WIP: Atmosphere PBR support #10

mate-h · 2025-07-27T07:39:06Z

Work in progress pull request for rebasing bevyengine#19037 onto Bevy main branch and address some feedback.

crates/bevy_pbr/src/render/light.rs

crates/bevy_pbr/src/atmosphere/functions.wgsl

mate-h · 2025-07-28T05:49:11Z

atlv24 · 2025-07-28T13:39:12Z

black pixels repro on main
why is function called setup_camera_fog in atmosphere example?

mate-h · 2025-08-01T08:00:43Z

Reusing the shadow functions requires significant changes, my initial approach completely failed when I attempted to pass the directional light as a reference / pointer to the wgsl functions. still the shadow samplers bindings are conflicting, and without some hacky work arounds like shader defs or simply copying things, not the right approach.

My second approach is this PR #11 which could work in theory, but is still a large refactoring change and probably out of scope for just the PBR integration of the atmosphere.

atlv24 · 2025-08-01T23:42:15Z

crates/bevy_light/src/probe.rs

+    pub rotation: Quat,
+    /// Whether the diffuse contribution should affect meshes that already have lightmaps.
+    pub affects_lightmapped_mesh_diffuse: bool,
+    /// Cubemap resolution in pixels (must be a power-of-two).


Is POT-ness verified/asserted anywhere?

Nope, good idea to add that assert

assert is panic, hold off on this for now, theres a pattern we gotta make for this

atlv24 · 2025-08-01T23:50:14Z

crates/bevy_light/src/probe.rs

+pub struct AtmosphereEnvironmentMapLight {
+    /// Luminance multiplier in cd/m².
+    pub intensity: f32,
+    /// World-space rotation applied to the generated cubemap.


what is this for? is there a practical application, or is it just cus you can? if i set this to like, 45 degrees around X, will the sky be sideways, or will the pixelation be sideways (but the sky still the right way)

Not needed, let's remove this. The probe should always follow the orientation of the light being received from the atmosphere. So it should be relative to that and not relative to anything else.

I think I was merely copying the same attributes as the GeneratedEnvironmentMapLight

atlv24 · 2025-08-01T23:52:30Z

crates/bevy_pbr/src/atmosphere/bruneton_functions.wgsl

+fn transmittance_lut_r_mu_to_uv(atm: Atmosphere, r: f32, mu: f32) -> vec2<f32> {
  // Distance along a horizontal ray from the ground to the top atmosphere boundary
-    let H = sqrt(atmosphere.top_radius * atmosphere.top_radius - atmosphere.bottom_radius * atmosphere.bottom_radius);
+    let H = sqrt(atm.top_radius * atm.top_radius - atm.bottom_radius * atm.bottom_radius);


is this atm now to avoid name conflict with a binding of the same name? i prefer atmosphere

Yes that's right I was getting name conflicts. Any tips on how to keep it atmosphere ?I would also prefer that :D

i think putting this function in a file that doesnt import the binding should work, not sure though

atlv24 · 2025-08-01T23:55:17Z

crates/bevy_pbr/src/atmosphere/mod.rs

+        origin: Vec3::new(0.0, 3_389_500.0, 0.0),
+        ground_albedo: Vec3::splat(0.1),
+        rayleigh_density_exp_scale: 1.0 / 10_430.0,
+        rayleigh_scattering: Vec3::new(0.019918e-3, 0.01357e-3, 0.00575e-3),


why are these using a mix of both e-3 and 0.00xx? why not 1.9918e-5 etc

The values are copied from scientific papers so I kept the units roughly the same as cited in the paper. While we could shorten these, it does degrade the readability and make them seem like arbitrarily-chosen constants. Maybe it would be helpful to add references.

lets keep them yeah and add reference

atlv24 · 2025-08-01T23:57:01Z

crates/bevy_pbr/src/atmosphere/node.rs

+                env_map_light.size.x / 8,
+                env_map_light.size.y / 8,


are we relying on this being divisible by 8? i usually write div_ceil here. iirc theres a check in the shader to discard threads that are outside the useful range, is that path ever hit

Good point, should add an assert for it, and stress test it with different POT values for the env map size.

atlv24 · 2025-08-01T23:57:51Z

crates/bevy_pbr/src/atmosphere/render_sky.wgsl

    }
+
+    // exposure compensation
+    inscattering *= view.exposure;


is this physically based or is it a "look good" factor? what is it compensating for

Previously the exposure multiplication was inside of the functions, see other comment
Exposure compensation is needed for compositing the atmosphere over the rendered scene both before/after these changes.

crates/bevy_pbr/src/atmosphere/render_sky.wgsl

atlv24 · 2025-08-01T23:59:23Z

crates/bevy_pbr/src/atmosphere/resources.rs

                    (1, uniform_buffer::<AtmosphereSettings>(true)),
                    (2, uniform_buffer::<AtmosphereTransform>(true)),
                    (3, uniform_buffer::<ViewUniform>(true)),
                    (4, uniform_buffer::<GpuLights>(true)),


this can be ::sequential now

mate-h · 2025-08-01T23:59:31Z

crates/bevy_pbr/src/atmosphere/functions.wgsl


        inscattering += (*light).color.rgb * (scattering_factor + multiscattering_factor);
    }
-    return inscattering * view.exposure;


Removed exposure here @atlv24

atlv24 · 2025-08-01T23:59:53Z

crates/bevy_pbr/src/atmosphere/resources.rs

                    (1, uniform_buffer::<AtmosphereSettings>(true)),
                    (2, uniform_buffer::<AtmosphereTransform>(true)),
                    (3, uniform_buffer::<ViewUniform>(true)),
                    (4, uniform_buffer::<GpuLights>(true)),


same here, ::sequential

For the sequential, I've thought about cleaning this up, yet another separate, small PR

atlv24 · 2025-08-02T00:00:59Z

crates/bevy_pbr/src/atmosphere/resources.rs

                (1, settings_binding.clone()),
                (2, transforms_binding.clone()),
                (3, view_binding.clone()),
                (4, lights_binding.clone()),


Co-authored-by: atlv <[email protected]>

crates/bevy_pbr/src/atmosphere/sky_view_lut.wgsl

atlv24 · 2025-08-02T00:02:36Z

crates/bevy_pbr/src/atmosphere/sky_view_lut.wgsl

-    let r = view_radius();
+    // let world_pos = vec3<f32>(0.0, 0.0, 0.0); // get_view_position();
+    let r = view_radius_constant();
+    let world_pos = vec3<f32>(0.0, r, 0.0);


should this be get_view_position?

Good question, I have a long answer to this. No, the sky view lut calculation assumes a non-spherical coordinate system because of the way it is parameterized. This is a limitation of this optimization technique. The result is that the horizon may be out of alignment for renderings that rely on the sky view lut. at the present we use it for environment maps, and if the raymarching is turned on this out of alignment of the horizon is not noticeable at high altitudes. That's thanks to the spherical coordinate system that the raymarched rendering introduces.

atlv24 · 2025-08-02T00:03:34Z

examples/3d/atmosphere.rs

        AtmosphereSettings {
            aerial_view_lut_max_distance: 3.2e5,
            scene_units_to_m: 1e+4,
+            // set this to Raymarching for high quality volumetric shadows


can we have a runtime toggle for this? maybe pressing R or something

Yes good idea, will add that.

Co-authored-by: atlv <[email protected]>

mate-h · 2025-08-02T00:20:51Z

After some discussion on Discord it's best if I split this into multiple smaller PRs:

add SunLight to bevy_light
add origin to atmosphere settings
spherical coordinate changes
PBR directional light support
environment map
raymarched rendering
martian atmosphere

# Objective - Add generated environment lighting to the atmosphere for reflections and diffuse - Tracking issue: #20374 ## Solution - created a new atmosphere node type called environment - up-sampling the existing sky-view lookup texture and tied to the current view only. this atmosphere cube map pass has negligible performance impact, however the filtering pipeline does have a small performance impact to the example. - using the new filtering pipeline, creates mip chain for different roughness levels as well as diffuse ## Testing - ran the atmosphere example --- ## Showcase <img width="1281" height="752" alt="Screenshot 2025-08-12 at 12 19 08 PM" src="https://github.com/user-attachments/assets/8f00ff25-5f48-4c51-b67e-abcbf421abc4" /> ```rs commands.spawn(( Camera3d::default(), // ... // Renders the atmosphere to the environment map from the perspective of the camera AtmosphereEnvironmentMapLight::default(), )); ``` ## Limitations, out of scope - The generation does not support light probes (yet). This allows to render the atmosphere as a light probe from any "location" within the scene and within the overall atmosphere. - therefore we assume that the relative scale of the scene to the atmosphere are orders of magnitude different , this is a safe assumption now with the current impl not officially supporting space views, yet - the PBR directional light is still unaffected by the atmosphere (not tinted by it) out of scope for this PR - unrelated issue to this PR: small black pixels around the fully reflective sphere. narrowed it down, this is a problem in the inscattering calculation based on the depth value in the render sky shader. however the addition of the env map light makes this problem more "apparent". to be addressed separately. - past work staged for further PRs, up next: mate-h#10

atlv24 reviewed Jul 28, 2025

View reviewed changes

crates/bevy_pbr/src/render/light.rs Outdated Show resolved Hide resolved

atlv24 reviewed Jul 28, 2025

View reviewed changes

crates/bevy_pbr/src/atmosphere/functions.wgsl Outdated Show resolved Hide resolved

mate-h force-pushed the atmosphere-pbr branch from a9559f9 to 1f5fd6e Compare July 28, 2025 04:20

mate-h changed the title ~~Atmosphere pbr~~ WIP: Atmosphere PBR support Jul 28, 2025

mate-h added 7 commits July 29, 2025 00:27

Rebase atmosphere pbr changes

a7d7c8e

Fix missing bindings and shader errors

600cbe1

Fix runtime error and formatting

81d6b40

Radiometric units

b1ce73f

Extract global transform and simplify to unwrap or default

b48003d

Remove divide by w

3108faf

Shuffle and reuse wgsl functions

a7a65ac

mate-h force-pushed the atmosphere-pbr branch from 3ce565a to a7a65ac Compare July 29, 2025 07:27

mate-h added 2 commits July 29, 2025 16:49

Add repeating sampler to bluenoise texture

2eaaf84

Move Atmosphere envmap light to bevy_light and relax query filters

a7ad0ca

mate-h force-pushed the atmosphere-pbr branch from 401b2d6 to a7ad0ca Compare August 1, 2025 07:56

Fix CI formatting

969418d

atlv24 reviewed Aug 1, 2025

View reviewed changes

crates/bevy_pbr/src/atmosphere/render_sky.wgsl Outdated Show resolved Hide resolved

atlv24 reviewed Aug 1, 2025

View reviewed changes

mate-h commented Aug 1, 2025

View reviewed changes

atlv24 reviewed Aug 1, 2025

View reviewed changes

atlv24 reviewed Aug 2, 2025

View reviewed changes

Update crates/bevy_pbr/src/atmosphere/render_sky.wgsl

d060aa7

Co-authored-by: atlv <[email protected]>

atlv24 reviewed Aug 2, 2025

View reviewed changes

crates/bevy_pbr/src/atmosphere/sky_view_lut.wgsl Outdated Show resolved Hide resolved

atlv24 reviewed Aug 2, 2025

View reviewed changes

Update crates/bevy_pbr/src/atmosphere/sky_view_lut.wgsl

73f49f3

Co-authored-by: atlv <[email protected]>

Remove rotation from env map light

975dd97

This was referenced Aug 2, 2025

Procedural Atmosphere PBR integration bevyengine/bevy#19037

Closed

Atmosphere PBR support and volumetric shadows tracking issue bevyengine/bevy#20374

Open

Add bevy_light::SunDisk bevyengine/bevy#20434

Merged

mate-h mentioned this pull request Aug 12, 2025

Atmosphere generated environment map lighting bevyengine/bevy#20529

Merged

mate-h closed this Oct 23, 2025

WIP: Atmosphere PBR support #10

WIP: Atmosphere PBR support #10

Uh oh!

Conversation

mate-h commented Jul 27, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mate-h commented Jul 28, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

atlv24 commented Jul 28, 2025

Uh oh!

mate-h commented Aug 1, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mate-h Aug 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

mate-h commented Aug 2, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

mate-h commented Jul 27, 2025 •

edited

Loading

mate-h commented Jul 28, 2025 •

edited

Loading

mate-h Aug 1, 2025 •

edited

Loading