Change the light attenuation formulas.

[reduz]

-Much better looking, physically based.
-Almost negligible extra cost.

Pictures (Old Left, New Right):

Videos:

Old:

old_attenuation.mp4

New:

new_attenuation.mp4

[fire]

Is it possible if you also changed the formulas in the gltf document for importing and exporting lights at the same time.

godot/modules/gltf/gltf_document.cpp

Line 4936 in 6e23ab4

Light3D *GLTFDocument::_generate_light(Ref<GLTFState> state, Node *scene_parent, const GLTFNodeIndex node_index) {

godot/modules/gltf/gltf_document.cpp

Line 5027 in 6e23ab4

GLTFLightIndex GLTFDocument::_convert_light(Ref<GLTFState> state, Light3D *p_light) {

See original test cases from gltf assets in #39450

Edited: Had the wrong link for exporting lights.

[akien-mga]

There are similar looking equations in other places, are they meant to change too?

(Non exhaustive list, I just looked for spot_attenuation and for some pow(1.0 - x) equations.)

modules/lightmapper_rd/lm_compute.glsl
322:                            attenuation *= pow(1.0 - d, lights.data[i].spot_attenuation);
servers/rendering/renderer_rd/shaders/giprobe_write.glsl
155:                    attenuation *= pow(1.0 - d, lights.data[light].spot_attenuation);
servers/rendering/renderer_rd/shaders/sdfgi_direct_light.glsl
211:                                    attenuation *= pow(1.0 - d, lights.data[i].spot_attenuation);
servers/rendering/renderer_rd/shaders/scene_forward.glsl
1222:   spot_attenuation *= 1.0 - pow(spot_rim, spot_att_angle.x);
servers/rendering/renderer_rd/shaders/volumetric_fog.glsl
280:                    float attenuation = pow(max(1.0 - d, 0.0), attenuation_energy.x);
336:                    float attenuation = pow(max(1.0 - d, 0.0), attenuation_energy.x);

[clayjohn]

Looks good to me. Subject to Akien's comment above. It looks like you accidentally removed a line.

[reduz]

There are similar looking equations in other places, are they meant to change too?

No, spot attenuation is ok, this will never be physically very correct anyway, we need to invert it though. @JFonS was planning to work on this, we can do it a separate PR, and this can be a good guide of all the places that need to be touched (though probably in this case, it needs to be inverted on the CPP)

[clayjohn]

@reduz the volumetric fog attenuation should change though.

[JFonS]

Yup, I already found a way to properly invert the angle attenuation. I will also make sure this attenuation is consistent across LM/SDFGI/GIProbe etc. because currently there are some discrepancies.

[reduz]

@clayjohn ok, added the missing fog one. Should probably be good to merge once CI passes.