Initial Wayland support

Signed-off-by: Slendi <slendi@socopon.com>

.gitignore

@@ -3,3 +3,4 @@ result
 .cache
 .direnv
 .clangd
+screenshot*

flake.lock

@@ -20,12 +20,12 @@
     },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1764517877,
+        "lastModified": 1767892417,
-        "narHash": "sha256-pp3uT4hHijIC8JUK5MEqeAWmParJrgBVzHLNfJDZxg4=",
+        "narHash": "sha256-dhhvQY67aboBk8b0/u0XB6vwHdgbROZT3fJAjyNh5Ww=",
-        "rev": "2d293cbfa5a793b4c50d17c05ef9e385b90edf6c",
+        "rev": "3497aa5c9457a9d88d71fa93a4a8368816fbeeba",
-        "revCount": 904649,
+        "revCount": 924538,
         "type": "tarball",
-        "url": "https://api.flakehub.com/f/pinned/NixOS/nixpkgs/0.1.904649%2Brev-2d293cbfa5a793b4c50d17c05ef9e385b90edf6c/019ad7f2-e8f3-79e9-ad92-dd7a45c069d3/source.tar.gz?rev=2d293cbfa5a793b4c50d17c05ef9e385b90edf6c&revCount=904649"
+        "url": "https://api.flakehub.com/f/pinned/NixOS/nixpkgs/0.1.924538%2Brev-3497aa5c9457a9d88d71fa93a4a8368816fbeeba/019ba2f6-9b41-7674-b81c-5f768968b13a/source.tar.gz?rev=3497aa5c9457a9d88d71fa93a4a8368816fbeeba&revCount=924538"
       },
       "original": {
         "type": "tarball",

flake.nix

@@ -23,6 +23,7 @@
           pkg-config
           glslang
           shaderc
+          wayland-scanner
         ];
         buildInputs = with pkgs; [
           vulkan-loader
@@ -51,6 +52,7 @@
                   doxygen
                   gtest
                   cppcheck
+                  tracy
                 ]
                 ++ buildInputs
                 ++ nativeBuildInputs

meson.build

@@ -1,4 +1,4 @@
-project('vr-compositor', 'cpp',
+project('vr-compositor', 'c', 'cpp',
 	version: '0.1',
 	default_options: [
 		'cpp_std=c++26',
@@ -17,14 +17,32 @@ fastgltf_opts.set_override_option('werror', 'false')
 fastgltf = cmake.subproject('fastgltf', options: fastgltf_opts)
 
 cc = meson.get_compiler('cpp')
+wl_mod = import('wayland')
 
-wayland_dep = dependency('wayland-server')
+wayland_dep = dependency('wayland-server', include_type: 'system')
-vulkan_dep = dependency('vulkan')
+wayland_client_dep = dependency('wayland-client', include_type: 'system')
-openxr_dep = dependency('openxr')
+vulkan_dep = dependency('vulkan', include_type: 'system')
-zlib_dep = dependency('zlib')
+openxr_dep = dependency('openxr', include_type: 'system')
-sdl3_dep = dependency('sdl3')
+zlib_dep = dependency('zlib', include_type: 'system')
-libinput_dep = dependency('libinput')
+sdl3_dep = dependency('sdl3', include_type: 'system')
-libudev_dep = dependency('libudev')
+libinput_dep = dependency('libinput', include_type: 'system')
+libudev_dep = dependency('libudev', include_type: 'system')
+wayland_protocol = wl_mod.scan_xml(
+	'protocols/wayland.xml',
+	client: false,
+	server: true,
+)
+wayland_protocol_source = wayland_protocol[0]
+wayland_server_header = wayland_protocol[1]
+
+xdg_shell_protocol = wl_mod.scan_xml(
+	'protocols/xdg-shell.xml',
+	client: true,
+	server: true,
+)
+xdg_shell_protocol_source = xdg_shell_protocol[0]
+xdg_shell_client_header = xdg_shell_protocol[1]
+xdg_shell_server_header = xdg_shell_protocol[2]
 imgui_src = files(
 	'thirdparty/imgui/imgui.cpp',
 	'thirdparty/imgui/imgui_draw.cpp',
@@ -37,9 +55,25 @@ fastgltf_dep = fastgltf.dependency('fastgltf')
 
 vkbootstrap_dev = get_option('vkbootstrap_dev')
 vkbootstrap_lib = get_option('vkbootstrap_lib')
+tracy_enable = get_option('tracy_enable')
+
+if tracy_enable and get_option('buildtype') != 'debugoptimized'
+	warning('Profiling builds should set --buildtype=debugoptimized')
+endif
+
+tracy = disabler()
+if tracy_enable
+	tracy_proj = subproject('tracy', default_options: [
+		'default_library=static',
+		'warning_level=0',
+		'werror=false',
+	])
+	tracy = tracy_proj.get_variable('tracy_dep')
+endif
 
 vkbootstrap_inc = include_directories(
-	join_paths(vkbootstrap_dev, 'include')
+	join_paths(vkbootstrap_dev, 'include'),
+	is_system: true,
 )
 
 vkbootstrap_dep = cc.find_library(
@@ -78,6 +112,8 @@ add_project_arguments(
 		'-Wno-exit-time-destructors',
 		'-Wno-zero-as-null-pointer-constant',
 		'-Wno-unused-macros',
+		'-Wno-reserved-macro-identifier',
+		'-Wno-reserved-identifier',
 		'-Wno-suggest-override',
 		'-Wno-macro-redefined',
 		'-DVULKAN_HPP_DEFAULT_DISPATCH_LOADER_DYNAMIC_STORAGE',
@@ -114,6 +150,21 @@ imgui_lib = static_library('imgui',
 	],
 )
 
+exe_deps = [
+	wayland_dep,
+	vulkan_dep,
+	openxr_dep,
+	vkbootstrap_dep,
+	zlib_dep,
+	sdl3_dep,
+	fastgltf_dep,
+	libinput_dep,
+	libudev_dep,
+]
+if tracy_enable
+	exe_deps += tracy
+endif
+
 exe = executable('vr-compositor',
 	[
 		'src/main.cpp',
@@ -127,27 +178,39 @@ exe = executable('vr-compositor',
 		'src/Pipeline.cpp',
 		'src/Loader.cpp',
 		'src/DescriptorWriter.cpp',
+		'src/CPUTexture.cpp',
+		'src/wayland/WaylandServer.cpp',
+		'src/wayland/Surface.cpp',
+		'src/wayland/Shm.cpp',
+		'src/wayland/protocols/CompositorProtocol.cpp',
+		'src/wayland/protocols/ShmProtocol.cpp',
+		'src/wayland/protocols/XdgShellProtocol.cpp',
+		'src/Skybox.cpp',
 		'src/VulkanRenderer.cpp',
 		'src/Application.cpp',
+		wayland_protocol_source,
+		xdg_shell_protocol_source,
+		wayland_server_header,
+		xdg_shell_server_header,
 	],
+	c_args: ['-Wno-missing-variable-declarations'],
 	include_directories: [
 		vkbootstrap_inc,
+		include_directories('.'),
 		imgui_inc,
 		'thirdparty/smath/include'
 	],
 	link_with: imgui_lib,
-	dependencies: [
+	dependencies: exe_deps,
-		wayland_dep,
-		vulkan_dep,
-		openxr_dep,
-		vkbootstrap_dep,
-		zlib_dep,
-		sdl3_dep,
-		fastgltf_dep,
-		libinput_dep,
-		libudev_dep,
-	],
 	cpp_args: [
-		'--embed-dir=' + join_paths(meson.project_build_root(), 'shaders')
+		'--embed-dir=' + join_paths(meson.project_build_root(), 'shaders'),
 	],
 )
+
+executable('shm-life',
+	'tools/shm_life.cpp',
+	dependencies: [wayland_client_dep],
+	sources: [xdg_shell_protocol_source, xdg_shell_client_header],
+	c_args: ['-Wno-missing-variable-declarations'],
+	cpp_args: ['-Wno-cast-qual'],
+)

meson_options.txt

@@ -1,2 +1,3 @@
 option('vkbootstrap_dev', type: 'string', description: 'vk-bootstrap dev output path')
 option('vkbootstrap_lib', type: 'string', description: 'vk-bootstrap lib output path')
+option('tracy_enable', type: 'boolean', value: false, description: 'Enable profiling')

shaders/gradient.comp

@@ -1,23 +0,0 @@
-#version 460
-
-layout (local_size_x = 16, local_size_y = 16) in;
-layout(rgba16f, set = 0, binding = 0) uniform image2D image;
-
-void main() {
-	ivec2 texelCoord = ivec2(gl_GlobalInvocationID.xy);
-	ivec2 size = imageSize(image);
-
-	if (texelCoord.x >= size.x || texelCoord.y >= size.y)
-		return;
-
-	vec2 uv = (vec2(texelCoord) + 0.5) / vec2(size);
-
-	float v = sin(uv.x * 10.0) + cos(uv.y * 10.0);
-
-	float r = 0.5 + 0.5 * cos(6.2831 * (uv.x + v));
-	float g = 0.5 + 0.5 * cos(6.2831 * (uv.y + v + 0.33));
-	float b = 0.5 + 0.5 * cos(6.2831 * (uv.x - uv.y + 0.66));
-
-	vec4 color = vec4(r, g, b, 1.0);
-	imageStore(image, texelCoord, color);
-}

shaders/gradient_fullscreen.frag

@@ -1,21 +0,0 @@
-#version 450
-
-layout(location = 0) out vec4 out_color;
-
-layout(push_constant) uniform PushConstants {
-	vec2 resolution;
-} pc;
-
-void main()
-{
-	vec2 uv = (gl_FragCoord.xy + vec2(0.5)) / pc.resolution;
-
-	float v = sin(uv.x * 10.0) + cos(uv.y * 10.0);
-
-	float r = 0.5 + 0.5 * cos(6.2831 * (uv.x + v));
-	float g = 0.5 + 0.5 * cos(6.2831 * (uv.y + v + 0.33));
-	float b = 0.5 + 0.5 * cos(6.2831 * (uv.x - uv.y + 0.66));
-
-	out_color = vec4(r, g, b, 1.0);
-}
-

shaders/gradient_fullscreen.vert

@@ -1,13 +0,0 @@
-#version 450
-
-void main()
-{
-	vec2 positions[3] = vec2[](
-		vec2(-1.0, -1.0),
-		vec2( 3.0, -1.0),
-		vec2(-1.0,  3.0)
-	);
-
-	gl_Position = vec4(positions[gl_VertexIndex], 0.0, 1.0);
-}
-

shaders/meson.build

@@ -19,6 +19,8 @@ shader_sources = files(
 	'triangle_mesh.frag',
 	'triangle_mesh.vert',
 	'tex_image.frag',
+	'skybox.frag',
+	'skybox.vert',
 )
 
 spirv_shaders = []

shaders/skybox.frag

@@ -0,0 +1,12 @@
+#version 450
+
+layout (location = 0) in vec3 in_dir;
+
+layout (location = 0) out vec4 out_frag_color;
+
+layout (set = 0, binding = 0) uniform samplerCube environment_map;
+
+void main() {
+	vec3 dir = normalize(in_dir);
+	out_frag_color = vec4(texture(environment_map, dir).rgb, 1.0f);
+}

shaders/skybox.vert

@@ -0,0 +1,14 @@
+#version 450
+
+layout (location = 0) in vec3 in_position;
+layout (location = 0) out vec3 out_dir;
+
+layout(push_constant) uniform constants {
+	mat4 mvp;
+} PushConstants;
+
+void main() {
+	out_dir = in_position;
+	vec4 pos = PushConstants.mvp * vec4(in_position, 1.0f);
+	gl_Position = vec4(pos.xy, pos.w, pos.w);
+}

src/Application.h

@@ -1,12 +1,24 @@
 #pragma once
 
+#include <array>
+#include <cstdint>
+#include <filesystem>
+#include <functional>
 #include <memory>
+#include <vector>
 
 #include <SDL3/SDL_video.h>
-
-#include "Logger.h"
-
 #include <imgui.h>
+#include <linux/input-event-codes.h>
+#include <openxr/openxr.h>
+
+#include "smath.hpp"
+
+#include "Loader.h"
+#include "Logger.h"
+#include "Skybox.h"
+#include "Types.h"
+#include "wayland/Display.h"
 
 struct libinput;
 struct libinput_event_keyboard;
@@ -16,38 +28,99 @@ struct udev;
 namespace Lunar {
 
 struct VulkanRenderer;
+struct OpenXrState;
+namespace Wayland {
+struct WaylandServer;
+}
 
 struct Application {
-	Application();
-	~Application();
-
 	auto run() -> void;
+	auto binary_directory() const -> std::filesystem::path;
 
 	auto mouse_captured(bool new_state) -> void;
 	auto mouse_captured() const -> bool { return m_mouse_captured; }
 	auto toggle_mouse_captured() -> void { mouse_captured(!m_mouse_captured); }
+	auto is_key_down(uint32_t key) const -> bool;
+	auto is_key_up(uint32_t key) const -> bool;
+	auto is_key_pressed(uint32_t key) const -> bool;
+	auto is_key_released(uint32_t key) const -> bool;
+
+	static auto the() -> Application &;
 
 private:
+	enum class Backend {
+		SDL,
+		KMS,
+	};
+
+	Application();
+	~Application();
+
 	auto init_input() -> void;
+	auto init_test_meshes() -> void;
+	auto init_wayland() -> void;
+	auto asset_directory() -> std::filesystem::path;
 	auto shutdown_input() -> void;
 	auto process_libinput_events() -> void;
 	auto handle_keyboard_event(libinput_event_keyboard *event) -> void;
+	auto handle_pointer_motion(libinput_event_pointer *event) -> void;
+	auto handle_pointer_button(libinput_event_pointer *event) -> void;
+	auto handle_pointer_axis(libinput_event_pointer *event) -> void;
+	auto handle_pointer_frame() -> void;
+	auto handle_pointer_end_frame() -> void;
+	auto handle_pointer_cancel() -> void;
+	auto handle_keyboard_key(std::optional<uint32_t> key, bool pressed) -> void;
 	auto clamp_mouse_to_window(int width, int height) -> void;
+	auto init_openxr() -> void;
+	auto init_openxr_session() -> void;
+	auto shutdown_openxr() -> void;
+	auto poll_openxr_events() -> void;
+	auto render_openxr_frame(
+	    std::function<void(VulkanRenderer::GL &)> const &record,
+	    float dt_seconds) -> bool;
+	auto update_camera_from_xr_view(XrView const &view) -> void;
+	auto update_hands(XrTime display_time) -> void;
+	auto render_hands(
+	    VulkanRenderer::GL &gl, smath::Mat4 const &view_projection) -> void;
 
 	SDL_Window *m_window { nullptr };
+	Backend m_backend { Backend::SDL };
 	Logger m_logger { "Lunar" };
 	std::unique_ptr<VulkanRenderer> m_renderer;
+	Skybox m_skybox;
+	std::vector<std::shared_ptr<Mesh>> m_test_meshes;
 
 	udev *m_udev { nullptr };
 	libinput *m_libinput { nullptr };
 
+	std::unique_ptr<OpenXrState> m_openxr {};
+	std::unique_ptr<Wayland::WaylandServer> m_wayland {};
+
 	bool m_running { true };
 	bool m_mouse_captured { false };
 	bool m_show_imgui { false };
+	bool m_window_focused { true };
 	int m_ctrl_pressed_count { 0 };
+	std::uint32_t m_screenshot_index { 0 };
 
 	double m_mouse_x { 0.0 };
 	double m_mouse_y { 0.0 };
+	double m_mouse_dx { 0.0 };
+	double m_mouse_dy { 0.0 };
+	float m_mouse_sensitivity { 0.001f };
+
+	std::array<bool, KEY_MAX + 1> m_key_state {};
+	std::array<bool, KEY_MAX + 1> m_key_state_previous {};
+
+	Camera m_camera;
+	PolarCoordinate m_cursor;
+
+	static inline std::array<smath::Vec3, XR_HAND_JOINT_COUNT_EXT>
+	    m_left_joints {};
+	static inline std::array<smath::Vec3, XR_HAND_JOINT_COUNT_EXT>
+	    m_right_joints {};
+	static inline bool m_left_hand_valid { false };
+	static inline bool m_right_hand_valid { false };
 };
 
 } // namespace Lunar

src/CPUTexture.cpp

@@ -0,0 +1,88 @@
+#include "CPUTexture.h"
+
+#include <format>
+#include <stdexcept>
+
+#if defined(__clang__)
+#	pragma clang diagnostic push
+#	pragma clang diagnostic ignored "-Wreserved-identifier"
+#	pragma clang diagnostic ignored "-Wcast-qual"
+#	pragma clang diagnostic ignored "-Wimplicit-fallthrough"
+#	pragma clang diagnostic ignored "-Wmissing-field-initializers"
+#	pragma clang diagnostic ignored "-Wused-but-marked-unused"
+#	pragma clang diagnostic ignored "-Wmissing-prototypes"
+#	pragma clang diagnostic ignored "-Wextra-semi-stmt"
+#	pragma clang diagnostic ignored "-Wimplicit-int-conversion"
+#	pragma clang diagnostic ignored "-Wsign-conversion"
+#	pragma clang diagnostic ignored "-Wshorten-64-to-32"
+#	pragma clang diagnostic ignored "-Wconversion"
+#	pragma clang diagnostic ignored "-Wcomma"
+#	pragma clang diagnostic ignored "-Wdouble-promotion"
+#	pragma clang diagnostic ignored "-Wimplicit-float-conversion"
+#	pragma clang diagnostic ignored "-Wunsafe-buffer-usage"
+#	pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#	pragma clang diagnostic ignored "-Wdisabled-macro-expansion"
+#	pragma clang diagnostic ignored "-Wsign-compare"
+#	pragma clang diagnostic ignored "-Wfloat-equal"
+#	pragma clang diagnostic ignored "-Wpacked"
+#	pragma clang diagnostic ignored "-Wold-style-cast"
+#	pragma clang diagnostic ignored "-Wexit-time-destructors"
+#	pragma clang diagnostic ignored "-Wglobal-constructors"
+#	pragma clang diagnostic ignored "-Wzero-as-null-pointer-constant"
+#	pragma clang diagnostic ignored "-Wcast-align"
+#	pragma clang diagnostic ignored "-Wcast-qual"
+#	pragma clang diagnostic ignored "-Wshadow"
+#	pragma clang diagnostic ignored "-Wnewline-eof"
+#	pragma clang diagnostic ignored "-Wformat-nonliteral"
+#	pragma clang diagnostic ignored "-Wswitch-default"
+#	pragma clang diagnostic ignored "-Wswitch-enum"
+#	pragma clang diagnostic ignored "-Wcovered-switch-default"
+#	pragma clang diagnostic ignored "-Wdocumentation"
+#	pragma clang diagnostic ignored "-Wdocumentation-unknown-command"
+#	pragma clang diagnostic ignored "-Wextra-semi"
+#	pragma clang diagnostic ignored "-Wundef"
+#	pragma clang diagnostic ignored "-Wreserved-macro-identifier"
+#	pragma clang diagnostic ignored "-Wc++98-compat"
+#	pragma clang diagnostic ignored "-Wc++98-compat-pedantic"
+#	pragma clang diagnostic ignored "-Wweak-vtables"
+#	pragma clang diagnostic ignored "-Wswitch"
+#	pragma clang diagnostic ignored "-Wunused-macros"
+#	pragma clang diagnostic ignored "-Wextra"
+#endif
+#define STB_IMAGE_IMPLEMENTATION
+#include "../thirdparty/stb/stb_image.h"
+#if defined(__clang__)
+#	pragma clang diagnostic pop
+#endif
+
+namespace Lunar {
+
+CPUTexture::CPUTexture(std::filesystem::path const &path)
+{
+	int width_out { 0 };
+	int height_out { 0 };
+	int channels_out { 0 };
+	stbi_uc *data { stbi_load(path.string().c_str(), &width_out, &height_out,
+		&channels_out, STBI_rgb_alpha) };
+	if (!data) {
+		throw std::runtime_error(
+		    std::format("Failed to load texture: {}", path.string()));
+	}
+
+	width = static_cast<uint32_t>(width_out);
+	height = static_cast<uint32_t>(height_out);
+	format = vk::Format::eR8G8B8A8Unorm;
+	pixels.assign(data, data + (width * height * 4));
+	stbi_image_free(data);
+}
+
+CPUTexture::CPUTexture(std::vector<uint8_t> pixels, uint32_t width,
+    uint32_t height, vk::Format format)
+    : pixels(std::move(pixels))
+    , width(width)
+    , height(height)
+    , format(format)
+{
+}
+
+} // namespace Lunar

src/CPUTexture.h

@@ -0,0 +1,22 @@
+#pragma once
+
+#include <cstdint>
+#include <filesystem>
+#include <vector>
+
+#include <vulkan/vulkan.hpp>
+
+namespace Lunar {
+
+struct CPUTexture {
+	std::vector<uint8_t> pixels;
+	uint32_t width { 0 };
+	uint32_t height { 0 };
+	vk::Format format { vk::Format::eR8G8B8A8Unorm };
+
+	explicit CPUTexture(std::filesystem::path const &path);
+	CPUTexture(std::vector<uint8_t> pixels, uint32_t width, uint32_t height,
+	    vk::Format format);
+};
+
+} // namespace Lunar

src/DescriptorAllocatorGrowable.cpp

@@ -78,7 +78,7 @@ auto DescriptorAllocatorGrowable::destroy_pools(VkDevice dev) -> void
 auto DescriptorAllocatorGrowable::allocate(Logger &logger, VkDevice dev,
     VkDescriptorSetLayout layout, void *p_next) -> VkDescriptorSet
 {
-	auto pool_to_use = get_pool(dev);
+	auto pool_to_use { get_pool(dev) };
 
 	VkDescriptorSetAllocateInfo alloci {};
 	alloci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO;

src/GraphicsPipelineBuilder.cpp

@@ -32,6 +32,8 @@ auto GraphicsPipelineBuilder::clear() -> GraphicsPipelineBuilder &
 	m_render_info.sType = VK_STRUCTURE_TYPE_PIPELINE_RENDERING_CREATE_INFO;
 
 	m_shader_stages.clear();
+	m_vertex_bindings.clear();
+	m_vertex_attributes.clear();
 
 	return *this;
 }
@@ -168,6 +170,17 @@ auto GraphicsPipelineBuilder::set_pipeline_layout(VkPipelineLayout layout)
 	return *this;
 }
 
+auto GraphicsPipelineBuilder::set_vertex_input(
+    std::span<VkVertexInputBindingDescription const> bindings,
+    std::span<VkVertexInputAttributeDescription const> attributes)
+    -> GraphicsPipelineBuilder &
+{
+	m_vertex_bindings.assign(bindings.begin(), bindings.end());
+	m_vertex_attributes.assign(attributes.begin(), attributes.end());
+
+	return *this;
+}
+
 auto GraphicsPipelineBuilder::disable_depth_testing()
     -> GraphicsPipelineBuilder &
 {
@@ -223,6 +236,12 @@ auto GraphicsPipelineBuilder::build(VkDevice dev) -> VkPipeline
 	VkPipelineVertexInputStateCreateInfo vertex_input_ci {};
 	vertex_input_ci.sType
 	    = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO;
+	vertex_input_ci.vertexBindingDescriptionCount
+	    = static_cast<uint32_t>(m_vertex_bindings.size());
+	vertex_input_ci.pVertexBindingDescriptions = m_vertex_bindings.data();
+	vertex_input_ci.vertexAttributeDescriptionCount
+	    = static_cast<uint32_t>(m_vertex_attributes.size());
+	vertex_input_ci.pVertexAttributeDescriptions = m_vertex_attributes.data();
 
 	VkGraphicsPipelineCreateInfo pipeline_ci {};
 	pipeline_ci.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO;

src/GraphicsPipelineBuilder.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <span>
 #include <vector>
 
 #include <vulkan/vulkan_core.h>
@@ -35,6 +36,10 @@ struct GraphicsPipelineBuilder {
 	auto set_depth_format(VkFormat format) -> GraphicsPipelineBuilder &;
 	auto set_pipeline_layout(VkPipelineLayout layout)
 	    -> GraphicsPipelineBuilder &;
+	auto set_vertex_input(
+	    std::span<VkVertexInputBindingDescription const> bindings,
+	    std::span<VkVertexInputAttributeDescription const> attributes)
+	    -> GraphicsPipelineBuilder &;
 	auto disable_depth_testing() -> GraphicsPipelineBuilder &;
 	auto enable_depth_testing(bool depth_write_enable = true,
 	    VkCompareOp op = VK_COMPARE_OP_LESS_OR_EQUAL)
@@ -52,6 +57,8 @@ private:
 	VkFormat m_color_attachment_format {};
 
 	std::vector<VkPipelineShaderStageCreateInfo> m_shader_stages {};
+	std::vector<VkVertexInputBindingDescription> m_vertex_bindings {};
+	std::vector<VkVertexInputAttributeDescription> m_vertex_attributes {};
 
 	Logger &m_logger;
 };

src/Loader.cpp

@@ -61,7 +61,7 @@ auto Mesh::load_gltf_meshes(
 {
 	renderer.logger().debug("Loading GLTF from file: {}", path);
 
-	auto data = fastgltf::GltfDataBuffer::FromPath(path);
+	auto data { fastgltf::GltfDataBuffer::FromPath(path) };
 	if (data.error() != fastgltf::Error::None) {
 		renderer.logger().err("Failed to open glTF file: {} (error {})", path,
 		    fastgltf::to_underlying(data.error()));
@@ -98,7 +98,7 @@ auto Mesh::load_gltf_meshes(
 			new_surface.count = static_cast<uint32_t>(
 			    gltf.accessors[p.indicesAccessor.value()].count);
 
-			size_t initial_vertex = vertices.size();
+			size_t initial_vertex { vertices.size() };
 
 			{ // Indices
 				auto &accessor = gltf.accessors[p.indicesAccessor.value()];
@@ -128,7 +128,7 @@ auto Mesh::load_gltf_meshes(
 
 			if (auto attr = p.findAttribute("NORMAL")) { // Normals
 				auto &accessor = gltf.accessors[attr->accessorIndex];
-				size_t local_index = 0;
+				size_t local_index { 0 };
 				for (auto normal :
 				    fastgltf::iterateAccessor<smath::Vec3>(gltf, accessor)) {
 					vertices[initial_vertex + local_index].normal = normal;
@@ -138,7 +138,7 @@ auto Mesh::load_gltf_meshes(
 
 			if (auto attr = p.findAttribute("TEXCOORD_0")) { // UVs
 				auto &accessor = gltf.accessors[attr->accessorIndex];
-				size_t local_index = 0;
+				size_t local_index { 0 };
 				for (auto uv :
 				    fastgltf::iterateAccessor<smath::Vec2>(gltf, accessor)) {
 					uv.unpack(vertices[initial_vertex + local_index].u,
@@ -149,7 +149,7 @@ auto Mesh::load_gltf_meshes(
 
 			if (auto attr = p.findAttribute("COLOR_0")) { // Colors
 				auto &accessor = gltf.accessors[attr->accessorIndex];
-				size_t local_index = 0;
+				size_t local_index { 0 };
 
 				switch (accessor.type) {
 				case fastgltf::AccessorType::Vec3: {

src/Logger.cpp

@@ -34,7 +34,7 @@
 static std::filesystem::path get_log_path(std::string_view app_name)
 {
 #ifdef _WIN32
-	PWSTR path = nullptr;
+	PWSTR path { nullptr };
 	SHGetKnownFolderPath(FOLDERID_LocalAppData, 0, nullptr, &path);
 	std::wstring wpath(path);
 	CoTaskMemFree(path);
@@ -70,7 +70,7 @@ static int compress_file(std::filesystem::path const &input_path,
 	std::vector<char> buffer(chunk_size);
 	while (in) {
 		in.read(buffer.data(), static_cast<std::streamsize>(buffer.size()));
-		std::streamsize bytes = in.gcount();
+		std::streamsize bytes { in.gcount() };
 		if (bytes > 0)
 			gzwrite(out, buffer.data(), static_cast<unsigned int>(bytes));
 	}
@@ -99,20 +99,20 @@ Logger::Logger(std::string_view app_name)
 		if (!file.is_regular_file())
 			continue;
 
-		auto name = file.path().filename().stem().string();
+		auto name { file.path().filename().stem().string() };
 		constexpr std::string_view prefix = "log_";
 
 		if (name.rfind(prefix, 0) != 0) {
 			continue;
 		}
 
-		int v = std::stoi(name.substr(prefix.size()));
+		int v { std::stoi(name.substr(prefix.size())) };
 		if (v > max)
 			max = v;
 
-		auto ext = file.path().filename().extension().string();
+		auto ext { file.path().filename().extension().string() };
 		if (ext == ".txt") {
-			auto np = file.path();
+			auto np { file.path() };
 			np.replace_extension(ext + ".gz");
 			compress_file(file.path(), np);
 		}
@@ -153,7 +153,7 @@ static std::string get_current_time_string()
 
 void Logger::log(Level level, std::string_view msg)
 {
-	auto time_str = get_current_time_string();
+	auto time_str { get_current_time_string() };
 	std::string level_str;
 	switch (level) {
 	case Logger::Level::Debug:

src/Pipeline.cpp

@@ -27,13 +27,13 @@ auto Pipeline::Builder::set_push_constant_ranges(
 auto Pipeline::Builder::build_compute(
     vk::PipelineShaderStageCreateInfo const &stage) -> Pipeline
 {
-	auto pipeline_layout = build_layout();
+	auto pipeline_layout { build_layout() };
 
 	vk::ComputePipelineCreateInfo pipeline_ci {};
 	pipeline_ci.layout = pipeline_layout.get();
 	pipeline_ci.stage = stage;
 
-	auto pipeline_ret = m_device.createComputePipelineUnique({}, pipeline_ci);
+	auto pipeline_ret { m_device.createComputePipelineUnique({}, pipeline_ci) };
 	VK_CHECK(m_logger, pipeline_ret.result);
 
 	return Pipeline {
@@ -46,14 +46,14 @@ auto Pipeline::Builder::build_graphics(
     std::function<GraphicsPipelineBuilder &(GraphicsPipelineBuilder &)> const
         &configure) -> Pipeline
 {
-	auto pipeline_layout = build_layout();
+	auto pipeline_layout { build_layout() };
 
-	auto builder = GraphicsPipelineBuilder { m_logger };
+	auto builder { GraphicsPipelineBuilder { m_logger } };
 	builder.set_pipeline_layout(
 	    static_cast<VkPipelineLayout>(pipeline_layout.get()));
 	configure(builder);
 
-	auto pipeline_handle = builder.build(static_cast<VkDevice>(m_device));
+	auto pipeline_handle { builder.build(static_cast<VkDevice>(m_device)) };
 	vk::UniquePipeline pipeline_unique(pipeline_handle,
 	    vk::detail::ObjectDestroy<vk::Device,
 	        VULKAN_HPP_DEFAULT_DISPATCHER_TYPE>(m_device));

src/Skybox.cpp

@@ -0,0 +1,314 @@
+#include "Skybox.h"
+
+#include <array>
+#include <cstddef>
+#include <span>
+#include <vector>
+
+#include "CPUTexture.h"
+#include "DescriptorWriter.h"
+#include "GraphicsPipelineBuilder.h"
+#include "Util.h"
+
+namespace Lunar {
+
+namespace {
+
+struct SkyboxPushConstants {
+	smath::Mat4 mvp;
+};
+
+struct FaceOffset {
+	uint32_t x;
+	uint32_t y;
+};
+
+constexpr std::array<FaceOffset, 6> CROSS_OFFSETS {
+	FaceOffset { 2, 1 }, // +X
+	FaceOffset { 0, 1 }, // -X
+	FaceOffset { 1, 0 }, // +Y
+	FaceOffset { 1, 2 }, // -Y
+	FaceOffset { 1, 1 }, // +Z
+	FaceOffset { 3, 1 }, // -Z
+};
+
+} // namespace
+
+auto Skybox::rebuild_pipeline(VulkanRenderer &renderer) -> bool
+{
+	Pipeline::Builder pipeline_builder { renderer.device(), renderer.logger() };
+
+	uint8_t skybox_vert_shader_data[] {
+#embed "skybox_vert.spv"
+	};
+	auto skybox_vert_shader
+	    = vkutil::load_shader_module(std::span<uint8_t>(skybox_vert_shader_data,
+	                                     sizeof(skybox_vert_shader_data)),
+	        renderer.device());
+	if (!skybox_vert_shader) {
+		renderer.logger().err("Failed to load skybox vert shader");
+		return false;
+	}
+
+	uint8_t skybox_frag_shader_data[] {
+#embed "skybox_frag.spv"
+	};
+	auto skybox_frag_shader
+	    = vkutil::load_shader_module(std::span<uint8_t>(skybox_frag_shader_data,
+	                                     sizeof(skybox_frag_shader_data)),
+	        renderer.device());
+	if (!skybox_frag_shader) {
+		renderer.logger().err("Failed to load skybox frag shader");
+		return false;
+	}
+
+	vk::PushConstantRange push_constant_range {};
+	push_constant_range.stageFlags = vk::ShaderStageFlagBits::eVertex;
+	push_constant_range.offset = 0;
+	push_constant_range.size = sizeof(SkyboxPushConstants);
+
+	std::array push_constant_ranges { push_constant_range };
+	pipeline_builder.set_push_constant_ranges(push_constant_ranges);
+	std::array descriptor_set_layouts {
+		renderer.single_image_descriptor_layout()
+	};
+	pipeline_builder.set_descriptor_set_layouts(descriptor_set_layouts);
+
+	VkVertexInputBindingDescription binding {};
+	binding.binding = 0;
+	binding.stride = sizeof(Vertex);
+	binding.inputRate = VK_VERTEX_INPUT_RATE_VERTEX;
+
+	VkVertexInputAttributeDescription attribute {};
+	attribute.location = 0;
+	attribute.binding = 0;
+	attribute.format = VK_FORMAT_R32G32B32_SFLOAT;
+	attribute.offset = offsetof(Vertex, position);
+
+	std::array bindings { binding };
+	std::array attributes { attribute };
+
+	m_pipeline = pipeline_builder.build_graphics(
+	    [&](GraphicsPipelineBuilder &builder) -> GraphicsPipelineBuilder & {
+		    builder.set_vertex_input(bindings, attributes);
+		    return builder
+		        .set_shaders(skybox_vert_shader.get(), skybox_frag_shader.get())
+		        .set_input_topology(VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST)
+		        .set_polygon_mode(VK_POLYGON_MODE_FILL)
+		        .set_cull_mode(
+		            VK_CULL_MODE_FRONT_BIT, VK_FRONT_FACE_COUNTER_CLOCKWISE)
+		        .set_multisampling(
+		            static_cast<VkSampleCountFlagBits>(renderer.msaa_samples()))
+		        .disable_blending()
+		        .enable_depth_testing(false, VK_COMPARE_OP_LESS_OR_EQUAL)
+		        .set_color_attachment_format(
+		            static_cast<VkFormat>(renderer.draw_image_format()))
+		        .set_depth_format(
+		            static_cast<VkFormat>(renderer.depth_image_format()));
+	    });
+
+	m_pipeline_samples = renderer.msaa_samples();
+	return true;
+}
+
+auto Skybox::init(VulkanRenderer &renderer, std::filesystem::path const &path)
+    -> void
+{
+	if (ok) {
+		destroy(renderer);
+		ok = false;
+	}
+
+	CPUTexture texture { path };
+
+	if (texture.width == 0 || texture.height == 0) {
+		renderer.logger().err("Skybox texture is empty: {}", path.string());
+		ok = false;
+		return;
+	}
+
+	if (texture.width % 4 != 0 || texture.height % 3 != 0
+	    || texture.width / 4 != texture.height / 3) {
+		renderer.logger().err(
+		    "Skybox texture must be 4x3 faces: {}", path.string());
+		ok = false;
+		return;
+	}
+
+	uint32_t const face_size = texture.width / 4;
+	size_t const face_bytes = static_cast<size_t>(face_size) * face_size * 4;
+
+	std::vector<uint8_t> cubemap_pixels(face_bytes * CROSS_OFFSETS.size());
+
+	for (size_t face = 0; face < CROSS_OFFSETS.size(); ++face) {
+		auto const offset = CROSS_OFFSETS[face];
+		for (uint32_t y = 0; y < face_size; ++y) {
+			for (uint32_t x = 0; x < face_size; ++x) {
+				uint32_t const src_x = offset.x * face_size + x;
+				uint32_t const src_y = offset.y * face_size + y;
+				size_t const src_index
+				    = (static_cast<size_t>(src_y) * texture.width + src_x) * 4;
+				size_t const dst_index = face * face_bytes
+				    + (static_cast<size_t>(y) * face_size + x) * 4;
+				std::copy_n(texture.pixels.data() + src_index, 4,
+				    cubemap_pixels.data() + dst_index);
+			}
+		}
+	}
+
+	m_cubemap = renderer.create_cubemap(cubemap_pixels, face_size,
+	    texture.format, vk::ImageUsageFlagBits::eSampled);
+	if (!m_cubemap.image) {
+		renderer.logger().err("Failed to create cubemap image");
+		return;
+	}
+
+	vk::SamplerCreateInfo sampler_ci {};
+	sampler_ci.magFilter = vk::Filter::eLinear;
+	sampler_ci.minFilter = vk::Filter::eLinear;
+	sampler_ci.mipmapMode = vk::SamplerMipmapMode::eLinear;
+	sampler_ci.addressModeU = vk::SamplerAddressMode::eClampToEdge;
+	sampler_ci.addressModeV = vk::SamplerAddressMode::eClampToEdge;
+	sampler_ci.addressModeW = vk::SamplerAddressMode::eClampToEdge;
+	m_sampler = renderer.device().createSamplerUnique(sampler_ci);
+
+	vk::DescriptorPoolSize pool_size {};
+	pool_size.type = vk::DescriptorType::eCombinedImageSampler;
+	pool_size.descriptorCount = 1;
+
+	vk::DescriptorPoolCreateInfo pool_ci {};
+	pool_ci.maxSets = 1;
+	pool_ci.poolSizeCount = 1;
+	pool_ci.pPoolSizes = &pool_size;
+	m_descriptor_pool = renderer.device().createDescriptorPoolUnique(pool_ci);
+
+	vk::DescriptorSetAllocateInfo alloc_info {};
+	alloc_info.descriptorPool = m_descriptor_pool.get();
+	alloc_info.descriptorSetCount = 1;
+	vk::DescriptorSetLayout layout {
+		renderer.single_image_descriptor_layout()
+	};
+	alloc_info.pSetLayouts = &layout;
+	m_descriptor_set
+	    = renderer.device().allocateDescriptorSets(alloc_info).front();
+
+	DescriptorWriter()
+	    .write_image(0, m_cubemap.image_view, m_sampler.get(),
+	        static_cast<VkImageLayout>(vk::ImageLayout::eShaderReadOnlyOptimal),
+	        VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER)
+	    .update_set(renderer.device(), m_descriptor_set);
+
+	std::vector<Vertex> vertices;
+	vertices.reserve(8);
+
+	auto push_vertex = [&](smath::Vec3 const &pos) {
+		Vertex v {};
+		v.position = pos;
+		v.u = 0.0f;
+		v.v = 0.0f;
+		v.normal = smath::Vec3 { 0.0f, 0.0f, 1.0f };
+		v.color = smath::Vec4 { 1.0f, 1.0f, 1.0f, 1.0f };
+		vertices.emplace_back(v);
+	};
+
+	push_vertex(smath::Vec3 { -1.0f, -1.0f, -1.0f });
+	push_vertex(smath::Vec3 { 1.0f, -1.0f, -1.0f });
+	push_vertex(smath::Vec3 { 1.0f, 1.0f, -1.0f });
+	push_vertex(smath::Vec3 { -1.0f, 1.0f, -1.0f });
+	push_vertex(smath::Vec3 { -1.0f, -1.0f, 1.0f });
+	push_vertex(smath::Vec3 { 1.0f, -1.0f, 1.0f });
+	push_vertex(smath::Vec3 { 1.0f, 1.0f, 1.0f });
+	push_vertex(smath::Vec3 { -1.0f, 1.0f, 1.0f });
+
+	std::vector<uint32_t> indices {
+		4,
+		5,
+		6,
+		4,
+		6,
+		7, // +Z
+		1,
+		0,
+		3,
+		1,
+		3,
+		2, // -Z
+		5,
+		1,
+		2,
+		5,
+		2,
+		6, // +X
+		0,
+		4,
+		7,
+		0,
+		7,
+		3, // -X
+		7,
+		6,
+		2,
+		7,
+		2,
+		3, // +Y
+		0,
+		1,
+		5,
+		0,
+		5,
+		4, // -Y
+	};
+
+	m_index_count = static_cast<uint32_t>(indices.size());
+	m_cube_mesh = renderer.upload_mesh(indices, vertices);
+
+	if (!rebuild_pipeline(renderer)) {
+		ok = false;
+		return;
+	}
+
+	ok = true;
+}
+
+auto Skybox::destroy(VulkanRenderer &renderer) -> void
+{
+	if (m_cube_mesh.index_buffer.buffer) {
+		renderer.destroy_buffer(m_cube_mesh.index_buffer);
+	}
+	if (m_cube_mesh.vertex_buffer.buffer) {
+		renderer.destroy_buffer(m_cube_mesh.vertex_buffer);
+	}
+	if (m_cubemap.image) {
+		renderer.destroy_image(m_cubemap);
+	}
+	m_sampler.reset();
+	m_descriptor_pool.reset();
+	m_pipeline.reset();
+	m_pipeline_samples = vk::SampleCountFlagBits::e1;
+	m_descriptor_set = vk::DescriptorSet {};
+	m_cube_mesh = {};
+	m_cubemap = {};
+	m_index_count = 0;
+	ok = false;
+}
+
+auto Skybox::draw(VulkanRenderer::GL &gl, VulkanRenderer &renderer,
+    smath::Mat4 const &mvp) -> void
+{
+	if (!ok) {
+		return;
+	}
+
+	if (m_pipeline_samples != renderer.msaa_samples()) {
+		if (!rebuild_pipeline(renderer)) {
+			return;
+		}
+	}
+
+	SkyboxPushConstants push_constants { mvp };
+	auto bytes { std::as_bytes(std::span { &push_constants, 1 }) };
+	gl.draw_indexed(m_pipeline, m_descriptor_set, m_cube_mesh.vertex_buffer,
+	    m_cube_mesh.index_buffer, m_index_count, bytes);
+}
+
+} // namespace Lunar

src/Skybox.h

@@ -0,0 +1,36 @@
+#pragma once
+
+#include <cstdint>
+#include <filesystem>
+
+#include <vulkan/vulkan.hpp>
+
+#include "Pipeline.h"
+#include "Types.h"
+#include "VulkanRenderer.h"
+
+namespace Lunar {
+
+struct Skybox {
+	bool ok { false };
+
+	auto init(VulkanRenderer &renderer, std::filesystem::path const &path)
+	    -> void;
+	auto destroy(VulkanRenderer &renderer) -> void;
+	auto draw(VulkanRenderer::GL &gl, VulkanRenderer &renderer,
+	    smath::Mat4 const &mvp) -> void;
+
+private:
+	auto rebuild_pipeline(VulkanRenderer &renderer) -> bool;
+
+	Pipeline m_pipeline {};
+	GPUMeshBuffers m_cube_mesh {};
+	AllocatedImage m_cubemap {};
+	vk::UniqueSampler m_sampler {};
+	vk::UniqueDescriptorPool m_descriptor_pool {};
+	vk::DescriptorSet m_descriptor_set {};
+	vk::SampleCountFlagBits m_pipeline_samples { vk::SampleCountFlagBits::e1 };
+	uint32_t m_index_count { 0 };
+};
+
+} // namespace Lunar

src/Types.h

@@ -1,5 +1,9 @@
 #pragma once
 
+#include <cmath>
+#include <cstdint>
+#include <vector>
+
 #include <smath.hpp>
 #include <vk_mem_alloc.h>
 #include <vulkan/vulkan.hpp>
@@ -31,6 +35,15 @@ struct FrameData {
 
 	DeletionQueue deletion_queue;
 	DescriptorAllocatorGrowable frame_descriptors;
+	AllocatedBuffer frame_image_buffer {};
+	vk::Extent2D frame_image_extent {};
+	std::vector<std::uint8_t> frame_image_rgba;
+	bool frame_image_ready { false };
+	bool tracy_frame_ready { false };
+	AllocatedBuffer screenshot_buffer {};
+	vk::Extent2D screenshot_extent {};
+	std::vector<std::uint8_t> screenshot_rgba;
+	bool screenshot_ready { false };
 };
 
 struct Vertex {
@@ -55,4 +68,39 @@ struct GPUSceneData {
 	smath::Vec4 sunlight_color;
 };
 
+struct Camera {
+	smath::Vec3 position {};
+	smath::Vec3 target { 0, 0, -1 };
+	smath::Vec3 up { 0, 1, 0 };
+	float fovy { smath::deg(70.0f) };
+};
+
+struct PolarCoordinate {
+	float r, theta, phi;
+
+	static PolarCoordinate from_vec3(smath::Vec3 const &v)
+	{
+		PolarCoordinate p;
+		p.r = std::sqrt(v.x() * v.x() + v.y() * v.y() + v.z() * v.z());
+
+		if (p.r == 0.0f) {
+			p.theta = 0.0f;
+			p.phi = 0.0f;
+			return p;
+		}
+
+		p.theta = std::atan2(v.z(), v.x());
+		p.phi = std::acos(v.y() / p.r);
+		return p;
+	}
+
+	smath::Vec3 to_vec3() const
+	{
+		float sin_phi { std::sin(phi) };
+
+		return smath::Vec3 { r * sin_phi * std::cos(theta), r * std::cos(phi),
+			r * sin_phi * std::sin(theta) };
+	}
+};
+
 } // namespace Lunar

src/Util.h

@@ -2,7 +2,6 @@
 
 #include <span>
 
-#include <vulkan/vk_enum_string_helper.h>
 #include <vulkan/vulkan.hpp>
 
 template<typename F> struct privDefer {
@@ -30,13 +29,21 @@ template<typename F> privDefer<F> defer_func(F f) { return privDefer<F>(f); }
 #define VK_CHECK(logger, x) \
 	do { \
 		auto err { x }; \
-		auto result = vk::Result(err); \
+		auto result { vk::Result(err) }; \
 		if (result != vk::Result::eSuccess) { \
 			(logger).err("Detected Vulkan error: {}", vk::to_string(result)); \
 			throw std::runtime_error("Vulkan error"); \
 		} \
 	} while (0)
 
+#if defined(TRACY_ENABLE)
+#	define GZoneScopedN(name) ZoneScopedN(name)
+#else
+#	define GZoneScopedN(name) \
+		do { \
+		} while (0)
+#endif
+
 namespace vkutil {
 
 auto transition_image(vk::CommandBuffer cmd, vk::Image image,

src/VulkanRenderer.h

@@ -1,8 +1,14 @@
 #pragma once
 
 #include <array>
+#include <cstddef>
+#include <cstdint>
 #include <functional>
+#include <mutex>
 #include <optional>
+#include <span>
+#include <string>
+#include <variant>
 #include <vector>
 
 #include <SDL3/SDL_video.h>
@@ -11,6 +17,7 @@
 #include <vk_mem_alloc.h>
 #include <vulkan/vulkan.hpp>
 
+#include "CPUTexture.h"
 #include "Colors.h"
 #include "DeletionQueue.h"
 #include "Loader.h"
@@ -28,14 +35,13 @@ struct GPUDrawPushConstants {
 constexpr unsigned FRAME_OVERLAP = 2;
 
 struct VulkanRenderer {
-	enum class AntiAliasingKind {
+	struct ScreenshotPixels {
-		NONE,
+		std::span<std::uint8_t const> pixels;
-		MSAA_2X,
+		vk::Extent2D extent;
-		MSAA_4X,
-		MSAA_8X,
 	};
 
 	struct GL {
+
 		enum class GeometryKind {
 			Triangles,
 			TriangleStrip,
@@ -71,17 +77,27 @@ struct VulkanRenderer {
 		auto normal(smath::Vec3 const &normal) -> void;
 		auto set_texture(std::optional<AllocatedImage const *> texture
 		    = std::nullopt) -> void;
+		auto set_culling(bool enabled) -> void;
 		auto draw_rectangle(smath::Vec2 pos, smath::Vec2 size,
 		    smath::Vec4 color = smath::Vec4 { Colors::WHITE, 1.0f },
 		    float rotation = 0.0f) -> void;
+		auto draw_sphere(smath::Vec3 center, float radius, int rings = 16,
+		    int segments = 32, std::optional<smath::Vec4> sphere_color = {})
+		    -> void;
 		auto end() -> void;
 		auto flush() -> void;
 
 		auto use_pipeline(Pipeline &pipeline) -> void;
 		auto set_transform(smath::Mat4 const &transform) -> void;
+		auto push_transform() -> void;
+		auto pop_transform() -> void;
 		auto draw_mesh(GPUMeshBuffers const &mesh, smath::Mat4 const &transform,
 		    uint32_t index_count, uint32_t first_index = 0,
 		    int32_t vertex_offset = 0) -> void;
+		auto draw_indexed(Pipeline &pipeline, vk::DescriptorSet descriptor_set,
+		    AllocatedBuffer const &vertex_buffer,
+		    AllocatedBuffer const &index_buffer, uint32_t index_count,
+		    std::span<std::byte const> push_constants) -> void;
 
 	private:
 		auto push_vertex(smath::Vec3 const &pos) -> void;
@@ -96,7 +112,9 @@ struct VulkanRenderer {
 		bool m_inside_primitive { false };
 		bool m_drawing { false };
 		Pipeline *m_active_pipeline { nullptr };
+		bool m_culling_enabled { true };
 		smath::Mat4 m_transform { smath::Mat4::identity() };
+		std::vector<smath::Mat4> m_transform_stack;
 		smath::Vec4 m_current_color { 1.0f, 1.0f, 1.0f, 1.0f };
 		smath::Vec3 m_current_normal { 0.0f, 0.0f, 1.0f };
 		smath::Vec2 m_current_uv { 0.0f, 0.0f };
@@ -106,12 +124,30 @@ struct VulkanRenderer {
 		std::vector<uint32_t> m_indices;
 	};
 
-	VulkanRenderer(SDL_Window *window, Logger &logger);
+	enum class AntiAliasingKind {
+		NONE,
+		MSAA_2X,
+		MSAA_4X,
+		MSAA_8X,
+	};
+
+	struct KmsSurfaceConfig { };
+
+	VulkanRenderer(SDL_Window *window, Logger &logger,
+	    std::span<std::string const> instance_extensions = {},
+	    std::span<std::string const> device_extensions = {});
+	VulkanRenderer(KmsSurfaceConfig config, Logger &logger,
+	    std::span<std::string const> instance_extensions = {},
+	    std::span<std::string const> device_extensions = {});
 	~VulkanRenderer();
 
 	auto render(std::function<void(GL &)> const &record = {}) -> void;
+	auto render_to_image(vk::Image target_image, vk::Extent2D target_extent,
+	    std::function<void(GL &)> const &record = {}) -> void;
 	auto resize(uint32_t width, uint32_t height) -> void;
+	auto set_offscreen_extent(vk::Extent2D extent) -> void;
 	auto set_antialiasing(AntiAliasingKind kind) -> void;
+	auto set_antialiasing_immediate(AntiAliasingKind kind) -> void;
 	auto antialiasing() const -> AntiAliasingKind
 	{
 		return m_vk.antialiasing_kind;
@@ -122,14 +158,17 @@ struct VulkanRenderer {
 	    bool clear_frame_descriptors = true) -> void;
 	auto upload_mesh(std::span<uint32_t> indices, std::span<Vertex> vertices)
 	    -> GPUMeshBuffers;
+	auto destroy_buffer(AllocatedBuffer const &buffer) -> void;
+	auto create_image(CPUTexture const &texture, vk::ImageUsageFlags flags,
+	    bool mipmapped = false) -> AllocatedImage;
+	auto create_cubemap(std::span<uint8_t const> pixels, uint32_t face_size,
+	    vk::Format format, vk::ImageUsageFlags flags) -> AllocatedImage;
+	auto destroy_image(AllocatedImage const &img) -> void;
+	auto destroy_image_later(AllocatedImage img) -> void;
 	auto rectangle_mesh() const -> GPUMeshBuffers const &
 	{
 		return m_vk.rectangle;
 	}
-	auto test_meshes() const -> std::vector<std::shared_ptr<Mesh>> const &
-	{
-		return m_vk.test_meshes;
-	}
 	auto white_texture() const -> AllocatedImage const &
 	{
 		return m_vk.white_image;
@@ -148,16 +187,75 @@ struct VulkanRenderer {
 	}
 	auto draw_extent() const -> vk::Extent2D { return m_vk.draw_extent; }
 	auto mesh_pipeline() -> Pipeline & { return m_vk.mesh_pipeline; }
+	auto wayland_pipeline() -> Pipeline & { return m_vk.wayland_pipeline; }
 	auto triangle_pipeline() -> Pipeline & { return m_vk.triangle_pipeline; }
+	auto device() const -> vk::Device { return m_device; }
+	auto instance() const -> vk::Instance { return m_instance; }
+	auto physical_device() const -> vk::PhysicalDevice
+	{
+		return m_physical_device;
+	}
+	auto graphics_queue() const -> vk::Queue { return m_vk.graphics_queue; }
+	auto graphics_queue_family() const -> uint32_t
+	{
+		return m_vk.graphics_queue_family;
+	}
+	auto draw_image_format() const -> vk::Format
+	{
+		return m_vk.draw_image.format;
+	}
+	auto depth_image_format() const -> vk::Format
+	{
+		return m_vk.depth_image.format;
+	}
+	auto msaa_samples() const -> vk::SampleCountFlagBits
+	{
+		return m_vk.msaa_samples;
+	}
+	auto single_image_descriptor_layout() const -> vk::DescriptorSetLayout
+	{
+		return m_vk.single_image_descriptor_layout;
+	}
 	auto gl_api() -> GL & { return gl; }
+	auto get_screenshot() const -> std::optional<AllocatedImage>
+	{
+		return m_latest_screenshot;
+	}
+	auto get_screenshot_pixels() const
+	    -> std::optional<VulkanRenderer::ScreenshotPixels>
+	{
+		if (m_latest_screenshot_pixels.empty()
+		    || m_latest_screenshot_extent.width == 0
+		    || m_latest_screenshot_extent.height == 0) {
+			return {};
+		}
+
+		auto const span { std::span<std::uint8_t const> {
+			m_latest_screenshot_pixels.data(),
+			m_latest_screenshot_pixels.size() } };
+		return ScreenshotPixels { span, m_latest_screenshot_extent };
+	}
 
 	auto logger() const -> Logger & { return m_logger; }
 
 	GL gl;
 
+	std::optional<AllocatedImage> m_latest_screenshot {};
+	std::vector<std::uint8_t> m_latest_screenshot_pixels {};
+	vk::Extent2D m_latest_screenshot_extent {};
+
 private:
+	struct RenderCommand {
+		struct SetAntiAliasing {
+			AntiAliasingKind kind;
+		};
+
+		std::variant<SetAntiAliasing> payload;
+	};
+
 	auto vk_init() -> void;
 	auto swapchain_init() -> void;
+	auto setup_kms_surface() -> void;
 	auto commands_init() -> void;
 	auto sync_init() -> void;
 	auto descriptors_init() -> void;
@@ -179,17 +277,30 @@ private:
 	auto destroy_msaa_color_image() -> void;
 	auto recreate_swapchain(uint32_t width, uint32_t height) -> void;
 	auto destroy_swapchain() -> void;
+	auto ensure_screenshot_buffers(vk::Extent2D extent) -> void;
+	auto destroy_screenshot_buffers() -> void;
+	auto emit_frame_screenshot(FrameData &frame) -> void;
+#if defined(TRACY_ENABLE)
+	auto ensure_tracy_frame_buffers(vk::Extent2D extent) -> void;
+	auto destroy_tracy_frame_buffers() -> void;
+	auto emit_tracy_frame_image(FrameData &frame) -> void;
+#endif
 	auto create_image(vk::Extent3D size, vk::Format format,
 	    vk::ImageUsageFlags flags,
 	    vk::SampleCountFlagBits samples = vk::SampleCountFlagBits::e1,
 	    bool mipmapped = false) -> AllocatedImage;
+	auto create_image_no_view(vk::Extent3D size, vk::Format format,
+	    vk::ImageUsageFlags flags,
+	    vk::SampleCountFlagBits samples = vk::SampleCountFlagBits::e1,
+	    bool mipmapped = false) -> AllocatedImage;
 	auto create_image(void const *data, vk::Extent3D size, vk::Format format,
 	    vk::ImageUsageFlags flags, bool mipmapped = false) -> AllocatedImage;
-	auto destroy_image(AllocatedImage const &img) -> void;
 
 	auto create_buffer(size_t alloc_size, vk::BufferUsageFlags usage,
 	    VmaMemoryUsage memory_usage) -> AllocatedBuffer;
-	auto destroy_buffer(AllocatedBuffer const &buffer) -> void;
+	auto enqueue_render_command(RenderCommand &&command) -> void;
+	auto process_render_commands() -> void;
+	auto apply_antialiasing(AntiAliasingKind kind) -> void;
 
 	vk::Instance m_instance {};
 	vk::PhysicalDevice m_physical_device {};
@@ -227,6 +338,13 @@ private:
 		vk::ImageLayout msaa_color_image_layout { vk::ImageLayout::eUndefined };
 		AllocatedImage depth_image {};
 		vk::ImageLayout depth_image_layout { vk::ImageLayout::eUndefined };
+#if defined(TRACY_ENABLE)
+		AllocatedImage tracy_capture_image {};
+		vk::ImageLayout tracy_capture_image_layout {
+			vk::ImageLayout::eUndefined
+		};
+		vk::Extent2D tracy_capture_extent {};
+#endif
 		vk::Extent2D draw_extent {};
 		AntiAliasingKind antialiasing_kind { AntiAliasingKind::NONE };
 		vk::SampleCountFlagBits msaa_samples { vk::SampleCountFlagBits::e1 };
@@ -240,7 +358,10 @@ private:
 		vk::DescriptorSetLayout single_image_descriptor_layout {};
 
 		Pipeline triangle_pipeline;
+		Pipeline triangle_pipeline_culled;
 		Pipeline mesh_pipeline;
+		Pipeline mesh_pipeline_culled;
+		Pipeline wayland_pipeline;
 
 		GPUMeshBuffers rectangle;
 
@@ -254,8 +375,6 @@ private:
 
 		uint64_t frame_number { 0 };
 
-		std::vector<std::shared_ptr<Mesh>> test_meshes;
-
 		AllocatedImage white_image {};
 		AllocatedImage black_image {};
 		AllocatedImage gray_image {};
@@ -265,8 +384,27 @@ private:
 		vk::UniqueSampler default_sampler_nearest;
 	} m_vk;
 
+	struct KmsState {
+		vk::DisplayKHR display {};
+		vk::DisplayModeKHR mode {};
+		vk::Extent2D extent {};
+		uint32_t plane_index { 0 };
+		uint32_t plane_stack_index { 0 };
+		std::string display_name {};
+	};
+
 	SDL_Window *m_window { nullptr };
 	Logger &m_logger;
+	std::mutex m_command_mutex;
+	std::vector<RenderCommand> m_pending_render_commands;
+	bool m_use_kms { false };
+	bool m_imgui_enabled { true };
+	std::optional<KmsState> m_kms_state {};
+	vk::PhysicalDevice m_kms_physical_device {};
+	vk::Extent2D m_kms_extent {};
+	bool m_kms_physical_device_set { false };
+	std::vector<std::string> m_extra_instance_extensions {};
+	std::vector<std::string> m_extra_device_extensions {};
 };
 
 } // namespace Lunar

src/main.cpp

@@ -1,7 +1,3 @@
 #include "src/Application.h"
 
-auto main() -> int
+auto main() -> int { Lunar::Application::the().run(); }
-{
-	Lunar::Application app {};
-	app.run();
-}

src/wayland/Client.h

@@ -0,0 +1,113 @@
+#pragma once
+
+#include <cassert>
+#include <format>
+#include <functional>
+#include <optional>
+#include <tuple>
+
+#include <wayland-server-core.h>
+
+#include "Display.h"
+#include "List.h"
+
+namespace Lunar::Wayland {
+
+struct Client {
+	Client(wl_client *client)
+	    : m_client { std::move(client) }
+	{
+		assert(m_client);
+	}
+	~Client() = default;
+
+	inline auto c_ptr() const -> wl_client * { return m_client; }
+
+	static auto from_link(wl_list *link) -> Client
+	{
+		return Client { wl_client_from_link(link) };
+	}
+
+	auto flush() { wl_client_flush(m_client); }
+
+	auto get_display() -> Display
+	{
+		return Display { wl_client_get_display(m_client) };
+	}
+
+	auto get_credentials() noexcept -> std::tuple<pid_t, uid_t, gid_t>
+	{
+		std::tuple<pid_t, uid_t, gid_t> ret {};
+		auto &[pid, uid, gid] { ret };
+		wl_client_get_credentials(m_client, &pid, &uid, &gid);
+		return ret;
+	}
+
+	auto get_fd() noexcept -> int { return wl_client_get_fd(m_client); }
+
+	auto get_object(uint32_t id) -> std::optional<wl_resource *>
+	{
+		if (auto *res { wl_client_get_object(m_client, id) }; res != NULL) {
+			return res;
+		} else {
+			return {};
+		}
+	}
+
+	auto post_implementation_error(std::string_view string)
+	{
+		wl_client_post_implementation_error(
+		    m_client, "%.*s", static_cast<int>(string.size()), string.data());
+	}
+
+	template<typename... Args>
+	auto post_implementation_error(
+	    std::format_string<Args...> fmt, Args &&...args)
+	{
+		post_implementation_error(
+		    std::format(fmt, std::forward<Args>(args)...));
+	}
+
+	auto add_destroy_listener(wl_listener *listener)
+	{
+		wl_client_add_destroy_listener(m_client, listener);
+	}
+
+	auto add_destroy_late_listener(wl_listener *listener)
+	{
+		wl_client_add_destroy_late_listener(m_client, listener);
+	}
+
+	auto get_link() -> wl_list * { return wl_client_get_link(m_client); }
+
+	auto add_resource_created_listener(wl_listener *listener)
+	{
+		wl_client_add_resource_created_listener(m_client, listener);
+	}
+
+	auto for_each_resource(
+	    std::function<wl_iterator_result(wl_resource *)> const &fn) -> void
+	{
+		wl_client_for_each_resource(
+		    m_client,
+		    (wl_client_for_each_resource_iterator_func_t)[](
+		        wl_resource * res, void *user_data)
+		        ->wl_iterator_result {
+			        auto *f { static_cast<
+				        std::function<wl_iterator_result(wl_resource *)> *>(
+				        user_data) };
+			        return (*f)(res);
+		        },
+		    const_cast<void *>(static_cast<void const *>(&fn)));
+	}
+
+	auto set_max_buffer_size(size_t max_buffer_size)
+	{
+		wl_client_set_max_buffer_size(m_client, max_buffer_size);
+	}
+
+private:
+	wl_client *m_client {};
+};
+
+} // namespace Lunar::Wayland

src/wayland/Display.h

@@ -0,0 +1,103 @@
+#pragma once
+
+#include <cassert>
+#include <format>
+#include <stdexcept>
+#include <vector>
+
+#include <wayland-server-core.h>
+
+namespace Lunar::Wayland {
+
+struct Display {
+	Display()
+	    : m_display(wl_display_create())
+	{
+	}
+	Display(wl_display *display)
+	    : m_display { std::move(display) }
+	    , m_should_cleanup { false }
+	{
+	}
+	~Display()
+	{
+		if (!m_should_cleanup)
+			return;
+		wl_display_destroy_clients(m_display);
+		wl_display_destroy(m_display);
+	}
+
+	inline auto c_ptr() const -> wl_display * { return m_display; }
+
+	auto set_global_filter(
+	    wl_display_global_filter_func_t filter, void *data) noexcept
+	{
+		wl_display_set_global_filter(m_display, filter, data);
+	}
+
+	auto next_serial() noexcept -> uint32_t
+	{
+		return wl_display_next_serial(m_display);
+	}
+
+	auto set_default_max_buffer_size(size_t max_buffer_size) noexcept
+	{
+		wl_display_set_default_max_buffer_size(m_display, max_buffer_size);
+	}
+
+	auto add_socket_fd(int fd)
+	{
+		if (wl_display_add_socket_fd(m_display, fd) == -1) {
+			throw std::runtime_error(
+			    "Failed to add socket fd to Wayland display");
+		}
+	}
+
+	auto add_socket(char const *name)
+	{
+		if (wl_display_add_socket(m_display, name) == -1) {
+			throw std::runtime_error(std::format(
+			    "Failed to add socket `{}` to Wayland display", name));
+		}
+	}
+
+	auto add_protocol_logger(wl_protocol_logger_func_t func, void *user_data)
+	    -> wl_protocol_logger *
+	{
+		if (auto *logger {
+		        wl_display_add_protocol_logger(m_display, func, user_data) };
+		    logger != NULL) {
+			return logger;
+		} else {
+			throw std::runtime_error(
+			    "Failed to add protocol logger to Wayland display");
+		}
+	}
+
+	auto add_shm_format(uint32_t format) -> uint32_t *
+	{
+		if (auto *fmt { wl_display_add_shm_format(m_display, format) };
+		    fmt != NULL) {
+			return fmt;
+		} else {
+			throw std::runtime_error(
+			    "Failed to add SHM format to Wayland display");
+		}
+	}
+
+	auto get_client_list() -> std::vector<wl_client *>
+	{
+		std::vector<wl_client *> ret {};
+		auto const list { wl_display_get_client_list(m_display) };
+		assert(list);
+		wl_client *client {};
+		wl_client_for_each(client, list) { ret.push_back(client); }
+		return ret;
+	}
+
+private:
+	wl_display *m_display {};
+	bool m_should_cleanup { true };
+};
+
+} // namespace Lunar::Wayland

src/wayland/Global.h

@@ -0,0 +1,71 @@
+#pragma once
+
+#include <cassert>
+#include <optional>
+#include <utility>
+
+#include <wayland-server-core.h>
+
+#include "Client.h"
+#include "Display.h"
+
+namespace Lunar::Wayland {
+
+struct Global {
+	Global() = delete;
+
+	explicit Global(Display &display, wl_interface const *interface,
+	    int version, void *data, wl_global_bind_func_t bind)
+	    : m_global {
+		    wl_global_create(display.c_ptr(), interface, version, data, bind),
+	    }
+	{
+	}
+
+	Global(wl_global *global)
+	    : m_global { std::move(global) }
+	    , m_should_cleanup { false }
+	{
+		assert(m_global);
+	}
+
+	~Global()
+	{
+		if (!m_should_cleanup)
+			return;
+		wl_global_destroy(m_global);
+	}
+
+	inline auto c_ptr() const -> wl_global * { return m_global; }
+
+	auto get_name(Client &client) const -> std::optional<uint32_t>
+	{
+		if (auto const ret { wl_global_get_name(m_global, client.c_ptr()) };
+		    ret != 0) {
+			return ret;
+		} else {
+			return {};
+		}
+	}
+
+	inline auto get_version() const -> uint32_t
+	{
+		return wl_global_get_version(m_global);
+	}
+
+	inline auto get_display() const -> Display
+	{
+		return wl_global_get_display(m_global);
+	}
+
+	inline auto get_interface() const -> wl_interface const *
+	{
+		return wl_global_get_interface(m_global);
+	}
+
+private:
+	wl_global *m_global {};
+	bool m_should_cleanup { true };
+};
+
+} // namespace Lunar::Wayland

src/wayland/List.h

@@ -0,0 +1,287 @@
+#pragma once
+
+#include <cstddef>
+#include <iterator>
+
+#include <wayland-server-core.h>
+
+namespace Lunar::Wayland {
+
+namespace detail {
+
+template<typename T, wl_list T::*Member>
+constexpr std::ptrdiff_t member_offset() noexcept
+{
+	return reinterpret_cast<std::ptrdiff_t>(
+	    &(reinterpret_cast<T const volatile *>(0)->*Member));
+}
+
+template<typename T, wl_list T::*Member>
+inline T *container_of(wl_list *node) noexcept
+{
+	auto *p { reinterpret_cast<std::byte *>(node)
+		- member_offset<T, Member>() };
+	return reinterpret_cast<T *>(p);
+}
+
+template<typename T, wl_list T::*Member>
+inline T const *container_of(wl_list const *node) noexcept
+{
+	auto *p { reinterpret_cast<std::byte const *>(node)
+		- member_offset<T, Member>() };
+	return reinterpret_cast<T const *>(p);
+}
+
+} // namespace detail
+
+template<typename T, wl_list T::*Member> struct List {
+	struct Iterator {
+		using iterator_category = std::bidirectional_iterator_tag;
+		using value_type = T;
+		using difference_type = std::ptrdiff_t;
+		using pointer = T *;
+		using reference = T &;
+
+		Iterator() = default;
+		Iterator(wl_list *cur, wl_list *head)
+		    : m_cur(cur)
+		    , m_head(head)
+		{
+		}
+
+		auto operator*() const -> reference
+		{
+			return *detail::container_of<T, Member>(m_cur);
+		}
+		auto operator->() const -> pointer
+		{
+			return detail::container_of<T, Member>(m_cur);
+		}
+
+		auto operator++() -> Iterator &
+		{
+			m_cur = m_cur->next;
+			return *this;
+		}
+		auto operator++(int) -> Iterator
+		{
+			auto t { *this };
+			++(*this);
+			return t;
+		}
+
+		auto operator--() -> Iterator &
+		{
+			m_cur = (m_cur == m_head) ? m_head->prev : m_cur->prev;
+			return *this;
+		}
+		auto operator--(int) -> Iterator
+		{
+			auto t { *this };
+			--(*this);
+			return t;
+		}
+
+		friend auto operator==(Iterator a, Iterator b) -> bool
+		{
+			return a.m_cur == b.m_cur;
+		}
+		friend auto operator!=(Iterator a, Iterator b) -> bool
+		{
+			return !(a == b);
+		}
+
+	private:
+		wl_list *m_cur { nullptr };
+		wl_list *m_head { nullptr };
+	};
+
+	struct ConstIterator {
+		using iterator_category = std::bidirectional_iterator_tag;
+		using value_type = T const;
+		using difference_type = std::ptrdiff_t;
+		using pointer = T const *;
+		using reference = T const &;
+
+		ConstIterator() = default;
+		ConstIterator(wl_list const *cur, wl_list const *head)
+		    : m_cur(cur)
+		    , m_head(head)
+		{
+		}
+
+		auto operator*() const -> reference
+		{
+			return *detail::container_of<T, Member>(m_cur);
+		}
+		auto operator->() const -> pointer
+		{
+			return detail::container_of<T, Member>(m_cur);
+		}
+
+		auto operator++() -> ConstIterator &
+		{
+			m_cur = m_cur->next;
+			return *this;
+		}
+		auto operator++(int) -> ConstIterator
+		{
+			auto t { *this };
+			++(*this);
+			return t;
+		}
+
+		auto operator--() -> ConstIterator &
+		{
+			m_cur = (m_cur == m_head) ? m_head->prev : m_cur->prev;
+			return *this;
+		}
+		auto operator--(int) -> ConstIterator
+		{
+			auto t { *this };
+			--(*this);
+			return t;
+		}
+
+		friend auto operator==(ConstIterator a, ConstIterator b) -> bool
+		{
+			return a.m_cur == b.m_cur;
+		}
+		friend auto operator!=(ConstIterator a, ConstIterator b) -> bool
+		{
+			return !(a == b);
+		}
+
+	private:
+		wl_list const *m_cur { nullptr };
+		wl_list const *m_head { nullptr };
+	};
+
+	List()
+	    : m_head {}
+	    , m_external_head { nullptr }
+	    , m_should_cleanup { true }
+	{
+		wl_list_init(&m_head);
+	}
+
+	explicit List(
+	    wl_list *existing_head, bool should_cleanup = false, bool init = false)
+	    : m_head {}
+	    , m_external_head { existing_head }
+	    , m_should_cleanup { should_cleanup }
+	{
+		if (init && m_external_head)
+			wl_list_init(m_external_head);
+	}
+
+	~List()
+	{
+		if (!m_should_cleanup)
+			return;
+
+		clear();
+		if (auto *h { head_ptr() }; h)
+			wl_list_init(h);
+	}
+
+	List(List const &) = delete;
+	auto operator=(List const &) -> List & = delete;
+
+	List(List &&other) noexcept { move_from(other); }
+
+	auto operator=(List &&other) noexcept -> List &
+	{
+		if (this != &other) {
+			this->~List();
+			move_from(other);
+		}
+		return *this;
+	}
+
+	inline auto c_ptr() -> wl_list * { return head_ptr(); }
+	inline auto c_ptr() const -> wl_list const * { return head_ptr(); }
+
+	auto empty() const noexcept -> bool { return wl_list_empty(head_ptr()); }
+	auto length() const noexcept -> int { return wl_list_length(head_ptr()); }
+
+	auto push_front(T *elem) noexcept -> void
+	{
+		wl_list_insert(head_ptr(), &(elem->*Member));
+	}
+
+	auto push_back(T *elem) noexcept -> void
+	{
+		auto *h { head_ptr() };
+		wl_list_insert(h->prev, &(elem->*Member));
+	}
+
+	auto remove(T *elem) noexcept -> void
+	{
+		wl_list_remove(&(elem->*Member));
+		wl_list_init(&(elem->*Member));
+	}
+
+	auto clear() noexcept -> void
+	{
+		auto *h { head_ptr() };
+		while (!wl_list_empty(h)) {
+			auto *node { h->next };
+			wl_list_remove(node);
+			wl_list_init(node);
+		}
+	}
+
+	auto begin() noexcept -> Iterator
+	{
+		auto *h { head_ptr() };
+		return Iterator(h->next, h);
+	}
+
+	auto end() noexcept -> Iterator
+	{
+		auto *h { head_ptr() };
+		return Iterator(h, h);
+	}
+
+	auto begin() const noexcept -> ConstIterator
+	{
+		auto const *h { head_ptr() };
+		return ConstIterator(h->next, h);
+	}
+
+	auto end() const noexcept -> ConstIterator
+	{
+		auto const *h { head_ptr() };
+		return ConstIterator(h, h);
+	}
+
+private:
+	auto head_ptr() noexcept -> wl_list *
+	{
+		return m_external_head ? m_external_head : &m_head;
+	}
+
+	auto head_ptr() const noexcept -> wl_list const *
+	{
+		return m_external_head ? m_external_head : &m_head;
+	}
+
+	auto move_from(List &other) noexcept -> void
+	{
+		m_head = other.m_head;
+		m_external_head = other.m_external_head;
+		m_should_cleanup = other.m_should_cleanup;
+
+		other.m_external_head = nullptr;
+		other.m_should_cleanup = false;
+		wl_list_init(&other.m_head);
+	}
+
+private:
+	wl_list m_head {};
+	wl_list *m_external_head { nullptr };
+	bool m_should_cleanup { true };
+};
+
+} // namespace Lunar::Wayland

src/wayland/Region.h

@@ -0,0 +1,43 @@
+#pragma once
+
+#include <cstdint>
+#include <vector>
+
+#include <wayland-server-core.h>
+
+namespace Lunar::Wayland {
+
+struct Region {
+	struct Box {
+		std::int32_t x {};
+		std::int32_t y {};
+		std::int32_t width {};
+		std::int32_t height {};
+	};
+
+	explicit Region(wl_resource *resource)
+	    : m_resource(resource)
+	{
+	}
+
+	auto resource() const -> wl_resource * { return m_resource; }
+
+	auto add(std::int32_t x, std::int32_t y, std::int32_t width,
+	    std::int32_t height) -> void
+	{
+		m_boxes.push_back(Box { x, y, width, height });
+	}
+
+	auto subtract(std::int32_t x, std::int32_t y, std::int32_t width,
+	    std::int32_t height) -> void
+	{
+		m_subtract_boxes.push_back(Box { x, y, width, height });
+	}
+
+private:
+	wl_resource *m_resource {};
+	std::vector<Box> m_boxes {};
+	std::vector<Box> m_subtract_boxes {};
+};
+
+} // namespace Lunar::Wayland

src/wayland/Shm.cpp

@@ -0,0 +1,96 @@
+#include "Shm.h"
+
+#include <cerrno>
+#include <cstring>
+#include <sys/mman.h>
+#include <unistd.h>
+#include <utility>
+
+namespace Lunar::Wayland {
+
+ShmPool::ShmPool(int fd, std::size_t size, Logger &logger)
+    : m_logger(logger)
+    , m_fd(fd)
+    , m_size(size)
+{
+	m_data = mmap(nullptr, m_size, PROT_READ | PROT_WRITE, MAP_SHARED, m_fd, 0);
+	if (m_data == MAP_FAILED) {
+		m_data = nullptr;
+		m_logger.err("Failed to mmap shm pool: {}", std::strerror(errno));
+	}
+}
+
+ShmPool::~ShmPool()
+{
+	if (m_data) {
+		munmap(m_data, m_size);
+	}
+	if (m_fd >= 0) {
+		close(m_fd);
+	}
+}
+
+auto ShmPool::resize(std::size_t new_size) -> bool
+{
+	if (!m_data) {
+		return false;
+	}
+	void *new_data = mremap(m_data, m_size, new_size, MREMAP_MAYMOVE);
+	if (new_data == MAP_FAILED) {
+		m_logger.err("Failed to resize shm pool: {}", std::strerror(errno));
+		return false;
+	}
+	m_data = new_data;
+	m_size = new_size;
+	return true;
+}
+
+auto ShmPool::data() const -> std::byte *
+{
+	return static_cast<std::byte *>(m_data);
+}
+
+ShmBuffer::ShmBuffer(std::shared_ptr<ShmPool> pool, wl_resource *resource,
+    std::int32_t offset, std::int32_t width, std::int32_t height,
+    std::int32_t stride, std::uint32_t format)
+    : pool(std::move(pool))
+    , resource(resource)
+    , offset(offset)
+    , width(width)
+    , height(height)
+    , stride(stride)
+    , format(format)
+{
+}
+
+auto ShmBuffer::data() const -> std::byte *
+{
+	if (!pool || !pool->data()) {
+		return nullptr;
+	}
+	return pool->data() + offset;
+}
+
+auto ShmBuffer::byte_size() const -> std::size_t
+{
+	if (height <= 0 || stride <= 0) {
+		return 0;
+	}
+	return static_cast<std::size_t>(height) * static_cast<std::size_t>(stride);
+}
+
+auto shm_buffer_from_resource(wl_resource *resource)
+    -> std::shared_ptr<ShmBuffer>
+{
+	if (!resource) {
+		return {};
+	}
+	auto *handle { static_cast<std::shared_ptr<ShmBuffer> *>(
+		wl_resource_get_user_data(resource)) };
+	if (!handle) {
+		return {};
+	}
+	return *handle;
+}
+
+} // namespace Lunar::Wayland

src/wayland/Shm.h

@@ -0,0 +1,48 @@
+#pragma once
+
+#include <cstddef>
+#include <cstdint>
+#include <memory>
+
+#include <wayland-server-core.h>
+
+#include "../Logger.h"
+
+namespace Lunar::Wayland {
+
+struct ShmPool {
+	explicit ShmPool(int fd, std::size_t size, Logger &logger);
+	~ShmPool();
+
+	auto resize(std::size_t new_size) -> bool;
+	auto data() const -> std::byte *;
+	auto size() const -> std::size_t { return m_size; }
+
+private:
+	Logger &m_logger;
+	int m_fd { -1 };
+	std::size_t m_size { 0 };
+	void *m_data { nullptr };
+};
+
+struct ShmBuffer {
+	ShmBuffer(std::shared_ptr<ShmPool> pool, wl_resource *resource,
+	    std::int32_t offset, std::int32_t width, std::int32_t height,
+	    std::int32_t stride, std::uint32_t format);
+
+	auto data() const -> std::byte *;
+	auto byte_size() const -> std::size_t;
+
+	std::shared_ptr<ShmPool> pool;
+	wl_resource *resource {};
+	std::int32_t offset {};
+	std::int32_t width {};
+	std::int32_t height {};
+	std::int32_t stride {};
+	std::uint32_t format {};
+};
+
+auto shm_buffer_from_resource(wl_resource *resource)
+    -> std::shared_ptr<ShmBuffer>;
+
+} // namespace Lunar::Wayland

src/wayland/Signal.h

@@ -0,0 +1,28 @@
+#pragma once
+
+#include <cassert>
+
+#include <wayland-server-core.h>
+
+namespace Lunar::Wayland {
+
+struct Signal {
+	Signal(wl_signal *signal)
+	    : m_signal { std::move(signal) }
+	{
+		assert(m_signal);
+	}
+	~Signal() = default;
+
+	inline auto c_ptr() const -> wl_signal * { return m_signal; }
+
+	template<typename T = void> auto flush(T *data)
+	{
+		wl_signal_emit_mutable(m_signal, (void *)data);
+	}
+
+private:
+	wl_signal *m_signal {};
+};
+
+} // namespace Lunar::Wayland

src/wayland/Surface.cpp

@@ -0,0 +1,98 @@
+#include "Surface.h"
+
+#include <algorithm>
+#include <utility>
+
+#include "WaylandServer.h"
+
+#include "wayland-server-protocol.h"
+
+namespace Lunar::Wayland {
+
+Surface::Surface(WaylandServer &server, wl_resource *resource)
+    : m_server(server)
+    , m_resource(resource)
+{
+	m_server.register_surface(this);
+}
+
+Surface::~Surface()
+{
+	m_server.unregister_surface(this);
+	for (auto *callback : m_frame_callbacks) {
+		if (callback) {
+			wl_callback_send_done(callback, m_server.now_ms());
+			wl_resource_destroy(callback);
+		}
+	}
+}
+
+auto Surface::attach(
+    std::shared_ptr<ShmBuffer> buffer, std::int32_t x, std::int32_t y) -> void
+{
+	m_pending_buffer = std::move(buffer);
+	m_pending_offset_x = x;
+	m_pending_offset_y = y;
+}
+
+auto Surface::damage(std::int32_t, std::int32_t, std::int32_t, std::int32_t)
+    -> void
+{
+}
+
+auto Surface::damage_buffer(
+    std::int32_t, std::int32_t, std::int32_t, std::int32_t) -> void
+{
+}
+
+auto Surface::frame(wl_resource *callback) -> void
+{
+	if (!callback) {
+		return;
+	}
+	m_frame_callbacks.push_back(callback);
+}
+
+auto Surface::commit() -> void
+{
+	auto previous { m_current_buffer };
+	m_current_buffer = m_pending_buffer;
+	m_pending_buffer.reset();
+
+	if (previous && previous != m_current_buffer && previous->resource) {
+		wl_buffer_send_release(previous->resource);
+	}
+
+	if (!m_frame_callbacks.empty()) {
+		auto callbacks { std::move(m_frame_callbacks) };
+		auto done_time { m_server.now_ms() };
+		for (auto *callback : callbacks) {
+			if (callback) {
+				wl_callback_send_done(callback, done_time);
+				wl_resource_destroy(callback);
+			}
+		}
+	}
+}
+
+auto Surface::set_opaque_region(std::shared_ptr<Region> region) -> void
+{
+	m_opaque_region = std::move(region);
+}
+
+auto Surface::set_input_region(std::shared_ptr<Region> region) -> void
+{
+	m_input_region = std::move(region);
+}
+
+auto Surface::set_buffer_transform(std::int32_t transform) -> void
+{
+	m_buffer_transform = transform;
+}
+
+auto Surface::set_buffer_scale(std::int32_t scale) -> void
+{
+	m_buffer_scale = std::max(1, scale);
+}
+
+} // namespace Lunar::Wayland

src/wayland/Surface.h

@@ -0,0 +1,53 @@
+#pragma once
+
+#include <cstdint>
+#include <memory>
+#include <vector>
+
+#include <wayland-server-core.h>
+
+#include "Region.h"
+#include "Shm.h"
+
+namespace Lunar::Wayland {
+
+struct WaylandServer;
+
+struct Surface {
+	explicit Surface(WaylandServer &server, wl_resource *resource);
+	~Surface();
+
+	auto resource() const -> wl_resource * { return m_resource; }
+	auto current_buffer() const -> std::shared_ptr<ShmBuffer> const &
+	{
+		return m_current_buffer;
+	}
+
+	auto attach(std::shared_ptr<ShmBuffer> buffer, std::int32_t x,
+	    std::int32_t y) -> void;
+	auto damage(std::int32_t x, std::int32_t y, std::int32_t width,
+	    std::int32_t height) -> void;
+	auto damage_buffer(std::int32_t x, std::int32_t y, std::int32_t width,
+	    std::int32_t height) -> void;
+	auto frame(wl_resource *callback) -> void;
+	auto commit() -> void;
+	auto set_opaque_region(std::shared_ptr<Region> region) -> void;
+	auto set_input_region(std::shared_ptr<Region> region) -> void;
+	auto set_buffer_transform(std::int32_t transform) -> void;
+	auto set_buffer_scale(std::int32_t scale) -> void;
+
+private:
+	WaylandServer &m_server;
+	wl_resource *m_resource {};
+	std::shared_ptr<ShmBuffer> m_pending_buffer {};
+	std::shared_ptr<ShmBuffer> m_current_buffer {};
+	std::shared_ptr<Region> m_opaque_region {};
+	std::shared_ptr<Region> m_input_region {};
+	std::vector<wl_resource *> m_frame_callbacks {};
+	std::int32_t m_buffer_transform { 0 };
+	std::int32_t m_buffer_scale { 1 };
+	std::int32_t m_pending_offset_x { 0 };
+	std::int32_t m_pending_offset_y { 0 };
+};
+
+} // namespace Lunar::Wayland

src/wayland/WaylandServer.cpp

@@ -0,0 +1,81 @@
+#include "WaylandServer.h"
+
+#include <algorithm>
+#include <chrono>
+#include <span>
+#include <stdexcept>
+
+namespace Lunar::Wayland {
+
+WaylandServer::WaylandServer(Logger &logger)
+    : m_logger(logger)
+{
+	m_loop = wl_display_get_event_loop(m_display.c_ptr());
+	if (!m_loop) {
+		throw std::runtime_error("Failed to get Wayland event loop");
+	}
+
+	auto *socket_name { wl_display_add_socket_auto(m_display.c_ptr()) };
+	if (!socket_name) {
+		throw std::runtime_error("Failed to create Wayland socket");
+	}
+	if (socket_name) {
+		m_socket_name = socket_name;
+	}
+
+	m_logger.info("Wayland listening on {}", m_socket_name);
+
+	m_compositor_global = create_compositor_global();
+	m_shm_global = create_shm_global();
+	m_xdg_wm_base_global = create_xdg_wm_base_global();
+}
+
+WaylandServer::~WaylandServer() = default;
+
+auto WaylandServer::dispatch() -> void
+{
+	if (!m_loop) {
+		return;
+	}
+	wl_event_loop_dispatch(m_loop, 0);
+}
+
+auto WaylandServer::flush() -> void
+{
+	wl_display_flush_clients(m_display.c_ptr());
+}
+
+auto WaylandServer::now_ms() const -> std::uint32_t
+{
+	using Clock = std::chrono::steady_clock;
+	auto now { Clock::now().time_since_epoch() };
+	auto ms {
+		std::chrono::duration_cast<std::chrono::milliseconds>(now).count()
+	};
+	return static_cast<std::uint32_t>(ms);
+}
+
+auto WaylandServer::register_surface(Surface *surface) -> void
+{
+	if (!surface) {
+		return;
+	}
+	m_surfaces.push_back(surface);
+}
+
+auto WaylandServer::unregister_surface(Surface *surface) -> void
+{
+	if (!surface) {
+		return;
+	}
+	auto it { std::remove(m_surfaces.begin(), m_surfaces.end(), surface) };
+	m_surfaces.erase(it, m_surfaces.end());
+}
+
+auto WaylandServer::surfaces() const -> std::span<Surface *const>
+{
+	return { m_surfaces.data(), m_surfaces.size() };
+}
+
+} // namespace Lunar::Wayland
+#include "Surface.h"

src/wayland/WaylandServer.h

@@ -0,0 +1,50 @@
+#pragma once
+
+#include <cstdint>
+#include <memory>
+#include <span>
+#include <string>
+#include <string_view>
+#include <vector>
+
+#include <wayland-server-core.h>
+
+#include "../Logger.h"
+
+#include "Display.h"
+#include "Global.h"
+
+namespace Lunar::Wayland {
+
+struct Surface;
+
+struct WaylandServer {
+	explicit WaylandServer(Logger &logger);
+	~WaylandServer();
+
+	auto display() -> Display & { return m_display; }
+	auto logger() -> Logger & { return m_logger; }
+	auto socket_name() const -> std::string_view { return m_socket_name; }
+	auto dispatch() -> void;
+	auto flush() -> void;
+	auto now_ms() const -> std::uint32_t;
+	auto register_surface(Surface *surface) -> void;
+	auto unregister_surface(Surface *surface) -> void;
+	auto surfaces() const -> std::span<Surface *const>;
+
+private:
+	auto create_compositor_global() -> std::unique_ptr<Global>;
+	auto create_shm_global() -> std::unique_ptr<Global>;
+	auto create_xdg_wm_base_global() -> std::unique_ptr<Global>;
+
+	Logger &m_logger;
+	Display m_display {};
+	wl_event_loop *m_loop { nullptr };
+	std::string m_socket_name {};
+	std::unique_ptr<Global> m_compositor_global {};
+	std::unique_ptr<Global> m_shm_global {};
+	std::unique_ptr<Global> m_xdg_wm_base_global {};
+	std::vector<Surface *> m_surfaces {};
+};
+
+} // namespace Lunar::Wayland

src/wayland/protocols/CompositorProtocol.cpp

@@ -0,0 +1,254 @@
+#include <algorithm>
+#include <cstdint>
+#include <memory>
+#include <utility>
+
+#include <wayland-server-core.h>
+
+#include "wayland-server-protocol.h"
+
+#include "../Region.h"
+#include "../Shm.h"
+#include "../Surface.h"
+#include "../WaylandServer.h"
+
+namespace Lunar::Wayland {
+namespace {
+
+constexpr std::uint32_t COMPOSITOR_VERSION = 4;
+
+auto resource_version(wl_resource *resource) -> std::uint32_t
+{
+	return std::min<std::uint32_t>(
+	    wl_resource_get_version(resource), COMPOSITOR_VERSION);
+}
+
+auto region_from_resource(wl_resource *resource) -> std::shared_ptr<Region>
+{
+	if (!resource) {
+		return {};
+	}
+	auto *handle { static_cast<std::shared_ptr<Region> *>(
+		wl_resource_get_user_data(resource)) };
+	if (!handle) {
+		return {};
+	}
+	return *handle;
+}
+
+void region_destroy(wl_resource *resource)
+{
+	auto *handle { static_cast<std::shared_ptr<Region> *>(
+		wl_resource_get_user_data(resource)) };
+	delete handle;
+}
+
+void region_handle_destroy(wl_client *, wl_resource *resource)
+{
+	wl_resource_destroy(resource);
+}
+
+void region_handle_add(wl_client *, wl_resource *resource, std::int32_t x,
+    std::int32_t y, std::int32_t width, std::int32_t height)
+{
+	if (auto region { region_from_resource(resource) }) {
+		region->add(x, y, width, height);
+	}
+}
+
+void region_handle_subtract(wl_client *, wl_resource *resource, std::int32_t x,
+    std::int32_t y, std::int32_t width, std::int32_t height)
+{
+	if (auto region { region_from_resource(resource) }) {
+		region->subtract(x, y, width, height);
+	}
+}
+
+struct wl_region_interface const REGION_INTERFACE = {
+	.destroy = region_handle_destroy,
+	.add = region_handle_add,
+	.subtract = region_handle_subtract,
+};
+
+auto surface_from_resource(wl_resource *resource) -> Surface *
+{
+	return static_cast<Surface *>(wl_resource_get_user_data(resource));
+}
+
+void surface_destroy_resource(wl_resource *resource)
+{
+	auto *surface { surface_from_resource(resource) };
+	delete surface;
+}
+
+void surface_handle_destroy(wl_client *, wl_resource *resource)
+{
+	wl_resource_destroy(resource);
+}
+
+void surface_handle_attach(wl_client *, wl_resource *resource,
+    wl_resource *buffer_resource, std::int32_t x, std::int32_t y)
+{
+	auto *surface { surface_from_resource(resource) };
+	if (!surface) {
+		return;
+	}
+	surface->attach(shm_buffer_from_resource(buffer_resource), x, y);
+}
+
+void surface_handle_damage(wl_client *, wl_resource *resource, std::int32_t x,
+    std::int32_t y, std::int32_t width, std::int32_t height)
+{
+	if (auto *surface { surface_from_resource(resource) }) {
+		surface->damage(x, y, width, height);
+	}
+}
+
+void surface_handle_frame(
+    wl_client *client, wl_resource *resource, std::uint32_t callback_id)
+{
+	auto *surface { surface_from_resource(resource) };
+	if (!surface) {
+		return;
+	}
+	auto version { wl_resource_get_version(resource) };
+	auto *callback_resource { wl_resource_create(
+		client, &wl_callback_interface, version, callback_id) };
+	if (!callback_resource) {
+		return;
+	}
+	wl_resource_set_implementation(
+	    callback_resource, nullptr, nullptr, nullptr);
+	surface->frame(callback_resource);
+}
+
+void surface_handle_set_opaque_region(
+    wl_client *, wl_resource *resource, wl_resource *region_resource)
+{
+	if (auto *surface { surface_from_resource(resource) }) {
+		surface->set_opaque_region(region_from_resource(region_resource));
+	}
+}
+
+void surface_handle_set_input_region(
+    wl_client *, wl_resource *resource, wl_resource *region_resource)
+{
+	if (auto *surface { surface_from_resource(resource) }) {
+		surface->set_input_region(region_from_resource(region_resource));
+	}
+}
+
+void surface_handle_commit(wl_client *, wl_resource *resource)
+{
+	if (auto *surface { surface_from_resource(resource) }) {
+		surface->commit();
+	}
+}
+
+void surface_handle_set_buffer_transform(
+    wl_client *, wl_resource *resource, std::int32_t transform)
+{
+	if (auto *surface { surface_from_resource(resource) }) {
+		surface->set_buffer_transform(transform);
+	}
+}
+
+void surface_handle_set_buffer_scale(
+    wl_client *, wl_resource *resource, std::int32_t scale)
+{
+	if (auto *surface { surface_from_resource(resource) }) {
+		surface->set_buffer_scale(scale);
+	}
+}
+
+void surface_handle_damage_buffer(wl_client *, wl_resource *resource,
+    std::int32_t x, std::int32_t y, std::int32_t width, std::int32_t height)
+{
+	if (auto *surface { surface_from_resource(resource) }) {
+		surface->damage_buffer(x, y, width, height);
+	}
+}
+
+void surface_handle_offset(
+    wl_client *, wl_resource *resource, std::int32_t x, std::int32_t y)
+{
+	(void)resource;
+	(void)x;
+	(void)y;
+}
+
+struct wl_surface_interface const SURFACE_INTERFACE = {
+	.destroy = surface_handle_destroy,
+	.attach = surface_handle_attach,
+	.damage = surface_handle_damage,
+	.frame = surface_handle_frame,
+	.set_opaque_region = surface_handle_set_opaque_region,
+	.set_input_region = surface_handle_set_input_region,
+	.commit = surface_handle_commit,
+	.set_buffer_transform = surface_handle_set_buffer_transform,
+	.set_buffer_scale = surface_handle_set_buffer_scale,
+	.damage_buffer = surface_handle_damage_buffer,
+	.offset = surface_handle_offset,
+};
+
+void compositor_handle_create_surface(
+    wl_client *client, wl_resource *resource, std::uint32_t id)
+{
+	auto *server { static_cast<WaylandServer *>(
+		wl_resource_get_user_data(resource)) };
+	if (!server) {
+		return;
+	}
+	auto version { resource_version(resource) };
+	auto *surface_resource { wl_resource_create(
+		client, &wl_surface_interface, version, id) };
+	if (!surface_resource) {
+		return;
+	}
+	auto *surface { new Surface(*server, surface_resource) };
+	wl_resource_set_implementation(surface_resource, &SURFACE_INTERFACE,
+	    surface, surface_destroy_resource);
+}
+
+void compositor_handle_create_region(
+    wl_client *client, wl_resource *resource, std::uint32_t id)
+{
+	auto version { wl_resource_get_version(resource) };
+	auto *region_resource { wl_resource_create(
+		client, &wl_region_interface, version, id) };
+	if (!region_resource) {
+		return;
+	}
+	auto region { std::make_shared<Region>(region_resource) };
+	auto *handle { new std::shared_ptr<Region>(std::move(region)) };
+	wl_resource_set_implementation(
+	    region_resource, &REGION_INTERFACE, handle, region_destroy);
+}
+
+struct wl_compositor_interface const COMPOSITOR_INTERFACE = {
+	.create_surface = compositor_handle_create_surface,
+	.create_region = compositor_handle_create_region,
+};
+
+void bind_compositor(
+    wl_client *client, void *data, std::uint32_t version, std::uint32_t id)
+{
+	auto *server { static_cast<WaylandServer *>(data) };
+	auto *resource { wl_resource_create(client, &wl_compositor_interface,
+		std::min(version, COMPOSITOR_VERSION), id) };
+	if (!resource) {
+		return;
+	}
+	wl_resource_set_implementation(
+	    resource, &COMPOSITOR_INTERFACE, server, nullptr);
+}
+
+} // namespace
+
+auto WaylandServer::create_compositor_global() -> std::unique_ptr<Global>
+{
+	return std::make_unique<Global>(display(), &wl_compositor_interface,
+	    COMPOSITOR_VERSION, this, bind_compositor);
+}
+
+} // namespace Lunar::Wayland

src/wayland/protocols/ShmProtocol.cpp

@@ -0,0 +1,171 @@
+#include <algorithm>
+#include <cstdint>
+#include <memory>
+#include <unistd.h>
+#include <utility>
+
+#include <wayland-server-core.h>
+
+#include "wayland-server-protocol.h"
+
+#include "../Shm.h"
+#include "../WaylandServer.h"
+
+namespace Lunar::Wayland {
+namespace {
+
+constexpr std::uint32_t SHM_VERSION = 2;
+
+auto shm_pool_from_resource(wl_resource *resource) -> std::shared_ptr<ShmPool>
+{
+	auto *handle { static_cast<std::shared_ptr<ShmPool> *>(
+		wl_resource_get_user_data(resource)) };
+	if (!handle) {
+		return {};
+	}
+	return *handle;
+}
+
+void shm_pool_destroy_resource(wl_resource *resource)
+{
+	auto *handle { static_cast<std::shared_ptr<ShmPool> *>(
+		wl_resource_get_user_data(resource)) };
+	delete handle;
+}
+
+void shm_buffer_destroy_resource(wl_resource *resource)
+{
+	auto *handle { static_cast<std::shared_ptr<ShmBuffer> *>(
+		wl_resource_get_user_data(resource)) };
+	delete handle;
+}
+
+void shm_pool_handle_destroy(wl_client *, wl_resource *resource)
+{
+	wl_resource_destroy(resource);
+}
+
+void shm_pool_handle_resize(
+    wl_client *, wl_resource *resource, std::int32_t size)
+{
+	auto pool { shm_pool_from_resource(resource) };
+	if (!pool) {
+		return;
+	}
+	if (size <= 0) {
+		return;
+	}
+	pool->resize(static_cast<std::size_t>(size));
+}
+
+void shm_handle_release(wl_client *, wl_resource *) { }
+
+void shm_pool_handle_create_buffer(wl_client *client, wl_resource *resource,
+    std::uint32_t id, std::int32_t offset, std::int32_t width,
+    std::int32_t height, std::int32_t stride, std::uint32_t format)
+{
+	auto pool { shm_pool_from_resource(resource) };
+	if (!pool) {
+		return;
+	}
+
+	if (width <= 0 || height <= 0 || stride <= 0 || offset < 0) {
+		wl_resource_post_error(resource, WL_SHM_ERROR_INVALID_STRIDE,
+		    "Invalid shm buffer geometry");
+		return;
+	}
+
+	if (format != WL_SHM_FORMAT_XRGB8888 && format != WL_SHM_FORMAT_ARGB8888) {
+		wl_resource_post_error(
+		    resource, WL_SHM_ERROR_INVALID_FORMAT, "Unsupported shm format");
+		return;
+	}
+
+	auto required { static_cast<std::size_t>(offset)
+		+ static_cast<std::size_t>(height) * static_cast<std::size_t>(stride) };
+	if (required > pool->size()) {
+		wl_resource_post_error(resource, WL_SHM_ERROR_INVALID_STRIDE,
+		    "Shm buffer size out of bounds");
+		return;
+	}
+
+	auto *buffer_resource { wl_resource_create(
+		client, &wl_buffer_interface, 1, id) };
+	if (!buffer_resource) {
+		return;
+	}
+
+	auto buffer { std::make_shared<ShmBuffer>(
+		pool, buffer_resource, offset, width, height, stride, format) };
+	auto *handle { new std::shared_ptr<ShmBuffer>(std::move(buffer)) };
+	wl_resource_set_implementation(
+	    buffer_resource, nullptr, handle, shm_buffer_destroy_resource);
+}
+
+struct wl_shm_pool_interface const SHM_POOL_INTERFACE = {
+	.create_buffer = shm_pool_handle_create_buffer,
+	.destroy = shm_pool_handle_destroy,
+	.resize = shm_pool_handle_resize,
+};
+
+void shm_handle_create_pool(wl_client *client, wl_resource *resource,
+    std::uint32_t id, int fd, std::int32_t size)
+{
+	auto *server
+	    = static_cast<WaylandServer *>(wl_resource_get_user_data(resource));
+	if (!server) {
+		return;
+	}
+	if (size <= 0) {
+		wl_resource_post_error(
+		    resource, WL_SHM_ERROR_INVALID_STRIDE, "Invalid shm pool size");
+		close(fd);
+		return;
+	}
+
+	auto pool { std::make_shared<ShmPool>(
+		fd, static_cast<std::size_t>(size), server->logger()) };
+	if (!pool->data()) {
+		wl_resource_post_error(
+		    resource, WL_SHM_ERROR_INVALID_FD, "Failed to mmap shm pool");
+		return;
+	}
+
+	auto *pool_resource
+	    = wl_resource_create(client, &wl_shm_pool_interface, 1, id);
+	if (!pool_resource) {
+		return;
+	}
+	auto *handle { new std::shared_ptr<ShmPool>(std::move(pool)) };
+	wl_resource_set_implementation(
+	    pool_resource, &SHM_POOL_INTERFACE, handle, shm_pool_destroy_resource);
+}
+
+struct wl_shm_interface const SHM_INTERFACE = {
+	.create_pool = shm_handle_create_pool,
+	.release = shm_handle_release,
+};
+
+void bind_shm(
+    wl_client *client, void *data, std::uint32_t version, std::uint32_t id)
+{
+	auto *server { static_cast<WaylandServer *>(data) };
+	auto *resource { wl_resource_create(
+		client, &wl_shm_interface, std::min(version, SHM_VERSION), id) };
+	if (!resource) {
+		return;
+	}
+	wl_resource_set_implementation(resource, &SHM_INTERFACE, server, nullptr);
+	wl_shm_send_format(resource, WL_SHM_FORMAT_XRGB8888);
+	wl_shm_send_format(resource, WL_SHM_FORMAT_ARGB8888);
+}
+
+} // namespace
+
+auto WaylandServer::create_shm_global() -> std::unique_ptr<Global>
+{
+	return std::make_unique<Global>(
+	    display(), &wl_shm_interface, SHM_VERSION, this, bind_shm);
+}
+
+} // namespace Lunar::Wayland

src/wayland/protocols/XdgShellProtocol.cpp

@@ -0,0 +1,327 @@
+#include <algorithm>
+#include <cstdint>
+
+#include <wayland-server-core.h>
+
+#include "xdg-shell-server-protocol.h"
+
+#include "../WaylandServer.h"
+
+namespace Lunar::Wayland {
+namespace {
+
+constexpr std::uint32_t XDG_WM_BASE_VERSION = 7;
+
+struct XdgSurface {
+	WaylandServer &server;
+	wl_resource *resource {};
+	wl_resource *surface_resource {};
+	wl_resource *toplevel_resource {};
+	std::uint32_t last_serial { 0 };
+};
+
+struct XdgToplevel {
+	WaylandServer &server;
+	wl_resource *resource {};
+	XdgSurface *surface {};
+};
+
+extern struct xdg_toplevel_interface const XDG_TOPLEVEL_INTERFACE_IMPL;
+
+auto xdg_surface_from_resource(wl_resource *resource) -> XdgSurface *
+{
+	return static_cast<XdgSurface *>(wl_resource_get_user_data(resource));
+}
+
+auto xdg_toplevel_from_resource(wl_resource *resource) -> XdgToplevel *
+{
+	return static_cast<XdgToplevel *>(wl_resource_get_user_data(resource));
+}
+
+auto send_configure(XdgSurface &surface) -> void
+{
+	std::uint32_t serial = surface.server.display().next_serial();
+	if (surface.toplevel_resource) {
+		wl_array states;
+		wl_array_init(&states);
+		xdg_toplevel_send_configure(surface.toplevel_resource, 0, 0, &states);
+		wl_array_release(&states);
+	}
+	surface.last_serial = serial;
+	xdg_surface_send_configure(surface.resource, serial);
+}
+
+auto xdg_positioner_handle_destroy(wl_client *, wl_resource *resource) -> void
+{
+	wl_resource_destroy(resource);
+}
+
+auto xdg_positioner_handle_set_size(
+    wl_client *, wl_resource *, std::int32_t, std::int32_t) -> void
+{
+}
+
+auto xdg_positioner_handle_set_anchor_rect(wl_client *, wl_resource *,
+    std::int32_t, std::int32_t, std::int32_t, std::int32_t) -> void
+{
+}
+
+auto xdg_positioner_handle_set_anchor(wl_client *, wl_resource *, std::uint32_t)
+    -> void
+{
+}
+
+auto xdg_positioner_handle_set_gravity(
+    wl_client *, wl_resource *, std::uint32_t) -> void
+{
+}
+
+auto xdg_positioner_handle_set_constraint_adjustment(
+    wl_client *, wl_resource *, std::uint32_t) -> void
+{
+}
+
+auto xdg_positioner_handle_set_offset(
+    wl_client *, wl_resource *, std::int32_t, std::int32_t) -> void
+{
+}
+
+auto xdg_positioner_handle_set_reactive(wl_client *, wl_resource *) -> void { }
+
+auto xdg_positioner_handle_set_parent_size(
+    wl_client *, wl_resource *, std::int32_t, std::int32_t) -> void
+{
+}
+
+auto xdg_positioner_handle_set_parent_configure(
+    wl_client *, wl_resource *, std::uint32_t) -> void
+{
+}
+
+struct xdg_positioner_interface const XDG_POSITIONER_INTERFACE_IMPL = {
+	.destroy = xdg_positioner_handle_destroy,
+	.set_size = xdg_positioner_handle_set_size,
+	.set_anchor_rect = xdg_positioner_handle_set_anchor_rect,
+	.set_anchor = xdg_positioner_handle_set_anchor,
+	.set_gravity = xdg_positioner_handle_set_gravity,
+	.set_constraint_adjustment
+	= xdg_positioner_handle_set_constraint_adjustment,
+	.set_offset = xdg_positioner_handle_set_offset,
+	.set_reactive = xdg_positioner_handle_set_reactive,
+	.set_parent_size = xdg_positioner_handle_set_parent_size,
+	.set_parent_configure = xdg_positioner_handle_set_parent_configure,
+};
+
+auto xdg_surface_destroy_resource(wl_resource *resource) -> void
+{
+	auto *surface { xdg_surface_from_resource(resource) };
+	delete surface;
+}
+
+auto xdg_surface_handle_destroy(wl_client *, wl_resource *resource) -> void
+{
+	wl_resource_destroy(resource);
+}
+
+auto xdg_surface_handle_get_toplevel(
+    wl_client *client, wl_resource *resource, std::uint32_t id) -> void
+{
+	auto *surface { xdg_surface_from_resource(resource) };
+	if (!surface) {
+		return;
+	}
+	auto version { wl_resource_get_version(resource) };
+	auto *toplevel_resource { wl_resource_create(
+		client, &::xdg_toplevel_interface, version, id) };
+	if (!toplevel_resource) {
+		return;
+	}
+	auto *toplevel { new XdgToplevel {
+		surface->server, toplevel_resource, surface } };
+	surface->toplevel_resource = toplevel_resource;
+	wl_resource_set_implementation(toplevel_resource,
+	    &XDG_TOPLEVEL_INTERFACE_IMPL, toplevel, [](wl_resource *res) {
+		    auto *tl { xdg_toplevel_from_resource(res) };
+		    if (tl && tl->surface && tl->surface->toplevel_resource == res) {
+			    tl->surface->toplevel_resource = nullptr;
+		    }
+		    delete tl;
+	    });
+
+	send_configure(*surface);
+}
+
+auto xdg_surface_handle_get_popup(wl_client *, wl_resource *, std::uint32_t,
+    wl_resource *, wl_resource *) -> void
+{
+}
+
+auto xdg_surface_handle_set_window_geometry(wl_client *, wl_resource *,
+    std::int32_t, std::int32_t, std::int32_t, std::int32_t) -> void
+{
+}
+
+auto xdg_surface_handle_ack_configure(
+    wl_client *, wl_resource *resource, std::uint32_t serial) -> void
+{
+	if (auto *surface { xdg_surface_from_resource(resource) }) {
+		surface->last_serial = serial;
+	}
+}
+
+struct xdg_surface_interface const XDG_SURFACE_INTERFACE_IMPL = {
+	.destroy = xdg_surface_handle_destroy,
+	.get_toplevel = xdg_surface_handle_get_toplevel,
+	.get_popup = xdg_surface_handle_get_popup,
+	.set_window_geometry = xdg_surface_handle_set_window_geometry,
+	.ack_configure = xdg_surface_handle_ack_configure,
+};
+
+auto xdg_toplevel_handle_destroy(wl_client *, wl_resource *resource) -> void
+{
+	wl_resource_destroy(resource);
+}
+
+auto xdg_toplevel_handle_set_parent(wl_client *, wl_resource *, wl_resource *)
+    -> void
+{
+}
+
+auto xdg_toplevel_handle_set_title(wl_client *, wl_resource *, char const *)
+    -> void
+{
+}
+
+auto xdg_toplevel_handle_set_app_id(wl_client *, wl_resource *, char const *)
+    -> void
+{
+}
+
+auto xdg_toplevel_handle_show_window_menu(wl_client *, wl_resource *,
+    wl_resource *, std::uint32_t, std::int32_t, std::int32_t) -> void
+{
+}
+
+auto xdg_toplevel_handle_move(
+    wl_client *, wl_resource *, wl_resource *, std::uint32_t) -> void
+{
+}
+
+auto xdg_toplevel_handle_resize(wl_client *, wl_resource *, wl_resource *,
+    std::uint32_t, std::uint32_t) -> void
+{
+}
+
+auto xdg_toplevel_handle_set_max_size(
+    wl_client *, wl_resource *, std::int32_t, std::int32_t) -> void
+{
+}
+
+auto xdg_toplevel_handle_set_min_size(
+    wl_client *, wl_resource *, std::int32_t, std::int32_t) -> void
+{
+}
+
+auto xdg_toplevel_handle_set_maximized(wl_client *, wl_resource *) -> void { }
+
+auto xdg_toplevel_handle_unset_maximized(wl_client *, wl_resource *) -> void { }
+
+auto xdg_toplevel_handle_set_fullscreen(
+    wl_client *, wl_resource *, wl_resource *) -> void
+{
+}
+
+auto xdg_toplevel_handle_unset_fullscreen(wl_client *, wl_resource *) -> void {
+}
+
+auto xdg_toplevel_handle_set_minimized(wl_client *, wl_resource *) -> void { }
+
+struct xdg_toplevel_interface const XDG_TOPLEVEL_INTERFACE_IMPL = {
+	.destroy = xdg_toplevel_handle_destroy,
+	.set_parent = xdg_toplevel_handle_set_parent,
+	.set_title = xdg_toplevel_handle_set_title,
+	.set_app_id = xdg_toplevel_handle_set_app_id,
+	.show_window_menu = xdg_toplevel_handle_show_window_menu,
+	.move = xdg_toplevel_handle_move,
+	.resize = xdg_toplevel_handle_resize,
+	.set_max_size = xdg_toplevel_handle_set_max_size,
+	.set_min_size = xdg_toplevel_handle_set_min_size,
+	.set_maximized = xdg_toplevel_handle_set_maximized,
+	.unset_maximized = xdg_toplevel_handle_unset_maximized,
+	.set_fullscreen = xdg_toplevel_handle_set_fullscreen,
+	.unset_fullscreen = xdg_toplevel_handle_unset_fullscreen,
+	.set_minimized = xdg_toplevel_handle_set_minimized,
+};
+
+auto xdg_wm_base_handle_destroy(wl_client *, wl_resource *resource) -> void
+{
+	wl_resource_destroy(resource);
+}
+
+auto xdg_wm_base_handle_create_positioner(
+    wl_client *client, wl_resource *resource, std::uint32_t id) -> void
+{
+	auto version { wl_resource_get_version(resource) };
+	auto *positioner { wl_resource_create(
+		client, &::xdg_positioner_interface, version, id) };
+	if (!positioner) {
+		return;
+	}
+	wl_resource_set_implementation(
+	    positioner, &XDG_POSITIONER_INTERFACE_IMPL, nullptr, nullptr);
+}
+
+auto xdg_wm_base_handle_get_xdg_surface(wl_client *client,
+    wl_resource *resource, std::uint32_t id, wl_resource *surface_resource)
+    -> void
+{
+	auto *server
+	    = static_cast<WaylandServer *>(wl_resource_get_user_data(resource));
+	if (!server) {
+		return;
+	}
+	auto version { wl_resource_get_version(resource) };
+	auto *xdg_surface_resource { wl_resource_create(
+		client, &::xdg_surface_interface, version, id) };
+	if (!xdg_surface_resource) {
+		return;
+	}
+	auto *surface { new XdgSurface {
+		*server, xdg_surface_resource, surface_resource, nullptr, 0 } };
+	wl_resource_set_implementation(xdg_surface_resource,
+	    &XDG_SURFACE_INTERFACE_IMPL, surface, xdg_surface_destroy_resource);
+}
+
+auto xdg_wm_base_handle_pong(wl_client *, wl_resource *, std::uint32_t) -> void
+{
+}
+
+struct xdg_wm_base_interface const XDG_WM_BASE_INTERFACE_IMPL = {
+	.destroy = xdg_wm_base_handle_destroy,
+	.create_positioner = xdg_wm_base_handle_create_positioner,
+	.get_xdg_surface = xdg_wm_base_handle_get_xdg_surface,
+	.pong = xdg_wm_base_handle_pong,
+};
+
+auto bind_xdg_wm_base(wl_client *client, void *data, std::uint32_t version,
+    std::uint32_t id) -> void
+{
+	auto *server { static_cast<WaylandServer *>(data) };
+	auto *resource { wl_resource_create(client, &::xdg_wm_base_interface,
+		std::min(version, XDG_WM_BASE_VERSION), id) };
+	if (!resource) {
+		return;
+	}
+	wl_resource_set_implementation(
+	    resource, &XDG_WM_BASE_INTERFACE_IMPL, server, nullptr);
+}
+
+} // namespace
+
+auto WaylandServer::create_xdg_wm_base_global() -> std::unique_ptr<Global>
+{
+	return std::make_unique<Global>(display(), &::xdg_wm_base_interface,
+	    XDG_WM_BASE_VERSION, this, bind_xdg_wm_base);
+}
+
+} // namespace Lunar::Wayland

subprojects/tracy/.clang-format

@@ -0,0 +1,18 @@
+# Empirical format config, based on observed style guide
+# Use this only as an help to fit the surrounding code style - don't reformat whole files at once
+---
+BasedOnStyle: LLVM
+AllowShortIfStatementsOnASingleLine: WithoutElse
+AllowShortLoopsOnASingleLine: true
+AlwaysBreakTemplateDeclarations: Yes
+BreakBeforeBraces: Allman
+BreakConstructorInitializers: BeforeComma
+BreakStringLiterals: false
+ColumnLimit: 120
+FixNamespaceComments: false
+IndentPPDirectives: AfterHash
+IndentWidth: 4
+PointerAlignment: Left
+SpaceBeforeParens: Never
+SpacesInParentheses: true
+TabWidth: 4

subprojects/tracy/.clang-tidy

@@ -0,0 +1,62 @@
+---
+Checks:
+'
+	clang-diagnostic-*,
+	clang-analyzer-*,
+	bugprone-*,
+	google-*,
+	misc-*,
+	modernize-*,
+	performance-*,
+	readability-*,
+	-bugprone-easily-swappable-parameters,
+	-bugprone-implicit-widening-of-multiplication-result,
+	-bugprone-narrowing-conversions,
+	-bugprone-reserved-identifier,
+	-google-readability-braces-around-statements,
+	-google-readability-casting,
+	-google-readability-function-size,
+	-google-readability-todo,
+	-google-readability-namespace-comments,
+	-misc-confusable-identifiers,
+	-misc-no-recursion,
+	-modernize-avoid-c-arrays,
+	-modernize-deprecated-headers,
+	-modernize-use-default-member-init,
+	-modernize-use-trailing-return-type,
+	-performance-no-int-to-ptr,
+	-readability-braces-around-statements,
+	-readability-else-after-return,
+	-readability-function-cognitive-complexity,
+	-readability-identifier-length,
+	-readability-implicit-bool-conversion,
+	-readability-isolate-declaration,
+	-readability-magic-numbers,
+	-readability-qualified-auto,
+	-readability-uppercase-literal-suffix
+'
+WarningsAsErrors: ''
+HeaderFilterRegex: ''
+AnalyzeTemporaryDtors: false
+FormatStyle:     none
+CheckOptions:
+  llvm-else-after-return.WarnOnConditionVariables: 'false'
+  modernize-loop-convert.MinConfidence: reasonable
+  modernize-replace-auto-ptr.IncludeStyle: llvm
+  modernize-pass-by-value.IncludeStyle: llvm
+  google-readability-namespace-comments.ShortNamespaceLines: '10'
+  google-readability-namespace-comments.SpacesBeforeComments: '2'
+  cppcoreguidelines-non-private-member-variables-in-classes.IgnoreClassesWithAllMemberVariablesBeingPublic: 'true'
+  google-readability-braces-around-statements.ShortStatementLines: '1'
+  cert-err33-c.CheckedFunctions: '::aligned_alloc;::asctime_s;::at_quick_exit;::atexit;::bsearch;::bsearch_s;::btowc;::c16rtomb;::c32rtomb;::calloc;::clock;::cnd_broadcast;::cnd_init;::cnd_signal;::cnd_timedwait;::cnd_wait;::ctime_s;::fclose;::fflush;::fgetc;::fgetpos;::fgets;::fgetwc;::fopen;::fopen_s;::fprintf;::fprintf_s;::fputc;::fputs;::fputwc;::fputws;::fread;::freopen;::freopen_s;::fscanf;::fscanf_s;::fseek;::fsetpos;::ftell;::fwprintf;::fwprintf_s;::fwrite;::fwscanf;::fwscanf_s;::getc;::getchar;::getenv;::getenv_s;::gets_s;::getwc;::getwchar;::gmtime;::gmtime_s;::localtime;::localtime_s;::malloc;::mbrtoc16;::mbrtoc32;::mbsrtowcs;::mbsrtowcs_s;::mbstowcs;::mbstowcs_s;::memchr;::mktime;::mtx_init;::mtx_lock;::mtx_timedlock;::mtx_trylock;::mtx_unlock;::printf_s;::putc;::putwc;::raise;::realloc;::remove;::rename;::scanf;::scanf_s;::setlocale;::setvbuf;::signal;::snprintf;::snprintf_s;::sprintf;::sprintf_s;::sscanf;::sscanf_s;::strchr;::strerror_s;::strftime;::strpbrk;::strrchr;::strstr;::strtod;::strtof;::strtoimax;::strtok;::strtok_s;::strtol;::strtold;::strtoll;::strtoul;::strtoull;::strtoumax;::strxfrm;::swprintf;::swprintf_s;::swscanf;::swscanf_s;::thrd_create;::thrd_detach;::thrd_join;::thrd_sleep;::time;::timespec_get;::tmpfile;::tmpfile_s;::tmpnam;::tmpnam_s;::tss_create;::tss_get;::tss_set;::ungetc;::ungetwc;::vfprintf;::vfprintf_s;::vfscanf;::vfscanf_s;::vfwprintf;::vfwprintf_s;::vfwscanf;::vfwscanf_s;::vprintf_s;::vscanf;::vscanf_s;::vsnprintf;::vsnprintf_s;::vsprintf;::vsprintf_s;::vsscanf;::vsscanf_s;::vswprintf;::vswprintf_s;::vswscanf;::vswscanf_s;::vwprintf_s;::vwscanf;::vwscanf_s;::wcrtomb;::wcschr;::wcsftime;::wcspbrk;::wcsrchr;::wcsrtombs;::wcsrtombs_s;::wcsstr;::wcstod;::wcstof;::wcstoimax;::wcstok;::wcstok_s;::wcstol;::wcstold;::wcstoll;::wcstombs;::wcstombs_s;::wcstoul;::wcstoull;::wcstoumax;::wcsxfrm;::wctob;::wctrans;::wctype;::wmemchr;::wprintf_s;::wscanf;::wscanf_s;'
+  modernize-loop-convert.MaxCopySize: '16'
+  cert-dcl16-c.NewSuffixes: 'L;LL;LU;LLU'
+  cert-oop54-cpp.WarnOnlyIfThisHasSuspiciousField: 'false'
+  cert-str34-c.DiagnoseSignedUnsignedCharComparisons: 'false'
+  modernize-use-nullptr.NullMacros: 'NULL'
+  llvm-qualified-auto.AddConstToQualified: 'false'
+  modernize-loop-convert.NamingStyle: CamelCase
+  llvm-else-after-return.WarnOnUnfixable: 'false'
+  google-readability-function-size.StatementThreshold: '800'
+...
+

subprojects/tracy/.github/FUNDING.yml

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

subprojects/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

subprojects/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

subprojects/tracy/.gitignore

@@ -0,0 +1,36 @@
+.vs
+_build
+_compiler
+tools/*
+*.d
+*.o
+*.so
+*.swp
+*.obj
+imgui.ini
+test/tracy_test
+test/tracy_test.exe
+*/build/unix/*-*
+manual/t*.aux
+manual/t*.log
+manual/t*.out
+manual/t*.pdf
+manual/t*.synctex.gz
+manual/t*.toc
+manual/t*.bbl
+manual/t*.blg
+manual/t*.fdb_latexmk
+manual/t*.fls
+profiler/build/win32/packages
+profiler/build/win32/Tracy.aps
+.deps/
+.dirstamp
+/_*/**
+/**/__pycache__/**
+extra/vswhere.exe
+extra/tracy-build
+.cache
+compile_commands.json
+profiler/build/wasm/Tracy-release.*
+profiler/build/wasm/Tracy-debug.*
+profiler/build/wasm/embed.tracy

subprojects/tracy/.vscode/extensions.json

@@ -0,0 +1,7 @@
+{
+    "recommendations": [
+        "llvm-vs-code-extensions.vscode-clangd",
+        "vadimcn.vscode-lldb",
+        "ms-vscode.cmake-tools"
+    ]
+}

subprojects/tracy/.vscode/launch.json

@@ -0,0 +1,14 @@
+{
+    "version": "0.2.0",
+    "configurations": [
+        {
+            "name": "Launch",
+            "type": "lldb",
+            "request": "launch",
+            "program": "${command:cmake.launchTargetPath}",
+            "args": [],
+            "cwd": "${workspaceFolder}",
+            "terminal": "console"
+        }
+    ]
+}

subprojects/tracy/.vscode/settings.json

@@ -0,0 +1,20 @@
+{
+    "cmake.configureOnOpen": true,
+    "cmake.sourceDirectory": [
+        "${workspaceFolder}/profiler",
+        "${workspaceFolder}/capture",
+        "${workspaceFolder}/csvexport",
+        "${workspaceFolder}/import",
+        "${workspaceFolder}/update",
+        "${workspaceFolder}/test",
+        "${workspaceFolder}",
+    ],
+    "cmake.buildDirectory": "${sourceDirectory}/build",
+    "cmake.autoSelectActiveFolder": false,
+    "cmake.options.advanced": {
+        "folder": { "statusBarVisibility": "visible" },
+        "variant": { "statusBarVisibility": "compact" }
+    },
+    "cmake.copyCompileCommands": "${workspaceFolder}/compile_commands.json",
+    "lldb.launch.initCommands": ["command script import ${workspaceRoot}/extra/natvis.py"],
+}

subprojects/tracy/CMakeLists.txt

@@ -0,0 +1,195 @@
+cmake_minimum_required(VERSION 3.10)
+
+# Run version helper script
+include(cmake/version.cmake)
+
+project(Tracy LANGUAGES CXX VERSION ${TRACY_VERSION_STRING})
+
+file(GENERATE OUTPUT .gitignore CONTENT "*")
+
+if(${BUILD_SHARED_LIBS})
+	set(DEFAULT_STATIC OFF)
+else()
+	set(DEFAULT_STATIC ON)
+endif()
+
+option(TRACY_STATIC "Whether to build Tracy as a static library" ${DEFAULT_STATIC})
+
+find_package(Threads REQUIRED)
+
+set(TRACY_PUBLIC_DIR ${CMAKE_CURRENT_SOURCE_DIR}/public)
+
+if(TRACY_STATIC)
+	set(TRACY_VISIBILITY "STATIC")
+else()
+	set(TRACY_VISIBILITY "SHARED")
+endif()
+
+add_library(TracyClient ${TRACY_VISIBILITY} "${TRACY_PUBLIC_DIR}/TracyClient.cpp")
+target_compile_features(TracyClient PUBLIC cxx_std_11)
+target_include_directories(TracyClient SYSTEM PUBLIC
+    $<BUILD_INTERFACE:${TRACY_PUBLIC_DIR}>
+    $<INSTALL_INTERFACE:include>)
+target_link_libraries(
+    TracyClient
+    PUBLIC
+        Threads::Threads
+        ${CMAKE_DL_LIBS}
+)
+
+# Public dependency on some libraries required when using Mingw
+if(WIN32 AND ${CMAKE_CXX_COMPILER_ID} MATCHES "GNU|Clang")
+    target_link_libraries(TracyClient PUBLIC ws2_32 dbghelp)
+endif()
+
+if(CMAKE_SYSTEM_NAME MATCHES "FreeBSD")
+    find_library(EXECINFO_LIBRARY NAMES execinfo REQUIRED)
+    target_link_libraries(TracyClient PUBLIC ${EXECINFO_LIBRARY})
+endif()
+
+if(TRACY_LIBUNWIND_BACKTRACE)
+    include(FindPkgConfig)
+    pkg_check_modules(unwind REQUIRED libunwind)
+    target_include_directories(TracyClient INTERFACE ${unwind_INCLUDE_DIRS})
+    target_link_libraries(TracyClient INTERFACE ${unwind_LINK_LIBRARIES})
+endif()
+
+add_library(Tracy::TracyClient ALIAS TracyClient)
+
+macro(set_option option help value)
+    option(${option} ${help} ${value})
+    if(${option})
+        message(STATUS "${option}: ON")
+        target_compile_definitions(TracyClient PUBLIC ${option})
+    else()
+        message(STATUS "${option}: OFF")
+    endif()
+endmacro()
+
+set_option(TRACY_ENABLE "Enable profiling" ON)
+set_option(TRACY_ON_DEMAND "On-demand profiling" OFF)
+set_option(TRACY_CALLSTACK "Enforce callstack collection for tracy regions" OFF)
+set_option(TRACY_NO_CALLSTACK "Disable all callstack related functionality" OFF)
+set_option(TRACY_NO_CALLSTACK_INLINES "Disables the inline functions in callstacks" OFF)
+set_option(TRACY_ONLY_LOCALHOST "Only listen on the localhost interface" OFF)
+set_option(TRACY_NO_BROADCAST "Disable client discovery by broadcast to local network" OFF)
+set_option(TRACY_ONLY_IPV4 "Tracy will only accept connections on IPv4 addresses (disable IPv6)" OFF)
+set_option(TRACY_NO_CODE_TRANSFER "Disable collection of source code" OFF)
+set_option(TRACY_NO_CONTEXT_SWITCH "Disable capture of context switches" OFF)
+set_option(TRACY_NO_EXIT "Client executable does not exit until all profile data is sent to server" OFF)
+set_option(TRACY_NO_SAMPLING "Disable call stack sampling" OFF)
+set_option(TRACY_NO_VERIFY "Disable zone validation for C API" OFF)
+set_option(TRACY_NO_VSYNC_CAPTURE "Disable capture of hardware Vsync events" OFF)
+set_option(TRACY_NO_FRAME_IMAGE  "Disable the frame image support and its thread" OFF)
+set_option(TRACY_NO_SYSTEM_TRACING  "Disable systrace sampling" OFF)
+set_option(TRACY_PATCHABLE_NOPSLEDS  "Enable nopsleds for efficient patching by system-level tools (e.g. rr)" OFF)
+set_option(TRACY_DELAYED_INIT "Enable delayed initialization of the library (init on first call)" OFF)
+set_option(TRACY_MANUAL_LIFETIME "Enable the manual lifetime management of the profile" OFF)
+set_option(TRACY_FIBERS "Enable fibers support" OFF)
+set_option(TRACY_NO_CRASH_HANDLER "Disable crash handling" OFF)
+set_option(TRACY_TIMER_FALLBACK "Use lower resolution timers" OFF)
+set_option(TRACY_LIBUNWIND_BACKTRACE "Use libunwind backtracing where supported" OFF)
+set_option(TRACY_SYMBOL_OFFLINE_RESOLVE "Instead of full runtime symbol resolution, only resolve the image path and offset to enable offline symbol resolution" OFF)
+set_option(TRACY_LIBBACKTRACE_ELF_DYNLOAD_SUPPORT "Enable libbacktrace to support dynamically loaded elfs in symbol resolution resolution after the first symbol resolve operation" OFF)
+
+# advanced
+set_option(TRACY_VERBOSE "[advanced] Verbose output from the profiler" OFF)
+mark_as_advanced(TRACY_VERBOSE)
+set_option(TRACY_DEMANGLE "[advanced] Don't use default demangling function - You'll need to provide your own" OFF)
+mark_as_advanced(TRACY_DEMANGLE)
+
+if(NOT TRACY_STATIC)
+    target_compile_definitions(TracyClient PRIVATE TRACY_EXPORTS)
+    target_compile_definitions(TracyClient PUBLIC TRACY_IMPORTS)
+endif()
+
+include(CMakePackageConfigHelpers)
+include(GNUInstallDirs)
+
+set_target_properties(TracyClient PROPERTIES VERSION ${PROJECT_VERSION})
+
+set(tracy_includes
+    ${TRACY_PUBLIC_DIR}/tracy/TracyC.h
+    ${TRACY_PUBLIC_DIR}/tracy/Tracy.hpp
+    ${TRACY_PUBLIC_DIR}/tracy/TracyD3D11.hpp
+    ${TRACY_PUBLIC_DIR}/tracy/TracyD3D12.hpp
+    ${TRACY_PUBLIC_DIR}/tracy/TracyLua.hpp
+    ${TRACY_PUBLIC_DIR}/tracy/TracyOpenCL.hpp
+    ${TRACY_PUBLIC_DIR}/tracy/TracyOpenGL.hpp
+    ${TRACY_PUBLIC_DIR}/tracy/TracyVulkan.hpp)
+
+set(client_includes
+    ${TRACY_PUBLIC_DIR}/client/tracy_concurrentqueue.h
+    ${TRACY_PUBLIC_DIR}/client/tracy_rpmalloc.hpp
+    ${TRACY_PUBLIC_DIR}/client/tracy_SPSCQueue.h
+    ${TRACY_PUBLIC_DIR}/client/TracyKCore.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyArmCpuTable.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyCallstack.h
+    ${TRACY_PUBLIC_DIR}/client/TracyCallstack.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyCpuid.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyDebug.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyDxt1.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyFastVector.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyLock.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyProfiler.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyRingBuffer.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyScoped.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyStringHelpers.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracySysPower.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracySysTime.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracySysTrace.hpp
+    ${TRACY_PUBLIC_DIR}/client/TracyThread.hpp)
+
+set(common_includes
+    ${TRACY_PUBLIC_DIR}/common/tracy_lz4.hpp
+    ${TRACY_PUBLIC_DIR}/common/tracy_lz4hc.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracyAlign.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracyAlloc.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracyApi.h
+    ${TRACY_PUBLIC_DIR}/common/TracyColor.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracyForceInline.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracyMutex.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracyProtocol.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracyQueue.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracySocket.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracyStackFrames.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracySystem.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracyUwp.hpp
+    ${TRACY_PUBLIC_DIR}/common/TracyYield.hpp)
+
+install(TARGETS TracyClient
+        EXPORT TracyConfig
+        RUNTIME  DESTINATION ${CMAKE_INSTALL_BINDIR}
+        LIBRARY  DESTINATION ${CMAKE_INSTALL_LIBDIR}
+        ARCHIVE  DESTINATION ${CMAKE_INSTALL_LIBDIR}
+        COMPONENT lib)
+# Export targets to build tree root
+export(TARGETS TracyClient
+       NAMESPACE Tracy::
+       FILE ${CMAKE_BINARY_DIR}/TracyTargets.cmake)
+install(FILES ${tracy_includes}
+        DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/tracy)
+install(FILES ${client_includes}
+        DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/client)
+install(FILES ${common_includes}
+        DESTINATION ${CMAKE_INSTALL_INCLUDEDIR}/common)
+install(EXPORT TracyConfig
+        NAMESPACE Tracy::
+        FILE TracyTargets.cmake
+        DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/Tracy)
+include(CMakePackageConfigHelpers)
+configure_package_config_file(${CMAKE_CURRENT_SOURCE_DIR}/Config.cmake.in
+        "${CMAKE_CURRENT_BINARY_DIR}/TracyConfig.cmake"
+        INSTALL_DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/Tracy)
+install(FILES ${CMAKE_CURRENT_BINARY_DIR}/TracyConfig.cmake
+        DESTINATION ${CMAKE_INSTALL_DATAROOTDIR}/Tracy)
+
+option(TRACY_CLIENT_PYTHON "Whether to build Tracy python client library" OFF)
+
+if(TRACY_CLIENT_PYTHON)
+    if(TRACY_STATIC)
+        message(FATAL_ERROR "Python-bindings require a shared client library")
+    endif()
+
+    add_subdirectory(python)
+endif()

subprojects/tracy/Config.cmake.in

@@ -0,0 +1,6 @@
+@PACKAGE_INIT@
+
+include(CMakeFindDependencyMacro)
+find_dependency(Threads REQUIRED)
+
+include("${CMAKE_CURRENT_LIST_DIR}/TracyTargets.cmake")

subprojects/tracy/LICENSE

@@ -0,0 +1,27 @@
+Tracy Profiler (https://github.com/wolfpld/tracy) is licensed under the
+3-clause BSD license.
+
+Copyright (c) 2017-2024, Bartosz Taudul <wolf@nereid.pl>
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+    * Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in the
+      documentation and/or other materials provided with the distribution.
+    * Neither the name of the <organization> nor the
+      names of its contributors may be used to endorse or promote products
+      derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL <COPYRIGHT HOLDER> 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.

subprojects/tracy/README.md

@@ -0,0 +1,28 @@
+# Tracy Profiler
+
+[![Sponsor](.github/sponsor.png)](https://github.com/sponsors/wolfpld/)
+
+### A real time, nanosecond resolution, remote telemetry, hybrid frame and sampling profiler for games and other applications.
+
+Tracy supports profiling CPU (Direct support is provided for C, C++, Lua and Python integration. At the same time, third-party bindings to many other languages exist on the internet, such as [Rust](https://github.com/nagisa/rust_tracy_client), [Zig](https://github.com/nektro/zig-tracy), [C#](https://github.com/clibequilibrium/Tracy-CSharp), [OCaml](https://github.com/imandra-ai/ocaml-tracy), [Odin](https://github.com/oskarnp/odin-tracy), etc.), GPU (All major graphic APIs: OpenGL, Vulkan, Direct3D 11/12, OpenCL.), memory allocations, locks, context switches, automatically attribute screenshots to captured frames, and much more.
+
+- [Documentation](https://github.com/wolfpld/tracy/releases/latest/download/tracy.pdf) for usage and build process instructions
+- [Releases](https://github.com/wolfpld/tracy/releases) containing the documentation (`tracy.pdf`) and compiled Windows x64 binaries (`Tracy-<version>.7z`) as assets
+- [Changelog](NEWS)
+- [Interactive demo](https://tracy.nereid.pl/)
+
+![](doc/profiler.png)
+
+![](doc/profiler2.png)
+
+![](doc/profiler3.png)
+
+[An Introduction to Tracy Profiler in C++ - Marcos Slomp - CppCon 2023](https://youtu.be/ghXk3Bk5F2U?t=37)
+
+[Introduction to Tracy Profiler v0.2](https://www.youtube.com/watch?v=fB5B46lbapc)  
+[New features in Tracy Profiler v0.3](https://www.youtube.com/watch?v=3SXpDpDh2Uo)  
+[New features in Tracy Profiler v0.4](https://www.youtube.com/watch?v=eAkgkaO8B9o)  
+[New features in Tracy Profiler v0.5](https://www.youtube.com/watch?v=P6E7qLMmzTQ)  
+[New features in Tracy Profiler v0.6](https://www.youtube.com/watch?v=uJkrFgriuOo)  
+[New features in Tracy Profiler v0.7](https://www.youtube.com/watch?v=_hU7vw00MZ4)  
+[New features in Tracy Profiler v0.8](https://www.youtube.com/watch?v=30wpRpHTTag)

subprojects/tracy/capture/CMakeLists.txt

@@ -0,0 +1,27 @@
+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_STATISTICS "Disable calculation of statistics" ON)
+option(NO_PARALLEL_STL "Disable parallel STL" OFF)
+
+include(${CMAKE_CURRENT_LIST_DIR}/../cmake/version.cmake)
+
+set(CMAKE_CXX_STANDARD 20)
+
+project(
+    tracy-capture
+    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/capture.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})

subprojects/tracy/capture/src/capture.cpp

@@ -0,0 +1,364 @@
+#ifdef _WIN32
+#  include <windows.h>
+#  include <io.h>
+#else
+#  include <unistd.h>
+#endif
+
+#include <atomic>
+#include <chrono>
+#include <inttypes.h>
+#include <mutex>
+#include <signal.h>
+#include <stdarg.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <sys/stat.h>
+
+#include "../../public/common/TracyProtocol.hpp"
+#include "../../public/common/TracyStackFrames.hpp"
+#include "../../server/TracyFileWrite.hpp"
+#include "../../server/TracyMemory.hpp"
+#include "../../server/TracyPrint.hpp"
+#include "../../server/TracySysUtil.hpp"
+#include "../../server/TracyWorker.hpp"
+
+#ifdef _WIN32
+#  include "../../getopt/getopt.h"
+#endif
+
+
+// This atomic is written by a signal handler (SigInt). Traditionally that would
+// have had to be `volatile sig_atomic_t`, and annoyingly, `bool` was
+// technically not allowed there, even though in practice it would work.
+// The good thing with C++11 atomics is that we can use atomic<bool> instead
+// here and be on the actually supported path.
+static std::atomic<bool> s_disconnect { false };
+
+void SigInt( int )
+{
+    // Relaxed order is closest to a traditional `volatile` write.
+    // We don't need stronger ordering since this signal handler doesn't do
+    // anything else that would need to be ordered relatively to this.
+    s_disconnect.store(true, std::memory_order_relaxed);
+}
+
+static bool s_isStdoutATerminal = false;
+
+void InitIsStdoutATerminal() {
+#ifdef _WIN32
+    s_isStdoutATerminal = _isatty( fileno( stdout ) );
+#else
+    s_isStdoutATerminal = isatty( fileno( stdout ) );
+#endif
+}
+
+bool IsStdoutATerminal() { return s_isStdoutATerminal; }
+
+#define ANSI_RESET "\033[0m"
+#define ANSI_BOLD "\033[1m"
+#define ANSI_BLACK "\033[30m"
+#define ANSI_RED "\033[31m"
+#define ANSI_GREEN "\033[32m"
+#define ANSI_YELLOW "\033[33m"
+#define ANSI_BLUE "\033[34m"
+#define ANSI_MAGENTA "\033[35m"
+#define ANSI_CYAN "\033[36m"
+#define ANSI_ERASE_LINE "\033[2K"
+
+// Like printf, but if stdout is a terminal, prepends the output with
+// the given `ansiEscape` and appends ANSI_RESET.
+void AnsiPrintf( const char* ansiEscape, const char* format, ... ) {
+    if( IsStdoutATerminal() )
+    {
+        // Prepend ansiEscape and append ANSI_RESET.
+        char buf[256];
+        va_list args;
+        va_start( args, format );
+        vsnprintf( buf, sizeof buf, format, args );
+        va_end( args );
+        printf( "%s%s" ANSI_RESET, ansiEscape, buf );
+    }
+    else
+    {
+        // Just a normal printf.
+        va_list args;
+        va_start( args, format );
+        vfprintf( stdout, format, args );
+        va_end( args );
+    }
+}
+
+[[noreturn]] void Usage()
+{
+    printf( "Usage: capture -o output.tracy [-a address] [-p port] [-f] [-s seconds] [-m memlimit]\n" );
+    exit( 1 );
+}
+
+int main( int argc, char** argv )
+{
+#ifdef _WIN32
+    if( !AttachConsole( ATTACH_PARENT_PROCESS ) )
+    {
+        AllocConsole();
+        SetConsoleMode( GetStdHandle( STD_OUTPUT_HANDLE ), 0x07 );
+    }
+#endif
+
+    InitIsStdoutATerminal();
+
+    bool overwrite = false;
+    const char* address = "127.0.0.1";
+    const char* output = nullptr;
+    int port = 8086;
+    int seconds = -1;
+    int64_t memoryLimit = -1;
+
+    int c;
+    while( ( c = getopt( argc, argv, "a:o:p:fs:m:" ) ) != -1 )
+    {
+        switch( c )
+        {
+        case 'a':
+            address = optarg;
+            break;
+        case 'o':
+            output = optarg;
+            break;
+        case 'p':
+            port = atoi( optarg );
+            break;
+        case 'f':
+            overwrite = true;
+            break;
+        case 's':
+            seconds = atoi(optarg);
+            break;
+        case 'm':
+            memoryLimit = std::clamp( atoll( optarg ), 1ll, 999ll ) * tracy::GetPhysicalMemorySize() / 100;
+            break;
+        default:
+            Usage();
+            break;
+        }
+    }
+
+    if( !address || !output ) Usage();
+
+    struct stat st;
+    if( stat( output, &st ) == 0 && !overwrite )
+    {
+        printf( "Output file %s already exists! Use -f to force overwrite.\n", output );
+        return 4;
+    }
+
+    FILE* test = fopen( output, "wb" );
+    if( !test )
+    {
+        printf( "Cannot open output file %s for writing!\n", output );
+        return 5;
+    }
+    fclose( test );
+    unlink( output );
+
+    printf( "Connecting to %s:%i...", address, port );
+    fflush( stdout );
+    tracy::Worker worker( address, port, memoryLimit );
+    while( !worker.HasData() )
+    {
+        const auto handshake = worker.GetHandshakeStatus();
+        if( handshake == tracy::HandshakeProtocolMismatch )
+        {
+            printf( "\nThe client you are trying to connect to uses incompatible protocol version.\nMake sure you are using the same Tracy version on both client and server.\n" );
+            return 1;
+        }
+        if( handshake == tracy::HandshakeNotAvailable )
+        {
+            printf( "\nThe client you are trying to connect to is no longer able to sent profiling data,\nbecause another server was already connected to it.\nYou can do the following:\n\n  1. Restart the client application.\n  2. Rebuild the client application with on-demand mode enabled.\n" );
+            return 2;
+        }
+        if( handshake == tracy::HandshakeDropped )
+        {
+            printf( "\nThe client you are trying to connect to has disconnected during the initial\nconnection handshake. Please check your network configuration.\n" );
+            return 3;
+        }
+        std::this_thread::sleep_for( std::chrono::milliseconds( 100 ) );
+    }
+    printf( "\nQueue delay: %s\nTimer resolution: %s\n", tracy::TimeToString( worker.GetDelay() ), tracy::TimeToString( worker.GetResolution() ) );
+
+#ifdef _WIN32
+    signal( SIGINT, SigInt );
+#else
+    struct sigaction sigint, oldsigint;
+    memset( &sigint, 0, sizeof( sigint ) );
+    sigint.sa_handler = SigInt;
+    sigaction( SIGINT, &sigint, &oldsigint );
+#endif
+
+    const auto firstTime = worker.GetFirstTime();
+    auto& lock = worker.GetMbpsDataLock();
+
+    const auto t0 = std::chrono::high_resolution_clock::now();
+    while( worker.IsConnected() )
+    {
+        // Relaxed order is sufficient here because `s_disconnect` is only ever
+        // set by this thread or by the SigInt handler, and that handler does
+        // nothing else than storing `s_disconnect`.
+        if( s_disconnect.load( std::memory_order_relaxed ) )
+        {
+            worker.Disconnect();
+            // Relaxed order is sufficient because only this thread ever reads
+            // this value.
+            s_disconnect.store(false, std::memory_order_relaxed );
+            break;
+        }
+
+        lock.lock();
+        const auto mbps = worker.GetMbpsData().back();
+        const auto compRatio = worker.GetCompRatio();
+        const auto netTotal = worker.GetDataTransferred();
+        lock.unlock();
+
+        // Output progress info only if destination is a TTY to avoid bloating
+        // log files (so this is not just about usage of ANSI color codes).
+        if( IsStdoutATerminal() )
+        {
+            const char* unit = "Mbps";
+            float unitsPerMbps = 1.f;
+            if( mbps < 0.1f )
+            {
+                unit = "Kbps";
+                unitsPerMbps = 1000.f;
+            }
+            AnsiPrintf( ANSI_ERASE_LINE ANSI_CYAN ANSI_BOLD, "\r%7.2f %s", mbps * unitsPerMbps, unit );
+            printf( " /");
+            AnsiPrintf( ANSI_CYAN ANSI_BOLD, "%5.1f%%", compRatio * 100.f );
+            printf( " =");
+            AnsiPrintf( ANSI_YELLOW ANSI_BOLD, "%7.2f Mbps", mbps / compRatio );
+            printf( " | ");
+            AnsiPrintf( ANSI_YELLOW, "Tx: ");
+            AnsiPrintf( ANSI_GREEN, "%s", tracy::MemSizeToString( netTotal ) );
+            printf( " | ");
+            AnsiPrintf( ANSI_RED ANSI_BOLD, "%s", tracy::MemSizeToString( tracy::memUsage.load( std::memory_order_relaxed ) ) );
+            if( memoryLimit > 0 )
+            {
+                printf( " / " );
+                AnsiPrintf( ANSI_BLUE ANSI_BOLD, "%s", tracy::MemSizeToString( memoryLimit ) );
+            }
+            printf( " | ");
+            AnsiPrintf( ANSI_RED, "%s", tracy::TimeToString( worker.GetLastTime() - firstTime ) );
+            fflush( stdout );
+        }
+
+        std::this_thread::sleep_for( std::chrono::milliseconds( 100 ) );
+        if( seconds != -1 )
+        {
+            const auto dur = std::chrono::high_resolution_clock::now() - t0;
+            if( std::chrono::duration_cast<std::chrono::seconds>(dur).count() >= seconds )
+            {
+                // Relaxed order is sufficient because only this thread ever reads
+                // this value.
+                s_disconnect.store(true, std::memory_order_relaxed );
+            }
+        }
+    }
+    const auto t1 = std::chrono::high_resolution_clock::now();
+
+    const auto& failure = worker.GetFailureType();
+    if( failure != tracy::Worker::Failure::None )
+    {
+        AnsiPrintf( ANSI_RED ANSI_BOLD, "\nInstrumentation failure: %s", tracy::Worker::GetFailureString( failure ) );
+        auto& fd = worker.GetFailureData();
+        if( !fd.message.empty() )
+        {
+            printf( "\nContext: %s", fd.message.c_str() );
+        }
+        if( fd.callstack != 0 )
+        {
+            AnsiPrintf( ANSI_BOLD, "\nFailure callstack:\n" );
+            auto& cs = worker.GetCallstack( fd.callstack );
+            int fidx = 0;
+            for( auto& entry : cs )
+            {
+                auto frameData = worker.GetCallstackFrame( entry );
+                if( !frameData )
+                {
+                    printf( "%3i. %p\n", fidx++, (void*)worker.GetCanonicalPointer( entry ) );
+                }
+                else
+                {
+                    const auto fsz = frameData->size;
+                    for( uint8_t f=0; f<fsz; f++ )
+                    {
+                        const auto& frame = frameData->data[f];
+                        auto txt = worker.GetString( frame.name );
+
+                        if( fidx == 0 && f != fsz-1 )
+                        {
+                            auto test = tracy::s_tracyStackFrames;
+                            bool match = false;
+                            do
+                            {
+                                if( strcmp( txt, *test ) == 0 )
+                                {
+                                    match = true;
+                                    break;
+                                }
+                            }
+                            while( *++test );
+                            if( match ) continue;
+                        }
+
+                        if( f == fsz-1 )
+                        {
+                            printf( "%3i. ", fidx++ );
+                        }
+                        else
+                        {
+                            AnsiPrintf( ANSI_BLACK ANSI_BOLD, "inl. " );
+                        }
+                        AnsiPrintf( ANSI_CYAN, "%s  ", txt );
+                        txt = worker.GetString( frame.file );
+                        if( frame.line == 0 )
+                        {
+                            AnsiPrintf( ANSI_YELLOW, "(%s)", txt );
+                        }
+                        else
+                        {
+                            AnsiPrintf( ANSI_YELLOW, "(%s:%" PRIu32 ")", txt, frame.line );
+                        }
+                        if( frameData->imageName.Active() )
+                        {
+                            AnsiPrintf( ANSI_MAGENTA, " %s\n", worker.GetString( frameData->imageName ) );
+                        }
+                        else
+                        {
+                            printf( "\n" );
+                        }
+                    }
+                }
+            }
+        }
+    }
+
+    printf( "\nFrames: %" PRIu64 "\nTime span: %s\nZones: %s\nElapsed time: %s\nSaving trace...",
+        worker.GetFrameCount( *worker.GetFramesBase() ), tracy::TimeToString( worker.GetLastTime() - firstTime ), tracy::RealToString( worker.GetZoneCount() ),
+        tracy::TimeToString( std::chrono::duration_cast<std::chrono::nanoseconds>( t1 - t0 ).count() ) );
+    fflush( stdout );
+    auto f = std::unique_ptr<tracy::FileWrite>( tracy::FileWrite::Open( output, tracy::FileCompression::Zstd, 3, 4 ) );
+    if( f )
+    {
+        worker.Write( *f, false );
+        AnsiPrintf( ANSI_GREEN ANSI_BOLD, " done!\n" );
+        f->Finish();
+        const auto stats = f->GetCompressionStatistics();
+        printf( "Trace size %s (%.2f%% ratio)\n", tracy::MemSizeToString( stats.second ), 100.f * stats.second / stats.first );
+    }
+    else
+    {
+        AnsiPrintf( ANSI_RED ANSI_BOLD, " failed!\n");
+    }
+
+    return 0;
+}

subprojects/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()

subprojects/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()

subprojects/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 "*")

subprojects/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()

subprojects/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()

subprojects/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}")

subprojects/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})

subprojects/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;
+}

subprojects/tracy/dtl/Diff.hpp

@@ -0,0 +1,706 @@
+/**
+   dtl -- Diff Template Library
+   
+   In short, Diff Template Library is distributed under so called "BSD license",
+   
+   Copyright (c) 2015 Tatsuhiko Kubo <cubicdaiya@gmail.com>
+   All rights reserved.
+   
+   Redistribution and use in source and binary forms, with or without modification,
+   are permitted provided that the following conditions are met:
+   
+   * Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+   
+   * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+   
+   * Neither the name of the authors nor the names of its contributors
+   may be used to endorse or promote products derived from this software 
+   without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+   TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+/* If you use this library, you must include dtl.hpp only. */
+
+#ifndef DTL_DIFF_H
+#define DTL_DIFF_H
+
+namespace dtl {
+    
+    /**
+     * diff class template
+     * sequence must support random_access_iterator.
+     */
+    template <typename elem, typename sequence = vector< elem >, typename comparator = Compare< elem > >
+    class Diff
+    {
+    private :
+        dtl_typedefs(elem, sequence)
+        sequence           A;
+        sequence           B;
+        size_t             M;
+        size_t             N;
+        size_t             delta;
+        size_t             offset;
+        long long          *fp;
+        long long          editDistance;
+        Lcs< elem >        lcs;
+        Ses< elem >        ses;
+        editPath           path;
+        editPathCordinates pathCordinates;
+        bool               swapped;
+        bool               huge;
+        bool               trivial;
+        bool               editDistanceOnly;
+        uniHunkVec         uniHunks;
+        comparator         cmp;
+        long long          ox;
+        long long          oy;
+    public :
+        Diff () {}
+        
+        Diff (const sequence& a, 
+              const sequence& b) : A(a), B(b), ses(false) {
+            init();
+        }
+        
+        Diff (const sequence& a,
+              const sequence& b,
+              bool deletesFirst) : A(a), B(b), ses(deletesFirst) {
+            init();
+        }
+        
+        Diff (const sequence& a,
+              const sequence& b, 
+              const comparator& comp) : A(a), B(b), ses(false), cmp(comp) {
+            init();
+        }
+        
+        Diff (const sequence& a, 
+              const sequence& b, 
+              bool deleteFirst,
+              const comparator& comp) : A(a), B(b), ses(deleteFirst), cmp(comp) {
+            init();
+        }
+        
+        ~Diff() {}
+        
+        long long getEditDistance () const {
+            return editDistance;
+        }
+        
+        Lcs< elem > getLcs () const {
+            return lcs;
+        }
+        
+        elemVec getLcsVec () const {
+            return lcs.getSequence();
+        }
+        
+        Ses< elem > getSes () const {
+            return ses;
+        }
+        
+        uniHunkVec getUniHunks () const {
+            return uniHunks;
+        }
+        
+        /* These should be deprecated */
+        bool isHuge () const {
+            return huge;
+        }
+        
+        void onHuge () {
+            this->huge = true;
+        }
+        
+        void offHuge () {
+            this->huge = false;
+        }
+        
+        bool isUnserious () const {
+            return trivial;
+        }
+        
+        void onUnserious () {
+            this->trivial = true;
+        }
+        
+        void offUnserious () {
+            this->trivial = false;
+        }
+        
+        void onOnlyEditDistance () {
+            this->editDistanceOnly = true;
+        }
+        
+        /* These are the replacements for the above */
+        bool hugeEnabled () const {
+            return huge;
+        }
+        
+        void enableHuge () {
+            this->huge = true;
+        }
+        
+        void disableHuge () {
+            this->huge = false;
+        }
+        
+        bool trivialEnabled () const {
+            return trivial;
+        }
+        
+        void enableTrivial () {
+            this->trivial = true;
+        }
+        
+        void disableTrivial () {
+            this->trivial = false;
+        }
+        
+        void editDistanceOnlyEnabled () {
+            this->editDistanceOnly = true;
+        }
+        
+        /**
+         * patching with Unified Format Hunks
+         */
+        sequence uniPatch (const sequence& seq) {
+            elemList        seqLst(seq.begin(), seq.end());
+            sesElemVec      shunk;
+            sesElemVec_iter vsesIt;
+            elemList_iter   lstIt         = seqLst.begin();
+            long long       inc_dec_total = 0;
+            long long       gap           = 1;
+            for (uniHunkVec_iter it=uniHunks.begin();it!=uniHunks.end();++it) {
+                joinSesVec(shunk, it->common[0]);
+                joinSesVec(shunk, it->change);
+                joinSesVec(shunk, it->common[1]);
+                it->a         += inc_dec_total;
+                inc_dec_total += it->inc_dec_count;
+                for (long long i=0;i<it->a - gap;++i) {
+                    ++lstIt;
+                }
+                gap = it->a + it->b + it->inc_dec_count;
+                vsesIt = shunk.begin();
+                while (vsesIt!=shunk.end()) {
+                    switch (vsesIt->second.type) {
+                    case SES_ADD :
+                        seqLst.insert(lstIt, vsesIt->first);
+                        break;
+                    case SES_DELETE :
+                        if (lstIt != seqLst.end()) {
+                            lstIt = seqLst.erase(lstIt);
+                        }
+                        break;
+                    case SES_COMMON :
+                        if (lstIt != seqLst.end()) {
+                            ++lstIt;
+                        }
+                        break;
+                    default :
+                        // no fall-through
+                        break;
+                    }
+                    ++vsesIt;
+                }
+                shunk.clear();
+            }
+            
+            sequence patchedSeq(seqLst.begin(), seqLst.end());
+            return patchedSeq;
+        }
+        
+        /**
+         * patching with Shortest Edit Script (SES)
+         */
+        sequence patch (const sequence& seq) const {
+            sesElemVec    sesSeq = ses.getSequence();
+            elemList      seqLst(seq.begin(), seq.end());
+            elemList_iter lstIt  = seqLst.begin();
+            for (sesElemVec_iter sesIt=sesSeq.begin();sesIt!=sesSeq.end();++sesIt) {
+                switch (sesIt->second.type) {
+                case SES_ADD :
+                    seqLst.insert(lstIt, sesIt->first);
+                    break;
+                case SES_DELETE :
+                    lstIt = seqLst.erase(lstIt);
+                    break;
+                case SES_COMMON :
+                    ++lstIt;
+                    break;
+                default :
+                    // no through
+                    break;
+                }
+            }
+            sequence patchedSeq(seqLst.begin(), seqLst.end());
+            return patchedSeq;
+        }
+        
+        /**
+         * compose Longest Common Subsequence and Shortest Edit Script.
+         * The algorithm implemented here is based on "An O(NP) Sequence Comparison Algorithm"
+         * described by Sun Wu, Udi Manber and Gene Myers
+         */
+        void compose() {
+            
+            if (isHuge()) {
+                pathCordinates.reserve(MAX_CORDINATES_SIZE);
+            }
+            ox = 0;
+            oy = 0;
+            long long p = -1;
+            fp = new long long[M + N + 3];
+            fill(&fp[0], &fp[M + N + 3], -1);
+            path = editPath(M + N + 3);
+            fill(path.begin(), path.end(), -1);
+        ONP:
+            do {
+                ++p;
+                for (long long k=-p;k<=static_cast<long long>(delta)-1;++k) {
+                    fp[k+offset] = snake(k, fp[k-1+offset]+1, fp[k+1+offset]);
+                }
+                for (long long k=static_cast<long long>(delta)+p;k>=static_cast<long long>(delta)+1;--k) {
+                    fp[k+offset] = snake(k, fp[k-1+offset]+1, fp[k+1+offset]);
+                }
+                fp[delta+offset] = snake(static_cast<long long>(delta), fp[delta-1+offset]+1, fp[delta+1+offset]);
+            } while (fp[delta+offset] != static_cast<long long>(N) && pathCordinates.size() < MAX_CORDINATES_SIZE);
+            
+            editDistance += static_cast<long long>(delta) + 2 * p;
+            long long r = path[delta+offset];
+            P cordinate;
+            editPathCordinates epc(0);
+            
+            // recording edit distance only
+            if (editDistanceOnly) {
+                delete[] this->fp;
+                return;
+            }
+            
+            while(r != -1) {
+                cordinate.x = pathCordinates[(size_t)r].x;
+                cordinate.y = pathCordinates[(size_t)r].y;
+                epc.push_back(cordinate);
+                r = pathCordinates[(size_t)r].k;
+            }
+            
+            // record Longest Common Subsequence & Shortest Edit Script
+            if (!recordSequence(epc)) {
+                pathCordinates.resize(0);
+                epc.resize(0);
+                p = -1;
+                goto ONP;
+            }
+            delete[] this->fp;
+        }
+
+        /**
+         * print difference between A and B as an SES
+         */
+        template < typename stream >
+        void printSES (stream& out) const {
+            sesElemVec ses_v = ses.getSequence();
+            for_each(ses_v.begin(), ses_v.end(), ChangePrinter< sesElem, stream >(out));
+        }
+        
+        void printSES (ostream& out = cout) const {
+            printSES< ostream >(out);
+        }
+        
+        /**
+         * print differences given an SES
+         */
+        template < typename stream >
+        static void printSES (const Ses< elem >& s, stream& out) {
+            sesElemVec ses_v = s.getSequence();
+            for_each(ses_v.begin(), ses_v.end(), ChangePrinter< sesElem, stream >(out));
+        }
+        
+        static void printSES (const Ses< elem >& s, ostream& out = cout) {
+            printSES< ostream >(s, out);
+        }
+
+        /**
+         * print difference between A and B as an SES with custom printer
+         */
+        template < typename stream, template < typename SEET, typename STRT > class PT >
+        void printSES (stream& out) const {
+            sesElemVec ses_v = ses.getSequence ();
+            for_each (ses_v.begin (), ses_v.end(), PT < sesElem, stream > (out));
+        }
+
+        /**
+         * store difference between A and B as an SES with custom storage
+         */
+        template < typename storedData, template < typename SEET, typename STRT > class ST >
+        void storeSES(storedData& sd) const {
+            sesElemVec ses_v = ses.getSequence();
+            for_each(ses_v.begin(), ses_v.end(), ST < sesElem, storedData >(sd));
+        }
+        
+        /**
+         * print difference between A and B in the Unified Format
+         */
+        template < typename stream >
+        void printUnifiedFormat (stream& out) const {
+            for_each(uniHunks.begin(), uniHunks.end(), UniHunkPrinter< sesElem, stream >(out));
+        }
+        
+        void printUnifiedFormat (ostream& out = cout) const {
+            printUnifiedFormat< ostream >(out);
+        }
+        
+        /**
+         * print unified format difference with given unified format hunks
+         */
+        template < typename stream >
+        static void printUnifiedFormat (const uniHunkVec& hunks, stream& out) {
+            for_each(hunks.begin(), hunks.end(), UniHunkPrinter< sesElem >(out));
+        }
+
+        static void printUnifiedFormat (const uniHunkVec& hunks, ostream& out = cout) {
+            printUnifiedFormat< ostream >(hunks, out);
+        }
+
+        /**
+         * compose Unified Format Hunks from Shortest Edit Script
+         */
+        void composeUnifiedHunks () {
+            sesElemVec         common[2];
+            sesElemVec         change;
+            sesElemVec         ses_v  = ses.getSequence();
+            long long          l_cnt  = 1;
+            long long          length = distance(ses_v.begin(), ses_v.end());
+            long long          middle = 0;
+            bool               isMiddle, isAfter;
+            elemInfo           einfo;
+            long long          a, b, c, d;        // @@ -a,b +c,d @@
+            long long          inc_dec_count = 0;
+            uniHunk< sesElem > hunk;
+            sesElemVec         adds;
+            sesElemVec         deletes;
+            
+            isMiddle = isAfter = false;
+            a = b = c = d = 0;
+            
+            for (sesElemVec_iter it=ses_v.begin();it!=ses_v.end();++it, ++l_cnt) {
+                einfo = it->second;
+                switch (einfo.type) {
+                case SES_ADD :
+                    middle = 0;
+                    ++inc_dec_count;
+                    adds.push_back(*it);
+                    if (!isMiddle)       isMiddle = true;
+                    if (isMiddle)        ++d;
+                    if (l_cnt >= length) {
+                        joinSesVec(change, deletes);
+                        joinSesVec(change, adds);
+                        isAfter = true;
+                    }
+                    break;
+                case SES_DELETE :
+                    middle = 0;
+                    --inc_dec_count;
+                    deletes.push_back(*it);
+                    if (!isMiddle)       isMiddle = true;
+                    if (isMiddle)        ++b;
+                    if (l_cnt >= length) {
+                        joinSesVec(change, deletes);
+                        joinSesVec(change, adds);
+                        isAfter = true;
+                    }
+                    break;
+                case SES_COMMON :
+                    ++b;++d;
+                    if (common[1].empty() && adds.empty() && deletes.empty() && change.empty()) {
+                        if (static_cast<long long>(common[0].size()) < DTL_CONTEXT_SIZE) {
+                            if (a == 0 && c == 0) {
+                                if (!wasSwapped()) {
+                                    a = einfo.beforeIdx;
+                                    c = einfo.afterIdx;
+                                } else {
+                                    a = einfo.afterIdx;
+                                    c = einfo.beforeIdx;
+                                }
+                            }
+                            common[0].push_back(*it);
+                        } else {
+                            rotate(common[0].begin(), common[0].begin() + 1, common[0].end());
+                            common[0].pop_back();
+                            common[0].push_back(*it);
+                            ++a;++c;
+                            --b;--d;
+                        }
+                    }
+                    if (isMiddle && !isAfter) {
+                        ++middle;
+                        joinSesVec(change, deletes);
+                        joinSesVec(change, adds);
+                        change.push_back(*it);
+                        if (middle >= DTL_SEPARATE_SIZE || l_cnt >= length) {
+                            isAfter = true;
+                        }
+                        adds.clear();
+                        deletes.clear();
+                    }
+                    break;
+                default :
+                    // no through
+                    break;
+                }
+                // compose unified format hunk
+                if (isAfter && !change.empty()) {
+                    sesElemVec_iter cit = it;
+                    long long       cnt = 0;
+                    for (long long i=0;i<DTL_SEPARATE_SIZE && (cit != ses_v.end());++i, ++cit) {
+                        if (cit->second.type == SES_COMMON) {
+                            ++cnt;
+                        }
+                    }
+                    if (cnt < DTL_SEPARATE_SIZE && l_cnt < length) {
+                        middle = 0;
+                        isAfter = false;
+                        continue;
+                    }
+                    if (static_cast<long long>(common[0].size()) >= DTL_SEPARATE_SIZE) {
+                        long long c0size = static_cast<long long>(common[0].size());
+                        rotate(common[0].begin(), 
+                               common[0].begin() + (size_t)c0size - DTL_SEPARATE_SIZE, 
+                               common[0].end());
+                        for (long long i=0;i<c0size - DTL_SEPARATE_SIZE;++i) {
+                            common[0].pop_back();
+                        }
+                        a += c0size - DTL_SEPARATE_SIZE;
+                        c += c0size - DTL_SEPARATE_SIZE;
+                    }
+                    if (a == 0) ++a;
+                    if (c == 0) ++c;
+                    if (wasSwapped()) swap(a, c);
+                    hunk.a = a;
+                    hunk.b = b;
+                    hunk.c = c;
+                    hunk.d = d;
+                    hunk.common[0]     = common[0];
+                    hunk.change        = change;
+                    hunk.common[1]     = common[1];
+                    hunk.inc_dec_count = inc_dec_count;
+                    uniHunks.push_back(hunk);
+                    isMiddle = false;
+                    isAfter  = false;
+                    common[0].clear();
+                    common[1].clear();
+                    adds.clear();
+                    deletes.clear();
+                    change.clear();
+                    a = b = c = d = middle = inc_dec_count = 0;
+                }
+            }
+        }
+        
+        /**
+         * compose ses from stream
+         */
+        template <typename stream>
+        static Ses< elem > composeSesFromStream (stream& st)
+        {
+            elem        line;
+            Ses< elem > ret;
+            long long   x_idx, y_idx;
+            x_idx = y_idx = 1;
+            while (getline(st, line)) {
+                elem mark(line.begin(), line.begin() + 1);
+                elem e(line.begin() + 1, line.end());
+                if (mark == SES_MARK_DELETE) {
+                    ret.addSequence(e, x_idx, 0, SES_DELETE);
+                    ++x_idx;
+                } else if (mark == SES_MARK_ADD) {
+                    ret.addSequence(e, y_idx, 0, SES_ADD);
+                    ++y_idx;
+                } else if (mark == SES_MARK_COMMON) {
+                    ret.addSequence(e, x_idx, y_idx, SES_COMMON);
+                    ++x_idx;
+                    ++y_idx;
+                }
+            }
+            return ret;
+        }
+        
+    private :
+        /**
+         * initialize
+         */
+        void init () {
+            M = distance(A.begin(), A.end());
+            N = distance(B.begin(), B.end());
+            if (M < N) {
+                swapped = false;
+            } else {
+                swap(A, B);
+                swap(M, N);
+                swapped = true;
+            }
+            editDistance     = 0;
+            delta            = N - M;
+            offset           = M + 1;
+            huge             = false;
+            trivial          = false;
+            editDistanceOnly = false;
+            fp               = NULL;
+        }
+        
+        /**
+         * search shortest path and record the path
+         */
+        long long snake(const long long& k, const long long& above, const long long& below) {
+            long long r = above > below ? path[(size_t)k-1+offset] : path[(size_t)k+1+offset];
+            long long y = max(above, below);
+            long long x = y - k;
+            while ((size_t)x < M && (size_t)y < N && (swapped ? cmp.impl(B[(size_t)y], A[(size_t)x]) : cmp.impl(A[(size_t)x], B[(size_t)y]))) {
+                ++x;++y;
+            }
+            
+            path[(size_t)k+offset] = static_cast<long long>(pathCordinates.size());
+            if (!editDistanceOnly) {
+                P p;
+                p.x = x;p.y = y;p.k = r;
+                pathCordinates.push_back(p);      
+            }
+            return y;
+        }
+        
+        /**
+         * record SES and LCS
+         */
+        bool recordSequence (const editPathCordinates& v) {
+            sequence_const_iter x(A.begin());
+            sequence_const_iter y(B.begin());
+            long long           x_idx,  y_idx;  // line number for Unified Format
+            long long           px_idx, py_idx; // cordinates
+            bool                complete = false;
+            x_idx  = y_idx  = 1;
+            px_idx = py_idx = 0;
+            for (size_t i=v.size()-1;!complete;--i) {
+                while(px_idx < v[i].x || py_idx < v[i].y) {
+                    if (v[i].y - v[i].x > py_idx - px_idx) {
+                        if (!wasSwapped()) {
+                            ses.addSequence(*y, 0, y_idx + oy, SES_ADD);
+                        } else {
+                            ses.addSequence(*y, y_idx + oy, 0, SES_DELETE);
+                        }
+                        ++y;
+                        ++y_idx;
+                        ++py_idx;
+                    } else if (v[i].y - v[i].x < py_idx - px_idx) {
+                        if (!wasSwapped()) {
+                            ses.addSequence(*x, x_idx + ox, 0, SES_DELETE);
+                        } else {
+                            ses.addSequence(*x, 0, x_idx + ox, SES_ADD);
+                        }
+                        ++x;
+                        ++x_idx;
+                        ++px_idx;
+                    } else {
+                        if (!wasSwapped()) {
+                            lcs.addSequence(*x);
+                            ses.addSequence(*x, x_idx + ox, y_idx + oy, SES_COMMON);
+                        } else {
+                            lcs.addSequence(*y);
+                            ses.addSequence(*y, y_idx + oy, x_idx + ox, SES_COMMON);
+                        }
+                        ++x;
+                        ++y;
+                        ++x_idx;
+                        ++y_idx;
+                        ++px_idx;
+                        ++py_idx;
+                    }
+                }
+                if (i == 0) complete = true;
+            }
+            
+            if (x_idx > static_cast<long long>(M) && y_idx > static_cast<long long>(N)) {
+                // all recording succeeded
+            } else {
+                // trivial difference
+                if (trivialEnabled()) {
+                    if (!wasSwapped()) {
+                        recordOddSequence(x_idx, M, x, SES_DELETE);
+                        recordOddSequence(y_idx, N, y, SES_ADD);
+                    } else {
+                        recordOddSequence(x_idx, M, x, SES_ADD);
+                        recordOddSequence(y_idx, N, y, SES_DELETE);
+                    }
+                    return true;
+                }
+                
+                // nontrivial difference
+                sequence A_(A.begin() + (size_t)x_idx - 1, A.end());
+                sequence B_(B.begin() + (size_t)y_idx - 1, B.end());
+                A        = A_;
+                B        = B_;
+                M        = distance(A.begin(), A.end());
+                N        = distance(B.begin(), B.end());
+                delta    = N - M;
+                offset   = M + 1;
+                delete[] fp;
+                fp = new long long[M + N + 3];
+                fill(&fp[0], &fp[M + N + 3], -1);
+                fill(path.begin(), path.end(), -1);
+                ox = x_idx - 1;
+                oy = y_idx - 1;
+                return false;
+            }
+            return true;
+        }
+        
+        /**
+         * record odd sequence in SES
+         */
+        void inline recordOddSequence (long long idx, long long length, sequence_const_iter it, const edit_t et) {
+            while(idx < length){
+                ses.addSequence(*it, idx, 0, et);
+                ++it;
+                ++idx;
+                ++editDistance;
+            }
+            ses.addSequence(*it, idx, 0, et);
+            ++editDistance;
+        }
+        
+        /**
+         * join SES vectors
+         */
+        void inline joinSesVec (sesElemVec& s1, sesElemVec& s2) const {
+            if (!s2.empty()) {
+                for (sesElemVec_iter vit=s2.begin();vit!=s2.end();++vit) {
+                    s1.push_back(*vit);
+                }
+            }      
+        }
+        
+        /**
+         * check if the sequences have been swapped
+         */
+        bool inline wasSwapped () const {
+            return swapped;
+        }
+
+    };
+}
+
+#endif // DTL_DIFF_H

subprojects/tracy/dtl/Diff3.hpp

@@ -0,0 +1,245 @@
+/**
+   dtl -- Diff Template Library
+   
+   In short, Diff Template Library is distributed under so called "BSD license",
+   
+   Copyright (c) 2015 Tatsuhiko Kubo <cubicdaiya@gmail.com>
+   All rights reserved.
+   
+   Redistribution and use in source and binary forms, with or without modification,
+   are permitted provided that the following conditions are met:
+   
+   * Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+   
+   * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+   
+   * Neither the name of the authors nor the names of its contributors
+   may be used to endorse or promote products derived from this software 
+   without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+   TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+/* If you use this library, you must include dtl.hpp only. */
+
+#ifndef DTL_DIFF3_H
+#define DTL_DIFF3_H
+
+namespace dtl {
+    
+    /**
+     * diff3 class template
+     * sequence must support random_access_iterator.
+     */
+    template <typename elem, typename sequence = vector< elem >, typename comparator = Compare< elem > >
+    class Diff3
+    {
+    private:
+        dtl_typedefs(elem, sequence)
+        sequence                           A;
+        sequence                           B;
+        sequence                           C;
+        sequence                           S;
+        Diff< elem, sequence, comparator > diff_ba;
+        Diff< elem, sequence, comparator > diff_bc;
+        bool                               conflict;
+        elem                               csepabegin;
+        elem                               csepa;
+        elem                               csepaend;
+    public :
+        Diff3 () {}
+        Diff3 (const sequence& a, 
+               const sequence& b, 
+               const sequence& c) : A(a), B(b), C(c), 
+                                    diff_ba(b, a), diff_bc(b, c), 
+                                    conflict(false) {} 
+        
+        ~Diff3 () {}
+        
+        bool isConflict () const {
+            return conflict;
+        }
+        
+        sequence getMergedSequence () const {
+            return S;
+        }
+        
+        /**
+         * merge changes B and C into A
+         */
+        bool merge () {
+            if (diff_ba.getEditDistance() == 0) {     // A == B
+                if (diff_bc.getEditDistance() == 0) { // A == B == C
+                    S = B;
+                    return true;
+                }
+                S = C;
+                return true;
+            } else {                                  // A != B
+                if (diff_bc.getEditDistance() == 0) { // A != B == C
+                    S = A;                              
+                    return true;
+                } else {                              // A != B != C
+                    S = merge_();
+                    if (isConflict()) {               // conflict occured
+                        return false;
+                    }
+                }
+            }
+            return true;
+        }
+        
+        /**
+         * compose differences
+         */
+        void compose () {
+            diff_ba.compose();
+            diff_bc.compose();
+        }
+        
+    private :
+        /**
+         * merge implementation
+         */
+        sequence merge_ () {
+            elemVec         seq;
+            Ses< elem >     ses_ba   = diff_ba.getSes();
+            Ses< elem >     ses_bc   = diff_bc.getSes();
+            sesElemVec      ses_ba_v = ses_ba.getSequence();
+            sesElemVec      ses_bc_v = ses_bc.getSequence();
+            sesElemVec_iter ba_it    = ses_ba_v.begin();
+            sesElemVec_iter bc_it    = ses_bc_v.begin();
+            sesElemVec_iter ba_end   = ses_ba_v.end();
+            sesElemVec_iter bc_end   = ses_bc_v.end();
+            
+            while (!isEnd(ba_end, ba_it) || !isEnd(bc_end, bc_it)) {
+                while (true) {
+                    if (!isEnd(ba_end, ba_it)            && 
+                        !isEnd(bc_end, bc_it)            &&
+                        ba_it->first == bc_it->first     && 
+                        ba_it->second.type == SES_COMMON && 
+                        bc_it->second.type == SES_COMMON) {
+                        // do nothing
+                    } else {
+                        break;
+                    }
+                    if      (!isEnd(ba_end, ba_it)) seq.push_back(ba_it->first);
+                    else if (!isEnd(bc_end, bc_it)) seq.push_back(bc_it->first);
+                    forwardUntilEnd(ba_end, ba_it);
+                    forwardUntilEnd(bc_end, bc_it);
+                }
+                if (isEnd(ba_end, ba_it) || isEnd(bc_end, bc_it)) break;
+                if (   ba_it->second.type == SES_COMMON 
+                       && bc_it->second.type == SES_DELETE) {
+                    forwardUntilEnd(ba_end, ba_it);
+                    forwardUntilEnd(bc_end, bc_it);
+                } else if (ba_it->second.type == SES_COMMON && 
+                           bc_it->second.type == SES_ADD) {
+                    seq.push_back(bc_it->first);
+                    forwardUntilEnd(bc_end, bc_it);
+                } else if (ba_it->second.type == SES_DELETE && 
+                           bc_it->second.type == SES_COMMON) {
+                    forwardUntilEnd(ba_end, ba_it);
+                    forwardUntilEnd(bc_end, bc_it);
+                } else if (ba_it->second.type == SES_DELETE && 
+                           bc_it->second.type == SES_DELETE) {
+                    if (ba_it->first == bc_it->first) {
+                        forwardUntilEnd(ba_end, ba_it);
+                        forwardUntilEnd(bc_end, bc_it);
+                    } else {
+                        // conflict
+                        conflict = true;
+                        return B;
+                    }
+                } else if (ba_it->second.type == SES_DELETE && 
+                           bc_it->second.type == SES_ADD) {
+                    // conflict
+                    conflict = true;
+                    return B;
+                } else if (ba_it->second.type == SES_ADD && 
+                           bc_it->second.type == SES_COMMON) {
+                    seq.push_back(ba_it->first);
+                    forwardUntilEnd(ba_end, ba_it);
+                } else if (ba_it->second.type == SES_ADD && 
+                           bc_it->second.type == SES_DELETE) {
+                    // conflict
+                    conflict = true;
+                    return B;
+                } else if (ba_it->second.type == SES_ADD && 
+                           bc_it->second.type == SES_ADD) {
+                    if (ba_it->first == bc_it->first) {
+                        seq.push_back(ba_it->first);
+                        forwardUntilEnd(ba_end, ba_it);
+                        forwardUntilEnd(bc_end, bc_it);
+                    } else {
+                        // conflict
+                        conflict = true;
+                        return B;
+                    }
+                }
+            }
+            
+            if (isEnd(ba_end, ba_it)) {
+                addDecentSequence(bc_end, bc_it, seq);
+            } else if (isEnd(bc_end, bc_it)) {
+                addDecentSequence(ba_end, ba_it, seq);
+            }
+            
+            sequence mergedSeq(seq.begin(), seq.end());
+            return mergedSeq;
+        }
+        
+        /**
+         * join elem vectors
+         */
+        void inline joinElemVec (elemVec& s1, elemVec& s2) const {
+            if (!s2.empty()) {
+                for (elemVec_iter vit=s2.begin();vit!=s2.end();++vit) {
+                    s1.push_back(*vit);
+                }
+            }
+        }
+        
+        /**
+         * check if sequence is at end
+         */
+        template <typename T_iter>
+        bool inline isEnd (const T_iter& end, const T_iter& it) const {
+            return it == end ? true : false;
+        }
+        
+        /**
+         * increment iterator until iterator is at end
+         */
+        template <typename T_iter>
+        void inline forwardUntilEnd (const T_iter& end, T_iter& it) const {
+            if (!isEnd(end, it)) ++it;
+        }
+        
+        /**
+         * add elements whose SES's type is ADD
+         */
+        void inline addDecentSequence (const sesElemVec_iter& end, sesElemVec_iter& it, elemVec& seq) const {
+            while (!isEnd(end, it)) {
+                if (it->second.type == SES_ADD) seq.push_back(it->first);
+                ++it;
+            }      
+        }
+        
+    };
+}
+
+#endif // DTL_DIFF3_H

subprojects/tracy/dtl/Lcs.hpp

@@ -0,0 +1,55 @@
+/**
+   dtl -- Diff Template Library
+   
+   In short, Diff Template Library is distributed under so called "BSD license",
+   
+   Copyright (c) 2015 Tatsuhiko Kubo <cubicdaiya@gmail.com>
+   All rights reserved.
+   
+   Redistribution and use in source and binary forms, with or without modification,
+   are permitted provided that the following conditions are met:
+   
+   * Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+   
+   * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+   
+   * Neither the name of the authors nor the names of its contributors
+   may be used to endorse or promote products derived from this software 
+   without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+   TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+/* If you use this library, you must include dtl.hpp only. */
+
+#ifndef DTL_LCS_H
+#define DTL_LCS_H
+
+namespace dtl {
+    
+    /**
+     * Longest Common Subsequence template class
+     */
+    template <typename elem>
+    class Lcs : public Sequence< elem >
+    {
+    public :
+        Lcs ()  {}
+        ~Lcs () {}
+    };
+}
+
+#endif // DTL_LCS_H

subprojects/tracy/dtl/Sequence.hpp

@@ -0,0 +1,65 @@
+/**
+   dtl -- Diff Template Library
+   
+   In short, Diff Template Library is distributed under so called "BSD license",
+   
+   Copyright (c) 2015 Tatsuhiko Kubo <cubicdaiya@gmail.com>
+   All rights reserved.
+   
+   Redistribution and use in source and binary forms, with or without modification,
+   are permitted provided that the following conditions are met:
+   
+   * Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+   
+   * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+   
+   * Neither the name of the authors nor the names of its contributors
+   may be used to endorse or promote products derived from this software 
+   without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+   TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+/* If you use this library, you must include dtl.hpp only. */
+
+#ifndef DTL_SEQUENCE_H
+#define DTL_SEQUENCE_H
+
+namespace dtl {
+    
+    /**  
+     * sequence class template
+     */
+    template <typename elem>
+    class Sequence
+    {
+    public :
+        typedef vector< elem > elemVec;
+        Sequence () {}
+        virtual ~Sequence () {}
+        
+        elemVec getSequence () const {
+            return sequence;
+        }
+        void addSequence (elem e) {
+            sequence.push_back(e);
+        }
+    protected :
+        elemVec sequence;
+    };
+}
+
+#endif // DTL_SEQUENCE_H

subprojects/tracy/dtl/Ses.hpp

@@ -0,0 +1,132 @@
+/**
+   dtl -- Diff Template Library
+   
+   In short, Diff Template Library is distributed under so called "BSD license",
+   
+   Copyright (c) 2015 Tatsuhiko Kubo <cubicdaiya@gmail.com>
+   All rights reserved.
+   
+   Redistribution and use in source and binary forms, with or without modification,
+   are permitted provided that the following conditions are met:
+   
+   * Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+   
+   * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+   
+   * Neither the name of the authors nor the names of its contributors
+   may be used to endorse or promote products derived from this software 
+   without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+   TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+/* If you use this library, you must include dtl.hpp only. */
+
+#ifndef DTL_SES_H
+#define DTL_SES_H
+
+namespace dtl {
+    
+    /**
+     * Shortest Edit Script template class
+     */
+    template <typename elem>
+    class Ses : public Sequence< elem >
+    {
+    private :
+        typedef pair< elem, elemInfo > sesElem;
+        typedef vector< sesElem >      sesElemVec;
+    public :
+        
+        Ses () : onlyAdd(true), onlyDelete(true), onlyCopy(true), deletesFirst(false) {
+            nextDeleteIdx = 0;
+        }
+        Ses (bool moveDel) : onlyAdd(true), onlyDelete(true), onlyCopy(true), deletesFirst(moveDel) {
+            nextDeleteIdx = 0;
+        }
+        ~Ses () {}
+        
+        bool isOnlyAdd () const {
+            return onlyAdd;
+        }
+        
+        bool isOnlyDelete () const {
+            return onlyDelete;
+        }
+        
+        bool isOnlyCopy () const {
+            return onlyCopy;
+        }
+        
+        bool isOnlyOneOperation () const {
+            return isOnlyAdd() || isOnlyDelete() || isOnlyCopy();
+        }
+        
+        bool isChange () const {
+            return !onlyCopy;
+        }
+        
+        using Sequence< elem >::addSequence;
+        void addSequence (elem e, long long beforeIdx, long long afterIdx, const edit_t type) {
+            elemInfo info;
+            info.beforeIdx = beforeIdx;
+            info.afterIdx  = afterIdx;
+            info.type      = type;
+            sesElem pe(e, info);
+            if (!deletesFirst) {
+                sequence.push_back(pe);
+            }
+            switch (type) {
+            case SES_DELETE:
+                onlyCopy   = false;
+                onlyAdd    = false;
+                if (deletesFirst) {
+                    sequence.insert(sequence.begin() + nextDeleteIdx, pe);
+                    nextDeleteIdx++;
+                }
+                break;
+            case SES_COMMON:
+                onlyAdd    = false;
+                onlyDelete = false;
+                if (deletesFirst) {
+                    sequence.push_back(pe);
+                    nextDeleteIdx = sequence.size();
+                }
+                break;
+            case SES_ADD:
+                onlyDelete = false;
+                onlyCopy   = false;
+                if (deletesFirst) {
+                    sequence.push_back(pe);
+                }
+                break;
+            }
+        }
+        
+        sesElemVec getSequence () const {
+            return sequence;
+        }
+    private :
+        sesElemVec sequence;
+        bool       onlyAdd;
+        bool       onlyDelete;
+        bool       onlyCopy;
+        bool       deletesFirst;
+        size_t     nextDeleteIdx;
+    };
+}
+
+#endif // DTL_SES_H

subprojects/tracy/dtl/dtl.hpp

@@ -0,0 +1,47 @@
+/**
+   dtl -- Diff Template Library
+   
+   In short, Diff Template Library is distributed under so called "BSD license",
+   
+   Copyright (c) 2015 Tatsuhiko Kubo <cubicdaiya@gmail.com>
+   All rights reserved.
+   
+   Redistribution and use in source and binary forms, with or without modification,
+   are permitted provided that the following conditions are met:
+   
+   * Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+   
+   * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+   
+   * Neither the name of the authors nor the names of its contributors
+   may be used to endorse or promote products derived from this software 
+   without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+   TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+#ifndef DTL_H
+#define DTL_H
+
+#include "variables.hpp"
+#include "functors.hpp"
+#include "Sequence.hpp"
+#include "Lcs.hpp"
+#include "Ses.hpp"
+#include "Diff.hpp"
+#include "Diff3.hpp"
+
+#endif // DTL_H

subprojects/tracy/dtl/functors.hpp

@@ -0,0 +1,151 @@
+/**
+   dtl -- Diff Template Library
+   
+   In short, Diff Template Library is distributed under so called "BSD license",
+   
+   Copyright (c) 2015 Tatsuhiko Kubo <cubicdaiya@gmail.com>
+   All rights reserved.
+   
+   Redistribution and use in source and binary forms, with or without modification,
+   are permitted provided that the following conditions are met:
+   
+   * Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+   
+   * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+   
+   * Neither the name of the authors nor the names of its contributors
+   may be used to endorse or promote products derived from this software 
+   without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+   TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+/* If you use this library, you must include dtl.hpp only. */
+
+#ifndef DTL_FUNCTORS_H
+#define DTL_FUNCTORS_H
+
+namespace dtl {
+    
+    /**
+     * printer class template
+     */
+    template <typename sesElem, typename stream = ostream >
+    class Printer
+    {
+    public :
+        Printer ()            : out_(cout) {}
+        Printer (stream& out) : out_(out)  {}
+        virtual ~Printer () {}
+        virtual void operator() (const sesElem& se) const = 0;
+    protected :
+        stream& out_;
+    };
+    
+    /**
+     * common element printer class template
+     */
+    template <typename sesElem, typename stream = ostream >
+    class CommonPrinter : public Printer < sesElem, stream >
+    {
+    public :
+        CommonPrinter  ()            : Printer < sesElem, stream > ()    {}
+        CommonPrinter  (stream& out) : Printer < sesElem, stream > (out) {}
+        ~CommonPrinter () {}
+        void operator() (const sesElem& se) const {
+            this->out_ << SES_MARK_COMMON << se.first << endl;    
+        }
+    };
+    
+    /**
+     * ses element printer class template
+     */
+    template <typename sesElem, typename stream = ostream >
+    class ChangePrinter : public Printer < sesElem, stream >
+    {
+    public :
+        ChangePrinter  ()            : Printer < sesElem, stream > ()    {}
+        ChangePrinter  (stream& out) : Printer < sesElem, stream > (out) {}
+        ~ChangePrinter () {}
+        void operator() (const sesElem& se) const {
+            switch (se.second.type) {
+            case SES_ADD:
+                this->out_ << SES_MARK_ADD    << se.first << endl;
+                break;
+            case SES_DELETE:
+                this->out_ << SES_MARK_DELETE << se.first << endl;
+                break;
+            case SES_COMMON:
+                this->out_ << SES_MARK_COMMON << se.first << endl;
+                break;
+            }
+        }
+    };
+    
+    /**
+     * unified format element printer class template
+     */
+    template <typename sesElem, typename stream = ostream >
+    class UniHunkPrinter
+    {
+    public :
+        UniHunkPrinter  ()            : out_(cout) {}
+        UniHunkPrinter  (stream& out) : out_(out)  {}
+        ~UniHunkPrinter () {}
+        void operator() (const uniHunk< sesElem >& hunk) const {
+            out_ << "@@"
+                 << " -"  << hunk.a << "," << hunk.b
+                 << " +"  << hunk.c << "," << hunk.d
+                 << " @@" << endl;
+            
+            for_each(hunk.common[0].begin(), hunk.common[0].end(), CommonPrinter< sesElem, stream >(out_));
+            for_each(hunk.change.begin(),    hunk.change.end(),    ChangePrinter< sesElem, stream >(out_));
+            for_each(hunk.common[1].begin(), hunk.common[1].end(), CommonPrinter< sesElem, stream >(out_));
+        }
+    private :
+        stream& out_;
+    };
+
+    /**
+     * storage class template
+     */
+    template <typename sesElem, typename storedData >
+    class Storage
+    {
+    public:
+        Storage(storedData& sd) : storedData_(sd) {}
+        virtual ~Storage() {}
+        virtual void operator() (const sesElem& se) const = 0;
+    protected:
+        storedData& storedData_;
+    };
+    
+    /**
+     * compare class template
+     */
+    template <typename elem>
+    class Compare
+    {
+    public :
+        Compare () {}
+        virtual ~Compare () {}
+        virtual inline bool impl (const elem& e1, const elem& e2) const {
+            return e1 == e2;
+        }
+    };
+}
+
+#endif // DTL_FUNCTORS_H

subprojects/tracy/dtl/variables.hpp

@@ -0,0 +1,142 @@
+/**
+   dtl -- Diff Template Library
+   
+   In short, Diff Template Library is distributed under so called "BSD license",
+   
+   Copyright (c) 2015 Tatsuhiko Kubo <cubicdaiya@gmail.com>
+   All rights reserved.
+   
+   Redistribution and use in source and binary forms, with or without modification,
+   are permitted provided that the following conditions are met:
+   
+   * Redistributions of source code must retain the above copyright notice,
+   this list of conditions and the following disclaimer.
+   
+   * Redistributions in binary form must reproduce the above copyright notice,
+   this list of conditions and the following disclaimer in the documentation
+   and/or other materials provided with the distribution.
+   
+   * Neither the name of the authors nor the names of its contributors
+   may be used to endorse or promote products derived from this software 
+   without specific prior written permission.
+   
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
+   TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+   PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+   LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+   NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+   SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+*/
+
+/* If you use this library, you must include dtl.hpp only. */
+
+#ifndef DTL_VARIABLES_H
+#define DTL_VARIABLES_H
+
+#include <vector>
+#include <list>
+#include <string>
+#include <algorithm>
+#include <iostream>
+
+namespace dtl {
+    
+    using std::vector;
+    using std::string;
+    using std::pair;
+    using std::ostream;
+    using std::list;
+    using std::for_each;
+    using std::distance;
+    using std::fill;
+    using std::cout;
+    using std::endl;
+    using std::rotate;
+    using std::swap;
+    using std::max;
+
+    /**
+     * version string
+     */
+    const string version = "1.20";
+
+    /**
+     * type of edit for SES
+     */
+    typedef int edit_t;
+    const   edit_t SES_DELETE = -1;
+    const   edit_t SES_COMMON = 0;
+    const   edit_t SES_ADD    = 1;
+    
+    /**
+     * mark of SES
+     */
+#define SES_MARK_DELETE "-"
+#define SES_MARK_COMMON " "
+#define SES_MARK_ADD    "+"
+
+    /**
+     * info for Unified Format
+     */
+    typedef struct eleminfo {
+        long long beforeIdx;           // index of prev sequence
+        long long afterIdx;            // index of after sequence
+        edit_t    type;                // type of edit(Add, Delete, Common)
+        bool operator==(const eleminfo& other) const{
+            return (this->beforeIdx == other.beforeIdx && this->afterIdx == other.afterIdx && this->type == other.type);
+        }
+    } elemInfo;
+    
+    const long long DTL_SEPARATE_SIZE = 3;
+    const long long DTL_CONTEXT_SIZE  = 3;
+    
+    /**
+     * cordinate for registering route
+     */
+    typedef struct Point {
+        long long x;                         // x cordinate
+        long long y;                         // y cordinate
+        long long k;                         // vertex
+    } P;
+    
+    /**
+     * limit of cordinate size
+     */
+    const unsigned long long MAX_CORDINATES_SIZE = 2000000;
+    
+    typedef vector< long long > editPath;
+    typedef vector< P >         editPathCordinates;
+    
+    /**
+     * Structure of Unified Format Hunk
+     */
+    template <typename sesElem>
+    struct uniHunk {
+        long long a, b, c, d;        // @@ -a,b +c,d @@
+        vector< sesElem > common[2]; // anteroposterior commons on changes
+        vector< sesElem > change;    // changes
+        long long inc_dec_count;     // count of increace and decrease
+    };
+
+#define dtl_typedefs(elem, sequence)                                    \
+    typedef pair< elem, elemInfo >            sesElem;                  \
+    typedef vector< sesElem >                 sesElemVec;               \
+    typedef vector< uniHunk< sesElem > >      uniHunkVec;               \
+    typedef list< elem >                      elemList;                 \
+    typedef vector< elem >                    elemVec;                  \
+    typedef typename uniHunkVec::iterator     uniHunkVec_iter;          \
+    typedef typename sesElemVec::iterator     sesElemVec_iter;          \
+    typedef typename elemList::iterator       elemList_iter;            \
+    typedef typename sequence::iterator       sequence_iter;            \
+    typedef typename sequence::const_iterator sequence_const_iter;      \
+    typedef typename elemVec::iterator        elemVec_iter;
+
+
+} 
+
+#endif // DTL_VARIABLES_H

subprojects/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)

subprojects/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;
+}

subprojects/tracy/examples/ToyPathTracer/.gitignore

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

subprojects/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.

subprojects/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)

subprojects/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

subprojects/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
+}

subprojects/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;
+};
+

subprojects/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;
+}

subprojects/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);

subprojects/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      
+    }
+
+}

subprojects/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;
+    }
+
+}

subprojects/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() );
+}

subprojects/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 );
+	};
+
+}

subprojects/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();
+}
+

subprojects/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

subprojects/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
+
+