Add some important profiling hooks.

Author: Ivorforce

core/profiling.cpp

@@ -30,7 +30,13 @@
 
 #include "profiling.h"
 
-#ifdef GODOT_USE_PERFETTO
+#if defined(GODOT_USE_TRACY)
+void godot_init_profiler() {
+	// Send our first event to tracy; otherwise it doesn't start collecting data.
+	// FrameMark is kind of fitting because it communicates "this is where we started tracing".
+	FrameMark;
+}
+#elif defined(GODOT_USE_PERFETTO)
 PERFETTO_TRACK_EVENT_STATIC_STORAGE();
 
 void godot_init_profiler() {
@@ -41,5 +47,8 @@ void godot_init_profiler() {
 	perfetto::Tracing::Initialize(args);
 	perfetto::TrackEvent::Register();
 }
-
+#else
+void godot_init_profiler() {
+	// Stub
+}
 #endif

core/profiling.h

@@ -57,11 +57,7 @@
 	static constexpr tracy::SourceLocationData TracyConcat(__tracy_source_location, TracyLine){ m_zone_name, TracyFunction, TracyFile, (uint32_t)TracyLine, 0 }; \
 	new (&__godot_tracy_zone_##m_group_name) tracy::ScopedZone(&TracyConcat(__tracy_source_location, TracyLine), TRACY_CALLSTACK, true)
 
-static void godot_init_profiler() {
-	// Send our first event to tracy; otherwise it doesn't start collecting data.
-	// FrameMark is kind of fitting because it communicates "this is where we started tracing".
-	FrameMark;
-}
+void godot_init_profiler();
 
 #elif defined(GODOT_USE_PERFETTO)
 // Use the perfetto profiler.
@@ -98,8 +94,7 @@ void godot_init_profiler();
 #else
 // No profiling; all macros are stubs.
 
-static void godot_init_profiler() {
-}
+void godot_init_profiler();
 
 #define GodotProfileFrameMark
 #define GodotProfileZone(m_zone_name)

main/main.cpp

@@ -50,6 +50,7 @@
 #include "core/object/script_language.h"
 #include "core/os/os.h"
 #include "core/os/time.h"
+#include "core/profiling.h"
 #include "core/register_core_types.h"
 #include "core/string/translation_server.h"
 #include "core/version.h"
@@ -957,6 +958,7 @@ int Main::test_entrypoint(int argc, char *argv[], bool &tests_need_run) {
  */
 
 Error Main::setup(const char *execpath, int argc, char *argv[], bool p_second_phase) {
+	GodotProfileZone("setup");
 	Thread::make_main_thread();
 	set_current_thread_safe_for_nodes(true);
 
@@ -2882,6 +2884,7 @@ Error _parse_resource_dummy(void *p_data, VariantParser::Stream *p_stream, Ref<R
 }
 
 Error Main::setup2(bool p_show_boot_logo) {
+	GodotProfileZone("setup2");
 	OS::get_singleton()->benchmark_begin_measure("Startup", "Main::Setup2");
 
 	Thread::make_main_thread(); // Make whatever thread call this the main thread.
@@ -3693,6 +3696,7 @@ Error Main::setup2(bool p_show_boot_logo) {
 }
 
 void Main::setup_boot_logo() {
+	GodotProfileZone("setup_boot_logo");
 	MAIN_PRINT("Main: Load Boot Image");
 
 #if !defined(TOOLS_ENABLED) && defined(WEB_ENABLED)
@@ -3783,6 +3787,7 @@ static MainTimerSync main_timer_sync;
 // and should move on to `OS::run`, and EXIT_FAILURE otherwise for
 // an early exit with that error code.
 int Main::start() {
+	GodotProfileZone("start");
 	OS::get_singleton()->benchmark_begin_measure("Startup", "Main::Start");
 
 	ERR_FAIL_COND_V(!_start_success, false);
@@ -4601,6 +4606,8 @@ static uint64_t navigation_process_max = 0;
 // will terminate the program. In case of failure, the OS exit code needs
 // to be set explicitly here (defaults to EXIT_SUCCESS).
 bool Main::iteration() {
+	GodotProfileZone("Main::iteration");
+	GodotProfileZoneGroupedFirst(_profile_zone, "prepare");
 	iterating++;
 
 	const uint64_t ticks = OS::get_singleton()->get_ticks_usec();
@@ -4646,6 +4653,8 @@ bool Main::iteration() {
 #endif // XR_DISABLED
 
 	for (int iters = 0; iters < advance.physics_steps; ++iters) {
+		GodotProfileZone("Physics Step");
+		GodotProfileZoneGroupedFirst(_physics_zone, "setup");
 		if (Input::get_singleton()->is_agile_input_event_flushing()) {
 			Input::get_singleton()->flush_buffered_events();
 		}
@@ -4658,18 +4667,22 @@ bool Main::iteration() {
 		// Prepare the fixed timestep interpolated nodes BEFORE they are updated
 		// by the physics server, otherwise the current and previous transforms
 		// may be the same, and no interpolation takes place.
+		GodotProfileZoneGrouped(_physics_zone, "main loop iteration prepare");
 		OS::get_singleton()->get_main_loop()->iteration_prepare();
 
 #ifndef PHYSICS_3D_DISABLED
+		GodotProfileZoneGrouped(_physics_zone, "PhysicsServer3D::sync");
 		PhysicsServer3D::get_singleton()->sync();
 		PhysicsServer3D::get_singleton()->flush_queries();
 #endif // PHYSICS_3D_DISABLED
 
 #ifndef PHYSICS_2D_DISABLED
+		GodotProfileZoneGrouped(_physics_zone, "PhysicsServer2D::sync");
 		PhysicsServer2D::get_singleton()->sync();
 		PhysicsServer2D::get_singleton()->flush_queries();
 #endif // PHYSICS_2D_DISABLED
 
+		GodotProfileZoneGrouped(_physics_zone, "physics_process");
 		if (OS::get_singleton()->get_main_loop()->physics_process(physics_step * time_scale)) {
 #ifndef PHYSICS_3D_DISABLED
 			PhysicsServer3D::get_singleton()->end_sync();
@@ -4687,9 +4700,11 @@ bool Main::iteration() {
 		uint64_t navigation_begin = OS::get_singleton()->get_ticks_usec();
 
 #ifndef NAVIGATION_2D_DISABLED
+		GodotProfileZoneGrouped(_profile_zone, "NavigationServer2D::physics_process");
 		NavigationServer2D::get_singleton()->physics_process(physics_step * time_scale);
 #endif // NAVIGATION_2D_DISABLED
 #ifndef NAVIGATION_3D_DISABLED
+		GodotProfileZoneGrouped(_profile_zone, "NavigationServer3D::physics_process");
 		NavigationServer3D::get_singleton()->physics_process(physics_step * time_scale);
 #endif // NAVIGATION_3D_DISABLED
 
@@ -4700,17 +4715,20 @@ bool Main::iteration() {
 #endif // !defined(NAVIGATION_2D_DISABLED) || !defined(NAVIGATION_3D_DISABLED)
 
 #ifndef PHYSICS_3D_DISABLED
+		GodotProfileZoneGrouped(_profile_zone, "3D physics");
 		PhysicsServer3D::get_singleton()->end_sync();
 		PhysicsServer3D::get_singleton()->step(physics_step * time_scale);
 #endif // PHYSICS_3D_DISABLED
 
 #ifndef PHYSICS_2D_DISABLED
+		GodotProfileZoneGrouped(_profile_zone, "2D physics");
 		PhysicsServer2D::get_singleton()->end_sync();
 		PhysicsServer2D::get_singleton()->step(physics_step * time_scale);
 #endif // PHYSICS_2D_DISABLED
 
 		message_queue->flush();
 
+		GodotProfileZoneGrouped(_profile_zone, "main loop iteration end");
 		OS::get_singleton()->get_main_loop()->iteration_end();
 
 		physics_process_ticks = MAX(physics_process_ticks, OS::get_singleton()->get_ticks_usec() - physics_begin); // keep the largest one for reference
@@ -4725,20 +4743,25 @@ bool Main::iteration() {
 
 	uint64_t process_begin = OS::get_singleton()->get_ticks_usec();
 
+	GodotProfileZoneGrouped(_profile_zone, "process");
 	if (OS::get_singleton()->get_main_loop()->process(process_step * time_scale)) {
 		exit = true;
 	}
 	message_queue->flush();
 
 #ifndef NAVIGATION_2D_DISABLED
+	GodotProfileZoneGrouped(_profile_zone, "process 2D navigation");
 	NavigationServer2D::get_singleton()->process(process_step * time_scale);
 #endif // NAVIGATION_2D_DISABLED
 #ifndef NAVIGATION_3D_DISABLED
+	GodotProfileZoneGrouped(_profile_zone, "process 3D navigation");
 	NavigationServer3D::get_singleton()->process(process_step * time_scale);
 #endif // NAVIGATION_3D_DISABLED
 
+	GodotProfileZoneGrouped(_profile_zone, "RenderingServer::sync");
 	RenderingServer::get_singleton()->sync(); //sync if still drawing from previous frames.
 
+	GodotProfileZoneGrouped(_profile_zone, "RenderingServer::draw");
 	const bool has_pending_resources_for_processing = RD::get_singleton() && RD::get_singleton()->has_pending_resources_for_processing();
 	bool wants_present = (DisplayServer::get_singleton()->can_any_window_draw() ||
 								 DisplayServer::get_singleton()->has_additional_outputs()) &&
@@ -4762,10 +4785,12 @@ bool Main::iteration() {
 	process_max = MAX(process_ticks, process_max);
 	uint64_t frame_time = OS::get_singleton()->get_ticks_usec() - ticks;
 
+	GodotProfileZoneGrouped(_profile_zone, "ScriptServer::frame");
 	for (int i = 0; i < ScriptServer::get_language_count(); i++) {
 		ScriptServer::get_language(i)->frame();
 	}
 
+	GodotProfileZoneGrouped(_profile_zone, "AudioServer::update");
 	AudioServer::get_singleton()->update();
 
 	if (EngineDebugger::is_active()) {
@@ -4804,6 +4829,7 @@ bool Main::iteration() {
 	iterating--;
 
 	if (movie_writer) {
+		GodotProfileZoneGrouped(_profile_zone, "movie_writer->add_frame");
 		movie_writer->add_frame();
 	}
 
@@ -4831,6 +4857,7 @@ bool Main::iteration() {
 	bool skip_delay = scene_tree && scene_tree->is_accessibility_enabled();
 
 	if (!skip_delay) {
+		GodotProfileZoneGrouped(_profile_zone, "OS::add_frame_delay");
 		OS::get_singleton()->add_frame_delay(DisplayServer::get_singleton()->window_can_draw());
 	}
 
@@ -4871,6 +4898,7 @@ void Main::force_redraw() {
  * The order matters as some of those steps are linked with each other.
  */
 void Main::cleanup(bool p_force) {
+	GodotProfileZone("cleanup");
 	OS::get_singleton()->benchmark_begin_measure("Shutdown", "Main::Cleanup");
 	if (!p_force) {
 		ERR_FAIL_COND(!_start_success);

platform/android/os_android.cpp

@@ -41,6 +41,7 @@
 #include "core/config/project_settings.h"
 #include "core/extension/gdextension_manager.h"
 #include "core/io/xml_parser.h"
+#include "core/profiling.h"
 #include "drivers/unix/dir_access_unix.h"
 #include "drivers/unix/file_access_unix.h"
 #ifdef TOOLS_ENABLED
@@ -347,6 +348,8 @@ void OS_Android::main_loop_begin() {
 }
 
 bool OS_Android::main_loop_iterate(bool *r_should_swap_buffers) {
+	GodotProfileFrameMark;
+	GodotProfileZone("OS_Android::main_loop_iterate");
 	if (!main_loop) {
 		return false;
 	}

platform/ios/os_ios.mm

@@ -42,6 +42,7 @@
 #include "core/io/dir_access.h"
 #include "core/io/file_access.h"
 #include "core/io/file_access_pack.h"
+#include "core/profiling.h"
 #include "drivers/unix/syslog_logger.h"
 #include "main/main.h"
 
@@ -203,6 +204,8 @@ Rect2 fit_keep_aspect_covered(const Vector2 &p_container, const Vector2 &p_rect)
 }
 
 bool OS_IOS::iterate() {
+	GodotProfileFrameMark;
+	GodotProfileZone("OS_IOS::iterate");
 	if (!main_loop) {
 		return true;
 	}

platform/linuxbsd/os_linuxbsd.cpp

@@ -32,6 +32,7 @@
 
 #include "core/io/certs_compressed.gen.h"
 #include "core/io/dir_access.h"
+#include "core/profiling.h"
 #include "main/main.h"
 #include "servers/display_server.h"
 #include "servers/rendering_server.h"
@@ -960,6 +961,8 @@ String OS_LinuxBSD::get_system_dir(SystemDir p_dir, bool p_shared_storage) const
 }
 
 void OS_LinuxBSD::run() {
+	GodotProfileFrameMark;
+	GodotProfileZone("OS_LinuxBSD::run");
 	if (!main_loop) {
 		return;
 	}

platform/macos/os_macos.mm

@@ -37,6 +37,7 @@
 #import "macos_terminal_logger.h"
 
 #include "core/crypto/crypto_core.h"
+#include "core/profiling.h"
 #include "core/version_generated.gen.h"
 #include "main/main.h"
 
@@ -51,6 +52,9 @@
 
 	@autoreleasepool {
 		@try {
+			GodotProfileFrameMark;
+			GodotProfileZone("OS_MacOS::pre_wait_observer_cb");
+
 			// Get rid of pending events.
 			DisplayServer *ds = DisplayServer::get_singleton();
 			DisplayServerMacOS *ds_mac = Object::cast_to<DisplayServerMacOS>(ds);

platform/web/os_web.cpp

@@ -38,6 +38,7 @@
 
 #include "core/config/project_settings.h"
 #include "core/debugger/engine_debugger.h"
+#include "core/profiling.h"
 #include "drivers/unix/dir_access_unix.h"
 #include "drivers/unix/file_access_unix.h"
 #include "main/main.h"
@@ -77,6 +78,8 @@ void OS_Web::fs_sync_callback() {
 }
 
 bool OS_Web::main_loop_iterate() {
+	GodotProfileFrameMark;
+	GodotProfileZone("OS_Web::main_loop_iterate");
 	if (is_userfs_persistent() && idb_needs_sync && !idb_is_syncing) {
 		idb_is_syncing = true;
 		idb_needs_sync = false;

platform/windows/os_windows.cpp

@@ -39,6 +39,7 @@
 #include "core/debugger/engine_debugger.h"
 #include "core/debugger/script_debugger.h"
 #include "core/io/marshalls.h"
+#include "core/profiling.h"
 #include "core/version_generated.gen.h"
 #include "drivers/windows/dir_access_windows.h"
 #include "drivers/windows/file_access_windows.h"
@@ -2268,6 +2269,8 @@ void OS_Windows::run() {
 	main_loop->initialize();
 
 	while (true) {
+		GodotProfileFrameMark;
+		GodotProfileZone("OS_Windows::run");
 		DisplayServer::get_singleton()->process_events(); // get rid of pending events
 		if (Main::iteration()) {
 			break;

scene/main/scene_tree.cpp

@@ -37,6 +37,7 @@
 #include "core/object/message_queue.h"
 #include "core/object/worker_thread_pool.h"
 #include "core/os/os.h"
+#include "core/profiling.h"
 #include "node.h"
 #include "scene/animation/tween.h"
 #include "scene/debugger/scene_debugger.h"
@@ -570,6 +571,7 @@ void SceneTree::set_group(const StringName &p_group, const String &p_name, const
 }
 
 void SceneTree::initialize() {
+	GodotProfileZone("SceneTree::initialize");
 	ERR_FAIL_NULL(root);
 	MainLoop::initialize();
 	root->_set_tree(this);