static inline void GenerateVertexShader(T& out, u32 components, API_TYPE api_type)
{
	// Non-uid template parameters will write to the dummy data (=> gets optimized out)
	vertex_shader_uid_data dummy_data;
	vertex_shader_uid_data* uid_data = out.template GetUidData<vertex_shader_uid_data>();
	if (uid_data == nullptr)
		uid_data = &dummy_data;

	out.SetBuffer(text);
	const bool is_writing_shadercode = (out.GetBuffer() != nullptr);

	if (is_writing_shadercode)
		text[sizeof(text) - 1] = 0x7C;  // canary

	_assert_(bpmem.genMode.numtexgens == xfmem.numTexGen.numTexGens);
	_assert_(bpmem.genMode.numcolchans == xfmem.numChan.numColorChans);

	if (DriverDetails::HasBug(DriverDetails::BUG_BROKENIVECSHIFTS))
	{
		// Add functions to do shifts on scalars and ivecs.
		// This is included in the vertex shader for lighting shader generation.
		out.Write("int ilshift(int a, int b) { return a << b; }\n"
		          "int irshift(int a, int b) { return a >> b; }\n"

		          "int2 ilshift(int2 a, int2 b) { return int2(a.x << b.x, a.y << b.y); }\n"
		          "int2 ilshift(int2 a, int b) { return int2(a.x << b, a.y << b); }\n"
		          "int2 irshift(int2 a, int2 b) { return int2(a.x >> b.x, a.y >> b.y); }\n"
		          "int2 irshift(int2 a, int b) { return int2(a.x >> b, a.y >> b); }\n"

		          "int3 ilshift(int3 a, int3 b) { return int3(a.x << b.x, a.y << b.y, a.z << b.z); }\n"
		          "int3 ilshift(int3 a, int b) { return int3(a.x << b, a.y << b, a.z << b); }\n"
		          "int3 irshift(int3 a, int3 b) { return int3(a.x >> b.x, a.y >> b.y, a.z >> b.z); }\n"
		          "int3 irshift(int3 a, int b) { return int3(a.x >> b, a.y >> b, a.z >> b); }\n"

		          "int4 ilshift(int4 a, int4 b) { return int4(a.x << b.x, a.y << b.y, a.z << b.z, a.w << b.w); }\n"
		          "int4 ilshift(int4 a, int b) { return int4(a.x << b, a.y << b, a.z << b, a.w << b); }\n"
		          "int4 irshift(int4 a, int4 b) { return int4(a.x >> b.x, a.y >> b.y, a.z >> b.z, a.w >> b.w); }\n"
		          "int4 irshift(int4 a, int b) { return int4(a.x >> b, a.y >> b, a.z >> b, a.w >> b); }\n\n");
	}

	out.Write("%s", s_lighting_struct);

	// uniforms
	if (api_type == API_OPENGL)
		out.Write("layout(std140%s) uniform VSBlock {\n", g_ActiveConfig.backend_info.bSupportsBindingLayout ? ", binding = 2" : "");
	else
		out.Write("cbuffer VSBlock {\n");
	out.Write(s_shader_uniforms);
	out.Write("};\n");

	out.Write("struct VS_OUTPUT {\n");
	GenerateVSOutputMembers<T>(out, api_type);
	out.Write("};\n");

	uid_data->numTexGens = xfmem.numTexGen.numTexGens;
	uid_data->components = components;
	uid_data->pixel_lighting = g_ActiveConfig.bEnablePixelLighting;

	if (api_type == API_OPENGL)
	{
		out.Write("in float4 rawpos; // ATTR%d,\n", SHADER_POSITION_ATTRIB);
		if (components & VB_HAS_POSMTXIDX)
			out.Write("in int posmtx; // ATTR%d,\n", SHADER_POSMTX_ATTRIB);
		if (components & VB_HAS_NRM0)
			out.Write("in float3 rawnorm0; // ATTR%d,\n", SHADER_NORM0_ATTRIB);
		if (components & VB_HAS_NRM1)
			out.Write("in float3 rawnorm1; // ATTR%d,\n", SHADER_NORM1_ATTRIB);
		if (components & VB_HAS_NRM2)
			out.Write("in float3 rawnorm2; // ATTR%d,\n", SHADER_NORM2_ATTRIB);

		if (components & VB_HAS_COL0)
			out.Write("in float4 color0; // ATTR%d,\n", SHADER_COLOR0_ATTRIB);
		if (components & VB_HAS_COL1)
			out.Write("in float4 color1; // ATTR%d,\n", SHADER_COLOR1_ATTRIB);

		for (int i = 0; i < 8; ++i)
		{
			u32 hastexmtx = (components & (VB_HAS_TEXMTXIDX0<<i));
			if ((components & (VB_HAS_UV0<<i)) || hastexmtx)
				out.Write("in float%d tex%d; // ATTR%d,\n", hastexmtx ? 3 : 2, i, SHADER_TEXTURE0_ATTRIB + i);
		}

		if (g_ActiveConfig.backend_info.bSupportsGeometryShaders)
		{
			out.Write("out VertexData {\n");
			GenerateVSOutputMembers<T>(out, api_type, g_ActiveConfig.backend_info.bSupportsBindingLayout ? "centroid" : "centroid out");
			out.Write("} vs;\n");
		}
		else
		{
			// Let's set up attributes
			for (size_t i = 0; i < 8; ++i)
			{
				if (i < xfmem.numTexGen.numTexGens)
				{
					out.Write("centroid out float3 uv%d;\n", i);
				}
			}
			out.Write("centroid out float4 clipPos;\n");
			if (g_ActiveConfig.bEnablePixelLighting)
			{
				out.Write("centroid out float3 Normal;\n");
				out.Write("centroid out float3 WorldPos;\n");
			}
			out.Write("centroid out float4 colors_0;\n");
			out.Write("centroid out float4 colors_1;\n");
		}

		out.Write("void main()\n{\n");
	}
	else // D3D
	{
		out.Write("VS_OUTPUT main(\n");

		// inputs
		if (components & VB_HAS_NRM0)
			out.Write("  float3 rawnorm0 : NORMAL0,\n");
		if (components & VB_HAS_NRM1)
			out.Write("  float3 rawnorm1 : NORMAL1,\n");
		if (components & VB_HAS_NRM2)
			out.Write("  float3 rawnorm2 : NORMAL2,\n");
		if (components & VB_HAS_COL0)
			out.Write("  float4 color0 : COLOR0,\n");
		if (components & VB_HAS_COL1)
			out.Write("  float4 color1 : COLOR1,\n");
		for (int i = 0; i < 8; ++i)
		{
			u32 hastexmtx = (components & (VB_HAS_TEXMTXIDX0<<i));
			if ((components & (VB_HAS_UV0<<i)) || hastexmtx)
				out.Write("  float%d tex%d : TEXCOORD%d,\n", hastexmtx ? 3 : 2, i, i);
		}
		if (components & VB_HAS_POSMTXIDX)
			out.Write("  int posmtx : BLENDINDICES,\n");
		out.Write("  float4 rawpos : POSITION) {\n");
	}

	out.Write("VS_OUTPUT o;\n");

	// transforms
	if (components & VB_HAS_POSMTXIDX)
	{
		if (is_writing_shadercode && (DriverDetails::HasBug(DriverDetails::BUG_NODYNUBOACCESS) && !DriverDetails::HasBug(DriverDetails::BUG_ANNIHILATEDUBOS)))
		{
			// This'll cause issues, but  it can't be helped
			out.Write("float4 pos = float4(dot(" I_TRANSFORMMATRICES"[0], rawpos), dot(" I_TRANSFORMMATRICES"[1], rawpos), dot(" I_TRANSFORMMATRICES"[2], rawpos), 1);\n");
			if (components & VB_HAS_NRMALL)
				out.Write("float3 N0 = " I_NORMALMATRICES"[0].xyz, N1 = " I_NORMALMATRICES"[1].xyz, N2 = " I_NORMALMATRICES"[2].xyz;\n");
		}
		else
		{
			out.Write("float4 pos = float4(dot(" I_TRANSFORMMATRICES"[posmtx], rawpos), dot(" I_TRANSFORMMATRICES"[posmtx+1], rawpos), dot(" I_TRANSFORMMATRICES"[posmtx+2], rawpos), 1);\n");

			if (components & VB_HAS_NRMALL)
			{
				out.Write("int normidx = posmtx >= 32 ? (posmtx-32) : posmtx;\n");
				out.Write("float3 N0 = " I_NORMALMATRICES"[normidx].xyz, N1 = " I_NORMALMATRICES"[normidx+1].xyz, N2 = " I_NORMALMATRICES"[normidx+2].xyz;\n");
			}
		}

		if (components & VB_HAS_NRM0)
			out.Write("float3 _norm0 = normalize(float3(dot(N0, rawnorm0), dot(N1, rawnorm0), dot(N2, rawnorm0)));\n");
		if (components & VB_HAS_NRM1)
			out.Write("float3 _norm1 = float3(dot(N0, rawnorm1), dot(N1, rawnorm1), dot(N2, rawnorm1));\n");
		if (components & VB_HAS_NRM2)
			out.Write("float3 _norm2 = float3(dot(N0, rawnorm2), dot(N1, rawnorm2), dot(N2, rawnorm2));\n");
	}
	else
	{
		out.Write("float4 pos = float4(dot(" I_POSNORMALMATRIX"[0], rawpos), dot(" I_POSNORMALMATRIX"[1], rawpos), dot(" I_POSNORMALMATRIX"[2], rawpos), 1.0);\n");
		if (components & VB_HAS_NRM0)
			out.Write("float3 _norm0 = normalize(float3(dot(" I_POSNORMALMATRIX"[3].xyz, rawnorm0), dot(" I_POSNORMALMATRIX"[4].xyz, rawnorm0), dot(" I_POSNORMALMATRIX"[5].xyz, rawnorm0)));\n");
		if (components & VB_HAS_NRM1)
			out.Write("float3 _norm1 = float3(dot(" I_POSNORMALMATRIX"[3].xyz, rawnorm1), dot(" I_POSNORMALMATRIX"[4].xyz, rawnorm1), dot(" I_POSNORMALMATRIX"[5].xyz, rawnorm1));\n");
		if (components & VB_HAS_NRM2)
			out.Write("float3 _norm2 = float3(dot(" I_POSNORMALMATRIX"[3].xyz, rawnorm2), dot(" I_POSNORMALMATRIX"[4].xyz, rawnorm2), dot(" I_POSNORMALMATRIX"[5].xyz, rawnorm2));\n");
	}

	if (!(components & VB_HAS_NRM0))
		out.Write("float3 _norm0 = float3(0.0, 0.0, 0.0);\n");


	out.Write("o.pos = float4(dot(" I_PROJECTION"[0], pos), dot(" I_PROJECTION"[1], pos), dot(" I_PROJECTION"[2], pos), dot(" I_PROJECTION"[3], pos));\n");

	out.Write("int4 lacc;\n"
			"float3 ldir, h, cosAttn, distAttn;\n"
			"float dist, dist2, attn;\n");

	uid_data->numColorChans = xfmem.numChan.numColorChans;
	if (xfmem.numChan.numColorChans == 0)
	{
		if (components & VB_HAS_COL0)
			out.Write("o.colors_0 = color0;\n");
		else
			out.Write("o.colors_0 = float4(1.0, 1.0, 1.0, 1.0);\n");
	}

	GenerateLightingShader<T>(out, uid_data->lighting, components, "color", "o.colors_");

	if (xfmem.numChan.numColorChans < 2)
	{
		if (components & VB_HAS_COL1)
			out.Write("o.colors_1 = color1;\n");
		else
			out.Write("o.colors_1 = o.colors_0;\n");
	}
	// special case if only pos and tex coord 0 and tex coord input is AB11
	// donko - this has caused problems in some games. removed for now.
	bool texGenSpecialCase = false;
	/*bool texGenSpecialCase =
		((g_main_cp_state.vtx_desc.Hex & 0x60600L) == g_main_cp_state.vtx_desc.Hex) && // only pos and tex coord 0
		(g_main_cp_state.vtx_desc.Tex0Coord != NOT_PRESENT) &&
		(xfmem.texcoords[0].texmtxinfo.inputform == XF_TEXINPUT_AB11);
		*/

	// transform texcoords
	out.Write("float4 coord = float4(0.0, 0.0, 1.0, 1.0);\n");
	for (unsigned int i = 0; i < xfmem.numTexGen.numTexGens; ++i)
	{
		TexMtxInfo& texinfo = xfmem.texMtxInfo[i];

		out.Write("{\n");
		out.Write("coord = float4(0.0, 0.0, 1.0, 1.0);\n");
		uid_data->texMtxInfo[i].sourcerow = xfmem.texMtxInfo[i].sourcerow;
		switch (texinfo.sourcerow)
		{
		case XF_SRCGEOM_INROW:
			// The following assert was triggered in Super Smash Bros. Project M 3.6.
			//_assert_(texinfo.inputform == XF_TEXINPUT_ABC1);
			out.Write("coord = rawpos;\n"); // pos.w is 1
			break;
		case XF_SRCNORMAL_INROW:
			if (components & VB_HAS_NRM0)
			{
				_assert_(texinfo.inputform == XF_TEXINPUT_ABC1);
				out.Write("coord = float4(rawnorm0.xyz, 1.0);\n");
			}
			break;
		case XF_SRCCOLORS_INROW:
			_assert_(texinfo.texgentype == XF_TEXGEN_COLOR_STRGBC0 || texinfo.texgentype == XF_TEXGEN_COLOR_STRGBC1);
			break;
		case XF_SRCBINORMAL_T_INROW:
			if (components & VB_HAS_NRM1)
			{
				_assert_(texinfo.inputform == XF_TEXINPUT_ABC1);
				out.Write("coord = float4(rawnorm1.xyz, 1.0);\n");
			}
			break;
		case XF_SRCBINORMAL_B_INROW:
			if (components & VB_HAS_NRM2)
			{
				_assert_(texinfo.inputform == XF_TEXINPUT_ABC1);
				out.Write("coord = float4(rawnorm2.xyz, 1.0);\n");
			}
			break;
		default:
			_assert_(texinfo.sourcerow <= XF_SRCTEX7_INROW);
			if (components & (VB_HAS_UV0 << (texinfo.sourcerow - XF_SRCTEX0_INROW)))
				out.Write("coord = float4(tex%d.x, tex%d.y, 1.0, 1.0);\n", texinfo.sourcerow - XF_SRCTEX0_INROW, texinfo.sourcerow - XF_SRCTEX0_INROW);
			break;
		}

		// first transformation
		uid_data->texMtxInfo[i].texgentype = xfmem.texMtxInfo[i].texgentype;
		switch (texinfo.texgentype)
		{
			case XF_TEXGEN_EMBOSS_MAP: // calculate tex coords into bump map

				if (components & (VB_HAS_NRM1|VB_HAS_NRM2))
				{
					// transform the light dir into tangent space
					uid_data->texMtxInfo[i].embosslightshift = xfmem.texMtxInfo[i].embosslightshift;
					uid_data->texMtxInfo[i].embosssourceshift = xfmem.texMtxInfo[i].embosssourceshift;
					out.Write("ldir = normalize(" LIGHT_POS".xyz - pos.xyz);\n", LIGHT_POS_PARAMS(texinfo.embosslightshift));
					out.Write("o.tex%d.xyz = o.tex%d.xyz + float3(dot(ldir, _norm1), dot(ldir, _norm2), 0.0);\n", i, texinfo.embosssourceshift);
				}
				else
				{
					// The following assert was triggered in House of the Dead Overkill and Star Wars Rogue Squadron 2
					//_assert_(0); // should have normals
					uid_data->texMtxInfo[i].embosssourceshift = xfmem.texMtxInfo[i].embosssourceshift;
					out.Write("o.tex%d.xyz = o.tex%d.xyz;\n", i, texinfo.embosssourceshift);
				}

				break;
			case XF_TEXGEN_COLOR_STRGBC0:
				_assert_(texinfo.sourcerow == XF_SRCCOLORS_INROW);
				out.Write("o.tex%d.xyz = float3(o.colors_0.x, o.colors_0.y, 1);\n", i);
				break;
			case XF_TEXGEN_COLOR_STRGBC1:
				_assert_(texinfo.sourcerow == XF_SRCCOLORS_INROW);
				out.Write("o.tex%d.xyz = float3(o.colors_1.x, o.colors_1.y, 1);\n", i);
				break;
			case XF_TEXGEN_REGULAR:
			default:
				uid_data->texMtxInfo_n_projection |= xfmem.texMtxInfo[i].projection << i;
				if (components & (VB_HAS_TEXMTXIDX0<<i))
				{
					out.Write("int tmp = int(tex%d.z);\n", i);
					if (texinfo.projection == XF_TEXPROJ_STQ)
						out.Write("o.tex%d.xyz = float3(dot(coord, " I_TRANSFORMMATRICES"[tmp]), dot(coord, " I_TRANSFORMMATRICES"[tmp+1]), dot(coord, " I_TRANSFORMMATRICES"[tmp+2]));\n", i);
					else
						out.Write("o.tex%d.xyz = float3(dot(coord, " I_TRANSFORMMATRICES"[tmp]), dot(coord, " I_TRANSFORMMATRICES"[tmp+1]), 1);\n", i);
				}
				else
				{
					if (texinfo.projection == XF_TEXPROJ_STQ)
						out.Write("o.tex%d.xyz = float3(dot(coord, " I_TEXMATRICES"[%d]), dot(coord, " I_TEXMATRICES"[%d]), dot(coord, " I_TEXMATRICES"[%d]));\n", i, 3*i, 3*i+1, 3*i+2);
					else
						out.Write("o.tex%d.xyz = float3(dot(coord, " I_TEXMATRICES"[%d]), dot(coord, " I_TEXMATRICES"[%d]), 1);\n", i, 3*i, 3*i+1);
				}
				break;
		}

		uid_data->dualTexTrans_enabled = xfmem.dualTexTrans.enabled;
		// CHECKME: does this only work for regular tex gen types?
		if (xfmem.dualTexTrans.enabled && texinfo.texgentype == XF_TEXGEN_REGULAR)
		{
			const PostMtxInfo& postInfo = xfmem.postMtxInfo[i];

			uid_data->postMtxInfo[i].index = xfmem.postMtxInfo[i].index;
			int postidx = postInfo.index;
			out.Write("float4 P0 = " I_POSTTRANSFORMMATRICES"[%d];\n"
				"float4 P1 = " I_POSTTRANSFORMMATRICES"[%d];\n"
				"float4 P2 = " I_POSTTRANSFORMMATRICES"[%d];\n",
				postidx & 0x3f, (postidx + 1) & 0x3f, (postidx + 2) & 0x3f);

			if (texGenSpecialCase)
			{
				// no normalization
				// q of input is 1
				// q of output is unknown

				// multiply by postmatrix
				out.Write("o.tex%d.xyz = float3(dot(P0.xy, o.tex%d.xy) + P0.z + P0.w, dot(P1.xy, o.tex%d.xy) + P1.z + P1.w, 0.0);\n", i, i, i);
			}
			else
			{
				uid_data->postMtxInfo[i].normalize = xfmem.postMtxInfo[i].normalize;
				if (postInfo.normalize)
					out.Write("o.tex%d.xyz = normalize(o.tex%d.xyz);\n", i, i);

				// multiply by postmatrix
				out.Write("o.tex%d.xyz = float3(dot(P0.xyz, o.tex%d.xyz) + P0.w, dot(P1.xyz, o.tex%d.xyz) + P1.w, dot(P2.xyz, o.tex%d.xyz) + P2.w);\n", i, i, i, i);
			}
		}

		out.Write("}\n");
	}

	// clipPos/w needs to be done in pixel shader, not here
	out.Write("o.clipPos = o.pos;\n");

	if (g_ActiveConfig.bEnablePixelLighting)
	{
		out.Write("o.Normal = _norm0;\n");
		out.Write("o.WorldPos = pos.xyz;\n");

		if (components & VB_HAS_COL0)
			out.Write("o.colors_0 = color0;\n");

		if (components & VB_HAS_COL1)
			out.Write("o.colors_1 = color1;\n");
	}

	//write the true depth value, if the game uses depth textures pixel shaders will override with the correct values
	//if not early z culling will improve speed
	if (g_ActiveConfig.backend_info.bSupportsClipControl)
	{
		out.Write("o.pos.z = -o.pos.z;\n");
	}
	else // OGL
	{
		// this results in a scale from -1..0 to -1..1 after perspective
		// divide
		out.Write("o.pos.z = o.pos.z * -2.0 - o.pos.w;\n");

		// the next steps of the OGL pipeline are:
		// (x_c,y_c,z_c,w_c) = o.pos  //switch to OGL spec terminology
		// clipping to -w_c <= (x_c,y_c,z_c) <= w_c
		// (x_d,y_d,z_d) = (x_c,y_c,z_c)/w_c//perspective divide
		// z_w = (f-n)/2*z_d + (n+f)/2
		// z_w now contains the value to go to the 0..1 depth buffer

		//trying to get the correct semantic while not using glDepthRange
		//seems to get rather complicated
	}

	// The console GPU places the pixel center at 7/12 in screen space unless
	// antialiasing is enabled, while D3D and OpenGL place it at 0.5. This results
	// in some primitives being placed one pixel too far to the bottom-right,
	// which in turn can be critical if it happens for clear quads.
	// Hence, we compensate for this pixel center difference so that primitives
	// get rasterized correctly.
	out.Write("o.pos.xy = o.pos.xy - o.pos.w * " I_PIXELCENTERCORRECTION".xy;\n");

	if (api_type == API_OPENGL)
	{
		if (g_ActiveConfig.backend_info.bSupportsGeometryShaders)
		{
			AssignVSOutputMembers(out, "vs", "o");
		}
		else
		{
			// TODO: Pass interface blocks between shader stages even if geometry shaders
			// are not supported, however that will require at least OpenGL 3.2 support.
			for (unsigned int i = 0; i < xfmem.numTexGen.numTexGens; ++i)
				out.Write("uv%d.xyz = o.tex%d;\n", i, i);
			out.Write("clipPos = o.clipPos;\n");
			if (g_ActiveConfig.bEnablePixelLighting)
			{
				out.Write("Normal = o.Normal;\n");
				out.Write("WorldPos = o.WorldPos;\n");
			}
			out.Write("colors_0 = o.colors_0;\n");
			out.Write("colors_1 = o.colors_1;\n");
		}

		out.Write("gl_Position = o.pos;\n");
	}
	else // D3D
	{
		out.Write("return o;\n");
	}
	out.Write("}\n");

	if (is_writing_shadercode)
	{
		if (text[sizeof(text) - 1] != 0x7C)
			PanicAlert("VertexShader generator - buffer too small, canary has been eaten!");
	}
}
Example #2
0
ShaderCode GenerateGeometryShaderCode(APIType ApiType, const geometry_shader_uid_data* uid_data)
{
  ShaderCode out;
  // Non-uid template parameters will write to the dummy data (=> gets optimized out)

  const unsigned int vertex_in = uid_data->primitive_type + 1;
  unsigned int vertex_out = uid_data->primitive_type == PRIMITIVE_TRIANGLES ? 3 : 4;

  if (uid_data->wireframe)
    vertex_out++;

  if (ApiType == APIType::OpenGL || ApiType == APIType::Vulkan)
  {
    // Insert layout parameters
    if (g_ActiveConfig.backend_info.bSupportsGSInstancing)
    {
      out.Write("layout(%s, invocations = %d) in;\n", primitives_ogl[uid_data->primitive_type],
                uid_data->stereo ? 2 : 1);
      out.Write("layout(%s_strip, max_vertices = %d) out;\n",
                uid_data->wireframe ? "line" : "triangle", vertex_out);
    }
    else
    {
      out.Write("layout(%s) in;\n", primitives_ogl[uid_data->primitive_type]);
      out.Write("layout(%s_strip, max_vertices = %d) out;\n",
                uid_data->wireframe ? "line" : "triangle",
                uid_data->stereo ? vertex_out * 2 : vertex_out);
    }
  }

  out.Write("%s", s_lighting_struct);

  // uniforms
  if (ApiType == APIType::OpenGL || ApiType == APIType::Vulkan)
    out.Write("UBO_BINDING(std140, 3) uniform GSBlock {\n");
  else
    out.Write("cbuffer GSBlock {\n");

  out.Write("\tfloat4 " I_STEREOPARAMS ";\n"
            "\tfloat4 " I_LINEPTPARAMS ";\n"
            "\tint4 " I_TEXOFFSET ";\n"
            "};\n");

  out.Write("struct VS_OUTPUT {\n");
  GenerateVSOutputMembers<ShaderCode>(out, ApiType, uid_data->numTexGens, uid_data->pixel_lighting,
                                      "");
  out.Write("};\n");

  if (ApiType == APIType::OpenGL || ApiType == APIType::Vulkan)
  {
    if (g_ActiveConfig.backend_info.bSupportsGSInstancing)
      out.Write("#define InstanceID gl_InvocationID\n");

    out.Write("VARYING_LOCATION(0) in VertexData {\n");
    GenerateVSOutputMembers<ShaderCode>(
        out, ApiType, uid_data->numTexGens, uid_data->pixel_lighting,
        GetInterpolationQualifier(uid_data->msaa, uid_data->ssaa, true, true));
    out.Write("} vs[%d];\n", vertex_in);

    out.Write("VARYING_LOCATION(0) out VertexData {\n");
    GenerateVSOutputMembers<ShaderCode>(
        out, ApiType, uid_data->numTexGens, uid_data->pixel_lighting,
        GetInterpolationQualifier(uid_data->msaa, uid_data->ssaa, false, true));

    if (uid_data->stereo)
      out.Write("\tflat int layer;\n");

    out.Write("} ps;\n");

    out.Write("void main()\n{\n");
  }
  else  // D3D
  {
    out.Write("struct VertexData {\n");
    out.Write("\tVS_OUTPUT o;\n");

    if (uid_data->stereo)
      out.Write("\tuint layer : SV_RenderTargetArrayIndex;\n");

    out.Write("};\n");

    if (g_ActiveConfig.backend_info.bSupportsGSInstancing)
    {
      out.Write("[maxvertexcount(%d)]\n[instance(%d)]\n", vertex_out, uid_data->stereo ? 2 : 1);
      out.Write("void main(%s VS_OUTPUT o[%d], inout %sStream<VertexData> output, in uint "
                "InstanceID : SV_GSInstanceID)\n{\n",
                primitives_d3d[uid_data->primitive_type], vertex_in,
                uid_data->wireframe ? "Line" : "Triangle");
    }
    else
    {
      out.Write("[maxvertexcount(%d)]\n", uid_data->stereo ? vertex_out * 2 : vertex_out);
      out.Write("void main(%s VS_OUTPUT o[%d], inout %sStream<VertexData> output)\n{\n",
                primitives_d3d[uid_data->primitive_type], vertex_in,
                uid_data->wireframe ? "Line" : "Triangle");
    }

    out.Write("\tVertexData ps;\n");
  }

  if (uid_data->primitive_type == PRIMITIVE_LINES)
  {
    if (ApiType == APIType::OpenGL || ApiType == APIType::Vulkan)
    {
      out.Write("\tVS_OUTPUT start, end;\n");
      AssignVSOutputMembers(out, "start", "vs[0]", uid_data->numTexGens, uid_data->pixel_lighting);
      AssignVSOutputMembers(out, "end", "vs[1]", uid_data->numTexGens, uid_data->pixel_lighting);
    }
    else
    {
      out.Write("\tVS_OUTPUT start = o[0];\n");
      out.Write("\tVS_OUTPUT end = o[1];\n");
    }

    // GameCube/Wii's line drawing algorithm is a little quirky. It does not
    // use the correct line caps. Instead, the line caps are vertical or
    // horizontal depending the slope of the line.
    out.Write("\tfloat2 offset;\n"
              "\tfloat2 to = abs(end.pos.xy / end.pos.w - start.pos.xy / start.pos.w);\n"
              // FIXME: What does real hardware do when line is at a 45-degree angle?
              // FIXME: Lines aren't drawn at the correct width. See Twilight Princess map.
              "\tif (" I_LINEPTPARAMS ".y * to.y > " I_LINEPTPARAMS ".x * to.x) {\n"
              // Line is more tall. Extend geometry left and right.
              // Lerp LineWidth/2 from [0..VpWidth] to [-1..1]
              "\t\toffset = float2(" I_LINEPTPARAMS ".z / " I_LINEPTPARAMS ".x, 0);\n"
              "\t} else {\n"
              // Line is more wide. Extend geometry up and down.
              // Lerp LineWidth/2 from [0..VpHeight] to [1..-1]
              "\t\toffset = float2(0, -" I_LINEPTPARAMS ".z / " I_LINEPTPARAMS ".y);\n"
              "\t}\n");
  }
  else if (uid_data->primitive_type == PRIMITIVE_POINTS)
  {
    if (ApiType == APIType::OpenGL || ApiType == APIType::Vulkan)
    {
      out.Write("\tVS_OUTPUT center;\n");
      AssignVSOutputMembers(out, "center", "vs[0]", uid_data->numTexGens, uid_data->pixel_lighting);
    }
    else
    {
      out.Write("\tVS_OUTPUT center = o[0];\n");
    }

    // Offset from center to upper right vertex
    // Lerp PointSize/2 from [0,0..VpWidth,VpHeight] to [-1,1..1,-1]
    out.Write("\tfloat2 offset = float2(" I_LINEPTPARAMS ".w / " I_LINEPTPARAMS
              ".x, -" I_LINEPTPARAMS ".w / " I_LINEPTPARAMS ".y) * center.pos.w;\n");
  }

  if (uid_data->stereo)
  {
    // If the GPU supports invocation we don't need a for loop and can simply use the
    // invocation identifier to determine which layer we're rendering.
    if (g_ActiveConfig.backend_info.bSupportsGSInstancing)
      out.Write("\tint eye = InstanceID;\n");
    else
      out.Write("\tfor (int eye = 0; eye < 2; ++eye) {\n");
  }

  if (uid_data->wireframe)
    out.Write("\tVS_OUTPUT first;\n");

  out.Write("\tfor (int i = 0; i < %d; ++i) {\n", vertex_in);

  if (ApiType == APIType::OpenGL || ApiType == APIType::Vulkan)
  {
    out.Write("\tVS_OUTPUT f;\n");
    AssignVSOutputMembers(out, "f", "vs[i]", uid_data->numTexGens, uid_data->pixel_lighting);

    if (g_ActiveConfig.backend_info.bSupportsDepthClamp &&
        DriverDetails::HasBug(DriverDetails::BUG_BROKENCLIPDISTANCE))
    {
      // On certain GPUs we have to consume the clip distance from the vertex shader
      // or else the other vertex shader outputs will get corrupted.
      out.Write("\tf.clipDist0 = gl_in[i].gl_ClipDistance[0];\n");
      out.Write("\tf.clipDist1 = gl_in[i].gl_ClipDistance[1];\n");
    }
  }
  else
  {
    out.Write("\tVS_OUTPUT f = o[i];\n");
  }

  if (uid_data->stereo)
  {
    // Select the output layer
    out.Write("\tps.layer = eye;\n");
    if (ApiType == APIType::OpenGL || ApiType == APIType::Vulkan)
      out.Write("\tgl_Layer = eye;\n");

    // For stereoscopy add a small horizontal offset in Normalized Device Coordinates proportional
    // to the depth of the vertex. We retrieve the depth value from the w-component of the projected
    // vertex which contains the negated z-component of the original vertex.
    // For negative parallax (out-of-screen effects) we subtract a convergence value from
    // the depth value. This results in objects at a distance smaller than the convergence
    // distance to seemingly appear in front of the screen.
    // This formula is based on page 13 of the "Nvidia 3D Vision Automatic, Best Practices Guide"
    out.Write("\tfloat hoffset = (eye == 0) ? " I_STEREOPARAMS ".x : " I_STEREOPARAMS ".y;\n");
    out.Write("\tf.pos.x += hoffset * (f.pos.w - " I_STEREOPARAMS ".z);\n");
  }

  if (uid_data->primitive_type == PRIMITIVE_LINES)
  {
    out.Write("\tVS_OUTPUT l = f;\n"
              "\tVS_OUTPUT r = f;\n");

    out.Write("\tl.pos.xy -= offset * l.pos.w;\n"
              "\tr.pos.xy += offset * r.pos.w;\n");

    out.Write("\tif (" I_TEXOFFSET "[2] != 0) {\n");
    out.Write("\tfloat texOffset = 1.0 / float(" I_TEXOFFSET "[2]);\n");

    for (unsigned int i = 0; i < uid_data->numTexGens; ++i)
    {
      out.Write("\tif (((" I_TEXOFFSET "[0] >> %d) & 0x1) != 0)\n", i);
      out.Write("\t\tr.tex%d.x += texOffset;\n", i);
    }
    out.Write("\t}\n");

    EmitVertex(out, uid_data, "l", ApiType, true);
    EmitVertex(out, uid_data, "r", ApiType);
  }
  else if (uid_data->primitive_type == PRIMITIVE_POINTS)
  {
    out.Write("\tVS_OUTPUT ll = f;\n"
              "\tVS_OUTPUT lr = f;\n"
              "\tVS_OUTPUT ul = f;\n"
              "\tVS_OUTPUT ur = f;\n");

    out.Write("\tll.pos.xy += float2(-1,-1) * offset;\n"
              "\tlr.pos.xy += float2(1,-1) * offset;\n"
              "\tul.pos.xy += float2(-1,1) * offset;\n"
              "\tur.pos.xy += offset;\n");

    out.Write("\tif (" I_TEXOFFSET "[3] != 0) {\n");
    out.Write("\tfloat2 texOffset = float2(1.0 / float(" I_TEXOFFSET
              "[3]), 1.0 / float(" I_TEXOFFSET "[3]));\n");

    for (unsigned int i = 0; i < uid_data->numTexGens; ++i)
    {
      out.Write("\tif (((" I_TEXOFFSET "[1] >> %d) & 0x1) != 0) {\n", i);
      out.Write("\t\tll.tex%d.xy += float2(0,1) * texOffset;\n", i);
      out.Write("\t\tlr.tex%d.xy += texOffset;\n", i);
      out.Write("\t\tur.tex%d.xy += float2(1,0) * texOffset;\n", i);
      out.Write("\t}\n");
    }
    out.Write("\t}\n");

    EmitVertex(out, uid_data, "ll", ApiType, true);
    EmitVertex(out, uid_data, "lr", ApiType);
    EmitVertex(out, uid_data, "ul", ApiType);
    EmitVertex(out, uid_data, "ur", ApiType);
  }
  else
  {
    EmitVertex(out, uid_data, "f", ApiType, true);
  }

  out.Write("\t}\n");

  EndPrimitive(out, uid_data, ApiType);

  if (uid_data->stereo && !g_ActiveConfig.backend_info.bSupportsGSInstancing)
    out.Write("\t}\n");

  out.Write("}\n");

  return out;
}
Example #3
0
static inline void GenerateGeometryShader(T& out, u32 primitive_type, API_TYPE ApiType)
{
	// Non-uid template parameters will write to the dummy data (=> gets optimized out)
	geometry_shader_uid_data dummy_data;
	geometry_shader_uid_data* uid_data = out.template GetUidData<geometry_shader_uid_data>();
	if (uid_data == nullptr)
		uid_data = &dummy_data;

	out.SetBuffer(text);
	const bool is_writing_shadercode = (out.GetBuffer() != nullptr);

	if (is_writing_shadercode)
		text[sizeof(text) - 1] = 0x7C;  // canary

	uid_data->primitive_type = primitive_type;
	const unsigned int vertex_in = primitive_type + 1;
	unsigned int vertex_out = primitive_type == PRIMITIVE_TRIANGLES ? 3 : 4;

	uid_data->wireframe = g_ActiveConfig.bWireFrame;
	if (g_ActiveConfig.bWireFrame)
		vertex_out++;

	uid_data->stereo = g_ActiveConfig.iStereoMode > 0;
	if (ApiType == API_OPENGL)
	{
		// Insert layout parameters
		if (g_ActiveConfig.backend_info.bSupportsGSInstancing)
		{
			out.Write("layout(%s, invocations = %d) in;\n", primitives_ogl[primitive_type], g_ActiveConfig.iStereoMode > 0 ? 2 : 1);
			out.Write("layout(%s_strip, max_vertices = %d) out;\n", g_ActiveConfig.bWireFrame ? "line" : "triangle", vertex_out);
		}
		else
		{
			out.Write("layout(%s) in;\n", primitives_ogl[primitive_type]);
			out.Write("layout(%s_strip, max_vertices = %d) out;\n", g_ActiveConfig.bWireFrame ? "line" : "triangle", g_ActiveConfig.iStereoMode > 0 ? vertex_out * 2 : vertex_out);
		}
	}

	out.Write("%s", s_lighting_struct);

	// uniforms
	if (ApiType == API_OPENGL)
		out.Write("layout(std140%s) uniform GSBlock {\n", g_ActiveConfig.backend_info.bSupportsBindingLayout ? ", binding = 3" : "");
	else
		out.Write("cbuffer GSBlock {\n");
	out.Write(
		"\tfloat4 " I_STEREOPARAMS";\n"
		"\tfloat4 " I_LINEPTPARAMS";\n"
		"\tint4 " I_TEXOFFSET";\n"
		"};\n");

	uid_data->numTexGens = xfmem.numTexGen.numTexGens;
	uid_data->pixel_lighting = g_ActiveConfig.bEnablePixelLighting;

	out.Write("struct VS_OUTPUT {\n");
	GenerateVSOutputMembers<T>(out, ApiType);
	out.Write("};\n");

	if (ApiType == API_OPENGL)
	{
		if (g_ActiveConfig.backend_info.bSupportsGSInstancing)
			out.Write("#define InstanceID gl_InvocationID\n");

		out.Write("in VertexData {\n");
		GenerateVSOutputMembers<T>(out, ApiType, GetInterpolationQualifier(ApiType, true, true));
		out.Write("} vs[%d];\n", vertex_in);

		out.Write("out VertexData {\n");
		GenerateVSOutputMembers<T>(out, ApiType, GetInterpolationQualifier(ApiType, false, true));

		if (g_ActiveConfig.iStereoMode > 0)
			out.Write("\tflat int layer;\n");

		out.Write("} ps;\n");

		out.Write("void main()\n{\n");
	}
	else // D3D
	{
		out.Write("struct VertexData {\n");
		out.Write("\tVS_OUTPUT o;\n");

		if (g_ActiveConfig.iStereoMode > 0)
			out.Write("\tuint layer : SV_RenderTargetArrayIndex;\n");

		out.Write("};\n");

		if (g_ActiveConfig.backend_info.bSupportsGSInstancing)
		{
			out.Write("[maxvertexcount(%d)]\n[instance(%d)]\n", vertex_out, g_ActiveConfig.iStereoMode > 0 ? 2 : 1);
			out.Write("void main(%s VS_OUTPUT o[%d], inout %sStream<VertexData> output, in uint InstanceID : SV_GSInstanceID)\n{\n", primitives_d3d[primitive_type], vertex_in, g_ActiveConfig.bWireFrame ? "Line" : "Triangle");
		}
		else
		{
			out.Write("[maxvertexcount(%d)]\n", g_ActiveConfig.iStereoMode > 0 ? vertex_out * 2 : vertex_out);
			out.Write("void main(%s VS_OUTPUT o[%d], inout %sStream<VertexData> output)\n{\n", primitives_d3d[primitive_type], vertex_in, g_ActiveConfig.bWireFrame ? "Line" : "Triangle");
		}

		out.Write("\tVertexData ps;\n");
	}

	if (primitive_type == PRIMITIVE_LINES)
	{
		if (ApiType == API_OPENGL)
		{
			out.Write("\tVS_OUTPUT start, end;\n");
			AssignVSOutputMembers(out, "start", "vs[0]");
			AssignVSOutputMembers(out, "end", "vs[1]");
		}
		else
		{
			out.Write("\tVS_OUTPUT start = o[0];\n");
			out.Write("\tVS_OUTPUT end = o[1];\n");
		}

		// GameCube/Wii's line drawing algorithm is a little quirky. It does not
		// use the correct line caps. Instead, the line caps are vertical or
		// horizontal depending the slope of the line.
		out.Write(
			"\tfloat2 offset;\n"
			"\tfloat2 to = abs(end.pos.xy / end.pos.w - start.pos.xy / start.pos.w);\n"
			// FIXME: What does real hardware do when line is at a 45-degree angle?
			// FIXME: Lines aren't drawn at the correct width. See Twilight Princess map.
			"\tif (" I_LINEPTPARAMS".y * to.y > " I_LINEPTPARAMS".x * to.x) {\n"
			// Line is more tall. Extend geometry left and right.
			// Lerp LineWidth/2 from [0..VpWidth] to [-1..1]
			"\t\toffset = float2(" I_LINEPTPARAMS".z / " I_LINEPTPARAMS".x, 0);\n"
			"\t} else {\n"
			// Line is more wide. Extend geometry up and down.
			// Lerp LineWidth/2 from [0..VpHeight] to [1..-1]
			"\t\toffset = float2(0, -" I_LINEPTPARAMS".z / " I_LINEPTPARAMS".y);\n"
			"\t}\n");
	}
	else if (primitive_type == PRIMITIVE_POINTS)
	{
		if (ApiType == API_OPENGL)
		{
			out.Write("\tVS_OUTPUT center;\n");
			AssignVSOutputMembers(out, "center", "vs[0]");
		}
		else
		{
			out.Write("\tVS_OUTPUT center = o[0];\n");
		}

		// Offset from center to upper right vertex
		// Lerp PointSize/2 from [0,0..VpWidth,VpHeight] to [-1,1..1,-1]
		out.Write("\tfloat2 offset = float2(" I_LINEPTPARAMS".w / " I_LINEPTPARAMS".x, -" I_LINEPTPARAMS".w / " I_LINEPTPARAMS".y) * center.pos.w;\n");
	}

	if (g_ActiveConfig.iStereoMode > 0)
	{
		// If the GPU supports invocation we don't need a for loop and can simply use the
		// invocation identifier to determine which layer we're rendering.
		if (g_ActiveConfig.backend_info.bSupportsGSInstancing)
			out.Write("\tint eye = InstanceID;\n");
		else
			out.Write("\tfor (int eye = 0; eye < 2; ++eye) {\n");
	}

	if (g_ActiveConfig.bWireFrame)
		out.Write("\tVS_OUTPUT first;\n");

	out.Write("\tfor (int i = 0; i < %d; ++i) {\n", vertex_in);

	if (ApiType == API_OPENGL)
	{
		out.Write("\tVS_OUTPUT f;\n");
		AssignVSOutputMembers(out, "f", "vs[i]");
	}
	else
	{
		out.Write("\tVS_OUTPUT f = o[i];\n");
	}

	if (g_ActiveConfig.iStereoMode > 0)
	{
		// Select the output layer
		out.Write("\tps.layer = eye;\n");
		if (ApiType == API_OPENGL)
			out.Write("\tgl_Layer = eye;\n");

		// For stereoscopy add a small horizontal offset in Normalized Device Coordinates proportional
		// to the depth of the vertex. We retrieve the depth value from the w-component of the projected
		// vertex which contains the negated z-component of the original vertex.
		// For negative parallax (out-of-screen effects) we subtract a convergence value from
		// the depth value. This results in objects at a distance smaller than the convergence
		// distance to seemingly appear in front of the screen.
		// This formula is based on page 13 of the "Nvidia 3D Vision Automatic, Best Practices Guide"
		out.Write("\tf.pos.x += " I_STEREOPARAMS"[eye] * (f.pos.w - " I_STEREOPARAMS"[2]);\n");
	}

	if (primitive_type == PRIMITIVE_LINES)
	{
		out.Write("\tVS_OUTPUT l = f;\n"
		          "\tVS_OUTPUT r = f;\n");

		out.Write("\tl.pos.xy -= offset * l.pos.w;\n"
		          "\tr.pos.xy += offset * r.pos.w;\n");

		out.Write("\tif (" I_TEXOFFSET"[2] != 0) {\n");
		out.Write("\tfloat texOffset = 1.0 / float(" I_TEXOFFSET"[2]);\n");

		for (unsigned int i = 0; i < xfmem.numTexGen.numTexGens; ++i)
		{
			out.Write("\tif (((" I_TEXOFFSET"[0] >> %d) & 0x1) != 0)\n", i);
			out.Write("\t\tr.tex%d.x += texOffset;\n", i);
		}
		out.Write("\t}\n");

		EmitVertex<T>(out, "l", ApiType, true);
		EmitVertex<T>(out, "r", ApiType);
	}
	else if (primitive_type == PRIMITIVE_POINTS)
	{
		out.Write("\tVS_OUTPUT ll = f;\n"
		          "\tVS_OUTPUT lr = f;\n"
		          "\tVS_OUTPUT ul = f;\n"
		          "\tVS_OUTPUT ur = f;\n");

		out.Write("\tll.pos.xy += float2(-1,-1) * offset;\n"
		          "\tlr.pos.xy += float2(1,-1) * offset;\n"
		          "\tul.pos.xy += float2(-1,1) * offset;\n"
		          "\tur.pos.xy += offset;\n");

		out.Write("\tif (" I_TEXOFFSET"[3] != 0) {\n");
		out.Write("\tfloat2 texOffset = float2(1.0 / float(" I_TEXOFFSET"[3]), 1.0 / float(" I_TEXOFFSET"[3]));\n");

		for (unsigned int i = 0; i < xfmem.numTexGen.numTexGens; ++i)
		{
			out.Write("\tif (((" I_TEXOFFSET"[1] >> %d) & 0x1) != 0) {\n", i);
			out.Write("\t\tll.tex%d.xy += float2(0,1) * texOffset;\n", i);
			out.Write("\t\tlr.tex%d.xy += texOffset;\n", i);
			out.Write("\t\tur.tex%d.xy += float2(1,0) * texOffset;\n", i);
			out.Write("\t}\n");
		}
		out.Write("\t}\n");

		EmitVertex<T>(out, "ll", ApiType, true);
		EmitVertex<T>(out, "lr", ApiType);
		EmitVertex<T>(out, "ul", ApiType);
		EmitVertex<T>(out, "ur", ApiType);
	}
	else
	{
		EmitVertex<T>(out, "f", ApiType, true);
	}

	out.Write("\t}\n");

	EndPrimitive<T>(out, ApiType);

	if (g_ActiveConfig.iStereoMode > 0 && !g_ActiveConfig.backend_info.bSupportsGSInstancing)
		out.Write("\t}\n");

	out.Write("}\n");

	if (is_writing_shadercode)
	{
		if (text[sizeof(text) - 1] != 0x7C)
			PanicAlert("GeometryShader generator - buffer too small, canary has been eaten!");
	}
}
Example #4
0
static T GenerateVertexShader(API_TYPE api_type)
{
  T out;
  const u32 components = VertexLoaderManager::g_current_components;
  // Non-uid template parameters will write to the dummy data (=> gets optimized out)
  vertex_shader_uid_data dummy_data;
  vertex_shader_uid_data* uid_data = out.template GetUidData<vertex_shader_uid_data>();
  if (uid_data != nullptr)
    memset(uid_data, 0, sizeof(*uid_data));
  else
    uid_data = &dummy_data;

  _assert_(bpmem.genMode.numtexgens == xfmem.numTexGen.numTexGens);
  _assert_(bpmem.genMode.numcolchans == xfmem.numChan.numColorChans);

  out.Write("%s", s_lighting_struct);

  // uniforms
  if (api_type == API_OPENGL)
    out.Write("layout(std140%s) uniform VSBlock {\n",
              g_ActiveConfig.backend_info.bSupportsBindingLayout ? ", binding = 2" : "");
  else
    out.Write("cbuffer VSBlock {\n");
  out.Write(s_shader_uniforms);
  out.Write("};\n");

  out.Write("struct VS_OUTPUT {\n");
  GenerateVSOutputMembers<T>(out, api_type, "");
  out.Write("};\n");

  uid_data->numTexGens = xfmem.numTexGen.numTexGens;
  uid_data->components = components;
  uid_data->pixel_lighting = g_ActiveConfig.bEnablePixelLighting;

  if (api_type == API_OPENGL)
  {
    out.Write("in float4 rawpos; // ATTR%d,\n", SHADER_POSITION_ATTRIB);
    if (components & VB_HAS_POSMTXIDX)
      out.Write("in int posmtx; // ATTR%d,\n", SHADER_POSMTX_ATTRIB);
    if (components & VB_HAS_NRM0)
      out.Write("in float3 rawnorm0; // ATTR%d,\n", SHADER_NORM0_ATTRIB);
    if (components & VB_HAS_NRM1)
      out.Write("in float3 rawnorm1; // ATTR%d,\n", SHADER_NORM1_ATTRIB);
    if (components & VB_HAS_NRM2)
      out.Write("in float3 rawnorm2; // ATTR%d,\n", SHADER_NORM2_ATTRIB);

    if (components & VB_HAS_COL0)
      out.Write("in float4 color0; // ATTR%d,\n", SHADER_COLOR0_ATTRIB);
    if (components & VB_HAS_COL1)
      out.Write("in float4 color1; // ATTR%d,\n", SHADER_COLOR1_ATTRIB);

    for (int i = 0; i < 8; ++i)
    {
      u32 hastexmtx = (components & (VB_HAS_TEXMTXIDX0 << i));
      if ((components & (VB_HAS_UV0 << i)) || hastexmtx)
        out.Write("in float%d tex%d; // ATTR%d,\n", hastexmtx ? 3 : 2, i,
                  SHADER_TEXTURE0_ATTRIB + i);
    }

    if (g_ActiveConfig.backend_info.bSupportsGeometryShaders)
    {
      out.Write("out VertexData {\n");
      GenerateVSOutputMembers<T>(out, api_type, GetInterpolationQualifier(true, false));
      out.Write("} vs;\n");
    }
    else
    {
      // Let's set up attributes
      for (u32 i = 0; i < 8; ++i)
      {
        if (i < xfmem.numTexGen.numTexGens)
        {
          out.Write("%s out float3 uv%u;\n", GetInterpolationQualifier(), i);
        }
      }
      out.Write("%s out float4 clipPos;\n", GetInterpolationQualifier());
      if (g_ActiveConfig.bEnablePixelLighting)
      {
        out.Write("%s out float3 Normal;\n", GetInterpolationQualifier());
        out.Write("%s out float3 WorldPos;\n", GetInterpolationQualifier());
      }
      out.Write("%s out float4 colors_0;\n", GetInterpolationQualifier());
      out.Write("%s out float4 colors_1;\n", GetInterpolationQualifier());
    }

    out.Write("void main()\n{\n");
  }
  else  // D3D
  {
    out.Write("VS_OUTPUT main(\n");

    // inputs
    if (components & VB_HAS_NRM0)
      out.Write("  float3 rawnorm0 : NORMAL0,\n");
    if (components & VB_HAS_NRM1)
      out.Write("  float3 rawnorm1 : NORMAL1,\n");
    if (components & VB_HAS_NRM2)
      out.Write("  float3 rawnorm2 : NORMAL2,\n");
    if (components & VB_HAS_COL0)
      out.Write("  float4 color0 : COLOR0,\n");
    if (components & VB_HAS_COL1)
      out.Write("  float4 color1 : COLOR1,\n");
    for (int i = 0; i < 8; ++i)
    {
      u32 hastexmtx = (components & (VB_HAS_TEXMTXIDX0 << i));
      if ((components & (VB_HAS_UV0 << i)) || hastexmtx)
        out.Write("  float%d tex%d : TEXCOORD%d,\n", hastexmtx ? 3 : 2, i, i);
    }
    if (components & VB_HAS_POSMTXIDX)
      out.Write("  int posmtx : BLENDINDICES,\n");
    out.Write("  float4 rawpos : POSITION) {\n");
  }

  out.Write("VS_OUTPUT o;\n");

  // transforms
  if (components & VB_HAS_POSMTXIDX)
  {
    out.Write("float4 pos = float4(dot(" I_TRANSFORMMATRICES
              "[posmtx], rawpos), dot(" I_TRANSFORMMATRICES
              "[posmtx+1], rawpos), dot(" I_TRANSFORMMATRICES "[posmtx+2], rawpos), 1);\n");

    if (components & VB_HAS_NRMALL)
    {
      out.Write("int normidx = posmtx & 31;\n");
      out.Write("float3 N0 = " I_NORMALMATRICES "[normidx].xyz, N1 = " I_NORMALMATRICES
                "[normidx+1].xyz, N2 = " I_NORMALMATRICES "[normidx+2].xyz;\n");
    }

    if (components & VB_HAS_NRM0)
      out.Write("float3 _norm0 = normalize(float3(dot(N0, rawnorm0), dot(N1, rawnorm0), dot(N2, "
                "rawnorm0)));\n");
    if (components & VB_HAS_NRM1)
      out.Write(
          "float3 _norm1 = float3(dot(N0, rawnorm1), dot(N1, rawnorm1), dot(N2, rawnorm1));\n");
    if (components & VB_HAS_NRM2)
      out.Write(
          "float3 _norm2 = float3(dot(N0, rawnorm2), dot(N1, rawnorm2), dot(N2, rawnorm2));\n");
  }
  else
  {
    out.Write("float4 pos = float4(dot(" I_POSNORMALMATRIX "[0], rawpos), dot(" I_POSNORMALMATRIX
              "[1], rawpos), dot(" I_POSNORMALMATRIX "[2], rawpos), 1.0);\n");
    if (components & VB_HAS_NRM0)
      out.Write("float3 _norm0 = normalize(float3(dot(" I_POSNORMALMATRIX
                "[3].xyz, rawnorm0), dot(" I_POSNORMALMATRIX
                "[4].xyz, rawnorm0), dot(" I_POSNORMALMATRIX "[5].xyz, rawnorm0)));\n");
    if (components & VB_HAS_NRM1)
      out.Write("float3 _norm1 = float3(dot(" I_POSNORMALMATRIX
                "[3].xyz, rawnorm1), dot(" I_POSNORMALMATRIX
                "[4].xyz, rawnorm1), dot(" I_POSNORMALMATRIX "[5].xyz, rawnorm1));\n");
    if (components & VB_HAS_NRM2)
      out.Write("float3 _norm2 = float3(dot(" I_POSNORMALMATRIX
                "[3].xyz, rawnorm2), dot(" I_POSNORMALMATRIX
                "[4].xyz, rawnorm2), dot(" I_POSNORMALMATRIX "[5].xyz, rawnorm2));\n");
  }

  if (!(components & VB_HAS_NRM0))
    out.Write("float3 _norm0 = float3(0.0, 0.0, 0.0);\n");

  out.Write("o.pos = float4(dot(" I_PROJECTION "[0], pos), dot(" I_PROJECTION
            "[1], pos), dot(" I_PROJECTION "[2], pos), dot(" I_PROJECTION "[3], pos));\n");

  out.Write("int4 lacc;\n"
            "float3 ldir, h, cosAttn, distAttn;\n"
            "float dist, dist2, attn;\n");

  uid_data->numColorChans = xfmem.numChan.numColorChans;
  if (xfmem.numChan.numColorChans == 0)
  {
    if (components & VB_HAS_COL0)
      out.Write("o.colors_0 = color0;\n");
    else
      out.Write("o.colors_0 = float4(1.0, 1.0, 1.0, 1.0);\n");
  }

  GenerateLightingShader<T>(out, uid_data->lighting, components, "color", "o.colors_");

  if (xfmem.numChan.numColorChans < 2)
  {
    if (components & VB_HAS_COL1)
      out.Write("o.colors_1 = color1;\n");
    else
      out.Write("o.colors_1 = o.colors_0;\n");
  }

  // transform texcoords
  out.Write("float4 coord = float4(0.0, 0.0, 1.0, 1.0);\n");
  for (unsigned int i = 0; i < xfmem.numTexGen.numTexGens; ++i)
  {
    TexMtxInfo& texinfo = xfmem.texMtxInfo[i];

    out.Write("{\n");
    out.Write("coord = float4(0.0, 0.0, 1.0, 1.0);\n");
    uid_data->texMtxInfo[i].sourcerow = xfmem.texMtxInfo[i].sourcerow;
    switch (texinfo.sourcerow)
    {
    case XF_SRCGEOM_INROW:
      out.Write("coord.xyz = rawpos.xyz;\n");
      break;
    case XF_SRCNORMAL_INROW:
      if (components & VB_HAS_NRM0)
      {
        out.Write("coord.xyz = rawnorm0.xyz;\n");
      }
      break;
    case XF_SRCCOLORS_INROW:
      _assert_(texinfo.texgentype == XF_TEXGEN_COLOR_STRGBC0 ||
               texinfo.texgentype == XF_TEXGEN_COLOR_STRGBC1);
      break;
    case XF_SRCBINORMAL_T_INROW:
      if (components & VB_HAS_NRM1)
      {
        out.Write("coord.xyz = rawnorm1.xyz;\n");
      }
      break;
    case XF_SRCBINORMAL_B_INROW:
      if (components & VB_HAS_NRM2)
      {
        out.Write("coord.xyz = rawnorm2.xyz;\n");
      }
      break;
    default:
      _assert_(texinfo.sourcerow <= XF_SRCTEX7_INROW);
      if (components & (VB_HAS_UV0 << (texinfo.sourcerow - XF_SRCTEX0_INROW)))
        out.Write("coord = float4(tex%d.x, tex%d.y, 1.0, 1.0);\n",
                  texinfo.sourcerow - XF_SRCTEX0_INROW, texinfo.sourcerow - XF_SRCTEX0_INROW);
      break;
    }
    // Input form of AB11 sets z element to 1.0
    uid_data->texMtxInfo[i].inputform = xfmem.texMtxInfo[i].inputform;
    if (texinfo.inputform == XF_TEXINPUT_AB11)
      out.Write("coord.z = 1.0;\n");

    // first transformation
    uid_data->texMtxInfo[i].texgentype = xfmem.texMtxInfo[i].texgentype;
    switch (texinfo.texgentype)
    {
    case XF_TEXGEN_EMBOSS_MAP:  // calculate tex coords into bump map

      if (components & (VB_HAS_NRM1 | VB_HAS_NRM2))
      {
        // transform the light dir into tangent space
        uid_data->texMtxInfo[i].embosslightshift = xfmem.texMtxInfo[i].embosslightshift;
        uid_data->texMtxInfo[i].embosssourceshift = xfmem.texMtxInfo[i].embosssourceshift;
        out.Write("ldir = normalize(" LIGHT_POS ".xyz - pos.xyz);\n",
                  LIGHT_POS_PARAMS(texinfo.embosslightshift));
        out.Write(
            "o.tex%d.xyz = o.tex%d.xyz + float3(dot(ldir, _norm1), dot(ldir, _norm2), 0.0);\n", i,
            texinfo.embosssourceshift);
      }
      else
      {
        // The following assert was triggered in House of the Dead Overkill and Star Wars Rogue
        // Squadron 2
        //_assert_(0); // should have normals
        uid_data->texMtxInfo[i].embosssourceshift = xfmem.texMtxInfo[i].embosssourceshift;
        out.Write("o.tex%d.xyz = o.tex%d.xyz;\n", i, texinfo.embosssourceshift);
      }

      break;
    case XF_TEXGEN_COLOR_STRGBC0:
      out.Write("o.tex%d.xyz = float3(o.colors_0.x, o.colors_0.y, 1);\n", i);
      break;
    case XF_TEXGEN_COLOR_STRGBC1:
      out.Write("o.tex%d.xyz = float3(o.colors_1.x, o.colors_1.y, 1);\n", i);
      break;
    case XF_TEXGEN_REGULAR:
    default:
      uid_data->texMtxInfo_n_projection |= xfmem.texMtxInfo[i].projection << i;
      if (components & (VB_HAS_TEXMTXIDX0 << i))
      {
        out.Write("int tmp = int(tex%d.z);\n", i);
        if (texinfo.projection == XF_TEXPROJ_STQ)
          out.Write("o.tex%d.xyz = float3(dot(coord, " I_TRANSFORMMATRICES
                    "[tmp]), dot(coord, " I_TRANSFORMMATRICES
                    "[tmp+1]), dot(coord, " I_TRANSFORMMATRICES "[tmp+2]));\n",
                    i);
        else
          out.Write("o.tex%d.xyz = float3(dot(coord, " I_TRANSFORMMATRICES
                    "[tmp]), dot(coord, " I_TRANSFORMMATRICES "[tmp+1]), 1);\n",
                    i);
      }
      else
      {
        if (texinfo.projection == XF_TEXPROJ_STQ)
          out.Write("o.tex%d.xyz = float3(dot(coord, " I_TEXMATRICES
                    "[%d]), dot(coord, " I_TEXMATRICES "[%d]), dot(coord, " I_TEXMATRICES
                    "[%d]));\n",
                    i, 3 * i, 3 * i + 1, 3 * i + 2);
        else
          out.Write("o.tex%d.xyz = float3(dot(coord, " I_TEXMATRICES
                    "[%d]), dot(coord, " I_TEXMATRICES "[%d]), 1);\n",
                    i, 3 * i, 3 * i + 1);
      }
      break;
    }

    uid_data->dualTexTrans_enabled = xfmem.dualTexTrans.enabled;
    // CHECKME: does this only work for regular tex gen types?
    if (xfmem.dualTexTrans.enabled && texinfo.texgentype == XF_TEXGEN_REGULAR)
    {
      const PostMtxInfo& postInfo = xfmem.postMtxInfo[i];

      uid_data->postMtxInfo[i].index = xfmem.postMtxInfo[i].index;
      int postidx = postInfo.index;
      out.Write("float4 P0 = " I_POSTTRANSFORMMATRICES "[%d];\n"
                "float4 P1 = " I_POSTTRANSFORMMATRICES "[%d];\n"
                "float4 P2 = " I_POSTTRANSFORMMATRICES "[%d];\n",
                postidx & 0x3f, (postidx + 1) & 0x3f, (postidx + 2) & 0x3f);

      uid_data->postMtxInfo[i].normalize = xfmem.postMtxInfo[i].normalize;
      if (postInfo.normalize)
        out.Write("o.tex%d.xyz = normalize(o.tex%d.xyz);\n", i, i);

      // multiply by postmatrix
      out.Write("o.tex%d.xyz = float3(dot(P0.xyz, o.tex%d.xyz) + P0.w, dot(P1.xyz, o.tex%d.xyz) + "
                "P1.w, dot(P2.xyz, o.tex%d.xyz) + P2.w);\n",
                i, i, i, i);
    }

    out.Write("}\n");
  }

  // clipPos/w needs to be done in pixel shader, not here
  out.Write("o.clipPos = o.pos;\n");

  if (g_ActiveConfig.bEnablePixelLighting)
  {
    out.Write("o.Normal = _norm0;\n");
    out.Write("o.WorldPos = pos.xyz;\n");

    if (components & VB_HAS_COL0)
      out.Write("o.colors_0 = color0;\n");

    if (components & VB_HAS_COL1)
      out.Write("o.colors_1 = color1;\n");
  }

  // write the true depth value, if the game uses depth textures pixel shaders will override with
  // the correct values
  // if not early z culling will improve speed
  if (g_ActiveConfig.backend_info.bSupportsClipControl)
  {
    out.Write("o.pos.z = -o.pos.z;\n");
  }
  else  // OGL
  {
    // this results in a scale from -1..0 to -1..1 after perspective
    // divide
    out.Write("o.pos.z = o.pos.z * -2.0 - o.pos.w;\n");

    // the next steps of the OGL pipeline are:
    // (x_c,y_c,z_c,w_c) = o.pos  //switch to OGL spec terminology
    // clipping to -w_c <= (x_c,y_c,z_c) <= w_c
    // (x_d,y_d,z_d) = (x_c,y_c,z_c)/w_c//perspective divide
    // z_w = (f-n)/2*z_d + (n+f)/2
    // z_w now contains the value to go to the 0..1 depth buffer

    // trying to get the correct semantic while not using glDepthRange
    // seems to get rather complicated
  }

  // The console GPU places the pixel center at 7/12 in screen space unless
  // antialiasing is enabled, while D3D and OpenGL place it at 0.5. This results
  // in some primitives being placed one pixel too far to the bottom-right,
  // which in turn can be critical if it happens for clear quads.
  // Hence, we compensate for this pixel center difference so that primitives
  // get rasterized correctly.
  out.Write("o.pos.xy = o.pos.xy - o.pos.w * " I_PIXELCENTERCORRECTION ".xy;\n");

  if (api_type == API_OPENGL)
  {
    if (g_ActiveConfig.backend_info.bSupportsGeometryShaders)
    {
      AssignVSOutputMembers(out, "vs", "o");
    }
    else
    {
      // TODO: Pass interface blocks between shader stages even if geometry shaders
      // are not supported, however that will require at least OpenGL 3.2 support.
      for (unsigned int i = 0; i < xfmem.numTexGen.numTexGens; ++i)
        out.Write("uv%d.xyz = o.tex%d;\n", i, i);
      out.Write("clipPos = o.clipPos;\n");
      if (g_ActiveConfig.bEnablePixelLighting)
      {
        out.Write("Normal = o.Normal;\n");
        out.Write("WorldPos = o.WorldPos;\n");
      }
      out.Write("colors_0 = o.colors_0;\n");
      out.Write("colors_1 = o.colors_1;\n");
    }

    out.Write("gl_Position = o.pos;\n");
  }
  else  // D3D
  {
    out.Write("return o;\n");
  }
  out.Write("}\n");

  return out;
}