Core: Skip duplicate frames when using frame advance

It used to be the case that frame advance skipped duplicate frames
(i.e. it would take 30 frame advances to get through one second
of emulated time in a 30 fps game), but this broke in 9c5c3c0.
Skipping duplicate frames making TASing less annoying.

Source/Core/Core/Core.cpp

@@ -77,6 +77,7 @@
 #include "InputCommon/ControllerInterface/ControllerInterface.h"
 #include "InputCommon/GCAdapter.h"
 
+#include "VideoCommon/AsyncRequests.h"
 #include "VideoCommon/Fifo.h"
 #include "VideoCommon/OnScreenDisplay.h"
 #include "VideoCommon/RenderBase.h"
@@ -104,6 +105,7 @@ static std::thread s_cpu_thread;
 static bool s_request_refresh_info = false;
 static bool s_is_throttler_temp_disabled = false;
 static bool s_frame_step = false;
+static std::atomic<bool> s_stop_frame_step;
 
 #ifdef USE_MEMORYWATCHER
 static std::unique_ptr<MemoryWatcher> s_memory_watcher;
@@ -863,18 +865,28 @@ void VideoThrottle()
 void Callback_FramePresented()
 {
   s_drawn_frame++;
+  s_stop_frame_step.store(true);
 }
 
 // Called from VideoInterface::Update (CPU thread) at emulated field boundaries
 void Callback_NewField()
 {
-  Movie::FrameUpdate();
   if (s_frame_step)
   {
-    s_frame_step = false;
-    CPU::Break();
-    if (s_on_state_changed_callback)
-      s_on_state_changed_callback(Core::GetState());
+    // To ensure that s_stop_frame_step is up to date, wait for the GPU thread queue to empty,
+    // since it is may contain a swap event (which will call Callback_FramePresented). This hurts
+    // the performance a little, but luckily, performance matters less when using frame stepping.
+    AsyncRequests::GetInstance()->WaitForEmptyQueue();
+
+    // Only stop the frame stepping if a new frame was displayed
+    // (as opposed to the previous frame being displayed for another frame).
+    if (s_stop_frame_step.load())
+    {
+      s_frame_step = false;
+      CPU::Break();
+      if (s_on_state_changed_callback)
+        s_on_state_changed_callback(Core::GetState());
+    }
   }
 }
 
@@ -1045,6 +1057,7 @@ void DoFrameStep()
   if (GetState() == State::Paused)
   {
     // if already paused, frame advance for 1 frame
+    s_stop_frame_step = false;
     s_frame_step = true;
     RequestRefreshInfo();
     SetState(State::Running);

Source/Core/Core/HW/VideoInterface.cpp

@@ -24,6 +24,7 @@
 #include "Core/HW/ProcessorInterface.h"
 #include "Core/HW/SI/SI.h"
 #include "Core/HW/SystemTimers.h"
+#include "Core/Movie.h"
 
 #include "DiscIO/Enums.h"
 
@@ -816,8 +817,34 @@ static void EndField()
 // Run when: When a frame is scanned (progressive/interlace)
 void Update(u64 ticks)
 {
-  // If this half-line is at a field boundary, potentially deal with frame-stepping
-  // and/or update movie state before dealing with anything else
+  // Movie's frame counter should be updated before actually rendering the frame,
+  // in case frame counter display is enabled
+
+  Movie::FrameUpdate();
+
+  // If this half-line is at some boundary of the "active video lines" in either field, we either
+  // need to (a) send a request to the GPU thread to actually render the XFB, or (b) increment
+  // the number of frames we've actually drawn
+
+  if (s_half_line_count == s_even_field_first_hl)
+  {
+    BeginField(FieldType::Even, ticks);
+  }
+  else if (s_half_line_count == s_odd_field_first_hl)
+  {
+    BeginField(FieldType::Odd, ticks);
+  }
+  else if (s_half_line_count == s_even_field_last_hl)
+  {
+    EndField();
+  }
+  else if (s_half_line_count == s_odd_field_last_hl)
+  {
+    EndField();
+  }
+
+  // If this half-line is at a field boundary, deal with updating movie state before potentially
+  // dealing with SI polls, but after potentially sending a swap request to the GPU thread
 
   if (s_half_line_count == 0 || s_half_line_count == GetHalfLinesPerEvenField())
     Core::Callback_NewField();
@@ -843,27 +870,6 @@ void Update(u64 ticks)
     s_half_line_of_next_si_poll = GetHalfLinesPerEvenField() + num_half_lines_for_si_poll;
   }
 
-  // If this half-line is at some boundary of the "active video lines" in either field, we either
-  // need to (a) send a request to the GPU thread to actually render the XFB, or (b) increment
-  // the number of frames we've actually drawn
-
-  if (s_half_line_count == s_even_field_first_hl)
-  {
-    BeginField(FieldType::Even, ticks);
-  }
-  else if (s_half_line_count == s_odd_field_first_hl)
-  {
-    BeginField(FieldType::Odd, ticks);
-  }
-  else if (s_half_line_count == s_even_field_last_hl)
-  {
-    EndField();
-  }
-  else if (s_half_line_count == s_odd_field_last_hl)
-  {
-    EndField();
-  }
-
   // Move to the next half-line and potentially roll-over the count to zero. If we've reached
   // the beginning of a new full-line, update the timer
 

Source/Core/VideoCommon/AsyncRequests.cpp

@@ -97,6 +97,12 @@ void AsyncRequests::PushEvent(const AsyncRequests::Event& event, bool blocking)
   }
 }
 
+void AsyncRequests::WaitForEmptyQueue()
+{
+  std::unique_lock<std::mutex> lock(m_mutex);
+  m_cond.wait(lock, [this] { return m_queue.empty(); });
+}
+
 void AsyncRequests::SetEnable(bool enable)
 {
   std::unique_lock<std::mutex> lock(m_mutex);

Source/Core/VideoCommon/AsyncRequests.h

@@ -82,6 +82,7 @@ public:
       PullEventsInternal();
   }
   void PushEvent(const Event& event, bool blocking = false);
+  void WaitForEmptyQueue();
   void SetEnable(bool enable);
   void SetPassthrough(bool enable);