Add: [Video] move GameLoop into its own thread

This allows drawing to happen while the GameLoop is doing an
iteration too.

Sadly, not much drawing currently can be done while the GameLoop
is running, as for example PollEvent() or UpdateWindows() can
influence the game-state. As such, they first need to acquire a
lock on the game-state before they can be called.

Currently, the main advantage is the time spend in Paint(), which
for non-OpenGL drivers can be a few milliseconds. For OpenGL this
is more like 0.05 milliseconds; in these instances this change
doesn't add any benefits for now.

This is an alternative to the former "draw-thread", which moved
the drawing in a thread for some OSes. It has similar performance
gain as this does, although this implementation allows for more
finer control over what suffers when the GameLoop takes too
long: drawing or the next GameLoop. For now they both suffer
equally.

src/video/allegro_v.cpp

@@ -412,7 +412,7 @@ bool VideoDriver_Allegro::PollEvent()
  */
 int _allegro_instance_count = 0;
 
-const char *VideoDriver_Allegro::Start(const StringList &parm)
+const char *VideoDriver_Allegro::Start(const StringList &param)
 {
 	if (_allegro_instance_count == 0 && install_allegro(SYSTEM_AUTODETECT, &errno, nullptr)) {
 		DEBUG(driver, 0, "allegro: install_allegro failed '%s'", allegro_error);
@@ -440,6 +440,8 @@ const char *VideoDriver_Allegro::Start(const StringList &parm)
 	MarkWholeScreenDirty();
 	set_close_button_callback(HandleExitGameRequest);
 
+	this->is_game_threaded = !GetDriverParamBool(param, "no_threads") && !GetDriverParamBool(param, "no_thread");
+
 	return nullptr;
 }
 
@@ -475,12 +477,16 @@ void VideoDriver_Allegro::InputLoop()
 
 void VideoDriver_Allegro::MainLoop()
 {
+	this->StartGameThread();
+
 	for (;;) {
-		if (_exit_game) return;
+		if (_exit_game) break;
 
 		this->Tick();
 		this->SleepTillNextTick();
 	}
+
+	this->StopGameThread();
 }
 
 bool VideoDriver_Allegro::ChangeResolution(int w, int h)

src/video/cocoa/cocoa_ogl.h

@@ -33,6 +33,8 @@ public:
 
 	void ClearSystemSprites() override;
 
+	void PopulateSystemSprites() override;
+
 	bool HasAnimBuffer() override { return true; }
 	uint8 *GetAnimBuffer() override { return this->anim_buffer; }
 

src/video/cocoa/cocoa_ogl.mm

@@ -214,6 +214,8 @@ const char *VideoDriver_CocoaOpenGL::Start(const StringList &param)
 	this->UpdateVideoModes();
 	MarkWholeScreenDirty();
 
+	this->is_game_threaded = !GetDriverParamBool(param, "no_threads") && !GetDriverParamBool(param, "no_thread");
+
 	return nullptr;
 
 }
@@ -227,6 +229,11 @@ void VideoDriver_CocoaOpenGL::Stop()
 	CGLReleaseContext(this->gl_context);
 }
 
+void VideoDriver_CocoaOpenGL::PopulateSystemSprites()
+{
+	OpenGLBackend::Get()->PopulateCursorCache();
+}
+
 void VideoDriver_CocoaOpenGL::ClearSystemSprites()
 {
 	CGLSetCurrentContext(this->gl_context);

src/video/cocoa/cocoa_v.mm

@@ -435,6 +435,8 @@ void VideoDriver_Cocoa::InputLoop()
 /** Main game loop. */
 void VideoDriver_Cocoa::MainLoopReal()
 {
+	this->StartGameThread();
+
 	for (;;) {
 		@autoreleasepool {
 			if (_exit_game) {
@@ -447,6 +449,8 @@ void VideoDriver_Cocoa::MainLoopReal()
 			this->SleepTillNextTick();
 		}
 	}
+
+	this->StopGameThread();
 }
 
 
@@ -558,6 +562,8 @@ const char *VideoDriver_CocoaQuartz::Start(const StringList &param)
 	this->GameSizeChanged();
 	this->UpdateVideoModes();
 
+	this->is_game_threaded = !GetDriverParamBool(param, "no_threads") && !GetDriverParamBool(param, "no_thread");
+
 	return nullptr;
 
 }

src/video/dedicated_v.cpp

@@ -267,6 +267,8 @@ void VideoDriver_Dedicated::MainLoop()
 		}
 	}
 
+	this->is_game_threaded = false;
+
 	/* Done loading, start game! */
 
 	if (!_networking) {

src/video/null_v.cpp

@@ -48,9 +48,9 @@ void VideoDriver_Null::MainLoop()
 	uint i;
 
 	for (i = 0; i < this->ticks; i++) {
-		GameLoop();
-		InputLoop();
-		UpdateWindows();
+		::GameLoop();
+		::InputLoop();
+		::UpdateWindows();
 	}
 }
 

src/video/opengl.cpp

@@ -1036,7 +1036,23 @@ void OpenGLBackend::DrawMouseCursor()
 	_cur_dpi = &_screen;
 	for (uint i = 0; i < _cursor.sprite_count; ++i) {
 		SpriteID sprite = _cursor.sprite_seq[i].sprite;
-		const Sprite *spr = GetSprite(sprite, ST_NORMAL);
+
+		/* Sprites are cached by PopulateCursorCache(). */
+		if (this->cursor_cache.Contains(sprite)) {
+			const Sprite *spr = GetSprite(sprite, ST_NORMAL);
+
+			this->RenderOglSprite((OpenGLSprite *)this->cursor_cache.Get(sprite)->data, _cursor.sprite_seq[i].pal,
+					_cursor.pos.x + _cursor.sprite_pos[i].x + UnScaleByZoom(spr->x_offs, ZOOM_LVL_GUI),
+					_cursor.pos.y + _cursor.sprite_pos[i].y + UnScaleByZoom(spr->y_offs, ZOOM_LVL_GUI),
+					ZOOM_LVL_GUI);
+		}
+	}
+}
+
+void OpenGLBackend::PopulateCursorCache()
+{
+	for (uint i = 0; i < _cursor.sprite_count; ++i) {
+		SpriteID sprite = _cursor.sprite_seq[i].sprite;
 
 		if (!this->cursor_cache.Contains(sprite)) {
 			Sprite *old = this->cursor_cache.Insert(sprite, (Sprite *)GetRawSprite(sprite, ST_NORMAL, &SimpleSpriteAlloc, this));
@@ -1046,11 +1062,6 @@ void OpenGLBackend::DrawMouseCursor()
 				free(old);
 			}
 		}
-
-		this->RenderOglSprite((OpenGLSprite *)this->cursor_cache.Get(sprite)->data, _cursor.sprite_seq[i].pal,
-				_cursor.pos.x + _cursor.sprite_pos[i].x + UnScaleByZoom(spr->x_offs, ZOOM_LVL_GUI),
-				_cursor.pos.y + _cursor.sprite_pos[i].y + UnScaleByZoom(spr->y_offs, ZOOM_LVL_GUI),
-				ZOOM_LVL_GUI);
 	}
 }
 

src/video/opengl.h

@@ -88,6 +88,7 @@ public:
 	void Paint();
 
 	void DrawMouseCursor();
+	void PopulateCursorCache();
 	void ClearCursorCache();
 
 	void *GetVideoBuffer();

src/video/sdl2_default_v.cpp

@@ -107,13 +107,7 @@ void VideoDriver_SDL_Default::Paint()
 				break;
 
 			case Blitter::PALETTE_ANIMATION_BLITTER: {
-				bool need_buf = _screen.dst_ptr == nullptr;
-				if (need_buf) _screen.dst_ptr = this->GetVideoPointer();
 				blitter->PaletteAnimate(this->local_palette);
-				if (need_buf) {
-					this->ReleaseVideoPointer();
-					_screen.dst_ptr = nullptr;
-				}
 				break;
 			}
 

src/video/sdl2_opengl_v.cpp

@@ -110,6 +110,11 @@ const char *VideoDriver_SDL_OpenGL::AllocateContext()
 	return OpenGLBackend::Create(&GetOGLProcAddressCallback);
 }
 
+void VideoDriver_SDL_OpenGL::PopulateSystemSprites()
+{
+	OpenGLBackend::Get()->PopulateCursorCache();
+}
+
 void VideoDriver_SDL_OpenGL::ClearSystemSprites()
 {
 	OpenGLBackend::Get()->ClearCursorCache();

src/video/sdl2_opengl_v.h

@@ -24,6 +24,8 @@ public:
 
 	void ClearSystemSprites() override;
 
+	void PopulateSystemSprites() override;
+
 	bool HasAnimBuffer() override { return true; }
 	uint8 *GetAnimBuffer() override { return this->anim_buffer; }
 

src/video/sdl2_v.cpp

@@ -541,14 +541,14 @@ const char *VideoDriver_SDL_Base::Initialize()
 	return nullptr;
 }
 
-const char *VideoDriver_SDL_Base::Start(const StringList &parm)
+const char *VideoDriver_SDL_Base::Start(const StringList &param)
 {
 	if (BlitterFactory::GetCurrentBlitter()->GetScreenDepth() == 0) return "Only real blitters supported";
 
 	const char *error = this->Initialize();
 	if (error != nullptr) return error;
 
-	this->startup_display = FindStartupDisplay(GetDriverParamInt(parm, "display", -1));
+	this->startup_display = FindStartupDisplay(GetDriverParamInt(param, "display", -1));
 
 	if (!CreateMainSurface(_cur_resolution.width, _cur_resolution.height, false)) {
 		return SDL_GetError();
@@ -562,6 +562,12 @@ const char *VideoDriver_SDL_Base::Start(const StringList &parm)
 	SDL_StopTextInput();
 	this->edit_box_focused = false;
 
+#ifdef __EMSCRIPTEN__
+	this->is_game_threaded = false;
+#else
+	this->is_game_threaded = !GetDriverParamBool(param, "no_threads") && !GetDriverParamBool(param, "no_thread");
+#endif
+
 	return nullptr;
 }
 
@@ -637,9 +643,13 @@ void VideoDriver_SDL_Base::MainLoop()
 	/* Run the main loop event-driven, based on RequestAnimationFrame. */
 	emscripten_set_main_loop_arg(&this->EmscriptenLoop, this, 0, 1);
 #else
+	this->StartGameThread();
+
 	while (!_exit_game) {
 		LoopOnce();
 	}
+
+	this->StopGameThread();
 #endif
 }
 

src/video/sdl_v.cpp

@@ -570,10 +570,10 @@ bool VideoDriver_SDL::PollEvent()
 	return true;
 }
 
-const char *VideoDriver_SDL::Start(const StringList &parm)
+const char *VideoDriver_SDL::Start(const StringList &param)
 {
 	char buf[30];
-	_use_hwpalette = GetDriverParamInt(parm, "hw_palette", 2);
+	_use_hwpalette = GetDriverParamInt(param, "hw_palette", 2);
 
 	/* Just on the offchance the audio subsystem started before the video system,
 	 * check whether any part of SDL has been initialised before getting here.
@@ -599,6 +599,8 @@ const char *VideoDriver_SDL::Start(const StringList &parm)
 	MarkWholeScreenDirty();
 	SetupKeyboard();
 
+	this->is_game_threaded = !GetDriverParamBool(param, "no_threads") && !GetDriverParamBool(param, "no_thread");
+
 	return nullptr;
 }
 
@@ -647,12 +649,16 @@ void VideoDriver_SDL::InputLoop()
 
 void VideoDriver_SDL::MainLoop()
 {
+	this->StartGameThread();
+
 	for (;;) {
 		if (_exit_game) break;
 
 		this->Tick();
 		this->SleepTillNextTick();
 	}
+
+	this->StopGameThread();
 }
 
 bool VideoDriver_SDL::ChangeResolution(int w, int h)

src/video/video_driver.cpp

@@ -8,9 +8,9 @@
 /** @file video_driver.cpp Common code between video driver implementations. */
 
 #include "../stdafx.h"
-#include "../debug.h"
 #include "../core/random_func.hpp"
 #include "../network/network.h"
+#include "../debug.h"
 #include "../gfx_func.h"
 #include "../progress.h"
 #include "../thread.h"
@@ -19,40 +19,84 @@
 
 bool _video_hw_accel; ///< Whether to consider hardware accelerated video drivers.
 
-void VideoDriver::Tick()
+void VideoDriver::GameLoop()
 {
-	auto cur_ticks = std::chrono::steady_clock::now();
+	this->next_game_tick += this->GetGameInterval();
 
-	if (cur_ticks >= this->next_game_tick) {
-		this->next_game_tick += this->GetGameInterval();
-		/* Avoid next_game_tick getting behind more and more if it cannot keep up. */
-		if (this->next_game_tick < cur_ticks - ALLOWED_DRIFT * this->GetGameInterval()) this->next_game_tick = cur_ticks;
+	/* Avoid next_game_tick getting behind more and more if it cannot keep up. */
+	auto now = std::chrono::steady_clock::now();
+	if (this->next_game_tick < now - ALLOWED_DRIFT * this->GetGameInterval()) this->next_game_tick = now;
+
+	{
+		std::lock_guard<std::mutex> lock(this->game_state_mutex);
 
-		/* The game loop is the part that can run asynchronously.
-		 * The rest except sleeping can't. */
-		this->UnlockVideoBuffer();
 		::GameLoop();
-		this->LockVideoBuffer();
+	}
+}
+
+void VideoDriver::GameThread()
+{
+	while (!_exit_game) {
+		this->GameLoop();
+
+		auto now = std::chrono::steady_clock::now();
+		if (this->next_game_tick > now) {
+			std::this_thread::sleep_for(this->next_game_tick - now);
+		} else {
+			/* Ensure we yield this thread if drawings wants to take a lock on
+			 * the game state. This is mainly because most OSes have an
+			 * optimization that if you unlock/lock a mutex in the same thread
+			 * quickly, it will never context switch even if there is another
+			 * thread waiting to take the lock on the same mutex. */
+			std::lock_guard<std::mutex> lock(this->game_thread_wait_mutex);
+		}
+	}
+}
+
+/* static */ void VideoDriver::GameThreadThunk(VideoDriver *drv)
+{
+	drv->GameThread();
+}
+
+void VideoDriver::StartGameThread()
+{
+	if (this->is_game_threaded) {
+		this->is_game_threaded = StartNewThread(&this->game_thread, "ottd:game", &VideoDriver::GameThreadThunk, this);
+	}
+
+	DEBUG(driver, 1, "using %sthread for game-loop", this->is_game_threaded ? "" : "no ");
+}
+
+void VideoDriver::StopGameThread()
+{
+	if (!this->is_game_threaded) return;
+
+	this->game_thread.join();
+}
+
+void VideoDriver::Tick()
+{
+	if (!this->is_game_threaded && std::chrono::steady_clock::now() >= this->next_game_tick) {
+		this->GameLoop();
 
 		/* For things like dedicated server, don't run a separate draw-tick. */
 		if (!this->HasGUI()) {
 			::InputLoop();
-			UpdateWindows();
+			::UpdateWindows();
 			this->next_draw_tick = this->next_game_tick;
 		}
 	}
 
-	/* Prevent drawing when switching mode, as windows can be removed when they should still appear. */
-	if (this->HasGUI() && cur_ticks >= this->next_draw_tick && (_switch_mode == SM_NONE || _game_mode == GM_BOOTSTRAP || HasModalProgress())) {
+	auto now = std::chrono::steady_clock::now();
+	if (this->HasGUI() && now >= this->next_draw_tick) {
 		this->next_draw_tick += this->GetDrawInterval();
 		/* Avoid next_draw_tick getting behind more and more if it cannot keep up. */
-		if (this->next_draw_tick < cur_ticks - ALLOWED_DRIFT * this->GetDrawInterval()) this->next_draw_tick = cur_ticks;
+		if (this->next_draw_tick < now - ALLOWED_DRIFT * this->GetDrawInterval()) this->next_draw_tick = now;
 
 		/* Keep the interactive randomizer a bit more random by requesting
 		 * new values when-ever we can. */
 		InteractiveRandom();
 
-		while (this->PollEvent()) {}
 		this->InputLoop();
 
 		/* Check if the fast-forward button is still pressed. */
@@ -64,23 +108,41 @@ void VideoDriver::Tick()
 			this->fast_forward_via_key = false;
 		}
 
-		::InputLoop();
-		UpdateWindows();
+		{
+			/* Tell the game-thread to stop so we can have a go. */
+			std::lock_guard<std::mutex> lock_wait(this->game_thread_wait_mutex);
+			std::lock_guard<std::mutex> lock_state(this->game_state_mutex);
+
+			this->LockVideoBuffer();
+
+			while (this->PollEvent()) {}
+			::InputLoop();
+
+			/* Prevent drawing when switching mode, as windows can be removed when they should still appear. */
+			if (_game_mode == GM_BOOTSTRAP || _switch_mode == SM_NONE || HasModalProgress()) {
+				::UpdateWindows();
+			}
+
+			this->PopulateSystemSprites();
+		}
 
 		this->CheckPaletteAnim();
 		this->Paint();
+
+		this->UnlockVideoBuffer();
 	}
 }
 
 void VideoDriver::SleepTillNextTick()
 {
-	/* See how much time there is till we have to process the next event, and try to hit that as close as possible. */
-	auto next_tick = std::min(this->next_draw_tick, this->next_game_tick);
+	auto next_tick = this->next_draw_tick;
 	auto now = std::chrono::steady_clock::now();
 
+	if (!this->is_game_threaded) {
+		next_tick = min(next_tick, this->next_game_tick);
+	}
+
 	if (next_tick > now) {
-		this->UnlockVideoBuffer();
 		std::this_thread::sleep_for(next_tick - now);
-		this->LockVideoBuffer();
 	}
 }

src/video/video_driver.hpp

@@ -16,7 +16,11 @@
 #include "../gfx_func.h"
 #include "../settings_type.h"
 #include "../zoom_type.h"
+#include <atomic>
 #include <chrono>
+#include <condition_variable>
+#include <mutex>
+#include <thread>
 #include <vector>
 
 extern std::string _ini_videodriver;
@@ -31,6 +35,8 @@ class VideoDriver : public Driver {
 	const uint DEFAULT_WINDOW_HEIGHT = 480u; ///< Default window height.
 
 public:
+	VideoDriver() : is_game_threaded(true) {}
+
 	/**
 	 * Mark a particular area dirty.
 	 * @param left   The left most line of the dirty area.
@@ -83,6 +89,11 @@ public:
 		return false;
 	}
 
+	/**
+	 * Populate all sprites in cache.
+	 */
+	virtual void PopulateSystemSprites() {}
+
 	/**
 	 * Clear all cached sprites.
 	 */
@@ -239,6 +250,16 @@ protected:
 	 */
 	virtual bool PollEvent() { return false; };
 
+	/**
+	 * Start the loop for game-tick.
+	 */
+	void StartGameThread();
+
+	/**
+	 * Stop the loop for the game-tick. This can still tick at most one time before truly shutting down.
+	 */
+	void StopGameThread();
+
 	/**
 	 * Give the video-driver a tick.
 	 * It will process any potential game-tick and/or draw-tick, and/or any
@@ -271,6 +292,17 @@ protected:
 
 	bool fast_forward_key_pressed; ///< The fast-forward key is being pressed.
 	bool fast_forward_via_key; ///< The fast-forward was enabled by key press.
+
+	bool is_game_threaded;
+	std::thread game_thread;
+	std::mutex game_state_mutex;
+	std::mutex game_thread_wait_mutex;
+
+	static void GameThreadThunk(VideoDriver *drv);
+
+private:
+	void GameLoop();
+	void GameThread();
 };
 
 #endif /* VIDEO_VIDEO_DRIVER_HPP */

src/video/win32_v.cpp

@@ -864,12 +864,16 @@ bool VideoDriver_Win32Base::PollEvent()
 
 void VideoDriver_Win32Base::MainLoop()
 {
+	this->StartGameThread();
+
 	for (;;) {
 		if (_exit_game) break;
 
 		this->Tick();
 		this->SleepTillNextTick();
 	}
+
+	this->StopGameThread();
 }
 
 void VideoDriver_Win32Base::ClientSizeChanged(int w, int h, bool force)
@@ -995,6 +999,8 @@ const char *VideoDriver_Win32GDI::Start(const StringList &param)
 
 	MarkWholeScreenDirty();
 
+	this->is_game_threaded = !GetDriverParamBool(param, "no_threads") && !GetDriverParamBool(param, "no_thread");
+
 	return nullptr;
 }
 
@@ -1115,13 +1121,7 @@ void VideoDriver_Win32GDI::Paint()
 				break;
 
 			case Blitter::PALETTE_ANIMATION_BLITTER: {
-				bool need_buf = _screen.dst_ptr == nullptr;
-				if (need_buf) _screen.dst_ptr = this->GetVideoPointer();
 				blitter->PaletteAnimate(_local_palette);
-				if (need_buf) {
-					this->ReleaseVideoPointer();
-					_screen.dst_ptr = nullptr;
-				}
 				break;
 			}
 
@@ -1291,6 +1291,8 @@ const char *VideoDriver_Win32OpenGL::Start(const StringList &param)
 
 	MarkWholeScreenDirty();
 
+	this->is_game_threaded = !GetDriverParamBool(param, "no_threads") && !GetDriverParamBool(param, "no_thread");
+
 	return nullptr;
 }
 
@@ -1371,6 +1373,11 @@ bool VideoDriver_Win32OpenGL::AfterBlitterChange()
 	return true;
 }
 
+void VideoDriver_Win32OpenGL::PopulateSystemSprites()
+{
+	OpenGLBackend::Get()->PopulateCursorCache();
+}
+
 void VideoDriver_Win32OpenGL::ClearSystemSprites()
 {
 	OpenGLBackend::Get()->ClearCursorCache();

src/video/win32_v.h

@@ -129,6 +129,8 @@ public:
 
 	bool UseSystemCursor() override { return true; }
 
+	void PopulateSystemSprites() override;
+
 	void ClearSystemSprites() override;
 
 	bool HasAnimBuffer() override { return true; }