static void EndPrimitive(ShaderCode& out, const geometry_shader_uid_data* uid_data, APIType ApiType)
{
if (uid_data->wireframe)
EmitVertex(out, uid_data, "first", ApiType);
if (ApiType == APIType::OpenGL || ApiType == APIType::Vulkan)
out.Write("\tEndPrimitive();\n");
else
out.Write("\toutput.RestartStrip();\n");
}
static void EmitVertex(ShaderCode& out, const geometry_shader_uid_data* uid_data,
const char* vertex, APIType ApiType, bool first_vertex)
{
if (uid_data->wireframe && first_vertex)
out.Write("\tif (i == 0) first = %s;\n", vertex);
if (ApiType == APIType::OpenGL)
{
out.Write("\tgl_Position = %s.pos;\n", vertex);
if (g_ActiveConfig.backend_info.bSupportsDepthClamp)
{
out.Write("\tgl_ClipDistance[0] = %s.clipDist0;\n", vertex);
out.Write("\tgl_ClipDistance[1] = %s.clipDist1;\n", vertex);
}
AssignVSOutputMembers(out, "ps", vertex, uid_data->numTexGens, uid_data->pixel_lighting);
}
else if (ApiType == APIType::Vulkan)
{
// Vulkan NDC space has Y pointing down (right-handed NDC space).
out.Write("\tgl_Position = %s.pos;\n", vertex);
out.Write("\tgl_Position.y = -gl_Position.y;\n");
AssignVSOutputMembers(out, "ps", vertex, uid_data->numTexGens, uid_data->pixel_lighting);
}
else
{
out.Write("\tps.o = %s;\n", vertex);
}
if (ApiType == APIType::OpenGL || ApiType == APIType::Vulkan)
out.Write("\tEmitVertex();\n");
else
out.Write("\toutput.Append(ps);\n");
}
bool LineGeometryShader::SetShader(u32 components, float lineWidth,
float texOffset, float vpWidth, float vpHeight, const bool* texOffsetEnable)
{
if (!m_ready)
return false;
// Make sure geometry shader for "components" is available
ComboMap::iterator shaderIt = m_shaders.find(components);
if (shaderIt == m_shaders.end())
{
// Generate new shader. Warning: not thread-safe.
static char buffer[16384];
ShaderCode code;
code.SetBuffer(buffer);
GenerateVSOutputStructForGS(code, API_D3D);
code.Write("\n%s", LINE_GS_COMMON);
std::stringstream numTexCoordsStream;
numTexCoordsStream << xfmem.numTexGen.numTexGens;
INFO_LOG(VIDEO, "Compiling line geometry shader for components 0x%.08X (num texcoords %d)",
components, xfmem.numTexGen.numTexGens);
const std::string& numTexCoordsStr = numTexCoordsStream.str();
D3D_SHADER_MACRO macros[] = {
{ "NUM_TEXCOORDS", numTexCoordsStr.c_str() },
{ nullptr, nullptr }
};
ID3D11GeometryShader* newShader = D3D::CompileAndCreateGeometryShader(code.GetBuffer(), macros);
if (!newShader)
{
WARN_LOG(VIDEO, "Line geometry shader for components 0x%.08X failed to compile", components);
// Add dummy shader to prevent trying to compile again
m_shaders[components] = nullptr;
return false;
}
shaderIt = m_shaders.insert(std::make_pair(components, newShader)).first;
}
if (shaderIt != m_shaders.end())
{
if (shaderIt->second)
{
D3D11_MAPPED_SUBRESOURCE map;
HRESULT hr = D3D::context->Map(m_paramsBuffer, 0, D3D11_MAP_WRITE_DISCARD, 0, &map);
if (SUCCEEDED(hr))
{
LineGSParams* params = (LineGSParams*)map.pData;
params->LineWidth = lineWidth;
params->TexOffset = texOffset;
params->VpWidth = vpWidth;
params->VpHeight = vpHeight;
for (int i = 0; i < 8; ++i)
params->TexOffsetEnable[i] = texOffsetEnable[i] ? 1.f : 0.f;
D3D::context->Unmap(m_paramsBuffer, 0);
}
else
ERROR_LOG(VIDEO, "Failed to map line gs params buffer");
DEBUG_LOG(VIDEO, "Line params: width %f, texOffset %f, vpWidth %f, vpHeight %f",
lineWidth, texOffset, vpWidth, vpHeight);
D3D::context->GSSetShader(shaderIt->second, nullptr, 0);
D3D::context->GSSetConstantBuffers(0, 1, &m_paramsBuffer);
return true;
}
else
return false;
}
else
return false;
}
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;
}
// vertex shader
// lights/colors
// materials name is I_MATERIALS in vs and I_PMATERIALS in ps
// inColorName is color in vs and colors_ in ps
// dest is o.colors_ in vs and colors_ in ps
void GenerateLightingShaderCode(ShaderCode& object, const LightingUidData& uid_data, int components,
const char* inColorName, const char* dest)
{
for (unsigned int j = 0; j < xfmem.numChan.numColorChans; j++)
{
object.Write("{\n");
bool colormatsource = !!(uid_data.matsource & (1 << j));
if (colormatsource) // from vertex
{
if (components & (VB_HAS_COL0 << j))
object.Write("int4 mat = int4(round(%s%d * 255.0));\n", inColorName, j);
else if (components & VB_HAS_COL0)
object.Write("int4 mat = int4(round(%s0 * 255.0));\n", inColorName);
else
object.Write("int4 mat = int4(255, 255, 255, 255);\n");
}
else // from color
{
object.Write("int4 mat = %s[%d];\n", I_MATERIALS, j + 2);
}
if (uid_data.enablelighting & (1 << j))
{
if (uid_data.ambsource & (1 << j)) // from vertex
{
if (components & (VB_HAS_COL0 << j))
object.Write("lacc = int4(round(%s%d * 255.0));\n", inColorName, j);
else if (components & VB_HAS_COL0)
object.Write("lacc = int4(round(%s0 * 255.0));\n", inColorName);
else
// TODO: this isn't verified. Here we want to read the ambient from the vertex,
// but the vertex itself has no color. So we don't know which value to read.
// Returning 1.0 is the same as disabled lightning, so this could be fine
object.Write("lacc = int4(255, 255, 255, 255);\n");
}
else // from color
{
object.Write("lacc = %s[%d];\n", I_MATERIALS, j);
}
}
else
{
object.Write("lacc = int4(255, 255, 255, 255);\n");
}
// check if alpha is different
bool alphamatsource = !!(uid_data.matsource & (1 << (j + 2)));
if (alphamatsource != colormatsource)
{
if (alphamatsource) // from vertex
{
if (components & (VB_HAS_COL0 << j))
object.Write("mat.w = int(round(%s%d.w * 255.0));\n", inColorName, j);
else if (components & VB_HAS_COL0)
object.Write("mat.w = int(round(%s0.w * 255.0));\n", inColorName);
else
object.Write("mat.w = 255;\n");
}
else // from color
{
object.Write("mat.w = %s[%d].w;\n", I_MATERIALS, j + 2);
}
}
if (uid_data.enablelighting & (1 << (j + 2)))
{
if (uid_data.ambsource & (1 << (j + 2))) // from vertex
{
if (components & (VB_HAS_COL0 << j))
object.Write("lacc.w = int(round(%s%d.w * 255.0));\n", inColorName, j);
else if (components & VB_HAS_COL0)
object.Write("lacc.w = int(round(%s0.w * 255.0));\n", inColorName);
else
// TODO: The same for alpha: We want to read from vertex, but the vertex has no color
object.Write("lacc.w = 255;\n");
}
else // from color
{
object.Write("lacc.w = %s[%d].w;\n", I_MATERIALS, j);
}
}
else
{
object.Write("lacc.w = 255;\n");
}
if (uid_data.enablelighting & (1 << j)) // Color lights
{
for (int i = 0; i < 8; ++i)
if (uid_data.light_mask & (1 << (i + 8 * j)))
GenerateLightShader(object, uid_data, i, j, false);
}
if (uid_data.enablelighting & (1 << (j + 2))) // Alpha lights
{
for (int i = 0; i < 8; ++i)
if (uid_data.light_mask & (1 << (i + 8 * (j + 2))))
GenerateLightShader(object, uid_data, i, j + 2, true);
}
object.Write("lacc = clamp(lacc, 0, 255);\n");
object.Write("%s%d = float4((mat * (lacc + (lacc >> 7))) >> 8) / 255.0;\n", dest, j);
object.Write("}\n");
}
}
static void GenerateLightShader(ShaderCode& object, const LightingUidData& uid_data, int index,
int litchan_index, bool alpha)
{
const char* swizzle = alpha ? "a" : "rgb";
const char* swizzle_components = (alpha) ? "" : "3";
int attnfunc = (uid_data.attnfunc >> (2 * litchan_index)) & 0x3;
int diffusefunc = (uid_data.diffusefunc >> (2 * litchan_index)) & 0x3;
switch (attnfunc)
{
case LIGHTATTN_NONE:
case LIGHTATTN_DIR:
object.Write("ldir = normalize(" LIGHT_POS ".xyz - pos.xyz);\n", LIGHT_POS_PARAMS(index));
object.Write("attn = 1.0;\n");
object.Write("if (length(ldir) == 0.0)\n\t ldir = _norm0;\n");
break;
case LIGHTATTN_SPEC:
object.Write("ldir = normalize(" LIGHT_POS ".xyz - pos.xyz);\n", LIGHT_POS_PARAMS(index));
object.Write("attn = (dot(_norm0, ldir) >= 0.0) ? max(0.0, dot(_norm0, " LIGHT_DIR
".xyz)) : 0.0;\n",
LIGHT_DIR_PARAMS(index));
object.Write("cosAttn = " LIGHT_COSATT ".xyz;\n", LIGHT_COSATT_PARAMS(index));
object.Write("distAttn = %s(" LIGHT_DISTATT ".xyz);\n",
(diffusefunc == LIGHTDIF_NONE) ? "" : "normalize", LIGHT_DISTATT_PARAMS(index));
object.Write("attn = max(0.0f, dot(cosAttn, float3(1.0, attn, attn*attn))) / dot(distAttn, "
"float3(1.0, attn, attn*attn));\n");
break;
case LIGHTATTN_SPOT:
object.Write("ldir = " LIGHT_POS ".xyz - pos.xyz;\n", LIGHT_POS_PARAMS(index));
object.Write("dist2 = dot(ldir, ldir);\n"
"dist = sqrt(dist2);\n"
"ldir = ldir / dist;\n"
"attn = max(0.0, dot(ldir, " LIGHT_DIR ".xyz));\n",
LIGHT_DIR_PARAMS(index));
// attn*attn may overflow
object.Write("attn = max(0.0, " LIGHT_COSATT ".x + " LIGHT_COSATT ".y*attn + " LIGHT_COSATT
".z*attn*attn) / dot(" LIGHT_DISTATT ".xyz, float3(1.0,dist,dist2));\n",
LIGHT_COSATT_PARAMS(index), LIGHT_COSATT_PARAMS(index), LIGHT_COSATT_PARAMS(index),
LIGHT_DISTATT_PARAMS(index));
break;
}
switch (diffusefunc)
{
case LIGHTDIF_NONE:
object.Write("lacc.%s += int%s(round(attn * float%s(" LIGHT_COL ")));\n", swizzle,
swizzle_components, swizzle_components, LIGHT_COL_PARAMS(index, swizzle));
break;
case LIGHTDIF_SIGN:
case LIGHTDIF_CLAMP:
object.Write("lacc.%s += int%s(round(attn * %sdot(ldir, _norm0)) * float%s(" LIGHT_COL ")));\n",
swizzle, swizzle_components, diffusefunc != LIGHTDIF_SIGN ? "max(0.0," : "(",
swizzle_components, LIGHT_COL_PARAMS(index, swizzle));
break;
default:
_assert_(0);
}
object.Write("\n");
}
