCHES "[/\\?]([a-zA-Z0-9_\\.-]+)\\.(tar|tar\\.gz|tar\\.bz2|zip|ZIP)(\\?|/|$)")
+    # We matched an archive
+    set(filename "${CMAKE_MATCH_1}")
+
+    if(filename MATCHES "([a-zA-Z0-9_\\.-]+)[_-]v?(([0-9]+\\.)*[0-9]+[a-zA-Z0-9]*)")
+      # We matched <name>-<version> (ie foo-1.2.3)
+      set(${outName}
+          "${CMAKE_MATCH_1}"
+          PARENT_SCOPE
+      )
+      set(${outVer}
+          "${CMAKE_MATCH_2}"
+          PARENT_SCOPE
+      )
+    elseif(filename MATCHES "(([0-9]+\\.)+[0-9]+[a-zA-Z0-9]*)")
+      # We couldn't find a name, but we found a version
+      #
+      # In many cases (which we don't handle here) the url would look something like
+      # `irrelevant/ACTUAL_PACKAGE_NAME/irrelevant/1.2.3.zip`. In such a case we can't possibly
+      # distinguish the package name from the irrelevant bits. Moreover if we try to match the
+      # package name from the filename, we'd get bogus at best.
+      unset(${outName} PARENT_SCOPE)
+      set(${outVer}
+          "${CMAKE_MATCH_1}"
+          PARENT_SCOPE
+      )
+    else()
+      # Boldly assume that the file name is the package name.
+      #
+      # Yes, something like `irrelevant/ACTUAL_NAME/irrelevant/download.zip` will ruin our day, but
+      # such cases should be quite rare. No popular service does this... we think.
+      set(${outName}
+          "${filename}"
+          PARENT_SCOPE
+      )
+      unset(${outVer} PARENT_SCOPE)
+    endif()
+  else()
+    # No ideas yet what to do with non-archives
+    unset(${outName} PARENT_SCOPE)
+    unset(${outVer} PARENT_SCOPE)
+  endif()
+endfunction()
+
+function(cpm_find_package NAME VERSION)
+  string(REPLACE " " ";" EXTRA_ARGS "${ARGN}")
+  find_package(${NAME} ${VERSION} ${EXTRA_ARGS} QUIET)
+  if(${CPM_ARGS_NAME}_FOUND)
+    if(DEFINED ${CPM_ARGS_NAME}_VERSION)
+      set(VERSION ${${CPM_ARGS_NAME}_VERSION})
+    endif()
+    cpm_message(STATUS "${CPM_INDENT} Using local package ${CPM_ARGS_NAME}@${VERSION}")
+    CPMRegisterPackage(${CPM_ARGS_NAME} "${VERSION}")
+    set(CPM_PACKAGE_FOUND
+        YES
+        PARENT_SCOPE
+    )
+  else()
+    set(CPM_PACKAGE_FOUND
+        NO
+        PARENT_SCOPE
+    )
+  endif()
+endfunction()
+
+# Create a custom FindXXX.cmake module for a CPM package This prevents `find_package(NAME)` from
+# finding the system library
+function(cpm_create_module_file Name)
+  if(NOT CPM_DONT_UPDATE_MODULE_PATH)
+    # erase any previous modules
+    file(WRITE ${CPM_MODULE_PATH}/Find${Name}.cmake
+         "include(\"${CPM_FILE}\")\n${ARGN}\nset(${Name}_FOUND TRUE)"
+    )
+  endif()
+endfunction()
+
+# Find a package locally or fallback to CPMAddPackage
+function(CPMFindPackage)
+  set(oneValueArgs NAME VERSION GIT_TAG FIND_PACKAGE_ARGUMENTS)
+
+  cmake_parse_arguments(CPM_ARGS "" "${oneValueArgs}" "" ${ARGN})
+
+  if(NOT DEFINED CPM_ARGS_VERSION)
+    if(DEFINED CPM_ARGS_GIT_TAG)
+      cpm_get_version_from_git_tag("${CPM_ARGS_GIT_TAG}" CPM_ARGS_VERSION)
+    endif()
+  endif()
+
+  set(downloadPackage ${CPM_DOWNLOAD_ALL})
+  if(DEFINED CPM_DOWNLOAD_${CPM_ARGS_NAME})
+    set(downloadPackage ${CPM_DOWNLOAD_${CPM_ARGS_NAME}})
+  elseif(DEFINED ENV{CPM_DOWNLOAD_${CPM_ARGS_NAME}})
+    set(downloadPackage $ENV{CPM_DOWNLOAD_${CPM_ARGS_NAME}})
+  endif()
+  if(downloadPackage)
+    CPMAddPackage(${ARGN})
+    cpm_export_variables(${CPM_ARGS_NAME})
+    return()
+  endif()
+
+  cpm_find_package(${CPM_ARGS_NAME} "${CPM_ARGS_VERSION}" ${CPM_ARGS_FIND_PACKAGE_ARGUMENTS})
+
+  if(NOT CPM_PACKAGE_FOUND)
+    CPMAddPackage(${ARGN})
+    cpm_export_variables(${CPM_ARGS_NAME})
+  endif()
+
+endfunction()
+
+# checks if a package has been added before
+function(cpm_check_if_package_already_added CPM_ARGS_NAME CPM_ARGS_VERSION)
+  if("${CPM_ARGS_NAME}" IN_LIST CPM_PACKAGES)
+    CPMGetPackageVersion(${CPM_ARGS_NAME} CPM_PACKAGE_VERSION)
+    if("${CPM_PACKAGE_VERSION}" VERSION_LESS "${CPM_ARGS_VERSION}")
+      message(
+        WARNING
+          "${CPM_INDENT} Requires a newer version of ${CPM_ARGS_NAME} (${CPM_ARGS_VERSION}) than currently included (${CPM_PACKAGE_VERSION})."
+      )
+    endif()
+    cpm_get_fetch_properties(${CPM_ARGS_NAME})
+    set(${CPM_ARGS_NAME}_ADDED NO)
+    set(CPM_PACKAGE_ALREADY_ADDED
+        YES
+        PARENT_SCOPE
+    )
+    cpm_export_variables(${CPM_ARGS_NAME})
+  else()
+    set(CPM_PACKAGE_ALREADY_ADDED
+        NO
+        PARENT_SCOPE
+    )
+  endif()
+endfunction()
+
+# Parse the argument of CPMAddPackage in case a single one was provided and convert it to a list of
+# arguments which can then be parsed idiomatically. For example gh:foo/bar@1.2.3 will be converted
+# to: GITHUB_REPOSITORY;foo/bar;VERSION;1.2.3
+function(cpm_parse_add_package_single_arg arg outArgs)
+  # Look for a scheme
+  if("${arg}" MATCHES "^([a-zA-Z]+):(.+)$")
+    string(TOLOWER "${CMAKE_MATCH_1}" scheme)
+    set(uri "${CMAKE_MATCH_2}")
+
+    # Check for CPM-specific schemes
+    if(scheme STREQUAL "gh")
+      set(out "GITHUB_REPOSITORY;${uri}")
+      set(packageType "git")
+    elseif(scheme STREQUAL "gl")
+      set(out "GITLAB_REPOSITORY;${uri}")
+      set(packageType "git")
+    elseif(scheme STREQUAL "bb")
+      set(out "BITBUCKET_REPOSITORY;${uri}")
+      set(packageType "git")
+      # A CPM-specific scheme was not found. Looks like this is a generic URL so try to determine
+      # type
+    elseif(arg MATCHES ".git/?(@|#|$)")
+      set(out "GIT_REPOSITORY;${arg}")
+      set(packageType "git")
+    else()
+      # Fall back to a URL
+      set(out "URL;${arg}")
+      set(packageType "archive")
+
+      # We could also check for SVN since FetchContent supports it, but SVN is so rare these days.
+      # We just won't bother with the additional complexity it will induce in this function. SVN is
+      # done by multi-arg
+    endif()
+  else()
+    if(arg MATCHES ".git/?(@|#|$)")
+      set(out "GIT_REPOSITORY;${arg}")
+      set(packageType "git")
+    else()
+      # Give up
+      message(FATAL_ERROR "${CPM_INDENT} Can't determine package type of '${arg}'")
+    endif()
+  endif()
+
+  # For all packages we interpret @... as version. Only replace the last occurrence. Thus URIs
+  # containing '@' can be used
+  string(REGEX REPLACE "@([^@]+)$" ";VERSION;\\1" out "${out}")
+
+  # Parse the rest according to package type
+  if(packageType STREQUAL "git")
+    # For git repos we interpret #... as a tag or branch or commit hash
+    string(REGEX REPLACE "#([^#]+)$" ";GIT_TAG;\\1" out "${out}")
+  elseif(packageType STREQUAL "archive")
+    # For archives we interpret #... as a URL hash.
+    string(REGEX REPLACE "#([^#]+)$" ";URL_HASH;\\1" out "${out}")
+    # We don't try to parse the version if it's not provided explicitly. cpm_get_version_from_url
+    # should do this at a later point
+  else()
+    # We should never get here. This is an assertion and hitting it means there's a problem with the
+    # code above. A packageType was set, but not handled by this if-else.
+    message(FATAL_ERROR "${CPM_INDENT} Unsupported package type '${packageType}' of '${arg}'")
+  endif()
+
+  set(${outArgs}
+      ${out}
+      PARENT_SCOPE
+  )
+endfunction()
+
+# Check that the working directory for a git repo is clean
+function(cpm_check_git_working_dir_is_clean repoPath gitTag isClean)
+
+  find_package(Git REQUIRED)
+
+  if(NOT GIT_EXECUTABLE)
+    # No git executable, assume directory is clean
+    set(${isClean}
+        TRUE
+        PARENT_SCOPE
+    )
+    return()
+  endif()
+
+  # check for uncommitted changes
+  execute_process(
+    COMMAND ${GIT_EXECUTABLE} status --porcelain
+    RESULT_VARIABLE resultGitStatus
+    OUTPUT_VARIABLE repoStatus
+    OUTPUT_STRIP_TRAILING_WHITESPACE ERROR_QUIET
+    WORKING_DIRECTORY ${repoPath}
+  )
+  if(resultGitStatus)
+    # not supposed to happen, assume clean anyway
+    message(WARNING "${CPM_INDENT} Calling git status on folder ${repoPath} failed")
+    set(${isClean}
+        TRUE
+        PARENT_SCOPE
+    )
+    return()
+  endif()
+
+  if(NOT "${repoStatus}" STREQUAL "")
+    set(${isClean}
+        FALSE
+        PARENT_SCOPE
+    )
+    return()
+  endif()
+
+  # check for committed changes
+  execute_process(
+    COMMAND ${GIT_EXECUTABLE} diff -s --exit-code ${gitTag}
+    RESULT_VARIABLE resultGitDiff
+    OUTPUT_STRIP_TRAILING_WHITESPACE OUTPUT_QUIET
+    WORKING_DIRECTORY ${repoPath}
+  )
+
+  if(${resultGitDiff} EQUAL 0)
+    set(${isClean}
+        TRUE
+        PARENT_SCOPE
+    )
+  else()
+    set(${isClean}
+        FALSE
+        PARENT_SCOPE
+    )
+  endif()
+
+endfunction()
+
+# Add PATCH_COMMAND to CPM_ARGS_UNPARSED_ARGUMENTS. This method consumes a list of files in ARGN
+# then generates a `PATCH_COMMAND` appropriate for `ExternalProject_Add()`. This command is appended
+# to the parent scope's `CPM_ARGS_UNPARSED_ARGUMENTS`.
+function(cpm_add_patches)
+  # Return if no patch files are supplied.
+  if(NOT ARGN)
+    return()
+  endif()
+
+  # Find the patch program.
+  find_program(PATCH_EXECUTABLE patch)
+  if(WIN32 AND NOT PATCH_EXECUTABLE)
+    # The Windows git executable is distributed with patch.exe. Find the path to the executable, if
+    # it exists, then search `../usr/bin` and `../../usr/bin` for patch.exe.
+    find_package(Git QUIET)
+    if(GIT_EXECUTABLE)
+      get_filename_component(extra_search_path ${GIT_EXECUTABLE} DIRECTORY)
+      get_filename_component(extra_search_path_1up ${extra_search_path} DIRECTORY)
+      get_filename_component(extra_search_path_2up ${extra_search_path_1up} DIRECTORY)
+      find_program(
+        PATCH_EXECUTABLE patch HINTS "${extra_search_path_1up}/usr/bin"
+                                     "${extra_search_path_2up}/usr/bin"
+      )
+    endif()
+  endif()
+  if(NOT PATCH_EXECUTABLE)
+    message(FATAL_ERROR "Couldn't find `patch` executable to use with PATCHES keyword.")
+  endif()
+
+  # Create a temporary
+  set(temp_list ${CPM_ARGS_UNPARSED_ARGUMENTS})
+
+  # Ensure each file exists (or error out) and add it to the list.
+  set(first_item True)
+  foreach(PATCH_FILE ${ARGN})
+    # Make sure the patch file exists, if we can't find it, try again in the current directory.
+    if(NOT EXISTS "${PATCH_FILE}")
+      if(NOT EXISTS "${CMAKE_CURRENT_LIST_DIR}/${PATCH_FILE}")
+        message(FATAL_ERROR "Couldn't find patch file: '${PATCH_FILE}'")
+      endif()
+      set(PATCH_FILE "${CMAKE_CURRENT_LIST_DIR}/${PATCH_FILE}")
+    endif()
+
+    # Convert to absolute path for use with patch file command.
+    get_filename_component(PATCH_FILE "${PATCH_FILE}" ABSOLUTE)
+
+    # The first patch entry must be preceded by "PATCH_COMMAND" while the following items are
+    # preceded by "&&".
+    if(first_item)
+      set(first_item False)
+      list(APPEND temp_list "PATCH_COMMAND")
+    else()
+      list(APPEND temp_list "&&")
+    endif()
+    # Add the patch command to the list
+    list(APPEND temp_list "${PATCH_EXECUTABLE}" "-p1" "<" "${PATCH_FILE}")
+  endforeach()
+
+  # Move temp out into parent scope.
+  set(CPM_ARGS_UNPARSED_ARGUMENTS
+      ${temp_list}
+      PARENT_SCOPE
+  )
+
+endfunction()
+
+# method to overwrite internal FetchContent properties, to allow using CPM.cmake to overload
+# FetchContent calls. As these are internal cmake properties, this method should be used carefully
+# and may need modification in future CMake versions. Source:
+# https://github.com/Kitware/CMake/blob/dc3d0b5a0a7d26d43d6cfeb511e224533b5d188f/Modules/FetchContent.cmake#L1152
+function(cpm_override_fetchcontent contentName)
+  cmake_parse_arguments(PARSE_ARGV 1 arg "" "SOURCE_DIR;BINARY_DIR" "")
+  if(NOT "${arg_UNPARSED_ARGUMENTS}" STREQUAL "")
+    message(FATAL_ERROR "${CPM_INDENT} Unsupported arguments: ${arg_UNPARSED_ARGUMENTS}")
+  endif()
+
+  string(TOLOWER ${contentName} contentNameLower)
+  set(prefix "_FetchContent_${contentNameLower}")
+
+  set(propertyName "${prefix}_sourceDir")
+  define_property(
+    GLOBAL
+    PROPERTY ${propertyName}
+    BRIEF_DOCS "Internal implementation detail of FetchContent_Populate()"
+    FULL_DOCS "Details used by FetchContent_Populate() for ${contentName}"
+  )
+  set_property(GLOBAL PROPERTY ${propertyName} "${arg_SOURCE_DIR}")
+
+  set(propertyName "${prefix}_binaryDir")
+  define_property(
+    GLOBAL
+    PROPERTY ${propertyName}
+    BRIEF_DOCS "Internal implementation detail of FetchContent_Populate()"
+    FULL_DOCS "Details used by FetchContent_Populate() for ${contentName}"
+  )
+  set_property(GLOBAL PROPERTY ${propertyName} "${arg_BINARY_DIR}")
+
+  set(propertyName "${prefix}_populated")
+  define_property(
+    GLOBAL
+    PROPERTY ${propertyName}
+    BRIEF_DOCS "Internal implementation detail of FetchContent_Populate()"
+    FULL_DOCS "Details used by FetchContent_Populate() for ${contentName}"
+  )
+  set_property(GLOBAL PROPERTY ${propertyName} TRUE)
+endfunction()
+
+# Download and add a package from source
+function(CPMAddPackage)
+  cpm_set_policies()
+
+  list(LENGTH ARGN argnLength)
+  if(argnLength EQUAL 1)
+    cpm_parse_add_package_single_arg("${ARGN}" ARGN)
+
+    # The shorthand syntax implies EXCLUDE_FROM_ALL and SYSTEM
+    set(ARGN "${ARGN};EXCLUDE_FROM_ALL;YES;SYSTEM;YES;")
+  endif()
+
+  set(oneValueArgs
+      NAME
+      FORCE
+      VERSION
+      GIT_TAG
+      DOWNLOAD_ONLY
+      GITHUB_REPOSITORY
+      GITLAB_REPOSITORY
+      BITBUCKET_REPOSITORY
+      GIT_REPOSITORY
+      SOURCE_DIR
+      FIND_PACKAGE_ARGUMENTS
+      NO_CACHE
+      SYSTEM
+      GIT_SHALLOW
+      EXCLUDE_FROM_ALL
+      SOURCE_SUBDIR
+      CUSTOM_CACHE_KEY
+  )
+
+  set(multiValueArgs URL OPTIONS DOWNLOAD_COMMAND PATCHES)
+
+  cmake_parse_arguments(CPM_ARGS "" "${oneValueArgs}" "${multiValueArgs}" "${ARGN}")
+
+  # Set default values for arguments
+
+  if(NOT DEFINED CPM_ARGS_VERSION)
+    if(DEFINED CPM_ARGS_GIT_TAG)
+      cpm_get_version_from_git_tag("${CPM_ARGS_GIT_TAG}" CPM_ARGS_VERSION)
+    endif()
+  endif()
+
+  if(CPM_ARGS_DOWNLOAD_ONLY)
+    set(DOWNLOAD_ONLY ${CPM_ARGS_DOWNLOAD_ONLY})
+  else()
+    set(DOWNLOAD_ONLY NO)
+  endif()
+
+  if(DEFINED CPM_ARGS_GITHUB_REPOSITORY)
+    set(CPM_ARGS_GIT_REPOSITORY "https://github.com/${CPM_ARGS_GITHUB_REPOSITORY}.git")
+  elseif(DEFINED CPM_ARGS_GITLAB_REPOSITORY)
+    set(CPM_ARGS_GIT_REPOSITORY "https://gitlab.com/${CPM_ARGS_GITLAB_REPOSITORY}.git")
+  elseif(DEFINED CPM_ARGS_BITBUCKET_REPOSITORY)
+    set(CPM_ARGS_GIT_REPOSITORY "https://bitbucket.org/${CPM_ARGS_BITBUCKET_REPOSITORY}.git")
+  endif()
+
+  if(DEFINED CPM_ARGS_GIT_REPOSITORY)
+    list(APPEND CPM_ARGS_UNPARSED_ARGUMENTS GIT_REPOSITORY ${CPM_ARGS_GIT_REPOSITORY})
+    if(NOT DEFINED CPM_ARGS_GIT_TAG)
+      set(CPM_ARGS_GIT_TAG v${CPM_ARGS_VERSION})
+    endif()
+
+    # If a name wasn't provided, try to infer it from the git repo
+    if(NOT DEFINED CPM_ARGS_NAME)
+      cpm_package_name_from_git_uri(${CPM_ARGS_GIT_REPOSITORY} CPM_ARGS_NAME)
+    endif()
+  endif()
+
+  set(CPM_SKIP_FETCH FALSE)
+
+  if(DEFINED CPM_ARGS_GIT_TAG)
+    list(APPEND CPM_ARGS_UNPARSED_ARGUMENTS GIT_TAG ${CPM_ARGS_GIT_TAG})
+    # If GIT_SHALLOW is explicitly specified, honor the value.
+    if(DEFINED CPM_ARGS_GIT_SHALLOW)
+      list(APPEND CPM_ARGS_UNPARSED_ARGUMENTS GIT_SHALLOW ${CPM_ARGS_GIT_SHALLOW})
+    endif()
+  endif()
+
+  if(DEFINED CPM_ARGS_URL)
+    # If a name or version aren't provided, try to infer them from the URL
+    list(GET CPM_ARGS_URL 0 firstUrl)
+    cpm_package_name_and_ver_from_url(${firstUrl} nameFromUrl verFromUrl)
+    # If we fail to obtain name and version from the first URL, we could try other URLs if any.
+    # However multiple URLs are expected to be quite rare, so for now we won't bother.
+
+    # If the caller provided their own name and version, they trump the inferred ones.
+    if(NOT DEFINED CPM_ARGS_NAME)
+      set(CPM_ARGS_NAME ${nameFromUrl})
+    endif()
+    if(NOT DEFINED CPM_ARGS_VERSION)
+      set(CPM_ARGS_VERSION ${verFromUrl})
+    endif()
+
+    list(APPEND CPM_ARGS_UNPARSED_ARGUMENTS URL "${CPM_ARGS_URL}")
+  endif()
+
+  # Check for required arguments
+
+  if(NOT DEFINED CPM_ARGS_NAME)
+    message(
+      FATAL_ERROR
+        "${CPM_INDENT} 'NAME' was not provided and couldn't be automatically inferred for package added with arguments: '${ARGN}'"
+    )
+  endif()
+
+  # Check if package has been added before
+  cpm_check_if_package_already_added(${CPM_ARGS_NAME} "${CPM_ARGS_VERSION}")
+  if(CPM_PACKAGE_ALREADY_ADDED)
+    cpm_export_variables(${CPM_ARGS_NAME})
+    return()
+  endif()
+
+  # Check for manual overrides
+  if(NOT CPM_ARGS_FORCE AND NOT "${CPM_${CPM_ARGS_NAME}_SOURCE}" STREQUAL "")
+    set(PACKAGE_SOURCE ${CPM_${CPM_ARGS_NAME}_SOURCE})
+    set(CPM_${CPM_ARGS_NAME}_SOURCE "")
+    CPMAddPackage(
+      NAME "${CPM_ARGS_NAME}"
+      SOURCE_DIR "${PACKAGE_SOURCE}"
+      EXCLUDE_FROM_ALL "${CPM_ARGS_EXCLUDE_FROM_ALL}"
+      SYSTEM "${CPM_ARGS_SYSTEM}"
+      PATCHES "${CPM_ARGS_PATCHES}"
+      OPTIONS "${CPM_ARGS_OPTIONS}"
+      SOURCE_SUBDIR "${CPM_ARGS_SOURCE_SUBDIR}"
+      DOWNLOAD_ONLY "${DOWNLOAD_ONLY}"
+      FORCE True
+    )
+    cpm_export_variables(${CPM_ARGS_NAME})
+    return()
+  endif()
+
+  # Check for available declaration
+  if(NOT CPM_ARGS_FORCE AND NOT "${CPM_DECLARATION_${CPM_ARGS_NAME}}" STREQUAL "")
+    set(declaration ${CPM_DECLARATION_${CPM_ARGS_NAME}})
+    set(CPM_DECLARATION_${CPM_ARGS_NAME} "")
+    CPMAddPackage(${declaration})
+    cpm_export_variables(${CPM_ARGS_NAME})
+    # checking again to ensure version and option compatibility
+    cpm_check_if_package_already_added(${CPM_ARGS_NAME} "${CPM_ARGS_VERSION}")
+    return()
+  endif()
+
+  if(NOT CPM_ARGS_FORCE)
+    if(CPM_USE_LOCAL_PACKAGES OR CPM_LOCAL_PACKAGES_ONLY)
+      cpm_find_package(${CPM_ARGS_NAME} "${CPM_ARGS_VERSION}" ${CPM_ARGS_FIND_PACKAGE_ARGUMENTS})
+
+      if(CPM_PACKAGE_FOUND)
+        cpm_export_variables(${CPM_ARGS_NAME})
+        return()
+      endif()
+
+      if(CPM_LOCAL_PACKAGES_ONLY)
+        message(
+          SEND_ERROR
+            "${CPM_INDENT} ${CPM_ARGS_NAME} not found via find_package(${CPM_ARGS_NAME} ${CPM_ARGS_VERSION})"
+        )
+      endif()
+    endif()
+  endif()
+
+  CPMRegisterPackage("${CPM_ARGS_NAME}" "${CPM_ARGS_VERSION}")
+
+  if(DEFINED CPM_ARGS_GIT_TAG)
+    set(PACKAGE_INFO "${CPM_ARGS_GIT_TAG}")
+  elseif(DEFINED CPM_ARGS_SOURCE_DIR)
+    set(PACKAGE_INFO "${CPM_ARGS_SOURCE_DIR}")
+  else()
+    set(PACKAGE_INFO "${CPM_ARGS_VERSION}")
+  endif()
+
+  if(DEFINED FETCHCONTENT_BASE_DIR)
+    # respect user's FETCHCONTENT_BASE_DIR if set
+    set(CPM_FETCHCONTENT_BASE_DIR ${FETCHCONTENT_BASE_DIR})
+  else()
+    set(CPM_FETCHCONTENT_BASE_DIR ${CMAKE_BINARY_DIR}/_deps)
+  endif()
+
+  cpm_add_patches(${CPM_ARGS_PATCHES})
+
+  if(DEFINED CPM_ARGS_DOWNLOAD_COMMAND)
+    list(APPEND CPM_ARGS_UNPARSED_ARGUMENTS DOWNLOAD_COMMAND ${CPM_ARGS_DOWNLOAD_COMMAND})
+  elseif(DEFINED CPM_ARGS_SOURCE_DIR)
+    list(APPEND CPM_ARGS_UNPARSED_ARGUMENTS SOURCE_DIR ${CPM_ARGS_SOURCE_DIR})
+    if(NOT IS_ABSOLUTE ${CPM_ARGS_SOURCE_DIR})
+      # Expand `CPM_ARGS_SOURCE_DIR` relative path. This is important because EXISTS doesn't work
+      # for relative paths.
+      get_filename_component(
+        source_directory ${CPM_ARGS_SOURCE_DIR} REALPATH BASE_DIR ${CMAKE_CURRENT_BINARY_DIR}
+      )
+    else()
+      set(source_directory ${CPM_ARGS_SOURCE_DIR})
+    endif()
+    if(NOT EXISTS ${source_directory})
+      string(TOLOWER ${CPM_ARGS_NAME} lower_case_name)
+      # remove timestamps so CMake will re-download the dependency
+      file(REMOVE_RECURSE "${CPM_FETCHCONTENT_BASE_DIR}/${lower_case_name}-subbuild")
+    endif()
+  elseif(CPM_SOURCE_CACHE AND NOT CPM_ARGS_NO_CACHE)
+    string(TOLOWER ${CPM_ARGS_NAME} lower_case_name)
+    set(origin_parameters ${CPM_ARGS_UNPARSED_ARGUMENTS})
+    list(SORT origin_parameters)
+    if(CPM_ARGS_CUSTOM_CACHE_KEY)
+      # Application set a custom unique directory name
+      set(download_directory ${CPM_SOURCE_CACHE}/${lower_case_name}/${CPM_ARGS_CUSTOM_CACHE_KEY})
+    elseif(CPM_USE_NAMED_CACHE_DIRECTORIES)
+      string(SHA1 origin_hash "${origin_parameters};NEW_CACHE_STRUCTURE_TAG")
+      set(download_directory ${CPM_SOURCE_CACHE}/${lower_case_name}/${origin_hash}/${CPM_ARGS_NAME})
+    else()
+      string(SHA1 origin_hash "${origin_parameters}")
+      set(download_directory ${CPM_SOURCE_CACHE}/${lower_case_name}/${origin_hash})
+    endif()
+    # Expand `download_directory` relative path. This is important because EXISTS doesn't work for
+    # relative paths.
+    get_filename_component(download_directory ${download_directory} ABSOLUTE)
+    list(APPEND CPM_ARGS_UNPARSED_ARGUMENTS SOURCE_DIR ${download_directory})
+
+    if(CPM_SOURCE_CACHE)
+      file(LOCK ${download_directory}/../cmake.lock)
+    endif()
+
+    if(EXISTS ${download_directory})
+      if(CPM_SOURCE_CACHE)
+        file(LOCK ${download_directory}/../cmake.lock RELEASE)
+      endif()
+
+      cpm_store_fetch_properties(
+        ${CPM_ARGS_NAME} "${download_directory}"
+        "${CPM_FETCHCONTENT_BASE_DIR}/${lower_case_name}-build"
+      )
+      cpm_get_fetch_properties("${CPM_ARGS_NAME}")
+
+      if(DEFINED CPM_ARGS_GIT_TAG AND NOT (PATCH_COMMAND IN_LIST CPM_ARGS_UNPARSED_ARGUMENTS))
+        # warn if cache has been changed since checkout
+        cpm_check_git_working_dir_is_clean(${download_directory} ${CPM_ARGS_GIT_TAG} IS_CLEAN)
+        if(NOT ${IS_CLEAN})
+          message(
+            WARNING "${CPM_INDENT} Cache for ${CPM_ARGS_NAME} (${download_directory}) is dirty"
+          )
+        endif()
+      endif()
+
+      cpm_add_subdirectory(
+        "${CPM_ARGS_NAME}"
+        "${DOWNLOAD_ONLY}"
+        "${${CPM_ARGS_NAME}_SOURCE_DIR}/${CPM_ARGS_SOURCE_SUBDIR}"
+        "${${CPM_ARGS_NAME}_BINARY_DIR}"
+        "${CPM_ARGS_EXCLUDE_FROM_ALL}"
+        "${CPM_ARGS_SYSTEM}"
+        "${CPM_ARGS_OPTIONS}"
+      )
+      set(PACKAGE_INFO "${PACKAGE_INFO} at ${download_directory}")
+
+      # As the source dir is already cached/populated, we override the call to FetchContent.
+      set(CPM_SKIP_FETCH TRUE)
+      cpm_override_fetchcontent(
+        "${lower_case_name}" SOURCE_DIR "${${CPM_ARGS_NAME}_SOURCE_DIR}/${CPM_ARGS_SOURCE_SUBDIR}"
+        BINARY_DIR "${${CPM_ARGS_NAME}_BINARY_DIR}"
+      )
+
+    else()
+      # Enable shallow clone when GIT_TAG is not a commit hash. Our guess may not be accurate, but
+      # it should guarantee no commit hash get mis-detected.
+      if(NOT DEFINED CPM_ARGS_GIT_SHALLOW)
+        cpm_is_git_tag_commit_hash("${CPM_ARGS_GIT_TAG}" IS_HASH)
+        if(NOT ${IS_HASH})
+          list(APPEND CPM_ARGS_UNPARSED_ARGUMENTS GIT_SHALLOW TRUE)
+        endif()
+      endif()
+
+      # remove timestamps so CMake will re-download the dependency
+      file(REMOVE_RECURSE ${CPM_FETCHCONTENT_BASE_DIR}/${lower_case_name}-subbuild)
+      set(PACKAGE_INFO "${PACKAGE_INFO} to ${download_directory}")
+    endif()
+  endif()
+
+  cpm_create_module_file(${CPM_ARGS_NAME} "CPMAddPackage(\"${ARGN}\")")
+
+  if(CPM_PACKAGE_LOCK_ENABLED)
+    if((CPM_ARGS_VERSION AND NOT CPM_ARGS_SOURCE_DIR) OR CPM_INCLUDE_ALL_IN_PACKAGE_LOCK)
+      cpm_add_to_package_lock(${CPM_ARGS_NAME} "${ARGN}")
+    elseif(CPM_ARGS_SOURCE_DIR)
+      cpm_add_comment_to_package_lock(${CPM_ARGS_NAME} "local directory")
+    else()
+      cpm_add_comment_to_package_lock(${CPM_ARGS_NAME} "${ARGN}")
+    endif()
+  endif()
+
+  cpm_message(
+    STATUS "${CPM_INDENT} Adding package ${CPM_ARGS_NAME}@${CPM_ARGS_VERSION} (${PACKAGE_INFO})"
+  )
+
+  if(NOT CPM_SKIP_FETCH)
+    # CMake 3.28 added EXCLUDE, SYSTEM (3.25), and SOURCE_SUBDIR (3.18) to FetchContent_Declare.
+    # Calling FetchContent_MakeAvailable will then internally forward these options to
+    # add_subdirectory. Up until these changes, we had to call FetchContent_Populate and
+    # add_subdirectory separately, which is no longer necessary and has been deprecated as of 3.30.
+    set(fetchContentDeclareExtraArgs "")
+    if(${CMAKE_VERSION} VERSION_GREATER_EQUAL "3.28.0")
+      if(${CPM_ARGS_EXCLUDE_FROM_ALL})
+        list(APPEND fetchContentDeclareExtraArgs EXCLUDE_FROM_ALL)
+      endif()
+      if(${CPM_ARGS_SYSTEM})
+        list(APPEND fetchContentDeclareExtraArgs SYSTEM)
+      endif()
+      if(DEFINED CPM_ARGS_SOURCE_SUBDIR)
+        list(APPEND fetchContentDeclareExtraArgs SOURCE_SUBDIR ${CPM_ARGS_SOURCE_SUBDIR})
+      endif()
+      # For CMake version <3.28 OPTIONS are parsed in cpm_add_subdirectory
+      if(CPM_ARGS_OPTIONS AND NOT DOWNLOAD_ONLY)
+        foreach(OPTION ${CPM_ARGS_OPTIONS})
+          cpm_parse_option("${OPTION}")
+          set(${OPTION_KEY} "${OPTION_VALUE}")
+        endforeach()
+      endif()
+    endif()
+    cpm_declare_fetch(
+      "${CPM_ARGS_NAME}" ${fetchContentDeclareExtraArgs} "${CPM_ARGS_UNPARSED_ARGUMENTS}"
+    )
+
+    cpm_fetch_package("${CPM_ARGS_NAME}" ${DOWNLOAD_ONLY} populated ${CPM_ARGS_UNPARSED_ARGUMENTS})
+    if(CPM_SOURCE_CACHE AND download_directory)
+      file(LOCK ${download_directory}/../cmake.lock RELEASE)
+    endif()
+    if(${populated} AND ${CMAKE_VERSION} VERSION_LESS "3.28.0")
+      cpm_add_subdirectory(
+        "${CPM_ARGS_NAME}"
+        "${DOWNLOAD_ONLY}"
+        "${${CPM_ARGS_NAME}_SOURCE_DIR}/${CPM_ARGS_SOURCE_SUBDIR}"
+        "${${CPM_ARGS_NAME}_BINARY_DIR}"
+        "${CPM_ARGS_EXCLUDE_FROM_ALL}"
+        "${CPM_ARGS_SYSTEM}"
+        "${CPM_ARGS_OPTIONS}"
+      )
+    endif()
+    cpm_get_fetch_properties("${CPM_ARGS_NAME}")
+  endif()
+
+  set(${CPM_ARGS_NAME}_ADDED YES)
+  cpm_export_variables("${CPM_ARGS_NAME}")
+endfunction()
+
+# Fetch a previously declared package
+macro(CPMGetPackage Name)
+  if(DEFINED "CPM_DECLARATION_${Name}")
+    CPMAddPackage(NAME ${Name})
+  else()
+    message(SEND_ERROR "${CPM_INDENT} Cannot retrieve package ${Name}: no declaration available")
+  endif()
+endmacro()
+
+# export variables available to the caller to the parent scope expects ${CPM_ARGS_NAME} to be set
+macro(cpm_export_variables name)
+  set(${name}_SOURCE_DIR
+      "${${name}_SOURCE_DIR}"
+      PARENT_SCOPE
+  )
+  set(${name}_BINARY_DIR
+      "${${name}_BINARY_DIR}"
+      PARENT_SCOPE
+  )
+  set(${name}_ADDED
+      "${${name}_ADDED}"
+      PARENT_SCOPE
+  )
+  set(CPM_LAST_PACKAGE_NAME
+      "${name}"
+      PARENT_SCOPE
+  )
+endmacro()
+
+# declares a package, so that any call to CPMAddPackage for the package name will use these
+# arguments instead. Previous declarations will not be overridden.
+macro(CPMDeclarePackage Name)
+  if(NOT DEFINED "CPM_DECLARATION_${Name}")
+    set("CPM_DECLARATION_${Name}" "${ARGN}")
+  endif()
+endmacro()
+
+function(cpm_add_to_package_lock Name)
+  if(NOT CPM_DONT_CREATE_PACKAGE_LOCK)
+    cpm_prettify_package_arguments(PRETTY_ARGN false ${ARGN})
+    file(APPEND ${CPM_PACKAGE_LOCK_FILE} "# ${Name}\nCPMDeclarePackage(${Name}\n${PRETTY_ARGN})\n")
+  endif()
+endfunction()
+
+function(cpm_add_comment_to_package_lock Name)
+  if(NOT CPM_DONT_CREATE_PACKAGE_LOCK)
+    cpm_prettify_package_arguments(PRETTY_ARGN true ${ARGN})
+    file(APPEND ${CPM_PACKAGE_LOCK_FILE}
+         "# ${Name} (unversioned)\n# CPMDeclarePackage(${Name}\n${PRETTY_ARGN}#)\n"
+    )
+  endif()
+endfunction()
+
+# includes the package lock file if it exists and creates a target `cpm-update-package-lock` to
+# update it
+macro(CPMUsePackageLock file)
+  if(NOT CPM_DONT_CREATE_PACKAGE_LOCK)
+    get_filename_component(CPM_ABSOLUTE_PACKAGE_LOCK_PATH ${file} ABSOLUTE)
+    if(EXISTS ${CPM_ABSOLUTE_PACKAGE_LOCK_PATH})
+      include(${CPM_ABSOLUTE_PACKAGE_LOCK_PATH})
+    endif()
+    if(NOT TARGET cpm-update-package-lock)
+      add_custom_target(
+        cpm-update-package-lock COMMAND ${CMAKE_COMMAND} -E copy ${CPM_PACKAGE_LOCK_FILE}
+                                        ${CPM_ABSOLUTE_PACKAGE_LOCK_PATH}
+      )
+    endif()
+    set(CPM_PACKAGE_LOCK_ENABLED true)
+  endif()
+endmacro()
+
+# registers a package that has been added to CPM
+function(CPMRegisterPackage PACKAGE VERSION)
+  list(APPEND CPM_PACKAGES ${PACKAGE})
+  set(CPM_PACKAGES
+      ${CPM_PACKAGES}
+      CACHE INTERNAL ""
+  )
+  set("CPM_PACKAGE_${PACKAGE}_VERSION"
+      ${VERSION}
+      CACHE INTERNAL ""
+  )
+endfunction()
+
+# retrieve the current version of the package to ${OUTPUT}
+function(CPMGetPackageVersion PACKAGE OUTPUT)
+  set(${OUTPUT}
+      "${CPM_PACKAGE_${PACKAGE}_VERSION}"
+      PARENT_SCOPE
+  )
+endfunction()
+
+# declares a package in FetchContent_Declare
+function(cpm_declare_fetch PACKAGE)
+  if(${CPM_DRY_RUN})
+    cpm_message(STATUS "${CPM_INDENT} Package not declared (dry run)")
+    return()
+  endif()
+
+  FetchContent_Declare(${PACKAGE} ${ARGN})
+endfunction()
+
+# returns properties for a package previously defined by cpm_declare_fetch
+function(cpm_get_fetch_properties PACKAGE)
+  if(${CPM_DRY_RUN})
+    return()
+  endif()
+
+  set(${PACKAGE}_SOURCE_DIR
+      "${CPM_PACKAGE_${PACKAGE}_SOURCE_DIR}"
+      PARENT_SCOPE
+  )
+  set(${PACKAGE}_BINARY_DIR
+      "${CPM_PACKAGE_${PACKAGE}_BINARY_DIR}"
+      PARENT_SCOPE
+  )
+endfunction()
+
+function(cpm_store_fetch_properties PACKAGE source_dir binary_dir)
+  if(${CPM_DRY_RUN})
+    return()
+  endif()
+
+  set(CPM_PACKAGE_${PACKAGE}_SOURCE_DIR
+      "${source_dir}"
+      CACHE INTERNAL ""
+  )
+  set(CPM_PACKAGE_${PACKAGE}_BINARY_DIR
+      "${binary_dir}"
+      CACHE INTERNAL ""
+  )
+endfunction()
+
+# adds a package as a subdirectory if viable, according to provided options
+function(
+  cpm_add_subdirectory
+  PACKAGE
+  DOWNLOAD_ONLY
+  SOURCE_DIR
+  BINARY_DIR
+  EXCLUDE
+  SYSTEM
+  OPTIONS
+)
+
+  if(NOT DOWNLOAD_ONLY AND EXISTS ${SOURCE_DIR}/CMakeLists.txt)
+    set(addSubdirectoryExtraArgs "")
+    if(EXCLUDE)
+      list(APPEND addSubdirectoryExtraArgs EXCLUDE_FROM_ALL)
+    endif()
+    if("${SYSTEM}" AND "${CMAKE_VERSION}" VERSION_GREATER_EQUAL "3.25")
+      # https://cmake.org/cmake/help/latest/prop_dir/SYSTEM.html#prop_dir:SYSTEM
+      list(APPEND addSubdirectoryExtraArgs SYSTEM)
+    endif()
+    if(OPTIONS)
+      foreach(OPTION ${OPTIONS})
+        cpm_parse_option("${OPTION}")
+        set(${OPTION_KEY} "${OPTION_VALUE}")
+      endforeach()
+    endif()
+    set(CPM_OLD_INDENT "${CPM_INDENT}")
+    set(CPM_INDENT "${CPM_INDENT} ${PACKAGE}:")
+    add_subdirectory(${SOURCE_DIR} ${BINARY_DIR} ${addSubdirectoryExtraArgs})
+    set(CPM_INDENT "${CPM_OLD_INDENT}")
+  endif()
+endfunction()
+
+# downloads a previously declared package via FetchContent and exports the variables
+# `${PACKAGE}_SOURCE_DIR` and `${PACKAGE}_BINARY_DIR` to the parent scope
+function(cpm_fetch_package PACKAGE DOWNLOAD_ONLY populated)
+  set(${populated}
+      FALSE
+      PARENT_SCOPE
+  )
+  if(${CPM_DRY_RUN})
+    cpm_message(STATUS "${CPM_INDENT} Package ${PACKAGE} not fetched (dry run)")
+    return()
+  endif()
+
+  FetchContent_GetProperties(${PACKAGE})
+
+  string(TOLOWER "${PACKAGE}" lower_case_name)
+
+  if(NOT ${lower_case_name}_POPULATED)
+    if(${CMAKE_VERSION} VERSION_GREATER_EQUAL "3.28.0")
+      if(DOWNLOAD_ONLY)
+        # MakeAvailable will call add_subdirectory internally which is not what we want when
+        # DOWNLOAD_ONLY is set. Populate will only download the dependency without adding it to the
+        # build
+        FetchContent_Populate(
+          ${PACKAGE}
+          SOURCE_DIR "${CPM_FETCHCONTENT_BASE_DIR}/${lower_case_name}-src"
+          BINARY_DIR "${CPM_FETCHCONTENT_BASE_DIR}/${lower_case_name}-build"
+          SUBBUILD_DIR "${CPM_FETCHCONTENT_BASE_DIR}/${lower_case_name}-subbuild"
+          ${ARGN}
+        )
+      else()
+        FetchContent_MakeAvailable(${PACKAGE})
+      endif()
+    else()
+      FetchContent_Populate(${PACKAGE})
+    endif()
+    set(${populated}
+        TRUE
+        PARENT_SCOPE
+    )
+  endif()
+
+  cpm_store_fetch_properties(
+    ${CPM_ARGS_NAME} ${${lower_case_name}_SOURCE_DIR} ${${lower_case_name}_BINARY_DIR}
+  )
+
+  set(${PACKAGE}_SOURCE_DIR
+      ${${lower_case_name}_SOURCE_DIR}
+      PARENT_SCOPE
+  )
+  set(${PACKAGE}_BINARY_DIR
+      ${${lower_case_name}_BINARY_DIR}
+      PARENT_SCOPE
+  )
+endfunction()
+
+# splits a package option
+function(cpm_parse_option OPTION)
+  string(REGEX MATCH "^[^ ]+" OPTION_KEY "${OPTION}")
+  string(LENGTH "${OPTION}" OPTION_LENGTH)
+  string(LENGTH "${OPTION_KEY}" OPTION_KEY_LENGTH)
+  if(OPTION_KEY_LENGTH STREQUAL OPTION_LENGTH)
+    # no value for key provided, assume user wants to set option to "ON"
+    set(OPTION_VALUE "ON")
+  else()
+    math(EXPR OPTION_KEY_LENGTH "${OPTION_KEY_LENGTH}+1")
+    string(SUBSTRING "${OPTION}" "${OPTION_KEY_LENGTH}" "-1" OPTION_VALUE)
+  endif()
+  set(OPTION_KEY
+      "${OPTION_KEY}"
+      PARENT_SCOPE
+  )
+  set(OPTION_VALUE
+      "${OPTION_VALUE}"
+      PARENT_SCOPE
+  )
+endfunction()
+
+# guesses the package version from a git tag
+function(cpm_get_version_from_git_tag GIT_TAG RESULT)
+  string(LENGTH ${GIT_TAG} length)
+  if(length EQUAL 40)
+    # GIT_TAG is probably a git hash
+    set(${RESULT}
+        0
+        PARENT_SCOPE
+    )
+  else()
+    string(REGEX MATCH "v?([0123456789.]*).*" _ ${GIT_TAG})
+    set(${RESULT}
+        ${CMAKE_MATCH_1}
+        PARENT_SCOPE
+    )
+  endif()
+endfunction()
+
+# guesses if the git tag is a commit hash or an actual tag or a branch name.
+function(cpm_is_git_tag_commit_hash GIT_TAG RESULT)
+  string(LENGTH "${GIT_TAG}" length)
+  # full hash has 40 characters, and short hash has at least 7 characters.
+  if(length LESS 7 OR length GREATER 40)
+    set(${RESULT}
+        0
+        PARENT_SCOPE
+    )
+  else()
+    if(${GIT_TAG} MATCHES "^[a-fA-F0-9]+$")
+      set(${RESULT}
+          1
+          PARENT_SCOPE
+      )
+    else()
+      set(${RESULT}
+          0
+          PARENT_SCOPE
+      )
+    endif()
+  endif()
+endfunction()
+
+function(cpm_prettify_package_arguments OUT_VAR IS_IN_COMMENT)
+  set(oneValueArgs
+      NAME
+      FORCE
+      VERSION
+      GIT_TAG
+      DOWNLOAD_ONLY
+      GITHUB_REPOSITORY
+      GITLAB_REPOSITORY
+      BITBUCKET_REPOSITORY
+      GIT_REPOSITORY
+      SOURCE_DIR
+      FIND_PACKAGE_ARGUMENTS
+      NO_CACHE
+      SYSTEM
+      GIT_SHALLOW
+      EXCLUDE_FROM_ALL
+      SOURCE_SUBDIR
+  )
+  set(multiValueArgs URL OPTIONS DOWNLOAD_COMMAND)
+  cmake_parse_arguments(CPM_ARGS "" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+
+  foreach(oneArgName ${oneValueArgs})
+    if(DEFINED CPM_ARGS_${oneArgName})
+      if(${IS_IN_COMMENT})
+        string(APPEND PRETTY_OUT_VAR "#")
+      endif()
+      if(${oneArgName} STREQUAL "SOURCE_DIR")
+        string(REPLACE ${CMAKE_SOURCE_DIR} "\${CMAKE_SOURCE_DIR}" CPM_ARGS_${oneArgName}
+                       ${CPM_ARGS_${oneArgName}}
+        )
+      endif()
+      string(APPEND PRETTY_OUT_VAR "  ${oneArgName} ${CPM_ARGS_${oneArgName}}\n")
+    endif()
+  endforeach()
+  foreach(multiArgName ${multiValueArgs})
+    if(DEFINED CPM_ARGS_${multiArgName})
+      if(${IS_IN_COMMENT})
+        string(APPEND PRETTY_OUT_VAR "#")
+      endif()
+      string(APPEND PRETTY_OUT_VAR "  ${multiArgName}\n")
+      foreach(singleOption ${CPM_ARGS_${multiArgName}})
+        if(${IS_IN_COMMENT})
+          string(APPEND PRETTY_OUT_VAR "#")
+        endif()
+        string(APPEND PRETTY_OUT_VAR "    \"${singleOption}\"\n")
+      endforeach()
+    endif()
+  endforeach()
+
+  if(NOT "${CPM_ARGS_UNPARSED_ARGUMENTS}" STREQUAL "")
+    if(${IS_IN_COMMENT})
+      string(APPEND PRETTY_OUT_VAR "#")
+    endif()
+    string(APPEND PRETTY_OUT_VAR " ")
+    foreach(CPM_ARGS_UNPARSED_ARGUMENT ${CPM_ARGS_UNPARSED_ARGUMENTS})
+      string(APPEND PRETTY_OUT_VAR " ${CPM_ARGS_UNPARSED_ARGUMENT}")
+    endforeach()
+    string(APPEND PRETTY_OUT_VAR "\n")
+  endif()
+
+  set(${OUT_VAR}
+      ${PRETTY_OUT_VAR}
+      PARENT_SCOPE
+  )
+
+endfunction()

third_party/tracy/cmake/ECMFindModuleHelpers.cmake

@@ -0,0 +1,300 @@
+#.rst:
+# ECMFindModuleHelpers
+# --------------------
+#
+# Helper macros for find modules: ecm_find_package_version_check(),
+# ecm_find_package_parse_components() and
+# ecm_find_package_handle_library_components().
+#
+# ::
+#
+#   ecm_find_package_version_check(<name>)
+#
+# Prints warnings if the CMake version or the project's required CMake version
+# is older than that required by extra-cmake-modules.
+#
+# ::
+#
+#   ecm_find_package_parse_components(<name>
+#       RESULT_VAR <variable>
+#       KNOWN_COMPONENTS <component1> [<component2> [...]]
+#       [SKIP_DEPENDENCY_HANDLING])
+#
+# This macro will populate <variable> with a list of components found in
+# <name>_FIND_COMPONENTS, after checking that all those components are in the
+# list of KNOWN_COMPONENTS; if there are any unknown components, it will print
+# an error or warning (depending on the value of <name>_FIND_REQUIRED) and call
+# return().
+#
+# The order of components in <variable> is guaranteed to match the order they
+# are listed in the KNOWN_COMPONENTS argument.
+#
+# If SKIP_DEPENDENCY_HANDLING is not set, for each component the variable
+# <name>_<component>_component_deps will be checked for dependent components.
+# If <component> is listed in <name>_FIND_COMPONENTS, then all its (transitive)
+# dependencies will also be added to <variable>.
+#
+# ::
+#
+#   ecm_find_package_handle_library_components(<name>
+#       COMPONENTS <component> [<component> [...]]
+#       [SKIP_DEPENDENCY_HANDLING])
+#       [SKIP_PKG_CONFIG])
+#
+# Creates an imported library target for each component.  The operation of this
+# macro depends on the presence of a number of CMake variables.
+#
+# The <name>_<component>_lib variable should contain the name of this library,
+# and <name>_<component>_header variable should contain the name of a header
+# file associated with it (whatever relative path is normally passed to
+# '#include'). <name>_<component>_header_subdir variable can be used to specify
+# which subdirectory of the include path the headers will be found in.
+# ecm_find_package_components() will then search for the library
+# and include directory (creating appropriate cache variables) and create an
+# imported library target named <name>::<component>.
+#
+# Additional variables can be used to provide additional information:
+#
+# If SKIP_PKG_CONFIG, the <name>_<component>_pkg_config variable is set, and
+# pkg-config is found, the pkg-config module given by
+# <name>_<component>_pkg_config will be searched for and used to help locate the
+# library and header file.  It will also be used to set
+# <name>_<component>_VERSION.
+#
+# Note that if version information is found via pkg-config,
+# <name>_<component>_FIND_VERSION can be set to require a particular version
+# for each component.
+#
+# If SKIP_DEPENDENCY_HANDLING is not set, the INTERFACE_LINK_LIBRARIES property
+# of the imported target for <component> will be set to contain the imported
+# targets for the components listed in <name>_<component>_component_deps.
+# <component>_FOUND will also be set to false if any of the components in
+# <name>_<component>_component_deps are not found.  This requires the components
+# in <name>_<component>_component_deps to be listed before <component> in the
+# COMPONENTS argument.
+#
+# The following variables will be set:
+#
+# ``<name>_TARGETS``
+#   the imported targets
+# ``<name>_LIBRARIES``
+#   the found libraries
+# ``<name>_INCLUDE_DIRS``
+#   the combined required include directories for the components
+# ``<name>_DEFINITIONS``
+#   the "other" CFLAGS provided by pkg-config, if any
+# ``<name>_VERSION``
+#   the value of ``<name>_<component>_VERSION`` for the first component that
+#   has this variable set (note that components are searched for in the order
+#   they are passed to the macro), although if it is already set, it will not
+#   be altered
+#
+# Note that these variables are never cleared, so if
+# ecm_find_package_handle_library_components() is called multiple times with
+# different components (typically because of multiple find_package() calls) then
+# ``<name>_TARGETS``, for example, will contain all the targets found in any
+# call (although no duplicates).
+#
+# Since pre-1.0.0.
+
+#=============================================================================
+# Copyright 2014 Alex Merry <alex.merry@kde.org>
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#
+# 1. Redistributions of source code must retain the copyright
+#    notice, this list of conditions and the following disclaimer.
+# 2. Redistributions in binary form must reproduce the copyright
+#    notice, this list of conditions and the following disclaimer in the
+#    documentation and/or other materials provided with the distribution.
+# 3. The name of the author may not be used to endorse or promote products
+#    derived from this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.
+
+macro(ecm_find_package_version_check module_name)
+    if(CMAKE_VERSION VERSION_LESS 2.8.12)
+        message(FATAL_ERROR "CMake 2.8.12 is required by Find${module_name}.cmake")
+    endif()
+    if(CMAKE_MINIMUM_REQUIRED_VERSION VERSION_LESS 2.8.12)
+        message(AUTHOR_WARNING "Your project should require at least CMake 2.8.12 to use Find${module_name}.cmake")
+    endif()
+endmacro()
+
+macro(ecm_find_package_parse_components module_name)
+    set(ecm_fppc_options SKIP_DEPENDENCY_HANDLING)
+    set(ecm_fppc_oneValueArgs RESULT_VAR)
+    set(ecm_fppc_multiValueArgs KNOWN_COMPONENTS DEFAULT_COMPONENTS)
+    cmake_parse_arguments(ECM_FPPC "${ecm_fppc_options}" "${ecm_fppc_oneValueArgs}" "${ecm_fppc_multiValueArgs}" ${ARGN})
+
+    if(ECM_FPPC_UNPARSED_ARGUMENTS)
+        message(FATAL_ERROR "Unexpected arguments to ecm_find_package_parse_components: ${ECM_FPPC_UNPARSED_ARGUMENTS}")
+    endif()
+    if(NOT ECM_FPPC_RESULT_VAR)
+        message(FATAL_ERROR "Missing RESULT_VAR argument to ecm_find_package_parse_components")
+    endif()
+    if(NOT ECM_FPPC_KNOWN_COMPONENTS)
+        message(FATAL_ERROR "Missing KNOWN_COMPONENTS argument to ecm_find_package_parse_components")
+    endif()
+    if(NOT ECM_FPPC_DEFAULT_COMPONENTS)
+        set(ECM_FPPC_DEFAULT_COMPONENTS ${ECM_FPPC_KNOWN_COMPONENTS})
+    endif()
+
+    if(${module_name}_FIND_COMPONENTS)
+        set(ecm_fppc_requestedComps ${${module_name}_FIND_COMPONENTS})
+
+        if(NOT ECM_FPPC_SKIP_DEPENDENCY_HANDLING)
+            # Make sure deps are included
+            foreach(ecm_fppc_comp ${ecm_fppc_requestedComps})
+                foreach(ecm_fppc_dep_comp ${${module_name}_${ecm_fppc_comp}_component_deps})
+                    list(FIND ecm_fppc_requestedComps "${ecm_fppc_dep_comp}" ecm_fppc_index)
+                    if("${ecm_fppc_index}" STREQUAL "-1")
+                        if(NOT ${module_name}_FIND_QUIETLY)
+                            message(STATUS "${module_name}: ${ecm_fppc_comp} requires ${${module_name}_${ecm_fppc_comp}_component_deps}")
+                        endif()
+                        list(APPEND ecm_fppc_requestedComps "${ecm_fppc_dep_comp}")
+                    endif()
+                endforeach()
+            endforeach()
+        else()
+            message(STATUS "Skipping dependency handling for ${module_name}")
+        endif()
+        list(REMOVE_DUPLICATES ecm_fppc_requestedComps)
+
+        # This makes sure components are listed in the same order as
+        # KNOWN_COMPONENTS (potentially important for inter-dependencies)
+        set(${ECM_FPPC_RESULT_VAR})
+        foreach(ecm_fppc_comp ${ECM_FPPC_KNOWN_COMPONENTS})
+            list(FIND ecm_fppc_requestedComps "${ecm_fppc_comp}" ecm_fppc_index)
+            if(NOT "${ecm_fppc_index}" STREQUAL "-1")
+                list(APPEND ${ECM_FPPC_RESULT_VAR} "${ecm_fppc_comp}")
+                list(REMOVE_AT ecm_fppc_requestedComps ${ecm_fppc_index})
+            endif()
+        endforeach()
+        # if there are any left, they are unknown components
+        if(ecm_fppc_requestedComps)
+            set(ecm_fppc_msgType STATUS)
+            if(${module_name}_FIND_REQUIRED)
+                set(ecm_fppc_msgType FATAL_ERROR)
+            endif()
+            if(NOT ${module_name}_FIND_QUIETLY)
+                message(${ecm_fppc_msgType} "${module_name}: requested unknown components ${ecm_fppc_requestedComps}")
+            endif()
+            return()
+        endif()
+    else()
+        set(${ECM_FPPC_RESULT_VAR} ${ECM_FPPC_DEFAULT_COMPONENTS})
+    endif()
+endmacro()
+
+macro(ecm_find_package_handle_library_components module_name)
+    set(ecm_fpwc_options SKIP_PKG_CONFIG SKIP_DEPENDENCY_HANDLING)
+    set(ecm_fpwc_oneValueArgs)
+    set(ecm_fpwc_multiValueArgs COMPONENTS)
+    cmake_parse_arguments(ECM_FPWC "${ecm_fpwc_options}" "${ecm_fpwc_oneValueArgs}" "${ecm_fpwc_multiValueArgs}" ${ARGN})
+
+    if(ECM_FPWC_UNPARSED_ARGUMENTS)
+        message(FATAL_ERROR "Unexpected arguments to ecm_find_package_handle_components: ${ECM_FPWC_UNPARSED_ARGUMENTS}")
+    endif()
+    if(NOT ECM_FPWC_COMPONENTS)
+        message(FATAL_ERROR "Missing COMPONENTS argument to ecm_find_package_handle_components")
+    endif()
+
+    include(FindPackageHandleStandardArgs)
+    find_package(PkgConfig QUIET)
+    foreach(ecm_fpwc_comp ${ECM_FPWC_COMPONENTS})
+        set(ecm_fpwc_dep_vars)
+        set(ecm_fpwc_dep_targets)
+        if(NOT SKIP_DEPENDENCY_HANDLING)
+            foreach(ecm_fpwc_dep ${${module_name}_${ecm_fpwc_comp}_component_deps})
+                list(APPEND ecm_fpwc_dep_vars "${module_name}_${ecm_fpwc_dep}_FOUND")
+                list(APPEND ecm_fpwc_dep_targets "${module_name}::${ecm_fpwc_dep}")
+            endforeach()
+        endif()
+
+        if(NOT ECM_FPWC_SKIP_PKG_CONFIG AND ${module_name}_${ecm_fpwc_comp}_pkg_config)
+            pkg_check_modules(PKG_${module_name}_${ecm_fpwc_comp} QUIET
+                              ${${module_name}_${ecm_fpwc_comp}_pkg_config})
+        endif()
+
+        find_path(${module_name}_${ecm_fpwc_comp}_INCLUDE_DIR
+            NAMES ${${module_name}_${ecm_fpwc_comp}_header}
+            HINTS ${PKG_${module_name}_${ecm_fpwc_comp}_INCLUDE_DIRS}
+            PATH_SUFFIXES ${${module_name}_${ecm_fpwc_comp}_header_subdir}
+        )
+        find_library(${module_name}_${ecm_fpwc_comp}_LIBRARY
+            NAMES ${${module_name}_${ecm_fpwc_comp}_lib}
+            HINTS ${PKG_${module_name}_${ecm_fpwc_comp}_LIBRARY_DIRS}
+        )
+
+        set(${module_name}_${ecm_fpwc_comp}_VERSION "${PKG_${module_name}_${ecm_fpwc_comp}_VERSION}")
+        if(NOT ${module_name}_VERSION)
+            set(${module_name}_VERSION ${${module_name}_${ecm_fpwc_comp}_VERSION})
+        endif()
+
+        set(_name_mismatched_arg)
+        if(NOT CMAKE_VERSION VERSION_LESS 3.17)
+            set(_name_mismatched_arg NAME_MISMATCHED)
+        endif()
+        find_package_handle_standard_args(${module_name}_${ecm_fpwc_comp}
+            FOUND_VAR
+                ${module_name}_${ecm_fpwc_comp}_FOUND
+            REQUIRED_VARS
+                ${module_name}_${ecm_fpwc_comp}_LIBRARY
+                ${module_name}_${ecm_fpwc_comp}_INCLUDE_DIR
+                ${ecm_fpwc_dep_vars}
+            VERSION_VAR
+                ${module_name}_${ecm_fpwc_comp}_VERSION
+            ${_name_mismatched_arg}
+            )
+
+        mark_as_advanced(
+            ${module_name}_${ecm_fpwc_comp}_LIBRARY
+            ${module_name}_${ecm_fpwc_comp}_INCLUDE_DIR
+        )
+
+        if(${module_name}_${ecm_fpwc_comp}_FOUND)
+            list(APPEND ${module_name}_LIBRARIES
+                        "${${module_name}_${ecm_fpwc_comp}_LIBRARY}")
+            list(APPEND ${module_name}_INCLUDE_DIRS
+                        "${${module_name}_${ecm_fpwc_comp}_INCLUDE_DIR}")
+            set(${module_name}_DEFINITIONS
+                    ${${module_name}_DEFINITIONS}
+                    ${PKG_${module_name}_${ecm_fpwc_comp}_DEFINITIONS})
+            if(NOT TARGET ${module_name}::${ecm_fpwc_comp})
+                add_library(${module_name}::${ecm_fpwc_comp} UNKNOWN IMPORTED)
+                set_target_properties(${module_name}::${ecm_fpwc_comp} PROPERTIES
+                    IMPORTED_LOCATION "${${module_name}_${ecm_fpwc_comp}_LIBRARY}"
+                    INTERFACE_COMPILE_OPTIONS "${PKG_${module_name}_${ecm_fpwc_comp}_DEFINITIONS}"
+                    INTERFACE_INCLUDE_DIRECTORIES "${${module_name}_${ecm_fpwc_comp}_INCLUDE_DIR}"
+                    INTERFACE_LINK_LIBRARIES "${ecm_fpwc_dep_targets}"
+                )
+            endif()
+            list(APPEND ${module_name}_TARGETS
+                        "${module_name}::${ecm_fpwc_comp}")
+        endif()
+    endforeach()
+    if(${module_name}_LIBRARIES)
+        list(REMOVE_DUPLICATES ${module_name}_LIBRARIES)
+    endif()
+    if(${module_name}_INCLUDE_DIRS)
+        list(REMOVE_DUPLICATES ${module_name}_INCLUDE_DIRS)
+    endif()
+    if(${module_name}_DEFINITIONS)
+        list(REMOVE_DUPLICATES ${module_name}_DEFINITIONS)
+    endif()
+    if(${module_name}_TARGETS)
+        list(REMOVE_DUPLICATES ${module_name}_TARGETS)
+    endif()
+endmacro()

third_party/tracy/cmake/FindWaylandScanner.cmake

@@ -0,0 +1,170 @@
+#.rst:
+# FindWaylandScanner
+# ------------------
+#
+# Try to find wayland-scanner.
+#
+# If the wayland-scanner executable is not in your PATH, you can provide
+# an alternative name or full path location with the ``WaylandScanner_EXECUTABLE``
+# variable.
+#
+# This will define the following variables:
+#
+# ``WaylandScanner_FOUND``
+#     True if wayland-scanner is available.
+#
+# ``WaylandScanner_EXECUTABLE``
+#     The wayland-scanner executable.
+#
+# If ``WaylandScanner_FOUND`` is TRUE, it will also define the following imported
+# target:
+#
+# ``Wayland::Scanner``
+#     The wayland-scanner executable.
+#
+# This module provides the following functions to generate C protocol
+# implementations:
+#
+#   - ``ecm_add_wayland_client_protocol``
+#   - ``ecm_add_wayland_server_protocol``
+#
+# ::
+#
+#   ecm_add_wayland_client_protocol(<source_files_var>
+#                                   PROTOCOL <xmlfile>
+#                                   BASENAME <basename>)
+#
+# Generate Wayland client protocol files from ``<xmlfile>`` XML
+# definition for the ``<basename>`` interface and append those files
+# to ``<source_files_var>``.
+#
+# ::
+#
+#   ecm_add_wayland_server_protocol(<source_files_var>
+#                                   PROTOCOL <xmlfile>
+#                                   BASENAME <basename>)
+#
+# Generate Wayland server protocol files from ``<xmlfile>`` XML
+# definition for the ``<basename>`` interface and append those files
+# to ``<source_files_var>``.
+#
+# Since 1.4.0.
+
+#=============================================================================
+# Copyright 2012-2014 Pier Luigi Fiorini <pierluigi.fiorini@gmail.com>
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions
+# are met:
+#
+# 1. Redistributions of source code must retain the copyright
+#    notice, this list of conditions and the following disclaimer.
+# 2. Redistributions in binary form must reproduce the copyright
+#    notice, this list of conditions and the following disclaimer in the
+#    documentation and/or other materials provided with the distribution.
+# 3. The name of the author may not be used to endorse or promote products
+#    derived from this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.
+#=============================================================================
+
+include(${CMAKE_CURRENT_LIST_DIR}/ECMFindModuleHelpers.cmake)
+
+ecm_find_package_version_check(WaylandScanner)
+
+# Find wayland-scanner
+find_program(WaylandScanner_EXECUTABLE NAMES wayland-scanner)
+
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(WaylandScanner
+    FOUND_VAR
+        WaylandScanner_FOUND
+    REQUIRED_VARS
+        WaylandScanner_EXECUTABLE
+)
+
+mark_as_advanced(WaylandScanner_EXECUTABLE)
+
+if(NOT TARGET Wayland::Scanner AND WaylandScanner_FOUND)
+    add_executable(Wayland::Scanner IMPORTED)
+    set_target_properties(Wayland::Scanner PROPERTIES
+        IMPORTED_LOCATION "${WaylandScanner_EXECUTABLE}"
+    )
+endif()
+
+include(FeatureSummary)
+set_package_properties(WaylandScanner PROPERTIES
+    URL "https://wayland.freedesktop.org/"
+    DESCRIPTION "Executable that converts XML protocol files to C code"
+)
+
+function(ecm_add_wayland_client_protocol out_var)
+    # Parse arguments
+    set(oneValueArgs PROTOCOL BASENAME)
+    cmake_parse_arguments(ARGS "" "${oneValueArgs}" "" ${ARGN})
+
+    if(ARGS_UNPARSED_ARGUMENTS)
+        message(FATAL_ERROR "Unknown keywords given to ecm_add_wayland_client_protocol(): \"${ARGS_UNPARSED_ARGUMENTS}\"")
+    endif()
+
+    get_filename_component(_infile ${ARGS_PROTOCOL} ABSOLUTE)
+    set(_client_header "${CMAKE_CURRENT_BINARY_DIR}/wayland-${ARGS_BASENAME}-client-protocol.h")
+    set(_code "${CMAKE_CURRENT_BINARY_DIR}/wayland-${ARGS_BASENAME}-protocol.c")
+
+    set_source_files_properties(${_client_header} GENERATED)
+    set_source_files_properties(${_code} GENERATED)
+    set_property(SOURCE ${_client_header} PROPERTY SKIP_AUTOMOC ON)
+
+    add_custom_command(OUTPUT "${_client_header}"
+        COMMAND ${WaylandScanner_EXECUTABLE} client-header ${_infile} ${_client_header}
+        DEPENDS ${WaylandScanner_EXECUTABLE} ${_infile}
+        VERBATIM
+    )
+
+    add_custom_command(OUTPUT "${_code}"
+        COMMAND ${WaylandScanner_EXECUTABLE} private-code ${_infile} ${_code}
+        DEPENDS ${WaylandScanner_EXECUTABLE} ${_infile} ${_client_header}
+        VERBATIM
+    )
+
+    list(APPEND ${out_var} "${_client_header}" "${_code}")
+    set(${out_var} ${${out_var}} PARENT_SCOPE)
+endfunction()
+
+
+function(ecm_add_wayland_server_protocol out_var)
+    # Parse arguments
+    set(oneValueArgs PROTOCOL BASENAME)
+    cmake_parse_arguments(ARGS "" "${oneValueArgs}" "" ${ARGN})
+
+    if(ARGS_UNPARSED_ARGUMENTS)
+        message(FATAL_ERROR "Unknown keywords given to ecm_add_wayland_server_protocol(): \"${ARGS_UNPARSED_ARGUMENTS}\"")
+    endif()
+
+    ecm_add_wayland_client_protocol(${out_var}
+                                    PROTOCOL ${ARGS_PROTOCOL}
+                                    BASENAME ${ARGS_BASENAME})
+
+    get_filename_component(_infile ${ARGS_PROTOCOL} ABSOLUTE)
+    set(_server_header "${CMAKE_CURRENT_BINARY_DIR}/wayland-${ARGS_BASENAME}-server-protocol.h")
+    set_property(SOURCE ${_server_header} PROPERTY SKIP_AUTOMOC ON)
+    set_source_files_properties(${_server_header} GENERATED)
+
+    add_custom_command(OUTPUT "${_server_header}"
+        COMMAND ${WaylandScanner_EXECUTABLE} server-header ${_infile} ${_server_header}
+        DEPENDS ${WaylandScanner_EXECUTABLE} ${_infile}
+        VERBATIM
+   )
+
+    list(APPEND ${out_var} "${_server_header}")
+    set(${out_var} ${${out_var}} PARENT_SCOPE)
+endfunction()

third_party/tracy/cmake/config.cmake

@@ -0,0 +1,56 @@
+if (NOT NO_ISA_EXTENSIONS)
+    include(CheckCXXCompilerFlag)
+    if (CMAKE_SYSTEM_PROCESSOR MATCHES "aarch64" OR CMAKE_SYSTEM_PROCESSOR MATCHES "arm64")
+        CHECK_CXX_COMPILER_FLAG("-mcpu=native" COMPILER_SUPPORTS_MCPU_NATIVE)
+        if(COMPILER_SUPPORTS_MARCH_NATIVE)
+            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mcpu=native")
+            set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -mcpu=native")
+        endif()
+    else()
+        CHECK_CXX_COMPILER_FLAG("-march=native" COMPILER_SUPPORTS_MARCH_NATIVE)
+        if(COMPILER_SUPPORTS_MARCH_NATIVE)
+            set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native")
+            set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -march=native")
+        endif()
+    endif()
+    if(WIN32)
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX2")
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /arch:AVX2")
+    endif()
+endif()
+
+if(CMAKE_SYSTEM_NAME STREQUAL "Linux" AND NOT LEGACY)
+    set(USE_WAYLAND ON)
+else()
+    set(USE_WAYLAND OFF)
+endif()
+
+if(WIN32)
+    add_definitions(-DNOMINMAX -DWIN32_LEAN_AND_MEAN)
+    add_compile_options(/MP)
+else()
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fdiagnostics-color=always")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fdiagnostics-color=always")
+endif()
+
+if(NOT CMAKE_BUILD_TYPE STREQUAL "Debug" AND NOT EMSCRIPTEN)
+    set(CMAKE_INTERPROCEDURAL_OPTIMIZATION ON)
+endif()
+
+if(EMSCRIPTEN)
+    add_compile_options(-pthread)
+    add_link_options(-pthread)
+endif()
+
+if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" AND NOT EMSCRIPTEN)
+    find_program(MOLD_LINKER mold)
+    if(MOLD_LINKER)
+        set(CMAKE_LINKER_TYPE "MOLD")
+    endif()
+    if (CMAKE_BUILD_TYPE STREQUAL "Debug")
+        set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-eliminate-unused-debug-types")
+        set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fno-eliminate-unused-debug-types")
+    endif()
+endif()
+
+file(GENERATE OUTPUT .gitignore CONTENT "*")

third_party/tracy/cmake/server.cmake

@@ -0,0 +1,39 @@
+set(TRACY_COMMON_DIR ${CMAKE_CURRENT_LIST_DIR}/../public/common)
+
+set(TRACY_COMMON_SOURCES
+    tracy_lz4.cpp
+    tracy_lz4hc.cpp
+    TracySocket.cpp
+    TracyStackFrames.cpp
+    TracySystem.cpp
+)
+
+list(TRANSFORM TRACY_COMMON_SOURCES PREPEND "${TRACY_COMMON_DIR}/")
+
+
+set(TRACY_SERVER_DIR ${CMAKE_CURRENT_LIST_DIR}/../server)
+
+set(TRACY_SERVER_SOURCES
+    TracyMemory.cpp
+    TracyMmap.cpp
+    TracyPrint.cpp
+    TracySysUtil.cpp
+    TracyTaskDispatch.cpp
+    TracyTextureCompression.cpp
+    TracyThreadCompress.cpp
+    TracyWorker.cpp
+)
+
+list(TRANSFORM TRACY_SERVER_SOURCES PREPEND "${TRACY_SERVER_DIR}/")
+
+
+add_library(TracyServer STATIC ${TRACY_COMMON_SOURCES} ${TRACY_SERVER_SOURCES})
+target_include_directories(TracyServer PUBLIC ${TRACY_COMMON_DIR} ${TRACY_SERVER_DIR})
+target_link_libraries(TracyServer PUBLIC TracyCapstone TracyZstd)
+if(NO_STATISTICS)
+    target_compile_definitions(TracyServer PUBLIC TRACY_NO_STATISTICS)
+endif()
+
+if(NOT NO_PARALLEL_STL AND UNIX AND NOT APPLE AND NOT EMSCRIPTEN)
+    target_link_libraries(TracyServer PRIVATE TracyTbb)
+endif()

third_party/tracy/cmake/vendor.cmake

@@ -0,0 +1,238 @@
+# Vendor Specific CMake
+# The Tracy project keeps most vendor source locally
+
+set (ROOT_DIR "${CMAKE_CURRENT_LIST_DIR}/../")
+
+# Dependencies are taken from the system first and if not found, they are pulled with CPM and built from source
+
+include(FindPkgConfig)
+include(${CMAKE_CURRENT_LIST_DIR}/CPM.cmake)
+
+option(DOWNLOAD_CAPSTONE "Force download capstone" ON)
+option(DOWNLOAD_GLFW "Force download glfw" OFF)
+option(DOWNLOAD_FREETYPE "Force download freetype" OFF)
+
+# capstone
+
+pkg_check_modules(CAPSTONE capstone)
+if(CAPSTONE_FOUND AND NOT DOWNLOAD_CAPSTONE)
+    message(STATUS "Capstone found: ${CAPSTONE}")
+    add_library(TracyCapstone INTERFACE)
+    target_include_directories(TracyCapstone INTERFACE ${CAPSTONE_INCLUDE_DIRS})
+    target_link_libraries(TracyCapstone INTERFACE ${CAPSTONE_LINK_LIBRARIES})
+else()
+    CPMAddPackage(
+        NAME capstone
+        GITHUB_REPOSITORY capstone-engine/capstone
+        GIT_TAG 5.0.3
+    )
+    add_library(TracyCapstone INTERFACE)
+    target_include_directories(TracyCapstone INTERFACE ${capstone_SOURCE_DIR}/include/capstone)
+    target_link_libraries(TracyCapstone INTERFACE capstone)
+endif()
+
+# GLFW
+
+if(NOT USE_WAYLAND AND NOT EMSCRIPTEN)
+    pkg_check_modules(GLFW glfw3)
+    if (GLFW_FOUND AND NOT DOWNLOAD_GLFW)
+        add_library(TracyGlfw3 INTERFACE)
+        target_include_directories(TracyGlfw3 INTERFACE ${GLFW_INCLUDE_DIRS})
+        target_link_libraries(TracyGlfw3 INTERFACE ${GLFW_LINK_LIBRARIES})
+    else()
+        CPMAddPackage(
+            NAME glfw
+            GITHUB_REPOSITORY glfw/glfw
+            GIT_TAG 3.4
+            OPTIONS
+                "GLFW_BUILD_EXAMPLES OFF"
+                "GLFW_BUILD_TESTS OFF"
+                "GLFW_BUILD_DOCS OFF"
+                "GLFW_INSTALL OFF"
+        )
+        add_library(TracyGlfw3 INTERFACE)
+        target_link_libraries(TracyGlfw3 INTERFACE glfw)
+    endif()
+endif()
+
+# freetype
+
+pkg_check_modules(FREETYPE freetype2)
+if (FREETYPE_FOUND AND NOT DOWNLOAD_FREETYPE)
+    add_library(TracyFreetype INTERFACE)
+    target_include_directories(TracyFreetype INTERFACE ${FREETYPE_INCLUDE_DIRS})
+    target_link_libraries(TracyFreetype INTERFACE ${FREETYPE_LINK_LIBRARIES})
+else()
+    CPMAddPackage(
+        NAME freetype
+        GITHUB_REPOSITORY freetype/freetype
+        GIT_TAG VER-2-13-2
+        OPTIONS
+            "FT_DISABLE_HARFBUZZ ON"
+            "FT_WITH_HARFBUZZ OFF"
+    )
+    add_library(TracyFreetype INTERFACE)
+    target_link_libraries(TracyFreetype INTERFACE freetype)
+endif()
+
+# zstd
+
+set(ZSTD_DIR "${ROOT_DIR}/zstd")
+
+set(ZSTD_SOURCES
+    decompress/zstd_ddict.c
+    decompress/zstd_decompress_block.c
+    decompress/huf_decompress.c
+    decompress/zstd_decompress.c
+    common/zstd_common.c
+    common/error_private.c
+    common/xxhash.c
+    common/entropy_common.c
+    common/debug.c
+    common/threading.c
+    common/pool.c
+    common/fse_decompress.c
+    compress/zstd_ldm.c
+    compress/zstd_compress_superblock.c
+    compress/zstd_opt.c
+    compress/zstd_compress_sequences.c
+    compress/fse_compress.c
+    compress/zstd_double_fast.c
+    compress/zstd_compress.c
+    compress/zstd_compress_literals.c
+    compress/hist.c
+    compress/zstdmt_compress.c
+    compress/zstd_lazy.c
+    compress/huf_compress.c
+    compress/zstd_fast.c
+    dictBuilder/zdict.c
+    dictBuilder/cover.c
+    dictBuilder/divsufsort.c
+    dictBuilder/fastcover.c
+)
+
+list(TRANSFORM ZSTD_SOURCES PREPEND "${ZSTD_DIR}/")
+
+set_property(SOURCE ${ZSTD_DIR}/decompress/huf_decompress_amd64.S APPEND PROPERTY COMPILE_OPTIONS "-x" "assembler-with-cpp")
+
+add_library(TracyZstd STATIC ${ZSTD_SOURCES})
+target_include_directories(TracyZstd PUBLIC ${ZSTD_DIR})
+target_compile_definitions(TracyZstd PRIVATE ZSTD_DISABLE_ASM)
+
+
+# Diff Template Library
+
+set(DTL_DIR "${ROOT_DIR}/dtl")
+file(GLOB_RECURSE DTL_HEADERS CONFIGURE_DEPENDS RELATIVE ${DTL_DIR} "*.hpp")
+add_library(TracyDtl INTERFACE)
+target_sources(TracyDtl INTERFACE ${DTL_HEADERS})
+target_include_directories(TracyDtl INTERFACE ${DTL_DIR})
+
+
+# Get Opt
+
+set(GETOPT_DIR "${ROOT_DIR}/getopt")
+set(GETOPT_SOURCES ${GETOPT_DIR}/getopt.c)
+set(GETOPT_HEADERS ${GETOPT_DIR}/getopt.h)
+add_library(TracyGetOpt STATIC ${GETOPT_SOURCES} ${GETOPT_HEADERS})
+target_include_directories(TracyGetOpt PUBLIC ${GETOPT_DIR})
+
+
+# ImGui
+
+set(IMGUI_DIR "${ROOT_DIR}/imgui")
+
+set(IMGUI_SOURCES
+    imgui_widgets.cpp
+    imgui_draw.cpp
+    imgui_demo.cpp
+    imgui.cpp
+    imgui_tables.cpp
+    misc/freetype/imgui_freetype.cpp
+)
+
+list(TRANSFORM IMGUI_SOURCES PREPEND "${IMGUI_DIR}/")
+
+add_definitions(-DIMGUI_ENABLE_FREETYPE)
+
+add_library(TracyImGui STATIC ${IMGUI_SOURCES})
+target_include_directories(TracyImGui PUBLIC ${IMGUI_DIR})
+target_link_libraries(TracyImGui PUBLIC TracyFreetype)
+
+# NFD
+
+if (NOT NO_FILESELECTOR AND NOT EMSCRIPTEN)
+    set(NFD_DIR "${ROOT_DIR}/nfd")
+
+    if (WIN32)
+        set(NFD_SOURCES "${NFD_DIR}/nfd_win.cpp")
+    elseif (APPLE)
+        set(NFD_SOURCES "${NFD_DIR}/nfd_cocoa.m")
+    else()
+        if (GTK_FILESELECTOR)
+            set(NFD_SOURCES "${NFD_DIR}/nfd_gtk.cpp")
+        else()
+            set(NFD_SOURCES "${NFD_DIR}/nfd_portal.cpp")
+        endif()
+    endif()
+
+    file(GLOB_RECURSE NFD_HEADERS CONFIGURE_DEPENDS RELATIVE ${NFD_DIR} "*.h")
+    add_library(TracyNfd STATIC ${NFD_SOURCES} ${NFD_HEADERS})
+    target_include_directories(TracyNfd PUBLIC ${NFD_DIR})
+
+    if (APPLE)
+        find_library(APPKIT_LIBRARY AppKit)
+        find_library(UNIFORMTYPEIDENTIFIERS_LIBRARY UniformTypeIdentifiers)
+        target_link_libraries(TracyNfd PUBLIC ${APPKIT_LIBRARY} ${UNIFORMTYPEIDENTIFIERS_LIBRARY})
+    elseif (UNIX)
+        if (GTK_FILESELECTOR)
+            pkg_check_modules(GTK3 gtk+-3.0)
+            if (NOT GTK3_FOUND)
+                message(FATAL_ERROR "GTK3 not found. Please install it or set TRACY_GTK_FILESELECTOR to OFF.")
+            endif()
+            add_library(TracyGtk3 INTERFACE)
+            target_include_directories(TracyGtk3 INTERFACE ${GTK3_INCLUDE_DIRS})
+            target_link_libraries(TracyGtk3 INTERFACE ${GTK3_LINK_LIBRARIES})
+            target_link_libraries(TracyNfd PUBLIC TracyGtk3)
+        else()
+            pkg_check_modules(DBUS dbus-1)
+            if (NOT DBUS_FOUND)
+                message(FATAL_ERROR "D-Bus not found. Please install it or set TRACY_GTK_FILESELECTOR to ON.")
+            endif()
+            add_library(TracyDbus INTERFACE)
+            target_include_directories(TracyDbus INTERFACE ${DBUS_INCLUDE_DIRS})
+            target_link_libraries(TracyDbus INTERFACE ${DBUS_LINK_LIBRARIES})
+            target_link_libraries(TracyNfd PUBLIC TracyDbus)
+        endif()
+    endif()
+endif()
+
+# TBB
+if (NO_PARALLEL_STL)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DNO_PARALLEL_SORT")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -DNO_PARALLEL_SORT")
+else()
+    if (UNIX AND NOT APPLE AND NOT EMSCRIPTEN)
+        # Tracy does not use TBB directly, but the implementation of parallel algorithms
+        # in some versions of libstdc++ depends on TBB. When it does, you must
+        # explicitly link against -ltbb.
+        #
+        # Some distributions have pgk-config files for TBB, others don't.
+
+        pkg_check_modules(TBB tbb)
+        if (TBB_FOUND)
+            add_library(TracyTbb INTERFACE)
+            target_include_directories(TracyTbb INTERFACE ${TBB_INCLUDE_DIRS})
+            target_link_libraries(TracyTbb INTERFACE ${TBB_LINK_LIBRARIES})
+        else()
+            CPMAddPackage(
+                NAME tbb
+                GITHUB_REPOSITORY oneapi-src/oneTBB
+                GIT_TAG v2021.12.0-rc2
+                OPTIONS "TBB_TEST OFF"
+            )
+            add_library(TracyTbb INTERFACE)
+            target_link_libraries(TracyTbb INTERFACE tbb)
+        endif()
+    endif()
+endif()

third_party/tracy/cmake/version.cmake

@@ -0,0 +1,24 @@
+cmake_minimum_required(VERSION 3.10)
+
+message("Parsing public/common/TracyVersion.hpp file")
+
+file(READ "${CMAKE_CURRENT_LIST_DIR}/../public/common/TracyVersion.hpp" version)
+
+# Note: This looks for a specific pattern in TracyVersion.hpp, if it changes
+# this needs updating.
+string(REGEX MATCH "Major = ([0-9]+)" _ ${version})
+
+# This works do to the above () subexpression selection. See
+# https://cmake.org/cmake/help/latest/command/string.html#regex-match for more
+# details
+set(TRACY_VERSION_MAJOR ${CMAKE_MATCH_1})
+
+string(REGEX MATCH "Minor = ([0-9]+)" _ ${version})
+set(TRACY_VERSION_MINOR ${CMAKE_MATCH_1})
+
+string(REGEX MATCH "Patch = ([0-9]+)" _ ${version})
+set(TRACY_VERSION_PATCH ${CMAKE_MATCH_1})
+
+set(TRACY_VERSION_STRING "${TRACY_VERSION_MAJOR}.${TRACY_VERSION_MINOR}.${TRACY_VERSION_PATCH}")
+
+message("VERSION ${TRACY_VERSION_STRING}")

third_party/tracy/csvexport/CMakeLists.txt

@@ -0,0 +1,28 @@
+cmake_minimum_required(VERSION 3.16)
+
+option(NO_ISA_EXTENSIONS "Disable ISA extensions (don't pass -march=native or -mcpu=native to the compiler)" OFF)
+option(NO_PARALLEL_STL "Disable parallel STL" OFF)
+
+set(NO_STATISTICS OFF)
+
+include(${CMAKE_CURRENT_LIST_DIR}/../cmake/version.cmake)
+
+set(CMAKE_CXX_STANDARD 20)
+
+project(
+    tracy-csvexport
+    LANGUAGES C CXX
+    VERSION ${TRACY_VERSION_STRING}
+)
+
+include(${CMAKE_CURRENT_LIST_DIR}/../cmake/config.cmake)
+include(${CMAKE_CURRENT_LIST_DIR}/../cmake/vendor.cmake)
+include(${CMAKE_CURRENT_LIST_DIR}/../cmake/server.cmake)
+
+set(PROGRAM_FILES
+    src/csvexport.cpp
+)
+
+add_executable(${PROJECT_NAME} ${PROGRAM_FILES} ${COMMON_FILES} ${SERVER_FILES})
+target_link_libraries(${PROJECT_NAME} PRIVATE TracyServer TracyGetOpt)
+set_property(DIRECTORY ${CMAKE_CURRENT_LIST_DIR} PROPERTY VS_STARTUP_PROJECT ${PROJECT_NAME})

third_party/tracy/csvexport/src/csvexport.cpp

@@ -0,0 +1,353 @@
+#ifdef _WIN32
+#  include <windows.h>
+#endif
+
+#include <algorithm>
+#include <cctype>
+#include <fstream>
+#include <iostream>
+#include <sstream>
+#include <string>
+
+#include <math.h>
+#include <stdio.h>
+#include <stdint.h>
+
+#include "../../server/TracyFileRead.hpp"
+#include "../../server/TracyWorker.hpp"
+#include "../../getopt/getopt.h"
+
+void print_usage_exit(int e)
+{
+    fprintf(stderr, "Extract statistics from a trace to a CSV format\n");
+    fprintf(stderr, "Usage:\n");
+    fprintf(stderr, "  extract [OPTION...] <trace file>\n");
+    fprintf(stderr, "\n");
+    fprintf(stderr, "  -h, --help        Print usage\n");
+    fprintf(stderr, "  -f, --filter arg  Filter zone names (default: "")\n");
+    fprintf(stderr, "  -s, --sep arg     CSV separator (default: ,)\n");
+    fprintf(stderr, "  -c, --case        Case sensitive filtering\n");
+    fprintf(stderr, "  -e, --self        Get self times\n");
+    fprintf(stderr, "  -u, --unwrap      Report each zone event\n");
+    fprintf(stderr, "  -m, --messages    Report only messages\n");
+
+    exit(e);
+}
+
+struct Args {
+    const char* filter;
+    const char* separator;
+    const char* trace_file;
+    bool case_sensitive;
+    bool self_time;
+    bool unwrap;
+    bool unwrapMessages;
+};
+
+Args parse_args(int argc, char** argv)
+{
+    if (argc == 1)
+    {
+        print_usage_exit(1);
+    }
+
+    Args args = { "", ",", "", false, false, false, false };
+
+    struct option long_opts[] = {
+        { "help", no_argument, NULL, 'h' },
+        { "filter", optional_argument, NULL, 'f' },
+        { "sep", optional_argument, NULL, 's' },
+        { "case", no_argument, NULL, 'c' },
+        { "self", no_argument, NULL, 'e' },
+        { "unwrap", no_argument, NULL, 'u' },
+        { "messages", no_argument, NULL, 'm' },
+        { NULL, 0, NULL, 0 }
+    };
+
+    int c;
+    while ((c = getopt_long(argc, argv, "hf:s:ceum", long_opts, NULL)) != -1)
+    {
+        switch (c)
+        {
+        case 'h':
+            print_usage_exit(0);
+            break;
+        case 'f':
+            args.filter = optarg;
+            break;
+        case 's':
+            args.separator = optarg;
+            break;
+        case 'c':
+            args.case_sensitive = true;
+            break;
+        case 'e':
+            args.self_time = true;
+            break;
+        case 'u':
+            args.unwrap = true;
+            break;
+        case 'm':
+            args.unwrapMessages = true;
+            break;
+        default:
+            print_usage_exit(1);
+            break;
+        }
+    }
+
+    if (argc != optind + 1)
+    {
+        print_usage_exit(1);
+    }
+
+    args.trace_file = argv[optind];
+
+    return args;
+}
+
+bool is_substring(
+    const char* term,
+    const char* s,
+    bool case_sensitive = false
+){
+    auto new_term = std::string(term);
+    auto new_s = std::string(s);
+
+    if (!case_sensitive) {
+        std::transform(
+            new_term.begin(),
+            new_term.end(),
+            new_term.begin(),
+            [](unsigned char c){ return std::tolower(c); }
+        );
+
+        std::transform(
+            new_s.begin(),
+            new_s.end(),
+            new_s.begin(),
+            [](unsigned char c){ return std::tolower(c); }
+        );
+    }
+
+    return new_s.find(new_term) != std::string::npos;
+}
+
+const char* get_name(int32_t id, const tracy::Worker& worker)
+{
+    auto& srcloc = worker.GetSourceLocation(id);
+    return worker.GetString(srcloc.name.active ? srcloc.name : srcloc.function);
+}
+
+template <typename T>
+std::string join(const T& v, const char* sep) {
+    std::ostringstream s;
+    for (const auto& i : v) {
+        if (&i != &v[0]) {
+            s << sep;
+        }
+        s << i;
+    }
+    return s.str();
+}
+
+// From TracyView.cpp
+int64_t GetZoneChildTimeFast(
+    const tracy::Worker& worker,
+    const tracy::ZoneEvent& zone
+){
+    int64_t time = 0;
+    if( zone.HasChildren() )
+    {
+        auto& children = worker.GetZoneChildren( zone.Child() );
+        if( children.is_magic() )
+        {
+            auto& vec = *(tracy::Vector<tracy::ZoneEvent>*)&children;
+            for( auto& v : vec )
+            {
+                assert( v.IsEndValid() );
+                time += v.End() - v.Start();
+            }
+        }
+        else
+        {
+            for( auto& v : children )
+            {
+                assert( v->IsEndValid() );
+                time += v->End() - v->Start();
+            }
+        }
+    }
+    return time;
+}
+
+int main(int argc, char** argv)
+{
+#ifdef _WIN32
+    if (!AttachConsole(ATTACH_PARENT_PROCESS))
+    {
+        AllocConsole();
+        SetConsoleMode(GetStdHandle(STD_OUTPUT_HANDLE), 0x07);
+    }
+#endif
+
+    Args args = parse_args(argc, argv);
+
+    auto f = std::unique_ptr<tracy::FileRead>(
+        tracy::FileRead::Open(args.trace_file)
+    );
+    if (!f)
+    {
+        fprintf(stderr, "Could not open file %s\n", args.trace_file);
+        return 1;
+    }
+
+    auto worker = tracy::Worker(*f);
+
+    if (args.unwrapMessages) 
+    {
+        const auto& msgs = worker.GetMessages();
+    
+        if (msgs.size() > 0)
+        {
+            std::vector<const char*> columnsForMessages;
+            columnsForMessages = {
+                    "MessageName", "total_ns"
+                };
+            std::string headerForMessages = join(columnsForMessages, args.separator);
+            printf("%s\n", headerForMessages.data());
+
+            for(auto& it : msgs)
+            {
+                std::vector<std::string> values(columnsForMessages.size());
+
+                values[0] = worker.GetString(it->ref);
+                values[1] = std::to_string(it->time);
+
+                std::string row = join(values, args.separator);
+                printf("%s\n", row.data());
+            }
+        }
+        else
+        {
+            printf("There are currently no messages!\n");
+        }
+    
+        return 0;
+    }
+
+    while (!worker.AreSourceLocationZonesReady())
+    {
+        std::this_thread::sleep_for(std::chrono::milliseconds(10));
+    }
+
+    auto& slz = worker.GetSourceLocationZones();
+    tracy::Vector<decltype(slz.begin())> slz_selected;
+    slz_selected.reserve(slz.size());
+
+    uint32_t total_cnt = 0;
+    for(auto it = slz.begin(); it != slz.end(); ++it)
+    {
+        if(it->second.total != 0)
+        {
+            ++total_cnt;
+            if(args.filter[0] == '\0')
+            {
+                slz_selected.push_back_no_space_check(it);
+            }
+            else
+            {
+                auto name = get_name(it->first, worker);
+                if(is_substring(args.filter, name, args.case_sensitive))
+                {
+                    slz_selected.push_back_no_space_check(it);
+                }
+            }
+        }
+    }
+
+    std::vector<const char*> columns;
+    if (args.unwrap)
+    {
+        columns = {
+            "name", "src_file", "src_line", "ns_since_start", "exec_time_ns", "thread"
+        };
+    }
+    else
+    {
+        columns = {
+            "name", "src_file", "src_line", "total_ns", "total_perc",
+            "counts", "mean_ns", "min_ns", "max_ns", "std_ns"
+        };
+    }
+    std::string header = join(columns, args.separator);
+    printf("%s\n", header.data());
+
+    const auto last_time = worker.GetLastTime();
+    for(auto& it : slz_selected)
+    {
+        std::vector<std::string> values(columns.size());
+
+        values[0] = get_name(it->first, worker);
+
+        const auto& srcloc = worker.GetSourceLocation(it->first);
+        values[1] = worker.GetString(srcloc.file);
+        values[2] = std::to_string(srcloc.line);
+
+        const auto& zone_data = it->second;
+
+        if (args.unwrap)
+        {
+            int i = 0;
+            for (const auto& zone_thread_data : zone_data.zones) {
+                const auto zone_event = zone_thread_data.Zone();
+                const auto tId = zone_thread_data.Thread();
+                const auto start = zone_event->Start();
+                const auto end = zone_event->End();
+
+                values[3] = std::to_string(start);
+
+                auto timespan = end - start;
+                if (args.self_time) {
+                    timespan -= GetZoneChildTimeFast(worker, *zone_event);
+                }
+                values[4] = std::to_string(timespan);
+                values[5] = std::to_string(tId);
+
+                std::string row = join(values, args.separator);
+                printf("%s\n", row.data());
+            }
+        }
+        else
+        {
+            const auto time = args.self_time ? zone_data.selfTotal : zone_data.total;
+            values[3] = std::to_string(time);
+            values[4] = std::to_string(100. * time / last_time);
+
+            values[5] = std::to_string(zone_data.zones.size());
+
+            const auto avg = (args.self_time ? zone_data.selfTotal : zone_data.total)
+                / zone_data.zones.size();
+            values[6] = std::to_string(avg);
+
+            const auto tmin = args.self_time ? zone_data.selfMin : zone_data.min;
+            const auto tmax = args.self_time ? zone_data.selfMax : zone_data.max;
+            values[7] = std::to_string(tmin);
+            values[8] = std::to_string(tmax);
+
+            const auto sz = zone_data.zones.size();
+            const auto ss = zone_data.sumSq
+                - 2. * zone_data.total * avg
+                + avg * avg * sz;
+            double std = 0;
+            if( sz > 1 )
+                std = sqrt(ss / (sz - 1));
+            values[9] = std::to_string(std);
+
+            std::string row = join(values, args.separator);
+            printf("%s\n", row.data());
+        }
+    }
+
+    return 0;
+}

third_party/tracy/examples/OpenCLVectorAdd/CMakeLists.txt

@@ -0,0 +1,17 @@
+cmake_minimum_required(VERSION 3.10)
+
+project(OpenCLVectorAdd)
+
+set(THREADS_PREFER_PTHREAD_FLAG ON)
+find_package(Threads REQUIRED)
+
+find_package(OpenCL REQUIRED)
+
+add_executable(OpenCLVectorAdd OpenCLVectorAdd.cpp)
+
+add_library(TracyClient STATIC ../../public/TracyClient.cpp
+                               ../../public/tracy/TracyOpenCL.hpp)
+target_include_directories(TracyClient PUBLIC ../../public/tracy)
+target_compile_definitions(TracyClient PUBLIC TRACY_ENABLE=1)
+
+target_link_libraries(OpenCLVectorAdd PUBLIC OpenCL::OpenCL TracyClient ${CMAKE_DL_LIBS} Threads::Threads)

third_party/tracy/examples/OpenCLVectorAdd/OpenCLVectorAdd.cpp

@@ -0,0 +1,220 @@
+#include <algorithm>
+#include <iostream>
+#include <cassert>
+#include <string>
+#include <vector>
+#include <numeric>
+
+#include <CL/cl.h>
+
+#include <Tracy.hpp>
+#include <TracyOpenCL.hpp>
+
+#define CL_ASSERT(err)                                              \
+    if((err) != CL_SUCCESS)                                         \
+    {                                                               \
+        std::cerr << "OpenCL Call Returned " << err << std::endl;   \
+        assert(false);                                              \
+    }
+
+const char kernelSource[] =
+"   void __kernel vectorAdd(global float* C, global float* A, global float* B, int N)  "
+"   {                                                                                  "
+"       int i = get_global_id(0);                                                      "
+"       if (i < N) {                                                                   "
+"           C[i] = A[i] + B[i];                                                        "
+"       }                                                                              "
+"   }                                                                                  ";
+
+int main()
+{
+    cl_platform_id platform;
+    cl_device_id device;
+    cl_context context;
+    cl_command_queue commandQueue;
+    cl_kernel vectorAddKernel;
+    cl_program program;
+    cl_int err;
+    cl_mem bufferA, bufferB, bufferC;
+
+    TracyCLCtx tracyCLCtx;
+
+    {
+        ZoneScopedN("OpenCL Init");
+
+        cl_uint numPlatforms = 0;
+        CL_ASSERT(clGetPlatformIDs(0, nullptr, &numPlatforms));
+
+        if (numPlatforms == 0)
+        {
+            std::cerr << "Cannot find OpenCL platform to run this application" << std::endl;
+            return 1;
+        }
+
+        CL_ASSERT(clGetPlatformIDs(1, &platform, nullptr));
+
+        size_t platformNameBufferSize = 0;
+        CL_ASSERT(clGetPlatformInfo(platform, CL_PLATFORM_NAME, 0, nullptr, &platformNameBufferSize));
+        std::string platformName(platformNameBufferSize, '\0');
+        CL_ASSERT(clGetPlatformInfo(platform, CL_PLATFORM_NAME, platformNameBufferSize, &platformName[0], nullptr));
+
+        std::cout << "OpenCL Platform: " << platformName << std::endl;
+
+        CL_ASSERT(clGetDeviceIDs(platform, CL_DEVICE_TYPE_ALL, 1, &device, nullptr));
+        size_t deviceNameBufferSize = 0;
+        CL_ASSERT(clGetDeviceInfo(device, CL_DEVICE_NAME, 0, nullptr, &deviceNameBufferSize));
+        std::string deviceName(deviceNameBufferSize, '\0');
+        CL_ASSERT(clGetDeviceInfo(device, CL_DEVICE_NAME, deviceNameBufferSize, &deviceName[0], nullptr));
+
+        std::cout << "OpenCL Device: " << deviceName << std::endl;
+
+        err = CL_SUCCESS;
+        context = clCreateContext(nullptr, 1, &device, nullptr, nullptr, &err);
+        CL_ASSERT(err);
+
+        size_t kernelSourceLength = sizeof(kernelSource);
+        const char* kernelSourceArray = { kernelSource };
+        program = clCreateProgramWithSource(context, 1, &kernelSourceArray, &kernelSourceLength, &err);
+        CL_ASSERT(err);
+
+        if (clBuildProgram(program, 1, &device, nullptr, nullptr, nullptr) != CL_SUCCESS)
+        {
+            size_t programBuildLogBufferSize = 0;
+            CL_ASSERT(clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, 0, nullptr, &programBuildLogBufferSize));
+            std::string programBuildLog(programBuildLogBufferSize, '\0');
+            CL_ASSERT(clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, programBuildLogBufferSize, &programBuildLog[0], nullptr));
+            std::clog << programBuildLog << std::endl;
+            return 1;
+        }
+
+        vectorAddKernel = clCreateKernel(program, "vectorAdd", &err);
+        CL_ASSERT(err);
+
+        commandQueue = clCreateCommandQueue(context, device, CL_QUEUE_PROFILING_ENABLE, &err);
+        CL_ASSERT(err);
+    }
+
+    tracyCLCtx = TracyCLContext(context, device);
+
+    size_t N = 10 * 1024 * 1024 / sizeof(float); // 10MB of floats
+    std::vector<float> hostA, hostB, hostC;
+
+    {
+        ZoneScopedN("Host Data Init");
+        hostA.resize(N);
+        hostB.resize(N);
+        hostC.resize(N);
+
+        std::iota(std::begin(hostA), std::end(hostA), 0.0f);
+        std::iota(std::begin(hostB), std::end(hostB), 0.0f);
+    }
+
+    {
+        ZoneScopedN("Host to Device Memory Copy");
+
+        bufferA = clCreateBuffer(context, CL_MEM_READ_WRITE, N * sizeof(float), nullptr, &err);
+        CL_ASSERT(err);
+        bufferB = clCreateBuffer(context, CL_MEM_READ_WRITE, N * sizeof(float), nullptr, &err);
+        CL_ASSERT(err);
+        bufferC = clCreateBuffer(context, CL_MEM_READ_WRITE, N * sizeof(float), nullptr, &err);
+        CL_ASSERT(err);
+
+        cl_event writeBufferAEvent, writeBufferBEvent;
+        {
+            ZoneScopedN("Write Buffer A");
+            TracyCLZoneS(tracyCLCtx, "Write BufferA", 5);
+
+            CL_ASSERT(clEnqueueWriteBuffer(commandQueue, bufferA, CL_FALSE, 0, N * sizeof(float), hostA.data(), 0, nullptr, &writeBufferAEvent));
+
+            TracyCLZoneSetEvent(writeBufferAEvent);
+        }
+        {
+            ZoneScopedN("Write Buffer B");
+            TracyCLZone(tracyCLCtx, "Write BufferB");
+
+            CL_ASSERT(clEnqueueWriteBuffer(commandQueue, bufferB, CL_FALSE, 0, N * sizeof(float), hostB.data(), 0, nullptr, &writeBufferBEvent));
+
+            TracyCLZoneSetEvent(writeBufferBEvent);
+        }
+    }
+
+    cl_int clN = static_cast<cl_int>(N);
+    const int numFrames = 10;
+    const int launchsPerFrame = 10;
+    constexpr int numLaunchs = numFrames * launchsPerFrame;
+    std::vector<cl_event> kernelLaunchEvts;
+    kernelLaunchEvts.reserve(numLaunchs);
+    for (int i = 0; i < numFrames; ++i)
+    {
+        FrameMark;
+        for (int j = 0; j < launchsPerFrame; ++j) {
+            ZoneScopedN("VectorAdd Kernel Launch");
+            TracyCLZoneC(tracyCLCtx, "VectorAdd Kernel", tracy::Color::Blue4);
+
+            CL_ASSERT(clSetKernelArg(vectorAddKernel, 0, sizeof(cl_mem), &bufferC));
+            CL_ASSERT(clSetKernelArg(vectorAddKernel, 1, sizeof(cl_mem), &bufferA));
+            CL_ASSERT(clSetKernelArg(vectorAddKernel, 2, sizeof(cl_mem), &bufferB));
+            CL_ASSERT(clSetKernelArg(vectorAddKernel, 3, sizeof(cl_int), &clN));
+
+            cl_event vectorAddKernelEvent;
+            CL_ASSERT(clEnqueueNDRangeKernel(commandQueue, vectorAddKernel, 1, nullptr, &N, nullptr, 0, nullptr, &vectorAddKernelEvent));
+            TracyCLZoneSetEvent(vectorAddKernelEvent);
+            CL_ASSERT(clRetainEvent(vectorAddKernelEvent));
+            kernelLaunchEvts.push_back(vectorAddKernelEvent);
+            std::cout << "VectorAdd Kernel Enqueued" << std::endl;
+        }
+        {
+            // Wait frame events to be finished
+            ZoneScopedN("clFinish");
+            CL_ASSERT(clFinish(commandQueue));
+        }
+        // You should collect on each 'frame' ends, so that streaming can be achieved.
+        TracyCLCollect(tracyCLCtx);
+    }
+
+    {
+        ZoneScopedN("Device to Host Memory Copy");
+        TracyCLZone(tracyCLCtx, "Read Buffer C");
+
+        cl_event readbufferCEvent;
+        CL_ASSERT(clEnqueueReadBuffer(commandQueue, bufferC, CL_TRUE, 0, N * sizeof(float), hostC.data(), 0, nullptr, &readbufferCEvent));
+        TracyCLZoneSetEvent(readbufferCEvent);
+    }
+
+    CL_ASSERT(clFinish(commandQueue));
+    std::vector<float> durations(kernelLaunchEvts.size());
+    for (int i=0; i<kernelLaunchEvts.size(); i++) {
+        cl_event evt = kernelLaunchEvts[i];
+        cl_ulong start;
+        cl_ulong end;
+        CL_ASSERT(clGetEventProfilingInfo(evt, CL_PROFILING_COMMAND_START, sizeof(cl_ulong), &start, nullptr));
+        CL_ASSERT(clGetEventProfilingInfo(evt, CL_PROFILING_COMMAND_END, sizeof(cl_ulong), &end, nullptr));
+        CL_ASSERT(clReleaseEvent(evt));
+        durations[i] = (end - start) * 0.001f;
+        std::cout << "VectorAdd Kernel " << i << " tooks " <<  static_cast<int>(durations[i]) << "us" << std::endl;
+    };
+    float avg = std::accumulate(durations.cbegin(), durations.cend(), 0.0f) / durations.size();
+    float stddev2 = std::accumulate(durations.cbegin(), durations.cend(), 0.0f, [avg](const float& acc, const float& v) {
+        auto d = v - avg;
+        return acc + d*d;
+    }) / (durations.size() - 1.0f);
+    std::cout << "VectorAdd runtime avg: " << avg << "us, std: " << sqrt(stddev2) << "us over " << numLaunchs << " runs." << std::endl;
+
+    // User should ensure all events are finished, in this case, collect after the clFinish will do the trick.
+    TracyCLCollect(tracyCLCtx);
+
+    {
+        ZoneScopedN("Checking results");
+
+        for (int i = 0; i < N; ++i)
+        {
+            assert(hostC[i] == hostA[i] + hostB[i]);
+        }
+    }
+
+    std::cout << "Results are correct!" << std::endl;
+
+    TracyCLDestroy(tracyCLCtx);
+
+    return 0;
+}

third_party/tracy/examples/ToyPathTracer/.gitignore

@@ -0,0 +1 @@
+Windows/Compiled*Shader.h

third_party/tracy/examples/ToyPathTracer/README

@@ -0,0 +1,4 @@
+https://github.com/aras-p/ToyPathTracer
+
+Modified to render only 10 frames. Client part requires 12 GB, server part
+requires 6.4 GB.

third_party/tracy/examples/ToyPathTracer/Source/Config.h

@@ -0,0 +1,33 @@
+
+#if defined(__APPLE__) && !defined(__METAL_VERSION__)
+#include <TargetConditionals.h>
+#endif
+
+#define kBackbufferWidth 1280
+#define kBackbufferHeight 720
+
+#if defined(__EMSCRIPTEN__)
+#define CPU_CAN_DO_SIMD 0
+#define CPU_CAN_DO_THREADS 0
+#else
+#define CPU_CAN_DO_SIMD 1
+#define CPU_CAN_DO_THREADS 1
+#endif
+
+
+#define DO_SAMPLES_PER_PIXEL 4
+#define DO_ANIMATE_SMOOTHING 0.9f
+#define DO_LIGHT_SAMPLING 1
+#define DO_MITSUBA_COMPARE 0
+
+// Should path tracing be done on the GPU with a compute shader?
+#define DO_COMPUTE_GPU 0
+#define kCSGroupSizeX 8
+#define kCSGroupSizeY 8
+#define kCSMaxObjects 64
+
+// Should float3 struct use SSE/NEON?
+#define DO_FLOAT3_WITH_SIMD (!(DO_COMPUTE_GPU) && CPU_CAN_DO_SIMD && 1)
+
+// Should HitSpheres function use SSE/NEON?
+#define DO_HIT_SPHERES_SIMD (CPU_CAN_DO_SIMD && 1)

third_party/tracy/examples/ToyPathTracer/Source/MathSimd.h

@@ -0,0 +1,192 @@
+#pragma once
+
+#if defined(_MSC_VER)
+#define VM_INLINE __forceinline
+#else
+#define VM_INLINE __attribute__((unused, always_inline, nodebug)) inline
+#endif
+
+#define kSimdWidth 4
+
+#if !defined(__arm__) && !defined(__arm64__) && !defined(__EMSCRIPTEN__)
+
+// ---- SSE implementation
+
+#include <xmmintrin.h>
+#include <emmintrin.h>
+#include <smmintrin.h>
+
+#define SHUFFLE4(V, X,Y,Z,W) float4(_mm_shuffle_ps((V).m, (V).m, _MM_SHUFFLE(W,Z,Y,X)))
+
+struct float4
+{
+    VM_INLINE float4() {}
+    VM_INLINE explicit float4(const float *p) { m = _mm_loadu_ps(p); }
+    VM_INLINE explicit float4(float x, float y, float z, float w) { m = _mm_set_ps(w, z, y, x); }
+    VM_INLINE explicit float4(float v) { m = _mm_set_ps1(v); }
+    VM_INLINE explicit float4(__m128 v) { m = v; }
+    
+    VM_INLINE float getX() const { return _mm_cvtss_f32(m); }
+    VM_INLINE float getY() const { return _mm_cvtss_f32(_mm_shuffle_ps(m, m, _MM_SHUFFLE(1, 1, 1, 1))); }
+    VM_INLINE float getZ() const { return _mm_cvtss_f32(_mm_shuffle_ps(m, m, _MM_SHUFFLE(2, 2, 2, 2))); }
+    VM_INLINE float getW() const { return _mm_cvtss_f32(_mm_shuffle_ps(m, m, _MM_SHUFFLE(3, 3, 3, 3))); }
+    
+    __m128 m;
+};
+
+typedef float4 bool4;
+
+VM_INLINE float4 operator+ (float4 a, float4 b) { a.m = _mm_add_ps(a.m, b.m); return a; }
+VM_INLINE float4 operator- (float4 a, float4 b) { a.m = _mm_sub_ps(a.m, b.m); return a; }
+VM_INLINE float4 operator* (float4 a, float4 b) { a.m = _mm_mul_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator==(float4 a, float4 b) { a.m = _mm_cmpeq_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator!=(float4 a, float4 b) { a.m = _mm_cmpneq_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator< (float4 a, float4 b) { a.m = _mm_cmplt_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator> (float4 a, float4 b) { a.m = _mm_cmpgt_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator<=(float4 a, float4 b) { a.m = _mm_cmple_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator>=(float4 a, float4 b) { a.m = _mm_cmpge_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator&(bool4 a, bool4 b) { a.m = _mm_and_ps(a.m, b.m); return a; }
+VM_INLINE bool4 operator|(bool4 a, bool4 b) { a.m = _mm_or_ps(a.m, b.m); return a; }
+VM_INLINE float4 operator- (float4 a) { a.m = _mm_xor_ps(a.m, _mm_set1_ps(-0.0f)); return a; }
+VM_INLINE float4 min(float4 a, float4 b) { a.m = _mm_min_ps(a.m, b.m); return a; }
+VM_INLINE float4 max(float4 a, float4 b) { a.m = _mm_max_ps(a.m, b.m); return a; }
+
+VM_INLINE float hmin(float4 v)
+{
+    v = min(v, SHUFFLE4(v, 2, 3, 0, 0));
+    v = min(v, SHUFFLE4(v, 1, 0, 0, 0));
+    return v.getX();
+}
+
+// Returns a 4-bit code where bit0..bit3 is X..W
+VM_INLINE unsigned mask(float4 v) { return _mm_movemask_ps(v.m); }
+// Once we have a comparison, we can branch based on its results:
+VM_INLINE bool any(bool4 v) { return mask(v) != 0; }
+VM_INLINE bool all(bool4 v) { return mask(v) == 15; }
+
+// "select", i.e. hibit(cond) ? b : a
+// on SSE4.1 and up this can be done easily via "blend" instruction;
+// on older SSEs has to do a bunch of hoops, see
+// https://fgiesen.wordpress.com/2016/04/03/sse-mind-the-gap/
+
+VM_INLINE float4 select(float4 a, float4 b, bool4 cond)
+{
+#if defined(__SSE4_1__) || defined(_MSC_VER) // on windows assume we always have SSE4.1
+    a.m = _mm_blendv_ps(a.m, b.m, cond.m);
+#else
+    __m128 d = _mm_castsi128_ps(_mm_srai_epi32(_mm_castps_si128(cond.m), 31));
+    a.m = _mm_or_ps(_mm_and_ps(d, b.m), _mm_andnot_ps(d, a.m));
+#endif
+    return a;
+}
+VM_INLINE __m128i select(__m128i a, __m128i b, bool4 cond)
+{
+#if defined(__SSE4_1__) || defined(_MSC_VER) // on windows assume we always have SSE4.1
+    return _mm_blendv_epi8(a, b, _mm_castps_si128(cond.m));
+#else
+    __m128i d = _mm_srai_epi32(_mm_castps_si128(cond.m), 31);
+    return _mm_or_si128(_mm_and_si128(d, b), _mm_andnot_si128(d, a));
+#endif
+}
+
+VM_INLINE float4 sqrtf(float4 v) { return float4(_mm_sqrt_ps(v.m)); }
+
+#elif !defined(__EMSCRIPTEN__)
+
+// ---- NEON implementation
+
+#define USE_NEON 1
+#include <arm_neon.h>
+
+struct float4
+{
+    VM_INLINE float4() {}
+    VM_INLINE explicit float4(const float *p) { m = vld1q_f32(p); }
+    VM_INLINE explicit float4(float x, float y, float z, float w) { float v[4] = {x, y, z, w}; m = vld1q_f32(v); }
+    VM_INLINE explicit float4(float v) { m = vdupq_n_f32(v); }
+    VM_INLINE explicit float4(float32x4_t v) { m = v; }
+    
+    VM_INLINE float getX() const { return vgetq_lane_f32(m, 0); }
+    VM_INLINE float getY() const { return vgetq_lane_f32(m, 1); }
+    VM_INLINE float getZ() const { return vgetq_lane_f32(m, 2); }
+    VM_INLINE float getW() const { return vgetq_lane_f32(m, 3); }
+    
+    float32x4_t m;
+};
+
+typedef float4 bool4;
+
+VM_INLINE float4 operator+ (float4 a, float4 b) { a.m = vaddq_f32(a.m, b.m); return a; }
+VM_INLINE float4 operator- (float4 a, float4 b) { a.m = vsubq_f32(a.m, b.m); return a; }
+VM_INLINE float4 operator* (float4 a, float4 b) { a.m = vmulq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator==(float4 a, float4 b) { a.m = vceqq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator!=(float4 a, float4 b) { a.m = a.m = vmvnq_u32(vceqq_f32(a.m, b.m)); return a; }
+VM_INLINE bool4 operator< (float4 a, float4 b) { a.m = vcltq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator> (float4 a, float4 b) { a.m = vcgtq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator<=(float4 a, float4 b) { a.m = vcleq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator>=(float4 a, float4 b) { a.m = vcgeq_f32(a.m, b.m); return a; }
+VM_INLINE bool4 operator&(bool4 a, bool4 b) { a.m = vandq_u32(a.m, b.m); return a; }
+VM_INLINE bool4 operator|(bool4 a, bool4 b) { a.m = vorrq_u32(a.m, b.m); return a; }
+VM_INLINE float4 operator- (float4 a) { a.m = vnegq_f32(a.m); return a; }
+VM_INLINE float4 min(float4 a, float4 b) { a.m = vminq_f32(a.m, b.m); return a; }
+VM_INLINE float4 max(float4 a, float4 b) { a.m = vmaxq_f32(a.m, b.m); return a; }
+
+VM_INLINE float hmin(float4 v)
+{
+    float32x2_t minOfHalfs = vpmin_f32(vget_low_f32(v.m), vget_high_f32(v.m));
+    float32x2_t minOfMinOfHalfs = vpmin_f32(minOfHalfs, minOfHalfs);
+    return vget_lane_f32(minOfMinOfHalfs, 0);
+}
+
+// Returns a 4-bit code where bit0..bit3 is X..W
+VM_INLINE unsigned mask(float4 v)
+{
+    static const uint32x4_t movemask = { 1, 2, 4, 8 };
+    static const uint32x4_t highbit = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+    uint32x4_t t0 = vreinterpretq_u32_f32(v.m);
+    uint32x4_t t1 = vtstq_u32(t0, highbit);
+    uint32x4_t t2 = vandq_u32(t1, movemask);
+    uint32x2_t t3 = vorr_u32(vget_low_u32(t2), vget_high_u32(t2));
+    return vget_lane_u32(t3, 0) | vget_lane_u32(t3, 1);
+}
+// Once we have a comparison, we can branch based on its results:
+VM_INLINE bool any(bool4 v) { return mask(v) != 0; }
+VM_INLINE bool all(bool4 v) { return mask(v) == 15; }
+
+// "select", i.e. hibit(cond) ? b : a
+// on SSE4.1 and up this can be done easily via "blend" instruction;
+// on older SSEs has to do a bunch of hoops, see
+// https://fgiesen.wordpress.com/2016/04/03/sse-mind-the-gap/
+
+VM_INLINE float4 select(float4 a, float4 b, bool4 cond)
+{
+    a.m = vbslq_f32(cond.m, b.m, a.m);
+    return a;
+}
+VM_INLINE int32x4_t select(int32x4_t a, int32x4_t b, bool4 cond)
+{
+    return vbslq_f32(cond.m, b, a);
+}
+
+VM_INLINE float4 sqrtf(float4 v)
+{
+    float32x4_t V = v.m;
+    float32x4_t S0 = vrsqrteq_f32(V);
+    float32x4_t P0 = vmulq_f32( V, S0 );
+    float32x4_t R0 = vrsqrtsq_f32( P0, S0 );
+    float32x4_t S1 = vmulq_f32( S0, R0 );
+    float32x4_t P1 = vmulq_f32( V, S1 );
+    float32x4_t R1 = vrsqrtsq_f32( P1, S1 );
+    float32x4_t S2 = vmulq_f32( S1, R1 );
+    float32x4_t P2 = vmulq_f32( V, S2 );
+    float32x4_t R2 = vrsqrtsq_f32( P2, S2 );
+    float32x4_t S3 = vmulq_f32( S2, R2 );
+    return float4(vmulq_f32(V, S3));
+}
+
+VM_INLINE float4 splatX(float32x4_t v) { return float4(vdupq_lane_f32(vget_low_f32(v), 0)); }
+VM_INLINE float4 splatY(float32x4_t v) { return float4(vdupq_lane_f32(vget_low_f32(v), 1)); }
+VM_INLINE float4 splatZ(float32x4_t v) { return float4(vdupq_lane_f32(vget_high_f32(v), 0)); }
+VM_INLINE float4 splatW(float32x4_t v) { return float4(vdupq_lane_f32(vget_high_f32(v), 1)); }
+
+#endif

third_party/tracy/examples/ToyPathTracer/Source/Maths.cpp

@@ -0,0 +1,203 @@
+#include "Maths.h"
+#include <stdlib.h>
+#include <stdint.h>
+
+static uint32_t XorShift32(uint32_t& state)
+{
+    uint32_t x = state;
+    x ^= x << 13;
+    x ^= x >> 17;
+    x ^= x << 15;
+    state = x;
+    return x;
+}
+
+float RandomFloat01(uint32_t& state)
+{
+    return (XorShift32(state) & 0xFFFFFF) / 16777216.0f;
+}
+
+float3 RandomInUnitDisk(uint32_t& state)
+{
+    float3 p;
+    do
+    {
+        p = 2.0 * float3(RandomFloat01(state),RandomFloat01(state),0) - float3(1,1,0);
+    } while (dot(p,p) >= 1.0);
+    return p;
+}
+
+float3 RandomInUnitSphere(uint32_t& state)
+{
+    float3 p;
+    do {
+        p = 2.0*float3(RandomFloat01(state),RandomFloat01(state),RandomFloat01(state)) - float3(1,1,1);
+    } while (sqLength(p) >= 1.0);
+    return p;
+}
+
+float3 RandomUnitVector(uint32_t& state)
+{
+    float z = RandomFloat01(state) * 2.0f - 1.0f;
+    float a = RandomFloat01(state) * 2.0f * kPI;
+    float r = sqrtf(1.0f - z * z);
+    float x = r * cosf(a);
+    float y = r * sinf(a);
+    return float3(x, y, z);
+}
+
+
+int HitSpheres(const Ray& r, const SpheresSoA& spheres, float tMin, float tMax, Hit& outHit)
+{
+#if DO_HIT_SPHERES_SIMD
+    float4 hitT = float4(tMax);
+#if USE_NEON
+    int32x4_t id = vdupq_n_s32(-1);
+#else
+    __m128i id = _mm_set1_epi32(-1);
+#endif
+
+#if DO_FLOAT3_WITH_SIMD && !USE_NEON
+    float4 rOrigX = SHUFFLE4(r.orig, 0, 0, 0, 0);
+    float4 rOrigY = SHUFFLE4(r.orig, 1, 1, 1, 1);
+    float4 rOrigZ = SHUFFLE4(r.orig, 2, 2, 2, 2);
+    float4 rDirX = SHUFFLE4(r.dir, 0, 0, 0, 0);
+    float4 rDirY = SHUFFLE4(r.dir, 1, 1, 1, 1);
+    float4 rDirZ = SHUFFLE4(r.dir, 2, 2, 2, 2);
+#elif DO_FLOAT3_WITH_SIMD
+    float4 rOrigX = splatX(r.orig.m);
+    float4 rOrigY = splatY(r.orig.m);
+    float4 rOrigZ = splatZ(r.orig.m);
+    float4 rDirX = splatX(r.dir.m);
+    float4 rDirY = splatY(r.dir.m);
+    float4 rDirZ = splatZ(r.dir.m);
+#else
+    float4 rOrigX = float4(r.orig.x);
+    float4 rOrigY = float4(r.orig.y);
+    float4 rOrigZ = float4(r.orig.z);
+    float4 rDirX = float4(r.dir.x);
+    float4 rDirY = float4(r.dir.y);
+    float4 rDirZ = float4(r.dir.z);
+#endif
+    float4 tMin4 = float4(tMin);
+#if USE_NEON
+    int32x4_t curId = vcombine_u32(vcreate_u32(0ULL | (1ULL<<32)), vcreate_u32(2ULL | (3ULL<<32)));
+#else
+    __m128i curId = _mm_set_epi32(3, 2, 1, 0);
+#endif
+    // process 4 spheres at once
+    for (int i = 0; i < spheres.simdCount; i += kSimdWidth)
+    {
+        // load data for 4 spheres
+        float4 sCenterX = float4(spheres.centerX + i);
+        float4 sCenterY = float4(spheres.centerY + i);
+        float4 sCenterZ = float4(spheres.centerZ + i);
+        float4 sSqRadius = float4(spheres.sqRadius + i);
+        // note: we flip this vector and calculate -b (nb) since that happens to be slightly preferable computationally
+        float4 coX = sCenterX - rOrigX;
+        float4 coY = sCenterY - rOrigY;
+        float4 coZ = sCenterZ - rOrigZ;
+        float4 nb = coX * rDirX + coY * rDirY + coZ * rDirZ;
+        float4 c = coX * coX + coY * coY + coZ * coZ - sSqRadius;
+        float4 discr = nb * nb - c;
+        bool4 discrPos = discr > float4(0.0f);
+        // if ray hits any of the 4 spheres
+        if (any(discrPos))
+        {
+            float4 discrSq = sqrtf(discr);
+
+            // ray could hit spheres at t0 & t1
+            float4 t0 = nb - discrSq;
+            float4 t1 = nb + discrSq;
+
+            float4 t = select(t1, t0, t0 > tMin4); // if t0 is above min, take it (since it's the earlier hit); else try t1.
+            bool4 msk = discrPos & (t > tMin4) & (t < hitT);
+            // if hit, take it
+            id = select(id, curId, msk);
+            hitT = select(hitT, t, msk);
+        }
+#if USE_NEON
+        curId = vaddq_s32(curId, vdupq_n_s32(kSimdWidth));
+#else
+        curId = _mm_add_epi32(curId, _mm_set1_epi32(kSimdWidth));
+#endif
+    }
+    // now we have up to 4 hits, find and return closest one
+    float minT = hmin(hitT);
+    if (minT < tMax) // any actual hits?
+    {
+        int minMask = mask(hitT == float4(minT));
+        if (minMask != 0)
+        {
+            int id_scalar[4];
+            float hitT_scalar[4];
+#if USE_NEON
+            vst1q_s32(id_scalar, id);
+            vst1q_f32(hitT_scalar, hitT.m);
+#else
+            _mm_storeu_si128((__m128i *)id_scalar, id);
+            _mm_storeu_ps(hitT_scalar, hitT.m);
+#endif
+
+            // In general, you would do this with a bit scan (first set/trailing zero count).
+            // But who cares, it's only 16 options.
+            static const int laneId[16] =
+            {
+                0, 0, 1, 0, // 00xx
+                2, 0, 1, 0, // 01xx
+                3, 0, 1, 0, // 10xx
+                2, 0, 1, 0, // 11xx
+            };
+
+            int lane = laneId[minMask];
+            int hitId = id_scalar[lane];
+            float finalHitT = hitT_scalar[lane];
+
+            outHit.pos = r.pointAt(finalHitT);
+            outHit.normal = (outHit.pos - float3(spheres.centerX[hitId], spheres.centerY[hitId], spheres.centerZ[hitId])) * spheres.invRadius[hitId];
+            outHit.t = finalHitT;
+            return hitId;
+        }
+    }
+
+    return -1;
+
+#else // #if DO_HIT_SPHERES_SIMD
+
+    float hitT = tMax;
+    int id = -1;
+    for (int i = 0; i < spheres.count; ++i)
+    {
+        float coX = spheres.centerX[i] - r.orig.getX();
+        float coY = spheres.centerY[i] - r.orig.getY();
+        float coZ = spheres.centerZ[i] - r.orig.getZ();
+        float nb = coX * r.dir.getX() + coY * r.dir.getY() + coZ * r.dir.getZ();
+        float c = coX * coX + coY * coY + coZ * coZ - spheres.sqRadius[i];
+        float discr = nb * nb - c;
+        if (discr > 0)
+        {
+            float discrSq = sqrtf(discr);
+
+            // Try earlier t
+            float t = nb - discrSq;
+            if (t <= tMin) // before min, try later t!
+                t = nb + discrSq;
+
+            if (t > tMin && t < hitT)
+            {
+                id = i;
+                hitT = t;
+            }
+        }
+    }
+    if (id != -1)
+    {
+        outHit.pos = r.pointAt(hitT);
+        outHit.normal = (outHit.pos - float3(spheres.centerX[id], spheres.centerY[id], spheres.centerZ[id])) * spheres.invRadius[id];
+        outHit.t = hitT;
+        return id;
+    }
+    else
+        return -1;
+#endif // #else of #if DO_HIT_SPHERES_SIMD
+}

third_party/tracy/examples/ToyPathTracer/Source/Maths.h

@@ -0,0 +1,436 @@
+#pragma once
+
+#include <math.h>
+#include <assert.h>
+#include <stdint.h>
+#include "Config.h"
+#include "MathSimd.h"
+
+#define kPI 3.1415926f
+
+// SSE/SIMD vector largely based on http://www.codersnotes.com/notes/maths-lib-2016/
+#if DO_FLOAT3_WITH_SIMD
+
+
+#if !defined(__arm__) && !defined(__arm64__)
+
+// ---- SSE implementation
+
+// SHUFFLE3(v, 0,1,2) leaves the vector unchanged (v.xyz).
+// SHUFFLE3(v, 0,0,0) splats the X (v.xxx).
+#define SHUFFLE3(V, X,Y,Z) float3(_mm_shuffle_ps((V).m, (V).m, _MM_SHUFFLE(Z,Z,Y,X)))
+
+struct float3
+{
+    VM_INLINE float3() {}
+    VM_INLINE explicit float3(const float *p) { m = _mm_set_ps(p[2], p[2], p[1], p[0]); }
+    VM_INLINE explicit float3(float x, float y, float z) { m = _mm_set_ps(z, z, y, x); }
+    VM_INLINE explicit float3(float v) { m = _mm_set1_ps(v); }
+    VM_INLINE explicit float3(__m128 v) { m = v; }
+
+    VM_INLINE float getX() const { return _mm_cvtss_f32(m); }
+    VM_INLINE float getY() const { return _mm_cvtss_f32(_mm_shuffle_ps(m, m, _MM_SHUFFLE(1, 1, 1, 1))); }
+    VM_INLINE float getZ() const { return _mm_cvtss_f32(_mm_shuffle_ps(m, m, _MM_SHUFFLE(2, 2, 2, 2))); }
+
+    VM_INLINE float3 yzx() const { return SHUFFLE3(*this, 1, 2, 0); }
+    VM_INLINE float3 zxy() const { return SHUFFLE3(*this, 2, 0, 1); }
+
+    VM_INLINE void store(float *p) const { p[0] = getX(); p[1] = getY(); p[2] = getZ(); }
+
+    void setX(float x)
+    {
+        m = _mm_move_ss(m, _mm_set_ss(x));
+    }
+    void setY(float y)
+    {
+        __m128 t = _mm_move_ss(m, _mm_set_ss(y));
+        t = _mm_shuffle_ps(t, t, _MM_SHUFFLE(3, 2, 0, 0));
+        m = _mm_move_ss(t, m);
+    }
+    void setZ(float z)
+    {
+        __m128 t = _mm_move_ss(m, _mm_set_ss(z));
+        t = _mm_shuffle_ps(t, t, _MM_SHUFFLE(3, 0, 1, 0));
+        m = _mm_move_ss(t, m);
+    }
+
+    __m128 m;
+};
+
+typedef float3 bool3;
+
+VM_INLINE float3 operator+ (float3 a, float3 b) { a.m = _mm_add_ps(a.m, b.m); return a; }
+VM_INLINE float3 operator- (float3 a, float3 b) { a.m = _mm_sub_ps(a.m, b.m); return a; }
+VM_INLINE float3 operator* (float3 a, float3 b) { a.m = _mm_mul_ps(a.m, b.m); return a; }
+VM_INLINE float3 operator/ (float3 a, float3 b) { a.m = _mm_div_ps(a.m, b.m); return a; }
+VM_INLINE float3 operator* (float3 a, float b) { a.m = _mm_mul_ps(a.m, _mm_set1_ps(b)); return a; }
+VM_INLINE float3 operator/ (float3 a, float b) { a.m = _mm_div_ps(a.m, _mm_set1_ps(b)); return a; }
+VM_INLINE float3 operator* (float a, float3 b) { b.m = _mm_mul_ps(_mm_set1_ps(a), b.m); return b; }
+VM_INLINE float3 operator/ (float a, float3 b) { b.m = _mm_div_ps(_mm_set1_ps(a), b.m); return b; }
+VM_INLINE float3& operator+= (float3 &a, float3 b) { a = a + b; return a; }
+VM_INLINE float3& operator-= (float3 &a, float3 b) { a = a - b; return a; }
+VM_INLINE float3& operator*= (float3 &a, float3 b) { a = a * b; return a; }
+VM_INLINE float3& operator/= (float3 &a, float3 b) { a = a / b; return a; }
+VM_INLINE float3& operator*= (float3 &a, float b) { a = a * b; return a; }
+VM_INLINE float3& operator/= (float3 &a, float b) { a = a / b; return a; }
+VM_INLINE bool3 operator==(float3 a, float3 b) { a.m = _mm_cmpeq_ps(a.m, b.m); return a; }
+VM_INLINE bool3 operator!=(float3 a, float3 b) { a.m = _mm_cmpneq_ps(a.m, b.m); return a; }
+VM_INLINE bool3 operator< (float3 a, float3 b) { a.m = _mm_cmplt_ps(a.m, b.m); return a; }
+VM_INLINE bool3 operator> (float3 a, float3 b) { a.m = _mm_cmpgt_ps(a.m, b.m); return a; }
+VM_INLINE bool3 operator<=(float3 a, float3 b) { a.m = _mm_cmple_ps(a.m, b.m); return a; }
+VM_INLINE bool3 operator>=(float3 a, float3 b) { a.m = _mm_cmpge_ps(a.m, b.m); return a; }
+VM_INLINE float3 min(float3 a, float3 b) { a.m = _mm_min_ps(a.m, b.m); return a; }
+VM_INLINE float3 max(float3 a, float3 b) { a.m = _mm_max_ps(a.m, b.m); return a; }
+
+VM_INLINE float3 operator- (float3 a) { return float3(_mm_setzero_ps()) - a; }
+
+VM_INLINE float hmin(float3 v)
+{
+    v = min(v, SHUFFLE3(v, 1, 0, 2));
+    return min(v, SHUFFLE3(v, 2, 0, 1)).getX();
+}
+VM_INLINE float hmax(float3 v)
+{
+    v = max(v, SHUFFLE3(v, 1, 0, 2));
+    return max(v, SHUFFLE3(v, 2, 0, 1)).getX();
+}
+
+VM_INLINE float3 cross(float3 a, float3 b)
+{
+    // x  <-  a.y*b.z - a.z*b.y
+    // y  <-  a.z*b.x - a.x*b.z
+    // z  <-  a.x*b.y - a.y*b.x
+    // We can save a shuffle by grouping it in this wacky order:
+    return (a.zxy()*b - a*b.zxy()).zxy();
+}
+
+// Returns a 3-bit code where bit0..bit2 is X..Z
+VM_INLINE unsigned mask(float3 v) { return _mm_movemask_ps(v.m) & 7; }
+// Once we have a comparison, we can branch based on its results:
+VM_INLINE bool any(bool3 v) { return mask(v) != 0; }
+VM_INLINE bool all(bool3 v) { return mask(v) == 7; }
+
+VM_INLINE float3 clamp(float3 t, float3 a, float3 b) { return min(max(t, a), b); }
+VM_INLINE float sum(float3 v) { return v.getX() + v.getY() + v.getZ(); }
+VM_INLINE float dot(float3 a, float3 b) { return sum(a*b); }
+
+#else // #if !defined(__arm__) && !defined(__arm64__)
+
+// ---- NEON implementation
+
+#include <arm_neon.h>
+
+struct float3
+{
+    VM_INLINE float3() {}
+    VM_INLINE explicit float3(const float *p) { float v[4] = {p[0], p[1], p[2], 0}; m = vld1q_f32(v); }
+    VM_INLINE explicit float3(float x, float y, float z) { float v[4] = {x, y, z, 0}; m = vld1q_f32(v); }
+    VM_INLINE explicit float3(float v) { m = vdupq_n_f32(v); }
+    VM_INLINE explicit float3(float32x4_t v) { m = v; }
+    
+    VM_INLINE float getX() const { return vgetq_lane_f32(m, 0); }
+    VM_INLINE float getY() const { return vgetq_lane_f32(m, 1); }
+    VM_INLINE float getZ() const { return vgetq_lane_f32(m, 2); }
+    
+    VM_INLINE float3 yzx() const
+    {
+        float32x2_t low = vget_low_f32(m);
+        float32x4_t yzx = vcombine_f32(vext_f32(low, vget_high_f32(m), 1), low);
+        return float3(yzx);
+    }
+    VM_INLINE float3 zxy() const
+    {
+        float32x4_t p = m;
+        p = vuzpq_f32(vreinterpretq_f32_s32(vextq_s32(vreinterpretq_s32_f32(p), vreinterpretq_s32_f32(p), 1)), p).val[1];
+        return float3(p);
+    }
+    
+    VM_INLINE void store(float *p) const { p[0] = getX(); p[1] = getY(); p[2] = getZ(); }
+    
+    void setX(float x)
+    {
+        m = vsetq_lane_f32(x, m, 0);
+    }
+    void setY(float y)
+    {
+        m = vsetq_lane_f32(y, m, 1);
+    }
+    void setZ(float z)
+    {
+        m = vsetq_lane_f32(z, m, 2);
+    }
+    
+    float32x4_t m;
+};
+
+typedef float3 bool3;
+
+VM_INLINE float32x4_t rcp_2(float32x4_t v)
+{
+    float32x4_t e = vrecpeq_f32(v);
+    e = vmulq_f32(vrecpsq_f32(e, v), e);
+    e = vmulq_f32(vrecpsq_f32(e, v), e);
+    return e;
+}
+
+VM_INLINE float3 operator+ (float3 a, float3 b) { a.m = vaddq_f32(a.m, b.m); return a; }
+VM_INLINE float3 operator- (float3 a, float3 b) { a.m = vsubq_f32(a.m, b.m); return a; }
+VM_INLINE float3 operator* (float3 a, float3 b) { a.m = vmulq_f32(a.m, b.m); return a; }
+VM_INLINE float3 operator/ (float3 a, float3 b) { float32x4_t recip = rcp_2(b.m); a.m = vmulq_f32(a.m, recip); return a; }
+VM_INLINE float3 operator* (float3 a, float b) { a.m = vmulq_f32(a.m, vdupq_n_f32(b)); return a; }
+VM_INLINE float3 operator/ (float3 a, float b) { float32x4_t recip = rcp_2(vdupq_n_f32(b)); a.m = vmulq_f32(a.m, recip); return a; }
+VM_INLINE float3 operator* (float a, float3 b) { b.m = vmulq_f32(vdupq_n_f32(a), b.m); return b; }
+VM_INLINE float3 operator/ (float a, float3 b) { float32x4_t recip = rcp_2(b.m); b.m = vmulq_f32(vdupq_n_f32(a), recip); return b; }
+VM_INLINE float3& operator+= (float3 &a, float3 b) { a = a + b; return a; }
+VM_INLINE float3& operator-= (float3 &a, float3 b) { a = a - b; return a; }
+VM_INLINE float3& operator*= (float3 &a, float3 b) { a = a * b; return a; }
+VM_INLINE float3& operator/= (float3 &a, float3 b) { a = a / b; return a; }
+VM_INLINE float3& operator*= (float3 &a, float b) { a = a * b; return a; }
+VM_INLINE float3& operator/= (float3 &a, float b) { a = a / b; return a; }
+VM_INLINE bool3 operator==(float3 a, float3 b) { a.m = vceqq_f32(a.m, b.m); return a; }
+VM_INLINE bool3 operator!=(float3 a, float3 b) { a.m = vmvnq_u32(vceqq_f32(a.m, b.m)); return a; }
+VM_INLINE bool3 operator< (float3 a, float3 b) { a.m = vcltq_f32(a.m, b.m); return a; }
+VM_INLINE bool3 operator> (float3 a, float3 b) { a.m = vcgtq_f32(a.m, b.m); return a; }
+VM_INLINE bool3 operator<=(float3 a, float3 b) { a.m = vcleq_f32(a.m, b.m); return a; }
+VM_INLINE bool3 operator>=(float3 a, float3 b) { a.m = vcgeq_f32(a.m, b.m); return a; }
+VM_INLINE float3 min(float3 a, float3 b) { a.m = vminq_f32(a.m, b.m); return a; }
+VM_INLINE float3 max(float3 a, float3 b) { a.m = vmaxq_f32(a.m, b.m); return a; }
+
+VM_INLINE float3 operator- (float3 a) { a.m = vnegq_f32(a.m); return a; }
+
+VM_INLINE float hmin(float3 v)
+{
+    float32x2_t minOfHalfs = vpmin_f32(vget_low_f32(v.m), vget_high_f32(v.m));
+    float32x2_t minOfMinOfHalfs = vpmin_f32(minOfHalfs, minOfHalfs);
+    return vget_lane_f32(minOfMinOfHalfs, 0);
+}
+VM_INLINE float hmax(float3 v)
+{
+    float32x2_t maxOfHalfs = vpmax_f32(vget_low_f32(v.m), vget_high_f32(v.m));
+    float32x2_t maxOfMaxOfHalfs = vpmax_f32(maxOfHalfs, maxOfHalfs);
+    return vget_lane_f32(maxOfMaxOfHalfs, 0);
+}
+
+VM_INLINE float3 cross(float3 a, float3 b)
+{
+    // x  <-  a.y*b.z - a.z*b.y
+    // y  <-  a.z*b.x - a.x*b.z
+    // z  <-  a.x*b.y - a.y*b.x
+    // We can save a shuffle by grouping it in this wacky order:
+    return (a.zxy()*b - a*b.zxy()).zxy();
+}
+
+// Returns a 3-bit code where bit0..bit2 is X..Z
+VM_INLINE unsigned mask(float3 v)
+{
+    static const uint32x4_t movemask = { 1, 2, 4, 8 };
+    static const uint32x4_t highbit = { 0x80000000, 0x80000000, 0x80000000, 0x80000000 };
+    uint32x4_t t0 = vreinterpretq_u32_f32(v.m);
+    uint32x4_t t1 = vtstq_u32(t0, highbit);
+    uint32x4_t t2 = vandq_u32(t1, movemask);
+    uint32x2_t t3 = vorr_u32(vget_low_u32(t2), vget_high_u32(t2));
+    return vget_lane_u32(t3, 0) | vget_lane_u32(t3, 1);
+}
+// Once we have a comparison, we can branch based on its results:
+VM_INLINE bool any(bool3 v) { return mask(v) != 0; }
+VM_INLINE bool all(bool3 v) { return mask(v) == 7; }
+
+VM_INLINE float3 clamp(float3 t, float3 a, float3 b) { return min(max(t, a), b); }
+VM_INLINE float sum(float3 v) { return v.getX() + v.getY() + v.getZ(); }
+VM_INLINE float dot(float3 a, float3 b) { return sum(a*b); }
+
+
+#endif // #else of #if !defined(__arm__) && !defined(__arm64__)
+
+#else // #if DO_FLOAT3_WITH_SIMD
+
+// ---- Simple scalar C implementation
+
+
+struct float3
+{
+    float3() : x(0), y(0), z(0) {}
+    float3(float x_, float y_, float z_) : x(x_), y(y_), z(z_) {}
+
+    float3 operator-() const { return float3(-x, -y, -z); }
+    float3& operator+=(const float3& o) { x+=o.x; y+=o.y; z+=o.z; return *this; }
+    float3& operator-=(const float3& o) { x-=o.x; y-=o.y; z-=o.z; return *this; }
+    float3& operator*=(const float3& o) { x*=o.x; y*=o.y; z*=o.z; return *this; }
+    float3& operator*=(float o) { x*=o; y*=o; z*=o; return *this; }
+
+    VM_INLINE float getX() const { return x; }
+    VM_INLINE float getY() const { return y; }
+    VM_INLINE float getZ() const { return z; }
+    VM_INLINE void setX(float x_) { x = x_; }
+    VM_INLINE void setY(float y_) { y = y_; }
+    VM_INLINE void setZ(float z_) { z = z_; }
+    VM_INLINE void store(float *p) const { p[0] = getX(); p[1] = getY(); p[2] = getZ(); }
+
+    float x, y, z;
+};
+
+VM_INLINE float3 operator+(const float3& a, const float3& b) { return float3(a.x+b.x,a.y+b.y,a.z+b.z); }
+VM_INLINE float3 operator-(const float3& a, const float3& b) { return float3(a.x-b.x,a.y-b.y,a.z-b.z); }
+VM_INLINE float3 operator*(const float3& a, const float3& b) { return float3(a.x*b.x,a.y*b.y,a.z*b.z); }
+VM_INLINE float3 operator*(const float3& a, float b) { return float3(a.x*b,a.y*b,a.z*b); }
+VM_INLINE float3 operator*(float a, const float3& b) { return float3(a*b.x,a*b.y,a*b.z); }
+VM_INLINE float dot(const float3& a, const float3& b) { return a.x*b.x+a.y*b.y+a.z*b.z; }
+VM_INLINE float3 cross(const float3& a, const float3& b)
+{
+    return float3(
+                  a.y*b.z - a.z*b.y,
+                  -(a.x*b.z - a.z*b.x),
+                  a.x*b.y - a.y*b.x
+                  );
+}
+#endif // #else of #if DO_FLOAT3_WITH_SIMD
+
+VM_INLINE float length(float3 v) { return sqrtf(dot(v, v)); }
+VM_INLINE float sqLength(float3 v) { return dot(v, v); }
+VM_INLINE float3 normalize(float3 v) { return v * (1.0f / length(v)); }
+VM_INLINE float3 lerp(float3 a, float3 b, float t) { return a + (b-a)*t; }
+
+
+inline void AssertUnit(float3 v)
+{
+    assert(fabsf(sqLength(v) - 1.0f) < 0.01f);
+}
+
+inline float3 reflect(float3 v, float3 n)
+{
+    return v - 2*dot(v,n)*n;
+}
+
+inline bool refract(float3 v, float3 n, float nint, float3& outRefracted)
+{
+    AssertUnit(v);
+    float dt = dot(v, n);
+    float discr = 1.0f - nint*nint*(1-dt*dt);
+    if (discr > 0)
+    {
+        outRefracted = nint * (v - n*dt) - n*sqrtf(discr);
+        return true;
+    }
+    return false;
+}
+inline float schlick(float cosine, float ri)
+{
+    float r0 = (1-ri) / (1+ri);
+    r0 = r0*r0;
+    return r0 + (1-r0)*powf(1-cosine, 5);
+}
+
+struct Ray
+{
+    Ray() {}
+    Ray(float3 orig_, float3 dir_) : orig(orig_), dir(dir_) { AssertUnit(dir); }
+
+    float3 pointAt(float t) const { return orig + dir * t; }
+
+    float3 orig;
+    float3 dir;
+};
+
+
+struct Hit
+{
+    float3 pos;
+    float3 normal;
+    float t;
+};
+
+
+struct Sphere
+{
+    Sphere() : radius(1.0f), invRadius(0.0f) {}
+    Sphere(float3 center_, float radius_) : center(center_), radius(radius_), invRadius(0.0f) {}
+
+    void UpdateDerivedData() { invRadius = 1.0f/radius; }
+
+    float3 center;
+    float radius;
+    float invRadius;
+};
+
+
+// data for all spheres in a "structure of arrays" layout
+struct SpheresSoA
+{
+    SpheresSoA(int c)
+    {
+        count = c;
+        // we'll be processing spheres in kSimdWidth chunks, so make sure to allocate
+        // enough space
+        simdCount = (c + (kSimdWidth - 1)) / kSimdWidth * kSimdWidth;
+        centerX = new float[simdCount];
+        centerY = new float[simdCount];
+        centerZ = new float[simdCount];
+        sqRadius = new float[simdCount];
+        invRadius = new float[simdCount];
+        // set all data to "impossible sphere" state
+        for (int i = count; i < simdCount; ++i)
+        {
+            centerX[i] = centerY[i] = centerZ[i] = 10000.0f;
+            sqRadius[i] = 0.0f;
+            invRadius[i] = 0.0f;
+        }
+    }
+    ~SpheresSoA()
+    {
+        delete[] centerX;
+        delete[] centerY;
+        delete[] centerZ;
+        delete[] sqRadius;
+        delete[] invRadius;
+    }
+    float* centerX;
+    float* centerY;
+    float* centerZ;
+    float* sqRadius;
+    float* invRadius;
+    int simdCount;
+    int count;
+};
+
+
+int HitSpheres(const Ray& r, const SpheresSoA& spheres, float tMin, float tMax, Hit& outHit);
+
+float RandomFloat01(uint32_t& state);
+float3 RandomInUnitDisk(uint32_t& state);
+float3 RandomInUnitSphere(uint32_t& state);
+float3 RandomUnitVector(uint32_t& state);
+
+struct Camera
+{
+    Camera() {}
+    // vfov is top to bottom in degrees
+    Camera(const float3& lookFrom, const float3& lookAt, const float3& vup, float vfov, float aspect, float aperture, float focusDist)
+    {
+        lensRadius = aperture / 2;
+        float theta = vfov*kPI/180;
+        float halfHeight = tanf(theta/2);
+        float halfWidth = aspect * halfHeight;
+        origin = lookFrom;
+        w = normalize(lookFrom - lookAt);
+        u = normalize(cross(vup, w));
+        v = cross(w, u);
+        lowerLeftCorner = origin - halfWidth*focusDist*u - halfHeight*focusDist*v - focusDist*w;
+        horizontal = 2*halfWidth*focusDist*u;
+        vertical = 2*halfHeight*focusDist*v;
+    }
+
+    Ray GetRay(float s, float t, uint32_t& state) const
+    {
+        float3 rd = lensRadius * RandomInUnitDisk(state);
+        float3 offset = u * rd.getX() + v * rd.getY();
+        return Ray(origin + offset, normalize(lowerLeftCorner + s*horizontal + t*vertical - origin - offset));
+    }
+
+    float3 origin;
+    float3 lowerLeftCorner;
+    float3 horizontal;
+    float3 vertical;
+    float3 u, v, w;
+    float lensRadius;
+};
+

third_party/tracy/examples/ToyPathTracer/Source/Test.cpp

@@ -0,0 +1,392 @@
+#include "Config.h"
+#include "Test.h"
+#include "Maths.h"
+#include <algorithm>
+#if CPU_CAN_DO_THREADS
+#include "enkiTS/TaskScheduler_c.h"
+#include <thread>
+#endif
+#include <atomic>
+
+#include "../../../public/tracy/Tracy.hpp"
+
+// 46 spheres (2 emissive) when enabled; 9 spheres (1 emissive) when disabled
+#define DO_BIG_SCENE 1
+
+static Sphere s_Spheres[] =
+{
+    {float3(0,-100.5,-1), 100},
+    {float3(2,0,-1), 0.5f},
+    {float3(0,0,-1), 0.5f},
+    {float3(-2,0,-1), 0.5f},
+    {float3(2,0,1), 0.5f},
+    {float3(0,0,1), 0.5f},
+    {float3(-2,0,1), 0.5f},
+    {float3(0.5f,1,0.5f), 0.5f},
+    {float3(-1.5f,1.5f,0.f), 0.3f},
+#if DO_BIG_SCENE
+    {float3(4,0,-3), 0.5f}, {float3(3,0,-3), 0.5f}, {float3(2,0,-3), 0.5f}, {float3(1,0,-3), 0.5f}, {float3(0,0,-3), 0.5f}, {float3(-1,0,-3), 0.5f}, {float3(-2,0,-3), 0.5f}, {float3(-3,0,-3), 0.5f}, {float3(-4,0,-3), 0.5f},
+    {float3(4,0,-4), 0.5f}, {float3(3,0,-4), 0.5f}, {float3(2,0,-4), 0.5f}, {float3(1,0,-4), 0.5f}, {float3(0,0,-4), 0.5f}, {float3(-1,0,-4), 0.5f}, {float3(-2,0,-4), 0.5f}, {float3(-3,0,-4), 0.5f}, {float3(-4,0,-4), 0.5f},
+    {float3(4,0,-5), 0.5f}, {float3(3,0,-5), 0.5f}, {float3(2,0,-5), 0.5f}, {float3(1,0,-5), 0.5f}, {float3(0,0,-5), 0.5f}, {float3(-1,0,-5), 0.5f}, {float3(-2,0,-5), 0.5f}, {float3(-3,0,-5), 0.5f}, {float3(-4,0,-5), 0.5f},
+    {float3(4,0,-6), 0.5f}, {float3(3,0,-6), 0.5f}, {float3(2,0,-6), 0.5f}, {float3(1,0,-6), 0.5f}, {float3(0,0,-6), 0.5f}, {float3(-1,0,-6), 0.5f}, {float3(-2,0,-6), 0.5f}, {float3(-3,0,-6), 0.5f}, {float3(-4,0,-6), 0.5f},
+    {float3(1.5f,1.5f,-2), 0.3f},
+#endif // #if DO_BIG_SCENE
+};
+const int kSphereCount = sizeof(s_Spheres) / sizeof(s_Spheres[0]);
+
+static SpheresSoA s_SpheresSoA(kSphereCount);
+
+struct Material
+{
+    enum Type { Lambert, Metal, Dielectric };
+    Type type;
+    float3 albedo;
+    float3 emissive;
+    float roughness;
+    float ri;
+};
+
+static Material s_SphereMats[kSphereCount] =
+{
+    { Material::Lambert, float3(0.8f, 0.8f, 0.8f), float3(0,0,0), 0, 0, },
+    { Material::Lambert, float3(0.8f, 0.4f, 0.4f), float3(0,0,0), 0, 0, },
+    { Material::Lambert, float3(0.4f, 0.8f, 0.4f), float3(0,0,0), 0, 0, },
+    { Material::Metal, float3(0.4f, 0.4f, 0.8f), float3(0,0,0), 0, 0 },
+    { Material::Metal, float3(0.4f, 0.8f, 0.4f), float3(0,0,0), 0, 0 },
+    { Material::Metal, float3(0.4f, 0.8f, 0.4f), float3(0,0,0), 0.2f, 0 },
+    { Material::Metal, float3(0.4f, 0.8f, 0.4f), float3(0,0,0), 0.6f, 0 },
+    { Material::Dielectric, float3(0.4f, 0.4f, 0.4f), float3(0,0,0), 0, 1.5f },
+    { Material::Lambert, float3(0.8f, 0.6f, 0.2f), float3(30,25,15), 0, 0 },
+#if DO_BIG_SCENE
+    { Material::Lambert, float3(0.1f, 0.1f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.2f, 0.2f, 0.2f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.3f, 0.3f, 0.3f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.4f, 0.4f, 0.4f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.5f, 0.5f, 0.5f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.6f, 0.6f, 0.6f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.7f, 0.7f, 0.7f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.8f, 0.8f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.9f, 0.9f, 0.9f), float3(0,0,0), 0, 0, },
+    { Material::Metal, float3(0.1f, 0.1f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.2f, 0.2f, 0.2f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.3f, 0.3f, 0.3f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.4f, 0.4f, 0.4f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.5f, 0.5f, 0.5f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.6f, 0.6f, 0.6f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.7f, 0.7f, 0.7f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.8f, 0.8f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.9f, 0.9f, 0.9f), float3(0,0,0), 0, 0, },
+    { Material::Metal, float3(0.8f, 0.1f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.8f, 0.5f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.8f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.4f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.1f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.1f, 0.8f, 0.5f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.1f, 0.8f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.1f, 0.1f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.5f, 0.1f, 0.8f), float3(0,0,0), 0, 0, },
+    { Material::Lambert, float3(0.8f, 0.1f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.8f, 0.5f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.8f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.4f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.1f, 0.8f, 0.1f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.1f, 0.8f, 0.5f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.1f, 0.8f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Lambert, float3(0.1f, 0.1f, 0.8f), float3(0,0,0), 0, 0, }, { Material::Metal, float3(0.5f, 0.1f, 0.8f), float3(0,0,0), 0, 0, },
+    { Material::Lambert, float3(0.1f, 0.2f, 0.5f), float3(3,10,20), 0, 0 },
+#endif
+};
+
+static int s_EmissiveSpheres[kSphereCount];
+static int s_EmissiveSphereCount;
+
+static Camera s_Cam;
+
+const float kMinT = 0.001f;
+const float kMaxT = 1.0e7f;
+const int kMaxDepth = 10;
+
+
+bool HitWorld(const Ray& r, float tMin, float tMax, Hit& outHit, int& outID)
+{
+    outID = HitSpheres(r, s_SpheresSoA, tMin, tMax, outHit);
+    return outID != -1;
+}
+
+
+static bool Scatter(const Material& mat, const Ray& r_in, const Hit& rec, float3& attenuation, Ray& scattered, float3& outLightE, int& inoutRayCount, uint32_t& state)
+{
+    ZoneScoped;
+    outLightE = float3(0,0,0);
+    if (mat.type == Material::Lambert)
+    {
+        // random point on unit sphere that is tangent to the hit point
+        float3 target = rec.pos + rec.normal + RandomUnitVector(state);
+        scattered = Ray(rec.pos, normalize(target - rec.pos));
+        attenuation = mat.albedo;
+
+        // sample lights
+#if DO_LIGHT_SAMPLING
+        for (int j = 0; j < s_EmissiveSphereCount; ++j)
+        {
+            int i = s_EmissiveSpheres[j];
+            const Material& smat = s_SphereMats[i];
+            if (&mat == &smat)
+                continue; // skip self
+            const Sphere& s = s_Spheres[i];
+
+            // create a random direction towards sphere
+            // coord system for sampling: sw, su, sv
+            float3 sw = normalize(s.center - rec.pos);
+            float3 su = normalize(cross(fabs(sw.getX())>0.01f ? float3(0,1,0):float3(1,0,0), sw));
+            float3 sv = cross(sw, su);
+            // sample sphere by solid angle
+            float cosAMax = sqrtf(1.0f - s.radius*s.radius / sqLength(rec.pos-s.center));
+            float eps1 = RandomFloat01(state), eps2 = RandomFloat01(state);
+            float cosA = 1.0f - eps1 + eps1 * cosAMax;
+            float sinA = sqrtf(1.0f - cosA*cosA);
+            float phi = 2 * kPI * eps2;
+            float3 l = su * (cosf(phi) * sinA) + sv * (sinf(phi) * sinA) + sw * cosA;
+            //l = normalize(l); // NOTE(fg): This is already normalized, by construction.
+
+            // shoot shadow ray
+            Hit lightHit;
+            int hitID;
+            ++inoutRayCount;
+            if (HitWorld(Ray(rec.pos, l), kMinT, kMaxT, lightHit, hitID) && hitID == i)
+            {
+                float omega = 2 * kPI * (1-cosAMax);
+
+                float3 rdir = r_in.dir;
+                AssertUnit(rdir);
+                float3 nl = dot(rec.normal, rdir) < 0 ? rec.normal : -rec.normal;
+                outLightE += (mat.albedo * smat.emissive) * (std::max(0.0f, dot(l, nl)) * omega / kPI);
+            }
+        }
+#endif
+        return true;
+    }
+    else if (mat.type == Material::Metal)
+    {
+        AssertUnit(r_in.dir); AssertUnit(rec.normal);
+        float3 refl = reflect(r_in.dir, rec.normal);
+        // reflected ray, and random inside of sphere based on roughness
+        float roughness = mat.roughness;
+#if DO_MITSUBA_COMPARE
+        roughness = 0; // until we get better BRDF for metals
+#endif
+        scattered = Ray(rec.pos, normalize(refl + roughness*RandomInUnitSphere(state)));
+        attenuation = mat.albedo;
+        return dot(scattered.dir, rec.normal) > 0;
+    }
+    else if (mat.type == Material::Dielectric)
+    {
+        AssertUnit(r_in.dir); AssertUnit(rec.normal);
+        float3 outwardN;
+        float3 rdir = r_in.dir;
+        float3 refl = reflect(rdir, rec.normal);
+        float nint;
+        attenuation = float3(1,1,1);
+        float3 refr;
+        float reflProb;
+        float cosine;
+        if (dot(rdir, rec.normal) > 0)
+        {
+            outwardN = -rec.normal;
+            nint = mat.ri;
+            cosine = mat.ri * dot(rdir, rec.normal);
+        }
+        else
+        {
+            outwardN = rec.normal;
+            nint = 1.0f / mat.ri;
+            cosine = -dot(rdir, rec.normal);
+        }
+        if (refract(rdir, outwardN, nint, refr))
+        {
+            reflProb = schlick(cosine, mat.ri);
+        }
+        else
+        {
+            reflProb = 1;
+        }
+        if (RandomFloat01(state) < reflProb)
+            scattered = Ray(rec.pos, normalize(refl));
+        else
+            scattered = Ray(rec.pos, normalize(refr));
+    }
+    else
+    {
+        attenuation = float3(1,0,1);
+        return false;
+    }
+    return true;
+}
+
+static float3 Trace(const Ray& r, int depth, int& inoutRayCount, uint32_t& state, bool doMaterialE = true)
+{
+    ZoneScoped;
+    Hit rec;
+    int id = 0;
+    ++inoutRayCount;
+    if (HitWorld(r, kMinT, kMaxT, rec, id))
+    {
+        Ray scattered;
+        float3 attenuation;
+        float3 lightE;
+        const Material& mat = s_SphereMats[id];
+        float3 matE = mat.emissive;
+        if (depth < kMaxDepth && Scatter(mat, r, rec, attenuation, scattered, lightE, inoutRayCount, state))
+        {
+#if DO_LIGHT_SAMPLING
+            if (!doMaterialE) matE = float3(0,0,0); // don't add material emission if told so
+            // dor Lambert materials, we just did explicit light (emissive) sampling and already
+            // for their contribution, so if next ray bounce hits the light again, don't add
+            // emission
+            doMaterialE = (mat.type != Material::Lambert);
+#endif
+            return matE + lightE + attenuation * Trace(scattered, depth+1, inoutRayCount, state, doMaterialE);
+        }
+        else
+        {
+            return matE;
+        }
+    }
+    else
+    {
+        // sky
+#if DO_MITSUBA_COMPARE
+        return float3(0.15f,0.21f,0.3f); // easier compare with Mitsuba's constant environment light
+#else
+        float3 unitDir = r.dir;
+        float t = 0.5f*(unitDir.getY() + 1.0f);
+        return ((1.0f-t)*float3(1.0f, 1.0f, 1.0f) + t*float3(0.5f, 0.7f, 1.0f)) * 0.3f;
+#endif
+    }
+}
+
+#if CPU_CAN_DO_THREADS
+static enkiTaskScheduler* g_TS;
+#endif
+
+void InitializeTest()
+{
+    ZoneScoped;
+    #if CPU_CAN_DO_THREADS
+    g_TS = enkiNewTaskScheduler();
+    enkiInitTaskSchedulerNumThreads(g_TS, std::max<int>( 2, std::thread::hardware_concurrency() - 2));
+    #endif
+}
+
+void ShutdownTest()
+{
+    ZoneScoped;
+    #if CPU_CAN_DO_THREADS
+    enkiDeleteTaskScheduler(g_TS);
+    #endif
+}
+
+struct JobData
+{
+    float time;
+    int frameCount;
+    int screenWidth, screenHeight;
+    float* backbuffer;
+    Camera* cam;
+    std::atomic<int> rayCount;
+    unsigned testFlags;
+};
+
+static void TraceRowJob(uint32_t start, uint32_t end, uint32_t threadnum, void* data_)
+{
+    ZoneScoped;
+    JobData& data = *(JobData*)data_;
+    float* backbuffer = data.backbuffer + start * data.screenWidth * 4;
+    float invWidth = 1.0f / data.screenWidth;
+    float invHeight = 1.0f / data.screenHeight;
+    float lerpFac = float(data.frameCount) / float(data.frameCount+1);
+    if (data.testFlags & kFlagAnimate)
+        lerpFac *= DO_ANIMATE_SMOOTHING;
+    if (!(data.testFlags & kFlagProgressive))
+        lerpFac = 0;
+    int rayCount = 0;
+    for (uint32_t y = start; y < end; ++y)
+    {
+        uint32_t state = (y * 9781 + data.frameCount * 6271) | 1;
+        for (int x = 0; x < data.screenWidth; ++x)
+        {
+            float3 col(0, 0, 0);
+            for (int s = 0; s < DO_SAMPLES_PER_PIXEL; s++)
+            {
+                float u = float(x + RandomFloat01(state)) * invWidth;
+                float v = float(y + RandomFloat01(state)) * invHeight;
+                Ray r = data.cam->GetRay(u, v, state);
+                col += Trace(r, 0, rayCount, state);
+            }
+            col *= 1.0f / float(DO_SAMPLES_PER_PIXEL);
+
+            float3 prev(backbuffer[0], backbuffer[1], backbuffer[2]);
+            col = prev * lerpFac + col * (1-lerpFac);
+            col.store(backbuffer);
+            backbuffer += 4;
+        }
+    }
+    data.rayCount += rayCount;
+}
+
+void UpdateTest(float time, int frameCount, int screenWidth, int screenHeight, unsigned testFlags)
+{
+    ZoneScoped;
+    if (testFlags & kFlagAnimate)
+    {
+        s_Spheres[1].center.setY(cosf(time) + 1.0f);
+        s_Spheres[8].center.setZ(sinf(time)*0.3f);
+    }
+    float3 lookfrom(0, 2, 3);
+    float3 lookat(0, 0, 0);
+    float distToFocus = 3;
+#if DO_MITSUBA_COMPARE
+    float aperture = 0.0f;
+#else
+    float aperture = 0.1f;
+#endif
+#if DO_BIG_SCENE
+    aperture *= 0.2f;
+#endif
+
+    s_EmissiveSphereCount = 0;
+    for (int i = 0; i < kSphereCount; ++i)
+    {
+        Sphere& s = s_Spheres[i];
+        s.UpdateDerivedData();
+        s_SpheresSoA.centerX[i] = s.center.getX();
+        s_SpheresSoA.centerY[i] = s.center.getY();
+        s_SpheresSoA.centerZ[i] = s.center.getZ();
+        s_SpheresSoA.sqRadius[i] = s.radius * s.radius;
+        s_SpheresSoA.invRadius[i] = s.invRadius;
+
+        // Remember IDs of emissive spheres (light sources)
+        const Material& smat = s_SphereMats[i];
+        if (smat.emissive.getX() > 0 || smat.emissive.getY() > 0 || smat.emissive.getZ() > 0)
+        {
+            s_EmissiveSpheres[s_EmissiveSphereCount] = i;
+            s_EmissiveSphereCount++;
+        }
+    }
+
+    s_Cam = Camera(lookfrom, lookat, float3(0, 1, 0), 60, float(screenWidth) / float(screenHeight), aperture, distToFocus);
+}
+
+void DrawTest(float time, int frameCount, int screenWidth, int screenHeight, float* backbuffer, int& outRayCount, unsigned testFlags)
+{
+    ZoneScoped;
+    JobData args;
+    args.time = time;
+    args.frameCount = frameCount;
+    args.screenWidth = screenWidth;
+    args.screenHeight = screenHeight;
+    args.backbuffer = backbuffer;
+    args.cam = &s_Cam;
+    args.testFlags = testFlags;
+    args.rayCount = 0;
+
+    #if CPU_CAN_DO_THREADS
+    enkiTaskSet* task = enkiCreateTaskSet(g_TS, TraceRowJob);
+    bool threaded = true;
+    enkiAddTaskSetToPipeMinRange(g_TS, task, &args, screenHeight, threaded ? 4 : screenHeight);
+    enkiWaitForTaskSet(g_TS, task);
+    enkiDeleteTaskSet(task);
+    #else
+    TraceRowJob(0, screenHeight, 0, &args);
+    #endif
+
+    outRayCount = args.rayCount;
+}
+
+void GetObjectCount(int& outCount, int& outObjectSize, int& outMaterialSize, int& outCamSize)
+{
+    ZoneScoped;
+    outCount = kSphereCount;
+    outObjectSize = sizeof(Sphere);
+    outMaterialSize = sizeof(Material);
+    outCamSize = sizeof(Camera);
+}
+
+void GetSceneDesc(void* outObjects, void* outMaterials, void* outCam, void* outEmissives, int* outEmissiveCount)
+{
+    ZoneScoped;
+    memcpy(outObjects, s_Spheres, kSphereCount * sizeof(s_Spheres[0]));
+    memcpy(outMaterials, s_SphereMats, kSphereCount * sizeof(s_SphereMats[0]));
+    memcpy(outCam, &s_Cam, sizeof(s_Cam));
+    memcpy(outEmissives, s_EmissiveSpheres, s_EmissiveSphereCount * sizeof(s_EmissiveSpheres[0]));
+    *outEmissiveCount = s_EmissiveSphereCount;
+}

third_party/tracy/examples/ToyPathTracer/Source/Test.h

@@ -0,0 +1,17 @@
+#pragma once
+#include <stdint.h>
+
+enum TestFlags
+{
+    kFlagAnimate = (1 << 0),
+    kFlagProgressive = (1 << 1),
+};
+
+void InitializeTest();
+void ShutdownTest();
+
+void UpdateTest(float time, int frameCount, int screenWidth, int screenHeight, unsigned testFlags);
+void DrawTest(float time, int frameCount, int screenWidth, int screenHeight, float* backbuffer, int& outRayCount, unsigned testFlags);
+
+void GetObjectCount(int& outCount, int& outObjectSize, int& outMaterialSize, int& outCamSize);
+void GetSceneDesc(void* outObjects, void* outMaterials, void* outCam, void* outEmissives, int* outEmissiveCount);

third_party/tracy/examples/ToyPathTracer/Source/enkiTS/Atomics.h

@@ -0,0 +1,79 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#pragma once
+
+#include <stdint.h>
+
+#ifdef _WIN32
+    #define WIN32_LEAN_AND_MEAN
+    #include <Windows.h>
+	#undef GetObject
+    #include <intrin.h>
+
+    extern "C" void _ReadWriteBarrier();
+    #pragma intrinsic(_ReadWriteBarrier)
+    #pragma intrinsic(_InterlockedCompareExchange)
+    #pragma intrinsic(_InterlockedExchangeAdd)
+
+    // Memory Barriers to prevent CPU and Compiler re-ordering
+    #define BASE_MEMORYBARRIER_ACQUIRE() _ReadWriteBarrier()
+    #define BASE_MEMORYBARRIER_RELEASE() _ReadWriteBarrier()
+    #define BASE_ALIGN(x) __declspec( align( x ) ) 
+
+#else
+    #define BASE_MEMORYBARRIER_ACQUIRE() __asm__ __volatile__("": : :"memory")  
+    #define BASE_MEMORYBARRIER_RELEASE() __asm__ __volatile__("": : :"memory")  
+	#define BASE_ALIGN(x)  __attribute__ ((aligned( x )))
+#endif
+
+namespace enki
+{
+    // Atomically performs: if( *pDest == compareWith ) { *pDest = swapTo; }
+    // returns old *pDest (so if successfull, returns compareWith)
+    inline uint32_t AtomicCompareAndSwap( volatile uint32_t* pDest, uint32_t swapTo, uint32_t compareWith )
+    {
+       #ifdef _WIN32
+			// assumes two's complement - unsigned / signed conversion leads to same bit pattern
+            return _InterlockedCompareExchange( (volatile long*)pDest,swapTo, compareWith );
+        #else
+            return __sync_val_compare_and_swap( pDest, compareWith, swapTo );
+        #endif      
+    }
+
+    inline uint64_t AtomicCompareAndSwap( volatile uint64_t* pDest, uint64_t swapTo, uint64_t compareWith )
+    {
+       #ifdef _WIN32
+			// assumes two's complement - unsigned / signed conversion leads to same bit pattern
+            return _InterlockedCompareExchange64( (__int64 volatile*)pDest, swapTo, compareWith );
+        #else
+            return __sync_val_compare_and_swap( pDest, compareWith, swapTo );
+        #endif      
+    }	
+
+    // Atomically performs: tmp = *pDest; *pDest += value; return tmp;
+    inline int32_t AtomicAdd( volatile int32_t* pDest, int32_t value )
+    {
+       #ifdef _WIN32
+            return _InterlockedExchangeAdd( (long*)pDest, value );
+        #else
+            return __sync_fetch_and_add( pDest, value );
+        #endif      
+    }
+
+}

third_party/tracy/examples/ToyPathTracer/Source/enkiTS/LockLessMultiReadPipe.h

@@ -0,0 +1,240 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#pragma once
+
+#include <stdint.h>
+#include <assert.h>
+
+#include "Atomics.h"
+#include <string.h>
+
+
+namespace enki
+{
+    // LockLessMultiReadPipe - Single writer, multiple reader thread safe pipe using (semi) lockless programming
+    // Readers can only read from the back of the pipe
+    // The single writer can write to the front of the pipe, and read from both ends (a writer can be a reader)
+    // for many of the principles used here, see http://msdn.microsoft.com/en-us/library/windows/desktop/ee418650(v=vs.85).aspx
+    // Note: using log2 sizes so we do not need to clamp (multi-operation)
+    // T is the contained type
+    // Note this is not true lockless as the use of flags as a form of lock state.
+    template<uint8_t cSizeLog2, typename T> class LockLessMultiReadPipe
+    {
+    public:
+        LockLessMultiReadPipe();
+        ~LockLessMultiReadPipe() {}
+
+        // ReaderTryReadBack returns false if we were unable to read
+        // This is thread safe for both multiple readers and the writer
+        bool ReaderTryReadBack(   T* pOut );
+
+        // WriterTryReadFront returns false if we were unable to read
+        // This is thread safe for the single writer, but should not be called by readers
+        bool WriterTryReadFront(  T* pOut );
+
+        // WriterTryWriteFront returns false if we were unable to write
+        // This is thread safe for the single writer, but should not be called by readers
+        bool WriterTryWriteFront( const T& in );
+
+        // IsPipeEmpty() is a utility function, not intended for general use
+        // Should only be used very prudently.
+        bool IsPipeEmpty() const
+        {
+            return 0 == m_WriteIndex - m_ReadCount;
+        }
+
+        void Clear()
+        {
+            m_WriteIndex = 0;
+            m_ReadIndex = 0;
+            m_ReadCount = 0;
+            memset( (void*)m_Flags, 0, sizeof( m_Flags ) );
+        }
+
+    private:
+        const static uint32_t           ms_cSize        = ( 1 << cSizeLog2 );
+        const static uint32_t           ms_cIndexMask   = ms_cSize - 1;
+        const static uint32_t           FLAG_INVALID    = 0xFFFFFFFF; // 32bit for CAS
+        const static uint32_t           FLAG_CAN_WRITE  = 0x00000000; // 32bit for CAS
+        const static uint32_t           FLAG_CAN_READ   = 0x11111111; // 32bit for CAS
+
+        T                               m_Buffer[ ms_cSize ];
+
+        // read and write indexes allow fast access to the pipe, but actual access
+        // controlled by the access flags. 
+        volatile uint32_t BASE_ALIGN(4) m_WriteIndex;
+        volatile uint32_t BASE_ALIGN(4) m_ReadCount;
+        volatile uint32_t               m_Flags[  ms_cSize ];
+        volatile uint32_t BASE_ALIGN(4) m_ReadIndex;
+    };
+
+    template<uint8_t cSizeLog2, typename T> inline
+        LockLessMultiReadPipe<cSizeLog2,T>::LockLessMultiReadPipe()
+        : m_WriteIndex(0)
+        , m_ReadIndex(0)
+        , m_ReadCount(0)
+    {
+        assert( cSizeLog2 < 32 );
+        memset( (void*)m_Flags, 0, sizeof( m_Flags ) );
+    }
+
+    template<uint8_t cSizeLog2, typename T> inline
+        bool LockLessMultiReadPipe<cSizeLog2,T>::ReaderTryReadBack(   T* pOut )
+    {
+
+        uint32_t actualReadIndex;
+
+        uint32_t readCount  = m_ReadCount;
+
+        // We get hold of read index for consistency,
+        // and do first pass starting at read count
+        uint32_t readIndexToUse  = readCount;
+
+
+        while(true)
+        {
+
+            uint32_t writeIndex = m_WriteIndex;
+            // power of two sizes ensures we can use a simple calc without modulus
+            uint32_t numInPipe = writeIndex - readCount;
+            if( 0 == numInPipe )
+            {
+                return false;
+            }
+            if( readIndexToUse >= writeIndex )
+            {
+                // move back to start
+                readIndexToUse = m_ReadIndex;
+            }
+
+
+            // power of two sizes ensures we can perform AND for a modulus
+            actualReadIndex    = readIndexToUse & ms_cIndexMask;
+
+            // Multiple potential readers mean we should check if the data is valid,
+            // using an atomic compare exchange
+            uint32_t previous = AtomicCompareAndSwap( &m_Flags[  actualReadIndex ], FLAG_INVALID, FLAG_CAN_READ );
+            if( FLAG_CAN_READ == previous )
+            {
+                break;
+            }
+            ++readIndexToUse;
+
+            //update known readcount
+            readCount  = m_ReadCount;
+        }
+
+        // we update the read index using an atomic add, as we've only read one piece of data.
+        // this ensure consistency of the read index, and the above loop ensures readers
+        // only read from unread data
+        AtomicAdd(  (volatile int32_t*)&m_ReadCount, 1 );
+
+        BASE_MEMORYBARRIER_ACQUIRE();
+        // now read data, ensuring we do so after above reads & CAS
+        *pOut = m_Buffer[ actualReadIndex ];
+
+        m_Flags[  actualReadIndex ] = FLAG_CAN_WRITE;
+
+        return true;
+    }
+
+    template<uint8_t cSizeLog2, typename T> inline
+        bool LockLessMultiReadPipe<cSizeLog2,T>::WriterTryReadFront(  T* pOut )
+    {
+        uint32_t writeIndex = m_WriteIndex;
+        uint32_t frontReadIndex  = writeIndex;
+
+        // Multiple potential readers mean we should check if the data is valid,
+        // using an atomic compare exchange - which acts as a form of lock (so not quite lockless really).
+        uint32_t previous = FLAG_INVALID;
+        uint32_t actualReadIndex = 0;
+        while( true )
+        {
+            // power of two sizes ensures we can use a simple calc without modulus
+            uint32_t readCount = m_ReadCount;
+            uint32_t numInPipe = writeIndex - readCount;
+            if( 0 == numInPipe || 0 == frontReadIndex )
+            {
+                // frontReadIndex can get to 0 here if that item was just being read by another thread.
+                m_ReadIndex = readCount;
+                return false;
+            }
+            --frontReadIndex;
+            actualReadIndex = frontReadIndex & ms_cIndexMask;
+            previous = AtomicCompareAndSwap( &m_Flags[  actualReadIndex ], FLAG_INVALID, FLAG_CAN_READ );
+            if( FLAG_CAN_READ == previous )
+            {
+                break;
+            }
+            else if( m_ReadIndex >= frontReadIndex  )
+            {
+                return false;
+            }
+        }
+
+        // now read data, ensuring we do so after above reads & CAS
+        *pOut = m_Buffer[ actualReadIndex ];
+
+        m_Flags[  actualReadIndex ] = FLAG_CAN_WRITE;
+
+        BASE_MEMORYBARRIER_RELEASE();
+
+        // 32-bit aligned stores are atomic, and writer owns the write index
+        // we only move one back as this is as many as we have read, not where we have read from.
+        --m_WriteIndex;
+        return true;
+    }
+
+
+    template<uint8_t cSizeLog2, typename T> inline
+        bool LockLessMultiReadPipe<cSizeLog2,T>::WriterTryWriteFront( const T& in )
+    {
+        // The writer 'owns' the write index, and readers can only reduce
+        // the amount of data in the pipe.
+        // We get hold of both values for consistency and to reduce false sharing
+        // impacting more than one access
+        uint32_t writeIndex = m_WriteIndex;
+
+
+        // power of two sizes ensures we can perform AND for a modulus
+        uint32_t actualWriteIndex    = writeIndex & ms_cIndexMask;
+
+        // a reader may still be reading this item, as there are multiple readers
+        if( m_Flags[ actualWriteIndex ] != FLAG_CAN_WRITE ) 
+        {
+            return false; // still being read, so have caught up with tail. 
+        }
+
+
+        // as we are the only writer we can update the data without atomics
+        //  whilst the write index has not been updated
+        m_Buffer[ actualWriteIndex ] = in;
+        m_Flags[  actualWriteIndex ] = FLAG_CAN_READ;
+
+        // We need to ensure the above writes occur prior to updating the write index,
+        // otherwise another thread might read before it's finished
+        BASE_MEMORYBARRIER_RELEASE();
+
+        // 32-bit aligned stores are atomic, and the writer controls the write index
+        ++writeIndex;
+        m_WriteIndex = writeIndex;
+        return true;
+    }
+
+}

third_party/tracy/examples/ToyPathTracer/Source/enkiTS/TaskScheduler.cpp

@@ -0,0 +1,437 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#include <assert.h>
+
+#include "TaskScheduler.h"
+#include "LockLessMultiReadPipe.h"
+
+
+
+using namespace enki;
+
+
+static const uint32_t PIPESIZE_LOG2              = 8;
+static const uint32_t SPIN_COUNT                 = 100;
+static const uint32_t SPIN_BACKOFF_MULTIPLIER    = 10;
+static const uint32_t MAX_NUM_INITIAL_PARTITIONS = 8;
+
+// each software thread gets it's own copy of gtl_threadNum, so this is safe to use as a static variable
+static THREAD_LOCAL uint32_t                             gtl_threadNum       = 0;
+
+namespace enki 
+{
+	struct SubTaskSet
+	{
+		ITaskSet*           pTask;
+		TaskSetPartition    partition;
+	};
+
+	// we derive class TaskPipe rather than typedef to get forward declaration working easily
+	class TaskPipe : public LockLessMultiReadPipe<PIPESIZE_LOG2,enki::SubTaskSet> {};
+
+	struct ThreadArgs
+	{
+		uint32_t		threadNum;
+		TaskScheduler*  pTaskScheduler;
+	};
+}
+
+namespace
+{
+	SubTaskSet       SplitTask( SubTaskSet& subTask_, uint32_t rangeToSplit_ )
+	{
+		SubTaskSet splitTask = subTask_;
+		uint32_t rangeLeft = subTask_.partition.end - subTask_.partition.start;
+
+        if( rangeToSplit_ > rangeLeft )
+        {
+            rangeToSplit_ = rangeLeft;
+        }
+        splitTask.partition.end = subTask_.partition.start + rangeToSplit_;
+		subTask_.partition.start = splitTask.partition.end;
+		return splitTask;
+	}
+
+	#if defined _WIN32
+		#if defined _M_IX86  || defined _M_X64
+			#pragma intrinsic(_mm_pause)
+			inline void Pause() { _mm_pause(); }
+		#endif
+	#elif defined __i386__ || defined __x86_64__
+		inline void Pause() { __asm__ __volatile__("pause;"); }
+	#else
+		inline void Pause() { ;} // may have NOP or yield equiv
+	#endif
+}
+
+
+static void SafeCallback(ProfilerCallbackFunc func_, uint32_t threadnum_)
+{
+	if( func_ )
+	{
+		func_(threadnum_);
+	}
+}
+
+ProfilerCallbacks* TaskScheduler::GetProfilerCallbacks()
+{
+	return &m_ProfilerCallbacks;
+}
+
+THREADFUNC_DECL TaskScheduler::TaskingThreadFunction( void* pArgs )
+{
+	ThreadArgs args					= *(ThreadArgs*)pArgs;
+	uint32_t threadNum				= args.threadNum;
+	TaskScheduler*  pTS				= args.pTaskScheduler;
+    gtl_threadNum      = threadNum;
+
+	SafeCallback( pTS->m_ProfilerCallbacks.threadStart, threadNum );
+    
+    uint32_t spinCount = 0;
+	uint32_t hintPipeToCheck_io = threadNum + 1;	// does not need to be clamped.
+    while( pTS->m_bRunning )
+    {
+        if(!pTS->TryRunTask( threadNum, hintPipeToCheck_io ) )
+        {
+            // no tasks, will spin then wait
+            ++spinCount;
+            if( spinCount > SPIN_COUNT )
+            {
+				pTS->WaitForTasks( threadNum );
+				spinCount = 0;
+            }
+			else
+			{
+				uint32_t spinBackoffCount = spinCount * SPIN_BACKOFF_MULTIPLIER;
+				while( spinBackoffCount )
+				{
+					Pause();
+					--spinBackoffCount;
+				}
+			}
+        }
+        else
+        {
+            spinCount = 0;
+        }
+    }
+
+    AtomicAdd( &pTS->m_NumThreadsRunning, -1 );
+	SafeCallback( pTS->m_ProfilerCallbacks.threadStop, threadNum );
+
+    return 0;
+}
+
+
+void TaskScheduler::StartThreads()
+{
+    if( m_bHaveThreads )
+    {
+        return;
+    }
+    m_bRunning = true;
+
+    SemaphoreCreate( m_NewTaskSemaphore );
+
+    // we create one less thread than m_NumThreads as the main thread counts as one
+    m_pThreadNumStore = new ThreadArgs[m_NumThreads];
+    m_pThreadIDs      = new threadid_t[m_NumThreads];
+	m_pThreadNumStore[0].threadNum      = 0;
+	m_pThreadNumStore[0].pTaskScheduler = this;
+	m_pThreadIDs[0] = 0;
+    m_NumThreadsWaiting = 0;
+    m_NumThreadsRunning = 1;// acount for main thread
+    for( uint32_t thread = 1; thread < m_NumThreads; ++thread )
+    {
+		m_pThreadNumStore[thread].threadNum      = thread;
+		m_pThreadNumStore[thread].pTaskScheduler = this;
+        ThreadCreate( &m_pThreadIDs[thread], TaskingThreadFunction, &m_pThreadNumStore[thread] );
+        ++m_NumThreadsRunning;
+    }
+
+    // ensure we have sufficient tasks to equally fill either all threads including main
+    // or just the threads we've launched, this is outside the firstinit as we want to be able
+    // to runtime change it
+	if( 1 == m_NumThreads )
+	{
+		m_NumPartitions = 1;
+		m_NumInitialPartitions = 1;
+	}
+	else
+	{
+		m_NumPartitions = m_NumThreads * (m_NumThreads - 1);
+		m_NumInitialPartitions = m_NumThreads - 1;
+		if( m_NumInitialPartitions > MAX_NUM_INITIAL_PARTITIONS )
+		{
+			m_NumInitialPartitions = MAX_NUM_INITIAL_PARTITIONS;
+		}
+	}
+
+    m_bHaveThreads = true;
+}
+
+void TaskScheduler::StopThreads( bool bWait_ )
+{
+    if( m_bHaveThreads )
+    {
+        // wait for them threads quit before deleting data
+        m_bRunning = false;
+        while( bWait_ && m_NumThreadsRunning > 1 )
+        {
+            // keep firing event to ensure all threads pick up state of m_bRunning
+            SemaphoreSignal( m_NewTaskSemaphore, m_NumThreadsRunning );
+        }
+
+        for( uint32_t thread = 1; thread < m_NumThreads; ++thread )
+        {
+            ThreadTerminate( m_pThreadIDs[thread] );
+        }
+
+		m_NumThreads = 0;
+        delete[] m_pThreadNumStore;
+        delete[] m_pThreadIDs;
+        m_pThreadNumStore = 0;
+        m_pThreadIDs = 0;
+        SemaphoreClose( m_NewTaskSemaphore );
+
+        m_bHaveThreads = false;
+		m_NumThreadsWaiting = 0;
+		m_NumThreadsRunning = 0;
+    }
+}
+
+bool TaskScheduler::TryRunTask( uint32_t threadNum, uint32_t& hintPipeToCheck_io_ )
+{
+    // check for tasks
+    SubTaskSet subTask;
+    bool bHaveTask = m_pPipesPerThread[ threadNum ].WriterTryReadFront( &subTask );
+
+	uint32_t threadToCheck = hintPipeToCheck_io_;
+	uint32_t checkCount = 0;
+    while( !bHaveTask && checkCount < m_NumThreads )
+    {
+		threadToCheck = ( hintPipeToCheck_io_ + checkCount ) % m_NumThreads;
+		if( threadToCheck != threadNum )
+		{
+			bHaveTask = m_pPipesPerThread[ threadToCheck ].ReaderTryReadBack( &subTask );
+		}
+		++checkCount;
+    }
+        
+    if( bHaveTask )
+    {
+		// update hint, will preserve value unless actually got task from another thread.
+		hintPipeToCheck_io_ = threadToCheck;
+
+		uint32_t partitionSize = subTask.partition.end - subTask.partition.start;
+		if( subTask.pTask->m_RangeToRun < partitionSize )
+		{
+			SubTaskSet taskToRun = SplitTask( subTask, subTask.pTask->m_RangeToRun );
+			SplitAndAddTask( gtl_threadNum, subTask, subTask.pTask->m_RangeToRun, 0 );
+			taskToRun.pTask->ExecuteRange( taskToRun.partition, threadNum );
+			AtomicAdd( &taskToRun.pTask->m_RunningCount, -1 );
+		}
+		else
+		{
+
+			// the task has already been divided up by AddTaskSetToPipe, so just run it
+			subTask.pTask->ExecuteRange( subTask.partition, threadNum );
+			AtomicAdd( &subTask.pTask->m_RunningCount, -1 );
+		}
+    }
+
+    return bHaveTask;
+
+}
+
+void TaskScheduler::WaitForTasks( uint32_t threadNum )
+{
+	// We incrememt the number of threads waiting here in order
+	// to ensure that the check for tasks occurs after the increment
+	// to prevent a task being added after a check, then the thread waiting.
+	// This will occasionally result in threads being mistakenly awoken,
+	// but they will then go back to sleep.
+	AtomicAdd( &m_NumThreadsWaiting, 1 );
+
+    bool bHaveTasks = false;
+    for( uint32_t thread = 0; thread < m_NumThreads; ++thread )
+    {
+        if( !m_pPipesPerThread[ thread ].IsPipeEmpty() )
+        {
+            bHaveTasks = true;
+            break;
+        }
+    }
+    if( !bHaveTasks )
+    {
+        SafeCallback( m_ProfilerCallbacks.waitStart, threadNum );
+        SemaphoreWait( m_NewTaskSemaphore );
+        SafeCallback( m_ProfilerCallbacks.waitStop, threadNum );
+    }
+
+    int32_t prev = AtomicAdd( &m_NumThreadsWaiting, -1 );
+    assert( prev != 0 );
+}
+
+void TaskScheduler::WakeThreads()
+{
+	SemaphoreSignal( m_NewTaskSemaphore, m_NumThreadsWaiting );
+}
+
+void TaskScheduler::SplitAndAddTask( uint32_t threadNum_, SubTaskSet subTask_,
+	uint32_t rangeToSplit_, int32_t runningCountOffset_ )
+{
+    int32_t numAdded = 0;
+    while( subTask_.partition.start != subTask_.partition.end )
+    {
+        SubTaskSet taskToAdd = SplitTask( subTask_, rangeToSplit_ );
+
+        // add the partition to the pipe
+        ++numAdded;
+        if( !m_pPipesPerThread[ gtl_threadNum ].WriterTryWriteFront( taskToAdd ) )
+        {
+			if( numAdded > 1 )
+			{
+				WakeThreads();
+			}
+			// alter range to run the appropriate fraction
+			if( taskToAdd.pTask->m_RangeToRun < rangeToSplit_ )
+			{
+				taskToAdd.partition.end = taskToAdd.partition.start + taskToAdd.pTask->m_RangeToRun;
+				subTask_.partition.start = taskToAdd.partition.end;
+			}
+            taskToAdd.pTask->ExecuteRange( taskToAdd.partition, threadNum_ );
+            --numAdded;
+        }
+    }
+
+    // increment running count by number added
+    AtomicAdd( &subTask_.pTask->m_RunningCount, numAdded + runningCountOffset_ );
+
+	WakeThreads();
+}
+
+void    TaskScheduler::AddTaskSetToPipe( ITaskSet* pTaskSet )
+{
+	// set running count to -1 to guarantee it won't be found complete until all subtasks added
+    pTaskSet->m_RunningCount = -1;
+
+    // divide task up and add to pipe
+    pTaskSet->m_RangeToRun = pTaskSet->m_SetSize / m_NumPartitions;
+    if( pTaskSet->m_RangeToRun < pTaskSet->m_MinRange ) { pTaskSet->m_RangeToRun = pTaskSet->m_MinRange; }
+
+	uint32_t rangeToSplit = pTaskSet->m_SetSize / m_NumInitialPartitions;
+	if( rangeToSplit < pTaskSet->m_MinRange ) { rangeToSplit = pTaskSet->m_MinRange; }
+
+    SubTaskSet subTask;
+    subTask.pTask = pTaskSet;
+    subTask.partition.start = 0;
+    subTask.partition.end = pTaskSet->m_SetSize;
+	SplitAndAddTask( gtl_threadNum, subTask, rangeToSplit, 1 );
+}
+
+void    TaskScheduler::WaitforTaskSet( const ITaskSet* pTaskSet )
+{
+	uint32_t hintPipeToCheck_io = gtl_threadNum + 1;	// does not need to be clamped.
+	if( pTaskSet )
+	{
+		while( pTaskSet->m_RunningCount )
+		{
+			TryRunTask( gtl_threadNum, hintPipeToCheck_io );
+			// should add a spin then wait for task completion event.
+		}
+	}
+	else
+	{
+			TryRunTask( gtl_threadNum, hintPipeToCheck_io );
+	}
+}
+
+void    TaskScheduler::WaitforAll()
+{
+    bool bHaveTasks = true;
+ 	uint32_t hintPipeToCheck_io = gtl_threadNum  + 1;	// does not need to be clamped.
+	int32_t threadsRunning = m_NumThreadsRunning - 1;
+    while( bHaveTasks || m_NumThreadsWaiting < threadsRunning )
+    {
+        TryRunTask( gtl_threadNum, hintPipeToCheck_io );
+        bHaveTasks = false;
+        for( uint32_t thread = 0; thread < m_NumThreads; ++thread )
+        {
+            if( !m_pPipesPerThread[ thread ].IsPipeEmpty() )
+            {
+                bHaveTasks = true;
+                break;
+            }
+        }
+     }
+}
+
+void    TaskScheduler::WaitforAllAndShutdown()
+{
+    WaitforAll();
+    StopThreads(true);
+	delete[] m_pPipesPerThread;
+    m_pPipesPerThread = 0;
+}
+
+uint32_t        TaskScheduler::GetNumTaskThreads() const
+{
+    return m_NumThreads;
+}
+
+TaskScheduler::TaskScheduler()
+		: m_pPipesPerThread(NULL)
+		, m_NumThreads(0)
+		, m_pThreadNumStore(NULL)
+		, m_pThreadIDs(NULL)
+		, m_bRunning(false)
+		, m_NumThreadsRunning(0)
+		, m_NumThreadsWaiting(0)
+		, m_NumPartitions(0)
+		, m_bHaveThreads(false)
+{
+	memset(&m_ProfilerCallbacks, 0, sizeof(m_ProfilerCallbacks));
+}
+
+TaskScheduler::~TaskScheduler()
+{
+    StopThreads( true ); // Stops threads, waiting for them.
+
+    delete[] m_pPipesPerThread;
+    m_pPipesPerThread = 0;
+}
+
+void    TaskScheduler::Initialize( uint32_t numThreads_ )
+{
+	assert( numThreads_ );
+    StopThreads( true ); // Stops threads, waiting for them.
+    delete[] m_pPipesPerThread;
+
+	m_NumThreads = numThreads_;
+
+    m_pPipesPerThread = new TaskPipe[ m_NumThreads ];
+
+    StartThreads();
+}
+
+void   TaskScheduler::Initialize()
+{
+	Initialize( GetNumHardwareThreads() );
+}

third_party/tracy/examples/ToyPathTracer/Source/enkiTS/TaskScheduler.h

@@ -0,0 +1,177 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#pragma once
+
+#include <stdint.h>
+#include "Threads.h"
+
+namespace enki
+{
+
+	struct TaskSetPartition
+	{
+		uint32_t start;
+		uint32_t end;
+	};
+
+	class  TaskScheduler;
+	class  TaskPipe;
+	struct ThreadArgs;
+	struct SubTaskSet;
+
+	// Subclass ITaskSet to create tasks.
+	// TaskSets can be re-used, but check
+	class ITaskSet
+	{
+	public:
+        ITaskSet()
+            : m_SetSize(1)
+			, m_MinRange(1)
+            , m_RunningCount(0)
+			, m_RangeToRun(1)
+        {}
+
+        ITaskSet( uint32_t setSize_ )
+            : m_SetSize( setSize_ )
+			, m_MinRange(1)
+            , m_RunningCount(0)
+			, m_RangeToRun(1)
+        {}
+
+		ITaskSet( uint32_t setSize_, uint32_t minRange_ )
+            : m_SetSize( setSize_ )
+			, m_MinRange( minRange_ )
+            , m_RunningCount(0)
+			, m_RangeToRun(minRange_)
+        {}
+
+		// Execute range should be overloaded to process tasks. It will be called with a
+		// range_ where range.start >= 0; range.start < range.end; and range.end < m_SetSize;
+		// The range values should be mapped so that linearly processing them in order is cache friendly
+		// i.e. neighbouring values should be close together.
+		// threadnum should not be used for changing processing of data, it's intended purpose
+		// is to allow per-thread data buckets for output.
+		virtual void            ExecuteRange( TaskSetPartition range, uint32_t threadnum  ) = 0;
+
+		// Size of set - usually the number of data items to be processed, see ExecuteRange. Defaults to 1
+		uint32_t                m_SetSize;
+
+		// Minimum size of of TaskSetPartition range when splitting a task set into partitions.
+		// This should be set to a value which results in computation effort of at least 10k
+		// clock cycles to minimize tast scheduler overhead.
+		// NOTE: The last partition will be smaller than m_MinRange if m_SetSize is not a multiple
+		// of m_MinRange.
+		// Also known as grain size in literature.
+		uint32_t                m_MinRange;
+
+		bool                    GetIsComplete()
+		{
+			return 0 == m_RunningCount;
+		}
+	private:
+		friend class           TaskScheduler;
+		volatile int32_t        m_RunningCount;
+		uint32_t                m_RangeToRun;
+	};
+
+	// TaskScheduler implements several callbacks intended for profilers
+	typedef void (*ProfilerCallbackFunc)( uint32_t threadnum_ );
+	struct ProfilerCallbacks
+	{
+		ProfilerCallbackFunc threadStart;
+		ProfilerCallbackFunc threadStop;
+		ProfilerCallbackFunc waitStart;
+		ProfilerCallbackFunc waitStop;
+	};
+
+	class TaskScheduler
+	{
+	public:
+		TaskScheduler();
+		~TaskScheduler();
+
+		// Call either Initialize() or Initialize( numThreads_ ) before adding tasks.
+
+		// Initialize() will create GetNumHardwareThreads()-1 threads, which is
+		// sufficient to fill the system when including the main thread.
+		// Initialize can be called multiple times - it will wait for completion
+		// before re-initializing.
+		void			Initialize();
+
+		// Initialize( numThreads_ ) - numThreads_ (must be > 0)
+		// will create numThreads_-1 threads, as thread 0 is
+		// the thread on which the initialize was called.
+		void			Initialize( uint32_t numThreads_ );
+
+
+		// Adds the TaskSet to pipe and returns if the pipe is not full.
+		// If the pipe is full, pTaskSet is run.
+		// should only be called from main thread, or within a task
+		void            AddTaskSetToPipe( ITaskSet* pTaskSet );
+
+		// Runs the TaskSets in pipe until true == pTaskSet->GetIsComplete();
+		// should only be called from thread which created the taskscheduler , or within a task
+		// if called with 0 it will try to run tasks, and return if none available.
+		void            WaitforTaskSet( const ITaskSet* pTaskSet );
+
+		// Waits for all task sets to complete - not guaranteed to work unless we know we
+		// are in a situation where tasks aren't being continuosly added.
+		void            WaitforAll();
+
+		// Waits for all task sets to complete and shutdown threads - not guaranteed to work unless we know we
+		// are in a situation where tasks aren't being continuosly added.
+		void            WaitforAllAndShutdown();
+
+		// Returns the number of threads created for running tasks + 1
+		// to account for the main thread.
+		uint32_t        GetNumTaskThreads() const;
+
+		// Returns the ProfilerCallbacks structure so that it can be modified to
+		// set the callbacks.
+		ProfilerCallbacks* GetProfilerCallbacks();
+
+	private:
+		static THREADFUNC_DECL  TaskingThreadFunction( void* pArgs );
+        void             WaitForTasks( uint32_t threadNum );
+		bool             TryRunTask( uint32_t threadNum, uint32_t& hintPipeToCheck_io_ );
+		void             StartThreads();
+		void             StopThreads( bool bWait_ );
+		void             SplitAndAddTask( uint32_t threadNum_, SubTaskSet subTask_,
+										  uint32_t rangeToSplit_, int32_t runningCountOffset_ );
+		void             WakeThreads();
+
+		TaskPipe*                                                m_pPipesPerThread;
+
+		uint32_t                                                 m_NumThreads;
+		ThreadArgs*                                              m_pThreadNumStore;
+		threadid_t*                                              m_pThreadIDs;
+		volatile bool                                            m_bRunning;
+		volatile int32_t                                         m_NumThreadsRunning;
+		volatile int32_t                                         m_NumThreadsWaiting;
+		uint32_t                                                 m_NumPartitions;
+		uint32_t                                                 m_NumInitialPartitions;
+		semaphoreid_t                                            m_NewTaskSemaphore;
+		bool                                                     m_bHaveThreads;
+		ProfilerCallbacks										 m_ProfilerCallbacks;
+
+		TaskScheduler( const TaskScheduler& nocopy );
+		TaskScheduler& operator=( const TaskScheduler& nocopy );
+	};
+
+}

third_party/tracy/examples/ToyPathTracer/Source/enkiTS/TaskScheduler_c.cpp

@@ -0,0 +1,122 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#include "TaskScheduler_c.h"
+#include "TaskScheduler.h"
+
+#include <assert.h>
+
+using namespace enki;
+
+struct enkiTaskScheduler : TaskScheduler
+{
+};
+
+struct enkiTaskSet : ITaskSet
+{
+	enkiTaskSet( enkiTaskExecuteRange taskFun_ ) : taskFun(taskFun_), pArgs(NULL) {}
+
+	virtual void ExecuteRange( TaskSetPartition range, uint32_t threadnum  )
+	{
+		taskFun( range.start, range.end, threadnum, pArgs );
+	}
+
+	enkiTaskExecuteRange taskFun;
+	void* pArgs;
+};
+
+enkiTaskScheduler*	enkiNewTaskScheduler()
+{
+	enkiTaskScheduler* pETS = new enkiTaskScheduler();
+    return pETS;
+}
+
+void	            enkiInitTaskScheduler(  enkiTaskScheduler* pETS_ )
+{
+    pETS_->Initialize();
+}
+
+void	            enkiInitTaskSchedulerNumThreads(  enkiTaskScheduler* pETS_, uint32_t numThreads_ )
+{
+    pETS_->Initialize( numThreads_ );
+}
+
+void				enkiDeleteTaskScheduler( enkiTaskScheduler* pETS_ )
+{
+	delete pETS_;
+}
+
+enkiTaskSet*		enkiCreateTaskSet( enkiTaskScheduler* pETS_, enkiTaskExecuteRange taskFunc_  )
+{
+	return new enkiTaskSet( taskFunc_ );
+}
+
+void                enkiDeleteTaskSet( enkiTaskSet* pTaskSet_ )
+{
+	delete pTaskSet_;
+}
+
+void				enkiAddTaskSetToPipe( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_, void* pArgs_, uint32_t setSize_ )
+{
+	assert( pTaskSet_ );
+	assert( pTaskSet_->taskFun );
+
+	pTaskSet_->m_SetSize = setSize_;
+	pTaskSet_->pArgs = pArgs_;
+	pETS_->AddTaskSetToPipe( pTaskSet_ );
+}
+
+void enkiAddTaskSetToPipeMinRange(enkiTaskScheduler * pETS_, enkiTaskSet * pTaskSet_, void * pArgs_, uint32_t setSize_, uint32_t minRange_)
+{
+	assert( pTaskSet_ );
+	assert( pTaskSet_->taskFun );
+
+	pTaskSet_->m_SetSize = setSize_;
+	pTaskSet_->m_MinRange = minRange_;
+	pTaskSet_->pArgs = pArgs_;
+	pETS_->AddTaskSetToPipe( pTaskSet_ );
+}
+
+int				enkiIsTaskSetComplete( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_ )
+{
+	assert( pTaskSet_ );
+	return ( pTaskSet_->GetIsComplete() ) ? 1 : 0;
+}
+
+void				enkiWaitForTaskSet( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_ )
+{
+	pETS_->WaitforTaskSet( pTaskSet_ );
+}
+
+void				enkiWaitForAll( enkiTaskScheduler* pETS_ )
+{
+	pETS_->WaitforAll();
+}
+
+
+uint32_t			enkiGetNumTaskThreads( enkiTaskScheduler* pETS_ )
+{
+	return pETS_->GetNumTaskThreads();
+}
+
+enkiProfilerCallbacks*	enkiGetProfilerCallbacks( enkiTaskScheduler* pETS_ )
+{
+    assert( sizeof(enkiProfilerCallbacks) == sizeof(enki::ProfilerCallbacks) );
+    return (enkiProfilerCallbacks*)pETS_->GetProfilerCallbacks();
+}
+

third_party/tracy/examples/ToyPathTracer/Source/enkiTS/TaskScheduler_c.h

@@ -0,0 +1,104 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#pragma once
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include <stdint.h>
+
+typedef struct enkiTaskScheduler enkiTaskScheduler;
+typedef struct enkiTaskSet		 enkiTaskSet;
+
+typedef void (* enkiTaskExecuteRange)( uint32_t start_, uint32_t end, uint32_t threadnum_, void* pArgs_ );
+
+
+// Create a new task scheduler
+enkiTaskScheduler*	enkiNewTaskScheduler();
+
+// Initialize task scheduler - will create GetNumHardwareThreads()-1 threads, which is
+// sufficient to fill the system when including the main thread.
+// Initialize can be called multiple times - it will wait for completion
+// before re-initializing.
+void	            enkiInitTaskScheduler(  enkiTaskScheduler* pETS_ );
+
+// Initialize a task scheduler with numThreads_ (must be > 0)
+// will create numThreads_-1 threads, as thread 0 is
+// the thread on which the initialize was called.
+void	            enkiInitTaskSchedulerNumThreads(  enkiTaskScheduler* pETS_, uint32_t numThreads_ );
+
+
+// Delete a task scheduler
+void				enkiDeleteTaskScheduler( enkiTaskScheduler* pETS_ );
+
+// Create a task set.
+enkiTaskSet*		enkiCreateTaskSet( enkiTaskScheduler* pETS_, enkiTaskExecuteRange taskFunc_  );
+
+// Delete a task set.
+void                enkiDeleteTaskSet( enkiTaskSet* pTaskSet_ );
+
+// Schedule the task
+void				enkiAddTaskSetToPipe( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_,
+										   void* pArgs_, uint32_t setSize_ );
+
+// Schedule the task with a minimum range.
+// This should be set to a value which results in computation effort of at least 10k
+// clock cycles to minimize tast scheduler overhead.
+// NOTE: The last partition will be smaller than m_MinRange if m_SetSize is not a multiple
+// of m_MinRange.
+// Also known as grain size in literature.
+void				enkiAddTaskSetToPipeMinRange( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_,
+												  void* pArgs_, uint32_t setSize_, uint32_t minRange_ );
+
+
+// Check if TaskSet is complete. Doesn't wait. Returns 1 if complete, 0 if not.
+int					enkiIsTaskSetComplete( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_ );
+
+
+// Wait for a given task.
+// should only be called from thread which created the taskscheduler , or within a task
+// if called with 0 it will try to run tasks, and return if none available.
+void				enkiWaitForTaskSet( enkiTaskScheduler* pETS_, enkiTaskSet* pTaskSet_ );
+
+
+// Waits for all task sets to complete - not guaranteed to work unless we know we
+// are in a situation where tasks aren't being continuosly added.
+void				enkiWaitForAll( enkiTaskScheduler* pETS_ );
+
+
+// get number of threads
+uint32_t			enkiGetNumTaskThreads( enkiTaskScheduler* pETS_ );
+
+// TaskScheduler implements several callbacks intended for profilers
+typedef void (*enkiProfilerCallbackFunc)( uint32_t threadnum_ );
+struct enkiProfilerCallbacks
+{
+    enkiProfilerCallbackFunc threadStart;
+    enkiProfilerCallbackFunc threadStop;
+    enkiProfilerCallbackFunc waitStart;
+    enkiProfilerCallbackFunc waitStop;
+};
+
+// Get the callback structure so it can be set 
+struct enkiProfilerCallbacks*	enkiGetProfilerCallbacks( enkiTaskScheduler* pETS_ );
+
+#ifdef __cplusplus
+}
+#endif

third_party/tracy/examples/ToyPathTracer/Source/enkiTS/Threads.h

@@ -0,0 +1,210 @@
+// Copyright (c) 2013 Doug Binks
+// 
+// This software is provided 'as-is', without any express or implied
+// warranty. In no event will the authors be held liable for any damages
+// arising from the use of this software.
+// 
+// Permission is granted to anyone to use this software for any purpose,
+// including commercial applications, and to alter it and redistribute it
+// freely, subject to the following restrictions:
+// 
+// 1. The origin of this software must not be misrepresented; you must not
+//    claim that you wrote the original software. If you use this software
+//    in a product, an acknowledgement in the product documentation would be
+//    appreciated but is not required.
+// 2. Altered source versions must be plainly marked as such, and must not be
+//    misrepresented as being the original software.
+// 3. This notice may not be removed or altered from any source distribution.
+
+#pragma once
+
+#include <stdint.h>
+#include <assert.h>
+
+#ifdef _WIN32
+
+	#include "Atomics.h"
+
+	#define WIN32_LEAN_AND_MEAN
+	#include <Windows.h>
+	
+	#define THREADFUNC_DECL DWORD WINAPI
+	#define THREAD_LOCAL __declspec( thread )
+
+namespace enki
+{
+    typedef HANDLE threadid_t;
+
+    // declare the thread start function as:
+    // THREADFUNC_DECL MyThreadStart( void* pArg );
+    inline bool ThreadCreate( threadid_t* returnid, DWORD ( WINAPI *StartFunc) (void* ), void* pArg )
+    {
+        // posix equiv pthread_create
+        DWORD threadid;
+        *returnid = CreateThread( 0, 0, StartFunc, pArg, 0, &threadid );
+        return  *returnid != NULL;
+    }
+
+    inline bool ThreadTerminate( threadid_t threadid )
+    {
+        // posix equiv pthread_cancel
+        return CloseHandle( threadid ) == 0;
+    }
+
+    inline uint32_t GetNumHardwareThreads()
+    {
+        SYSTEM_INFO sysInfo;
+        GetSystemInfo(&sysInfo);
+        return sysInfo.dwNumberOfProcessors;
+    }
+}
+
+#else // posix
+
+	#include <pthread.h>
+	#include <unistd.h>
+	#define THREADFUNC_DECL void*
+	#define THREAD_LOCAL __thread
+
+namespace enki
+{
+    typedef pthread_t threadid_t;  
+        
+    // declare the thread start function as:
+    // THREADFUNC_DECL MyThreadStart( void* pArg );
+    inline bool ThreadCreate( threadid_t* returnid, void* ( *StartFunc) (void* ), void* pArg )
+    {
+        // posix equiv pthread_create
+        int32_t retval = pthread_create( returnid, NULL, StartFunc, pArg );
+
+        return  retval == 0;
+    }
+    
+    inline bool ThreadTerminate( threadid_t threadid )
+    {
+        // posix equiv pthread_cancel
+        return pthread_cancel( threadid ) == 0;
+    }
+    
+    inline uint32_t GetNumHardwareThreads()
+    {
+        return (uint32_t)sysconf( _SC_NPROCESSORS_ONLN );
+    }
+}
+
+#endif // posix
+
+
+// Semaphore implementation
+#ifdef _WIN32
+
+namespace enki
+{
+    struct semaphoreid_t
+    {
+        HANDLE      sem;
+    };
+	
+	inline void SemaphoreCreate( semaphoreid_t& semaphoreid )
+    {
+        semaphoreid.sem = CreateSemaphore(NULL, 0, MAXLONG, NULL );
+    }
+
+    inline void SemaphoreClose( semaphoreid_t& semaphoreid )
+    {
+        CloseHandle( semaphoreid.sem );
+    }
+
+    inline void SemaphoreWait( semaphoreid_t& semaphoreid  )
+    {
+        DWORD retval = WaitForSingleObject( semaphoreid.sem, INFINITE );
+
+        assert( retval != WAIT_FAILED );
+    }
+
+    inline void SemaphoreSignal( semaphoreid_t& semaphoreid, int32_t countWaiting )
+    {
+		if( countWaiting )
+		{
+			ReleaseSemaphore( semaphoreid.sem, countWaiting, NULL );
+		}
+    }
+}
+#elif defined(__MACH__)
+
+// OS X does not have POSIX semaphores
+// see https://developer.apple.com/library/content/documentation/Darwin/Conceptual/KernelProgramming/synchronization/synchronization.html
+#include <mach/mach.h>
+
+namespace enki
+{
+    
+    struct semaphoreid_t
+    {
+        semaphore_t   sem;
+    };
+	
+	inline void SemaphoreCreate( semaphoreid_t& semaphoreid )
+    {
+		semaphore_create( mach_task_self(), &semaphoreid.sem, SYNC_POLICY_FIFO, 0 );
+    }
+    
+    inline void SemaphoreClose( semaphoreid_t& semaphoreid )
+    {
+        semaphore_destroy( mach_task_self(), semaphoreid.sem );
+    }
+    
+    inline void SemaphoreWait( semaphoreid_t& semaphoreid  )
+    {
+        semaphore_wait( semaphoreid.sem );
+    }
+    
+    inline void SemaphoreSignal( semaphoreid_t& semaphoreid, int32_t countWaiting )
+    {
+        while( countWaiting-- > 0 )
+		{
+			semaphore_signal( semaphoreid.sem );
+		}
+    }
+}
+
+#else // POSIX
+
+#include <semaphore.h>
+
+namespace enki
+{
+    
+    struct semaphoreid_t
+    {
+        sem_t   sem;
+    };
+	
+	inline void SemaphoreCreate( semaphoreid_t& semaphoreid )
+    {
+		int err = sem_init( &semaphoreid.sem, 0, 0 );
+		assert( err == 0 );
+    }
+    
+    inline void SemaphoreClose( semaphoreid_t& semaphoreid )
+    {
+        sem_destroy( &semaphoreid.sem );
+    }
+    
+    inline void SemaphoreWait( semaphoreid_t& semaphoreid  )
+    {
+        int err = sem_wait( &semaphoreid.sem );
+		assert( err == 0 );
+    }
+    
+    inline void SemaphoreSignal( semaphoreid_t& semaphoreid, int32_t countWaiting )
+    {
+        while( countWaiting-- > 0 )
+		{
+			sem_post( &semaphoreid.sem );
+		}
+    }
+}
+#endif
+
+

third_party/tracy/examples/ToyPathTracer/Windows/ComputeShader.hlsl

@@ -0,0 +1,395 @@
+#include "../Source/Config.h"
+
+inline uint RNG(inout uint state)
+{
+    uint x = state;
+    x ^= x << 13;
+    x ^= x >> 17;
+    x ^= x << 15;
+    state = x;
+    return x;
+}
+
+float RandomFloat01(inout uint state)
+{
+    return (RNG(state) & 0xFFFFFF) / 16777216.0f;
+}
+
+float3 RandomInUnitDisk(inout uint state)
+{
+    float a = RandomFloat01(state) * 2.0f * 3.1415926f;
+    float2 xy = float2(cos(a), sin(a));
+    xy *= sqrt(RandomFloat01(state));
+    return float3(xy, 0);
+}
+float3 RandomInUnitSphere(inout uint state)
+{
+    float z = RandomFloat01(state) * 2.0f - 1.0f;
+    float t = RandomFloat01(state) * 2.0f * 3.1415926f;
+    float r = sqrt(max(0.0, 1.0f - z * z));
+    float x = r * cos(t);
+    float y = r * sin(t);
+    float3 res = float3(x, y, z);
+    res *= pow(RandomFloat01(state), 1.0 / 3.0);
+    return res;
+}
+float3 RandomUnitVector(inout uint state)
+{
+    float z = RandomFloat01(state) * 2.0f - 1.0f;
+    float a = RandomFloat01(state) * 2.0f * 3.1415926f;
+    float r = sqrt(1.0f - z * z);
+    float x = r * cos(a);
+    float y = r * sin(a);
+    return float3(x, y, z);
+}
+
+
+
+struct Ray
+{
+    float3 orig;
+    float3 dir;
+};
+Ray MakeRay(float3 orig_, float3 dir_) { Ray r; r.orig = orig_; r.dir = dir_; return r; }
+float3 RayPointAt(Ray r, float t) { return r.orig + r.dir * t; }
+
+
+inline bool refract(float3 v, float3 n, float nint, out float3 outRefracted)
+{
+    float dt = dot(v, n);
+    float discr = 1.0f - nint * nint*(1 - dt * dt);
+    if (discr > 0)
+    {
+        outRefracted = nint * (v - n * dt) - n * sqrt(discr);
+        return true;
+    }
+    return false;
+}
+inline float schlick(float cosine, float ri)
+{
+    float r0 = (1 - ri) / (1 + ri);
+    r0 = r0 * r0;
+    // note: saturate to guard against possible tiny negative numbers
+    return r0 + (1 - r0)*pow(saturate(1 - cosine), 5);
+}
+
+struct Hit
+{
+    float3 pos;
+    float3 normal;
+    float t;
+};
+
+struct Sphere
+{
+    float3 center;
+    float radius;
+    float invRadius;
+};
+
+#define MatLambert 0
+#define MatMetal 1
+#define MatDielectric 2
+
+struct Material
+{
+    int type;
+    float3 albedo;
+    float3 emissive;
+    float roughness;
+    float ri;
+};
+
+groupshared Sphere s_GroupSpheres[kCSMaxObjects];
+groupshared Material s_GroupMaterials[kCSMaxObjects];
+groupshared int s_GroupEmissives[kCSMaxObjects];
+
+
+struct Camera
+{
+    float3 origin;
+    float3 lowerLeftCorner;
+    float3 horizontal;
+    float3 vertical;
+    float3 u, v, w;
+    float lensRadius;
+};
+
+Ray CameraGetRay(Camera cam, float s, float t, inout uint state)
+{
+    float3 rd = cam.lensRadius * RandomInUnitDisk(state);
+    float3 offset = cam.u * rd.x + cam.v * rd.y;
+    return MakeRay(cam.origin + offset, normalize(cam.lowerLeftCorner + s * cam.horizontal + t * cam.vertical - cam.origin - offset));
+}
+
+
+int HitSpheres(Ray r, int sphereCount, float tMin, float tMax, inout Hit outHit)
+{
+    float hitT = tMax;
+    int id = -1;
+    for (int i = 0; i < sphereCount; ++i)
+    {
+        Sphere s = s_GroupSpheres[i];
+        float3 co = s.center - r.orig;
+        float nb = dot(co, r.dir);
+        float c = dot(co, co) - s.radius*s.radius;
+        float discr = nb * nb - c;
+        if (discr > 0)
+        {
+            float discrSq = sqrt(discr);
+
+            // Try earlier t
+            float t = nb - discrSq;
+            if (t <= tMin) // before min, try later t!
+                t = nb + discrSq;
+
+            if (t > tMin && t < hitT)
+            {
+                id = i;
+                hitT = t;
+            }
+        }
+    }
+
+    if (id != -1)
+    {
+        outHit.pos = RayPointAt(r, hitT);
+        outHit.normal = (outHit.pos - s_GroupSpheres[id].center) * s_GroupSpheres[id].invRadius;
+        outHit.t = hitT;
+    }
+    return id;
+}
+
+struct Params
+{
+    Camera cam;
+    int sphereCount;
+    int screenWidth;
+    int screenHeight;
+    int frames;
+    float invWidth;
+    float invHeight;
+    float lerpFac;
+    int emissiveCount;
+};
+
+
+#define kMinT 0.001f
+#define kMaxT 1.0e7f
+#define kMaxDepth 10
+
+
+static int HitWorld(int sphereCount, Ray r, float tMin, float tMax, inout Hit outHit)
+{
+    return HitSpheres(r, sphereCount, tMin, tMax, outHit);
+}
+
+
+static bool Scatter(int sphereCount, int emissiveCount, int matID, Ray r_in, Hit rec, out float3 attenuation, out Ray scattered, out float3 outLightE, inout int inoutRayCount, inout uint state)
+{
+    outLightE = float3(0, 0, 0);
+    Material mat = s_GroupMaterials[matID];
+    if (mat.type == MatLambert)
+    {
+        // random point on unit sphere that is tangent to the hit point
+        float3 target = rec.pos + rec.normal + RandomUnitVector(state);
+        scattered = MakeRay(rec.pos, normalize(target - rec.pos));
+        attenuation = mat.albedo;
+
+        // sample lights
+#if DO_LIGHT_SAMPLING
+        for (int j = 0; j < emissiveCount; ++j)
+        {
+            int i = s_GroupEmissives[j];
+            if (matID == i)
+                continue; // skip self
+            Material smat = s_GroupMaterials[i];
+            Sphere s = s_GroupSpheres[i];
+
+            // create a random direction towards sphere
+            // coord system for sampling: sw, su, sv
+            float3 sw = normalize(s.center - rec.pos);
+            float3 su = normalize(cross(abs(sw.x)>0.01f ? float3(0, 1, 0) : float3(1, 0, 0), sw));
+            float3 sv = cross(sw, su);
+            // sample sphere by solid angle
+            float cosAMax = sqrt(1.0f - s.radius*s.radius / dot(rec.pos - s.center, rec.pos - s.center));
+            float eps1 = RandomFloat01(state), eps2 = RandomFloat01(state);
+            float cosA = 1.0f - eps1 + eps1 * cosAMax;
+            float sinA = sqrt(1.0f - cosA * cosA);
+            float phi = 2 * 3.1415926 * eps2;
+            float3 l = su * cos(phi) * sinA + sv * sin(phi) * sinA + sw * cosA;
+
+            // shoot shadow ray
+            Hit lightHit;
+            ++inoutRayCount;
+            int hitID = HitWorld(sphereCount, MakeRay(rec.pos, l), kMinT, kMaxT, lightHit);
+            if (hitID == i)
+            {
+                float omega = 2 * 3.1415926 * (1 - cosAMax);
+
+                float3 rdir = r_in.dir;
+                float3 nl = dot(rec.normal, rdir) < 0 ? rec.normal : -rec.normal;
+                outLightE += (mat.albedo * smat.emissive) * (max(0.0f, dot(l, nl)) * omega / 3.1415926);
+            }
+        }
+#endif
+        return true;
+    }
+    else if (mat.type == MatMetal)
+    {
+        float3 refl = reflect(r_in.dir, rec.normal);
+        // reflected ray, and random inside of sphere based on roughness
+        float roughness = mat.roughness;
+#if DO_MITSUBA_COMPARE
+        roughness = 0; // until we get better BRDF for metals
+#endif
+        scattered = MakeRay(rec.pos, normalize(refl + roughness*RandomInUnitSphere(state)));
+        attenuation = mat.albedo;
+        return dot(scattered.dir, rec.normal) > 0;
+    }
+    else if (mat.type == MatDielectric)
+    {
+        float3 outwardN;
+        float3 rdir = r_in.dir;
+        float3 refl = reflect(rdir, rec.normal);
+        float nint;
+        attenuation = float3(1, 1, 1);
+        float3 refr;
+        float reflProb;
+        float cosine;
+        if (dot(rdir, rec.normal) > 0)
+        {
+            outwardN = -rec.normal;
+            nint = mat.ri;
+            cosine = mat.ri * dot(rdir, rec.normal);
+        }
+        else
+        {
+            outwardN = rec.normal;
+            nint = 1.0f / mat.ri;
+            cosine = -dot(rdir, rec.normal);
+        }
+        if (refract(rdir, outwardN, nint, refr))
+        {
+            reflProb = schlick(cosine, mat.ri);
+        }
+        else
+        {
+            reflProb = 1;
+        }
+        if (RandomFloat01(state) < reflProb)
+            scattered = MakeRay(rec.pos, normalize(refl));
+        else
+            scattered = MakeRay(rec.pos, normalize(refr));
+    }
+    else
+    {
+        attenuation = float3(1, 0, 1);
+        scattered = MakeRay(float3(0,0,0), float3(0, 0, 1));
+        return false;
+    }
+    return true;
+}
+
+static float3 Trace(int sphereCount, int emissiveCount, Ray r, inout int inoutRayCount, inout uint state)
+{
+    float3 col = 0;
+    float3 curAtten = 1;
+    bool doMaterialE = true;
+    // GPUs don't support recursion, so do tracing iterations in a loop up to max depth
+    for (int depth = 0; depth < kMaxDepth; ++depth)
+    {
+        Hit rec;
+        ++inoutRayCount;
+        int id = HitWorld(sphereCount, r, kMinT, kMaxT, rec);
+        if (id >= 0)
+        {
+            Ray scattered;
+            float3 attenuation;
+            float3 lightE;
+            Material mat = s_GroupMaterials[id];
+            float3 matE = mat.emissive;
+            if (Scatter(sphereCount, emissiveCount, id, r, rec, attenuation, scattered, lightE, inoutRayCount, state))
+            {
+#if DO_LIGHT_SAMPLING
+                if (!doMaterialE) matE = 0;
+                doMaterialE = (mat.type != MatLambert);
+#endif
+                col += curAtten * (matE + lightE);
+                curAtten *= attenuation;
+                r = scattered;
+            }
+            else
+            {
+                col += curAtten * matE;
+                break;
+            }
+        }
+        else
+        {
+            // sky
+#if DO_MITSUBA_COMPARE
+            col += curAtten * float3(0.15f, 0.21f, 0.3f); // easier compare with Mitsuba's constant environment light
+#else
+            float3 unitDir = r.dir;
+            float t = 0.5f*(unitDir.y + 1.0f);
+            float3 skyCol = ((1.0f - t)*float3(1.0f, 1.0f, 1.0f) + t * float3(0.5f, 0.7f, 1.0f)) * 0.3f;
+            col += curAtten * skyCol;
+#endif
+            break;
+        }
+    }
+    return col;
+}
+
+Texture2D srcImage : register(t0);
+RWTexture2D<float4> dstImage : register(u0);
+StructuredBuffer<Sphere> g_Spheres : register(t1);
+StructuredBuffer<Material> g_Materials : register(t2);
+StructuredBuffer<Params> g_Params : register(t3);
+StructuredBuffer<int> g_Emissives : register(t4);
+RWByteAddressBuffer g_OutRayCount : register(u1);
+
+[numthreads(kCSGroupSizeX, kCSGroupSizeY, 1)]
+void main(uint3 gid : SV_DispatchThreadID, uint3 tid : SV_GroupThreadID)
+{
+    // First, move scene data (spheres, materials, emissive indices) into group shared
+    // memory. Do this in parallel; each thread in group copies its own chunk of data.
+    uint threadID = tid.y * kCSGroupSizeX + tid.x;
+    uint groupSize = kCSGroupSizeX * kCSGroupSizeY;
+    uint objCount = g_Params[0].sphereCount;
+    uint myObjCount = (objCount + groupSize - 1) / groupSize;
+    uint myObjStart = threadID * myObjCount;
+    for (uint io = myObjStart; io < myObjStart + myObjCount; ++io)
+    {
+        if (io < objCount)
+        {
+            s_GroupSpheres[io] = g_Spheres[io];
+            s_GroupMaterials[io] = g_Materials[io];
+        }
+        if (io < g_Params[0].emissiveCount)
+        {
+            s_GroupEmissives[io] = g_Emissives[io];
+        }
+    }
+    GroupMemoryBarrierWithGroupSync();
+
+    int rayCount = 0;
+    float3 col = 0;
+    Params params = g_Params[0];
+    uint rngState = (gid.x * 1973 + gid.y * 9277 + params.frames * 26699) | 1;
+    for (int s = 0; s < DO_SAMPLES_PER_PIXEL; s++)
+    {
+        float u = float(gid.x + RandomFloat01(rngState)) * params.invWidth;
+        float v = float(gid.y + RandomFloat01(rngState)) * params.invHeight;
+        Ray r = CameraGetRay(params.cam, u, v, rngState);
+        col += Trace(params.sphereCount, params.emissiveCount, r, rayCount, rngState);
+    }
+    col *= 1.0f / float(DO_SAMPLES_PER_PIXEL);
+
+    float3 prev = srcImage.Load(int3(gid.xy,0)).rgb;
+    col = lerp(col, prev, params.lerpFac);
+    dstImage[gid.xy] = float4(col, 1);
+
+    g_OutRayCount.InterlockedAdd(0, rayCount);
+}

third_party/tracy/examples/ToyPathTracer/Windows/PixelShader.hlsl

@@ -0,0 +1,15 @@
+float3 LinearToSRGB(float3 rgb)
+{
+    rgb = max(rgb, float3(0, 0, 0));
+    return max(1.055 * pow(rgb, 0.416666667) - 0.055, 0.0);
+}
+
+Texture2D tex : register(t0);
+SamplerState smp : register(s0);
+
+float4 main(float2 uv : TEXCOORD0) : SV_Target
+{
+    float3 col = tex.Sample(smp, uv).rgb;
+    col = LinearToSRGB(col);
+    return float4(col, 1.0f);
+}

third_party/tracy/examples/ToyPathTracer/Windows/TestCpu.sln

@@ -0,0 +1,31 @@
+
+Microsoft Visual Studio Solution File, Format Version 12.00
+# Visual Studio Version 16
+VisualStudioVersion = 16.0.30907.101
+MinimumVisualStudioVersion = 10.0.40219.1
+Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "TestCpu", "TestCpu.vcxproj", "{4F84B756-87F5-4B92-827B-DA087DAE1900}"
+EndProject
+Global
+	GlobalSection(SolutionConfigurationPlatforms) = preSolution
+		Debug|x64 = Debug|x64
+		Debug|x86 = Debug|x86
+		Release|x64 = Release|x64
+		Release|x86 = Release|x86
+	EndGlobalSection
+	GlobalSection(ProjectConfigurationPlatforms) = postSolution
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Debug|x64.ActiveCfg = Debug|x64
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Debug|x64.Build.0 = Debug|x64
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Debug|x86.ActiveCfg = Debug|Win32
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Debug|x86.Build.0 = Debug|Win32
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Release|x64.ActiveCfg = Release|x64
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Release|x64.Build.0 = Release|x64
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Release|x86.ActiveCfg = Release|Win32
+		{4F84B756-87F5-4B92-827B-DA087DAE1900}.Release|x86.Build.0 = Release|Win32
+	EndGlobalSection
+	GlobalSection(SolutionProperties) = preSolution
+		HideSolutionNode = FALSE
+	EndGlobalSection
+	GlobalSection(ExtensibilityGlobals) = postSolution
+		SolutionGuid = {067FB780-37B8-465E-AD7E-E7B238B9C04F}
+	EndGlobalSection
+EndGlobal

third_party/tracy/examples/ToyPathTracer/Windows/TestCpu.vcxproj

@@ -0,0 +1,245 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" ToolsVersion="15.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+  <ItemGroup Label="ProjectConfigurations">
+    <ProjectConfiguration Include="Debug|Win32">
+      <Configuration>Debug</Configuration>
+      <Platform>Win32</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Release|Win32">
+      <Configuration>Release</Configuration>
+      <Platform>Win32</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Debug|x64">
+      <Configuration>Debug</Configuration>
+      <Platform>x64</Platform>
+    </ProjectConfiguration>
+    <ProjectConfiguration Include="Release|x64">
+      <Configuration>Release</Configuration>
+      <Platform>x64</Platform>
+    </ProjectConfiguration>
+  </ItemGroup>
+  <PropertyGroup Label="Globals">
+    <VCProjectVersion>15.0</VCProjectVersion>
+    <ProjectGuid>{4F84B756-87F5-4B92-827B-DA087DAE1900}</ProjectGuid>
+    <Keyword>Win32Proj</Keyword>
+    <RootNamespace>TestCpu</RootNamespace>
+    <WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
+  </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>true</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+    <CharacterSet>Unicode</CharacterSet>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>false</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+    <WholeProgramOptimization>true</WholeProgramOptimization>
+    <CharacterSet>Unicode</CharacterSet>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>true</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+    <CharacterSet>Unicode</CharacterSet>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="Configuration">
+    <ConfigurationType>Application</ConfigurationType>
+    <UseDebugLibraries>false</UseDebugLibraries>
+    <PlatformToolset>v143</PlatformToolset>
+    <WholeProgramOptimization>true</WholeProgramOptimization>
+    <CharacterSet>Unicode</CharacterSet>
+  </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+  <ImportGroup Label="ExtensionSettings">
+  </ImportGroup>
+  <ImportGroup Label="Shared">
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <PropertyGroup Label="UserMacros" />
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <LinkIncremental>true</LinkIncremental>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <LinkIncremental>true</LinkIncremental>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <LinkIncremental>false</LinkIncremental>
+  </PropertyGroup>
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <LinkIncremental>false</LinkIncremental>
+  </PropertyGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>Disabled</Optimization>
+      <SDLCheck>true</SDLCheck>
+      <PreprocessorDefinitions>WIN32;_DEBUG;_WINDOWS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <ConformanceMode>true</ConformanceMode>
+      <CallingConvention>VectorCall</CallingConvention>
+      <FloatingPointModel>Fast</FloatingPointModel>
+      <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
+    </ClCompile>
+    <Link>
+      <SubSystem>Windows</SubSystem>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalDependencies>d3d11.lib;kernel32.lib;user32.lib;gdi32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>Disabled</Optimization>
+      <SDLCheck>true</SDLCheck>
+      <PreprocessorDefinitions>TRACY_ENABLE;_DEBUG;_WINDOWS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <ConformanceMode>true</ConformanceMode>
+      <CallingConvention>VectorCall</CallingConvention>
+      <FloatingPointModel>Fast</FloatingPointModel>
+      <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
+    </ClCompile>
+    <Link>
+      <SubSystem>Windows</SubSystem>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalDependencies>d3d11.lib;kernel32.lib;user32.lib;gdi32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>MaxSpeed</Optimization>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <PreprocessorDefinitions>WIN32;NDEBUG;_WINDOWS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <ConformanceMode>true</ConformanceMode>
+      <ExceptionHandling>false</ExceptionHandling>
+      <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+      <BufferSecurityCheck>false</BufferSecurityCheck>
+      <CallingConvention>VectorCall</CallingConvention>
+      <FloatingPointModel>Fast</FloatingPointModel>
+    </ClCompile>
+    <Link>
+      <SubSystem>Windows</SubSystem>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <OptimizeReferences>true</OptimizeReferences>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalDependencies>d3d11.lib;kernel32.lib;user32.lib;gdi32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+  </ItemDefinitionGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
+    <ClCompile>
+      <WarningLevel>Level3</WarningLevel>
+      <Optimization>MaxSpeed</Optimization>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <PreprocessorDefinitions>TRACY_ENABLE;NDEBUG;_WINDOWS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <ConformanceMode>true</ConformanceMode>
+      <ExceptionHandling>false</ExceptionHandling>
+      <RuntimeLibrary>MultiThreaded</RuntimeLibrary>
+      <BufferSecurityCheck>false</BufferSecurityCheck>
+      <CallingConvention>VectorCall</CallingConvention>
+      <FloatingPointModel>Fast</FloatingPointModel>
+    </ClCompile>
+    <Link>
+      <SubSystem>Windows</SubSystem>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <OptimizeReferences>true</OptimizeReferences>
+      <GenerateDebugInformation>true</GenerateDebugInformation>
+      <AdditionalDependencies>d3d11.lib;kernel32.lib;user32.lib;gdi32.lib;%(AdditionalDependencies)</AdditionalDependencies>
+    </Link>
+  </ItemDefinitionGroup>
+  <ItemGroup>
+    <ClCompile Include="..\..\..\public\TracyClient.cpp" />
+    <ClCompile Include="..\Source\enkiTS\TaskScheduler.cpp" />
+    <ClCompile Include="..\Source\enkiTS\TaskScheduler_c.cpp" />
+    <ClCompile Include="..\Source\Maths.cpp" />
+    <ClCompile Include="..\Source\Test.cpp" />
+    <ClCompile Include="TestWin.cpp" />
+  </ItemGroup>
+  <ItemGroup>
+    <ClInclude Include="..\Source\Config.h" />
+    <ClInclude Include="..\Source\enkiTS\Atomics.h" />
+    <ClInclude Include="..\Source\enkiTS\LockLessMultiReadPipe.h" />
+    <ClInclude Include="..\Source\enkiTS\TaskScheduler.h" />
+    <ClInclude Include="..\Source\enkiTS\TaskScheduler_c.h" />
+    <ClInclude Include="..\Source\enkiTS\Threads.h" />
+    <ClInclude Include="..\Source\Maths.h" />
+    <ClInclude Include="..\Source\MathSimd.h" />
+    <ClInclude Include="..\Source\Test.h" />
+    <ClInclude Include="..\Source\stb_image.h" />
+  </ItemGroup>
+  <ItemGroup>
+    <None Include="..\.editorconfig" />
+  </ItemGroup>
+  <ItemGroup>
+    <FxCompile Include="ComputeShader.hlsl">
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|x64'">Compute</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|x64'">5.0</ShaderModel>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Compute</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">5.0</ShaderModel>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">Compute</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">5.0</ShaderModel>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Compute</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">5.0</ShaderModel>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">g_CSBytecode</VariableName>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CompiledComputeShader.h</HeaderFileOutput>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">g_CSBytecode</VariableName>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CompiledComputeShader.h</HeaderFileOutput>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">g_CSBytecode</VariableName>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">CompiledComputeShader.h</HeaderFileOutput>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|x64'">g_CSBytecode</VariableName>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|x64'">CompiledComputeShader.h</HeaderFileOutput>
+    </FxCompile>
+    <FxCompile Include="PixelShader.hlsl">
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|x64'">Pixel</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Pixel</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">Pixel</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Pixel</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|x64'">5.0</ShaderModel>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CompiledPixelShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CompiledPixelShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">CompiledPixelShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|x64'">CompiledPixelShader.h</HeaderFileOutput>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">g_PSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">g_PSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">g_PSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|x64'">g_PSBytecode</VariableName>
+    </FxCompile>
+    <FxCompile Include="VertexShader.hlsl">
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|x64'">Vertex</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">Vertex</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">Vertex</ShaderType>
+      <ShaderType Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">Vertex</ShaderType>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">5.0</ShaderModel>
+      <ShaderModel Condition="'$(Configuration)|$(Platform)'=='Release|x64'">5.0</ShaderModel>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">CompiledVertexShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">CompiledVertexShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">CompiledVertexShader.h</HeaderFileOutput>
+      <HeaderFileOutput Condition="'$(Configuration)|$(Platform)'=='Release|x64'">CompiledVertexShader.h</HeaderFileOutput>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">g_VSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">g_VSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">g_VSBytecode</VariableName>
+      <VariableName Condition="'$(Configuration)|$(Platform)'=='Release|x64'">g_VSBytecode</VariableName>
+    </FxCompile>
+  </ItemGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+  <ImportGroup Label="ExtensionTargets">
+  </ImportGroup>
+</Project>

third_party/tracy/examples/ToyPathTracer/Windows/TestCpu.vcxproj.filters

@@ -0,0 +1,67 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+  <ItemGroup>
+    <ClCompile Include="TestWin.cpp" />
+    <ClCompile Include="..\Source\Test.cpp">
+      <Filter>Source</Filter>
+    </ClCompile>
+    <ClCompile Include="..\Source\enkiTS\TaskScheduler.cpp">
+      <Filter>Source\enkiTS</Filter>
+    </ClCompile>
+    <ClCompile Include="..\Source\enkiTS\TaskScheduler_c.cpp">
+      <Filter>Source\enkiTS</Filter>
+    </ClCompile>
+    <ClCompile Include="..\Source\Maths.cpp">
+      <Filter>Source</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\..\public\TracyClient.cpp" />
+  </ItemGroup>
+  <ItemGroup>
+    <Filter Include="Source">
+      <UniqueIdentifier>{5f19f217-c1c7-4eeb-be61-8b986fee9375}</UniqueIdentifier>
+    </Filter>
+    <Filter Include="Source\enkiTS">
+      <UniqueIdentifier>{38c448a8-1dcc-4116-9410-a9f8d068caff}</UniqueIdentifier>
+    </Filter>
+  </ItemGroup>
+  <ItemGroup>
+    <ClInclude Include="..\Source\Test.h">
+      <Filter>Source</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\stb_image.h">
+      <Filter>Source</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\enkiTS\Atomics.h">
+      <Filter>Source\enkiTS</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\enkiTS\LockLessMultiReadPipe.h">
+      <Filter>Source\enkiTS</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\enkiTS\TaskScheduler.h">
+      <Filter>Source\enkiTS</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\enkiTS\TaskScheduler_c.h">
+      <Filter>Source\enkiTS</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\enkiTS\Threads.h">
+      <Filter>Source\enkiTS</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\Maths.h">
+      <Filter>Source</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\Config.h">
+      <Filter>Source</Filter>
+    </ClInclude>
+    <ClInclude Include="..\Source\MathSimd.h">
+      <Filter>Source</Filter>
+    </ClInclude>
+  </ItemGroup>
+  <ItemGroup>
+    <None Include="..\.editorconfig" />
+  </ItemGroup>
+  <ItemGroup>
+    <FxCompile Include="VertexShader.hlsl" />
+    <FxCompile Include="PixelShader.hlsl" />
+    <FxCompile Include="ComputeShader.hlsl" />
+  </ItemGroup>
+</Project>

third_party/tracy/examples/ToyPathTracer/Windows/TestWin.cpp

@@ -0,0 +1,567 @@
+#include <stdint.h>
+#define WIN32_LEAN_AND_MEAN
+#define NOMINMAX
+#include <windows.h>
+#include <d3d11_1.h>
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <time.h>
+#include <algorithm>
+
+#include "../Source/Config.h"
+#include "../Source/Maths.h"
+#include "../Source/Test.h"
+#include "CompiledVertexShader.h"
+#include "CompiledPixelShader.h"
+
+#include "../../../public/tracy/Tracy.hpp"
+#include "../../../public/tracy/TracyD3D11.hpp"
+
+static HINSTANCE g_HInstance;
+static HWND g_Wnd;
+
+ATOM                MyRegisterClass(HINSTANCE hInstance);
+BOOL                InitInstance(HINSTANCE, int);
+LRESULT CALLBACK    WndProc(HWND, UINT, WPARAM, LPARAM);
+INT_PTR CALLBACK    About(HWND, UINT, WPARAM, LPARAM);
+
+static HRESULT InitD3DDevice();
+static void ShutdownD3DDevice();
+static void RenderFrame();
+
+static float* g_Backbuffer;
+
+static D3D_FEATURE_LEVEL g_D3D11FeatureLevel = D3D_FEATURE_LEVEL_11_0;
+static ID3D11Device* g_D3D11Device = nullptr;
+static ID3D11DeviceContext* g_D3D11Ctx = nullptr;
+static IDXGISwapChain* g_D3D11SwapChain = nullptr;
+static ID3D11RenderTargetView* g_D3D11RenderTarget = nullptr;
+static ID3D11VertexShader* g_VertexShader;
+static ID3D11PixelShader* g_PixelShader;
+static ID3D11Texture2D *g_BackbufferTexture, *g_BackbufferTexture2;
+static ID3D11ShaderResourceView *g_BackbufferSRV, *g_BackbufferSRV2;
+static ID3D11UnorderedAccessView *g_BackbufferUAV, *g_BackbufferUAV2;
+static ID3D11SamplerState* g_SamplerLinear;
+static ID3D11RasterizerState* g_RasterState;
+static int g_BackbufferIndex;
+static tracy::D3D11Ctx *g_tracyCtx;
+
+
+#if DO_COMPUTE_GPU
+#include "CompiledComputeShader.h"
+struct ComputeParams
+{
+    Camera cam;
+    int sphereCount;
+    int screenWidth;
+    int screenHeight;
+    int frames;
+    float invWidth;
+    float invHeight;
+    float lerpFac;
+    int emissiveCount;
+};
+static ID3D11ComputeShader* g_ComputeShader;
+static ID3D11Buffer* g_DataSpheres;     static ID3D11ShaderResourceView* g_SRVSpheres;
+static ID3D11Buffer* g_DataMaterials;   static ID3D11ShaderResourceView* g_SRVMaterials;
+static ID3D11Buffer* g_DataParams;      static ID3D11ShaderResourceView* g_SRVParams;
+static ID3D11Buffer* g_DataEmissives;   static ID3D11ShaderResourceView* g_SRVEmissives;
+static ID3D11Buffer* g_DataCounter;     static ID3D11UnorderedAccessView* g_UAVCounter;
+static int g_SphereCount, g_ObjSize, g_MatSize;
+static ID3D11Query *g_QueryBegin, *g_QueryEnd, *g_QueryDisjoint;
+#endif // #if DO_COMPUTE_GPU
+
+int APIENTRY wWinMain(_In_ HINSTANCE hInstance, _In_opt_ HINSTANCE, _In_ LPWSTR, _In_ int nCmdShow)
+{
+    g_Backbuffer = new float[kBackbufferWidth * kBackbufferHeight * 4];
+    memset(g_Backbuffer, 0, kBackbufferWidth * kBackbufferHeight * 4 * sizeof(g_Backbuffer[0]));
+
+    InitializeTest();
+
+    MyRegisterClass(hInstance);
+    if (!InitInstance (hInstance, nCmdShow))
+    {
+        return FALSE;
+    }
+
+    if (FAILED(InitD3DDevice()))
+    {
+        ShutdownD3DDevice();
+        return 0;
+    }
+
+    g_D3D11Device->CreateVertexShader(g_VSBytecode, ARRAYSIZE(g_VSBytecode), NULL, &g_VertexShader);
+    g_D3D11Device->CreatePixelShader(g_PSBytecode, ARRAYSIZE(g_PSBytecode), NULL, &g_PixelShader);
+#if DO_COMPUTE_GPU
+    g_D3D11Device->CreateComputeShader(g_CSBytecode, ARRAYSIZE(g_CSBytecode), NULL, &g_ComputeShader);
+#endif
+
+    D3D11_TEXTURE2D_DESC texDesc = {};
+    texDesc.Width = kBackbufferWidth;
+    texDesc.Height = kBackbufferHeight;
+    texDesc.MipLevels = 1;
+    texDesc.ArraySize = 1;
+    texDesc.Format = DXGI_FORMAT_R32G32B32A32_FLOAT;
+    texDesc.SampleDesc.Count = 1;
+    texDesc.SampleDesc.Quality = 0;
+#if DO_COMPUTE_GPU
+    texDesc.Usage = D3D11_USAGE_DEFAULT;
+    texDesc.BindFlags = D3D11_BIND_SHADER_RESOURCE | D3D11_BIND_UNORDERED_ACCESS;
+    texDesc.CPUAccessFlags = 0;
+#else
+    texDesc.Usage = D3D11_USAGE_DYNAMIC;
+    texDesc.BindFlags = D3D11_BIND_SHADER_RESOURCE;
+    texDesc.CPUAccessFlags = D3D11_CPU_ACCESS_WRITE;
+#endif
+    texDesc.MiscFlags = 0;
+    g_D3D11Device->CreateTexture2D(&texDesc, NULL, &g_BackbufferTexture);
+    g_D3D11Device->CreateTexture2D(&texDesc, NULL, &g_BackbufferTexture2);
+
+    D3D11_SHADER_RESOURCE_VIEW_DESC srvDesc = {};
+    srvDesc.Format = texDesc.Format;
+    srvDesc.ViewDimension = D3D11_SRV_DIMENSION_TEXTURE2D;
+    srvDesc.Texture2D.MipLevels = 1;
+    srvDesc.Texture2D.MostDetailedMip = 0;
+    g_D3D11Device->CreateShaderResourceView(g_BackbufferTexture, &srvDesc, &g_BackbufferSRV);
+    g_D3D11Device->CreateShaderResourceView(g_BackbufferTexture2, &srvDesc, &g_BackbufferSRV2);
+
+    D3D11_SAMPLER_DESC smpDesc = {};
+    smpDesc.Filter = D3D11_FILTER_MIN_MAG_LINEAR_MIP_POINT;
+    smpDesc.AddressU = smpDesc.AddressV = smpDesc.AddressW = D3D11_TEXTURE_ADDRESS_CLAMP;
+    g_D3D11Device->CreateSamplerState(&smpDesc, &g_SamplerLinear);
+
+    D3D11_RASTERIZER_DESC rasterDesc = {};
+    rasterDesc.FillMode = D3D11_FILL_SOLID;
+    rasterDesc.CullMode = D3D11_CULL_NONE;
+    g_D3D11Device->CreateRasterizerState(&rasterDesc, &g_RasterState);
+
+#if DO_COMPUTE_GPU
+    D3D11_UNORDERED_ACCESS_VIEW_DESC uavDesc = {};
+
+    int camSize;
+    GetObjectCount(g_SphereCount, g_ObjSize, g_MatSize, camSize);
+    assert(g_ObjSize == 20);
+    assert(g_MatSize == 36);
+    assert(camSize == 88);
+    D3D11_BUFFER_DESC bdesc = {};
+    bdesc.ByteWidth = g_SphereCount * g_ObjSize;
+    bdesc.Usage = D3D11_USAGE_DEFAULT;
+    bdesc.BindFlags = D3D11_BIND_SHADER_RESOURCE;
+    bdesc.CPUAccessFlags = 0;
+    bdesc.MiscFlags = D3D11_RESOURCE_MISC_BUFFER_STRUCTURED;
+    bdesc.StructureByteStride = g_ObjSize;
+    g_D3D11Device->CreateBuffer(&bdesc, NULL, &g_DataSpheres);
+    srvDesc.Format = DXGI_FORMAT_UNKNOWN;
+    srvDesc.ViewDimension = D3D11_SRV_DIMENSION_BUFFER;
+    srvDesc.Buffer.FirstElement = 0;
+    srvDesc.Buffer.NumElements = g_SphereCount;
+    g_D3D11Device->CreateShaderResourceView(g_DataSpheres, &srvDesc, &g_SRVSpheres);
+
+    bdesc.ByteWidth = g_SphereCount * g_MatSize;
+    bdesc.StructureByteStride = g_MatSize;
+    g_D3D11Device->CreateBuffer(&bdesc, NULL, &g_DataMaterials);
+    srvDesc.Buffer.NumElements = g_SphereCount;
+    g_D3D11Device->CreateShaderResourceView(g_DataMaterials, &srvDesc, &g_SRVMaterials);
+
+    bdesc.ByteWidth = sizeof(ComputeParams);
+    bdesc.StructureByteStride = sizeof(ComputeParams);
+    g_D3D11Device->CreateBuffer(&bdesc, NULL, &g_DataParams);
+    srvDesc.Buffer.NumElements = 1;
+    g_D3D11Device->CreateShaderResourceView(g_DataParams, &srvDesc, &g_SRVParams);
+
+    bdesc.ByteWidth = g_SphereCount * 4;
+    bdesc.StructureByteStride = 4;
+    g_D3D11Device->CreateBuffer(&bdesc, NULL, &g_DataEmissives);
+    srvDesc.Buffer.NumElements = g_SphereCount;
+    g_D3D11Device->CreateShaderResourceView(g_DataEmissives, &srvDesc, &g_SRVEmissives);
+
+    bdesc.ByteWidth = 4;
+    bdesc.BindFlags |= D3D11_BIND_UNORDERED_ACCESS;
+    bdesc.MiscFlags = D3D11_RESOURCE_MISC_BUFFER_ALLOW_RAW_VIEWS;
+    bdesc.CPUAccessFlags = D3D11_CPU_ACCESS_READ;
+    g_D3D11Device->CreateBuffer(&bdesc, NULL, &g_DataCounter);
+    uavDesc.Format = DXGI_FORMAT_R32_TYPELESS;
+    uavDesc.ViewDimension = D3D11_UAV_DIMENSION_BUFFER;
+    uavDesc.Buffer.FirstElement = 0;
+    uavDesc.Buffer.NumElements = 1;
+    uavDesc.Buffer.Flags = D3D11_BUFFER_UAV_FLAG_RAW;
+    g_D3D11Device->CreateUnorderedAccessView(g_DataCounter, &uavDesc, &g_UAVCounter);
+
+    uavDesc.Format = DXGI_FORMAT_R32G32B32A32_FLOAT;
+    uavDesc.ViewDimension = D3D11_UAV_DIMENSION_TEXTURE2D;
+    uavDesc.Texture2D.MipSlice = 0;
+    g_D3D11Device->CreateUnorderedAccessView(g_BackbufferTexture, &uavDesc, &g_BackbufferUAV);
+    g_D3D11Device->CreateUnorderedAccessView(g_BackbufferTexture2, &uavDesc, &g_BackbufferUAV2);
+
+    D3D11_QUERY_DESC qDesc = {};
+    qDesc.Query = D3D11_QUERY_TIMESTAMP;
+    g_D3D11Device->CreateQuery(&qDesc, &g_QueryBegin);
+    g_D3D11Device->CreateQuery(&qDesc, &g_QueryEnd);
+    qDesc.Query = D3D11_QUERY_TIMESTAMP_DISJOINT;
+    g_D3D11Device->CreateQuery(&qDesc, &g_QueryDisjoint);
+#endif // #if DO_COMPUTE_GPU
+
+
+    static int framesLeft = 10;
+
+    // Main message loop
+    MSG msg = { 0 };
+    while (msg.message != WM_QUIT)
+    {
+        if (PeekMessage(&msg, NULL, 0U, 0U, PM_REMOVE))
+        {
+            TranslateMessage(&msg);
+            DispatchMessage(&msg);
+        }
+        else
+        {
+            RenderFrame();
+            TracyD3D11Collect(g_tracyCtx);
+            if( --framesLeft == 0 ) break;
+        }
+    }
+
+    ShutdownTest();
+    ShutdownD3DDevice();
+
+    return (int) msg.wParam;
+}
+
+
+ATOM MyRegisterClass(HINSTANCE hInstance)
+{
+    ZoneScoped;
+
+    WNDCLASSEXW wcex;
+    memset(&wcex, 0, sizeof(wcex));
+    wcex.cbSize = sizeof(WNDCLASSEX);
+    wcex.style          = CS_HREDRAW | CS_VREDRAW;
+    wcex.lpfnWndProc    = WndProc;
+    wcex.cbClsExtra     = 0;
+    wcex.cbWndExtra     = 0;
+    wcex.hInstance      = hInstance;
+    wcex.hCursor        = LoadCursor(nullptr, IDC_ARROW);
+    wcex.hbrBackground  = (HBRUSH)(COLOR_WINDOW+1);
+    wcex.lpszClassName  = L"TestClass";
+    return RegisterClassExW(&wcex);
+}
+
+BOOL InitInstance(HINSTANCE hInstance, int nCmdShow)
+{
+    ZoneScoped;
+
+    g_HInstance = hInstance;
+    RECT rc = { 0, 0, kBackbufferWidth, kBackbufferHeight };
+    DWORD style = WS_OVERLAPPED | WS_CAPTION | WS_SYSMENU | WS_MINIMIZEBOX;
+    AdjustWindowRect(&rc, style, FALSE);
+    HWND hWnd = CreateWindowW(L"TestClass", L"Test", style, CW_USEDEFAULT, CW_USEDEFAULT, rc.right-rc.left, rc.bottom-rc.top, nullptr, nullptr, hInstance, nullptr);
+    if (!hWnd)
+        return FALSE;
+    g_Wnd = hWnd;
+    ShowWindow(hWnd, nCmdShow);
+    UpdateWindow(hWnd);
+    return TRUE;
+}
+
+static uint64_t s_Time;
+static int s_Count;
+static char s_Buffer[200];
+static unsigned s_Flags = kFlagProgressive;
+static int s_FrameCount = 0;
+
+
+static void RenderFrame()
+{
+    ZoneScoped;
+    TracyD3D11Zone(g_tracyCtx, "RenderFrame");
+
+    LARGE_INTEGER time1;
+
+#if DO_COMPUTE_GPU
+    QueryPerformanceCounter(&time1);
+    float t = float(clock()) / CLOCKS_PER_SEC;
+    UpdateTest(t, s_FrameCount, kBackbufferWidth, kBackbufferHeight, s_Flags);
+
+    g_BackbufferIndex = 1 - g_BackbufferIndex;
+    void* dataSpheres = alloca(g_SphereCount * g_ObjSize);
+    void* dataMaterials = alloca(g_SphereCount * g_MatSize);
+    void* dataEmissives = alloca(g_SphereCount * 4);
+    ComputeParams dataParams;
+    GetSceneDesc(dataSpheres, dataMaterials, &dataParams.cam, dataEmissives, &dataParams.emissiveCount);
+
+    dataParams.sphereCount = g_SphereCount;
+    dataParams.screenWidth = kBackbufferWidth;
+    dataParams.screenHeight = kBackbufferHeight;
+    dataParams.frames = s_FrameCount;
+    dataParams.invWidth = 1.0f / kBackbufferWidth;
+    dataParams.invHeight = 1.0f / kBackbufferHeight;
+    float lerpFac = float(s_FrameCount) / float(s_FrameCount + 1);
+    if (s_Flags & kFlagAnimate)
+        lerpFac *= DO_ANIMATE_SMOOTHING;
+    if (!(s_Flags & kFlagProgressive))
+        lerpFac = 0;
+    dataParams.lerpFac = lerpFac;
+
+    g_D3D11Ctx->UpdateSubresource(g_DataSpheres, 0, NULL, dataSpheres, 0, 0);
+    g_D3D11Ctx->UpdateSubresource(g_DataMaterials, 0, NULL, dataMaterials, 0, 0);
+    g_D3D11Ctx->UpdateSubresource(g_DataParams, 0, NULL, &dataParams, 0, 0);
+    g_D3D11Ctx->UpdateSubresource(g_DataEmissives, 0, NULL, dataEmissives, 0, 0);
+
+    ID3D11ShaderResourceView* srvs[] = {
+        g_BackbufferIndex == 0 ? g_BackbufferSRV2 : g_BackbufferSRV,
+        g_SRVSpheres,
+        g_SRVMaterials,
+        g_SRVParams,
+        g_SRVEmissives
+    };
+    g_D3D11Ctx->CSSetShaderResources(0, ARRAYSIZE(srvs), srvs);
+    ID3D11UnorderedAccessView* uavs[] = {
+        g_BackbufferIndex == 0 ? g_BackbufferUAV : g_BackbufferUAV2,
+        g_UAVCounter
+    };
+    g_D3D11Ctx->CSSetUnorderedAccessViews(0, ARRAYSIZE(uavs), uavs, NULL);
+    g_D3D11Ctx->CSSetShader(g_ComputeShader, NULL, 0);
+    g_D3D11Ctx->Begin(g_QueryDisjoint);
+    g_D3D11Ctx->End(g_QueryBegin);
+    g_D3D11Ctx->Dispatch(kBackbufferWidth/kCSGroupSizeX, kBackbufferHeight/kCSGroupSizeY, 1);
+    g_D3D11Ctx->End(g_QueryEnd);
+    uavs[0] = NULL;
+    g_D3D11Ctx->CSSetUnorderedAccessViews(0, ARRAYSIZE(uavs), uavs, NULL);
+    ++s_FrameCount;
+
+#else
+    QueryPerformanceCounter(&time1);
+    float t = float(clock()) / CLOCKS_PER_SEC;
+    static size_t s_RayCounter = 0;
+    int rayCount;
+    UpdateTest(t, s_FrameCount, kBackbufferWidth, kBackbufferHeight, s_Flags);
+    DrawTest(t, s_FrameCount, kBackbufferWidth, kBackbufferHeight, g_Backbuffer, rayCount, s_Flags);
+    s_FrameCount++;
+    s_RayCounter += rayCount;
+    LARGE_INTEGER time2;
+    QueryPerformanceCounter(&time2);
+    uint64_t dt = time2.QuadPart - time1.QuadPart;
+    ++s_Count;
+    s_Time += dt;
+    if (s_Count > 10)
+    {
+        LARGE_INTEGER frequency;
+        QueryPerformanceFrequency(&frequency);
+
+        double s = double(s_Time) / double(frequency.QuadPart) / s_Count;
+        sprintf_s(s_Buffer, sizeof(s_Buffer), "%.2fms (%.1f FPS) %.1fMrays/s %.2fMrays/frame frames %i\n", s * 1000.0f, 1.f / s, s_RayCounter / s_Count / s * 1.0e-6f, s_RayCounter / s_Count * 1.0e-6f, s_FrameCount);
+        SetWindowTextA(g_Wnd, s_Buffer);
+        OutputDebugStringA(s_Buffer);
+        s_Count = 0;
+        s_Time = 0;
+        s_RayCounter = 0;
+    }
+
+    D3D11_MAPPED_SUBRESOURCE mapped;
+    g_D3D11Ctx->Map(g_BackbufferTexture, 0, D3D11_MAP_WRITE_DISCARD, 0, &mapped);
+    const uint8_t* src = (const uint8_t*)g_Backbuffer;
+    uint8_t* dst = (uint8_t*)mapped.pData;
+    for (int y = 0; y < kBackbufferHeight; ++y)
+    {
+        memcpy(dst, src, kBackbufferWidth * 16);
+        src += kBackbufferWidth * 16;
+        dst += mapped.RowPitch;
+    }
+    g_D3D11Ctx->Unmap(g_BackbufferTexture, 0);
+#endif
+
+    g_D3D11Ctx->VSSetShader(g_VertexShader, NULL, 0);
+    g_D3D11Ctx->PSSetShader(g_PixelShader, NULL, 0);
+    g_D3D11Ctx->PSSetShaderResources(0, 1, g_BackbufferIndex == 0 ? &g_BackbufferSRV : &g_BackbufferSRV2);
+    g_D3D11Ctx->PSSetSamplers(0, 1, &g_SamplerLinear);
+    g_D3D11Ctx->IASetPrimitiveTopology(D3D11_PRIMITIVE_TOPOLOGY_TRIANGLELIST);
+    g_D3D11Ctx->RSSetState(g_RasterState);
+    g_D3D11Ctx->Draw(3, 0);
+    g_D3D11SwapChain->Present(0, 0);
+
+    FrameMark;
+
+#if DO_COMPUTE_GPU
+    g_D3D11Ctx->End(g_QueryDisjoint);
+
+    // get GPU times
+    while (g_D3D11Ctx->GetData(g_QueryDisjoint, NULL, 0, 0) == S_FALSE) { Sleep(0); }
+    D3D10_QUERY_DATA_TIMESTAMP_DISJOINT tsDisjoint;
+    g_D3D11Ctx->GetData(g_QueryDisjoint, &tsDisjoint, sizeof(tsDisjoint), 0);
+    if (!tsDisjoint.Disjoint)
+    {
+        UINT64 tsBegin, tsEnd;
+        // Note: on some GPUs/drivers, even when the disjoint query above already said "yeah I have data",
+        // might still not return "I have data" for timestamp queries before it.
+        while (g_D3D11Ctx->GetData(g_QueryBegin, &tsBegin, sizeof(tsBegin), 0) == S_FALSE) { Sleep(0); }
+        while (g_D3D11Ctx->GetData(g_QueryEnd, &tsEnd, sizeof(tsEnd), 0) == S_FALSE) { Sleep(0); }
+
+        float s = float(tsEnd - tsBegin) / float(tsDisjoint.Frequency);
+
+        static uint64_t s_RayCounter;
+        D3D11_MAPPED_SUBRESOURCE mapped;
+        g_D3D11Ctx->Map(g_DataCounter, 0, D3D11_MAP_READ, 0, &mapped);
+        s_RayCounter += *(const int*)mapped.pData;
+        g_D3D11Ctx->Unmap(g_DataCounter, 0);
+        int zeroCount = 0;
+        g_D3D11Ctx->UpdateSubresource(g_DataCounter, 0, NULL, &zeroCount, 0, 0);
+
+        static float s_Time;
+        ++s_Count;
+        s_Time += s;
+        if (s_Count > 150)
+        {
+            s = s_Time / s_Count;
+            sprintf_s(s_Buffer, sizeof(s_Buffer), "%.2fms (%.1f FPS) %.1fMrays/s %.2fMrays/frame frames %i\n", s * 1000.0f, 1.f / s, s_RayCounter / s_Count / s * 1.0e-6f, s_RayCounter / s_Count * 1.0e-6f, s_FrameCount);
+            SetWindowTextA(g_Wnd, s_Buffer);
+            s_Count = 0;
+            s_Time = 0;
+            s_RayCounter = 0;
+        }
+
+    }
+#endif // #if DO_COMPUTE_GPU
+}
+
+
+LRESULT CALLBACK WndProc(HWND hWnd, UINT message, WPARAM wParam, LPARAM lParam)
+{
+    switch (message)
+    {
+    case WM_PAINT:
+        {
+            PAINTSTRUCT ps;
+            HDC hdc = BeginPaint(hWnd, &ps);
+            EndPaint(hWnd, &ps);
+        }
+        break;
+    case WM_DESTROY:
+        PostQuitMessage(0);
+        break;
+    case WM_CHAR:
+        if (wParam == 'a')
+            s_Flags = s_Flags ^ kFlagAnimate;
+        if (wParam == 'p')
+        {
+            s_Flags = s_Flags ^ kFlagProgressive;
+            s_FrameCount = 0;
+        }
+        break;
+    default:
+        return DefWindowProc(hWnd, message, wParam, lParam);
+    }
+    return 0;
+}
+
+
+static HRESULT InitD3DDevice()
+{
+    ZoneScoped;
+
+    HRESULT hr = S_OK;
+
+    RECT rc;
+    GetClientRect(g_Wnd, &rc);
+    UINT width = rc.right - rc.left;
+    UINT height = rc.bottom - rc.top;
+
+    UINT createDeviceFlags = 0;
+#ifdef _DEBUG
+    createDeviceFlags |= D3D11_CREATE_DEVICE_DEBUG;
+#endif
+
+    D3D_FEATURE_LEVEL featureLevels[] =
+    {
+        D3D_FEATURE_LEVEL_11_0,
+    };
+    UINT numFeatureLevels = ARRAYSIZE(featureLevels);
+    hr = D3D11CreateDevice(nullptr, D3D_DRIVER_TYPE_HARDWARE, nullptr, createDeviceFlags, featureLevels, numFeatureLevels, D3D11_SDK_VERSION, &g_D3D11Device, &g_D3D11FeatureLevel, &g_D3D11Ctx);
+    if (FAILED(hr))
+        return hr;
+
+    // Get DXGI factory
+    IDXGIFactory1* dxgiFactory = nullptr;
+    {
+        IDXGIDevice* dxgiDevice = nullptr;
+        hr = g_D3D11Device->QueryInterface(__uuidof(IDXGIDevice), reinterpret_cast<void**>(&dxgiDevice));
+        if (SUCCEEDED(hr))
+        {
+            IDXGIAdapter* adapter = nullptr;
+            hr = dxgiDevice->GetAdapter(&adapter);
+            if (SUCCEEDED(hr))
+            {
+                hr = adapter->GetParent(__uuidof(IDXGIFactory1), reinterpret_cast<void**>(&dxgiFactory));
+                adapter->Release();
+            }
+            dxgiDevice->Release();
+        }
+    }
+    if (FAILED(hr))
+        return hr;
+
+    // Create swap chain
+    DXGI_SWAP_CHAIN_DESC sd;
+    ZeroMemory(&sd, sizeof(sd));
+    sd.BufferCount = 1;
+    sd.BufferDesc.Width = width;
+    sd.BufferDesc.Height = height;
+    sd.BufferDesc.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
+    sd.BufferDesc.RefreshRate.Numerator = 60;
+    sd.BufferDesc.RefreshRate.Denominator = 1;
+    sd.BufferUsage = DXGI_USAGE_RENDER_TARGET_OUTPUT;
+    sd.OutputWindow = g_Wnd;
+    sd.SampleDesc.Count = 1;
+    sd.SampleDesc.Quality = 0;
+    sd.Windowed = TRUE;
+    hr = dxgiFactory->CreateSwapChain(g_D3D11Device, &sd, &g_D3D11SwapChain);
+
+    // Prevent Alt-Enter
+    dxgiFactory->MakeWindowAssociation(g_Wnd, DXGI_MWA_NO_ALT_ENTER);
+    dxgiFactory->Release();
+
+    if (FAILED(hr))
+        return hr;
+
+    // RTV
+    ID3D11Texture2D* pBackBuffer = nullptr;
+    hr = g_D3D11SwapChain->GetBuffer(0, __uuidof(ID3D11Texture2D), reinterpret_cast<void**>(&pBackBuffer));
+    if (FAILED(hr))
+        return hr;
+    hr = g_D3D11Device->CreateRenderTargetView(pBackBuffer, nullptr, &g_D3D11RenderTarget);
+    pBackBuffer->Release();
+    if (FAILED(hr))
+        return hr;
+
+    g_D3D11Ctx->OMSetRenderTargets(1, &g_D3D11RenderTarget, nullptr);
+
+    // Viewport
+    D3D11_VIEWPORT vp;
+    vp.Width = (float)width;
+    vp.Height = (float)height;
+    vp.MinDepth = 0.0f;
+    vp.MaxDepth = 1.0f;
+    vp.TopLeftX = 0;
+    vp.TopLeftY = 0;
+    g_D3D11Ctx->RSSetViewports(1, &vp);
+
+    g_tracyCtx = TracyD3D11Context(g_D3D11Device, g_D3D11Ctx);
+    const char* tracyD3D11CtxName = "D3D11";
+    TracyD3D11ContextName(g_tracyCtx, tracyD3D11CtxName, (uint16_t)strlen(tracyD3D11CtxName));
+
+    return S_OK;
+}
+
+static void ShutdownD3DDevice()
+{
+    ZoneScoped;
+
+    if (g_tracyCtx) TracyD3D11Destroy(g_tracyCtx);
+
+    if (g_D3D11Ctx) g_D3D11Ctx->ClearState();
+
+    if (g_D3D11RenderTarget) g_D3D11RenderTarget->Release();
+    if (g_D3D11SwapChain) g_D3D11SwapChain->Release();
+    if (g_D3D11Ctx) g_D3D11Ctx->Release();
+    if (g_D3D11Device) g_D3D11Device->Release();
+}

third_party/tracy/examples/ToyPathTracer/Windows/VertexShader.hlsl

@@ -0,0 +1,13 @@
+struct vs2ps
+{
+    float2 uv : TEXCOORD0;
+    float4 pos : SV_Position;
+};
+
+vs2ps main(uint vid : SV_VertexID)
+{
+    vs2ps o;
+    o.uv = float2((vid << 1) & 2, vid & 2);
+    o.pos = float4(o.uv * float2(2, 2) + float2(-1, -1), 0, 1);
+    return o;
+}

third_party/tracy/examples/ToyPathTracer/license.md

@@ -0,0 +1,24 @@
+This is free and unencumbered software released into the public domain.
+
+Anyone is free to copy, modify, publish, use, compile, sell, or
+distribute this software, either in source code form or as a compiled
+binary, for any purpose, commercial or non-commercial, and by any
+means.
+
+In jurisdictions that recognize copyright laws, the author or authors
+of this software dedicate any and all copyright interest in the
+software to the public domain. We make this dedication for the benefit
+of the public at large and to the detriment of our heirs and
+successors. We intend this dedication to be an overt act of
+relinquishment in perpetuity of all present and future rights to this
+software under copyright law.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+OTHER DEALINGS IN THE SOFTWARE.
+
+For more information, please refer to <http://unlicense.org>

third_party/tracy/examples/fibers.cpp

@@ -0,0 +1,30 @@
+// g++ fibers.cpp ../public/TracyClient.cpp -I../public/tracy -DTRACY_ENABLE -DTRACY_FIBERS -lpthread -ldl
+
+#include <thread>
+#include <unistd.h>
+
+#include "Tracy.hpp"
+#include "TracyC.h"
+
+const char* fiber = "job1";
+TracyCZoneCtx zone;
+
+int main()
+{
+    std::thread t1( [] {
+        TracyFiberEnter( fiber );
+        TracyCZone( ctx, 1 );
+        zone = ctx;
+        sleep( 1 );
+        TracyFiberLeave;
+    });
+    t1.join();
+
+    std::thread t2( [] {
+        TracyFiberEnter( fiber );
+        sleep( 1 );
+        TracyCZoneEnd( zone );
+        TracyFiberLeave;
+    });
+    t2.join();
+}

third_party/tracy/public/TracyClient.cpp

@@ -0,0 +1,61 @@
+//
+//          Tracy profiler
+//         ----------------
+//
+// For fast integration, compile and
+// link with this source file (and none
+// other) in your executable (or in the
+// main DLL / shared object on multi-DLL
+// projects).
+//
+
+// Define TRACY_ENABLE to enable profiler.
+
+#include "common/TracySystem.cpp"
+
+#ifdef TRACY_ENABLE
+
+#ifdef _MSC_VER
+#  pragma warning(push, 0)
+#endif
+
+#include "common/tracy_lz4.cpp"
+#include "client/TracyProfiler.cpp"
+#include "client/TracyCallstack.cpp"
+#include "client/TracySysPower.cpp"
+#include "client/TracySysTime.cpp"
+#include "client/TracySysTrace.cpp"
+#include "common/TracySocket.cpp"
+#include "client/tracy_rpmalloc.cpp"
+#include "client/TracyDxt1.cpp"
+#include "client/TracyAlloc.cpp"
+#include "client/TracyOverride.cpp"
+#include "client/TracyKCore.cpp"
+
+#if defined(TRACY_HAS_CALLSTACK)
+#  if TRACY_HAS_CALLSTACK == 2 || TRACY_HAS_CALLSTACK == 3 || TRACY_HAS_CALLSTACK == 4 || TRACY_HAS_CALLSTACK == 6
+#    include "libbacktrace/alloc.cpp"
+#    include "libbacktrace/dwarf.cpp"
+#    include "libbacktrace/fileline.cpp"
+#    include "libbacktrace/mmapio.cpp"
+#    include "libbacktrace/posix.cpp"
+#    include "libbacktrace/sort.cpp"
+#    include "libbacktrace/state.cpp"
+#    if TRACY_HAS_CALLSTACK == 4
+#      include "libbacktrace/macho.cpp"
+#    else
+#      include "libbacktrace/elf.cpp"
+#    endif
+#    include "common/TracyStackFrames.cpp"
+#  endif
+#endif
+
+#ifdef _MSC_VER
+#  pragma comment(lib, "ws2_32.lib")
+#  pragma comment(lib, "dbghelp.lib")
+#  pragma comment(lib, "advapi32.lib")
+#  pragma comment(lib, "user32.lib")
+#  pragma warning(pop)
+#endif
+
+#endif

third_party/tracy/public/libbacktrace/alloc.cpp

@@ -0,0 +1,174 @@
+/* alloc.c -- Memory allocation without mmap.
+   Copyright (C) 2012-2021 Free Software Foundation, Inc.
+   Written by Ian Lance Taylor, Google.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+    (1) Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+
+    (2) 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.
+
+    (3) The name of the author may not be used to
+    endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.  */
+
+#include "config.h"
+
+#include <errno.h>
+#include <stdlib.h>
+#include <sys/types.h>
+
+#include "backtrace.hpp"
+#include "internal.hpp"
+
+#include "../common/TracyAlloc.hpp"
+
+namespace tracy
+{
+
+/* Allocation routines to use on systems that do not support anonymous
+   mmap.  This implementation just uses malloc, which means that the
+   backtrace functions may not be safely invoked from a signal
+   handler.  */
+
+/* Allocate memory like malloc.  If ERROR_CALLBACK is NULL, don't
+   report an error.  */
+
+void *
+backtrace_alloc (struct backtrace_state *state ATTRIBUTE_UNUSED,
+		 size_t size, backtrace_error_callback error_callback,
+		 void *data)
+{
+  void *ret;
+
+  ret = tracy_malloc (size);
+  if (ret == NULL)
+    {
+      if (error_callback)
+	error_callback (data, "malloc", errno);
+    }
+  return ret;
+}
+
+/* Free memory.  */
+
+void
+backtrace_free (struct backtrace_state *state ATTRIBUTE_UNUSED,
+		void *p, size_t size ATTRIBUTE_UNUSED,
+		backtrace_error_callback error_callback ATTRIBUTE_UNUSED,
+		void *data ATTRIBUTE_UNUSED)
+{
+  tracy_free (p);
+}
+
+/* Grow VEC by SIZE bytes.  */
+
+void *
+backtrace_vector_grow (struct backtrace_state *state ATTRIBUTE_UNUSED,
+		       size_t size, backtrace_error_callback error_callback,
+		       void *data, struct backtrace_vector *vec)
+{
+  void *ret;
+
+  if (size > vec->alc)
+    {
+      size_t alc;
+      void *base;
+
+      if (vec->size == 0)
+	alc = 32 * size;
+      else if (vec->size >= 4096)
+	alc = vec->size + 4096;
+      else
+	alc = 2 * vec->size;
+
+      if (alc < vec->size + size)
+	alc = vec->size + size;
+
+      base = tracy_realloc (vec->base, alc);
+      if (base == NULL)
+	{
+	  error_callback (data, "realloc", errno);
+	  return NULL;
+	}
+
+      vec->base = base;
+      vec->alc = alc - vec->size;
+    }
+
+  ret = (char *) vec->base + vec->size;
+  vec->size += size;
+  vec->alc -= size;
+  return ret;
+}
+
+/* Finish the current allocation on VEC.  */
+
+void *
+backtrace_vector_finish (struct backtrace_state *state,
+			 struct backtrace_vector *vec,
+			 backtrace_error_callback error_callback,
+			 void *data)
+{
+  void *ret;
+
+  /* With this allocator we call realloc in backtrace_vector_grow,
+     which means we can't easily reuse the memory here.  So just
+     release it.  */
+  if (!backtrace_vector_release (state, vec, error_callback, data))
+    return NULL;
+  ret = vec->base;
+  vec->base = NULL;
+  vec->size = 0;
+  vec->alc = 0;
+  return ret;
+}
+
+/* Release any extra space allocated for VEC.  */
+
+int
+backtrace_vector_release (struct backtrace_state *state ATTRIBUTE_UNUSED,
+			  struct backtrace_vector *vec,
+			  backtrace_error_callback error_callback,
+			  void *data)
+{
+  vec->alc = 0;
+
+  if (vec->size == 0)
+    {
+      /* As of C17, realloc with size 0 is marked as an obsolescent feature, use
+	 free instead.  */
+      tracy_free (vec->base);
+      vec->base = NULL;
+      return 1;
+    }
+
+  vec->base = tracy_realloc (vec->base, vec->size);
+  if (vec->base == NULL)
+    {
+      error_callback (data, "realloc", errno);
+      return 0;
+    }
+
+  return 1;
+}
+
+}

third_party/tracy/public/libbacktrace/mmapio.cpp

@@ -0,0 +1,115 @@
+/* mmapio.c -- File views using mmap.
+   Copyright (C) 2012-2021 Free Software Foundation, Inc.
+   Written by Ian Lance Taylor, Google.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+    (1) Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+
+    (2) 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.
+
+    (3) The name of the author may not be used to
+    endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.  */
+
+#include "config.h"
+
+#include <errno.h>
+#include <sys/types.h>
+#include <sys/mman.h>
+#include <unistd.h>
+
+#include "backtrace.hpp"
+#include "internal.hpp"
+
+#ifndef HAVE_DECL_GETPAGESIZE
+extern int getpagesize (void);
+#endif
+
+#ifndef MAP_FAILED
+#define MAP_FAILED ((void *)-1)
+#endif
+
+namespace tracy
+{
+
+/* This file implements file views and memory allocation when mmap is
+   available.  */
+
+/* Create a view of SIZE bytes from DESCRIPTOR at OFFSET.  */
+
+int
+backtrace_get_view (struct backtrace_state *state ATTRIBUTE_UNUSED,
+		    int descriptor, off_t offset, uint64_t size,
+		    backtrace_error_callback error_callback,
+		    void *data, struct backtrace_view *view)
+{
+  size_t pagesize;
+  unsigned int inpage;
+  off_t pageoff;
+  void *map;
+
+  if ((uint64_t) (size_t) size != size)
+    {
+      error_callback (data, "file size too large", 0);
+      return 0;
+    }
+
+  pagesize = getpagesize ();
+  inpage = offset % pagesize;
+  pageoff = offset - inpage;
+
+  size += inpage;
+  size = (size + (pagesize - 1)) & ~ (pagesize - 1);
+
+  map = mmap (NULL, size, PROT_READ, MAP_PRIVATE, descriptor, pageoff);
+  if (map == MAP_FAILED)
+    {
+      error_callback (data, "mmap", errno);
+      return 0;
+    }
+
+  view->data = (char *) map + inpage;
+  view->base = map;
+  view->len = size;
+
+  return 1;
+}
+
+/* Release a view read by backtrace_get_view.  */
+
+void
+backtrace_release_view (struct backtrace_state *state ATTRIBUTE_UNUSED,
+			struct backtrace_view *view,
+			backtrace_error_callback error_callback,
+			void *data)
+{
+  union {
+    const void *cv;
+    void *v;
+  } cc;
+
+  cc.cv = view->base;
+  if (munmap (cc.v, view->len) < 0)
+    error_callback (data, "munmap", errno);
+}
+
+}

third_party/tracy/public/libbacktrace/sort.cpp

@@ -0,0 +1,113 @@
+/* sort.c -- Sort without allocating memory
+   Copyright (C) 2012-2021 Free Software Foundation, Inc.
+   Written by Ian Lance Taylor, Google.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+    (1) Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+
+    (2) 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.
+
+    (3) The name of the author may not be used to
+    endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.  */
+
+#include "config.h"
+
+#include <stddef.h>
+#include <sys/types.h>
+
+#include "backtrace.hpp"
+#include "internal.hpp"
+
+namespace tracy
+{
+
+/* The GNU glibc version of qsort allocates memory, which we must not
+   do if we are invoked by a signal handler.  So provide our own
+   sort.  */
+
+static void
+swap (char *a, char *b, size_t size)
+{
+  size_t i;
+
+  for (i = 0; i < size; i++, a++, b++)
+    {
+      char t;
+
+      t = *a;
+      *a = *b;
+      *b = t;
+    }
+}
+
+void
+backtrace_qsort (void *basearg, size_t count, size_t size,
+		 int (*compar) (const void *, const void *))
+{
+  char *base = (char *) basearg;
+  size_t i;
+  size_t mid;
+
+ tail_recurse:
+  if (count < 2)
+    return;
+
+  /* The symbol table and DWARF tables, which is all we use this
+     routine for, tend to be roughly sorted.  Pick the middle element
+     in the array as our pivot point, so that we are more likely to
+     cut the array in half for each recursion step.  */
+  swap (base, base + (count / 2) * size, size);
+
+  mid = 0;
+  for (i = 1; i < count; i++)
+    {
+      if ((*compar) (base, base + i * size) > 0)
+	{
+	  ++mid;
+	  if (i != mid)
+	    swap (base + mid * size, base + i * size, size);
+	}
+    }
+
+  if (mid > 0)
+    swap (base, base + mid * size, size);
+
+  /* Recurse with the smaller array, loop with the larger one.  That
+     ensures that our maximum stack depth is log count.  */
+  if (2 * mid < count)
+    {
+      backtrace_qsort (base, mid, size, compar);
+      base += (mid + 1) * size;
+      count -= mid + 1;
+      goto tail_recurse;
+    }
+  else
+    {
+      backtrace_qsort (base + (mid + 1) * size, count - (mid + 1),
+		       size, compar);
+      count = mid;
+      goto tail_recurse;
+    }
+}
+
+}

third_party/tracy/public/libbacktrace/state.cpp

@@ -0,0 +1,76 @@
+/* state.c -- Create the backtrace state.
+   Copyright (C) 2012-2021 Free Software Foundation, Inc.
+   Written by Ian Lance Taylor, Google.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are
+met:
+
+    (1) Redistributions of source code must retain the above copyright
+    notice, this list of conditions and the following disclaimer.
+
+    (2) 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.
+
+    (3) The name of the author may not be used to
+    endorse or promote products derived from this software without
+    specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``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 AUTHOR 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.  */
+
+#include "config.h"
+
+#include <string.h>
+#include <sys/types.h>
+
+#include "backtrace.hpp"
+#include "internal.hpp"
+
+namespace tracy
+{
+
+/* Create the backtrace state.  This will then be passed to all the
+   other routines.  */
+
+struct backtrace_state *
+backtrace_create_state (const char *filename, int threaded,
+			backtrace_error_callback error_callback,
+			void *data)
+{
+  struct backtrace_state init_state;
+  struct backtrace_state *state;
+
+#ifndef HAVE_SYNC_FUNCTIONS
+  if (threaded)
+    {
+      error_callback (data, "backtrace library does not support threads", 0);
+      return NULL;
+    }
+#endif
+
+  memset (&init_state, 0, sizeof init_state);
+  init_state.filename = filename;
+  init_state.threaded = threaded;
+
+  state = ((struct backtrace_state *)
+	   backtrace_alloc (&init_state, sizeof *state, error_callback, data));
+  if (state == NULL)
+    return NULL;
+  *state = init_state;
+
+  return state;
+}
+
+}

third_party/tracy/server/TracyCharUtil.hpp

@@ -0,0 +1,76 @@
+#ifndef __TRACY__CHARUTIL_HPP__
+#define __TRACY__CHARUTIL_HPP__
+
+#include <stddef.h>
+#include <stdint.h>
+#include <string.h>
+
+#define XXH_INLINE_ALL
+#include "tracy_xxhash.h"
+
+namespace tracy
+{
+namespace charutil
+{
+
+static inline size_t hash( const char* str )
+{
+    const auto sz = strlen( str );
+    return XXH3_64bits( str, sz );
+}
+
+static inline size_t hash( const char* str, size_t sz )
+{
+    return XXH3_64bits( str, sz );
+}
+
+struct Hasher
+{
+    size_t operator()( const char* key ) const
+    {
+        return hash( key );
+    }
+};
+
+struct Comparator
+{
+    bool operator()( const char* lhs, const char* rhs ) const
+    {
+        return strcmp( lhs, rhs ) == 0;
+    }
+};
+
+struct LessComparator
+{
+    bool operator()( const char* lhs, const char* rhs ) const
+    {
+        return strcmp( lhs, rhs ) < 0;
+    }
+};
+
+struct StringKey
+{
+    const char* ptr;
+    size_t sz;
+
+    struct Hasher
+    {
+        size_t operator()( const StringKey& key ) const
+        {
+            return hash( key.ptr, key.sz );
+        }
+    };
+
+    struct Comparator
+    {
+        bool operator()( const StringKey& lhs, const StringKey& rhs ) const
+        {
+            return lhs.sz == rhs.sz && memcmp( lhs.ptr, rhs.ptr, lhs.sz ) == 0;
+        }
+    };
+};
+
+}
+}
+
+#endif

third_party/tracy/server/TracyEvent.hpp

@@ -0,0 +1,850 @@
+#ifndef __TRACYEVENT_HPP__
+#define __TRACYEVENT_HPP__
+
+#include <assert.h>
+#include <limits>
+#include <stdint.h>
+#include <string>
+#include <string.h>
+
+#include "TracyCharUtil.hpp"
+#include "TracyShortPtr.hpp"
+#include "TracySortedVector.hpp"
+#include "TracyVector.hpp"
+#include "tracy_robin_hood.h"
+#include "../public/common/TracyForceInline.hpp"
+#include "../public/common/TracyQueue.hpp"
+
+namespace tracy
+{
+
+#pragma pack( push, 1 )
+
+struct StringRef
+{
+    enum Type { Ptr, Idx };
+
+    tracy_force_inline StringRef() : str( 0 ), __data( 0 ) {}
+    tracy_force_inline StringRef( Type t, uint64_t data )
+        : str( data )
+        , __data( 0 )
+    {
+        isidx = t == Idx;
+        active = 1;
+    }
+
+    uint64_t str;
+
+    union
+    {
+        struct
+        {
+            uint8_t isidx   : 1;
+            uint8_t active  : 1;
+        };
+        uint8_t __data;
+    };
+};
+
+struct StringRefHasher
+{
+    size_t operator()( const StringRef& key ) const
+    {
+        return charutil::hash( (const char*)&key, sizeof( StringRef ) );
+    }
+};
+
+struct StringRefComparator
+{
+    bool operator()( const StringRef& lhs, const StringRef& rhs ) const
+    {
+        return memcmp( &lhs, &rhs, sizeof( StringRef ) ) == 0;
+    }
+};
+
+class StringIdx
+{
+public:
+    tracy_force_inline StringIdx() { memset( m_idx, 0, sizeof( m_idx ) ); }
+    tracy_force_inline StringIdx( uint32_t idx )
+    {
+        SetIdx( idx );
+    }
+
+    tracy_force_inline void SetIdx( uint32_t idx )
+    {
+        idx++;
+        memcpy( m_idx, &idx, 3 );
+    }
+
+    tracy_force_inline uint32_t Idx() const
+    {
+        uint32_t idx = 0;
+        memcpy( &idx, m_idx, 3 );
+        assert( idx != 0 );
+        return idx - 1;
+    }
+
+    tracy_force_inline bool Active() const
+    {
+        uint32_t zero = 0;
+        return memcmp( m_idx, &zero, 3 ) != 0;
+    }
+
+private:
+    uint8_t m_idx[3];
+};
+
+struct StringIdxHasher
+{
+    size_t operator()( const StringIdx& key ) const
+    {
+        return charutil::hash( (const char*)&key, sizeof( StringIdx ) );
+    }
+};
+
+struct StringIdxComparator
+{
+    bool operator()( const StringIdx& lhs, const StringIdx& rhs ) const
+    {
+        return memcmp( &lhs, &rhs, sizeof( StringIdx ) ) == 0;
+    }
+};
+
+class Int24
+{
+public:
+    tracy_force_inline Int24() { memset( m_val, 0, sizeof( m_val ) ); }
+    tracy_force_inline Int24( uint32_t val )
+    {
+        SetVal( val );
+    }
+
+    tracy_force_inline void SetVal( uint32_t val )
+    {
+        memcpy( m_val, &val, 2 );
+        val >>= 16;
+        memcpy( m_val+2, &val, 1 );
+    }
+
+    tracy_force_inline uint32_t Val() const
+    {
+        uint8_t hi;
+        memcpy( &hi, m_val+2, 1 );
+        uint16_t lo;
+        memcpy( &lo, m_val, 2 );
+        return ( uint32_t( hi ) << 16 ) | lo;
+    }
+
+private:
+    uint8_t m_val[3];
+};
+
+class Int48
+{
+public:
+    tracy_force_inline Int48() {}
+    tracy_force_inline Int48( int64_t val )
+    {
+        SetVal( val );
+    }
+
+    tracy_force_inline void Clear()
+    {
+        memset( m_val, 0, 6 );
+    }
+
+    tracy_force_inline void SetVal( int64_t val )
+    {
+        memcpy( m_val, &val, 4 );
+        val >>= 32;
+        memcpy( m_val+4, &val, 2 );
+    }
+
+    tracy_force_inline int64_t Val() const
+    {
+        int16_t hi;
+        memcpy( &hi, m_val+4, 2 );
+        uint32_t lo;
+        memcpy( &lo, m_val, 4 );
+        return ( int64_t( uint64_t( hi ) << 32 ) ) | lo;
+    }
+
+    tracy_force_inline bool IsNonNegative() const
+    {
+        return ( m_val[5] >> 7 ) == 0;
+    }
+
+private:
+    uint8_t m_val[6];
+};
+
+struct Int48Sort { bool operator()( const Int48& lhs, const Int48& rhs ) { return lhs.Val() < rhs.Val(); }; };
+
+
+struct SourceLocationBase
+{
+    StringRef name;
+    StringRef function;
+    StringRef file;
+    uint32_t line;
+    uint32_t color;
+};
+
+struct SourceLocation : public SourceLocationBase
+{
+    mutable uint32_t namehash;
+};
+
+enum { SourceLocationSize = sizeof( SourceLocation ) };
+
+
+struct ZoneEvent
+{
+    tracy_force_inline ZoneEvent() {};
+
+    tracy_force_inline int64_t Start() const { return int64_t( _start_srcloc ) >> 16; }
+    tracy_force_inline void SetStart( int64_t start ) { assert( start < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_start_srcloc)+2, &start, 4 ); memcpy( ((char*)&_start_srcloc)+6, ((char*)&start)+4, 2 ); }
+    tracy_force_inline int64_t End() const { return int64_t( _end_child1 ) >> 16; }
+    tracy_force_inline void SetEnd( int64_t end ) { assert( end < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_end_child1)+2, &end, 4 ); memcpy( ((char*)&_end_child1)+6, ((char*)&end)+4, 2 ); }
+    tracy_force_inline bool IsEndValid() const { return ( _end_child1 >> 63 ) == 0; }
+    tracy_force_inline int16_t SrcLoc() const { return int16_t( _start_srcloc & 0xFFFF ); }
+    tracy_force_inline void SetSrcLoc( int16_t srcloc ) { memcpy( &_start_srcloc, &srcloc, 2 ); }
+    tracy_force_inline int32_t Child() const { int32_t child; memcpy( &child, &_child2, 4 ); return child; }
+    tracy_force_inline void SetChild( int32_t child ) { memcpy( &_child2, &child, 4 ); }
+    tracy_force_inline bool HasChildren() const { uint8_t tmp; memcpy( &tmp, ((char*)&_end_child1)+1, 1 ); return ( tmp >> 7 ) == 0; }
+
+    tracy_force_inline void SetStartSrcLoc( int64_t start, int16_t srcloc ) { assert( start < (int64_t)( 1ull << 47 ) ); start <<= 16; start |= uint16_t( srcloc ); memcpy( &_start_srcloc, &start, 8 ); }
+
+    uint64_t _start_srcloc;
+    uint16_t _child2;
+    uint64_t _end_child1;
+    uint32_t extra;
+};
+
+enum { ZoneEventSize = sizeof( ZoneEvent ) };
+static_assert( std::is_standard_layout<ZoneEvent>::value, "ZoneEvent is not standard layout" );
+
+
+struct ZoneExtra
+{
+    Int24 callstack;
+    StringIdx text;
+    StringIdx name;
+    Int24 color;
+};
+
+enum { ZoneExtraSize = sizeof( ZoneExtra ) };
+
+
+// This union exploits the fact that the current implementations of x64 and arm64 do not provide
+// full 64 bit address space. The high bits must be bit-extended, so 0x80... is an invalid pointer.
+// This allows using the highest bit as a selector between a native pointer and a table index here.
+union CallstackFrameId
+{
+    struct
+    {
+        uint64_t idx : 62;
+        uint64_t sel : 1;
+        uint64_t custom : 1;
+    };
+    uint64_t data;
+};
+
+enum { CallstackFrameIdSize = sizeof( CallstackFrameId ) };
+
+static tracy_force_inline bool operator==( const CallstackFrameId& lhs, const CallstackFrameId& rhs ) { return lhs.data == rhs.data; }
+
+
+struct SampleData
+{
+    Int48 time;
+    Int24 callstack;
+};
+
+enum { SampleDataSize = sizeof( SampleData ) };
+
+struct SampleDataSort { bool operator()( const SampleData& lhs, const SampleData& rhs ) { return lhs.time.Val() < rhs.time.Val(); }; };
+
+
+struct SampleDataRange
+{
+    Int48 time;
+    uint16_t thread;
+    CallstackFrameId ip;
+};
+
+enum { SampleDataRangeSize = sizeof( SampleDataRange ) };
+
+
+struct HwSampleData
+{
+    SortedVector<Int48, Int48Sort> cycles;
+    SortedVector<Int48, Int48Sort> retired;
+    SortedVector<Int48, Int48Sort> cacheRef;
+    SortedVector<Int48, Int48Sort> cacheMiss;
+    SortedVector<Int48, Int48Sort> branchRetired;
+    SortedVector<Int48, Int48Sort> branchMiss;
+
+    bool is_sorted() const
+    {
+        return
+            cycles.is_sorted() &&
+            retired.is_sorted() &&
+            cacheRef.is_sorted() &&
+            cacheMiss.is_sorted() &&
+            branchRetired.is_sorted() &&
+            branchMiss.is_sorted();
+    }
+
+    void sort()
+    {
+        if( !cycles.is_sorted() ) cycles.sort();
+        if( !retired.is_sorted() ) retired.sort();
+        if( !cacheRef.is_sorted() ) cacheRef.sort();
+        if( !cacheMiss.is_sorted() ) cacheMiss.sort();
+        if( !branchRetired.is_sorted() ) branchRetired.sort();
+        if( !branchMiss.is_sorted() ) branchMiss.sort();
+    }
+};
+
+enum { HwSampleDataSize = sizeof( HwSampleData ) };
+
+
+struct LockEvent
+{
+    enum class Type : uint8_t
+    {
+        Wait,
+        Obtain,
+        Release,
+        WaitShared,
+        ObtainShared,
+        ReleaseShared
+    };
+
+    tracy_force_inline int64_t Time() const { return int64_t( _time_srcloc ) >> 16; }
+    tracy_force_inline void SetTime( int64_t time ) { assert( time < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_time_srcloc)+2, &time, 4 ); memcpy( ((char*)&_time_srcloc)+6, ((char*)&time)+4, 2 ); }
+    tracy_force_inline int16_t SrcLoc() const { return int16_t( _time_srcloc & 0xFFFF ); }
+    tracy_force_inline void SetSrcLoc( int16_t srcloc ) { memcpy( &_time_srcloc, &srcloc, 2 ); }
+
+    uint64_t _time_srcloc;
+    uint8_t thread;
+    Type type;
+};
+
+struct LockEventShared : public LockEvent
+{
+    uint64_t waitShared;
+    uint64_t sharedList;
+};
+
+struct LockEventPtr
+{
+    short_ptr<LockEvent> ptr;
+    uint8_t lockingThread;
+    uint8_t lockCount;
+    uint64_t waitList;
+};
+
+enum { LockEventSize = sizeof( LockEvent ) };
+enum { LockEventSharedSize = sizeof( LockEventShared ) };
+enum { LockEventPtrSize = sizeof( LockEventPtr ) };
+
+enum { MaxLockThreads = sizeof( LockEventPtr::waitList ) * 8 };
+static_assert( std::numeric_limits<decltype(LockEventPtr::lockCount)>::max() >= MaxLockThreads, "Not enough space for lock count." );
+
+
+enum class LockType : uint8_t;
+
+struct LockMap
+{
+    struct TimeRange
+    {
+        int64_t start = std::numeric_limits<int64_t>::max();
+        int64_t end = std::numeric_limits<int64_t>::min();
+    };
+
+    StringIdx customName;
+    int16_t srcloc;
+    Vector<LockEventPtr> timeline;
+    unordered_flat_map<uint64_t, uint8_t> threadMap;
+    std::vector<uint64_t> threadList;
+    LockType type;
+    int64_t timeAnnounce;
+    int64_t timeTerminate;
+    bool valid;
+    bool isContended;
+    uint64_t lockingThread;
+
+    TimeRange range[64];
+};
+
+struct LockHighlight
+{
+    int64_t id;
+    int64_t begin;
+    int64_t end;
+    uint8_t thread;
+    bool blocked;
+};
+
+
+struct GpuEvent
+{
+    tracy_force_inline int64_t CpuStart() const { return int64_t( _cpuStart_srcloc ) >> 16; }
+    tracy_force_inline void SetCpuStart( int64_t cpuStart ) { assert( cpuStart < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_cpuStart_srcloc)+2, &cpuStart, 4 ); memcpy( ((char*)&_cpuStart_srcloc)+6, ((char*)&cpuStart)+4, 2 ); }
+    tracy_force_inline int64_t CpuEnd() const { return int64_t( _cpuEnd_thread ) >> 16; }
+    tracy_force_inline void SetCpuEnd( int64_t cpuEnd ) { assert( cpuEnd < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_cpuEnd_thread)+2, &cpuEnd, 4 ); memcpy( ((char*)&_cpuEnd_thread)+6, ((char*)&cpuEnd)+4, 2 ); }
+    tracy_force_inline int64_t GpuStart() const { return int64_t( _gpuStart_child1 ) >> 16; }
+    tracy_force_inline void SetGpuStart( int64_t gpuStart ) { /*assert( gpuStart < (int64_t)( 1ull << 47 ) );*/ memcpy( ((char*)&_gpuStart_child1)+2, &gpuStart, 4 ); memcpy( ((char*)&_gpuStart_child1)+6, ((char*)&gpuStart)+4, 2 ); }
+    tracy_force_inline int64_t GpuEnd() const { return int64_t( _gpuEnd_child2 ) >> 16; }
+    tracy_force_inline void SetGpuEnd( int64_t gpuEnd ) { assert( gpuEnd < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_gpuEnd_child2)+2, &gpuEnd, 4 ); memcpy( ((char*)&_gpuEnd_child2)+6, ((char*)&gpuEnd)+4, 2 ); }
+    tracy_force_inline int16_t SrcLoc() const { return int16_t( _cpuStart_srcloc & 0xFFFF ); }
+    tracy_force_inline void SetSrcLoc( int16_t srcloc ) { memcpy( &_cpuStart_srcloc, &srcloc, 2 ); }
+    tracy_force_inline uint16_t Thread() const { return uint16_t( _cpuEnd_thread & 0xFFFF ); }
+    tracy_force_inline void SetThread( uint16_t thread ) { memcpy( &_cpuEnd_thread, &thread, 2 ); }
+    tracy_force_inline int32_t Child() const { return int32_t( uint32_t( _gpuStart_child1 & 0xFFFF ) | ( uint32_t( _gpuEnd_child2 & 0xFFFF ) << 16 ) ); }
+    tracy_force_inline void SetChild( int32_t child ) { memcpy( &_gpuStart_child1, &child, 2 ); memcpy( &_gpuEnd_child2, ((char*)&child)+2, 2 ); }
+
+    uint64_t _cpuStart_srcloc;
+    uint64_t _cpuEnd_thread;
+    uint64_t _gpuStart_child1;
+    uint64_t _gpuEnd_child2;
+    Int24 callstack;
+};
+
+enum { GpuEventSize = sizeof( GpuEvent ) };
+static_assert( std::is_standard_layout<GpuEvent>::value, "GpuEvent is not standard layout" );
+
+
+struct MemEvent
+{
+    tracy_force_inline uint64_t Ptr() const { return uint64_t( int64_t( _ptr_csalloc1 ) >> 8 ); }
+    tracy_force_inline void SetPtr( uint64_t ptr ) { memcpy( ((char*)&_ptr_csalloc1)+1, &ptr, 4 ); memcpy( ((char*)&_ptr_csalloc1)+5, ((char*)&ptr)+4, 2 ); memcpy( ((char*)&_ptr_csalloc1)+7, ((char*)&ptr)+6, 1 ); }
+    tracy_force_inline uint64_t Size() const { return _size_csalloc2 >> 16; }
+    tracy_force_inline void SetSize( uint64_t size ) { assert( size < ( 1ull << 47 ) ); memcpy( ((char*)&_size_csalloc2)+2, &size, 4 ); memcpy( ((char*)&_size_csalloc2)+6, ((char*)&size)+4, 2 ); }
+    tracy_force_inline uint32_t CsAlloc() const { return uint8_t( _ptr_csalloc1 ) | ( uint16_t( _size_csalloc2 ) << 8 ); }
+    tracy_force_inline void SetCsAlloc( uint32_t csAlloc ) { memcpy( &_ptr_csalloc1, &csAlloc, 1 ); memcpy( &_size_csalloc2, ((char*)&csAlloc)+1, 2 ); }
+    tracy_force_inline int64_t TimeAlloc() const { return int64_t( _time_thread_alloc ) >> 16; }
+    tracy_force_inline void SetTimeAlloc( int64_t time ) { assert( time < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_time_thread_alloc)+2, &time, 4 ); memcpy( ((char*)&_time_thread_alloc)+6, ((char*)&time)+4, 2 ); }
+    tracy_force_inline int64_t TimeFree() const { return int64_t( _time_thread_free ) >> 16; }
+    tracy_force_inline void SetTimeFree( int64_t time ) { assert( time < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_time_thread_free)+2, &time, 4 ); memcpy( ((char*)&_time_thread_free)+6, ((char*)&time)+4, 2 ); }
+    tracy_force_inline uint16_t ThreadAlloc() const { return uint16_t( _time_thread_alloc ); }
+    tracy_force_inline void SetThreadAlloc( uint16_t thread ) { memcpy( &_time_thread_alloc, &thread, 2 ); }
+    tracy_force_inline uint16_t ThreadFree() const { return uint16_t( _time_thread_free ); }
+    tracy_force_inline void SetThreadFree( uint16_t thread ) { memcpy( &_time_thread_free, &thread, 2 ); }
+
+    tracy_force_inline void SetTimeThreadAlloc( int64_t time, uint16_t thread ) { time <<= 16; time |= thread; memcpy( &_time_thread_alloc, &time, 8 ); }
+    tracy_force_inline void SetTimeThreadFree( int64_t time, uint16_t thread ) { uint64_t t; memcpy( &t, &time, 8 ); t <<= 16; t |= thread; memcpy( &_time_thread_free, &t, 8 ); }
+
+    uint64_t _ptr_csalloc1;
+    uint64_t _size_csalloc2;
+    Int24 csFree;
+    uint64_t _time_thread_alloc;
+    uint64_t _time_thread_free;
+};
+
+enum { MemEventSize = sizeof( MemEvent ) };
+static_assert( std::is_standard_layout<MemEvent>::value, "MemEvent is not standard layout" );
+
+
+struct CallstackFrameBasic
+{
+    StringIdx name;
+    StringIdx file;
+    uint32_t line;
+};
+
+struct CallstackFrame : public CallstackFrameBasic
+{
+    uint64_t symAddr;
+};
+
+struct SymbolData : public CallstackFrameBasic
+{
+    StringIdx imageName;
+    StringIdx callFile;
+    uint32_t callLine;
+    uint8_t isInline;
+    Int24 size;
+};
+
+enum { CallstackFrameBasicSize = sizeof( CallstackFrameBasic ) };
+enum { CallstackFrameSize = sizeof( CallstackFrame ) };
+enum { SymbolDataSize = sizeof( SymbolData ) };
+
+
+struct SymbolLocation
+{
+    uint64_t addr;
+    uint32_t len;
+};
+
+enum { SymbolLocationSize = sizeof( SymbolLocation ) };
+
+
+struct CallstackFrameData
+{
+    short_ptr<CallstackFrame> data;
+    uint8_t size;
+    StringIdx imageName;
+};
+
+enum { CallstackFrameDataSize = sizeof( CallstackFrameData ) };
+
+
+struct MemCallstackFrameTree
+{
+    MemCallstackFrameTree( CallstackFrameId id ) : frame( id ), alloc( 0 ), count( 0 ) {}
+
+    CallstackFrameId frame;
+    uint64_t alloc;
+    uint32_t count;
+    unordered_flat_map<uint64_t, MemCallstackFrameTree> children;
+    unordered_flat_set<uint32_t> callstacks;
+};
+
+enum { MemCallstackFrameTreeSize = sizeof( MemCallstackFrameTree ) };
+
+
+struct CallstackFrameTree
+{
+    CallstackFrameTree( CallstackFrameId id ) : frame( id ), count( 0 ) {}
+
+    CallstackFrameId frame;
+    uint32_t count;
+    unordered_flat_map<uint64_t, CallstackFrameTree> children;
+};
+
+enum { CallstackFrameTreeSize = sizeof( CallstackFrameTree ) };
+
+
+struct CrashEvent
+{
+    uint64_t thread = 0;
+    int64_t time = 0;
+    uint64_t message = 0;
+    uint32_t callstack = 0;
+};
+
+enum { CrashEventSize = sizeof( CrashEvent ) };
+
+
+struct ContextSwitchData
+{
+    enum : int8_t { Fiber = 99 };
+    enum : int8_t { NoState = 100 };
+    enum : int8_t { Wakeup = -2 };
+
+    tracy_force_inline int64_t Start() const { return int64_t( _start_cpu ) >> 16; }
+    tracy_force_inline void SetStart( int64_t start ) { assert( start < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_start_cpu)+2, &start, 4 ); memcpy( ((char*)&_start_cpu)+6, ((char*)&start)+4, 2 ); }
+    tracy_force_inline int64_t End() const { return int64_t( _end_reason_state ) >> 16; }
+    tracy_force_inline void SetEnd( int64_t end ) { assert( end < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_end_reason_state)+2, &end, 4 ); memcpy( ((char*)&_end_reason_state)+6, ((char*)&end)+4, 2 ); }
+    tracy_force_inline bool IsEndValid() const { return ( _end_reason_state >> 63 ) == 0; }
+    tracy_force_inline uint8_t Cpu() const { return uint8_t( _start_cpu & 0xFF ); }
+    tracy_force_inline void SetCpu( uint8_t cpu ) { memcpy( &_start_cpu, &cpu, 1 ); }
+    tracy_force_inline int8_t Reason() const { return int8_t( (_end_reason_state >> 8) & 0xFF ); }
+    tracy_force_inline void SetReason( int8_t reason ) { memcpy( ((char*)&_end_reason_state)+1, &reason, 1 ); }
+    tracy_force_inline int8_t State() const { return int8_t( _end_reason_state & 0xFF ); }
+    tracy_force_inline void SetState( int8_t state ) { memcpy( &_end_reason_state, &state, 1 ); }
+    tracy_force_inline int64_t WakeupVal() const { return _wakeup.Val(); }
+    tracy_force_inline void SetWakeup( int64_t wakeup ) { assert( wakeup < (int64_t)( 1ull << 47 ) ); _wakeup.SetVal( wakeup ); }
+    tracy_force_inline uint16_t Thread() const { return _thread; }
+    tracy_force_inline void SetThread( uint16_t thread ) { _thread = thread; }
+
+    tracy_force_inline void SetStartCpu( int64_t start, uint8_t cpu ) { assert( start < (int64_t)( 1ull << 47 ) ); _start_cpu = ( uint64_t( start ) << 16 ) | cpu; }
+    tracy_force_inline void SetEndReasonState( int64_t end, int8_t reason, int8_t state ) { assert( end < (int64_t)( 1ull << 47 ) ); _end_reason_state = ( uint64_t( end ) << 16 ) | ( uint64_t( reason ) << 8 ) | uint8_t( state ); }
+
+    uint64_t _start_cpu;
+    uint64_t _end_reason_state;
+    Int48 _wakeup;
+    uint16_t _thread;
+};
+
+enum { ContextSwitchDataSize = sizeof( ContextSwitchData ) };
+
+
+struct ContextSwitchCpu
+{
+    tracy_force_inline int64_t Start() const { return int64_t( _start_thread ) >> 16; }
+    tracy_force_inline void SetStart( int64_t start ) { assert( start < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_start_thread)+2, &start, 4 ); memcpy( ((char*)&_start_thread)+6, ((char*)&start)+4, 2 ); }
+    tracy_force_inline int64_t End() const { int64_t v; memcpy( &v, ((char*)&_end)-2, 8 ); return v >> 16; }
+    tracy_force_inline void SetEnd( int64_t end ) { assert( end < (int64_t)( 1ull << 47 ) ); _end.SetVal( end ); }
+    tracy_force_inline bool IsEndValid() const { return _end.IsNonNegative(); }
+    tracy_force_inline uint16_t Thread() const { return uint16_t( _start_thread ); }
+    tracy_force_inline void SetThread( uint16_t thread ) { memcpy( &_start_thread, &thread, 2 ); }
+
+    tracy_force_inline void SetStartThread( int64_t start, uint16_t thread ) { assert( start < (int64_t)( 1ull << 47 ) ); _start_thread = ( uint64_t( start ) << 16 ) | thread; }
+
+    uint64_t _start_thread;
+    Int48 _end;
+};
+
+enum { ContextSwitchCpuSize = sizeof( ContextSwitchCpu ) };
+
+
+struct ContextSwitchUsage
+{
+    ContextSwitchUsage() {}
+    ContextSwitchUsage( int64_t time, uint8_t other, uint8_t own ) { SetTime( time ); SetOther( other ); SetOwn( own ); }
+
+    tracy_force_inline int64_t Time() const { return int64_t( _time_other_own ) >> 16; }
+    tracy_force_inline void SetTime( int64_t time ) { assert( time < (int64_t)( 1ull << 47 ) ); memcpy( ((char*)&_time_other_own)+2, &time, 4 ); memcpy( ((char*)&_time_other_own)+6, ((char*)&time)+4, 2 ); }
+    tracy_force_inline uint8_t Other() const { return uint8_t( _time_other_own ); }
+    tracy_force_inline void SetOther( uint8_t other ) { memcpy( &_time_other_own, &other, 1 ); }
+    tracy_force_inline uint8_t Own() const { uint8_t v; memcpy( &v, ((char*)&_time_other_own)+1, 1 );return v; }
+    tracy_force_inline void SetOwn( uint8_t own ) { memcpy( ((char*)&_time_other_own)+1, &own, 1 ); }
+
+    uint64_t _time_other_own;
+};
+
+enum { ContextSwitchUsageSize = sizeof( ContextSwitchUsage ) };
+
+
+struct MessageData
+{
+    int64_t time;
+    StringRef ref;
+    uint16_t thread;
+    uint32_t color;
+    Int24 callstack;
+};
+
+enum { MessageDataSize = sizeof( MessageData ) };
+
+
+struct PlotItem
+{
+    Int48 time;
+    double val;
+};
+
+enum { PlotItemSize = sizeof( PlotItem ) };
+
+
+struct FrameEvent
+{
+    int64_t start;
+    int64_t end;
+    int32_t frameImage;
+};
+
+enum { FrameEventSize = sizeof( FrameEvent ) };
+
+
+struct FrameImage
+{
+    short_ptr<const char> ptr;
+    uint32_t csz;
+    uint16_t w, h;
+    uint32_t frameRef;
+    uint8_t flip;
+};
+
+enum { FrameImageSize = sizeof( FrameImage ) };
+
+
+struct GhostZone
+{
+    Int48 start, end;
+    Int24 frame;
+    int32_t child;
+};
+
+enum { GhostZoneSize = sizeof( GhostZone ) };
+
+
+struct ChildSample
+{
+    Int48 time;
+    uint64_t addr;
+};
+
+enum { ChildSampleSize = sizeof( ChildSample ) };
+
+#pragma pack( pop )
+
+
+struct ThreadData
+{
+    uint64_t id;
+    uint64_t count;
+    Vector<short_ptr<ZoneEvent>> timeline;
+    Vector<short_ptr<ZoneEvent>> stack;
+    Vector<short_ptr<MessageData>> messages;
+    uint32_t nextZoneId;
+    Vector<uint32_t> zoneIdStack;
+#ifndef TRACY_NO_STATISTICS
+    Vector<int64_t> childTimeStack;
+    Vector<GhostZone> ghostZones;
+    uint64_t ghostIdx;
+    SortedVector<SampleData, SampleDataSort> postponedSamples;
+#endif
+    Vector<SampleData> samples;
+    SampleData pendingSample;
+    Vector<SampleData> ctxSwitchSamples;
+    uint64_t kernelSampleCnt;
+    uint8_t isFiber;
+    ThreadData* fiber;
+    uint8_t* stackCount;
+    int32_t groupHint;
+
+    tracy_force_inline void IncStackCount( int16_t srcloc ) { stackCount[uint16_t(srcloc)]++; }
+    tracy_force_inline bool DecStackCount( int16_t srcloc ) { return --stackCount[uint16_t(srcloc)] != 0; }
+};
+
+struct GpuCtxThreadData
+{
+    Vector<short_ptr<GpuEvent>> timeline;
+    Vector<short_ptr<GpuEvent>> stack;
+};
+
+struct GpuCtxData
+{
+    int64_t timeDiff;
+    uint64_t thread;
+    uint64_t count;
+    float period;
+    GpuContextType type;
+    bool hasPeriod;
+    bool hasCalibration;
+    int64_t calibratedGpuTime;
+    int64_t calibratedCpuTime;
+    double calibrationMod;
+    int64_t lastGpuTime;
+    uint64_t overflow;
+    uint32_t overflowMul;
+    StringIdx name;
+    unordered_flat_map<uint64_t, GpuCtxThreadData> threadData;
+    short_ptr<GpuEvent> query[64*1024];
+};
+
+enum { GpuCtxDataSize = sizeof( GpuCtxData ) };
+
+
+enum class PlotType : uint8_t
+{
+    User,
+    Memory,
+    SysTime,
+    Power
+};
+
+// Keep this in sync with enum in TracyC.h
+enum class PlotValueFormatting : uint8_t
+{
+    Number,
+    Memory,
+    Percentage,
+    Watt
+};
+
+struct PlotData
+{
+    struct PlotItemSort { bool operator()( const PlotItem& lhs, const PlotItem& rhs ) { return lhs.time.Val() < rhs.time.Val(); }; };
+
+    uint64_t name;
+    double min;
+    double max;
+    double sum;
+    SortedVector<PlotItem, PlotItemSort> data;
+    PlotType type;
+    PlotValueFormatting format;
+    uint8_t showSteps;
+    uint8_t fill;
+    uint32_t color;
+
+    double rMin, rMax, num;
+};
+
+struct MemData
+{
+    Vector<MemEvent> data;
+    Vector<uint32_t> frees;
+    unordered_flat_map<uint64_t, size_t> active;
+    uint64_t high = std::numeric_limits<uint64_t>::min();
+    uint64_t low = std::numeric_limits<uint64_t>::max();
+    uint64_t usage = 0;
+    PlotData* plot = nullptr;
+    bool reconstruct = false;
+    uint64_t name = 0;
+};
+
+struct FrameData
+{
+    uint64_t name;
+    Vector<FrameEvent> frames;
+    uint8_t continuous;
+
+    int64_t min = std::numeric_limits<int64_t>::max();
+    int64_t max = std::numeric_limits<int64_t>::min();
+    int64_t total = 0;
+    double sumSq = 0;
+};
+
+struct StringLocation
+{
+    const char* ptr;
+    uint32_t idx;
+};
+
+struct SourceLocationHasher
+{
+    size_t operator()( const SourceLocation* ptr ) const
+    {
+        return charutil::hash( (const char*)ptr, sizeof( SourceLocationBase ) );
+    }
+};
+
+struct SourceLocationComparator
+{
+    bool operator()( const SourceLocation* lhs, const SourceLocation* rhs ) const
+    {
+        return memcmp( lhs, rhs, sizeof( SourceLocationBase ) ) == 0;
+    }
+};
+
+
+struct ContextSwitch
+{
+    Vector<ContextSwitchData> v;
+    int64_t runningTime = 0;
+};
+
+struct CpuData
+{
+    Vector<ContextSwitchCpu> cs;
+};
+
+struct CpuThreadData
+{
+    int64_t runningTime = 0;
+    uint32_t runningRegions = 0;
+    uint32_t migrations = 0;
+};
+
+enum { CpuThreadDataSize = sizeof( CpuThreadData ) };
+
+
+struct Parameter
+{
+    uint32_t idx;
+    StringRef name;
+    bool isBool;
+    int32_t val;
+};
+
+
+struct SymbolStats
+{
+    uint32_t incl, excl;
+    unordered_flat_map<uint32_t, uint32_t> parents;
+    unordered_flat_map<uint32_t, uint32_t> baseParents;
+};
+
+enum { SymbolStatsSize = sizeof( SymbolStats ) };
+
+}
+
+#endif

third_party/tracy/server/TracyFileHeader.hpp

@@ -0,0 +1,22 @@
+#ifndef __TRACYFILEHEADER_HPP__
+#define __TRACYFILEHEADER_HPP__
+
+#include <stdint.h>
+
+#include "../public/common/TracyForceInline.hpp"
+
+namespace tracy
+{
+
+static const uint8_t TracyHeader[4] = { 't', 'r', 253, 'P' };
+static const uint8_t Lz4Header[4]  = { 't', 'l', 'Z', 4 };
+static const uint8_t ZstdHeader[4] = { 't', 'Z', 's', 't' };
+
+static constexpr tracy_force_inline int FileVersion( uint8_t h5, uint8_t h6, uint8_t h7 )
+{
+    return ( h5 << 16 ) | ( h6 << 8 ) | h7;
+}
+
+}
+
+#endif