Merge pull request #3530 from degasus/syncgpufix

Fifo: Fix SyncGPU.

6 files changed, 40 insertions, 27 deletions

diff --git a/Source/Core/Core/Core.cpp b/Source/Core/Core/Core.cpp
index ee7e1141d7..d8af296679 100644
--- a/Source/Core/Core/Core.cpp
+++ b/Source/Core/Core/Core.cpp
@@ -498,6 +498,9 @@ void EmuThread()
 
 	CBoot::BootUp();
 
+	// This adds the SyncGPU handler to CoreTiming, so now CoreTiming::Advance might block.
+	Fifo::Prepare();
+
 	// Thread is no longer acting as CPU Thread
 	UndeclareAsCPUThread();
 
diff --git a/Source/Core/Core/CoreTiming.cpp b/Source/Core/Core/CoreTiming.cpp
index 6a1c76b186..59d40727ae 100644
--- a/Source/Core/Core/CoreTiming.cpp
+++ b/Source/Core/Core/CoreTiming.cpp
@@ -446,7 +446,7 @@ void Idle()
 		//the VI will be desynchronized. So, We are waiting until the FIFO finish and
 		//while we process only the events required by the FIFO.
 		ProcessFifoWaitEvents();
-		Fifo::Update(0);
+		Fifo::FlushGpu();
 	}
 
 	idledCycles += DowncountToCycles(PowerPC::ppcState.downcount);
diff --git a/Source/Core/Core/HW/SystemTimers.cpp b/Source/Core/Core/HW/SystemTimers.cpp
index 82e6962bf2..e900e33762 100644
--- a/Source/Core/Core/HW/SystemTimers.cpp
+++ b/Source/Core/Core/HW/SystemTimers.cpp
@@ -66,7 +66,6 @@ namespace SystemTimers
 
 static int et_Dec;
 static int et_VI;
-static int et_CP;
 static int et_AudioDMA;
 static int et_DSP;
 static int et_IPC_HLE;
@@ -74,7 +73,6 @@ static int et_PatchEngine; // PatchEngine updates every 1/60th of a second by de
 static int et_Throttle;
 
 static u32 s_cpu_core_clock = 486000000u;             // 486 mhz (its not 485, stop bugging me!)
-static u64 s_last_sync_gpu_tick;
 
 // These two are badly educated guesses.
 // Feel free to experiment. Set them in Init below.
@@ -123,16 +121,6 @@ static void VICallback(u64 userdata, int cyclesLate)
 	CoreTiming::ScheduleEvent(VideoInterface::GetTicksPerHalfLine() - cyclesLate, et_VI);
 }
 
-static void CPCallback(u64 userdata, int cyclesLate)
-{
-	u64 now = CoreTiming::GetTicks();
-	int next = Fifo::Update((int)(now - s_last_sync_gpu_tick));
-	s_last_sync_gpu_tick = now;
-
-	if (next > 0)
-		CoreTiming::ScheduleEvent(next, et_CP);
-}
-
 static void DecrementerCallback(u64 userdata, int cyclesLate)
 {
 	PowerPC::ppcState.spr[SPR_DEC] = 0xFFFFFFFF;
@@ -238,8 +226,6 @@ void Init()
 
 	et_Dec = CoreTiming::RegisterEvent("DecCallback", DecrementerCallback);
 	et_VI = CoreTiming::RegisterEvent("VICallback", VICallback);
-	if (SConfig::GetInstance().bCPUThread && SConfig::GetInstance().bSyncGPU)
-		et_CP = CoreTiming::RegisterEvent("CPCallback", CPCallback);
 	et_DSP = CoreTiming::RegisterEvent("DSPCallback", DSPCallback);
 	et_AudioDMA = CoreTiming::RegisterEvent("AudioDMACallback", AudioDMACallback);
 	et_IPC_HLE = CoreTiming::RegisterEvent("IPC_HLE_UpdateCallback", IPC_HLE_UpdateCallback);
@@ -250,9 +236,6 @@ void Init()
 	CoreTiming::ScheduleEvent(0, et_DSP);
 	CoreTiming::ScheduleEvent(s_audio_dma_period, et_AudioDMA);
 	CoreTiming::ScheduleEvent(0, et_Throttle, Common::Timer::GetTimeMs());
-	if (SConfig::GetInstance().bCPUThread && SConfig::GetInstance().bSyncGPU)
-		CoreTiming::ScheduleEvent(0, et_CP);
-	s_last_sync_gpu_tick = CoreTiming::GetTicks();
 
 	CoreTiming::ScheduleEvent(VideoInterface::GetTicksPerField(), et_PatchEngine);
 
diff --git a/Source/Core/Core/State.cpp b/Source/Core/Core/State.cpp
index 33a7454083..df97809e16 100644
--- a/Source/Core/Core/State.cpp
+++ b/Source/Core/Core/State.cpp
@@ -71,7 +71,7 @@ static Common::Event g_compressAndDumpStateSyncEvent;
 static std::thread g_save_thread;
 
 // Don't forget to increase this after doing changes on the savestate system
-static const u32 STATE_VERSION = 50; // Last changed in PR 3457
+static const u32 STATE_VERSION = 51; // Last changed in PR 3530
 
 // Maps savestate versions to Dolphin versions.
 // Versions after 42 don't need to be added to this list,
diff --git a/Source/Core/VideoCommon/Fifo.cpp b/Source/Core/VideoCommon/Fifo.cpp
index 506024c193..e04f3ffcbf 100644
--- a/Source/Core/VideoCommon/Fifo.cpp
+++ b/Source/Core/VideoCommon/Fifo.cpp
@@ -15,6 +15,7 @@
 #include "Common/MsgHandler.h"
 
 #include "Core/ConfigManager.h"
+#include "Core/CoreTiming.h"
 #include "Core/NetPlayProto.h"
 #include "Core/HW/Memmap.h"
 
@@ -45,6 +46,9 @@ static u8* s_fifo_aux_read_ptr;
 
 bool g_use_deterministic_gpu_thread;
 
+static u64 s_last_sync_gpu_tick;
+static int s_event_sync_gpu;
+
 // STATE_TO_SAVE
 static u8* s_video_buffer;
 static u8* s_video_buffer_read_ptr;
@@ -79,6 +83,7 @@ void DoState(PointerWrap &p)
 		s_video_buffer_seen_ptr = s_video_buffer_pp_read_ptr = s_video_buffer_read_ptr;
 	}
 	p.Do(g_bSkipCurrentFrame);
+	p.Do(s_last_sync_gpu_tick);
 }
 
 void PauseAndLock(bool doLock, bool unpauseOnUnlock)
@@ -492,16 +497,15 @@ void UpdateWantDeterminism(bool want)
 	}
 }
 
-int Update(int ticks)
+/* This function checks the emulated CPU - GPU distance and may wake up the GPU,
+ * or block the CPU if required. It should be called by the CPU thread regulary.
+ * @ticks The gone emulated CPU time.
+ * @return A good time to call Update() next.
+ */
+static int Update(int ticks)
 {
 	const SConfig& param = SConfig::GetInstance();
 
-	if (ticks == 0)
-	{
-		FlushGpu();
-		return param.iSyncGpuMaxDistance;
-	}
-
 	// GPU is sleeping, so no need for synchronization
 	if (s_gpu_mainloop.IsDone() || g_use_deterministic_gpu_thread)
 	{
@@ -514,10 +518,12 @@ int Update(int ticks)
 		return param.iSyncGpuMaxDistance;
 	}
 
+	// Wakeup GPU
 	int old = s_sync_ticks.fetch_add(ticks);
 	if (old < param.iSyncGpuMinDistance && old + ticks >= param.iSyncGpuMinDistance)
 		RunGpu();
 
+	// Wait for GPU
 	if (s_sync_ticks.load() >= param.iSyncGpuMaxDistance)
 	{
 		while (s_sync_ticks.load() > 0)
@@ -529,4 +535,25 @@ int Update(int ticks)
 	return param.iSyncGpuMaxDistance - s_sync_ticks.load();
 }
 
+static void SyncGPUCallback(u64 userdata, int cyclesLate)
+{
+	u64 now = CoreTiming::GetTicks();
+	int next = Fifo::Update((int)(now - s_last_sync_gpu_tick));
+	s_last_sync_gpu_tick = now;
+
+	if (next > 0)
+		CoreTiming::ScheduleEvent(next, s_event_sync_gpu);
+}
+
+// Initialize GPU - CPU thread syncing, this gives us a deterministic way to start the GPU thread.
+void Prepare()
+{
+	if (SConfig::GetInstance().bCPUThread && SConfig::GetInstance().bSyncGPU)
+	{
+		s_event_sync_gpu = CoreTiming::RegisterEvent("SyncGPUCallback", SyncGPUCallback);
+		CoreTiming::ScheduleEvent(0, s_event_sync_gpu);
+		s_last_sync_gpu_tick = CoreTiming::GetTicks();
+	}
+}
+
 }
diff --git a/Source/Core/VideoCommon/Fifo.h b/Source/Core/VideoCommon/Fifo.h
index d342b535df..b30d41069e 100644
--- a/Source/Core/VideoCommon/Fifo.h
+++ b/Source/Core/VideoCommon/Fifo.h
@@ -22,6 +22,7 @@ extern std::atomic<u8*> g_video_buffer_write_ptr_xthread;
 
 void Init();
 void Shutdown();
+void Prepare(); // Must be called from the CPU thread.
 void DoState(PointerWrap &f);
 void PauseAndLock(bool doLock, bool unpauseOnUnlock);
 void UpdateWantDeterminism(bool want);
@@ -52,6 +53,5 @@ void EmulatorState(bool running);
 bool AtBreakpoint();
 void ResetVideoBuffer();
 void SetRendering(bool bEnabled);
-int Update(int ticks);
 
 };