Improved Lambert shadowmaps + Hemisphere light

[bunnybones1]

I changed the indirect lighting on the Lambert material so that shadows aren't pitch black when using hemispherical light. I accomplished this with an extra varying for ambient light, but only if using shadows and a hemisphere light. Supports double-sided materials.

Fixes #15530.

[mrdoob]

Nice! Do you mind removing the builds from the PR though?

[WestLangley]

There is no need to condition on the presence of shadows. Keep it simple:

vLightFront = vec3( 0.0 );
vIndirectFront = vec3( 0.0 );

#ifdef DOUBLE_SIDED
	vLightBack = vec3( 0.0 );
	vIndirectBack = vec3( 0.0 );
#endif

[bunnybones1]

I like calling the light indirect rather than ambient.
As for the conditions, it's to prevent most uses of Lambert from doing the extra varying color, and also to prevent adding the light colors in the fragment shader. You're sure the simplicity trumps the small performance difference? I'm keen to agree, but just want to make sure.

[WestLangley]

We can optimize for performance later if there is a performance issue. I doubt there will be.

[WestLangley]

No need to condition on shadows

for ( int i = 0; i < NUM_HEMI_LIGHTS; i ++ ) {

	vIndirectFront += getHemisphereLightIrradiance( hemisphereLights[ i ], geometry );

	#ifdef DOUBLE_SIDED

		vIndirectBack += getHemisphereLightIrradiance( hemisphereLights[ i ], backGeometry );

	#endif

}

[WestLangley]

You can remove the conditional.

[WestLangley]

Something like this:

// accumulation
reflectedLight.indirectDiffuse = getAmbientLightIrradiance( ambientLightColor );

#include <lightmap_fragment>

#ifdef DOUBLE_SIDED

	reflectedLight.indirectDiffuse += ( gl_FrontFacing ) ? vIndirectFront : vIndirectBack;

	reflectedLight.directDiffuse = ( gl_FrontFacing ) ? vLightFront : vLightBack;

#else

	reflectedLight.indirectDiffuse += vIndirectFront;

	reflectedLight.directDiffuse = vLightFront;

#endif

reflectedLight.indirectDiffuse *= BRDF_Diffuse_Lambert( diffuseColor.rgb );

reflectedLight.directDiffuse *= BRDF_Diffuse_Lambert( diffuseColor.rgb ) * getShadowMask();

[bunnybones1]

I just want to clarify that the reason for the shadow and hemisphere light check are because its only when we use these features together that we need to move the final light color calculation to the fragment shader. Otherwise it's fine to precalculate the final light color entirely in the vertex shader. Let me know if you'd prefer the simplicity wins over the small performance save and I'll submit a simplified commit.

[WestLangley]

I prefer to model the lighting correctly, and keep indirect separate from direct.

[bunnybones1]

Ok cool, I'll have the modified PR ready tomorrow morning. Thanks!

[WestLangley]

@bunnybones1 Thanks for this. You are correct, this needs to be fixed, but the basic lighting equations should not be changed just because shadows are present. See #15530 (comment).

Please see if my suggestions look correct to you.

[bunnybones1]

Nice! Do you mind removing the builds from the PR though?

Done! I'll remove the conditions as @WestLangley suggested and we should be good to go.

[bunnybones1]

Should be good to go!

[mrdoob]

Thanks!