gsdx-ogl: better support of palette

The purpose of the code is to support alpha channel
of RT uses as an index for a palette texure.

I'm afraid that code will likely break pure palette texture. Only used
if paltex is enabled

It fixes missing shadow in Star Ocean 3 (issue #374) in Native resolution
with filter = 0  (no filtering) or = 2 (normal fitering)

Rendering explanation:
The game emulates a stencil buffer with the alpha channel

The alpha channel of the RT can contains a palette texture index (format 4HH)
The idea is to have a gradient of value in the palette (16/32/48/...).
This way you can implement a +16/-16 and even wrap the alpha value every time
you hit the pixel.

Bilinear filtering breaks the rendering because it interpolates between counts
so you doesn't have the exact count

Upscaling breaks the rendering because the RT is reused as an input texture. It means
that we need to scale it down which again create some interpolations.

plugins/GSdx/GSDeviceOGL.cpp

@@ -626,6 +626,7 @@ GLuint GSDeviceOGL::CompilePS(PSSelector sel)
 		+ format("#define PS_WMS %d\n", sel.wms)
 		+ format("#define PS_WMT %d\n", sel.wmt)
 		+ format("#define PS_FMT %d\n", sel.fmt)
+		+ format("#define PS_IFMT %d\n", sel.ifmt)
 		+ format("#define PS_AEM %d\n", sel.aem)
 		+ format("#define PS_TFX %d\n", sel.tfx)
 		+ format("#define PS_TCC %d\n", sel.tcc)

plugins/GSdx/GSDeviceOGL.h

@@ -315,8 +315,9 @@ class GSDeviceOGL : public GSDevice
 				uint32 wms:2;
 				uint32 wmt:2;
 				uint32 ltf:1;
+				uint32 ifmt:2;
 
-				uint32 _free1:4;
+				uint32 _free1:2;
 
 				// Word 2
 				uint32 blend:8;

plugins/GSdx/GSRendererOGL.cpp

@@ -506,7 +506,16 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
 
 		ps_sel.wms = context->CLAMP.WMS;
 		ps_sel.wmt = context->CLAMP.WMT;
-		ps_sel.fmt = tex->m_palette ? cpsm.fmt | 4 : cpsm.fmt;
+		if (tex->m_palette) {
+			ps_sel.fmt = cpsm.fmt | 4;
+			ps_sel.ifmt = (context->TEX0.PSM == 0x1B) ? 3
+				: (context->TEX0.PSM == 0x24) ? 2
+				: (context->TEX0.PSM == 0x2C) ? 1
+				: 0;
+			GL_INS("Use palette with format %d and index format %d", ps_sel.fmt, ps_sel.ifmt);
+		} else {
+			ps_sel.fmt = cpsm.fmt;
+		}
 		ps_sel.aem = env.TEXA.AEM;
 		ps_sel.tfx = context->TEX0.TFX;
 		ps_sel.tcc = context->TEX0.TCC;
@@ -573,6 +582,7 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
 #ifdef ENABLE_OGL_DEBUG
 		// Unattach texture to avoid noise in debugger
 		dev->PSSetShaderResource(0, NULL);
+		dev->PSSetShaderResource(1, NULL);
 #endif
 	}
 

plugins/GSdx/res/glsl/tfx_fs.glsl

@@ -99,11 +99,9 @@ vec4 sample_c(vec2 uv)
 	return texture(TextureSampler, uv);
 }
 
-vec4 sample_p(float u)
+vec4 sample_p(uint idx)
 {
-	//FIXME do we need a 1D sampler. Big impact on opengl to find 1 dim
-	// So for the moment cheat with 0.0f dunno if it work
-	return texture(PaletteSampler, vec2(u, 0.0f));
+	return texelFetch(PaletteSampler, ivec2(idx, 0u), 0);
 }
 
 vec4 wrapuv(vec4 uv)
@@ -168,21 +166,47 @@ mat4 sample_4c(vec4 uv)
 	return c;
 }
 
-vec4 sample_4a(vec4 uv)
+uvec4 sample_4_index(vec4 uv)
 {
 	vec4 c;
 
-	// Dx used the alpha channel.
-	// Opengl is only 8 bits on red channel.
+	// Either GSdx will send a texture that contains a single channel
+	// in this case we must use the red channel (whereas Dx uses alpha)
+	//
+	// Or we have an old RT (ie RGBA8) that contains index (4/8) in the alpha channel
+
+#if PS_IFMT == 0
+	// Single channel texture
 	c.x = sample_c(uv.xy).r;
 	c.y = sample_c(uv.zy).r;
 	c.z = sample_c(uv.xw).r;
 	c.w = sample_c(uv.zw).r;
+	//return c * 255.0/256.0 + 0.5/256.0;
+#else
+	// 4 channels texture
+	c.x = sample_c(uv.xy).a;
+	c.y = sample_c(uv.zy).a;
+	c.z = sample_c(uv.xw).a;
+	c.w = sample_c(uv.zw).a;
+#endif
+
+	uvec4 i = uvec4(c * 255.0f + 0.5f); // Denormalize value
+	//return (i/uint(16)) & uint(0xF);
+
+#if PS_IFMT == 1
+	// 4HH alpha
+	return i >> 4u;
+#elif PS_IFMT == 2
+	// 4HL alpha
+	return i & 16u;
+#else
+	// 8 bits alpha or red
+	return i;
+#endif
 
-	return c * 255.0/256.0 + 0.5/256.0;
 }
 
-mat4 sample_4p(vec4 u)
+mat4 sample_4p(uvec4 u)
 {
 	mat4 c;
 
@@ -231,7 +255,7 @@ vec4 sample_color(vec2 st, float q)
 
 	if((PS_FMT & FMT_PAL) != 0)
 	{
-		c = sample_4p(sample_4a(uv));
+		c = sample_4p(sample_4_index(uv));
 	}
 	else
 	{

plugins/GSdx/res/glsl_source.h

@@ -852,11 +852,9 @@ static const char* tfx_fs_all_glsl =
 	"	return texture(TextureSampler, uv);\n"
 	"}\n"
 	"\n"
-	"vec4 sample_p(float u)\n"
+	"vec4 sample_p(uint idx)\n"
 	"{\n"
-	"	//FIXME do we need a 1D sampler. Big impact on opengl to find 1 dim\n"
-	"	// So for the moment cheat with 0.0f dunno if it work\n"
-	"	return texture(PaletteSampler, vec2(u, 0.0f));\n"
+	"	return texelFetch(PaletteSampler, ivec2(idx, 0u), 0);\n"
 	"}\n"
 	"\n"
 	"vec4 wrapuv(vec4 uv)\n"
@@ -921,21 +919,47 @@ static const char* tfx_fs_all_glsl =
 	"	return c;\n"
 	"}\n"
 	"\n"
-	"vec4 sample_4a(vec4 uv)\n"
+	"uvec4 sample_4_index(vec4 uv)\n"
 	"{\n"
 	"	vec4 c;\n"
 	"\n"
-	"	// Dx used the alpha channel.\n"
-	"	// Opengl is only 8 bits on red channel.\n"
+	"	// Either GSdx will send a texture that contains a single channel\n"
+	"	// in this case we must use the red channel (whereas Dx uses alpha)\n"
+	"	//\n"
+	"	// Or we have an old RT (ie RGBA8) that contains index (4/8) in the alpha channel\n"
+	"\n"
+	"#if PS_IFMT == 0\n"
+	"	// Single channel texture\n"
 	"	c.x = sample_c(uv.xy).r;\n"
 	"	c.y = sample_c(uv.zy).r;\n"
 	"	c.z = sample_c(uv.xw).r;\n"
 	"	c.w = sample_c(uv.zw).r;\n"
+	"	//return c * 255.0/256.0 + 0.5/256.0;\n"
+	"#else\n"
+	"	// 4 channels texture\n"
+	"	c.x = sample_c(uv.xy).a;\n"
+	"	c.y = sample_c(uv.zy).a;\n"
+	"	c.z = sample_c(uv.xw).a;\n"
+	"	c.w = sample_c(uv.zw).a;\n"
+	"#endif\n"
+	"\n"
+	"	uvec4 i = uvec4(c * 255.0f + 0.5f); // Denormalize value\n"
+	"	//return (i/uint(16)) & uint(0xF);\n"
+	"\n"
+	"#if PS_IFMT == 1\n"
+	"	// 4HH alpha\n"
+	"	return i >> 4u;\n"
+	"#elif PS_IFMT == 2\n"
+	"	// 4HL alpha\n"
+	"	return i & 16u;\n"
+	"#else\n"
+	"	// 8 bits alpha or red\n"
+	"	return i;\n"
+	"#endif\n"
 	"\n"
-	"	return c * 255.0/256.0 + 0.5/256.0;\n"
 	"}\n"
 	"\n"
-	"mat4 sample_4p(vec4 u)\n"
+	"mat4 sample_4p(uvec4 u)\n"
 	"{\n"
 	"	mat4 c;\n"
 	"\n"
@@ -984,7 +1008,7 @@ static const char* tfx_fs_all_glsl =
 	"\n"
 	"	if((PS_FMT & FMT_PAL) != 0)\n"
 	"	{\n"
-	"		c = sample_4p(sample_4a(uv));\n"
+	"		c = sample_4p(sample_4_index(uv));\n"
 	"	}\n"
 	"	else\n"
 	"	{\n"