glsl: redo color/alpha management correction

Please test it!

GS supports 3 formats for the output:

32 bits: normal case
=> no change

24 bits: like 32 bits but without alpha channel
=> mask alpha channel (ie don't write it anymore)
=> Always uses 1.0f as blending coefficient

16 bits: RGB5A1, emulated by a 32 bits openGL texture. I think it will be more correct to use
a real 16 bits GL texture. Unfortunately it would cost several (slow) target conversions.
Anyway as a current solution
=>  apply a mask of 0xF8 on color when SW blending is used (improve Castlevania shadow)
unfortunately normal blending mode still uses the full range of colors!

This commit also corrects a couple of blending factor. 128/255 is equivalent to 1.0f in PS2, whereas GPU uses 1.0f. So the blending factor must be 255/128 instead of 2

Note: disable CRC hack and enable accurate_colclip to see Castlevania shadow ^^
(issue #380).
Note2: SW renderer is darker on Castlevania. I don't know why maybe linked to the 16 bits format poorly emulated

plugins/GSdx/GSRendererOGL.cpp

@@ -288,6 +288,13 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
 	}
 
 	om_csel.wrgba = ~GSVector4i::load((int)context->FRAME.FBMSK).eq8(GSVector4i::xffffffff()).mask();
+	if (ps_sel.dfmt == 1) {
+		// 24 bits no alpha channel so use 1.0f fix factor as equivalent
+		ALPHA.C = 2;
+		afix = 1.0f;
+		// Disable writing of the alpha channel
+		om_csel.wa = 0;
+	}
 
 	if (DATE) {
 		if (GLLoader::found_GL_ARB_texture_barrier && !PrimitiveOverlap()) {

plugins/GSdx/res/glsl/tfx_fs.glsl

@@ -404,8 +404,13 @@ vec4 ps_color()
 void ps_blend(inout vec4 c, in float As)
 {
 	vec4 rt = texelFetch(RtSampler, ivec2(gl_FragCoord.xy), 0);
+#if PS_DFMT == FMT_24
+	float Ad = 1.0f;
+#else
+	// FIXME FMT_16 case
 	// FIXME Ad or Ad * 2?
-	float Ad = rt.a;
+	float Ad = rt.a * 255.0f / 128.0f;
+#endif
 	// Let the compiler do its jobs !
 	vec3 Cd = rt.rgb;
 	vec3 Cs = c.rgb;
@@ -640,12 +645,26 @@ void ps_blend(inout vec4 c, in float As)
 
 #endif
 
-#if PS_COLCLIP == 3
+	// FIXME dithering
+
+	// Correct the Color value based on the output format
+#if PS_COLCLIP != 3
+	// Standard Clamp
+	c.rgb = clamp(c.rgb, vec3(0.0f), vec3(1.0f));
+#endif
+
+#if PS_DFMT == FMT_16
+	// In 16 bits format, only 5 bits of colors are used. It impacts shadows computation of Castlevania
+
+	// Basically we want to do 'c.rgb &= 0xF8' in denormalized mode
+	c.rgb = vec3(uvec3((c.rgb * 255.0f) + 256.5f) & uvec3(0xF8)) / 255.0f;
+#elif PS_COLCLIP == 3
+	// Basically we want to do 'c.rgb &= 0xFF' in denormalized mode
 	c.rgb = vec3(uvec3((c.rgb * 255.0f) + 256.5f) & uvec3(0xFF)) / 255.0f;
+#endif
 
 	// Don't compile => unable to find compatible overloaded function "mod(vec3)"
 	//c.rgb = mod((c.rgb * 255.0f) + 256.5f) / 255.0f;
-#endif
 }
 
 void ps_main()
@@ -700,14 +719,16 @@ void ps_main()
 	c.a = 0.5f;
 #endif
 
-	float alpha = c.a * 2.0;
+	// Must be done before alpha correction
+	float alpha = c.a * 255.0f / 128.0f;
 
-#if (PS_AOUT != 0) // 16 bit output
+	// Correct the ALPHA value based on the output format
+	// FIXME add support of alpha mask to replace properly PS_AOUT
+#if (PS_DFMT == FMT_16) || (PS_AOUT)
 	float a = 128.0f / 255.0; // alpha output will be 0x80
-
 	c.a = (PS_FBA != 0) ? a : step(0.5, c.a) * a;
-#elif (PS_FBA != 0)
+#elif (PS_DFMT == FMT_32) && (PS_FBA != 0)
-	if(c.a < 0.5) c.a += 0.5;
+	if(c.a < 0.5) c.a += 128.0f/255.0f;
 #endif
 
 	// Get first primitive that will write a failling alpha value

plugins/GSdx/res/glsl_source.h

@@ -1157,8 +1157,13 @@ static const char* tfx_fs_all_glsl =
 	"void ps_blend(inout vec4 c, in float As)\n"
 	"{\n"
 	"	vec4 rt = texelFetch(RtSampler, ivec2(gl_FragCoord.xy), 0);\n"
+	"#if PS_DFMT == FMT_24\n"
+	"	float Ad = 1.0f;\n"
+	"#else\n"
+	"	// FIXME FMT_16 case\n"
 	"	// FIXME Ad or Ad * 2?\n"
-	"	float Ad = rt.a;\n"
+	"	float Ad = rt.a * 255.0f / 128.0f;\n"
+	"#endif\n"
 	"	// Let the compiler do its jobs !\n"
 	"	vec3 Cd = rt.rgb;\n"
 	"	vec3 Cs = c.rgb;\n"
@@ -1393,12 +1398,26 @@ static const char* tfx_fs_all_glsl =
 	"\n"
 	"#endif\n"
 	"\n"
-	"#if PS_COLCLIP == 3\n"
+	"	// FIXME dithering\n"
+	"\n"
+	"	// Correct the Color value based on the output format\n"
+	"#if PS_COLCLIP != 3\n"
+	"	// Standard Clamp\n"
+	"	c.rgb = clamp(c.rgb, vec3(0.0f), vec3(1.0f));\n"
+	"#endif\n"
+	"\n"
+	"#if PS_DFMT == FMT_16\n"
+	"	// In 16 bits format, only 5 bits of colors are used. It impacts shadows computation of Castlevania\n"
+	"\n"
+	"	// Basically we want to do 'c.rgb &= 0xF8' in denormalized mode\n"
+	"	c.rgb = vec3(uvec3((c.rgb * 255.0f) + 256.5f) & uvec3(0xF8)) / 255.0f;\n"
+	"#elif PS_COLCLIP == 3\n"
+	"	// Basically we want to do 'c.rgb &= 0xFF' in denormalized mode\n"
 	"	c.rgb = vec3(uvec3((c.rgb * 255.0f) + 256.5f) & uvec3(0xFF)) / 255.0f;\n"
+	"#endif\n"
 	"\n"
 	"	// Don't compile => unable to find compatible overloaded function \"mod(vec3)\"\n"
 	"	//c.rgb = mod((c.rgb * 255.0f) + 256.5f) / 255.0f;\n"
-	"#endif\n"
 	"}\n"
 	"\n"
 	"void ps_main()\n"
@@ -1453,14 +1472,16 @@ static const char* tfx_fs_all_glsl =
 	"	c.a = 0.5f;\n"
 	"#endif\n"
 	"\n"
-	"	float alpha = c.a * 2.0;\n"
+	"	// Must be done before alpha correction\n"
+	"	float alpha = c.a * 255.0f / 128.0f;\n"
 	"\n"
-	"#if (PS_AOUT != 0) // 16 bit output\n"
+	"	// Correct the ALPHA value based on the output format\n"
+	"	// FIXME add support of alpha mask to replace properly PS_AOUT\n"
+	"#if (PS_DFMT == FMT_16) || (PS_AOUT)\n"
 	"	float a = 128.0f / 255.0; // alpha output will be 0x80\n"
-	"\n"
 	"	c.a = (PS_FBA != 0) ? a : step(0.5, c.a) * a;\n"
-	"#elif (PS_FBA != 0)\n"
+	"#elif (PS_DFMT == FMT_32) && (PS_FBA != 0)\n"
-	"	if(c.a < 0.5) c.a += 0.5;\n"
+	"	if(c.a < 0.5) c.a += 128.0f/255.0f;\n"
 	"#endif\n"
 	"\n"
 	"	// Get first primitive that will write a failling alpha value\n"