Fifo: rewrite sync on idle skipping hack

Now it's done without a busy loop

Source/Core/Core/CoreTiming.cpp

@@ -480,11 +480,8 @@ void Idle()
 		//When the FIFO is processing data we must not advance because in this way
 		//the VI will be desynchronized. So, We are waiting until the FIFO finish and
 		//while we process only the events required by the FIFO.
-		while (g_video_backend->Video_IsPossibleWaitingSetDrawDone())
-		{
-			ProcessFifoWaitEvents();
-			Common::YieldCPU();
-		}
+		ProcessFifoWaitEvents();
+		g_video_backend->Video_Sync();
 	}
 
 	idledCycles += DowncountToCycles(PowerPC::ppcState.downcount);

Source/Core/VideoBackends/Software/SWmain.cpp

@@ -360,11 +360,6 @@ void VideoSoftware::Video_GatherPipeBursted()
 	SWCommandProcessor::GatherPipeBursted();
 }
 
-bool VideoSoftware::Video_IsPossibleWaitingSetDrawDone()
-{
-	return false;
-}
-
 void VideoSoftware::RegisterCPMMIO(MMIO::Mapping* mmio, u32 base)
 {
 	SWCommandProcessor::RegisterMMIO(mmio, base);

Source/Core/VideoBackends/Software/VideoBackend.h

@@ -45,7 +45,7 @@ class VideoSoftware : public VideoBackend
 	void Video_SetRendering(bool bEnabled) override;
 
 	void Video_GatherPipeBursted() override;
-	bool Video_IsPossibleWaitingSetDrawDone() override;
+	void Video_Sync() override {}
 
 	void RegisterCPMMIO(MMIO::Mapping* mmio, u32 base) override;
 

Source/Core/VideoCommon/CommandProcessor.cpp

@@ -40,7 +40,6 @@ static u16 m_bboxright;
 static u16 m_bboxbottom;
 static u16 m_tokenReg;
 
-volatile bool isPossibleWaitingSetDrawDone = false;
 volatile bool interruptSet= false;
 volatile bool interruptWaiting= false;
 volatile bool interruptTokenWaiting = false;
@@ -70,7 +69,6 @@ void DoState(PointerWrap &p)
 	p.Do(m_tokenReg);
 	p.Do(fifo);
 
-	p.Do(isPossibleWaitingSetDrawDone);
 	p.Do(interruptSet);
 	p.Do(interruptWaiting);
 	p.Do(interruptTokenWaiting);
@@ -123,8 +121,6 @@ void Init()
 	interruptFinishWaiting = false;
 	interruptTokenWaiting = false;
 
-	isPossibleWaitingSetDrawDone = false;
-
 	et_UpdateInterrupts = CoreTiming::RegisterEvent("CPInterrupt", UpdateInterrupts_Wrapper);
 }
 
@@ -319,7 +315,7 @@ void GatherPipeBursted()
 			    (ProcessorInterface::Fifo_CPUBase == fifo.CPBase) &&
 			    fifo.CPReadWriteDistance > 0)
 			{
-				ProcessFifoAllDistance();
+				FlushGpu();
 			}
 		}
 		RunGpu();
@@ -468,15 +464,6 @@ void SetCPStatusFromCPU()
 	}
 }
 
-void ProcessFifoAllDistance()
-{
-	if (IsOnThread())
-	{
-		while (!interruptWaiting && fifo.bFF_GPReadEnable && fifo.CPReadWriteDistance && !AtBreakpoint())
-			Common::YieldCPU();
-	}
-}
-
 void ProcessFifoEvents()
 {
 	if (IsOnThread() && (interruptWaiting || interruptFinishWaiting || interruptTokenWaiting))
@@ -518,7 +505,7 @@ void SetCpControlRegister()
 	if (fifo.bFF_GPReadEnable && !m_CPCtrlReg.GPReadEnable)
 	{
 		fifo.bFF_GPReadEnable = m_CPCtrlReg.GPReadEnable;
-		while (fifo.isGpuReadingData) Common::YieldCPU();
+		FlushGpu();
 	}
 	else
 	{

Source/Core/VideoCommon/CommandProcessor.h

@@ -17,7 +17,6 @@ namespace CommandProcessor
 
 extern SCPFifoStruct fifo; //This one is shared between gfx thread and emulator thread.
 
-extern volatile bool isPossibleWaitingSetDrawDone; //This one is used for sync gfx thread and emulator thread.
 extern volatile bool interruptSet;
 extern volatile bool interruptWaiting;
 extern volatile bool interruptTokenWaiting;
@@ -145,7 +144,6 @@ void UpdateInterruptsFromVideoBackend(u64 userdata);
 void SetCpClearRegister();
 void SetCpControlRegister();
 void SetCpStatusRegister();
-void ProcessFifoAllDistance();
 void ProcessFifoEvents();
 
 void Update();

Source/Core/VideoCommon/Fifo.cpp

@@ -62,6 +62,9 @@ static u8* s_video_buffer_pp_read_ptr;
 static Common::Flag s_gpu_is_running; // If this one is set, the gpu loop will be called at least once again
 static Common::Event s_gpu_new_work_event;
 
+static Common::Flag s_gpu_is_pending; // If this one is set, there might still be work to do
+static Common::Event s_gpu_done_event;
+
 void Fifo_DoState(PointerWrap &p)
 {
 	p.DoArray(s_video_buffer, FIFO_SIZE);
@@ -85,7 +88,6 @@ void Fifo_PauseAndLock(bool doLock, bool unpauseOnUnlock)
 		EmulatorState(false);
 		if (!Core::IsGPUThread())
 			m_csHWVidOccupied.lock();
-		_dbg_assert_(COMMON, !CommandProcessor::fifo.isGpuReadingData);
 	}
 	else
 	{
@@ -131,9 +133,8 @@ void ExitGpuLoop()
 {
 	// This should break the wait loop in CPU thread
 	CommandProcessor::fifo.bFF_GPReadEnable = false;
-	SCPFifoStruct &fifo = CommandProcessor::fifo;
-	while (fifo.isGpuReadingData)
-		Common::YieldCPU();
+	FlushGpu();
+
 	// Terminate GPU thread loop
 	GpuRunningState = false;
 	EmuRunningState = true;
@@ -317,7 +318,6 @@ void RunGpuLoop()
 			while (GpuRunningState && EmuRunningState && run_loop && !CommandProcessor::interruptWaiting && fifo.bFF_GPReadEnable && fifo.CPReadWriteDistance && !AtBreakpoint())
 			{
 				fifo.isGpuReadingData = true;
-				CommandProcessor::isPossibleWaitingSetDrawDone = fifo.bFF_GPLinkEnable ? true : false;
 
 				if (!SConfig::GetInstance().m_LocalCoreStartupParameter.bSyncGPU || Common::AtomicLoad(CommandProcessor::VITicks) > CommandProcessor::m_cpClockOrigin)
 				{
@@ -356,18 +356,21 @@ void RunGpuLoop()
 				// If we don't, s_swapRequested or s_efbAccessRequested won't be set to false
 				// leading the CPU thread to wait in Video_BeginField or Video_AccessEFB thus slowing things down.
 				AsyncRequests::GetInstance()->PullEvents();
-				CommandProcessor::isPossibleWaitingSetDrawDone = false;
 			}
 
 			// don't release the GPU running state on sync GPU waits
 			fifo.isGpuReadingData = !run_loop;
 		}
 
+		s_gpu_is_pending.Clear();
+		s_gpu_done_event.Set();
+
 		if (EmuRunningState)
 		{
 			if (s_gpu_is_running.IsSet())
 			{
 				// reset the atomic flag. But as the CPU thread might have pushed some new data, we have to rerun the GPU loop
+				s_gpu_is_pending.Set();
 				s_gpu_is_running.Clear();
 			}
 			else
@@ -393,6 +396,16 @@ void RunGpuLoop()
 	AsyncRequests::GetInstance()->SetPassthrough(true);
 }
 
+void FlushGpu()
+{
+	if (!SConfig::GetInstance().m_LocalCoreStartupParameter.bCPUThread || g_use_deterministic_gpu_thread)
+		return;
+
+	while (s_gpu_is_running.IsSet() || s_gpu_is_pending.IsSet())
+	{
+		s_gpu_done_event.Wait();
+	}
+}
 
 bool AtBreakpoint()
 {
@@ -437,6 +450,7 @@ void RunGpu()
 	// wake up GPU thread
 	if (SConfig::GetInstance().m_LocalCoreStartupParameter.bCPUThread && !s_gpu_is_running.IsSet())
 	{
+		s_gpu_is_pending.Set();
 		s_gpu_is_running.Set();
 		s_gpu_new_work_event.Set();
 	}

Source/Core/VideoCommon/Fifo.h

@@ -41,6 +41,7 @@ void SyncGPU(SyncGPUReason reason, bool may_move_read_ptr = true);
 void PushFifoAuxBuffer(void* ptr, size_t size);
 void* PopFifoAuxBuffer(size_t size);
 
+void FlushGpu();
 void RunGpu();
 void RunGpuLoop();
 void ExitGpuLoop();

Source/Core/VideoCommon/MainBase.cpp

@@ -233,9 +233,9 @@ void VideoBackendHardware::Video_GatherPipeBursted()
 	CommandProcessor::GatherPipeBursted();
 }
 
-bool VideoBackendHardware::Video_IsPossibleWaitingSetDrawDone()
+void VideoBackendHardware::Video_Sync()
 {
-	return CommandProcessor::isPossibleWaitingSetDrawDone;
+	FlushGpu();
 }
 
 void VideoBackendHardware::RegisterCPMMIO(MMIO::Mapping* mmio, u32 base)

Source/Core/VideoCommon/PixelEngine.cpp

@@ -287,7 +287,6 @@ void SetFinish_OnMainThread(u64 userdata, int cyclesLate)
 	Common::AtomicStore(*(volatile u32*)&g_bSignalFinishInterrupt, 1);
 	UpdateInterrupts();
 	CommandProcessor::interruptFinishWaiting = false;
-	CommandProcessor::isPossibleWaitingSetDrawDone = false;
 }
 
 // SetToken

Source/Core/VideoCommon/VideoBackendBase.h

@@ -99,7 +99,7 @@ public:
 
 	virtual void Video_GatherPipeBursted() = 0;
 
-	virtual bool Video_IsPossibleWaitingSetDrawDone() = 0;
+	virtual void Video_Sync() = 0;
 
 	// Registers MMIO handlers for the CommandProcessor registers.
 	virtual void RegisterCPMMIO(MMIO::Mapping* mmio, u32 base) = 0;
@@ -148,7 +148,7 @@ class VideoBackendHardware : public VideoBackend
 
 	void Video_GatherPipeBursted() override;
 
-	bool Video_IsPossibleWaitingSetDrawDone() override;
+	void Video_Sync() override;
 
 	void RegisterCPMMIO(MMIO::Mapping* mmio, u32 base) override;