this commit is divided in 4 parts:

1 - Optimize pixelshadergen to avoid redundant register overflow math, with this if a game don't need this will be not applied.
this must bring some fill rate back and improve speed a little in fill rate limited systems.
2- some corrections to vertexshadergen to avoid uninitialized texture coordinates, dono if is the correct way to fix it but t least it will make house of the dead overkill playable in dx11.
the bad thing: still missing geometri in dx9, in dx11 it works exactly as on opengl.
3 - some optimization made to improve fps a little wih the latests changes made to fifo.
* back to the original code in beginfield as now it will work right.
* check for efb access more often as a lot of time is lost waiting for efb access
4 - apply a little fix for missing textures in nvidia opengl tanks to  Wagnard28 for finding that nvidia does not like invalid shader id :)
please test for any regression

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@5812 8ced0084-cf51-0410-be5f-012b33b47a6e

2 files changed, 114 insertions, 55 deletions

diff --git a/Source/Core/VideoCommon/Src/PixelShaderGen.cpp b/Source/Core/VideoCommon/Src/PixelShaderGen.cpp
index afc1400c3a..388e16ffab 100644
--- a/Source/Core/VideoCommon/Src/PixelShaderGen.cpp
+++ b/Source/Core/VideoCommon/Src/PixelShaderGen.cpp
@@ -345,6 +345,15 @@ static char swapModeTable[4][5];
 
 static char text[16384];
 
+struct RegisterState
+{
+	bool ColorNeedOverflowControl;
+	bool AlphaNeedOverflowControl;
+	bool AuxStored;
+};
+
+static RegisterState RegisterStates[4];
+
 static void BuildSwapModeTable()
 {
     for (int i = 0; i < 4; i++)
@@ -383,21 +392,10 @@ const char *GeneratePixelShaderCode(u32 texture_mask, bool dstAlphaEnable,  API_
     }
 
     // Declare samplers
-    if (texture_mask)
+    if (texture_mask && ApiType == API_OPENGL)
 	{
-		if (ApiType == API_D3D11)
-		{
-			WRITE(p, "sampler ");
-		}
-		else if (ApiType == API_D3D9)
-		{
-			WRITE(p, "uniform sampler ");
-		}
-		else
-		{
-			WRITE(p, "uniform samplerRECT ");
-		}
-        bool bfirst = true;
+		WRITE(p, "uniform samplerRECT ");
+		bool bfirst = true;
 		for (int i = 0; i < 8; ++i)
 		{
             if (texture_mask & (1<<i))
@@ -406,21 +404,7 @@ const char *GeneratePixelShaderCode(u32 texture_mask, bool dstAlphaEnable,  API_
                 bfirst = false;
             }
 		}
-		WRITE(p, ";\n");
-		if(ApiType == API_D3D11)
-		{
-			bfirst = true;
-			WRITE(p, "Texture2D ");
-			for (int i = 0; i < 8; ++i)
-			{
-				if (texture_mask & (1<<i))
-				{
-					WRITE(p, "%s Tex%d : register(t%d)", bfirst?"":",", i, i);
-					bfirst = false;
-				}
-			}
-			WRITE(p, ";\n");
-		}        
+		WRITE(p, ";\n");		
 	}
 
     if (texture_mask != 0xff) 
@@ -549,19 +533,39 @@ const char *GeneratePixelShaderCode(u32 texture_mask, bool dstAlphaEnable,  API_
         }
     }
 
+	RegisterStates[0].AlphaNeedOverflowControl = false;
+	RegisterStates[0].ColorNeedOverflowControl = false;
+	RegisterStates[0].AuxStored = false;
+	for(int i = 0 ; i < 4 ; i++)
+	{
+		RegisterStates[i].AlphaNeedOverflowControl = true;
+		RegisterStates[i].ColorNeedOverflowControl = true;
+		RegisterStates[i].AuxStored = false;
+	}
 
 	for (int i = 0; i < numStages; i++)
 		WriteStage(p, i, texture_mask,ApiType); //build the equation for this stage
+
 	if(numStages)
 	{
 		// The results of the last texenv stage are put onto the screen,
 		// regardless of the used destination register
-		WRITE(p, "prev.rgb = %s;\n",tevCOutputTable[bpmem.combiners[numStages-1].colorC.dest]);
-		WRITE(p, "prev.a = %s;\n",tevAOutputTable[bpmem.combiners[numStages-1].alphaC.dest]);
+		if(bpmem.combiners[numStages - 1].colorC.dest != 0)
+		{
+			bool retrieveFromAuxRegister = !RegisterStates[bpmem.combiners[numStages - 1].colorC.dest].ColorNeedOverflowControl && RegisterStates[bpmem.combiners[numStages - 1].colorC.dest].AuxStored;
+			WRITE(p, "prev.rgb = %s%s;\n", retrieveFromAuxRegister ? "c" : "" ,tevCOutputTable[bpmem.combiners[numStages - 1].colorC.dest]);
+			RegisterStates[0].ColorNeedOverflowControl = RegisterStates[bpmem.combiners[numStages - 1].colorC.dest].ColorNeedOverflowControl;
+		}
+		if(bpmem.combiners[numStages - 1].alphaC.dest != 0)
+		{
+			bool retrieveFromAuxRegister = !RegisterStates[bpmem.combiners[numStages - 1].alphaC.dest].AlphaNeedOverflowControl && RegisterStates[bpmem.combiners[numStages - 1].alphaC.dest].AuxStored;
+			WRITE(p, "prev.a = %s%s;\n", retrieveFromAuxRegister ? "c" : "" ,tevAOutputTable[bpmem.combiners[numStages - 1].alphaC.dest]);		
+			RegisterStates[0].AlphaNeedOverflowControl = RegisterStates[bpmem.combiners[numStages - 1].alphaC.dest].AlphaNeedOverflowControl;		
+		}		
 	}
-
-	// emulation of unisgned 8 overflow when casting
-	WRITE(p, "prev = frac(4.0f + prev * (255.0f/256.0f)) * (256.0f/255.0f);\n");
+	// emulation of unisgned 8 overflow when casting if needed
+	if(RegisterStates[0].AlphaNeedOverflowControl || RegisterStates[0].ColorNeedOverflowControl)
+		WRITE(p, "prev = frac(4.0f + prev * (255.0f/256.0f)) * (256.0f/255.0f);\n");	
 		
     if (!WriteAlphaTest(p, ApiType))
 	{
@@ -782,7 +786,19 @@ static void WriteStage(char *&p, int n, u32 texture_mask, API_TYPE ApiType)
 	|| ac.a == TEVALPHAARG_APREV 
 	|| ac.b == TEVALPHAARG_APREV 
 	|| ac.c == TEVALPHAARG_APREV)
-		WRITE(p, "cprev = frac(4.0f + prev * (255.0f/256.0f)) * (256.0f/255.0f);\n");  
+	{
+		if(RegisterStates[0].AlphaNeedOverflowControl || RegisterStates[0].ColorNeedOverflowControl)
+		{
+			WRITE(p, "cprev = frac(4.0f + prev * (255.0f/256.0f)) * (256.0f/255.0f);\n");  
+			RegisterStates[0].AlphaNeedOverflowControl = false;
+			RegisterStates[0].ColorNeedOverflowControl = false;			
+		}
+		else
+		{
+			WRITE(p, "cprev = prev;\n");  
+		}
+		RegisterStates[0].AuxStored = true;
+	}
 
 	if(cc.a == TEVCOLORARG_C0 
 	|| cc.a == TEVCOLORARG_A0 
@@ -793,7 +809,20 @@ static void WriteStage(char *&p, int n, u32 texture_mask, API_TYPE ApiType)
 	|| ac.a == TEVALPHAARG_A0 
 	|| ac.b == TEVALPHAARG_A0 
 	|| ac.c == TEVALPHAARG_A0)
-		WRITE(p, "cc0 = frac(4.0f + c0 * (255.0f/256.0f)) * (256.0f/255.0f);\n");
+	{
+		if(RegisterStates[1].AlphaNeedOverflowControl || RegisterStates[1].ColorNeedOverflowControl)
+		{
+			WRITE(p, "cc0 = frac(4.0f + c0 * (255.0f/256.0f)) * (256.0f/255.0f);\n");
+			RegisterStates[1].AlphaNeedOverflowControl = false;
+			RegisterStates[1].ColorNeedOverflowControl = false;
+		}
+		else
+		{
+			WRITE(p, "cc0 = c0;\n");
+		}
+		RegisterStates[1].AuxStored = true;
+	}
+		
 
 	if(cc.a == TEVCOLORARG_C1 
 	|| cc.a == TEVCOLORARG_A1 
@@ -804,7 +833,19 @@ static void WriteStage(char *&p, int n, u32 texture_mask, API_TYPE ApiType)
 	|| ac.a == TEVALPHAARG_A1 
 	|| ac.b == TEVALPHAARG_A1 
 	|| ac.c == TEVALPHAARG_A1)
-		WRITE(p, "cc1 = frac(4.0f + c1 * (255.0f/256.0f)) * (256.0f/255.0f);\n");  
+	{
+		if(RegisterStates[2].AlphaNeedOverflowControl || RegisterStates[2].ColorNeedOverflowControl)
+		{
+			WRITE(p, "cc1 = frac(4.0f + c1 * (255.0f/256.0f)) * (256.0f/255.0f);\n");
+			RegisterStates[2].AlphaNeedOverflowControl = false;
+			RegisterStates[2].ColorNeedOverflowControl = false;
+		}
+		else
+		{
+			WRITE(p, "cc1 = c1;\n");
+		}
+		RegisterStates[2].AuxStored = true;
+	}
 	
 	if(cc.a == TEVCOLORARG_C2 
 	|| cc.a == TEVCOLORARG_A2 
@@ -815,8 +856,22 @@ static void WriteStage(char *&p, int n, u32 texture_mask, API_TYPE ApiType)
 	|| ac.a == TEVALPHAARG_A2 
 	|| ac.b == TEVALPHAARG_A2 
 	|| ac.c == TEVALPHAARG_A2)
-		WRITE(p, "cc2 = frac(4.0f + c2 * (255.0f/256.0f)) * (256.0f/255.0f);\n");  	
+	{
+		if(RegisterStates[3].AlphaNeedOverflowControl || RegisterStates[3].ColorNeedOverflowControl)
+		{
+			WRITE(p, "cc2 = frac(4.0f + c2 * (255.0f/256.0f)) * (256.0f/255.0f);\n");
+			RegisterStates[3].AlphaNeedOverflowControl = false;
+			RegisterStates[3].ColorNeedOverflowControl = false;
+		}
+		else
+		{
+			WRITE(p, "cc2 = c2;\n");
+		}
+		RegisterStates[3].AuxStored = true;
+	}
 
+	RegisterStates[cc.dest].ColorNeedOverflowControl = (cc.clamp == 0);
+	RegisterStates[cc.dest].AuxStored = false;
     if (cc.clamp)
 		WRITE(p, "%s=saturate(", tevCOutputTable[cc.dest]);
 	else
@@ -863,6 +918,8 @@ static void WriteStage(char *&p, int n, u32 texture_mask, API_TYPE ApiType)
 		WRITE(p,")");
 	WRITE(p,";\n");	
     
+	RegisterStates[ac.dest].AlphaNeedOverflowControl = (ac.clamp == 0);
+	RegisterStates[ac.dest].AuxStored = false;
     // combine the alpha channel
     if (ac.clamp)
 	    WRITE(p, "%s=saturate(", tevAOutputTable[ac.dest]);
diff --git a/Source/Core/VideoCommon/Src/VertexShaderGen.cpp b/Source/Core/VideoCommon/Src/VertexShaderGen.cpp
index 7f43508b02..53a2682314 100644
--- a/Source/Core/VideoCommon/Src/VertexShaderGen.cpp
+++ b/Source/Core/VideoCommon/Src/VertexShaderGen.cpp
@@ -332,7 +332,7 @@ const char *GenerateVertexShaderCode(u32 components, API_TYPE api_type)
 
     // zero left over channels
     for (int i = xfregs.nNumChans; i < 2; ++i)
-		WRITE(p, "o.colors[%d] = float4(0.0f,0.0f,0.0f,0.0f);\n", i);
+		WRITE(p, "o.colors[%d] = float4(0.0f,0.0f,0.0f,1.0f);\n", i);
 
 	// special case if only pos and tex coord 0 and tex coord input is AB11	
 	// donko - this has caused problems in some games. removed for now.
@@ -349,6 +349,7 @@ const char *GenerateVertexShaderCode(u32 components, API_TYPE api_type)
         TexMtxInfo& texinfo = xfregs.texcoords[i].texmtxinfo;
 
         WRITE(p, "{\n");
+		WRITE(p, "coord = float4(0.0f,0.0f,1.0f,1.0f);\n"); 
         switch (texinfo.sourcerow) {
         case XF_SRCGEOM_INROW:
             _assert_( texinfo.inputform == XF_TEXINPUT_ABC1 );
@@ -384,27 +385,12 @@ const char *GenerateVertexShaderCode(u32 components, API_TYPE api_type)
 
         // firs transformation
         switch (texinfo.texgentype) {
-            case XF_TEXGEN_REGULAR:
-                if (components & (VB_HAS_TEXMTXIDX0<<i)) {
-                    if (texinfo.projection == XF_TEXPROJ_STQ )
-                        WRITE(p, "o.tex%d.xyz = float3(dot(coord, "I_TRANSFORMMATRICES".T[tex%d.z].t), dot(coord, "I_TRANSFORMMATRICES".T[tex%d.z+1].t), dot(coord, "I_TRANSFORMMATRICES".T[tex%d.z+2].t));\n", i, i, i, i);
-                    else {
-                        WRITE(p, "o.tex%d.xyz = float3(dot(coord, "I_TRANSFORMMATRICES".T[tex%d.z].t), dot(coord, "I_TRANSFORMMATRICES".T[tex%d.z+1].t), 1);\n", i, i, i);
-                    }
-                }
-                else {
-                    if (texinfo.projection == XF_TEXPROJ_STQ )
-                        WRITE(p, "o.tex%d.xyz = float3(dot(coord, "I_TEXMATRICES".T[%d].t), dot(coord, "I_TEXMATRICES".T[%d].t), dot(coord, "I_TEXMATRICES".T[%d].t));\n", i, 3*i, 3*i+1, 3*i+2);
-                    else
-                        WRITE(p, "o.tex%d.xyz = float3(dot(coord, "I_TEXMATRICES".T[%d].t), dot(coord, "I_TEXMATRICES".T[%d].t), 1);\n", i, 3*i, 3*i+1);
-                }
-                break;
             case XF_TEXGEN_EMBOSS_MAP: // calculate tex coords into bump map
 
                 if (components & (VB_HAS_NRM1|VB_HAS_NRM2)) {
                     // transform the light dir into tangent space
                     WRITE(p, "ldir = normalize("I_LIGHTS".lights[%d].pos.xyz - pos.xyz);\n", texinfo.embosslightshift); 
-                    WRITE(p, "o.tex%d.xyz = o.tex%d.xyz + float3(dot(ldir, _norm1), dot(ldir, _norm2), 0.0f);\n", i, texinfo.embosssourceshift);
+                    WRITE(p, "o.tex%d.xyz = coord.xyz + float3(dot(ldir, _norm1), dot(ldir, _norm2), 0.0f);\n", i, texinfo.embosssourceshift);
                 }
                 else _assert_(0); // should have normals
 
@@ -417,6 +403,22 @@ const char *GenerateVertexShaderCode(u32 components, API_TYPE api_type)
                 _assert_(texinfo.sourcerow == XF_SRCCOLORS_INROW);
                 WRITE(p, "o.tex%d.xyz = float3(o.colors[1].x, o.colors[1].y, 1);\n", i);
                 break;
+			case XF_TEXGEN_REGULAR:
+			default:
+                if (components & (VB_HAS_TEXMTXIDX0<<i)) {
+                    if (texinfo.projection == XF_TEXPROJ_STQ )
+                        WRITE(p, "o.tex%d.xyz = float3(dot(coord, "I_TRANSFORMMATRICES".T[tex%d.z].t), dot(coord, "I_TRANSFORMMATRICES".T[tex%d.z+1].t), dot(coord, "I_TRANSFORMMATRICES".T[tex%d.z+2].t));\n", i, i, i, i);
+                    else {
+                        WRITE(p, "o.tex%d.xyz = float3(dot(coord, "I_TRANSFORMMATRICES".T[tex%d.z].t), dot(coord, "I_TRANSFORMMATRICES".T[tex%d.z+1].t), 1);\n", i, i, i);
+                    }
+                }
+                else {
+                    if (texinfo.projection == XF_TEXPROJ_STQ )
+                        WRITE(p, "o.tex%d.xyz = float3(dot(coord, "I_TEXMATRICES".T[%d].t), dot(coord, "I_TEXMATRICES".T[%d].t), dot(coord, "I_TEXMATRICES".T[%d].t));\n", i, 3*i, 3*i+1, 3*i+2);
+                    else
+                        WRITE(p, "o.tex%d.xyz = float3(dot(coord, "I_TEXMATRICES".T[%d].t), dot(coord, "I_TEXMATRICES".T[%d].t), 1);\n", i, 3*i, 3*i+1);
+                }
+                break;
         }
 
         if (xfregs.bEnableDualTexTransform && texinfo.texgentype == XF_TEXGEN_REGULAR) { // only works for regular tex gen types?