GPU: Use old frame when presenting a skip.

If we flip using a skipped frame, we may show an even older frame causing
weird flickering.

Core/HLE/sceDisplay.cpp

@@ -781,7 +781,7 @@ void __DisplayFlip(int cyclesLate) {
 			// Check first though, might've just quit / been paused.
 			if (coreState == CORE_RUNNING) {
 				coreState = CORE_NEXTFRAME;
-				gpu->CopyDisplayToOutput();
+				gpu->CopyDisplayToOutput(fbReallyDirty);
 				if (fbReallyDirty) {
 					actualFlips++;
 				}

GPU/Common/FramebufferCommon.cpp

@@ -846,7 +846,7 @@ void FramebufferManagerCommon::SetViewport2D(int x, int y, int w, int h) {
 	draw_->SetViewports(1, &vp);
 }
 
-void FramebufferManagerCommon::CopyDisplayToOutput() {
+void FramebufferManagerCommon::CopyDisplayToOutput(bool reallyDirty) {
 	DownloadFramebufferOnSwitch(currentRenderVfb_);
 	shaderManager_->DirtyLastShader();
 
@@ -869,12 +869,16 @@ void FramebufferManagerCommon::CopyDisplayToOutput() {
 	CardboardSettings cardboardSettings;
 	GetCardboardSettings(&cardboardSettings);
 
-	VirtualFramebuffer *vfb = GetVFBAt(displayFramebufPtr_);
+	// If it's not really dirty, we're probably frameskipping.  Use the last working one.
+	u32 fbaddr = reallyDirty ? displayFramebufPtr_ : prevDisplayFramebufPtr_;
+	prevDisplayFramebufPtr_ = fbaddr;
+
+	VirtualFramebuffer *vfb = GetVFBAt(fbaddr);
 	if (!vfb) {
 		// Let's search for a framebuf within this range. Note that we also look for
 		// "framebuffers" sitting in RAM (created from block transfer or similar) so we only take off the kernel
 		// and uncached bits of the address when comparing.
-		const u32 addr = displayFramebufPtr_ & 0x3FFFFFFF;
+		const u32 addr = fbaddr & 0x3FFFFFFF;
 		for (size_t i = 0; i < vfbs_.size(); ++i) {
 			VirtualFramebuffer *v = vfbs_[i];
 			const u32 v_addr = v->fb_address & 0x3FFFFFFF;
@@ -912,7 +916,7 @@ void FramebufferManagerCommon::CopyDisplayToOutput() {
 	}
 
 	if (!vfb) {
-		if (Memory::IsValidAddress(displayFramebufPtr_)) {
+		if (Memory::IsValidAddress(fbaddr)) {
 			// The game is displaying something directly from RAM. In GTA, it's decoded video.
 			if (!vfb) {
 				shaderManager_->DirtyLastShader();
@@ -923,12 +927,12 @@ void FramebufferManagerCommon::CopyDisplayToOutput() {
 				// Just a pointer to plain memory to draw. We should create a framebuffer, then draw to it.
 				SetViewport2D(0, 0, pixelWidth_, pixelHeight_);
 				draw_->SetScissorRect(0, 0, pixelWidth_, pixelHeight_);
-				DrawFramebufferToOutput(Memory::GetPointer(displayFramebufPtr_), displayFormat_, displayStride_, true);
+				DrawFramebufferToOutput(Memory::GetPointer(fbaddr), displayFormat_, displayStride_, true);
 				gstate_c.Dirty(DIRTY_BLEND_STATE);
 				return;
 			}
 		} else {
-			DEBUG_LOG(FRAMEBUF, "Found no FBO to display! displayFBPtr = %08x", displayFramebufPtr_);
+			DEBUG_LOG(FRAMEBUF, "Found no FBO to display! displayFBPtr = %08x", fbaddr);
 			// No framebuffer to display! Clear to black.
 			if (useBufferedRendering_) {
 				shaderManager_->DirtyLastShader();

GPU/Common/FramebufferCommon.h

@@ -228,7 +228,7 @@ public:
 	void RebindFramebuffer();
 	std::vector<FramebufferInfo> GetFramebufferList();
 
-	void CopyDisplayToOutput();
+	void CopyDisplayToOutput(bool reallyDirty);
 
 	bool NotifyFramebufferCopy(u32 src, u32 dest, int size, bool isMemset, u32 skipDrawReason);
 	void NotifyVideoUpload(u32 addr, int size, int width, GEBufferFormat fmt);
@@ -383,6 +383,7 @@ protected:
 	u32 displayFramebufPtr_ = 0;
 	u32 displayStride_ = 0;
 	GEBufferFormat displayFormat_;
+	u32 prevDisplayFramebufPtr_ = 0;
 
 	VirtualFramebuffer *displayFramebuf_ = nullptr;
 	VirtualFramebuffer *prevDisplayFramebuf_ = nullptr;

GPU/D3D11/GPU_D3D11.cpp

@@ -262,13 +262,13 @@ void GPU_D3D11::SetDisplayFramebuffer(u32 framebuf, u32 stride, GEBufferFormat f
 	framebufferManagerD3D11_->SetDisplayFramebuffer(framebuf, stride, format);
 }
 
-void GPU_D3D11::CopyDisplayToOutput() {
+void GPU_D3D11::CopyDisplayToOutput(bool reallyDirty) {
 	float blendColor[4]{};
 	context_->OMSetBlendState(stockD3D11.blendStateDisabledWithColorMask[0xF], blendColor, 0xFFFFFFFF);
 
 	drawEngine_.Flush();
 
-	framebufferManagerD3D11_->CopyDisplayToOutput();
+	framebufferManagerD3D11_->CopyDisplayToOutput(reallyDirty);
 	framebufferManagerD3D11_->EndFrame();
 
 	// shaderManager_->EndFrame();

GPU/D3D11/GPU_D3D11.h

@@ -71,7 +71,7 @@ private:
 
 	void InitClear() override;
 	void BeginFrame() override;
-	void CopyDisplayToOutput() override;
+	void CopyDisplayToOutput(bool reallyDirty) override;
 
 	ID3D11Device *device_;
 	ID3D11DeviceContext *context_;

GPU/Directx9/GPU_DX9.cpp

@@ -299,13 +299,13 @@ void GPU_DX9::SetDisplayFramebuffer(u32 framebuf, u32 stride, GEBufferFormat for
 	framebufferManagerDX9_->SetDisplayFramebuffer(framebuf, stride, format);
 }
 
-void GPU_DX9::CopyDisplayToOutput() {
+void GPU_DX9::CopyDisplayToOutput(bool reallyDirty) {
 	dxstate.depthWrite.set(true);
 	dxstate.colorMask.set(true, true, true, true);
 
 	drawEngine_.Flush();
 
-	framebufferManagerDX9_->CopyDisplayToOutput();
+	framebufferManagerDX9_->CopyDisplayToOutput(reallyDirty);
 	framebufferManagerDX9_->EndFrame();
 
 	// shaderManager_->EndFrame();

GPU/Directx9/GPU_DX9.h

@@ -70,7 +70,7 @@ private:
 
 	void InitClear() override;
 	void BeginFrame() override;
-	void CopyDisplayToOutput() override;
+	void CopyDisplayToOutput(bool reallyDirty) override;
 
 	LPDIRECT3DDEVICE9 device_;
 	LPDIRECT3DDEVICE9EX deviceEx_;

GPU/GLES/GPU_GLES.cpp

@@ -400,14 +400,14 @@ void GPU_GLES::SetDisplayFramebuffer(u32 framebuf, u32 stride, GEBufferFormat fo
 	framebufferManagerGL_->SetDisplayFramebuffer(framebuf, stride, format);
 }
 
-void GPU_GLES::CopyDisplayToOutput() {
+void GPU_GLES::CopyDisplayToOutput(bool reallyDirty) {
 	// Flush anything left over.
 	framebufferManagerGL_->RebindFramebuffer();
 	drawEngine_.Flush();
 
 	shaderManagerGL_->DirtyLastShader();
 
-	framebufferManagerGL_->CopyDisplayToOutput();
+	framebufferManagerGL_->CopyDisplayToOutput(reallyDirty);
 	framebufferManagerGL_->EndFrame();
 
 	// If buffered, discard the depth buffer of the backbuffer. Don't even know if we need one.

GPU/GLES/GPU_GLES.h

@@ -76,7 +76,7 @@ private:
 
 	void InitClear() override;
 	void BeginFrame() override;
-	void CopyDisplayToOutput() override;
+	void CopyDisplayToOutput(bool reallyDirty) override;
 	void Reinitialize() override;
 
 	FramebufferManagerGLES *framebufferManagerGL_;

GPU/GPUCommon.h

@@ -111,7 +111,7 @@ public:
 	u32  Break(int mode) override;
 	void ReapplyGfxState() override;
 
-	void CopyDisplayToOutput() override = 0;
+	void CopyDisplayToOutput(bool reallyDirty) override = 0;
 	void InitClear() override = 0;
 	bool PerformMemoryCopy(u32 dest, u32 src, int size) override;
 	bool PerformMemorySet(u32 dest, u8 v, int size) override;

GPU/GPUInterface.h

@@ -200,7 +200,7 @@ public:
 	// Framebuffer management
 	virtual void SetDisplayFramebuffer(u32 framebuf, u32 stride, GEBufferFormat format) = 0;
 	virtual void BeginFrame() = 0;  // Can be a good place to draw the "memory" framebuffer for accelerated plugins
-	virtual void CopyDisplayToOutput() = 0;
+	virtual void CopyDisplayToOutput(bool reallyDirty) = 0;
 
 	// Tells the GPU to update the gpuStats structure.
 	virtual void GetStats(char *buffer, size_t bufsize) = 0;

GPU/Null/NullGpu.h

@@ -32,7 +32,7 @@ public:
 	void ExecuteOp(u32 op, u32 diff) override;
 
 	void SetDisplayFramebuffer(u32 framebuf, u32 stride, GEBufferFormat format) override {}
-	void CopyDisplayToOutput() override {}
+	void CopyDisplayToOutput(bool reallyDirty) override {}
 	void GetStats(char *buffer, size_t bufsize) override;
 	void InvalidateCache(u32 addr, int size, GPUInvalidationType type) override;
 	void NotifyVideoUpload(u32 addr, int size, int width, int format) override;

GPU/Software/SoftGpu.cpp

@@ -353,7 +353,7 @@ void SoftGPU::CopyToCurrentFboFromDisplayRam(int srcwidth, int srcheight) {
 	draw_->BindIndexBuffer(nullptr, 0);
 }
 
-void SoftGPU::CopyDisplayToOutput() {
+void SoftGPU::CopyDisplayToOutput(bool reallyDirty) {
 	// The display always shows 480x272.
 	CopyToCurrentFboFromDisplayRam(FB_WIDTH, FB_HEIGHT);
 	framebufferDirty_ = false;

GPU/Software/SoftGpu.h

@@ -62,7 +62,7 @@ public:
 	void ExecuteOp(u32 op, u32 diff) override;
 
 	void SetDisplayFramebuffer(u32 framebuf, u32 stride, GEBufferFormat format) override;
-	void CopyDisplayToOutput() override;
+	void CopyDisplayToOutput(bool reallyDirty) override;
 	void GetStats(char *buffer, size_t bufsize) override;
 	void InvalidateCache(u32 addr, int size, GPUInvalidationType type) override;
 	void NotifyVideoUpload(u32 addr, int size, int width, int format) override;

GPU/Vulkan/GPU_Vulkan.cpp

@@ -431,13 +431,13 @@ void GPU_Vulkan::SetDisplayFramebuffer(u32 framebuf, u32 stride, GEBufferFormat
 	framebufferManager_->SetDisplayFramebuffer(framebuf, stride, format);
 }
 
-void GPU_Vulkan::CopyDisplayToOutput() {
+void GPU_Vulkan::CopyDisplayToOutput(bool reallyDirty) {
 	// Flush anything left over.
 	drawEngine_.Flush();
 
 	shaderManagerVulkan_->DirtyLastShader();
 
-	framebufferManagerVulkan_->CopyDisplayToOutput();
+	framebufferManagerVulkan_->CopyDisplayToOutput(reallyDirty);
 
 	gstate_c.Dirty(DIRTY_TEXTURE_IMAGE);
 }

GPU/Vulkan/GPU_Vulkan.h

@@ -78,7 +78,7 @@ private:
 	void CheckFlushOp(int cmd, u32 diff);
 	void BuildReportingInfo();
 	void InitClear() override;
-	void CopyDisplayToOutput() override;
+	void CopyDisplayToOutput(bool reallyDirty) override;
 	void Reinitialize() override;
 
 	void InitDeviceObjects();

UI/EmuScreen.cpp

@@ -1315,7 +1315,7 @@ void EmuScreen::render() {
 		thin3d->BindFramebufferAsRenderTarget(nullptr, { RPAction::CLEAR, RPAction::DONT_CARE, RPAction::DONT_CARE });
 		// Just to make sure.
 		if (PSP_IsInited()) {
-			gpu->CopyDisplayToOutput();
+			gpu->CopyDisplayToOutput(true);
 		}
 	} else {
 		// Didn't actually reach the end of the frame, ran out of the blockTicks cycles.