Merge pull request #823 from KScorp/depthmatrixshaders

Fixed depth matrix shaders in OpenGL and Direct3D to be more precise.

Source/Core/VideoBackends/D3D/PixelShaderCache.cpp

@@ -119,8 +119,28 @@ const char depth_matrix_program[] = {
 	" in float4 pos : SV_Position,\n"
 	" in float2 uv0 : TEXCOORD0){\n"
 	"	float4 texcol = Tex0.Sample(samp0,uv0);\n"
-	"float4 EncodedDepth = frac((texcol.r * (16777215.0f/16777216.0f)) * float4(1.0f,256.0f,256.0f*256.0f,1.0f));\n"
+
-	"texcol = round(EncodedDepth * (16777216.0f/16777215.0f) * float4(255.0f,255.0f,255.0f,15.0f)) / float4(255.0f,255.0f,255.0f,15.0f);\n"
+	// 255.99998474121 = 16777215/16777216*256
+	"	float workspace = texcol.x * 255.99998474121;\n"
+
+	"	texcol.x = floor(workspace);\n"         // x component
+
+	"	workspace = workspace - texcol.x;\n"    // subtract x component out
+	"	workspace = workspace * 256.0;\n"       // shift left 8 bits
+	"	texcol.y = floor(workspace);\n"         // y component
+
+	"	workspace = workspace - texcol.y;\n"    // subtract y component out
+	"	workspace = workspace * 256.0;\n"       // shift left 8 bits
+	"	texcol.z = floor(workspace);\n"         // z component
+
+	"	texcol.w = texcol.x;\n"                 // duplicate x into w
+
+	"	texcol = texcol / 255.0;\n"             // normalize components to [0.0..1.0]
+
+	"	texcol.w = texcol.w * 15.0;\n"
+	"	texcol.w = floor(texcol.w);\n"
+	"	texcol.w = texcol.w / 15.0;\n"          // w component
+
 	"	ocol0 = float4(dot(texcol,cColMatrix[0]),dot(texcol,cColMatrix[1]),dot(texcol,cColMatrix[2]),dot(texcol,cColMatrix[3])) + cColMatrix[4];\n"
 	"}\n"
 };
@@ -139,8 +159,28 @@ const char depth_matrix_program_msaa[] = {
 	"	for(int i = 0; i < samples; ++i)\n"
 	"		texcol += Tex0.Load(int2(uv0.x*(width), uv0.y*(height)), i);\n"
 	"	texcol /= samples;\n"
-	"float4 EncodedDepth = frac((texcol.r * (16777215.0f/16777216.0f)) * float4(1.0f,256.0f,256.0f*256.0f,16.0f));\n"
+
-	"texcol = round(EncodedDepth * (16777216.0f/16777215.0f) * float4(255.0f,255.0f,255.0f,15.0f)) / float4(255.0f,255.0f,255.0f,15.0f);\n"
+	// 255.99998474121 = 16777215/16777216*256
+	"	float workspace = texcol.x * 255.99998474121;\n"
+
+	"	texcol.x = floor(workspace);\n"         // x component
+
+	"	workspace = workspace - texcol.x;\n"    // subtract x component out
+	"	workspace = workspace * 256.0;\n"       // shift left 8 bits
+	"	texcol.y = floor(workspace);\n"         // y component
+
+	"	workspace = workspace - texcol.y;\n"    // subtract y component out
+	"	workspace = workspace * 256.0;\n"       // shift left 8 bits
+	"	texcol.z = floor(workspace);\n"         // z component
+
+	"	texcol.w = texcol.x;\n"                 // duplicate x into w
+
+	"	texcol = texcol / 255.0;\n"             // normalize components to [0.0..1.0]
+
+	"	texcol.w = texcol.w * 15.0;\n"
+	"	texcol.w = floor(texcol.w);\n"
+	"	texcol.w = texcol.w / 15.0;\n"          // w component
+
 	"	ocol0 = float4(dot(texcol,cColMatrix[0]),dot(texcol,cColMatrix[1]),dot(texcol,cColMatrix[2]),dot(texcol,cColMatrix[3])) + cColMatrix[4];\n"
 	"}\n"
 };

Source/Core/VideoBackends/OGL/TextureCache.cpp

@@ -358,9 +358,29 @@ TextureCache::TextureCache()
 		"\n"
 		"void main(){\n"
 		"	vec4 texcol = texture(samp9, uv0);\n"
-		"	vec4 EncodedDepth = fract((texcol.r * (16777215.0/16777216.0)) * vec4(1.0,256.0,256.0*256.0,1.0));\n"
+
-		"	texcol = round(EncodedDepth * (16777216.0/16777215.0) * vec4(255.0,255.0,255.0,15.0)) / vec4(255.0,255.0,255.0,15.0);\n"
+		// 255.99998474121 = 16777215/16777216*256
-		"	ocol0 = texcol * mat4(colmat[0], colmat[1], colmat[2], colmat[3]) + colmat[4];"
+		"	float workspace = texcol.x * 255.99998474121;\n"
+
+		"	texcol.x = floor(workspace);\n"         // x component
+
+		"	workspace = workspace - texcol.x;\n"    // subtract x component out
+		"	workspace = workspace * 256.0;\n"       // shift left 8 bits
+		"	texcol.y = floor(workspace);\n"         // y component
+
+		"	workspace = workspace - texcol.y;\n"    // subtract y component out
+		"	workspace = workspace * 256.0;\n"       // shift left 8 bits
+		"	texcol.z = floor(workspace);\n"         // z component
+
+		"	texcol.w = texcol.x;\n"                 // duplicate x into w
+
+		"	texcol = texcol / 255.0;\n"             // normalize components to [0.0..1.0]
+
+		"	texcol.w = texcol.w * 15.0;\n"
+		"	texcol.w = floor(texcol.w);\n"
+		"	texcol.w = texcol.w / 15.0;\n"          // w component
+
+		"	ocol0 = texcol * mat4(colmat[0], colmat[1], colmat[2], colmat[3]) + colmat[4];\n"
 		"}\n";
 
 	const char *VProgram =