More work towards unifying CopyDisplayToOutput

GPU/Common/FramebufferCommon.cpp

@@ -109,6 +109,7 @@ void CenterDisplayOutputRect(float *x, float *y, float *w, float *h, float origW
 
 FramebufferManagerCommon::FramebufferManagerCommon(Draw::DrawContext *draw) :
 	draw_(draw),
+	needBackBufferYSwap_(false),
 	displayFramebufPtr_(0),
 	displayStride_(0),
 	displayFormat_(GE_FORMAT_565),

GPU/Common/FramebufferCommon.h

@@ -260,7 +260,9 @@ public:
 protected:
 	virtual void SetViewport2D(int x, int y, int w, int h) = 0;
 	void CalculatePostShaderUniforms(int bufferWidth, int bufferHeight, int renderWidth, int renderHeight, PostShaderUniforms *uniforms);
+	virtual void MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height) = 0;
 	virtual void DrawActiveTexture(float x, float y, float w, float h, float destW, float destH, float u0, float v0, float u1, float v1, int uvRotation, bool linearFilter) = 0;
+	virtual void Bind2DShader() = 0;
 	virtual void BindPostShader(const PostShaderUniforms &uniforms) {}
 
 	// Cardboard Settings Calculator
@@ -314,6 +316,7 @@ protected:
 
 	Draw::DrawContext *draw_;
 	TextureCacheCommon *textureCache_;
+	bool needBackBufferYSwap_;
 
 	u32 displayFramebufPtr_;
 	u32 displayStride_;

GPU/D3D11/FramebufferManagerD3D11.cpp

@@ -238,6 +238,7 @@ void FramebufferManagerD3D11::MakePixelTexture(const u8 *srcPixels, GEBufferForm
 	}
 
 	context_->Unmap(drawPixelsTex_, 0);
+	context_->PSSetShaderResources(0, 1, &drawPixelsTexView_);
 	// D3DXSaveTextureToFile("game:\\cc.png", D3DXIFF_PNG, drawPixelsTex_, NULL);
 }
 
@@ -259,16 +260,15 @@ void FramebufferManagerD3D11::DrawPixels(VirtualFramebuffer *vfb, int dstX, int
 	}
 	MakePixelTexture(srcPixels, srcPixelFormat, srcStride, width, height);
 	DisableState();
-	context_->PSSetShaderResources(0, 1, &drawPixelsTexView_);
+	Bind2DShader();
 	DrawActiveTexture(dstX, dstY, width, height, vfb->bufferWidth, vfb->bufferHeight, 0.0f, 0.0f, 1.0f, 1.0f, ROTATION_LOCKED_HORIZONTAL, true);
 	textureCacheD3D11_->ForgetLastTexture();
-	//	context_->RSSetViewports(1, &vp);
+	shaderManager_->DirtyLastShader();
 }
 
 void FramebufferManagerD3D11::DrawFramebufferToOutput(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, bool applyPostShader) {
 	MakePixelTexture(srcPixels, srcPixelFormat, srcStride, 512, 272);
 	DisableState();
-
 	// This might draw directly at the backbuffer (if so, applyPostShader is set) so if there's a post shader, we need to apply it here.
 	// Should try to unify this path with the regular path somehow, but this simple solution works for most of the post shaders
 	// (it always runs at output resolution so FXAA may look odd).
@@ -276,7 +276,10 @@ void FramebufferManagerD3D11::DrawFramebufferToOutput(const u8 *srcPixels, GEBuf
 	int uvRotation = (g_Config.iRenderingMode != FB_NON_BUFFERED_MODE) ? g_Config.iInternalScreenRotation : ROTATION_LOCKED_HORIZONTAL;
 	CenterDisplayOutputRect(&x, &y, &w, &h, 480.0f, 272.0f, (float)pixelWidth_, (float)pixelHeight_, uvRotation);
 	context_->PSSetShaderResources(0, 1, &drawPixelsTexView_);
+	Bind2DShader();
 	DrawActiveTexture(x, y, w, h, (float)pixelWidth_, (float)pixelHeight_, 0.0f, 0.0f, 480.0f / 512.0f, 1.0f, uvRotation, g_Config.iBufFilter == SCALE_LINEAR);
+	textureCacheD3D11_->ForgetLastTexture();
+	shaderManager_->DirtyLastShader();
 }
 
 void FramebufferManagerD3D11::DrawActiveTexture(float x, float y, float w, float h, float destW, float destH, float u0, float v0, float u1, float v1, int uvRotation, bool linearFilter) {
@@ -331,17 +334,22 @@ void FramebufferManagerD3D11::DrawActiveTexture(float x, float y, float w, float
 	context_->RSSetState(stockD3D11.rasterStateNoCull);
 	context_->OMSetBlendState(stockD3D11.blendStateDisabledWithColorMask[0xF], nullptr, 0xFFFFFFFF);
 	context_->OMSetDepthStencilState(stockD3D11.depthStencilDisabled, 0);
-	context_->IASetInputLayout(quadInputLayout_);
-	context_->PSSetShader(quadPixelShader_, 0, 0);
-	context_->VSSetShader(quadVertexShader_, 0, 0);
 	context_->IASetPrimitiveTopology(D3D11_PRIMITIVE_TOPOLOGY_TRIANGLESTRIP);
 	context_->PSSetSamplers(0, 1, linearFilter ? &stockD3D11.samplerLinear2DClamp : &stockD3D11.samplerPoint2DClamp);
 	UINT stride = 20;
 	UINT offset = 0;
 	context_->IASetVertexBuffers(0, 1, &quadBuffer_, &stride, &offset);
 	context_->Draw(4, 0);
+}
+
+void FramebufferManagerD3D11::Bind2DShader() {
+	context_->IASetInputLayout(quadInputLayout_);
+	context_->PSSetShader(quadPixelShader_, 0, 0);
+	context_->VSSetShader(quadVertexShader_, 0, 0);
+}
+
+void FramebufferManagerD3D11::BindPostShader(const PostShaderUniforms &uniforms) {
 
-	shaderManager_->DirtyLastShader();
 }
 
 void FramebufferManagerD3D11::RebindFramebuffer() {
@@ -603,6 +611,7 @@ void FramebufferManagerD3D11::CopyDisplayToOutput() {
 		if (1) {
 			const u32 rw = PSP_CoreParameter().pixelWidth;
 			const u32 rh = PSP_CoreParameter().pixelHeight;
+			Bind2DShader();
 			DrawActiveTexture(x, y, w, h, (float)PSP_CoreParameter().pixelWidth, (float)PSP_CoreParameter().pixelHeight, u0, v0, u1, v1, uvRotation, g_Config.iBufFilter == SCALE_LINEAR);
 		}
 

GPU/D3D11/FramebufferManagerD3D11.h

@@ -100,7 +100,9 @@ protected:
 	void UpdateDownloadTempBuffer(VirtualFramebuffer *nvfb) override;
 
 private:
-	void MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height);
+	void BindPostShader(const PostShaderUniforms &uniforms) override;
+	void Bind2DShader() override;
+	void MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height) override;
 	void PackFramebufferD3D11_(VirtualFramebuffer *vfb, int x, int y, int w, int h);
 	void PackDepthbuffer(VirtualFramebuffer *vfb, int x, int y, int w, int h);
 

GPU/Directx9/FramebufferDX9.cpp

@@ -279,6 +279,7 @@ static void DXSetViewport(float x, float y, float w, float h, float minZ, float
 		device_->SetTexture(0, drawPixelsTex_);
 		DrawActiveTexture(dstX, dstY, width, height, vfb->bufferWidth, vfb->bufferHeight, 0.0f, 0.0f, 1.0f, 1.0f, ROTATION_LOCKED_HORIZONTAL, true);
 		textureCache_->ForgetLastTexture();
+		shaderManager_->DirtyLastShader();
 		dxstate.viewport.restore();
 	}
 
@@ -294,6 +295,8 @@ static void DXSetViewport(float x, float y, float w, float h, float minZ, float
 		CenterDisplayOutputRect(&x, &y, &w, &h, 480.0f, 272.0f, (float)pixelWidth_, (float)pixelHeight_, uvRotation);
 		device_->SetTexture(0, drawPixelsTex_);
 		DrawActiveTexture(x, y, w, h, (float)pixelWidth_, (float)pixelHeight_, 0.0f, 0.0f, 480.0f / 512.0f, 1.0f, uvRotation, true);
+		textureCache_->ForgetLastTexture();
+		shaderManager_->DirtyLastShader();
 	}
 
 	void FramebufferManagerDX9::SetViewport2D(int x, int y, int w, int h) {
@@ -349,10 +352,6 @@ static void DXSetViewport(float x, float y, float w, float h, float minZ, float
 			dxstate.texMinFilter.set(D3DTEXF_POINT);
 		}
 		pD3Ddevice->SetRenderState(D3DRS_CULLMODE, D3DCULL_NONE);
-		pD3Ddevice->SetVertexDeclaration(pFramebufferVertexDecl);
-		pD3Ddevice->SetPixelShader(pFramebufferPixelShader);
-		pD3Ddevice->SetVertexShader(pFramebufferVertexShader);
-		shaderManager_->DirtyLastShader();
 		HRESULT hr = pD3Ddevice->DrawPrimitiveUP(D3DPT_TRIANGLEFAN, 2, coord, 5 * sizeof(float));
 		if (FAILED(hr)) {
 			ERROR_LOG_REPORT(G3D, "DrawActiveTexture() failed: %08x", hr);
@@ -367,6 +366,12 @@ static void DXSetViewport(float x, float y, float w, float h, float minZ, float
 		}
 	}
 
+	void FramebufferManagerDX9::Bind2DShader() {
+		pD3Ddevice->SetVertexDeclaration(pFramebufferVertexDecl);
+		pD3Ddevice->SetPixelShader(pFramebufferPixelShader);
+		pD3Ddevice->SetVertexShader(pFramebufferVertexShader);
+	}
+
 	void FramebufferManagerDX9::ReformatFramebufferFrom(VirtualFramebuffer *vfb, GEBufferFormat old) {
 		if (!useBufferedRendering_ || !vfb->fbo) {
 			return;
@@ -743,6 +748,7 @@ static void DXSetViewport(float x, float y, float w, float h, float minZ, float
 			DrawActiveTexture(colorTexture, x, y, w, h, (float)PSP_CoreParameter().pixelWidth, (float)PSP_CoreParameter().pixelHeight, true, 480.0f / (float)vfb->width, 272.0f / (float)vfb->height, postShaderProgram_);
 			}
 			*/
+			shaderManager_->DirtyLastShader();
 			pD3Ddevice->SetTexture(0, NULL);
 		}
 		dxstate.viewport.restore();

GPU/Directx9/FramebufferDX9.h

@@ -88,6 +88,7 @@ public:
 	LPDIRECT3DSURFACE9 GetOffscreenSurface(D3DFORMAT fmt, u32 w, u32 h);
 
 protected:
+	void Bind2DShader() override;
 	void SetViewport2D(int x, int y, int w, int h) override;
 	void DisableState() override;
 	void ClearBuffer(bool keepState = false) override;
@@ -101,7 +102,7 @@ protected:
 	void UpdateDownloadTempBuffer(VirtualFramebuffer *nvfb) override;
 
 private:
-	void MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height);
+	void MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height) override;
 	void PackFramebufferDirectx9_(VirtualFramebuffer *vfb, int x, int y, int w, int h);
 	void PackDepthbuffer(VirtualFramebuffer *vfb, int x, int y, int w, int h);
 	static bool GetRenderTargetFramebuffer(LPDIRECT3DSURFACE9 renderTarget, LPDIRECT3DSURFACE9 offscreen, int w, int h, GPUDebugBuffer &buffer);

GPU/GLES/FramebufferManagerGLES.cpp

@@ -181,6 +181,10 @@ void FramebufferManagerGLES::CompileDraw2DProgram() {
 	}
 }
 
+void FramebufferManagerGLES::Bind2DShader() {
+	glsl_bind(draw2dprogram_);
+}
+
 void FramebufferManagerGLES::BindPostShader(const PostShaderUniforms &uniforms) {
 	glsl_bind(postShaderProgram_);
 	if (deltaLoc_ != -1)
@@ -220,6 +224,7 @@ FramebufferManagerGLES::FramebufferManagerGLES(Draw::DrawContext *draw) :
 	pixelBufObj_(nullptr),
 	currentPBO_(0)
 {
+	needBackBufferYSwap_ = true;
 }
 
 void FramebufferManagerGLES::Init() {
@@ -356,7 +361,7 @@ void FramebufferManagerGLES::DrawPixels(VirtualFramebuffer *vfb, int dstX, int d
 	DisableState();
 
 	bool linearFilter = vfb || g_Config.iBufFilter == SCALE_LINEAR;
-	glsl_bind(draw2dprogram_);
+	Bind2DShader();
 	DrawActiveTexture(dstX, dstY, width, height, vfb->bufferWidth, vfb->bufferHeight, 0.0f, v0, 1.0f, v1, ROTATION_LOCKED_HORIZONTAL, linearFilter);
 }
 
@@ -385,16 +390,17 @@ void FramebufferManagerGLES::DrawFramebufferToOutput(const u8 *srcPixels, GEBuff
 			CalculatePostShaderUniforms(480, 272, renderWidth_, renderHeight_, &uniforms);
 			BindPostShader(uniforms);
 		} else {
-			glsl_bind(draw2dprogram_);
+			Bind2DShader();
 		}
 	} else {
-		glsl_bind(draw2dprogram_);
+		Bind2DShader();
 	}
 	float u0 = 0.0f, u1 = 480.0f / 512.0f;
 	float v0 = 0.0f, v1 = 1.0f;
 
 	// We are drawing directly to the back buffer.
-	std::swap(v0, v1);
+	if (needBackBufferYSwap_)
+		std::swap(v0, v1);
 
 	bool linearFilter = g_Config.iBufFilter == SCALE_LINEAR;
 	if (cardboardSettings.enabled) {
@@ -633,15 +639,7 @@ void FramebufferManagerGLES::CopyDisplayToOutput() {
 	}
 
 	if (useBufferedRendering_) {
-		glClearColor(0.0f, 0.0f, 0.0f, 0.0f);
-#ifdef USING_GLES2
-		glClearDepthf(0.0f);
-#else
-		glClearDepth(0.0);
-#endif
-		glClearStencil(0);
-		// Hardly necessary to clear depth and stencil I guess...
-		glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT);
+		draw_->Clear(Draw::FB_COLOR_BIT | Draw::FB_STENCIL_BIT | Draw::FB_DEPTH_BIT, 0, 0, 0);
 	}
 
 	u32 offsetX = 0;
@@ -750,9 +748,9 @@ void FramebufferManagerGLES::CopyDisplayToOutput() {
 			bool linearFilter = g_Config.iBufFilter == SCALE_LINEAR;
 			// We are doing the DrawActiveTexture call directly to the backbuffer here. Hence, we must
 			// flip V.
-			GLSLProgram *program = draw2dprogram_;
-			glsl_bind(program);
-			std::swap(v0, v1);
+			Bind2DShader();
+			if (needBackBufferYSwap_)
+				std::swap(v0, v1);
 			if (cardboardSettings.enabled) {
 				// Left Eye Image
 				SetViewport2D(cardboardSettings.leftEyeXPosition, cardboardSettings.screenYPosition, cardboardSettings.screenWidth, cardboardSettings.screenHeight);
@@ -791,8 +789,9 @@ void FramebufferManagerGLES::CopyDisplayToOutput() {
 
 			// We are doing the DrawActiveTexture call directly to the backbuffer after here. Hence, we must
 			// flip V.
-			std::swap(v0, v1);
-			glsl_bind(draw2dprogram_);
+			if (needBackBufferYSwap_)
+				std::swap(v0, v1);
+			Bind2DShader();
 			linearFilter = !postShaderIsUpscalingFilter_ && g_Config.iBufFilter == SCALE_LINEAR;
 			if (g_Config.bEnableCardboard) {
 				// Left Eye Image
@@ -816,7 +815,8 @@ void FramebufferManagerGLES::CopyDisplayToOutput() {
 		} else {
 			// We are doing the DrawActiveTexture call directly to the backbuffer here. Hence, we must
 			// flip V.
-			std::swap(v0, v1);
+			if (needBackBufferYSwap_)
+				std::swap(v0, v1);
 			bool linearFilter = g_Config.iBufFilter == SCALE_LINEAR;
 
 			shaderManager_->DirtyLastShader();  // dirty lastShader_

GPU/GLES/FramebufferManagerGLES.h

@@ -113,7 +113,8 @@ protected:
 	void UpdateDownloadTempBuffer(VirtualFramebuffer *nvfb) override;
 
 private:
-	void MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height);
+	void MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height) override;
+	void Bind2DShader() override;
 	void BindPostShader(const PostShaderUniforms &uniforms) override;
 	void CompileDraw2DProgram();
 	void DestroyDraw2DProgram();

GPU/Vulkan/FramebufferVulkan.cpp

@@ -330,7 +330,7 @@ void FramebufferManagerVulkan::Init() {
 	resized_ = true;
 }
 
-VulkanTexture *FramebufferManagerVulkan::MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height) {
+void FramebufferManagerVulkan::MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height) {
 	if (drawPixelsTex_ && (drawPixelsTexFormat_ != srcPixelFormat || drawPixelsTex_->GetWidth() != width || drawPixelsTex_->GetHeight() != height)) {
 		delete drawPixelsTex_;
 		drawPixelsTex_ = nullptr;
@@ -399,7 +399,6 @@ VulkanTexture *FramebufferManagerVulkan::MakePixelTexture(const u8 *srcPixels, G
 	size_t offset = frameData_[curFrame_].push_->Push(data, width * height * 4, &buffer);
 	drawPixelsTex_->UploadMip(0, width, height, buffer, (uint32_t)offset, width);
 	drawPixelsTex_->EndCreate();
-	return drawPixelsTex_;
 }
 
 void FramebufferManagerVulkan::SetViewport2D(int x, int y, int w, int h) {
@@ -428,13 +427,15 @@ void FramebufferManagerVulkan::DrawPixels(VirtualFramebuffer *vfb, int dstX, int
 	// TODO: Don't use the viewport mechanism for this.
 	vkCmdSetViewport(curCmd_, 0, 1, &vp);
 
-	VulkanTexture *pixelTex = MakePixelTexture(srcPixels, srcPixelFormat, srcStride, width, height);
+	MakePixelTexture(srcPixels, srcPixelFormat, srcStride, width, height);
+	VulkanTexture *pixelTex = drawPixelsTex_;
 	DrawTexture(pixelTex, dstX, dstY, width, height, vfb->bufferWidth, vfb->bufferHeight, 0.0f, 0.0f, 1.0f, 1.0f, pipelineBasicTex_, ROTATION_LOCKED_HORIZONTAL);
 	textureCacheVulkan_->ForgetLastTexture();
 }
 
 void FramebufferManagerVulkan::DrawFramebufferToOutput(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, bool applyPostShader) {
-	VulkanTexture *pixelTex = MakePixelTexture(srcPixels, srcPixelFormat, srcStride, 512, 272);
+	MakePixelTexture(srcPixels, srcPixelFormat, srcStride, 512, 272);
+	VulkanTexture *pixelTex = drawPixelsTex_;
 	// glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, g_Config.iTexFiltering == TEX_FILTER_NEAREST ? GL_NEAREST : GL_LINEAR);
 
 	// This might draw directly at the backbuffer (if so, applyPostShader is set) so if there's a post shader, we need to apply it here.
@@ -545,6 +546,10 @@ void FramebufferManagerVulkan::DrawTexture(VulkanTexture *texture, float x, floa
 	vkCmdDraw(cmd, 4, 1, 0, 0);
 }
 
+void FramebufferManagerVulkan::Bind2DShader() {
+
+}
+
 void FramebufferManagerVulkan::RebindFramebuffer() {
 	if (currentRenderVfb_ && currentRenderVfb_->fbo) {
 		draw_->BindFramebufferAsRenderTarget(currentRenderVfb_->fbo);

GPU/Vulkan/FramebufferVulkan.h

@@ -126,6 +126,7 @@ public:
 	}
 
 protected:
+	void Bind2DShader() override;
 	void SetViewport2D(int x, int y, int w, int h);
 	void DisableState() override {}
 	void ClearBuffer(bool keepState = false) override;
@@ -139,7 +140,7 @@ protected:
 private:
 
 	// The returned texture does not need to be free'd, might be returned from a pool (currently single entry)
-	VulkanTexture *MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height);
+	void MakePixelTexture(const u8 *srcPixels, GEBufferFormat srcPixelFormat, int srcStride, int width, int height);
 	void DoNotifyDraw();
 
 	VkCommandBuffer AllocFrameCommandBuffer();