3d clouds

[DanielPSchenk]

Fully 3D clouds on a global scale. Built on ideas presented by Andrew Schneider at SIGGRAPH advances in real-time-graphics workshops. Adaped for cloud volumes the size of the earth.

The new clouds are fully 3D. To switch to the new cloud model toggle Advanced settings->Atmosphere->Advanced clouds.

Because ray marching is performed at a rather high resolution with secondary rays, the new system is really heavy on the GPU.

For designing cloudscapes, the new model uses additional textures. Defaults are provided and the configuration templates were modified to include those defaults.

One error occurs when building, but it is not caused by the new code
/cosmoscout-vr/src/cs-graphics/LuminanceMipMap.cpp(28,19): error C2065: "GL_SUBGROUP_SIZE_KHR" undeclared identifier

[Schneegans]

Thanks a ton! I'll look into it. Regarding the GL_SUBGROUP_SIZE_KHR error, you need to delete your build/*-externals-Release/glew directory and rebuild the externals. The glew version was updated recently.

[Schneegans]

I just played with moving the underlying cloud texture with time. At local scale, this actually gives quite a convincing effect! Maybe we could add this as an option in the future.

Screencast.from.2025-04-09.09-05-36.mp4

Screencast.from.2025-04-09.09-14-00.mp4

[Schneegans]

Thanks again! While maybe not looking great for every vantage point, there are definitely perspectives where the clouds look pretty awesome! And on my 4070 they work quite decently. Before I look into the details, here are some high-level observations:

• Could you remove all the debug logging? The read top texture at ... even causes a crash if the texture is not defined
• I think the old clouds do not work if no cloudTypeTexture is set.
• Could you add some notes to the README what the cloudTypeTexture contains?
• I think it could be beneficial to move all the cloud shader code to a separate file at this point. What do you think? I mean it's just very few methods which are called by the main.

[DanielPSchenk]

@Schneegans Are you changing the 2D or 3D noise or both?
About the points you brought up

1. sure, that should be removed
2. that's an oversight on my side, will fix this early next week
3. I can write a small documentation on what the texture does
4. that would be possible. GetCloudColor and everything it calls could be moved relatively easily, I will look into this possiblity next week
   Good to hear that it works well on your machine

[Schneegans]

@Schneegans Are you changing the 2D or 3D noise or both?
About the points you brought up

Neither, just the uCloudTexture. All I did was replacing

float density = remap(textureLod(uCloudTexture, texCoords, 0).r, 0, CLOUD_COVER_MAX, 0, 1);

with

float density = remap(textureLod(uCloudTexture, texCoords + vec2(uTime * 0.00002, 0.0), 0).r, 0, CLOUD_COVER_MAX, 0, 1);

[DanielPSchenk]

I changed Atmosphere.cpp to only block the use of the default cloud model when no cloud texture is present. The advanced cloud model only works if both cloud texture and cloud type texture are present. There are now warnings when a cloud model is activated but the required texture was not specified.

Also removed excessive debug printing of all the uniforms.

[DanielPSchenk]

I wrote a short documentation of what the different channels of the cloudType texture do

[coveralls]

Pull Request Test Coverage Report for Build 17213558452

Warning: This coverage report may be inaccurate.

This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.

• For more information on this, see Tracking coverage changes with pull request builds.
• To avoid this issue with future PRs, see these Recommended CI Configurations.
• For a quick fix, rebase this PR at GitHub. Your next report should be accurate.

Details

• 0 of 287 (0.0%) changed or added relevant lines in 4 files are covered.
• 359 unchanged lines in 9 files lost coverage.
• Overall coverage decreased (-0.02%) to 1.17%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
plugins/csp-web-api/src/Plugin.cpp	0	3	0.0%
plugins/csp-atmospheres/src/TileableVolumeNoise.cpp	0	30	0.0%
plugins/csp-atmospheres/src/Atmosphere.cpp	0	104	0.0%
plugins/csp-atmospheres/src/Plugin.cpp	0	150	0.0%

Files with Coverage Reduction	New Missed Lines	%
plugins/csp-sharad/src/Plugin.cpp	1	0.0%
plugins/csp-atmospheres/src/Atmosphere.cpp	2	0.0%
plugins/csp-atmospheres/src/Plugin.cpp	2	0.0%
src/cs-utils/DefaultProperty.hpp	3	0.0%
plugins/csp-lod-bodies/src/PlanetShader.cpp	20	0.0%
src/cs-core/Settings.hpp	39	0.0%
plugins/csp-simple-bodies/src/SimpleBody.cpp	57	0.0%
src/cs-core/Settings.cpp	114	0.0%
plugins/csp-lod-bodies/src/Plugin.cpp	121	0.0%

Totals
Change from base Build 14053686187:	-0.02%
Covered Lines:	201
Relevant Lines:	17175

---

💛 - Coveralls

[DanielPSchenk]

Cloud System Rework V2

I again reworked he cloud system to now fully feature the procedural noise-based techniques from the Nubis cubed talked (SIGGRAPH 2022). Cloud prior density is now purely read from a global 2D texture and the cloud type is completely based on a 2D noise. This enables more complex multi-layered clouds than before.

Clouds are denser now, with the density being a function of height and cloud type. This allows for cirrus clouds and high altitude fog.

Anisotropic noises are still missing but I currently do not have the capcacity to implement them soon.

There are lots of quality of life improvements, for example all the most relevant parameters of the cloud model can be set through the GUI and WebAPI.

For example the distribution of cloud types can be biased towards extremely tall clouds.

The cloud type texture is now easier to customize than before because it is automatically reloaded when the shader is compiled (causing a slight performance penalty but allowing for much faster iteration times).
There are new global quality settings that are accessible through the GUI and WebAPI: a multiplier on all step sizes, the maximum number of steps taken during integration is exposed now and the sample jittering used to break up banding artifacts can be easily tuned.

The ray marching clouds blend naturally with the environment.

Performance is bad at high quality on a powerful laptop, but using the quality slider and by turning down sample jittering, stable (if banding-prone) results with good frame rates can be achieved.

low quality, high performance

high quality, low performance

An obvious next direction for performance would be half-resolution rendering, even very high quality runs at a stable 30fps+ on a high-end Ampere generation laptop if the resolution is only 960x560.