16 Shading Models16 种光照模型

Model Registry — Key Values in GBuffer模型注册表 — GBuffer 中的键值

Each shading model has a numeric key stored in the GBuffer flags byte (bits 0-3). The enum is declared in forward_shading.render. New engine adds 4 models and removes Mobile_pbr:

每种着色模型都有一个存储在 GBuffer 标志字节（位 0-3）中的数字键。枚举在 forward_shading.render 中声明。新引擎新增 4 个模型，移除 Mobile_pbr：

Detailed Usage & Performance Guide详细使用与性能指南

Isotropy — Standard GGX PBR各向同性 — 标准 GGX PBR

isotropy.hlsl is the default shading model and the most feature-rich — it doubles as the SSSS 4S model when SKIN4S_ENABLE is set, and supports the TOON_EFFECT variant for cartoon diffuse ramps:

isotropy.hlsl 是默认着色模型，也是功能最丰富的——当设置 SKIN4S_ENABLE 时它充当 SSSS 4S 模型，并支持 TOON_EFFECT 变体用于卡通漫反射渐变：

// Energy compensation for multi-scattering (ENABLE_ENERGY_COMPENSATION)
// Prevents too-dark appearance at high roughness
float2 approxAB = EnvBRDFLUTApprox(roughness_pow2, n_dot_v);
energy_compensation = 1.0 + spec * (1.0 / (approxAB.x + approxAB.y) - 1.0);

// Toon diffuse (TOON_EFFECT)
float The = smoothstep(toon_edge.x, toon_edge.y, n_dot_l) * 0.5
           + smoothstep(toon_edge.y, toon_edge.z, n_dot_l);
light_diffuse = Lerp3(Dark_Color, Mid_Color, Light_Color, 0.5, The);

// SKIN4S dual-lobe specular (SKIN4S_SPECULARPASS)
float3 lobe0 = LightingSpecular(..., roughness0);
float3 lobe1 = LightingSpecular(..., roughness1);
light_specular = lerp(lobe0, lobe1, dual_lobe_mix);

Skin & Hair — Biologically-Grounded Models皮肤与头发 — 基于生物学的模型

skin.hlsl — Pre-Integrated Skinskin.hlsl — 预积分皮肤

Skin scattering is precomputed into a 2D LUT indexed by (0.5*NoL+0.5, curvature). A second LUT handles shadow coloring — skin shadows have a reddish fringe where light enters at grazing angles. LightNotAffect() always returns false so backface contributions are included:

皮肤散射预计算到以 (0.5*NoL+0.5, curvature) 为索引的 2D LUT 中。第二张 LUT 处理阴影着色——皮肤阴影在掠射角进光处有红色边缘。LightNotAffect() 始终返回 false，因此包含背面贡献：

// skin.hlsl — Diffuse from pre-integrated LUT
float3 LightingDiffuse(...) {
  float2 lut_uv = float2(0.5 * n_dot_l_raw + 0.5, mtl.sss_curvature);
  float3 skin   = GetPreintergratedSkin(lut_uv);    // Tex3.Sample(s1, lut_uv)
  float2 s_uv   = float2(shadow, mtl.sss_curvature);
  float3 shadow_color = GetShadowSkinColor(s_uv);     // Tex4.Sample(s1, s_uv)
  return skin * shadow_color;
}

hair.hlsl — Marschner 3-Lobe Modelhair.hlsl — Marschner 三波瓣模型

Hair uses a full Marschner physical model with three scattering lobes: R (specular reflection off the cuticle), TT (light transmitted through the hair), and TRT (transmitted → internal reflection → transmitted). Melanin-based absorption correctly derives the natural hair color from the underlying physics:

头发使用完整的 Marschner 物理模型，具有三个散射波瓣：R（角质层镜面反射）、TT（穿过发丝的透射光）和 TRT（透射→内反射→透射）。基于黑色素的吸收从底层物理中正确推导出自然发色：

// Marschner NDF — Gaussian in the longitudinal plane
float Hair_g(float B, float Theta) {
  return exp(-0.5 * Theta*Theta / (B*B)) / (sqrt(2*PI) * B);
}
// Melanin color model (Eumelanin + Pheomelanin absorption coefficients)
// Eumelanin:  sigma_a = (0.506, 0.841, 1.653) * concentration
// Pheomelanin sigma_a = (0.343, 0.733, 1.924) * concentration
float3 GetHairColorFromMelanin(float melanin, float redness, float3 dyeColor) {
  float3 sigma_a = melanin * lerp(eu, pheo, redness);
  return dyeColor * exp(-sigma_a);
}

Toon & ClearCoat卡通与清漆

toon.hlsl — Lerp3 Color Ramptoon.hlsl — Lerp3 色阶渐变

The toon shading model uses a 3-segment linear interpolation function Lerp3 to map the N·L value to a cel-shaded diffuse. The threshold and range parameters come from the material's sss_color.xy (packed by material_toon.hlsl):

卡通着色模型使用三段线性插值函数 Lerp3 将 N·L 值映射到色块漫反射。阈值和范围参数来自材质的 sss_color.xy（由 material_toon.hlsl 打包）：

// Lerp3: piecewise linear ramp over 3 color bands
float Lerp3(float C, float B, float A, float Middle, float T) {
  return  lerp(C, B, T / Middle)          * step(T, Middle)
        + lerp(B, A, (T - Middle) / (1 - Middle)) * step(Middle, T);
}
// Usage: T = smoothstep(..., n_dot_l), Middle = 0.5
// C=Dark_Color, B=Mid_Color, A=Light_Color read from CB

clearcoat.hlsl — Kelemen Visibilityclearcoat.hlsl — Kelemen 可见性

The clear coat layer uses Vis_Kelemen (a simpler visibility term) and fixes IOR at 1.5 (typical polymer). Energy absorbed by the coat layer attenuates the base layer contribution:

清漆层使用 Vis_Kelemen（更简单的可见性项），IOR 固定为 1.5（典型聚合物）。涂层吸收的能量会衰减底层贡献：

// Coat specular: D*Vis_Kelemen*F where F uses IOR=1.5 (F0=0.04)
float Vis_Kelemen(float VdotH) { return 0.25 / (VdotH * VdotH); }
float Fc = Pow5(1 - v_dot_h);
float F  = Fc + (1 - Fc) * 0.04;             // F0 = 0.04
float coat_fr = D_GGX(coat_rough_pow2, n_dot_h) * Vis_Kelemen(v_dot_h) * F;
float layer_atten = 1.0 - F;                  // energy reaching the base layer
// Final: base_spec * layer_atten + coat_fr

New Models: Cloth, Sheen, Glint新增模型：布料、Sheen、闪光

cloth.hlsl — Inverted GGX for Fabriccloth.hlsl — 用于织物的反转 GGX

Fabric fibers produce a specular highlight that is strongest at grazing angles (the opposite of smooth surfaces). This is modeled with an inverted GGX NDF and the Neubelt-Pettineo visibility function:

织物纤维产生在掠射角最强的高光（与光滑表面相反）。这用反转 GGX NDF 和 Neubelt-Pettineo 可见性函数建模：

// D_InvGGX: peaks at GRAZING angles (opposite of normal GGX)
float D_InvGGX(float a2, float NoH) {
  float d = (NoH - a2 * NoH) * NoH + a2;
  return rcp(PI * (1 + 4*a2)) * (1 + 4*a2*a2 / (d*d));
}
// Vis_Cloth: energy-conserving (Neubelt & Pettineo 2013)
float Vis_Cloth(float NoV, float NoL) {
  return rcp(4 * (NoL + NoV - NoL * NoV));
}
// Blend between GGX and cloth via 'Cloth' parameter:
float3 spec = lerp(normal_spec, D_InvGGX * Vis_Cloth * F_Schlick(FuzzColor, VoH), Cloth);

sheen.hlsl — Charlie BRDF for Velvetsheen.hlsl — 用于丝绒的 Charlie BRDF

Sheen/velvet surfaces use the Charlie BRDF (Estevez & Kulla 2017). The NDF is derived from sin-power distribution to produce the soft glancing highlight characteristic of velvet and microfiber:

Sheen/丝绒表面使用 Charlie BRDF（Estevez & Kulla 2017）。NDF 来自 sin 幂分布，产生丝绒和超细纤维特有的柔和掠射高光：

// D_Charlie: sin-power NDF peaking at 90° — the "inverted dome" highlight
float D_Charlie(float roughness, float NoH) {
  float invAlpha = 1.0 / (roughness * roughness);
  float cos2h = NoH * NoH;
  float sin2h = max(1.0 - cos2h, 0.0078125);
  return (2 + invAlpha) * pow(sin2h, invAlpha * 0.5) / (2 * PI);
}
// Environment BRDF: polynomial fit avoids texture LUT lookup
// float EnvClothSheenLUT3rd(R, NoV) = a*R / (b*R + c*NoV + NoV*R) + Offset

glint.hlsl — Stochastic Sparkle Flakes (P22 Method)glint.hlsl — 随机闪片（P22 方法）

The most complex shading model — implements the Stochastic Flakes BRDF for sparkling metallic paint, iridescent surfaces, and glitter. It generates thousands of statistically-correct microfacet highlights without explicit geometry:

最复杂的着色模型——为闪光金属漆、虹彩表面和亮片实现 随机闪片 BRDF。无需显式几何体即可生成数千个统计正确的微表面高光：

// Core: P22 hierarchical discrete NDF
// Queries a 3D GlintDictionary texture (precomputed distribution tables)
// hashIQ: integer hash → Gaussian-distributed microfacet slopes per cell
float3 P22_theta_alpha(float2 slope_h, float l, float2 s0, float2 t0) {
  // For each cell (i,j) in a grid over UV space:
  //   1. Hash cell ID to get random slope distribution parameters
  //   2. Evaluate Gaussian NDF at query slope vs cell's slope distribution
  //   3. Weight by cell microfacet density * cell area
  // Return: sum of all nearby cell contributions
}
// EWA filtering prevents temporal aliasing as the view changes
// Multi-lob path for filter footprint > 1 cell (far-away glints)
// Direct path (ndf_beckmann_anisotropic) for filter footprint ~0

New: water.hlsl — Slope-Space Microfacet Model新增：water.hlsl — 斜率空间微表面模型

water.hlsl implements a physically rigorous ocean rendering model based on slope-space statistics. Instead of a standard roughness value, it takes a statistical moment texture (LEAN map) encoding the full 2D slope covariance matrix:

water.hlsl 实现了基于斜率空间统计的物理严格海洋渲染模型。它接受编码完整 2D 斜率协方差矩阵的统计矩纹理（LEAN 贴图），而非标准粗糙度值：

// Full IBL path: multi-sample integration over slope space
float3 MicrofacetReflection(Material mtl, float3 V, ...) {
  // NumSamples == 1: direct environment sample (cheapest)
  // NumSamples > 1: jittered Gaussian grid in slope space
  for(int i = 0; i < NumSamples; i++) for(int j = 0; j < NumSamples; j++) {
    float2 slope = mean + jitter[i][j];           // Gaussian-distributed slope
    float3 NN    = normalize(float3(-slope.x, 1, -slope.y));
    float3 R     = reflect(-V, NN);              // per-sample reflection
    float  mip   = ComputeReflectionMip(roughness);
    color += SampleEnvCube(R, mip) * (VdotNN / ZdotNN); // projected area weight
  }
}
// Fresnel: roughness-modulated with exp(-2.69 * sigma^2) correction
// IOR = 1.33 (water at visible light frequencies)

Special Models: SSS, Eye, Glass, Transparent, Unlit特殊模型：SSS、眼睛、玻璃、透明、无光照

subsurface.hlsl & subsurface_two_sided.hlslsubsurface.hlsl 与 subsurface_two_sided.hlsl

Simple analytical SSS approximation. The single-sided variant approximates forward scattering via a tight power function; the two-sided variant adds a wrap-diffuse term for thin surfaces like leaves:

简单的解析 SSS 近似。单面变体通过紧凑幂函数近似正向散射；双面变体为树叶等薄表面添加 wrap-diffuse 项：

// subsurface: InScatter = pow(saturate(-dot(L,V)), 12) * lerp(3, 0.1, Opacity)
// subsurface_two_sided: WrapNoL = saturate((-NoL + 0.5) / 1.5^2)
//   Scatter = D_GGX(0.36, saturate(-VoL))  (wide GGX as scatter kernel)

eye.hlsl — Caustic + Scleraeye.hlsl — 焦散 + 巩膜

Eye rendering needs separate handling for iris and sclera. The iris uses a caustic term to simulate the focusing effect of the cornea; the sclera uses wrapped Lambert diffuse. A micro-shadow term handles sub-surface AO:

眼睛渲染需要分别处理虹膜和巩膜。虹膜使用焦散项模拟角膜的聚焦效果；巩膜使用 Wrapped Lambert 漫反射。微阴影项处理次表面 AO：

// Caustic: simulates iris focus via bent CausticNormal
CausticNormal = normalize(lerp(IrisNormal, -N, IrisMask * IrisDistance));
Caustic = 0.6 + 0.2*(Power+1) * pow(dot(CausticNormal, L), Power);
// Micro-shadow (SHOW_UE4_MODE)
micro_shadow = saturate(abs(n_dot_l) + 2*occlusion*occlusion - 1);

glass.hlsl — Blended Refractionglass.hlsl — 混合折射

Glass pre-computes both a reflected and a refracted world direction, then blends environment samples in the EnvLighting() pass — no screen-space refraction:

玻璃预计算反射和折射世界方向，然后在 EnvLighting() 通道中混合环境采样——没有屏幕空间折射：

// refract_world precomputed in CalcLightingData(view)
refract_world = refract(v.direction, N, 1.0);
// EnvLighting blends: reflected_env + refracted_env * (1 - opacity)

transparent.hlsl — Alpha Energy Compensationtransparent.hlsl — Alpha 能量补偿

Thin transparent surfaces (e.g. stained glass, colored film) would appear too dark at low opacity if standard GGX is used. Transparent divides specular by pow(opacity, 2.2) to compensate — less opaque = brighter reflection to conserve energy:

薄透明表面（如彩色玻璃、色膜）在低不透明度时使用标准 GGX 会显得太暗。Transparent 将高光除以 pow(opacity, 2.2) 补偿——透明度越低 = 反射越亮以守恒能量：

light_specular /= pow(mtl.opacity, 2.2);
env_spec       /= pow(mtl.opacity, 2.2);

Shared Infrastructure — shading_models_common.hlsl共享基础设施 — shading_models_common.hlsl

All shading models include shading_models_common.hlsl, which provides the shared environment sampling, reflection probe, and IBL functions:

所有着色模型都包含 shading_models_common.hlsl，提供共享的环境采样、反射探针和 IBL 函数：