Use dual-source blending for rendering the sky #17672
Conversation
ecoskey
added
A-Rendering
|
@mate-h this one is for you :) |
|
I've had an initial look at this, and using dual source blending is a great idea, the diff looks good to me. I went to test things out and noticed some inaccuracies between what I expect to see ground truth, vs the results seem to be shifting colors in the rendered geometry toward orange. I pushed all my testing code to the https://github.com/mate-h/bevy/tree/atmosphere-test branch and will keep working on that. 
|
thanks @mate-h for more debugging help and your fancy reference raymarcher!
|
After this last iteration and more thorough testing we were able to identify the source of the issue. I've tested both cases of the geometry above and below the horizon and the math checks out compared to the raymarched reference. Test case for when the ray does not intersect the ground (else block): |
| @@ -118,6 +118,23 @@ fn sample_transmittance_lut(r: f32, mu: f32) -> vec3<f32> { | |||
| return textureSampleLevel(transmittance_lut, transmittance_lut_sampler, uv, 0.0).rgb; | |||
| } | |||
|
|
|||
| //should be in bruneton_functions, but wouldn't work in that module bc imports. What to do wrt licensing? | |||
There was a problem hiding this comment.
I think we should just leave a comment saying that the license in the bruneton_functions.wgsl file applies to this function, or come up with some way to reorganize the imports. Also just in general, since the code is not exaclty the same as bruneton's verison, it is a WGSL re-write, not sure how the licensing works because I am no legal expert. but does it still apply?
There was a problem hiding this comment.
If it's a direct port with no significant practical changes then I'd say the license still applies.
| return sample_aerial_view_lut(in.uv, depth); | ||
| let t = ndc_to_camera_dist(vec3(uv_to_ndc(in.uv), depth)); | ||
| inscattering = sample_aerial_view_lut(in.uv, t); | ||
| transmittance = sample_transmittance_lut_segment(r, mu, t); |
There was a problem hiding this comment.
Just a note on performance, we should start profiling this since we are now taking 3 texture samples per fragment here to determine both inscattering and transmittance, as opposed to 1 that we used before. Although it seems negligible.
There was a problem hiding this comment.
Compared to full per-pixel raymarching I'm not too concerned. If it ends up being an issue we can always add a fallback to the old way
|
@alice-i-cecile this one is ready for final review and merge when you get a chance! |
|
Needs a second rendering review :) |
| @@ -118,6 +118,23 @@ fn sample_transmittance_lut(r: f32, mu: f32) -> vec3<f32> { | |||
| return textureSampleLevel(transmittance_lut, transmittance_lut_sampler, uv, 0.0).rgb; | |||
| } | |||
|
|
|||
| //should be in bruneton_functions, but wouldn't work in that module bc imports. What to do wrt licensing? | |||
There was a problem hiding this comment.
This needs to be resolved before we can merge. I'd leave a little comment here.
| @@ -130,23 +147,31 @@ fn sample_sky_view_lut(r: f32, ray_dir_as: vec3<f32>) -> vec3<f32> { | |||
| return textureSampleLevel(sky_view_lut, sky_view_lut_sampler, uv, 0.0).rgb; | |||
| } | |||
|
|
|||
| fn ndc_to_camera_dist(ndc: vec3<f32>) -> f32 { | |||
There was a problem hiding this comment.
Note here that there is also a bevy PBR builtin for this bevy_pbr::view_transformations::depth_ndc_to_view_z and then the return value could be just multiplied by settings.scene_units_to_m Also seems to consider projection in that function.
There was a problem hiding this comment.
I believe that function only returns the "parallel" distance from the camera to the object, i.e. |P - C| cosθ where θ is the angle between P - C and the camera forward vector. So we could use it, but we'd have to divide out that cosθ term, which isn't too difficult. There's also the issue of bindings, so we'd still have to copy the function into the local shaders
There was a problem hiding this comment.
Oh true I didn't think about the bindings, scratch that :D
Objective
Since previously we only had the alpha channel available, we stored the mean of the transmittance in the aerial view lut, resulting in a grayer fog than should be expected.
Solution
render_skywith two samples from the transmittance lut