big changes here:

- Eliminate the useless check for cpu modifications option from efb to ram as it must be enabled always
- use constant names in dx11 for buffer length calculation instead to simplify code reading
- implemented scaled efb copy in opengl, still bugy in some games, the option is not in the gui but will add it when it works perfect
- Change the depth calculation behavior:
if the game use z textures is exactly the same as before.
if the game do not use z texture calculate z values in the vertex shader. the advantage id this approach is that early z culling is applied,
improving fill rate. this mus speed up things, even with ssaa and msaa enabled.
please test for regression and enjoy.

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5896 8ced0084-cf51-0410-be5f-012b33b47a6e

7 files changed, 36 insertions, 59 deletions

diff --git a/Source/Core/VideoCommon/Src/PixelShaderGen.cpp b/Source/Core/VideoCommon/Src/PixelShaderGen.cpp
index 4fb871daf5..6cc3029ff2 100644
--- a/Source/Core/VideoCommon/Src/PixelShaderGen.cpp
+++ b/Source/Core/VideoCommon/Src/PixelShaderGen.cpp
@@ -342,6 +342,7 @@ static const char *swapColors = "rgba";
 static char swapModeTable[4][5];
 
 static char text[16384];
+static bool DepthTextureEnable;
 
 struct RegisterState
 {
@@ -388,7 +389,7 @@ const char *GeneratePixelShaderCode(u32 texture_mask, bool dstAlphaEnable, API_T
 				nIndirectStagesUsed |= 1 << bpmem.tevind[i].bt;
 		}
 	}
-
+	DepthTextureEnable = bpmem.ztex2.op != ZTEXTURE_DISABLE && !bpmem.zcontrol.zcomploc && bpmem.zmode.testenable && bpmem.zmode.updateenable;
 	// Declare samplers
 	if (texture_mask && ApiType == API_OPENGL)
 	{
@@ -454,10 +455,10 @@ const char *GeneratePixelShaderCode(u32 texture_mask, bool dstAlphaEnable, API_T
 
 	WRITE(p, "void main(\n");
 	if(ApiType != API_D3D11)
-		WRITE(p, "  out float4 ocol0 : COLOR0,\n  out float depth : DEPTH,\n  in float4 rawpos : POSITION,\n");
+		WRITE(p, "  out float4 ocol0 : COLOR0,%s\n  in float4 rawpos : POSITION,\n",DepthTextureEnable ? "\n  out float depth : DEPTH," : "");
 	else
-		WRITE(p, "  out float4 ocol0 : SV_Target,\n  out float depth : SV_Depth,\n  in float4 rawpos : SV_Position,\n");
-
+		WRITE(p, "  out float4 ocol0 : SV_Target,%s\n  in float4 rawpos : SV_Position,\n",DepthTextureEnable ? "\n  out float depth : SV_Depth," : "");
+	
 	WRITE(p, "  in float4 colors_0 : COLOR0,\n");
 	WRITE(p, "  in float4 colors_1 : COLOR1");
 
@@ -477,7 +478,8 @@ const char *GeneratePixelShaderCode(u32 texture_mask, bool dstAlphaEnable, API_T
 	}
 	WRITE(p, "        ) {\n");
 
-	char* pmainstart = p;
+	char* pmainstart = p;	
+	
 
 	WRITE(p, "  float4 c0 = "I_COLORS"[1], c1 = "I_COLORS"[2], c2 = "I_COLORS"[3], prev = float4(0.0f, 0.0f, 0.0f, 0.0f), textemp = float4(0.0f, 0.0f, 0.0f, 0.0f), rastemp = float4(0.0f, 0.0f, 0.0f, 0.0f), konsttemp = float4(0.0f, 0.0f, 0.0f, 0.0f);\n"
 			"  float3 comp16 = float3(1.0f, 255.0f, 0.0f), comp24 = float3(1.0f, 255.0f, 255.0f*255.0f);\n"
@@ -561,20 +563,23 @@ const char *GeneratePixelShaderCode(u32 texture_mask, bool dstAlphaEnable, API_T
 		// alpha test will always fail, so restart the shader and just make it an empty function
 		p = pmainstart;
 		WRITE(p, "ocol0 = 0;\n");
-		WRITE(p, "depth = 1.f;\n");
+		if(DepthTextureEnable)
+			WRITE(p, "depth = 1.f;\n");
 		WRITE(p, "discard;\n");
 		if(ApiType != API_D3D11)
 			WRITE(p, "return;\n");
 	}
 	else
 	{
-		if (numTexgen >= 7)
-			WRITE(p, "float4 clipPos = float4(uv0.w, uv1.w, uv2.w, uv3.w);\n");
-
-		// the screen space depth value = far z + (clip z / clip w) * z range
-		WRITE(p, "float zCoord = "I_ZBIAS"[1].x + (clipPos.z / clipPos.w) * "I_ZBIAS"[1].y;\n");
+		if((bpmem.fog.c_proj_fsel.fsel != 0) || DepthTextureEnable)
+		{
+			if (numTexgen >= 7)
+				WRITE(p, "float4 clipPos = float4(uv0.w, uv1.w, uv2.w, uv3.w);\n");
+			// the screen space depth value = far z + (clip z / clip w) * z range
+			WRITE(p, "float zCoord = "I_ZBIAS"[1].x + (clipPos.z / clipPos.w) * "I_ZBIAS"[1].y;\n");
+		}
 
-		if (bpmem.ztex2.op != ZTEXTURE_DISABLE && !bpmem.zcontrol.zcomploc && bpmem.zmode.testenable && bpmem.zmode.updateenable)
+		if (DepthTextureEnable)
 		{
 			// use the texture input of the last texture stage (textemp), hopefully this has been read and is in correct format...
 			if (bpmem.ztex2.op == ZTEXTURE_ADD)
@@ -586,8 +591,8 @@ const char *GeneratePixelShaderCode(u32 texture_mask, bool dstAlphaEnable, API_T
 			WRITE(p, "zCoord = zCoord * (16777215.0f/16777216.0f);\n");
 			WRITE(p, "zCoord = frac(zCoord);\n");
 			WRITE(p, "zCoord = zCoord * (16777216.0f/16777215.0f);\n");
+			WRITE(p, "depth = zCoord;\n");
 		}
-		WRITE(p, "depth = zCoord;\n");
 
 		if (dstAlphaEnable)
 			WRITE(p, "  ocol0 = float4(prev.rgb, "I_ALPHA"[0].a);\n");
@@ -954,7 +959,7 @@ static void WriteStage(char *&p, int n, u32 texture_mask, API_TYPE ApiType)
 
 void SampleTexture(char *&p, const char *destination, const char *texcoords, const char *texswap, int texmap, u32 texture_mask, API_TYPE ApiType)
 {
-	if (texture_mask & (1<<texmap)) {
+	if (texture_mask & (1<<texmap)) {// opengl only
 		// non pow 2
 		bool bwraps = (texture_mask & (0x100<<texmap)) ? true : false;
 		bool bwrapt = (texture_mask & (0x10000<<texmap)) ? true : false;
@@ -973,21 +978,11 @@ void SampleTexture(char *&p, const char *destination, const char *texcoords, con
 			else {
 				WRITE(p, "tempcoord.y = %s.y;\n", texcoords);
 			}
-			 if (ApiType == API_D3D11)
-				WRITE(p, "%s= Tex%d.Sample(samp%d,tempcoord.xy).%s;\n", destination, texmap,texmap, texswap);
-			 else if (ApiType == API_D3D9)
-				 WRITE(p, "%s=tex2D(samp%d,tempcoord.xy).%s;\n", destination, texmap, texswap);
-			 else
-				 WRITE(p, "%s=texRECT(samp%d,tempcoord.xy).%s;\n", destination, texmap, texswap);
+			WRITE(p, "%s=texRECT(samp%d,tempcoord.xy).%s;\n", destination, texmap, texswap);
 		}
 		else
 		{
-			 if (ApiType == API_D3D11)
-				 WRITE(p, "%s=Tex%d.Sample(samp%d,%s.xy).%s;\n", destination,texmap,texmap, texcoords, texswap);
-			 else if (ApiType == API_D3D9)
-				 WRITE(p, "%s=tex2D(samp%d,%s.xy).%s;\n", destination, texmap, texcoords, texswap);
-			 else
-				 WRITE(p, "%s=texRECT(samp%d,%s.xy).%s;\n", destination, texmap, texcoords, texswap);
+			 WRITE(p, "%s=texRECT(samp%d,%s.xy).%s;\n", destination, texmap, texcoords, texswap);
 		}
 	}
 	else
@@ -1061,9 +1056,7 @@ static bool WriteAlphaTest(char *&p, API_TYPE ApiType)
 	
 	compindex = bpmem.alphaFunc.comp1 % 8;
 	WRITE(p, tevAlphaFuncsTable[compindex],alphaRef[1]);//lookup the second component from the alpha function table
-
-	WRITE(p, ")){ocol0 = 0;depth = 1.f;discard;%s}\n",(ApiType != API_D3D11)? "return;" : "");
-	
+	WRITE(p, ")){ocol0 = 0;%sdiscard;%s}\n",DepthTextureEnable ? "depth = 1.f;" : "",(ApiType != API_D3D11)? "return;" : "");
 	return true;
 }
 
@@ -1087,13 +1080,13 @@ static void WriteFog(char *&p)
 	{
 		// perspective
 		// ze = A/(B - Zs)
-		WRITE (p, "  float ze = "I_FOG"[1].x / ("I_FOG"[1].y - depth);\n");
+		WRITE (p, "  float ze = "I_FOG"[1].x / ("I_FOG"[1].y - zCoord);\n");
 	}
 	else
 	{
 		// orthographic
 		// ze = a*Zs
-		WRITE (p, "  float ze = "I_FOG"[1].x * depth;\n");
+		WRITE (p, "  float ze = "I_FOG"[1].x * zCoord;\n");
 	}
 
 	WRITE (p, "  float fog = saturate(ze - "I_FOG"[1].z);\n");
diff --git a/Source/Core/VideoCommon/Src/PixelShaderGen.h b/Source/Core/VideoCommon/Src/PixelShaderGen.h
index 06c4e164d7..61d329a4f6 100644
--- a/Source/Core/VideoCommon/Src/PixelShaderGen.h
+++ b/Source/Core/VideoCommon/Src/PixelShaderGen.h
@@ -37,9 +37,8 @@
 #define C_INDTEXSCALE  (C_ZBIAS + 2)
 #define C_INDTEXMTX    (C_INDTEXSCALE + 2)
 #define C_FOG          (C_INDTEXMTX + 6)
-#define C_ENVCONST_END (C_FOG + 2)
-
 #define C_COLORMATRIX  (C_FOG + 2)
+#define C_PENVCONST_END (C_COLORMATRIX + 16)
 #define PIXELSHADERUID_MAX_VALUES (5 + 32 + 6 + 11)
 
 // DO NOT make anything in this class virtual.
diff --git a/Source/Core/VideoCommon/Src/VertexShaderGen.cpp b/Source/Core/VideoCommon/Src/VertexShaderGen.cpp
index 582cae9d64..02a482229b 100644
--- a/Source/Core/VideoCommon/Src/VertexShaderGen.cpp
+++ b/Source/Core/VideoCommon/Src/VertexShaderGen.cpp
@@ -135,6 +135,7 @@ const char *GenerateVertexShaderCode(u32 components, API_TYPE api_type)
 	WRITE(p, "uniform s_"I_LIGHTS" "I_LIGHTS" : register(c%d);\n", C_LIGHTS);
 	WRITE(p, "uniform s_"I_MATERIALS" "I_MATERIALS" : register(c%d);\n", C_MATERIALS);
 	WRITE(p, "uniform s_"I_PROJECTION" "I_PROJECTION" : register(c%d);\n", C_PROJECTION);
+	WRITE(p, "uniform float4 "I_DEPTHPARAMS" : register(c%d);\n", C_DEPTHPARAMS);
 
 	WRITE(p, "VS_OUTPUT main(\n");
 	
@@ -462,14 +463,11 @@ const char *GenerateVertexShaderCode(u32 components, API_TYPE api_type)
 		WRITE(p, "o.tex3.w = o.pos.w;\n");
 	}
 
+	//write the true depth value, if the game uses depth textures pixel shaders will override with the correct values
+	//if not early z culling will improve speed
 	if (is_d3d)
 	{
-		WRITE(p, "o.pos.z = o.pos.z + o.pos.w;\n");
-	}
-	else
-	{
-		// scale to gl clip space - which is -o.pos.w to o.pos.w, hence the addition.
-		WRITE(p, "o.pos.z = (o.pos.z * 2.0f) + o.pos.w;\n");
+		WRITE(p, "o.pos.z = "I_DEPTHPARAMS".x * o.pos.w + o.pos.z * "I_DEPTHPARAMS".y;\n");
 	}
 
 	WRITE(p, "return o;\n}\n");
diff --git a/Source/Core/VideoCommon/Src/VertexShaderGen.h b/Source/Core/VideoCommon/Src/VertexShaderGen.h
index 0ca2ea181f..1b39c248f0 100644
--- a/Source/Core/VideoCommon/Src/VertexShaderGen.h
+++ b/Source/Core/VideoCommon/Src/VertexShaderGen.h
@@ -35,7 +35,7 @@
 #define I_TRANSFORMMATRICES     "ctrmtx"
 #define I_NORMALMATRICES        "cnmtx"
 #define I_POSTTRANSFORMMATRICES "cpostmtx"
-#define I_FOGPARAMS             "cfog"
+#define I_DEPTHPARAMS           "cDepth"
 
 #define C_POSNORMALMATRIX        0
 #define C_PROJECTION            (C_POSNORMALMATRIX + 6)
@@ -45,8 +45,8 @@
 #define C_TRANSFORMMATRICES     (C_TEXMATRICES + 24)
 #define C_NORMALMATRICES        (C_TRANSFORMMATRICES + 64)
 #define C_POSTTRANSFORMMATRICES (C_NORMALMATRICES + 32)
-#define C_FOGPARAMS             (C_POSTTRANSFORMMATRICES + 64)
-
+#define C_DEPTHPARAMS           (C_POSTTRANSFORMMATRICES + 64)
+#define C_VENVCONST_END			(C_DEPTHPARAMS + 4)
 
 class VERTEXSHADERUID
 {
diff --git a/Source/Core/VideoCommon/Src/VertexShaderManager.cpp b/Source/Core/VideoCommon/Src/VertexShaderManager.cpp
index d7d1d83df2..fead1e0797 100644
--- a/Source/Core/VideoCommon/Src/VertexShaderManager.cpp
+++ b/Source/Core/VideoCommon/Src/VertexShaderManager.cpp
@@ -276,7 +276,7 @@ void VertexShaderManager::SetConstants()
 	if (bViewportChanged)
 	{
 		bViewportChanged = false;
-
+		SetVSConstant4f(C_DEPTHPARAMS,xfregs.rawViewport[5]/ 16777216.0f,xfregs.rawViewport[2]/ 16777216.0f,0.0f,0.0f);
 		// This is so implementation-dependent that we can't have it here.
 		UpdateViewport();
 	}
@@ -521,14 +521,6 @@ void VertexShaderManager::SetViewport(float* _Viewport, int constantIndex)
 	{
 		xfregs.rawViewport[constantIndex] = _Viewport[0];
 	}
-	/*//Tino: i think this is not needed so leave this commented till confirmed
-	// Workaround for paper mario, yep this is bizarre.
-	for (size_t i = 0; i < ARRAYSIZE(xfregs.rawViewport); ++i)
-	{
-		if (*(u32*)(_Viewport + i) == 0x7f800000)  // invalid fp number
-			return;
-	}
-	memcpy(xfregs.rawViewport, _Viewport, sizeof(xfregs.rawViewport));*/
 	bViewportChanged = true;
 }
 
diff --git a/Source/Core/VideoCommon/Src/VideoConfig.cpp b/Source/Core/VideoCommon/Src/VideoConfig.cpp
index a2cacffc10..6df9bdd756 100644
--- a/Source/Core/VideoCommon/Src/VideoConfig.cpp
+++ b/Source/Core/VideoCommon/Src/VideoConfig.cpp
@@ -93,7 +93,6 @@ void VideoConfig::Load(const char *ini_file)
 	iniFile.Get("Hacks", "EFBCopyDisable", &bEFBCopyDisable, false);
 	iniFile.Get("Hacks", "EFBCopyDisableHotKey", &bOSDHotKey, 0);
 	iniFile.Get("Hacks", "EFBToTextureEnable", &bCopyEFBToTexture, false);
-	iniFile.Get("Hacks", "EFBVerifyTextureModificationsByCPU",&bVerifyTextureModificationsByCPU,false);
 	iniFile.Get("Hacks", "EFBScaledCopy", &bCopyEFBScaled, true);
 	iniFile.Get("Hacks", "FIFOBPHack", &bFIFOBPhack, false);
 	iniFile.Get("Hacks", "ProjectionHack", &iPhackvalue, 0);
@@ -125,9 +124,7 @@ void VideoConfig::GameIniLoad(const char *ini_file)
 	if (iniFile.Exists("Video", "EFBCopyDisableHotKey"))
 		iniFile.Get("Video", "EFBCopyDisableHotKey", &bOSDHotKey);
 	if (iniFile.Exists("Video", "EFBToTextureEnable"))
-		iniFile.Get("Video", "EFBToTextureEnable", &bCopyEFBToTexture);
-	if (iniFile.Exists("Video", "EFBVerifyTextureModificationsByCPU"))
-		iniFile.Get("Video", "EFBVerifyTextureModificationsByCPU", &bVerifyTextureModificationsByCPU);
+		iniFile.Get("Video", "EFBToTextureEnable", &bCopyEFBToTexture);	
 	if (iniFile.Exists("Video", "EFBScaledCopy"))
 		iniFile.Get("Video", "EFBScaledCopy", &bCopyEFBScaled);
 	if (iniFile.Exists("Video", "SafeTextureCache"))
@@ -202,8 +199,7 @@ void VideoConfig::Save(const char *ini_file)
 	iniFile.Set("Hacks", "EFBAccessEnable", bEFBAccessEnable);
 	iniFile.Set("Hacks", "EFBCopyDisable", bEFBCopyDisable);
 	iniFile.Set("Hacks", "EFBCopyDisableHotKey", bOSDHotKey);
-	iniFile.Set("Hacks", "EFBToTextureEnable", bCopyEFBToTexture);
-	iniFile.Set("Hacks", "EFBVerifyTextureModificationsByCPU", bVerifyTextureModificationsByCPU);
+	iniFile.Set("Hacks", "EFBToTextureEnable", bCopyEFBToTexture);	
 	iniFile.Set("Hacks", "EFBScaledCopy", bCopyEFBScaled);
 	iniFile.Set("Hacks", "FIFOBPHack", bFIFOBPhack);
 	iniFile.Set("Hacks", "ProjectionHack", iPhackvalue);
diff --git a/Source/Core/VideoCommon/Src/VideoConfig.h b/Source/Core/VideoCommon/Src/VideoConfig.h
index 1e9db7b5a2..1184d2a6b4 100644
--- a/Source/Core/VideoCommon/Src/VideoConfig.h
+++ b/Source/Core/VideoCommon/Src/VideoConfig.h
@@ -115,8 +115,7 @@ struct VideoConfig
 	bool bEFBCopyDisable;  // should reverse polarity of this one :) true=disabled can be confusing
 	bool bOSDHotKey;
 	bool bHack;
-	bool bCopyEFBToTexture;
-	bool bVerifyTextureModificationsByCPU;
+	bool bCopyEFBToTexture;	
 	bool bCopyEFBScaled;
 	bool bSafeTextureCache;
 	int iSafeTextureCache_ColorSamples;