Main change: Implemented EFB pokes in DX9/DX11.

Games affected by this change: Mario Smash Football, Mario Strikers Charged Football, Monster Hunter Tri.
Other games possibly affected: Shaun White Snowboarding, Resident Evil Code: Veronica, Baten Kaitos.
This implementation will decrease performance if the game uses this feature, but the glitches will be gone. I'll add an option for this in a later commit. EFB pokes are somewhat slow in DX11 right now, speed should be okayish in DX9 though.

Other changes:
- SOMEWHAT cleaned up the EFB access code in DX9
- Fixed incompatible parameter list of AccessEFB and TVideo_AccessEFB.
- Fixed a theoretical bug in ReplaceRGBATexture2D, add support for STAGING textures
- Removed unused parameters in various DX9 functions


git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@6300 8ced0084-cf51-0410-be5f-012b33b47a6e

Source/Core/VideoCommon/Src/Render.h

@@ -79,7 +79,7 @@ public:
 
 	static TargetRectangle ConvertEFBRectangle(const EFBRectangle& rc);
 
-	static u32 AccessEFB(EFBAccessType type, int x, int y);
+	static u32 AccessEFB(EFBAccessType type, u32 x, u32 y, u32 poke_data);
 
 	// Random utilities
     static void RenderText(const char* pstr, int left, int top, u32 color);

Source/Plugins/Plugin_VideoDX11/Src/D3DTexture.cpp

@@ -24,7 +24,7 @@ namespace D3D
 
 void ReplaceRGBATexture2D(ID3D11Texture2D* pTexture, const u8* buffer, unsigned int width, unsigned int height, unsigned int pitch, unsigned int level, D3D11_USAGE usage)
 {
-	if (usage == D3D11_USAGE_DYNAMIC)
+	if (usage == D3D11_USAGE_DYNAMIC || usage == D3D11_USAGE_STAGING)
 	{
 		D3D11_MAPPED_SUBRESOURCE map;
 		D3D::context->Map(pTexture, level, D3D11_MAP_WRITE_DISCARD, 0, &map);
@@ -35,7 +35,7 @@ void ReplaceRGBATexture2D(ID3D11Texture2D* pTexture, const u8* buffer, unsigned
 		else
 		{
 			for (unsigned int y = 0; y < height; ++y)
-				memcpy((u8*)map.pData + y * map.RowPitch, (u32*)buffer + y * pitch, map.RowPitch);
+				memcpy((u8*)map.pData + y * map.RowPitch, (u32*)buffer + y * pitch, 4 * pitch);
 		}
 		D3D::context->Unmap(pTexture, level);
 	}

Source/Plugins/Plugin_VideoDX11/Src/D3DUtil.cpp

@@ -429,11 +429,13 @@ ID3D11SamplerState* linear_copy_sampler = NULL;
 ID3D11SamplerState* point_copy_sampler = NULL;
 ID3D11Buffer* stqvb = NULL;
 ID3D11Buffer* stsqvb = NULL;
+ID3D11Buffer* quadvb = NULL;
 ID3D11Buffer* clearvb = NULL;
 
 typedef struct { float x,y,z,u,v; } STQVertex;
 typedef struct { float x,y,z,u,v; } STSQVertex;
 typedef struct { float x,y,z; u32 col; } ClearVertex;
+typedef struct { float x,y,z; u32 col; } ColVertex;
 
 struct
 {
@@ -446,6 +448,12 @@ struct
 	float u1, v1, u2, v2;
 } tex_sub_quad_data;
 
+struct
+{
+	float x1, y1, x2, y2;
+	u32 col;
+} draw_quad_data;
+
 struct
 {
 	u32 col;
@@ -468,6 +476,7 @@ void InitUtils()
 	// cached data used to avoid unnecessarily reloading the vertex buffers
 	memset(&tex_quad_data, 0, sizeof(tex_quad_data));
 	memset(&tex_sub_quad_data, 0, sizeof(tex_sub_quad_data));
+	memset(&draw_quad_data, 0, sizeof(draw_quad_data));
 	memset(&clear_quad_data, 0, sizeof(clear_quad_data));
 
 	STQVertex stqcoords[4] = {
@@ -480,6 +489,9 @@ void InitUtils()
 	STSQVertex stsqcoords[4];
 	memset(stsqcoords, 0, sizeof(stsqcoords));
 
+	ColVertex colcoords[4];
+	memset(colcoords, 0, sizeof(colcoords));
+
 	ClearVertex cqcoords[4] = {
 		{-1.0f,  1.0f, 0, 0},
 		{ 1.0f,  1.0f, 0, 0},
@@ -495,6 +507,10 @@ void InitUtils()
 	CHECK(stsqvb!=NULL, "Create vertex buffer of drawShadedTexSubQuad");
 	SetDebugObjectName((ID3D11DeviceChild*)stsqvb, "vertex buffer of drawShadedTexSubQuad");
 
+	quadvb = CreateQuadVertexBuffer(4*sizeof(ColVertex), colcoords);
+	CHECK(quadvb!=NULL, "Create vertex buffer of drawColorQuad");
+	SetDebugObjectName((ID3D11DeviceChild*)quadvb, "vertex buffer of drawColorQuad");
+
 	clearvb = CreateQuadVertexBuffer(4*sizeof(ClearVertex), cqcoords);
 	CHECK(clearvb!=NULL, "Create vertex buffer of drawClearQuad");
 	SetDebugObjectName((ID3D11DeviceChild*)clearvb, "vertex buffer of drawClearQuad");
@@ -509,6 +525,7 @@ void ShutdownUtils()
 	SAFE_RELEASE(linear_copy_sampler);
 	SAFE_RELEASE(stqvb);
 	SAFE_RELEASE(stsqvb);
+	SAFE_RELEASE(quadvb);
 	SAFE_RELEASE(clearvb);
 }
 
@@ -627,6 +644,45 @@ void drawShadedTexSubQuad(ID3D11ShaderResourceView* texture,
 	context->PSSetShaderResources(0, 1, &texres); // immediately unbind the texture
 }
 
+// Fills a certain area of the current render target with the specified color
+// destination coordinates normalized to (-1;1)
+void drawColorQuad(u32 Color, float x1, float y1, float x2, float y2)
+{
+	if(draw_quad_data.x1 != x1 || draw_quad_data.y1 != y1 ||
+		draw_quad_data.x2 != x2 || draw_quad_data.y2 != y2 ||
+		draw_quad_data.col != Color)
+	{
+		ColVertex coords[4] = {
+			{ x1, y2, 0.f, Color },
+			{ x2, y2, 0.f, Color },
+			{ x1, y1, 0.f, Color },
+			{ x2, y1, 0.f, Color },
+		};
+
+		D3D11_MAPPED_SUBRESOURCE map;
+		context->Map(quadvb, 0, D3D11_MAP_WRITE_DISCARD, 0, &map);
+		memcpy(map.pData, coords, sizeof(coords));
+		context->Unmap(quadvb, 0);
+
+		draw_quad_data.x1 = x1;
+		draw_quad_data.y1 = y1;
+		draw_quad_data.x2 = x2;
+		draw_quad_data.y2 = y2;
+		draw_quad_data.col = Color;
+	}
+
+	context->VSSetShader(VertexShaderCache::GetClearVertexShader(), NULL, 0);
+	context->PSSetShader(PixelShaderCache::GetClearProgram(), NULL, 0);
+	context->IASetInputLayout(VertexShaderCache::GetClearInputLayout());
+
+	UINT stride = sizeof(ColVertex);
+	UINT offset = 0;
+	context->IASetPrimitiveTopology(D3D11_PRIMITIVE_TOPOLOGY_TRIANGLESTRIP);
+	context->IASetVertexBuffers(0, 1, &quadvb, &stride, &offset);
+
+	context->Draw(4, 0);
+}
+
 void drawClearQuad(u32 Color, float z, ID3D11PixelShader* PShader, ID3D11VertexShader* Vshader, ID3D11InputLayout* layout)
 {
 	if (clear_quad_data.col != Color || clear_quad_data.z != z)
@@ -657,5 +713,4 @@ void drawClearQuad(u32 Color, float z, ID3D11PixelShader* PShader, ID3D11VertexS
 	context->Draw(4, 0);
 }
 
-
 }  // namespace

Source/Plugins/Plugin_VideoDX11/Src/D3DUtil.h

@@ -80,6 +80,5 @@ namespace D3D
 							ID3D11VertexShader* Vshader,
 							ID3D11InputLayout* layout);
 	void drawClearQuad(u32 Color, float z, ID3D11PixelShader* PShader, ID3D11VertexShader* Vshader, ID3D11InputLayout* layout);
-	void SaveRenderStates();
+	void drawColorQuad(u32 Color, float x1, float y1, float x2, float y2);
-	void RestoreRenderStates();
 }

Source/Plugins/Plugin_VideoDX11/Src/Render.cpp

@@ -594,19 +594,19 @@ void Renderer::SetColorMask()
 	D3D::gfxstate->SetRenderTargetWriteMask(color_mask);
 }
 
-u32 Renderer::AccessEFB(EFBAccessType type, int x, int y)
+u32 Renderer::AccessEFB(EFBAccessType type, u32 x, u32 y, u32 poke_data)
 {
+	D3D11_MAPPED_SUBRESOURCE map;
 	ID3D11Texture2D* read_tex;
-	u32 ret = 0;
 
 	if (!g_ActiveConfig.bEFBAccessEnable)
 		return 0;
 
-	if (type == POKE_Z || type == POKE_COLOR)
+	if (type == POKE_Z)
 	{
 		static bool alert_only_once = true;
 		if (!alert_only_once) return 0;
-		PanicAlert("Poke EFB not implemented");
+		PanicAlert("EFB: Poke Z not implemented (tried to poke z value %#x at (%d,%d))", poke_data, x, y);
 		alert_only_once = false;
 		return 0;
 	}
@@ -619,12 +619,23 @@ u32 Renderer::AccessEFB(EFBAccessType type, int x, int y)
 	efbPixelRc.bottom = y + 1;
 	TargetRectangle targetPixelRc = Renderer::ConvertEFBRectangle(efbPixelRc);
 
-	// Take the mean of the resulting dimensions; TODO: check whether this causes any bugs compared to taking the average color of the target area
+	// Take the mean of the resulting dimensions; TODO: Don't use the center pixel, compute the average color instead
 	D3D11_RECT RectToLock;
+	if(type == PEEK_COLOR || type == PEEK_Z)
+	{
 		RectToLock.left = (targetPixelRc.left + targetPixelRc.right) / 2;
 		RectToLock.top = (targetPixelRc.top + targetPixelRc.bottom) / 2;
 		RectToLock.right = RectToLock.left + 1;
 		RectToLock.bottom = RectToLock.top + 1;
+	}
+	else
+	{
+		RectToLock.left = targetPixelRc.left;
+		RectToLock.right = targetPixelRc.right;
+		RectToLock.top = targetPixelRc.top;
+		RectToLock.bottom = targetPixelRc.bottom;
+	}
+
 	if (type == PEEK_Z)
 	{
 		ResetAPIState(); // Reset any game specific settings
@@ -651,41 +662,60 @@ u32 Renderer::AccessEFB(EFBAccessType type, int x, int y)
 		D3D::context->CopySubresourceRegion(read_tex, 0, 0, 0, 0, g_framebufferManager.GetEFBDepthReadTexture()->GetTex(), 0, &box);
 
 		RestoreAPIState(); // restore game state
+
+		// read the data from system memory
+		D3D::context->Map(read_tex, 0, D3D11_MAP_READ, 0, &map);
+
+		float val = *(float*)map.pData;
+		u32 ret = ((u32)(val * 0xffffff));
+		D3D::context->Unmap(read_tex, 0);
+
+		// TODO: in RE0 this value is often off by one in Video_DX9 (where this code is derived from), which causes lighting to disappear
+		return ret;
 	}
-	else
+	else if (type == PEEK_COLOR)
 	{
 		// we can directly copy to system memory here
 		read_tex = g_framebufferManager.GetEFBColorStagingBuffer();
 		D3D11_BOX box = CD3D11_BOX(RectToLock.left, RectToLock.top, 0, RectToLock.right, RectToLock.bottom, 1);
 		D3D::context->CopySubresourceRegion(read_tex, 0, 0, 0, 0, g_framebufferManager.GetEFBColorTexture()->GetTex(), 0, &box);
-	}
 
 		// read the data from system memory
-	D3D11_MAPPED_SUBRESOURCE map;
 		D3D::context->Map(read_tex, 0, D3D11_MAP_READ, 0, &map);
-
+		u32 ret = *(u32*)map.pData;
-	switch(type)
-	{
-	case PEEK_Z:
-		{
-			float val = *(float*)map.pData;
-			ret = ((u32)(val * 0xffffff));
-			break;
-		}
-
-	case PEEK_COLOR:
-		ret = *(u32*)map.pData;
-		break;
-
-	default:
-		// TODO: Implement POKE_Z and POKE_COLOR
-		break;
-	}
-
 		D3D::context->Unmap(read_tex, 0);
-	// TODO: in RE0 this value is often off by one in Video_DX9 (where this code is derived from), which causes lighting to disappear
 		return ret;
 	}
+	else //if(type == POKE_COLOR)
+	{
+		// TODO: Speed this up by batching pokes?
+		// If we're only writing one pixel (native resolution), we can directly copy the data into the target. TODO: Check if this is faster than drawing quads
+		u32 rgbaColor = (poke_data & 0xFF00FF00) | ((poke_data >> 16) & 0xFF) | ((poke_data << 16) & 0xFF0000);
+		if(RectToLock.right <= RectToLock.left + 1 && RectToLock.bottom <= RectToLock.top + 1)
+		{
+			D3D::context->Map(g_framebufferManager.GetEFBColorStagingBuffer(), 0, D3D11_MAP_WRITE, 0, &map);
+			*(u32*)map.pData = rgbaColor;
+			D3D::context->Unmap(g_framebufferManager.GetEFBColorStagingBuffer(), 0);
+
+			D3D11_BOX box = CD3D11_BOX(0, 0, 0, 1, 1, 1);
+			D3D::context->CopySubresourceRegion(g_framebufferManager.GetEFBColorTexture()->GetTex(), 0, RectToLock.left, RectToLock.top, 0, g_framebufferManager.GetEFBColorStagingBuffer(), 0, &box);
+			return 0;
+		}
+		else
+		{
+			ResetAPIState(); // Reset any game specific settings
+
+			D3D::context->OMSetRenderTargets(1, &g_framebufferManager.GetEFBColorTexture()->GetRTV(), NULL);
+			D3D::drawColorQuad(rgbaColor, (float)RectToLock.left   * 2.f / (float)Renderer::GetFullTargetWidth()  - 1.f,
+										- (float)RectToLock.top    * 2.f / (float)Renderer::GetFullTargetHeight() + 1.f,
+										  (float)RectToLock.right  * 2.f / (float)Renderer::GetFullTargetWidth()  - 1.f,
+										- (float)RectToLock.bottom * 2.f / (float)Renderer::GetFullTargetHeight() + 1.f);
+
+			RestoreAPIState();
+			return 0;
+		}
+	}
+}
 
 // Called from VertexShaderManager
 void UpdateViewport()

Source/Plugins/Plugin_VideoDX11/Src/main.cpp

@@ -378,7 +378,7 @@ void VideoFifo_CheckEFBAccess()
 {
 	if (Common::AtomicLoadAcquire(s_efbAccessRequested))
 	{
-		s_AccessEFBResult = Renderer::AccessEFB(s_accessEFBArgs.type, s_accessEFBArgs.x, s_accessEFBArgs.y);
+		s_AccessEFBResult = Renderer::AccessEFB(s_accessEFBArgs.type, s_accessEFBArgs.x, s_accessEFBArgs.y, s_accessEFBArgs.Data);
 
 		Common::AtomicStoreRelease(s_efbAccessRequested, FALSE);
 	}

Source/Plugins/Plugin_VideoDX9/Src/D3DUtil.cpp

@@ -21,6 +21,8 @@
 #include "D3DBase.h"
 #include "D3DUtil.h"
 #include "Render.h"
+#include "PixelShaderCache.h"
+#include "VertexShaderCache.h"
 
 namespace D3D
 {
@@ -423,6 +425,23 @@ void drawShadedTexSubQuad(IDirect3DTexture9 *texture,
 	RestoreShaders();
 }
 
+// Fills a certain area of the current render target with the specified color
+// Z buffer disabled; destination coordinates normalized to (-1;1)
+void drawColorQuad(u32 Color, float x1, float y1, float x2, float y2)
+{
+	struct CQVertex { float x, y, z, rhw; u32 col; } coords[4] = {
+		{ x1, y2, 0.f, 1.f, Color },
+		{ x2, y2, 0.f, 1.f, Color },
+		{ x1, y1, 0.f, 1.f, Color },
+		{ x2, y1, 0.f, 1.f, Color },
+	};
+	dev->SetVertexShader(VertexShaderCache::GetClearVertexShader());
+	dev->SetPixelShader(PixelShaderCache::GetClearProgram());
+	dev->SetFVF(D3DFVF_XYZW | D3DFVF_DIFFUSE);
+	dev->DrawPrimitiveUP(D3DPT_TRIANGLEFAN, 2, coords, sizeof(CQVertex));
+	RestoreShaders();
+}
+
 void drawClearQuad(u32 Color,float z,IDirect3DPixelShader9 *PShader,IDirect3DVertexShader9 *Vshader)
 {
 	struct Q2DVertex { float x,y,z,rhw;u32 color;} coords[4] = {

Source/Plugins/Plugin_VideoDX9/Src/D3DUtil.h

@@ -78,6 +78,8 @@ namespace D3D
 							IDirect3DPixelShader9 *PShader,
 							IDirect3DVertexShader9 *Vshader);
 	void drawClearQuad(u32 Color, float z, IDirect3DPixelShader9 *PShader, IDirect3DVertexShader9 *Vshader);
+	void drawColorQuad(u32 Color, float x1, float y1, float x2, float y2);
+
 	void SaveRenderStates();
 	void RestoreRenderStates();
 }

Source/Plugins/Plugin_VideoDX9/Src/FramebufferManager.cpp

@@ -73,12 +73,12 @@ D3DFORMAT FramebufferManager::GetEFBColorRTSurfaceFormat()
 	return s_efb_color_surface_Format;
 }
 
-LPDIRECT3DTEXTURE9 FramebufferManager::GetEFBColorTexture(const EFBRectangle& sourceRc)
+LPDIRECT3DTEXTURE9 FramebufferManager::GetEFBColorTexture()
 {
 	return s_efb_color_texture;
 }
 
-LPDIRECT3DTEXTURE9 FramebufferManager::GetEFBDepthTexture(const EFBRectangle &sourceRc)
+LPDIRECT3DTEXTURE9 FramebufferManager::GetEFBDepthTexture()
 {
 	return s_efb_depth_texture;
 }
@@ -305,7 +305,7 @@ void FramebufferManager::copyToRealXFB(u32 xfbAddr, u32 fbWidth, u32 fbHeight, c
 	}
 
 	TargetRectangle targetRc = Renderer::ConvertEFBRectangle(sourceRc);
-	TextureConverter::EncodeToRamYUYV(GetEFBColorTexture(sourceRc), targetRc, xfb_in_ram, fbWidth, fbHeight);
+	TextureConverter::EncodeToRamYUYV(GetEFBColorTexture(), targetRc, xfb_in_ram, fbWidth, fbHeight);
 }
 
 void FramebufferManager::copyToVirtualXFB(u32 xfbAddr, u32 fbWidth, u32 fbHeight, const EFBRectangle& sourceRc)
@@ -395,7 +395,7 @@ void FramebufferManager::copyToVirtualXFB(u32 xfbAddr, u32 fbWidth, u32 fbHeight
 	// Copy EFB data to XFB and restore render target again
 	if(!xfbTexture)
 		return;
-	LPDIRECT3DTEXTURE9 read_texture = GetEFBColorTexture(sourceRc);
+	LPDIRECT3DTEXTURE9 read_texture = GetEFBColorTexture();
 	
 	Renderer::ResetAPIState(); // Reset any game specific settings
 	LPDIRECT3DSURFACE9 Rendersurf = NULL;

Source/Plugins/Plugin_VideoDX9/Src/FramebufferManager.h

@@ -109,8 +109,8 @@ public:
 	void CopyToXFB(u32 xfbAddr, u32 fbWidth, u32 fbHeight, const EFBRectangle& sourceRc);
 	const XFBSource** GetXFBSource(u32 xfbAddr, u32 fbWidth, u32 fbHeight, u32 &xfbCount);
 
-	LPDIRECT3DTEXTURE9 GetEFBColorTexture(const EFBRectangle& sourceRc);
+	LPDIRECT3DTEXTURE9 GetEFBColorTexture();
-	LPDIRECT3DTEXTURE9 GetEFBDepthTexture(const EFBRectangle& sourceRc);
+	LPDIRECT3DTEXTURE9 GetEFBDepthTexture();
 
 	LPDIRECT3DSURFACE9 GetEFBColorRTSurface();
 	LPDIRECT3DSURFACE9 GetEFBDepthRTSurface();

Source/Plugins/Plugin_VideoDX9/Src/Render.cpp

@@ -691,46 +691,45 @@ void Renderer::SetColorMask()
 	D3D::SetRenderState(D3DRS_COLORWRITEENABLE, color_mask);
 }
 
-u32 Renderer::AccessEFB(EFBAccessType type, int x, int y)
+u32 Renderer::AccessEFB(EFBAccessType type, u32 x, u32 y, u32 poke_data)
 {
 	if (!g_ActiveConfig.bEFBAccessEnable)
 		return 0;
 
-	if (type == POKE_Z || type == POKE_COLOR)
+	if (type == POKE_Z)
 	{
 		static bool alert_only_once = true;
 		if (!alert_only_once) return 0;
-		PanicAlert("Poke EFB not implemented");
+		PanicAlert("EFB: Poke Z not implemented (tried to poke z value %#x at (%d,%d))", poke_data, x, y);
 		alert_only_once = false;
 		return 0;
 	}
 
-	// Get the working buffer
+	// We're using three surfaces here:
-	LPDIRECT3DSURFACE9 pBuffer = (type == PEEK_Z || type == POKE_Z) ? 
+	// - pEFBSurf: EFB Surface. Source surface when peeking, destination surface when poking.
-		g_framebufferManager.GetEFBDepthRTSurface() : g_framebufferManager.GetEFBColorRTSurface();
+	// - pBufferRT: A render target surface. When peeking, we render a textured quad to this surface.
-	// Get the temporal buffer to move 1pixel data
+	// - pSystemBuf: An offscreen surface. Used to retrieve the pixel data from pBufferRT.
-	LPDIRECT3DSURFACE9 RBuffer = (type == PEEK_Z || type == POKE_Z) ? 
+	LPDIRECT3DSURFACE9 pEFBSurf, pBufferRT, pSystemBuf;
-		g_framebufferManager.GetEFBDepthReadSurface() : g_framebufferManager.GetEFBColorReadSurface();
+	if(type == PEEK_Z || type == POKE_Z)
-	// Get the memory buffer that can be locked
+	{
-	LPDIRECT3DSURFACE9 pOffScreenBuffer = (type == PEEK_Z || type == POKE_Z) ? 
+		pEFBSurf = g_framebufferManager.GetEFBDepthRTSurface();
-		g_framebufferManager.GetEFBDepthOffScreenRTSurface() : g_framebufferManager.GetEFBColorOffScreenRTSurface();
+		pBufferRT = g_framebufferManager.GetEFBDepthReadSurface();
-	// Get the buffer format
+		pSystemBuf = g_framebufferManager.GetEFBDepthOffScreenRTSurface();
-	D3DFORMAT BufferFormat = (type == PEEK_Z || type == POKE_Z) ? 
+	}
-		g_framebufferManager.GetEFBDepthRTSurfaceFormat() : g_framebufferManager.GetEFBColorRTSurfaceFormat();
+	else //if(type == PEEK_COLOR || type == POKE_COLOR)
-	D3DFORMAT ReadBufferFormat = (type == PEEK_Z || type == POKE_Z) ? 
+	{
-		g_framebufferManager.GetEFBDepthReadSurfaceFormat() : BufferFormat;
+		pEFBSurf = g_framebufferManager.GetEFBColorRTSurface();
-	
+		pBufferRT = g_framebufferManager.GetEFBColorReadSurface();
-	if (BufferFormat == D3DFMT_D24X8)
+		pSystemBuf = g_framebufferManager.GetEFBColorOffScreenRTSurface();
-		return 0;
+	}
-
-	D3DLOCKED_RECT drect;
 
 	// Buffer not found alert
-	if (!pBuffer) {
+	if (!pEFBSurf) {
 		PanicAlert("No %s!", (type == PEEK_Z || type == POKE_Z) ? "Z-Buffer" : "Color EFB");
 		return 0;
 	}
-	// Get the rectangular target region covered by the EFB pixel
+
+	// Convert EFB dimensions to the ones of our render target
 	EFBRectangle efbPixelRc;
 	efbPixelRc.left = x;
 	efbPixelRc.top = y;
@@ -740,7 +739,6 @@ u32 Renderer::AccessEFB(EFBAccessType type, int x, int y)
 	TargetRectangle targetPixelRc = ConvertEFBRectangle(efbPixelRc);
 
 	u32 z = 0;
-	float val = 0.0f;
 	HRESULT hr;
 	RECT RectToLock;
 	RectToLock.bottom = targetPixelRc.bottom;
@@ -749,6 +747,9 @@ u32 Renderer::AccessEFB(EFBAccessType type, int x, int y)
 	RectToLock.top = targetPixelRc.top;	
 	if (type == PEEK_Z)
 	{
+		if (g_framebufferManager.GetEFBDepthRTSurfaceFormat() == D3DFMT_D24X8)
+			return 0;
+
 		RECT PixelRect;
 		PixelRect.bottom = 4;
 		PixelRect.left = 0;
@@ -764,33 +765,24 @@ u32 Renderer::AccessEFB(EFBAccessType type, int x, int y)
 			RectToLock.left++;
 
 		ResetAPIState(); // Reset any game specific settings
-		hr = D3D::dev->SetDepthStencilSurface(NULL);
+		D3D::dev->SetDepthStencilSurface(NULL);
-		hr = D3D::dev->SetRenderTarget(0, RBuffer);
+		D3D::dev->SetRenderTarget(0, pBufferRT);
-		if (FAILED(hr))
+
-		{
-			PanicAlert("unable to set pixel render buffer");
-			return 0;
-		}
-		D3DVIEWPORT9 vp;
 		// Stretch picture with increased internal resolution
+		D3DVIEWPORT9 vp;
 		vp.X = 0;
 		vp.Y = 0;
 		vp.Width  = 4;
 		vp.Height = 4;
 		vp.MinZ = 0.0f;
 		vp.MaxZ = 1.0f;
-		hr = D3D::dev->SetViewport(&vp);
+		D3D::dev->SetViewport(&vp);
-		if (FAILED(hr))
+
-		{
-			PanicAlert("unable to set pixel viewport");
-			return 0;
-		}
 		float colmat[16] = {0.0f};
 		float fConstAdd[4] = {0.0f};
 		colmat[0] = colmat[5] = colmat[10] = 1.0f;
 		PixelShaderManager::SetColorMatrix(colmat, fConstAdd); // set transformation
-		EFBRectangle source_rect;
+		LPDIRECT3DTEXTURE9 read_texture = g_framebufferManager.GetEFBDepthTexture();
-		LPDIRECT3DTEXTURE9 read_texture = g_framebufferManager.GetEFBDepthTexture(source_rect);
 		
 		D3D::ChangeSamplerState(0, D3DSAMP_MINFILTER, D3DTEXF_POINT);
 
@@ -800,54 +792,29 @@ u32 Renderer::AccessEFB(EFBAccessType type, int x, int y)
 			Renderer::GetFullTargetWidth(),
 			Renderer::GetFullTargetHeight(),
 			4, 4,
-			(BufferFormat == FOURCC_RAWZ) ? PixelShaderCache::GetColorMatrixProgram(0) : PixelShaderCache::GetDepthMatrixProgram(0), 
+			(g_framebufferManager.GetEFBDepthRTSurfaceFormat() == FOURCC_RAWZ) ? PixelShaderCache::GetColorMatrixProgram(0) : PixelShaderCache::GetDepthMatrixProgram(0),
 			VertexShaderCache::GetSimpleVertexShader(0));
 
 		D3D::RefreshSamplerState(0, D3DSAMP_MINFILTER);
 
-		hr = D3D::dev->SetRenderTarget(0, g_framebufferManager.GetEFBColorRTSurface());
+		D3D::dev->SetRenderTarget(0, g_framebufferManager.GetEFBColorRTSurface());
-		hr = D3D::dev->SetDepthStencilSurface(g_framebufferManager.GetEFBDepthRTSurface());
+		D3D::dev->SetDepthStencilSurface(g_framebufferManager.GetEFBDepthRTSurface());
 		RestoreAPIState();
+
+		// Retrieve the pixel data to the local memory buffer
 		RectToLock.bottom = 4;
 		RectToLock.left = 0;
 		RectToLock.right = 4;
 		RectToLock.top = 0;
-	}
+		D3D::dev->GetRenderTargetData(pBufferRT, pSystemBuf);
-	else
-	{
-		hr = D3D::dev->StretchRect(pBuffer, &RectToLock, RBuffer, NULL, D3DTEXF_NONE);
-		//change the rect to lock the entire one pixel buffer
-		RectToLock.bottom = 1;
-		RectToLock.left = 0;
-		RectToLock.right = 1;
-		RectToLock.top = 0;
-	}
-	if (FAILED(hr))
-	{
-		PanicAlert("Unable to stretch data to buffer");
-		return 0;
-	}
-	// Retrieve the pixel data to the local memory buffer
-	D3D::dev->GetRenderTargetData(RBuffer, pOffScreenBuffer);
-	if (FAILED(hr))
-	{
-		PanicAlert("Unable to copy data to mem buffer");
-		return 0;
-	}
 
+		// EFB data successfully retrieved, now get the pixel data
+		D3DLOCKED_RECT drect;
+		pSystemBuf->LockRect(&drect, &RectToLock, D3DLOCK_READONLY);
 
-	// The surface is good.. lock it
+		float val = 0.0f;
-	if ((hr = pOffScreenBuffer->LockRect(&drect, &RectToLock, D3DLOCK_READONLY)) != D3D_OK)
-	{
-		PanicAlert("ERROR: %s", hr == D3DERR_WASSTILLDRAWING ? "Still drawing" : hr == D3DERR_INVALIDCALL ? "Invalid call" : "w00t");
-		return 0;
-	}
 
-	switch (type)
+		switch (g_framebufferManager.GetEFBDepthReadSurfaceFormat())
-	{
-	case PEEK_Z:
-		{
-			switch (ReadBufferFormat)
 		{
 		case D3DFMT_R32F:
 			val = ((float*)drect.pBits)[6];
@@ -863,33 +830,41 @@ u32 Renderer::AccessEFB(EFBAccessType type, int x, int y)
 			break;
 		};
 		z = ((u32)(val * 0xffffff));
-		}
-		break;
 
-	case POKE_Z:
+		pSystemBuf->UnlockRect();
-		// TODO: Implement
-		break;
-
-	case PEEK_COLOR:
-		z = ((u32*)drect.pBits)[0];
-		break;
-
-	case POKE_COLOR:
-		// TODO: Implement. One way is to draw a tiny pixel-sized rectangle at
-		// the exact location. Note: EFB pokes are susceptible to Z-buffering
-		// and perhaps blending.
-		//WARN_LOG(VIDEOINTERFACE, "This is probably some kind of software rendering");
-		break;
-
-		// TODO: Implement POKE_Z and POKE_COLOR
-	default:
-		break;
-	}
-
-	pOffScreenBuffer->UnlockRect();
 		// TODO: in RE0 this value is often off by one, which causes lighting to disappear
 		return z;
 	}
+	else if(type == PEEK_COLOR)
+	{
+		// TODO: Can't we directly StretchRect to System buf?
+		hr = D3D::dev->StretchRect(pEFBSurf, &RectToLock, pBufferRT, NULL, D3DTEXF_NONE);
+		D3D::dev->GetRenderTargetData(pBufferRT, pSystemBuf);
+
+		// EFB data successfully retrieved, now get the pixel data
+		RectToLock.bottom = 1;
+		RectToLock.left = 0;
+		RectToLock.right = 1;
+		RectToLock.top = 0;
+		D3DLOCKED_RECT drect;
+		pSystemBuf->LockRect(&drect, &RectToLock, D3DLOCK_READONLY);
+
+		z = ((u32*)drect.pBits)[0];
+		pSystemBuf->UnlockRect();
+		return z;
+	}
+	else //if(type == POKE_COLOR)
+	{
+		// TODO: Speed this up by batching pokes?
+		ResetAPIState();
+		D3D::drawColorQuad(poke_data, (float)RectToLock.left   * 2.f / (float)Renderer::GetFullTargetWidth()  - 1.f,
+									- (float)RectToLock.top    * 2.f / (float)Renderer::GetFullTargetHeight() + 1.f,
+									  (float)RectToLock.right  * 2.f / (float)Renderer::GetFullTargetWidth()  - 1.f,
+									- (float)RectToLock.bottom * 2.f / (float)Renderer::GetFullTargetHeight() + 1.f);
+		RestoreAPIState();
+		return 0;
+	}
+}
 
 // Called from VertexShaderManager
 void UpdateViewport()
@@ -1190,7 +1165,7 @@ void Renderer::Swap(u32 xfbAddr, FieldType field, u32 fbWidth, u32 fbHeight,cons
 	else
 	{
 		TargetRectangle targetRc = ConvertEFBRectangle(rc);
-		LPDIRECT3DTEXTURE9 read_texture = g_framebufferManager.GetEFBColorTexture(rc);
+		LPDIRECT3DTEXTURE9 read_texture = g_framebufferManager.GetEFBColorTexture();
 		D3D::drawShadedTexQuad(read_texture,targetRc.AsRECT(),Renderer::GetFullTargetWidth(),Renderer::GetFullTargetHeight(),Width,Height,PixelShaderCache::GetColorCopyProgram(g_Config.iMultisampleMode),VertexShaderCache::GetSimpleVertexShader(g_Config.iMultisampleMode));
 	}
 	D3D::RefreshSamplerState(0, D3DSAMP_MINFILTER);

Source/Plugins/Plugin_VideoDX9/Src/TextureCache.cpp

@@ -74,8 +74,8 @@ void TextureCache::TCacheEntry::FromRenderTarget(bool bFromZBuffer, bool bScaleB
 	bool bIsIntensityFmt, u32 copyfmt)
 {
 	const LPDIRECT3DTEXTURE9 read_texture = bFromZBuffer ?
-		g_framebufferManager.GetEFBDepthTexture(source_rect) :
+		g_framebufferManager.GetEFBDepthTexture() :
-		g_framebufferManager.GetEFBColorTexture(source_rect);
+		g_framebufferManager.GetEFBColorTexture();
 
 	if (!isDynamic || g_ActiveConfig.bCopyEFBToTexture)
 	{

Source/Plugins/Plugin_VideoDX9/Src/TextureConverter.cpp

@@ -325,7 +325,7 @@ void EncodeToRam(u32 address, bool bFromZBuffer, bool bIsIntensityFmt, u32 copyf
 
 	u8 *dest_ptr = Memory_GetPtr(address);
 
-	LPDIRECT3DTEXTURE9 source_texture = bFromZBuffer ? g_framebufferManager.GetEFBDepthTexture(source) : g_framebufferManager.GetEFBColorTexture(source);
+	LPDIRECT3DTEXTURE9 source_texture = bFromZBuffer ? g_framebufferManager.GetEFBDepthTexture() : g_framebufferManager.GetEFBColorTexture();
 	int width = (source.right - source.left) >> bScaleByHalf;
 	int height = (source.bottom - source.top) >> bScaleByHalf;
 

Source/Plugins/Plugin_VideoDX9/Src/main.cpp

@@ -400,7 +400,7 @@ void VideoFifo_CheckEFBAccess()
 {
 	if (Common::AtomicLoadAcquire(s_efbAccessRequested))
 	{
-		s_AccessEFBResult = Renderer::AccessEFB(s_accessEFBArgs.type, s_accessEFBArgs.x, s_accessEFBArgs.y);
+		s_AccessEFBResult = Renderer::AccessEFB(s_accessEFBArgs.type, s_accessEFBArgs.x, s_accessEFBArgs.y, s_accessEFBArgs.Data);
 
 		Common::AtomicStoreRelease(s_efbAccessRequested, FALSE);
 	}

Source/Plugins/Plugin_VideoOGL/Src/Render.cpp

@@ -854,7 +854,7 @@ void Renderer::SetColorMask()
 	glColorMask(ColorMask,  ColorMask,  ColorMask,  AlphaMask);
 }
 
-u32 Renderer::AccessEFB(EFBAccessType type, int x, int y)
+u32 Renderer::AccessEFB(EFBAccessType type, u32 x, u32 y, u32 poke_data)
 {
 	if (!g_ActiveConfig.bEFBAccessEnable)
 		return 0;

Source/Plugins/Plugin_VideoOGL/Src/main.cpp

@@ -430,7 +430,7 @@ void VideoFifo_CheckEFBAccess()
 {
 	if (Common::AtomicLoadAcquire(s_efbAccessRequested))
 	{
-		s_AccessEFBResult = Renderer::AccessEFB(s_accessEFBArgs.type, s_accessEFBArgs.x, s_accessEFBArgs.y);
+		s_AccessEFBResult = Renderer::AccessEFB(s_accessEFBArgs.type, s_accessEFBArgs.x, s_accessEFBArgs.y, s_accessEFBArgs.Data);
 
 		Common::AtomicStoreRelease(s_efbAccessRequested, FALSE);
 	}