GS: Add heuristic-based internal FPS

pcsx2/Counters.cpp

@@ -552,8 +552,6 @@ static __fi void VSyncStart(u32 sCycle)
 	}
 #endif
 
-	PerformanceMetrics::Update();
-
 	frameLimit(); // limit FPS
 	gsPostVsyncStart(); // MUST be after framelimit; doing so before causes funk with frame times!
 

pcsx2/Frontend/ImGuiManager.cpp

@@ -509,7 +509,21 @@ static void DrawPerformanceOverlay()
 		const float speed = PerformanceMetrics::GetSpeed();
 		if (GSConfig.OsdShowFPS)
 		{
-			text.Write("%.2f", PerformanceMetrics::GetFPS());
+			switch (PerformanceMetrics::GetInternalFPSMethod())
+			{
+			case PerformanceMetrics::InternalFPSMethod::GSPrivilegedRegister:
+				text.Write("G: %.2f [P] | V: %.2f", PerformanceMetrics::GetInternalFPS(), PerformanceMetrics::GetFPS());
+				break;
+
+			case PerformanceMetrics::InternalFPSMethod::DISPFBBlit:
+				text.Write("G: %.2f [B] | V: %.2f", PerformanceMetrics::GetInternalFPS(), PerformanceMetrics::GetFPS());
+				break;
+
+			case PerformanceMetrics::InternalFPSMethod::None:
+			default:
+				text.Write("V: %.2f", PerformanceMetrics::GetFPS());
+				break;
+			}
 			first = false;
 		}
 		if (GSConfig.OsdShowSpeed)

pcsx2/GS.cpp

@@ -26,6 +26,7 @@ using namespace Threading;
 using namespace R5900;
 
 alignas(16) u8 g_RealGSMem[Ps2MemSize::GSregs];
+static bool s_GSRegistersWritten = false;
 
 void gsSetVideoMode(GS_VideoMode mode)
 {
@@ -222,6 +223,8 @@ void __fastcall gsWrite64_generic( u32 mem, const mem64_t* value )
 
 void __fastcall gsWrite64_page_00( u32 mem, const mem64_t* value )
 {
+	s_GSRegistersWritten |= (mem == GS_DISPFB1 || mem == GS_DISPFB2 || mem == GS_PMODE);
+
 	gsWrite64_generic( mem, value );
 }
 
@@ -433,7 +436,9 @@ void gsPostVsyncStart()
 {
 	//gifUnit.FlushToMTGS();  // Needed for some (broken?) homebrew game loaders
 	
-	GetMTGS().PostVsyncStart();
+	const bool registers_written = s_GSRegistersWritten;
+	s_GSRegistersWritten = false;
+	GetMTGS().PostVsyncStart(registers_written);
 }
 
 void _gs_ResetFrameskip()

pcsx2/GS.h

@@ -385,7 +385,7 @@ public:
 
 	u8* GetDataPacketPtr() const;
 	void SetEvent();
-	void PostVsyncStart();
+	void PostVsyncStart(bool registers_written);
 
 	bool IsGSOpened() const { return m_Opened; }
 

pcsx2/GS/GS.cpp

@@ -464,11 +464,11 @@ void GSgifTransfer3(u8* mem, u32 size)
 	}
 }
 
-void GSvsync(u32 field)
+void GSvsync(u32 field, bool registers_written)
 {
 	try
 	{
-		s_gs->VSync(field);
+		s_gs->VSync(field, registers_written);
 	}
 	catch (GSRecoverableError)
 	{

pcsx2/GS/GS.h

@@ -68,7 +68,7 @@ void GSgifTransfer(const u8* mem, u32 size);
 void GSgifTransfer1(u8* mem, u32 addr);
 void GSgifTransfer2(u8* mem, u32 size);
 void GSgifTransfer3(u8* mem, u32 size);
-void GSvsync(u32 field);
+void GSvsync(u32 field, bool registers_written);
 u32 GSmakeSnapshot(char* path);
 int GSfreeze(FreezeAction mode, freezeData* data);
 #ifndef PCSX2_CORE

pcsx2/GS/GSPerfMon.h

@@ -52,6 +52,7 @@ protected:
 	u64 m_frame;
 	clock_t m_lastframe;
 	int m_count;
+	int m_disp_fb_sprite_blits;
 
 	friend class GSPerfMonAutoTimer;
 
@@ -69,6 +70,14 @@ public:
 
 	void Start(int timer = Main);
 	void Stop(int timer = Main);
+
+	__fi void AddDisplayFramebufferSpriteBlit() { m_disp_fb_sprite_blits++; }
+	__fi int GetDisplayFramebufferSpriteBlits()
+	{
+		const int blits = m_disp_fb_sprite_blits;
+		m_disp_fb_sprite_blits = 0;
+		return blits;
+	}
 };
 
 class GSPerfMonAutoTimer

pcsx2/GS/GSState.cpp

@@ -1406,6 +1406,16 @@ void GSState::FlushPrim()
 	{
 		GL_REG("FlushPrim ctxt %d", PRIM->CTXT);
 
+		// internal frame rate detection based on sprite blits to the display framebuffer
+		{
+			const u32 FRAME_FBP = m_context->FRAME.FBP;
+			if ((m_regs->DISP[0].DISPFB.FBP == FRAME_FBP && m_regs->PMODE.EN1) ||
+				(m_regs->DISP[1].DISPFB.FBP == FRAME_FBP && m_regs->PMODE.EN2))
+			{
+				g_perfmon.AddDisplayFramebufferSpriteBlit();
+			}
+		}
+
 		GSVertex buff[2];
 		s_n++;
 

pcsx2/GS/Renderers/Common/GSRenderer.cpp

@@ -388,7 +388,7 @@ static GSVector4 CalculateDrawRect(s32 window_width, s32 window_height, s32 text
 	return ret;
 }
 
-void GSRenderer::VSync(u32 field)
+void GSRenderer::VSync(u32 field, bool registers_written)
 {
 	GSPerfMonAutoTimer pmat(&g_perfmon);
 
@@ -399,6 +399,10 @@ void GSRenderer::VSync(u32 field)
 		m_regs->Dump(root_sw + format("%05d_f%lld_gs_reg.txt", s_n, g_perfmon.GetFrame()));
 	}
 
+	const int fb_sprite_blits = g_perfmon.GetDisplayFramebufferSpriteBlits();
+	const bool fb_sprite_frame = (fb_sprite_blits > 0);
+	PerformanceMetrics::Update(registers_written, fb_sprite_frame);
+
 	g_gs_device->AgePool();
 
 	const bool blank_frame = !Merge(field ? 1 : 0);

pcsx2/GS/Renderers/Common/GSRenderer.h

@@ -44,7 +44,7 @@ public:
 
 	virtual void Destroy();
 
-	virtual void VSync(u32 field);
+	virtual void VSync(u32 field, bool registers_written);
 	virtual bool MakeSnapshot(const std::string& path);
 	virtual void KeyEvent(const HostKeyEvent& e);
 	virtual bool CanUpscale() { return false; }

pcsx2/GS/Renderers/HW/GSRendererHW.cpp

@@ -291,7 +291,7 @@ void GSRendererHW::Reset()
 	GSRenderer::Reset();
 }
 
-void GSRendererHW::VSync(u32 field)
+void GSRendererHW::VSync(u32 field, bool registers_written)
 {
 	//Check if the frame buffer width or display width has changed
 	SetScaling();
@@ -303,7 +303,7 @@ void GSRendererHW::VSync(u32 field)
 		m_reset = false;
 	}
 
-	GSRenderer::VSync(field);
+	GSRenderer::VSync(field, registers_written);
 
 	m_tc->IncAge();
 

pcsx2/GS/Renderers/HW/GSRendererHW.h

@@ -182,7 +182,7 @@ public:
 	GSVector2 GetTextureScaleFactor() override;
 
 	void Reset() override;
-	void VSync(u32 field) override;
+	void VSync(u32 field, bool registers_written) override;
 
 	GSTexture* GetOutput(int i, int& y_offset) override;
 	GSTexture* GetFeedbackOutput() override;

pcsx2/GS/Renderers/SW/GSRendererSW.cpp

@@ -91,7 +91,7 @@ void GSRendererSW::Reset()
 	GSRenderer::Reset();
 }
 
-void GSRendererSW::VSync(u32 field)
+void GSRendererSW::VSync(u32 field, bool registers_written)
 {
 	Sync(0); // IncAge might delete a cached texture in use
 
@@ -128,7 +128,7 @@ void GSRendererSW::VSync(u32 field)
 	//
 	*/
 
-	GSRenderer::VSync(field);
+	GSRenderer::VSync(field, registers_written);
 
 	m_tc->IncAge();
 

pcsx2/GS/Renderers/SW/GSRendererSW.h

@@ -80,7 +80,7 @@ protected:
 	std::atomic<u16> m_tex_pages[512];
 
 	void Reset() override;
-	void VSync(u32 field) override;
+	void VSync(u32 field, bool registers_written) override;
 	GSTexture* GetOutput(int i, int& y_offset) override;
 	GSTexture* GetFeedbackOutput() override;
 

pcsx2/MTGS.cpp

@@ -129,9 +129,13 @@ struct RingCmdPacket_Vsync
 	u32 csr;
 	u32 imr;
 	GSRegSIGBLID siglblid;
+
+	// must be 16 byte aligned
+	u32 registers_written;
+	u32 pad[3];
 };
 
-void SysMtgsThread::PostVsyncStart()
+void SysMtgsThread::PostVsyncStart(bool registers_written)
 {
 	// Optimization note: Typically regset1 isn't needed.  The regs in that area are typically
 	// changed infrequently, usually during video mode changes.  However, on modern systems the
@@ -146,7 +150,8 @@ void SysMtgsThread::PostVsyncStart()
 	remainder[0] = GSCSRr;
 	remainder[1] = GSIMR._u32;
 	(GSRegSIGBLID&)remainder[2] = GSSIGLBLID;
-	m_packet_writepos = (m_packet_writepos + 1) & RingBufferMask;
+	remainder[4] = static_cast<u32>(registers_written);
+	m_packet_writepos = (m_packet_writepos + 2) & RingBufferMask;
 
 	SendDataPacket();
 
@@ -436,7 +441,7 @@ void SysMtgsThread::ExecuteTaskInThread()
 							((GSRegSIGBLID&)RingBuffer.Regs[0x1080]) = (GSRegSIGBLID&)remainder[2];
 
 							// CSR & 0x2000; is the pageflip id.
-							GSvsync(((u32&)RingBuffer.Regs[0x1000]) & 0x2000);
+							GSvsync(((u32&)RingBuffer.Regs[0x1000]) & 0x2000, remainder[4] != 0);
 							gsFrameSkip();
 
 							m_QueuedFrameCount.fetch_sub(1);

pcsx2/PerformanceMetrics.cpp

@@ -28,6 +28,7 @@ static const float UPDATE_INTERVAL = 0.5f;
 
 static float s_vertical_frequency = 0.0f;
 static float s_fps = 0.0f;
+static float s_internal_fps = 0.0f;
 static float s_worst_frame_time = 0.0f;
 static float s_average_frame_time = 0.0f;
 static float s_average_frame_time_accumulator = 0.0f;
@@ -36,6 +37,14 @@ static u32 s_frames_since_last_update = 0;
 static Common::Timer s_last_update_time;
 static Common::Timer s_last_frame_time;
 
+// frame number, updated by the GS thread
+static u64 s_frame_number = 0;
+
+// internal fps heuristics
+static PerformanceMetrics::InternalFPSMethod s_internal_fps_method = PerformanceMetrics::InternalFPSMethod::None;
+static u32 s_gs_framebuffer_blits_since_last_update = 0;
+static u32 s_gs_privileged_register_writes_since_last_update = 0;
+
 static Common::ThreadCPUTimer s_cpu_thread_timer;
 static u64 s_last_gs_time = 0;
 static u64 s_last_vu_time = 0;
@@ -53,8 +62,10 @@ void PerformanceMetrics::Clear()
 	Reset();
 
 	s_fps = 0.0f;
+	s_internal_fps = 0.0f;
 	s_worst_frame_time = 0.0f;
 	s_average_frame_time = 0.0f;
+	s_internal_fps_method = PerformanceMetrics::InternalFPSMethod::None;
 
 	s_cpu_thread_usage = 0.0f;
 	s_cpu_thread_time = 0.0f;
@@ -62,10 +73,15 @@ void PerformanceMetrics::Clear()
 	s_gs_thread_time = 0.0f;
 	s_vu_thread_usage = 0.0f;
 	s_vu_thread_time = 0.0f;
+
+	s_frame_number = 0;
 }
 
 void PerformanceMetrics::Reset()
 {
+	s_frames_since_last_update = 0;
+	s_gs_framebuffer_blits_since_last_update = 0;
+	s_gs_privileged_register_writes_since_last_update = 0;
 	s_average_frame_time_accumulator = 0.0f;
 	s_worst_frame_time_accumulator = 0.0f;
 
@@ -78,12 +94,15 @@ void PerformanceMetrics::Reset()
 	s_last_ticks = GetCPUTicks();
 }
 
-void PerformanceMetrics::Update()
+void PerformanceMetrics::Update(bool gs_register_write, bool fb_blit)
 {
 	const float frame_time = s_last_frame_time.GetTimeMillisecondsAndReset();
 	s_average_frame_time_accumulator += frame_time;
 	s_worst_frame_time_accumulator = std::max(s_worst_frame_time_accumulator, frame_time);
 	s_frames_since_last_update++;
+	s_gs_privileged_register_writes_since_last_update += static_cast<u32>(gs_register_write);
+	s_gs_framebuffer_blits_since_last_update += static_cast<u32>(fb_blit);
+	s_frame_number++;
 
 	const Common::Timer::Value now_ticks = Common::Timer::GetCurrentValue();
 	const Common::Timer::Value ticks_diff = now_ticks - s_last_update_time.GetStartValue();
@@ -97,6 +116,26 @@ void PerformanceMetrics::Update()
 	s_average_frame_time_accumulator = 0.0f;
 	s_fps = static_cast<float>(s_frames_since_last_update) / time;
 
+	// prefer privileged register write based framerate detection, it's less likely to have false positives
+	if (s_gs_privileged_register_writes_since_last_update > 0)
+	{
+		s_internal_fps = static_cast<float>(s_gs_privileged_register_writes_since_last_update) / time;
+		s_internal_fps_method = InternalFPSMethod::GSPrivilegedRegister;
+	}
+	else if (s_gs_framebuffer_blits_since_last_update > 0)
+	{
+		s_internal_fps = static_cast<float>(s_gs_framebuffer_blits_since_last_update) / time;
+		s_internal_fps_method = InternalFPSMethod::DISPFBBlit;
+	}
+	else
+	{
+		s_internal_fps = 0;
+		s_internal_fps_method = InternalFPSMethod::None;
+	}
+
+	s_gs_privileged_register_writes_since_last_update = 0;
+	s_gs_framebuffer_blits_since_last_update = 0;
+
 	s_cpu_thread_timer.GetUsageInMillisecondsAndReset(ticks_diff, &s_cpu_thread_time, &s_cpu_thread_usage);
 	s_cpu_thread_time /= static_cast<double>(s_frames_since_last_update);
 
@@ -137,11 +176,31 @@ void PerformanceMetrics::SetVerticalFrequency(float rate)
 	s_vertical_frequency = rate;
 }
 
+u64 PerformanceMetrics::GetFrameNumber()
+{
+	return s_frame_number;
+}
+
+PerformanceMetrics::InternalFPSMethod PerformanceMetrics::GetInternalFPSMethod()
+{
+	return s_internal_fps_method;
+}
+
+bool PerformanceMetrics::IsInternalFPSValid()
+{
+	return s_internal_fps_method != InternalFPSMethod::None;
+}
+
 float PerformanceMetrics::GetFPS()
 {
 	return s_fps;
 }
 
+float PerformanceMetrics::GetInternalFPS()
+{
+	return s_internal_fps;
+}
+
 float PerformanceMetrics::GetSpeed()
 {
 	return (s_fps / s_vertical_frequency) * 100.0;

pcsx2/PerformanceMetrics.h

@@ -18,9 +18,16 @@
 
 namespace PerformanceMetrics
 {
+	enum class InternalFPSMethod
+	{
+		None,
+		GSPrivilegedRegister,
+		DISPFBBlit
+	};
+
 	void Clear();
 	void Reset();
-	void Update();
+	void Update(bool gs_register_write, bool fb_blit);
 
 	/// Sets the EE thread for CPU usage calculations.
 	void SetCPUThreadTimer(Common::ThreadCPUTimer timer);
@@ -28,7 +35,13 @@ namespace PerformanceMetrics
 	/// Sets the vertical frequency, used in speed calculations.
 	void SetVerticalFrequency(float rate);
 
+	u64 GetFrameNumber();
+
+	InternalFPSMethod GetInternalFPSMethod();
+	bool IsInternalFPSValid();
+
 	float GetFPS();
+	float GetInternalFPS();
 	float GetSpeed();
 	float GetAverageFrameTime();
 	float GetWorstFrameTime();

pcsx2/gui/Dialogs/GSDumpDialog.cpp

@@ -662,8 +662,7 @@ void Dialogs::GSDumpDialog::ProcessDumpEvent(const GSData& event, char* regs)
 		}
 		case VSync:
 		{
-			GSvsync((*((int*)(regs + 4096)) & 0x2000) > 0 ? (u8)1 : (u8)0);
+			GSvsync((*((int*)(regs + 4096)) & 0x2000) > 0 ? (u8)1 : (u8)0, false);
-			PerformanceMetrics::Update();
 			g_FrameCount++;
 			break;
 		}
@@ -813,7 +812,7 @@ void Dialogs::GSDumpDialog::GSThread::ExecuteTaskInThread()
 
 	if (GSfreeze(FreezeAction::Load, &fd))
 		GSDump::isRunning = false;
-	GSvsync(1);
+	GSvsync(1, false);
 	GSreset();
 	GSfreeze(FreezeAction::Load, &fd);