GSdx: GPU accelerate 8 bits texture conversion port from OpenGL to Di…

…rect3D11.

Commit:
d29e375

Only native res is supported currently, but it's still great progress.
Someone needs to port the ScalingFactor to
D3D from commit: 6121677

Credits to KrossX for porting the shader.

plugins/GSdx/GSDevice11.h

@@ -95,7 +95,7 @@ class GSDevice11 : public GSDeviceDX
 	{
 		CComPtr<ID3D11InputLayout> il;
 		CComPtr<ID3D11VertexShader> vs;
-		CComPtr<ID3D11PixelShader> ps[10];
+		CComPtr<ID3D11PixelShader> ps[18];
 		CComPtr<ID3D11SamplerState> ln;
 		CComPtr<ID3D11SamplerState> pt;
 		CComPtr<ID3D11DepthStencilState> dss;

plugins/GSdx/GSTextureCache.cpp

@@ -23,13 +23,15 @@
 #include "GSTextureCache.h"
 #include "GSUtil.h"
 
+bool s_IS_DIRECT3D11 = false;
 bool s_IS_OPENGL = false;
 bool GSTextureCache::m_disable_partial_invalidation = false;
 bool GSTextureCache::m_wrap_gs_mem = false;
 
 GSTextureCache::GSTextureCache(GSRenderer* r)
 	: m_renderer(r)
 {
+	s_IS_DIRECT3D11 = theApp.GetCurrentRendererType() == GSRendererType::DX1011_HW;
 	s_IS_OPENGL = theApp.GetCurrentRendererType() == GSRendererType::OGL_HW;
 
 	if (theApp.GetConfigB("UserHacks")) {
@@ -1187,7 +1189,7 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
 		// TODO: clean up this mess
 
 		int shader = dst->m_type != RenderTarget ? ShaderConvert_FLOAT32_TO_RGBA8 : ShaderConvert_COPY;
-		bool is_8bits = TEX0.PSM == PSM_PSMT8 && s_IS_OPENGL;
+		bool is_8bits = TEX0.PSM == PSM_PSMT8 && (s_IS_DIRECT3D11 || s_IS_OPENGL);
 
 		if (is_8bits) {
 			GL_INS("Reading RT as a packed-indexed 8 bits format");

plugins/GSdx/res/convert.fx

@@ -219,6 +219,162 @@ PS_OUTPUT ps_main9(PS_INPUT input) // triangular
 	return output;
 }
 
+// DUMMY SHADERS START
+PS_OUTPUT ps_main10(PS_INPUT input)
+{
+	PS_OUTPUT output;
+
+	output.c = 0;
+
+	return output;
+}
+
+PS_OUTPUT ps_main11(PS_INPUT input)
+{
+	PS_OUTPUT output;
+
+	output.c = 0;
+
+	return output;
+}
+
+PS_OUTPUT ps_main12(PS_INPUT input)
+{
+	PS_OUTPUT output;
+
+	output.c = 0;
+
+	return output;
+}
+
+PS_OUTPUT ps_main13(PS_INPUT input)
+{
+	PS_OUTPUT output;
+
+	output.c = 0;
+
+	return output;
+}
+
+PS_OUTPUT ps_main14(PS_INPUT input)
+{
+	PS_OUTPUT output;
+
+	output.c = 0;
+
+	return output;
+}
+
+PS_OUTPUT ps_main15(PS_INPUT input)
+{
+	PS_OUTPUT output;
+
+	output.c = 0;
+
+	return output;
+}
+
+PS_OUTPUT ps_main16(PS_INPUT input)
+{
+	PS_OUTPUT output;
+
+	output.c = 0;
+
+	return output;
+}
+// DUMMY SHADERS END
+
+PS_OUTPUT ps_main17(PS_INPUT input)
+{
+	PS_OUTPUT output;
+
+	// Potential speed optimization. There is a high probability that
+	// game only want to extract a single channel (blue). It will allow
+	// to remove most of the conditional operation and yield a +2/3 fps
+	// boost on MGS3
+	//
+	// Hypothesis wrong in Prince of Persia ... Seriously WTF !
+	//#define ONLY_BLUE;
+
+	// Convert a RGBA texture into a 8 bits packed texture
+	// Input column: 8x2 RGBA pixels
+	// 0: 8 RGBA
+	// 1: 8 RGBA
+	// Output column: 16x4 Index pixels
+	// 0: 8 R | 8 B
+	// 1: 8 R | 8 B
+	// 2: 8 G | 8 A
+	// 3: 8 G | 8 A
+	float c;
+
+	uint2 sel = uint2(input.p.xy) % uint2(16u, 16u);
+	int2  tb  = ((int2(input.p.xy) & ~int2(15, 3)) >> 1);
+
+	int ty   = tb.y | (int(input.p.y) & 1);
+	int txN  = tb.x | (int(input.p.x) & 7);
+	int txH  = tb.x | ((int(input.p.x) + 4) & 7);
+
+	//txN *= ScalingFactor.x;
+	//txH *= ScalingFactor.x;
+	//ty  *= ScalingFactor.y;
+
+	// TODO investigate texture gather
+	float4 cN = Texture.Load(int3(txN, ty, 0));
+	float4 cH = Texture.Load(int3(txH, ty, 0));
+
+
+	if ((sel.y & 4u) == 0u)
+	{
+#ifdef ONLY_BLUE
+		c = cN.b;
+#else
+		// Column 0 and 2
+		if ((sel.y & 3u) < 2u)
+		{
+			// First 2 lines of the col
+			if (sel.x < 8u)
+				c = cN.r;
+			else
+				c = cN.b;
+		}
+		else
+		{
+			if (sel.x < 8u)
+				c = cH.g;
+			else
+				c = cH.a;
+		}
+#endif
+	}
+	else
+	{
+#ifdef ONLY_BLUE
+		c = cH.b;
+#else
+		// Column 1 and 3
+		if ((sel.y & 3u) < 2u)
+		{
+			// First 2 lines of the col
+			if (sel.x < 8u)
+				c = cH.r;
+			else
+				c = cH.b;
+		}
+		else
+		{
+			if (sel.x < 8u)
+				c = cN.g;
+			else
+				c = cN.a;
+		}
+#endif
+	}
+
+	output.c = (float4)(c); // Divide by something here?
+
+	return output;
+}
+
 #elif SHADER_MODEL <= 0x300
 
 PS_OUTPUT ps_main1(PS_INPUT input)