gsdx-ogl: replace old colclip algo with the HDR algo

Similar speed but more accurate

Allow to clean the code

plugins/GSdx/GSDeviceOGL.cpp

@@ -667,7 +667,7 @@ GLuint GSDeviceOGL::CompilePS(PSSelector sel)
 		+ format("#define PS_SHUFFLE %d\n", sel.shuffle)
 		+ format("#define PS_READ_BA %d\n", sel.read_ba)
 		+ format("#define PS_FBMASK %d\n", sel.fbmask)
-		+ format("#define PS_BLEND_ACCU %d\n", sel.blend_accu)
+		+ format("#define PS_HDR %d\n", sel.hdr)
 		;
 
 	return m_shader->Compile("tfx.glsl", "ps_main", GL_FRAGMENT_SHADER, tfx_fs_all_glsl, macro);
@@ -1551,13 +1551,13 @@ const GSDeviceOGL::D3D9Blend GSDeviceOGL::m_blendMapD3D9[3*3*3*3] =
 	{ 17                 , D3DBLENDOP_ADD         , D3DBLEND_BLENDFACTOR    , D3DBLEND_INVBLENDFACTOR} , // 0121: (Cs - Cd)*F  + Cd ==> Cs*F + Cd*(1 - F)
 	{ 18                 , D3DBLENDOP_SUBTRACT    , D3DBLEND_BLENDFACTOR    , D3DBLEND_BLENDFACTOR}    , // 0122: (Cs - Cd)*F  +  0 ==> Cs*F - Cd*F
 	{ NO_BAR | A_MAX | 7 , D3DBLENDOP_ADD         , D3DBLEND_SRCALPHA       , D3DBLEND_ZERO}           , //*0200: (Cs -  0)*As + Cs ==> Cs*(As + 1)
-	{ 19                 , D3DBLENDOP_ADD         , D3DBLEND_SRCALPHA       , D3DBLEND_ONE}            , // 0201: (Cs -  0)*As + Cd ==> Cs*As + Cd
+	{ BLEND_ACCU | 19    , D3DBLENDOP_ADD         , D3DBLEND_ONE            , D3DBLEND_ONE}            , // 0201: (Cs -  0)*As + Cd ==> Cs*As + Cd
 	{ NO_BAR | 20        , D3DBLENDOP_ADD         , D3DBLEND_SRCALPHA       , D3DBLEND_ZERO}           , // 0202: (Cs -  0)*As +  0 ==> Cs*As
 	{ A_MAX | 8          , D3DBLENDOP_ADD         , D3DBLEND_DESTALPHA      , D3DBLEND_ZERO}           , //*0210: (Cs -  0)*Ad + Cs ==> Cs*(Ad + 1)
 	{ 21                 , D3DBLENDOP_ADD         , D3DBLEND_DESTALPHA      , D3DBLEND_ONE}            , // 0211: (Cs -  0)*Ad + Cd ==> Cs*Ad + Cd
 	{ 22                 , D3DBLENDOP_ADD         , D3DBLEND_DESTALPHA      , D3DBLEND_ZERO}           , // 0212: (Cs -  0)*Ad +  0 ==> Cs*Ad
 	{ NO_BAR| A_MAX | 9  , D3DBLENDOP_ADD         , D3DBLEND_BLENDFACTOR    , D3DBLEND_ZERO}           , //*0220: (Cs -  0)*F  + Cs ==> Cs*(F + 1)
-	{ 23                 , D3DBLENDOP_ADD         , D3DBLEND_BLENDFACTOR    , D3DBLEND_ONE}            , // 0221: (Cs -  0)*F  + Cd ==> Cs*F + Cd
+	{ BLEND_ACCU | 23    , D3DBLENDOP_ADD         , D3DBLEND_ONE            , D3DBLEND_ONE}            , // 0221: (Cs -  0)*F  + Cd ==> Cs*F + Cd
 	{ NO_BAR | 24        , D3DBLENDOP_ADD         , D3DBLEND_BLENDFACTOR    , D3DBLEND_ZERO}           , // 0222: (Cs -  0)*F  +  0 ==> Cs*F
 	{ 25                 , D3DBLENDOP_ADD         , D3DBLEND_INVSRCALPHA    , D3DBLEND_SRCALPHA}       , // 1000: (Cd - Cs)*As + Cs ==> Cd*As + Cs*(1 - As)
 	{ A_MAX | 10         , D3DBLENDOP_REVSUBTRACT , D3DBLEND_SRCALPHA       , D3DBLEND_SRCALPHA}       , //*1001: (Cd - Cs)*As + Cd ==> Cd*(As + 1) - Cs*As

plugins/GSdx/GSDeviceOGL.h

@@ -33,6 +33,7 @@
 #define A_MAX	(0x100)	 // Impossible blending uses coeff bigger than 1
 #define C_CLR	(0x200)	 // Clear color blending (use directly the destination color as blending factor)
 #define NO_BAR  (0x400)  // don't require texture barrier for the blending (because the RT is not used)
+#define BLEND_ACCU (0x800) // Allow to use a mix of SW and HW blending to keep the best of the 2 worlds
 
 #ifdef ENABLE_OGL_DEBUG_MEM_BW
 extern uint64 g_real_texture_upload_byte;
@@ -350,8 +351,8 @@ class GSDeviceOGL : public GSDevice
 				uint32 blend_b:2;
 				uint32 blend_c:2;
 				uint32 blend_d:2;
-				uint32 blend_accu:1;
 				uint32 dfmt:2;
+				uint32 hdr:1;
 
 				uint32 _free2:21;
 			};
@@ -451,9 +452,8 @@ class GSDeviceOGL : public GSDevice
 				uint32 c:2;
 				uint32 d:2;
 				uint32 negative:1;
-				uint32 accu:1;
 
-				uint32 _free:21;
+				uint32 _free:22;
 			};
 
 			struct

plugins/GSdx/GSRendererOGL.cpp

@@ -357,30 +357,26 @@ bool GSRendererOGL::EmulateBlending(GSDeviceOGL::PSSelector& ps_sel, GSDeviceOGL
 	// We really need SW blending for this one, barely used
 	bool impossible_blend = (blend_flag & A_MAX);
 	// Do the multiplication in shader for blending accumulation: Cs*As + Cd or Cs*Af + Cd
-	bool accumulation_blend = (ALPHA.A == 0 && ALPHA.B == 2 && ALPHA.C != 1 && ALPHA.D == 1);
+	bool accumulation_blend = (blend_flag & BLEND_ACCU);
 
-	bool sw_blending_base = m_sw_blending && (free_blend || impossible_blend || ps_sel.blend_accu);
+	bool sw_blending_base = m_sw_blending && (free_blend || impossible_blend);
 
 	// Color clip
-	bool acc_colclip_wrap = false;
 	if (m_env.COLCLAMP.CLAMP == 0) {
-		acc_colclip_wrap =  (m_sw_blending >= ACC_BLEND_CCLIP || sw_blending_base);
-		if (acc_colclip_wrap) {
+		if (accumulation_blend) {
+			ps_sel.hdr = 1;
+			GL_INS("COLCLIP Fast HDR mode ENABLED");
+		} else if (m_sw_blending >= ACC_BLEND_CCLIP || sw_blending_base) {
 			ps_sel.colclip = 3;
+			sw_blending_base = true;
 			GL_INS("COLCLIP SW ENABLED (blending is %d/%d/%d/%d)", ALPHA.A, ALPHA.B, ALPHA.C, ALPHA.D);
-		} else if (!PRIM->TME && PRIM->PRIM != GS_POINTLIST) {
-			// Standard (inaccurate) colclip
-			ps_sel.colclip = 1;
-			accumulation_blend = false;
-			GL_INS("COLCLIP ENABLED (blending is %d/%d/%d/%d)", ALPHA.A, ALPHA.B, ALPHA.C, ALPHA.D);
+		} else {
+			fprintf(stderr, "Sorry colclip isn't supported\n");
 		}
 	}
 
 	// Note: Option is duplicated, one impact the blend unit / the other the shader.
-	if (accumulation_blend && m_sw_blending) {
-		om_bsel.accu = ps_sel.blend_accu = 1;
-		sw_blending_base = true;
-	}
+	sw_blending_base |= accumulation_blend;
 
 	bool all_blend_sw;
 	switch (m_sw_blending) {
@@ -393,7 +389,6 @@ bool GSRendererOGL::EmulateBlending(GSDeviceOGL::PSSelector& ps_sel, GSDeviceOGL
 
 	bool sw_blending = sw_blending_base // Free case or Impossible blend
 		|| all_blend_sw // all blend
-		|| acc_colclip_wrap // accurate colclip
 		|| ps_sel.fbmask; // accurate fbmask
 
 
@@ -402,6 +397,9 @@ bool GSRendererOGL::EmulateBlending(GSDeviceOGL::PSSelector& ps_sel, GSDeviceOGL
 	// that write the bad alpha value. Sw blending will force the draw to run primitive by primitive
 	// (therefore primitiveID will be constant to 1)
 	sw_blending &= !DATE_GL42;
+	// Seriously don't expect me to support this kind of crazyness.
+	// No mix of COLCLIP + accumulation_blend + DATE GL42
+	ASSERT(!(ps_sel.hdr && DATE_GL42));
 
 	// For stat to optimize accurate option
 #if 0
@@ -409,20 +407,28 @@ bool GSRendererOGL::EmulateBlending(GSDeviceOGL::PSSelector& ps_sel, GSDeviceOGL
 			om_bsel.a, om_bsel.b,  om_bsel.c, om_bsel.d, m_env.COLCLAMP.CLAMP, m_vt.m_primclass, m_vertex.next, sw_blending);
 #endif
 	if (sw_blending) {
-		// Disable HW blending except in accu mode
-		om_bsel.abe = ps_sel.blend_accu;
 		ps_sel.blend_a = om_bsel.a;
 		ps_sel.blend_b = om_bsel.b;
 		ps_sel.blend_c = om_bsel.c;
 		ps_sel.blend_d = om_bsel.d;
 
+		if (accumulation_blend) {
+			// Keep HW blending to do the addition
+			om_bsel.abe = 1;
+			// Remove the addition from the SW blending
+			ps_sel.blend_d = 2;
+		} else {
+			// Disable HW blending
+			om_bsel.abe = 0;
+		}
+
 		// Require the fix alpha vlaue
 		if (ALPHA.C == 2) {
 			ps_cb.AlphaCoeff.a = afix;
 		}
 
 		// No need to flush for every primitive
-		require_barrier |= !(blend_flag & NO_BAR) && !ps_sel.blend_accu;
+		require_barrier |= !(blend_flag & NO_BAR) && !accumulation_blend;
 	} else {
 		ps_sel.clr1 = om_bsel.IsCLR1();
 		if (ps_sel.dfmt == 1 && ALPHA.C == 1) {
@@ -924,7 +930,7 @@ void GSRendererOGL::DrawPrims(GSTexture* rt, GSTexture* ds, GSTextureCache::Sour
 		GL_POP();
 	}
 
-	if (m_env.COLCLAMP.CLAMP == 0 && om_bsel.accu) {
+	if (ps_sel.hdr) {
 		hdr_rt = dev->CreateTexture(rtsize.x, rtsize.y, GL_RGBA16F);
 
 		dev->CopyRectConv(rt, hdr_rt, ComputeBoundingBox(rtscale, rtsize), false);

plugins/GSdx/GSTextureFXOGL.cpp

@@ -105,10 +105,7 @@ GSBlendStateOGL* GSDeviceOGL::CreateBlend(OMBlendSelector bsel, float afix)
 	{
 		int i = ((bsel.a * 3 + bsel.b) * 3 + bsel.c) * 3 + bsel.d;
 
-		if (bsel.accu)
-			bs->SetRGB(GL_FUNC_ADD, GL_ONE, GL_ONE);
-		else
-			bs->SetRGB(m_blendMapD3D9[i].op, m_blendMapD3D9[i].src, m_blendMapD3D9[i].dst);
+		bs->SetRGB(m_blendMapD3D9[i].op, m_blendMapD3D9[i].src, m_blendMapD3D9[i].dst);
 
 		if (m_blendMapD3D9[i].bogus & A_MAX) {
 			if (!theApp.GetConfig("accurate_blending_unit", 1)) {

plugins/GSdx/res/glsl/tfx_fs.glsl

@@ -453,9 +453,6 @@ void ps_blend(inout vec4 Color, float As)
 
 #if PS_BLEND_A == PS_BLEND_B
     Color.rgb = D;
-#elif PS_BLEND_ACCU == 1
-	// The D addition will be done in the blending unit
-	Color.rgb = trunc(A * C);
 #else
     Color.rgb = trunc((A - B) * C + D);
 #endif
@@ -612,8 +609,8 @@ void ps_main()
 
 	ps_fbmask(C);
 
-#if PS_BLEND_ACCU && PS_COLCLIP
-	// Use negative value to avoid overflow of the texture
+#if PS_HDR == 1
+	// Use negative value to avoid overflow of the texture (in accumulation mode)
 	if (any(greaterThan(C.rgb, vec3(128.0f)))) {
 		C.rgb = (C.rgb - 256.0f);
 	}

plugins/GSdx/res/glsl_source.h

@@ -1312,9 +1312,6 @@ static const char* tfx_fs_all_glsl =
 	"\n"
 	"#if PS_BLEND_A == PS_BLEND_B\n"
 	"    Color.rgb = D;\n"
-	"#elif PS_BLEND_ACCU == 1\n"
-	"	// The D addition will be done in the blending unit\n"
-	"	Color.rgb = trunc(A * C);\n"
 	"#else\n"
 	"    Color.rgb = trunc((A - B) * C + D);\n"
 	"#endif\n"
@@ -1471,8 +1468,8 @@ static const char* tfx_fs_all_glsl =
 	"\n"
 	"	ps_fbmask(C);\n"
 	"\n"
-	"#if PS_BLEND_ACCU && PS_COLCLIP\n"
-	"	// Use negative value to avoid overflow of the texture\n"
+	"#if PS_HDR == 1\n"
+	"	// Use negative value to avoid overflow of the texture (in accumulation mode)\n"
 	"	if (any(greaterThan(C.rgb, vec3(128.0f)))) {\n"
 	"		C.rgb = (C.rgb - 256.0f);\n"
 	"	}\n"