Finish the convolution shader

libraries/pbrlib/genglsl/lib/mx_pre_convolve_environment.glsl

@@ -62,8 +62,7 @@ void sample_ggx_dir(
  out vec3 L,
  out float NdotL,
  out float NdotH,
- out float VdotH,
- bool VeqN = false
+ out float VdotH
 )
 {
  // GGX NDF sampling
@@ -76,17 +75,9 @@ void sample_ggx_dir(
 
  vec3 localV;
 
- if (VeqN)
- {
- // localV == localN
- localV = vec3(0.0, 0.0, 1.0);
- VdotH = NdotH;
- }
- else
- {
- localV = transpose(localToWorld) * V;
- VdotH = saturate(dot(localV, localH));
- }
+ // localV == localN
+ localV = vec3(0.0, 0.0, 1.0);
+ VdotH = NdotH;
 
  // Compute { localL = reflect(-localV, localH) }
  vec3 localL = -localV + 2.0 * VdotH * localH;
@@ -115,38 +106,6 @@ float v_smith_joint_ggx(float NdotL, float NdotV, float alpha, float partLambdaV
  return 0.5 / max(lambdaV + lambdaL, FLT_MIN);
 }
 
-// // Use lower MIP-map levels for fetching samples with low probabilities
-// // in order to reduce the variance.
-// // Ref: http://http.developer.nvidia.com/GPUGems3/gpugems3_ch20.html
-// float get_prefiltered_mip_level(float NdotH, float alpha, float sampleCount, vec3 L)
-// {
-// // OmegaS: Solid angle associated with the sample
-// // OmegaP: Solid angle associated with the texel of the cubemap
-// float omegaS;
-
-// // float PDF = D * NdotH * Jacobian, where Jacobian = 1 / (4 * LdotH).
-// // Since (N == V), NdotH == LdotH.
-// // Thus PDF = D / 4.
-// float a2 = alpha * alpha;
-// float s = (NdotH * a2 - NdotH) * NdotH + 1.0;
-// float d_GGX = M_PI_INV * a2 / s * s;
-// float pdf = d_GGX * 0.25;
-// omegaS = 1.0 / (sampleCount * pdf);
-
-// const float mipBias = alpha;
-
-// vec3 envDir = normalize(($envMatrix * vec4(L,0.0)).xyz);
-// vec2 uv = mx_latlong_projection(envDir);
-// float deltaLongitude = 2.0 * M_PI / 2048.0;
-// float deltaLatitude = M_PI / 1024.0;
-// float latitudeStart = (uv.y - 0.5) * M_PI - 0.5 * deltaLatitude;
-// float latitudeEnd = latitudeStart + deltaLatitude;
-// float omegaP = deltaLongitude * (sin(latitudeEnd) - sin(latitudeStart));
-// float invOmegaP = 1.0 / omegaP;
-
-// return 0.5 * log2(omegaS * invOmegaP) + mipBias;
-// }
-
 vec3 mx_pre_convolve_environment()
 {
  vec2 uv = gl_FragCoord.xy * pow(2.0, (float)$convolutionMipLevel) / vec2(2048.0, 1024.0);
@@ -162,8 +121,10 @@ vec3 mx_pre_convolve_environment()
  const float NdotV = 1; // Because N == V
  float alpha = mx_lod_to_alpha(float($convolutionMipLevel));
  float partLambdaV = get_smith_joint_ggx_part_lambda_v(NdotV, alpha);
- // int sampleCount = 89; // Must be a Fibonacci number
- int sampleCount = 144; // Must be a Fibonacci number
+ // If we use prefiltering, we can have a smaller sample count, since pre-filtering will reduce
+ // the variance of the samples by choosing higher mip levels where necessary. We haven't
+ // implemented prefiltering yet, so we use a high sample count.
+ int sampleCount = 610; // Must be a Fibonacci number
 
  float3 lightInt = float3(0.0, 0.0, 0.0);
  float cbsdfInt = 0.0;
@@ -176,9 +137,11 @@ vec3 mx_pre_convolve_environment()
  float NdotL;
  float NdotH;
  float LdotH;
- sample_ggx_dir(Xi, V, localToWorld, alpha, L, NdotL, NdotH, LdotH, true);
+ sample_ggx_dir(Xi, V, localToWorld, alpha, L, NdotL, NdotH, LdotH);
+ if (NdotL <= 0) continue; // Note that some samples will have 0 contribution
 
- float mipLevel = 0.0; // get_prefiltered_mip_level(NdotH, alpha, sampleCount, L);
+ // If we were to implement pre-filtering, we would do so here.
+ float mipLevel = 0;
  vec3 val = mx_latlong_map_lookup(L, $envMatrix, mipLevel, $envRadiance);
  const float F = 1.0;
  float G = v_smith_joint_ggx(NdotL, NdotV, alpha, partLambdaV) * NdotL * NdotV; // 4 cancels out