import lib-pbr.glsl

float normal_distrib( 
  vec3 localH, 
  vec2 alpha) 
{ 
  localH.xy /= alpha; 
  float tmp = dot(localH, localH); 
  return 1.0 / (M_PI * alpha.x * alpha.y * tmp * tmp); 
} 
 
float G1( 
  vec3 localW, 
  vec2 alpha) 
{ 
  // One generic factor of the geometry function divided by ndw 
  localW.xy *= alpha; 
  return 2.0 / (localW.z + length(localW)); 
} 
 
float visibility( 
  vec3 localL, 
  vec3 localV, 
  vec2 alpha) 
{ 
  // visibility is a Cook-Torrance geometry function divided by (n.l)*(n.v) 
  return G1(localL, alpha) * G1(localV, alpha); 
} 
 
vec3 cook_torrance_contrib( 
  float vdh, 
  float ndh, 
  vec3 localL, 
  vec3 localE, 
  vec3 Ks, 
  vec2 alpha) 
{ 
  // This is the contribution when using importance sampling with the GGX based 
  // sample distribution. This means ct_contrib = ct_brdf / ggx_probability 
  return fresnel(vdh, Ks) * (visibility(localL, localE, alpha) * vdh * localL.z / ndh); 
} 
 
vec3 importanceSampleGGX(vec2 Xi, vec2 alpha) 
{ 
  float phi = 2.0 * M_PI * Xi.x; 
  vec2 slope = sqrt(Xi.y / (1.0 - Xi.y)) * alpha * vec2(cos(phi), sin(phi)); 
  return normalize(vec3(slope, 1.0)); 
} 
 
float probabilityGGX(vec3 localH, float vdh, vec2 alpha) 
{ 
  return normal_distrib(localH, alpha) * localH.z / (4.0 * vdh); 
} 
 
vec3 pbrComputeSpecularAnisotropic(LocalVectors vectors, vec3 specColor, vec2 roughness) 
{ 
  vec3 radiance = vec3(0.0); 
  vec2 alpha = roughness * roughness; 
  mat3 TBN = mat3(vectors.tangent, vectors.bitangent, vectors.normal); 
  vec3 localE = vectors.eye * TBN; 
 
  for(int i=0; i<nbSamples; ++i) 
  { 
    vec2 Xi = fibonacci2DDitheredTemporal(i, nbSamples); 
    vec3 localH = importanceSampleGGX(Xi, alpha); 
    vec3 localL = reflect(-localE, localH); 
 
    if (localL.z > 0.0) 
    { 
      vec3 Ln = TBN * localL; 
      float vdh = max(1e-8, dot(localE, localH)); 
 
      float fade = horizonFading(dot(vectors.vertexNormal, Ln), horizonFade); 
      float pdf = probabilityGGX(localH, vdh, alpha); 
      float lodS = max(roughness.x, roughness.y) < 0.01 ? 0.0 : computeLOD(Ln, pdf); 
      // Offset lodS to trade bias for more noise 
      lodS -= 1.0; 
      vec3 preconvolvedSample = envSampleLOD(Ln, lodS); 
 
      radiance += 
        fade * cook_torrance_contrib(vdh, localH.z, localL, localE, specColor, alpha) * 
        preconvolvedSample; 
    } 
  } 
 
  return radiance / float(nbSamples); 
}