1// Inputs a depth texture and outputs a MIP-chain of depths.
2//
3// Because SSAO's performance is bound by texture reads, this increases
4// performance over using the full resolution depth for every sample.
5
6// Reference: https://research.nvidia.com/sites/default/files/pubs/2012-06_Scalable-Ambient-Obscurance/McGuire12SAO.pdf, section 2.2
7
8#import bevy_render::view::View
9
10@group(0) @binding(0) var input_depth: texture_depth_2d;
11@group(0) @binding(1) var preprocessed_depth_mip0: texture_storage_2d<r16float, write>;
12@group(0) @binding(2) var preprocessed_depth_mip1: texture_storage_2d<r16float, write>;
13@group(0) @binding(3) var preprocessed_depth_mip2: texture_storage_2d<r16float, write>;
14@group(0) @binding(4) var preprocessed_depth_mip3: texture_storage_2d<r16float, write>;
15@group(0) @binding(5) var preprocessed_depth_mip4: texture_storage_2d<r16float, write>;
16@group(1) @binding(0) var point_clamp_sampler: sampler;
17@group(1) @binding(1) var linear_clamp_sampler: sampler;
18@group(1) @binding(2) var<uniform> view: View;
19
20
21// Using 4 depths from the previous MIP, compute a weighted average for the depth of the current MIP
22fn weighted_average(depth0: f32, depth1: f32, depth2: f32, depth3: f32) -> f32 {
23    let depth_range_scale_factor = 0.75;
24    let effect_radius = depth_range_scale_factor * 0.5 * 1.457;
25    let falloff_range = 0.615 * effect_radius;
26    let falloff_from = effect_radius * (1.0 - 0.615);
27    let falloff_mul = -1.0 / falloff_range;
28    let falloff_add = falloff_from / falloff_range + 1.0;
29
30    let min_depth = min(min(depth0, depth1), min(depth2, depth3));
31    let weight0 = saturate((depth0 - min_depth) * falloff_mul + falloff_add);
32    let weight1 = saturate((depth1 - min_depth) * falloff_mul + falloff_add);
33    let weight2 = saturate((depth2 - min_depth) * falloff_mul + falloff_add);
34    let weight3 = saturate((depth3 - min_depth) * falloff_mul + falloff_add);
35    let weight_total = weight0 + weight1 + weight2 + weight3;
36
37    return ((weight0 * depth0) + (weight1 * depth1) + (weight2 * depth2) + (weight3 * depth3)) / weight_total;
38}
39
40// Used to share the depths from the previous MIP level between all invocations in a workgroup
41var<workgroup> previous_mip_depth: array<array<f32, 8>, 8>;
42
43@compute
44@workgroup_size(8, 8, 1)
45fn preprocess_depth(@builtin(global_invocation_id) global_id: vec3<u32>, @builtin(local_invocation_id) local_id: vec3<u32>) {
46    let base_coordinates = vec2<i32>(global_id.xy);
47
48    // MIP 0 - Copy 4 texels from the input depth (per invocation, 8x8 invocations per workgroup)
49    let pixel_coordinates0 = base_coordinates * 2i;
50    let pixel_coordinates1 = pixel_coordinates0 + vec2<i32>(1i, 0i);
51    let pixel_coordinates2 = pixel_coordinates0 + vec2<i32>(0i, 1i);
52    let pixel_coordinates3 = pixel_coordinates0 + vec2<i32>(1i, 1i);
53    let depths_uv = vec2<f32>(pixel_coordinates0) / view.viewport.zw;
54    let depths = textureGather(0, input_depth, point_clamp_sampler, depths_uv, vec2<i32>(1i, 1i));
55    textureStore(preprocessed_depth_mip0, pixel_coordinates0, vec4<f32>(depths.w, 0.0, 0.0, 0.0));
56    textureStore(preprocessed_depth_mip0, pixel_coordinates1, vec4<f32>(depths.z, 0.0, 0.0, 0.0));
57    textureStore(preprocessed_depth_mip0, pixel_coordinates2, vec4<f32>(depths.x, 0.0, 0.0, 0.0));
58    textureStore(preprocessed_depth_mip0, pixel_coordinates3, vec4<f32>(depths.y, 0.0, 0.0, 0.0));
59
60    // MIP 1 - Weighted average of MIP 0's depth values (per invocation, 8x8 invocations per workgroup)
61    let depth_mip1 = weighted_average(depths.w, depths.z, depths.x, depths.y);
62    textureStore(preprocessed_depth_mip1, base_coordinates, vec4<f32>(depth_mip1, 0.0, 0.0, 0.0));
63    previous_mip_depth[local_id.x][local_id.y] = depth_mip1;
64
65    workgroupBarrier();
66
67    // MIP 2 - Weighted average of MIP 1's depth values (per invocation, 4x4 invocations per workgroup)
68    if all(local_id.xy % vec2<u32>(2u) == vec2<u32>(0u)) {
69        let depth0 = previous_mip_depth[local_id.x + 0u][local_id.y + 0u];
70        let depth1 = previous_mip_depth[local_id.x + 1u][local_id.y + 0u];
71        let depth2 = previous_mip_depth[local_id.x + 0u][local_id.y + 1u];
72        let depth3 = previous_mip_depth[local_id.x + 1u][local_id.y + 1u];
73        let depth_mip2 = weighted_average(depth0, depth1, depth2, depth3);
74        textureStore(preprocessed_depth_mip2, base_coordinates / 2i, vec4<f32>(depth_mip2, 0.0, 0.0, 0.0));
75        previous_mip_depth[local_id.x][local_id.y] = depth_mip2;
76    }
77
78    workgroupBarrier();
79
80    // MIP 3 - Weighted average of MIP 2's depth values (per invocation, 2x2 invocations per workgroup)
81    if all(local_id.xy % vec2<u32>(4u) == vec2<u32>(0u)) {
82        let depth0 = previous_mip_depth[local_id.x + 0u][local_id.y + 0u];
83        let depth1 = previous_mip_depth[local_id.x + 2u][local_id.y + 0u];
84        let depth2 = previous_mip_depth[local_id.x + 0u][local_id.y + 2u];
85        let depth3 = previous_mip_depth[local_id.x + 2u][local_id.y + 2u];
86        let depth_mip3 = weighted_average(depth0, depth1, depth2, depth3);
87        textureStore(preprocessed_depth_mip3, base_coordinates / 4i, vec4<f32>(depth_mip3, 0.0, 0.0, 0.0));
88        previous_mip_depth[local_id.x][local_id.y] = depth_mip3;
89    }
90
91    workgroupBarrier();
92
93    // MIP 4 - Weighted average of MIP 3's depth values (per invocation, 1 invocation per workgroup)
94    if all(local_id.xy % vec2<u32>(8u) == vec2<u32>(0u)) {
95        let depth0 = previous_mip_depth[local_id.x + 0u][local_id.y + 0u];
96        let depth1 = previous_mip_depth[local_id.x + 4u][local_id.y + 0u];
97        let depth2 = previous_mip_depth[local_id.x + 0u][local_id.y + 4u];
98        let depth3 = previous_mip_depth[local_id.x + 4u][local_id.y + 4u];
99        let depth_mip4 = weighted_average(depth0, depth1, depth2, depth3);
100        textureStore(preprocessed_depth_mip4, base_coordinates / 8i, vec4<f32>(depth_mip4, 0.0, 0.0, 0.0));
101    }
102}