Revert brdf

[vorg]

While adding clear coat big restructuring happened in parallel. This commit cleans up and simplifies the clear coat code:

• [x] switch to our orginal ggx brdf as the lighting was more balanced and to avoid breaking older projects
• [x] simplify clear coat code by moving NdotL and other common contributions outside of if checks
• [x] remove TARGET_MOBILE, some optimisations are hard to compare and measure, as we are targeting web tech stack we should probably decide on one simpler option
• [x] remove specular AO
• [x] unify using .a instead of .w for light intensity
• [x] decide on f0 vs specularColor (decided on f0)
• [x] add comments explaining clear coat normal mapping
• [x] calculate values in the same order they are used in equations
• [x] add clear coat example
• [x] update example screenshots
• [x] Clarify alphaRoughness vs linearRoughness vs roughness

[vorg]

[vorg]

After removing specularColor some materials got darker

Investigating:

lightColor = same exposure = same diffuseColor = same f0 = same directColor color = is different directColor lightColor = is darker -> old tested version had double light intensity bug, once fixed they are the same

indirectDiffuse = same indirectSpecular = is darker specularReflectance = is darker prefilteredRadiance = same f0 = same

Here is the reason. We use new reflectance based formula for f0 but old f0 mixing for specularColor.

data.f0 = 0.16 * uReflectance * uReflectance * (1.0 - data.metallic) + data.baseColor.rgb * data.metallic;
data.specularColor = mix(data.f0, data.baseColor, data.metallic);

But is it? For default reflectance of 0.5 it should produce identical result. Hmm. Another clue only mid values for metallic and clearCoat examples are impacted Replacing F0 formulate with specularColor mix based one, fixed metallic material but not clearCoat.

Fixed: as specularColor is based on f0 that already takes metallic ratio into account, blending it again using metallic moves it even closer to 1. The solution is to use uncorrelated version of F0. So new version is corect and old version was too bright. This confirms yet another reason to keep only F0

data.specularColor = mix(vec3(0.16 * uReflectance * uReflectance), data.baseColor, data.metallic);