godot/modules/lightmapper_rd/lm_compute.glsl

#[versions]

primary = "#define MODE_DIRECT_LIGHT";
secondary = "#define MODE_BOUNCE_LIGHT";
dilate = "#define MODE_DILATE";
unocclude = "#define MODE_UNOCCLUDE";
light_probes = "#define MODE_LIGHT_PROBES";
denoise = "#define MODE_DENOISE";
pack_coeffs = "#define MODE_PACK_L1_COEFFS";

#[compute]

#version 450

#VERSION_DEFINES

#extension GL_EXT_samplerless_texture_functions : enable

// One 2D local group focusing in one layer at a time, though all
// in parallel (no barriers) makes more sense than a 3D local group
// as this can take more advantage of the cache for each group.

#ifdef MODE_LIGHT_PROBES

layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in;

#else

layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;

#endif

#include "lm_common_inc.glsl"

#ifdef MODE_LIGHT_PROBES

layout(set = 1, binding = 0, std430) restrict buffer LightProbeData {
	vec4 data[];
}
light_probes;

layout(set = 1, binding = 1) uniform texture2DArray source_light;
layout(set = 1, binding = 2) uniform texture2D environment;
#endif

#ifdef MODE_UNOCCLUDE

layout(rgba32f, set = 1, binding = 0) uniform restrict image2DArray position;
layout(rgba32f, set = 1, binding = 1) uniform restrict readonly image2DArray unocclude;

#endif

#if defined(MODE_DIRECT_LIGHT) || defined(MODE_BOUNCE_LIGHT)

layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly image2DArray dest_light;
layout(set = 1, binding = 1) uniform texture2DArray source_light;
layout(set = 1, binding = 2) uniform texture2DArray source_position;
layout(set = 1, binding = 3) uniform texture2DArray source_normal;
layout(rgba16f, set = 1, binding = 4) uniform restrict image2DArray accum_light;

#endif

#if defined(MODE_DIRECT_LIGHT) && defined(USE_SHADOWMASK)
layout(rgba8, set = 1, binding = 5) uniform restrict writeonly image2DArray shadowmask;
#elif defined(MODE_BOUNCE_LIGHT)
layout(set = 1, binding = 5) uniform texture2D environment;
#endif

#if defined(MODE_DILATE) || defined(MODE_DENOISE) || defined(MODE_PACK_L1_COEFFS)
layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly image2DArray dest_light;
layout(set = 1, binding = 1) uniform texture2DArray source_light;
#endif

#ifdef MODE_DENOISE
layout(set = 1, binding = 2) uniform texture2DArray source_normal;
layout(set = 1, binding = 3) uniform DenoiseParams {
	float spatial_bandwidth;
	float light_bandwidth;
	float albedo_bandwidth;
	float normal_bandwidth;

	int half_search_window;
	float filter_strength;
}
denoise_params;
#endif

layout(push_constant, std430) uniform Params {
	uint atlas_slice;
	uint ray_count;
	uint ray_from;
	uint ray_to;

	ivec2 region_ofs;
	uint probe_count;
}
params;

//check it, but also return distance and barycentric coords (for uv lookup)
bool ray_hits_triangle(vec3 from, vec3 dir, float max_dist, vec3 p0, vec3 p1, vec3 p2, out float r_distance, out vec3 r_barycentric) {
	const float EPSILON = 0.00001;
	const vec3 e0 = p1 - p0;
	const vec3 e1 = p0 - p2;
	vec3 triangle_normal = cross(e1, e0);

	float n_dot_dir = dot(triangle_normal, dir);

	if (abs(n_dot_dir) < EPSILON) {
		return false;
	}

	const vec3 e2 = (p0 - from) / n_dot_dir;
	const vec3 i = cross(dir, e2);

	r_barycentric.y = dot(i, e1);
	r_barycentric.z = dot(i, e0);
	r_barycentric.x = 1.0 - (r_barycentric.z + r_barycentric.y);
	r_distance = dot(triangle_normal, e2);

	return (r_distance > bake_params.bias) && (r_distance < max_dist) && all(greaterThanEqual(r_barycentric, vec3(0.0)));
}

const uint RAY_MISS = 0;
const uint RAY_FRONT = 1;
const uint RAY_BACK = 2;
const uint RAY_ANY = 3;

bool ray_box_test(vec3 p_from, vec3 p_inv_dir, vec3 p_box_min, vec3 p_box_max) {
	vec3 t0 = (p_box_min - p_from) * p_inv_dir;
	vec3 t1 = (p_box_max - p_from) * p_inv_dir;
	vec3 tmin = min(t0, t1), tmax = max(t0, t1);
	return max(tmin.x, max(tmin.y, tmin.z)) <= min(tmax.x, min(tmax.y, tmax.z));
}

#if CLUSTER_SIZE > 32
#define CLUSTER_TRIANGLE_ITERATION
#endif

uint trace_ray(vec3 p_from, vec3 p_to, bool p_any_hit, out float r_distance, out vec3 r_normal, out uint r_triangle, out vec3 r_barycentric) {
	// World coordinates.
	vec3 rel = p_to - p_from;
	float rel_len = length(rel);
	vec3 dir = normalize(rel);
	vec3 inv_dir = 1.0 / dir;

	// Cell coordinates.
	vec3 from_cell = (p_from - bake_params.to_cell_offset) * bake_params.to_cell_size;
	vec3 to_cell = (p_to - bake_params.to_cell_offset) * bake_params.to_cell_size;

	// Prepare DDA.
	vec3 rel_cell = to_cell - from_cell;
	ivec3 icell = ivec3(from_cell);
	ivec3 iendcell = ivec3(to_cell);
	vec3 dir_cell = normalize(rel_cell);
	vec3 delta = min(abs(1.0 / dir_cell), bake_params.grid_size); // Use bake_params.grid_size as max to prevent infinity values.
	ivec3 step = ivec3(sign(rel_cell));
	const vec3 init_next_cell = vec3(icell) + max(vec3(0), sign(step));
	vec3 t_max = mix(vec3(0), (init_next_cell - from_cell) / dir_cell, notEqual(step, vec3(0))); // Distance to next boundary.

	uint iters = 0;
	while (all(greaterThanEqual(icell, ivec3(0))) && all(lessThan(icell, ivec3(bake_params.grid_size))) && (iters < 1000)) {
		uvec2 cell_data = texelFetch(grid, icell, 0).xy;
		uint triangle_count = cell_data.x;
		if (triangle_count > 0) {
			uint hit = RAY_MISS;
			float best_distance = 1e20;
			uint cluster_start = cluster_indices.data[cell_data.y * 2];
			uint cell_triangle_start = cluster_indices.data[cell_data.y * 2 + 1];
			uint cluster_count = (triangle_count + CLUSTER_SIZE - 1) / CLUSTER_SIZE;
			uint cluster_base_index = 0;
			while (cluster_base_index < cluster_count) {
				// To minimize divergence, all Ray-AABB tests on the clusters contained in the cell are performed
				// before checking against the triangles. We do this 32 clusters at a time and store the intersected
				// clusters on each bit of the 32-bit integer.
				uint cluster_test_count = min(32, cluster_count - cluster_base_index);
				uint cluster_hits = 0;
				for (uint i = 0; i < cluster_test_count; i++) {
					uint cluster_index = cluster_start + cluster_base_index + i;
					ClusterAABB cluster_aabb = cluster_aabbs.data[cluster_index];
					if (ray_box_test(p_from, inv_dir, cluster_aabb.min_bounds, cluster_aabb.max_bounds)) {
						cluster_hits |= (1 << i);
					}
				}

				// Check the triangles in any of the clusters that were intersected by toggling off the bits in the
				// 32-bit integer counter until no bits are left.
				while (cluster_hits > 0) {
					uint cluster_index = findLSB(cluster_hits);
					cluster_hits &= ~(1 << cluster_index);
					cluster_index += cluster_base_index;

					// Do the same divergence execution trick with triangles as well.
					uint triangle_base_index = 0;
#ifdef CLUSTER_TRIANGLE_ITERATION
					while (triangle_base_index < triangle_count)
#endif
					{
						uint triangle_start_index = cell_triangle_start + cluster_index * CLUSTER_SIZE + triangle_base_index;
						uint triangle_test_count = min(CLUSTER_SIZE, triangle_count - triangle_base_index);
						uint triangle_hits = 0;
						for (uint i = 0; i < triangle_test_count; i++) {
							uint triangle_index = triangle_indices.data[triangle_start_index + i];
							if (ray_box_test(p_from, inv_dir, triangles.data[triangle_index].min_bounds, triangles.data[triangle_index].max_bounds)) {
								triangle_hits |= (1 << i);
							}
						}

						while (triangle_hits > 0) {
							uint cluster_triangle_index = findLSB(triangle_hits);
							triangle_hits &= ~(1 << cluster_triangle_index);
							cluster_triangle_index += triangle_start_index;

							uint triangle_index = triangle_indices.data[cluster_triangle_index];
							Triangle triangle = triangles.data[triangle_index];

							// Gather the triangle vertex positions.
							vec3 vtx0 = vertices.data[triangle.indices.x].position;
							vec3 vtx1 = vertices.data[triangle.indices.y].position;
							vec3 vtx2 = vertices.data[triangle.indices.z].position;
							vec3 normal = -normalize(cross((vtx0 - vtx1), (vtx0 - vtx2)));
							bool backface = dot(normal, dir) >= 0.0;
							float distance;
							vec3 barycentric;
							if (ray_hits_triangle(p_from, dir, rel_len, vtx0, vtx1, vtx2, distance, barycentric)) {
								if (p_any_hit) {
									// Return early if any hit was requested.
									return RAY_ANY;
								}
								vec3 position = p_from + dir * distance;
								vec3 hit_cell = (position - bake_params.to_cell_offset) * bake_params.to_cell_size;
								if (icell != ivec3(hit_cell)) {
									// It's possible for the ray to hit a triangle in a position outside the bounds of the cell
									// if it's large enough to cover multiple ones. The hit must be ignored if this is the case.
									continue;
								}

								if (!backface) {
									// The case of meshes having both a front and back face in the same plane is more common than
									// expected, so if this is a front-face, bias it closer to the ray origin, so it always wins
									// over the back-face.
									distance = max(bake_params.bias, distance - bake_params.bias);
								}

								if (distance < best_distance) {
									switch (triangle.cull_mode) {
										case CULL_DISABLED:
											backface = false;
											break;
										case CULL_FRONT:
											backface = !backface;
											break;
										case CULL_BACK: // Default behavior.
											break;
									}

									hit = backface ? RAY_BACK : RAY_FRONT;
									best_distance = distance;
									r_distance = distance;
									r_normal = normal;
									r_triangle = triangle_index;
									r_barycentric = barycentric;
								}
							}
						}

#ifdef CLUSTER_TRIANGLE_ITERATION
						triangle_base_index += CLUSTER_SIZE;
#endif
					}
				}

				cluster_base_index += 32;
			}

			if (hit != RAY_MISS) {
				return hit;
			}
		}

		if (icell == iendcell) {
			break;
		}

		// There should be only one axis updated at a time for DDA to work properly.
		if (t_max.x < t_max.y && t_max.x < t_max.z) {
			icell.x += step.x;
			t_max.x += delta.x;
		} else if (t_max.y < t_max.z) {
			icell.y += step.y;
			t_max.y += delta.y;
		} else {
			icell.z += step.z;
			t_max.z += delta.z;
		}
		iters++;
	}

	return RAY_MISS;
}

uint trace_ray_closest_hit_triangle(vec3 p_from, vec3 p_to, out uint r_triangle, out vec3 r_barycentric) {
	float distance;
	vec3 normal;
	return trace_ray(p_from, p_to, false, distance, normal, r_triangle, r_barycentric);
}

uint trace_ray_closest_hit_triangle_albedo_alpha(vec3 p_from, vec3 p_to, out vec4 albedo_alpha, out vec3 hit_position) {
	float distance;
	vec3 normal;
	uint tidx;
	vec3 barycentric;

	uint ret = trace_ray(p_from, p_to, false, distance, normal, tidx, barycentric);
	if (ret != RAY_MISS) {
		Vertex vert0 = vertices.data[triangles.data[tidx].indices.x];
		Vertex vert1 = vertices.data[triangles.data[tidx].indices.y];
		Vertex vert2 = vertices.data[triangles.data[tidx].indices.z];

		vec3 uvw = vec3(barycentric.x * vert0.uv + barycentric.y * vert1.uv + barycentric.z * vert2.uv, float(triangles.data[tidx].slice));

		albedo_alpha = textureLod(sampler2DArray(albedo_tex, linear_sampler), uvw, 0);
		hit_position = barycentric.x * vert0.position + barycentric.y * vert1.position + barycentric.z * vert2.position;
	}

	return ret;
}

uint trace_ray_closest_hit_distance(vec3 p_from, vec3 p_to, out float r_distance, out vec3 r_normal) {
	uint triangle;
	vec3 barycentric;
	return trace_ray(p_from, p_to, false, r_distance, r_normal, triangle, barycentric);
}

uint trace_ray_any_hit(vec3 p_from, vec3 p_to) {
	float distance;
	vec3 normal;
	uint triangle;
	vec3 barycentric;
	return trace_ray(p_from, p_to, true, distance, normal, triangle, barycentric);
}

// https://www.reedbeta.com/blog/hash-functions-for-gpu-rendering/
uint hash(uint value) {
	uint state = value * 747796405u + 2891336453u;
	uint word = ((state >> ((state >> 28u) + 4u)) ^ state) * 277803737u;
	return (word >> 22u) ^ word;
}

uint random_seed(ivec3 seed) {
	return hash(seed.x ^ hash(seed.y ^ hash(seed.z)));
}

// generates a random value in range [0.0, 1.0)
float randomize(inout uint value) {
	value = hash(value);
	return float(value / 4294967296.0);
}

const float PI = 3.14159265f;

// http://www.realtimerendering.com/raytracinggems/unofficial_RayTracingGems_v1.4.pdf (chapter 15)
vec3 generate_hemisphere_cosine_weighted_direction(inout uint noise) {
	float noise1 = randomize(noise);
	float noise2 = randomize(noise) * 2.0 * PI;

	return vec3(sqrt(noise1) * cos(noise2), sqrt(noise1) * sin(noise2), sqrt(1.0 - noise1));
}

// Distribution generation adapted from "Generating uniformly distributed numbers on a sphere"
// <http://corysimon.github.io/articles/uniformdistn-on-sphere/>
vec3 generate_sphere_uniform_direction(inout uint noise) {
	float theta = 2.0 * PI * randomize(noise);
	float phi = acos(1.0 - 2.0 * randomize(noise));
	return vec3(sin(phi) * cos(theta), sin(phi) * sin(theta), cos(phi));
}

vec3 generate_ray_dir_from_normal(vec3 normal, inout uint noise) {
	vec3 v0 = abs(normal.z) < 0.999 ? vec3(0.0, 0.0, 1.0) : vec3(0.0, 1.0, 0.0);
	vec3 tangent = normalize(cross(v0, normal));
	vec3 bitangent = normalize(cross(tangent, normal));
	mat3 normal_mat = mat3(tangent, bitangent, normal);
	return normal_mat * generate_hemisphere_cosine_weighted_direction(noise);
}

#if defined(MODE_DIRECT_LIGHT) || defined(MODE_BOUNCE_LIGHT) || defined(MODE_LIGHT_PROBES)

float get_omni_attenuation(float distance, float inv_range, float decay) {
	float nd = distance * inv_range;
	nd *= nd;
	nd *= nd; // nd^4
	nd = max(1.0 - nd, 0.0);
	nd *= nd; // nd^2
	return nd * pow(max(distance, 0.0001), -decay);
}

const int AA_SAMPLES = 16;

const vec2 halton_map[AA_SAMPLES] = vec2[](
		vec2(0.5, 0.33333333),
		vec2(0.25, 0.66666667),
		vec2(0.75, 0.11111111),
		vec2(0.125, 0.44444444),
		vec2(0.625, 0.77777778),
		vec2(0.375, 0.22222222),
		vec2(0.875, 0.55555556),
		vec2(0.0625, 0.88888889),
		vec2(0.5625, 0.03703704),
		vec2(0.3125, 0.37037037),
		vec2(0.8125, 0.7037037),
		vec2(0.1875, 0.14814815),
		vec2(0.6875, 0.48148148),
		vec2(0.4375, 0.81481481),
		vec2(0.9375, 0.25925926),
		vec2(0.03125, 0.59259259));

vec2 get_vogel_disk(float p_i, float p_rotation, float p_sample_count_sqrt) {
	const float golden_angle = 2.4;

	float r = sqrt(p_i + 0.5) / p_sample_count_sqrt;
	float theta = p_i * golden_angle + p_rotation;

	return vec2(cos(theta), sin(theta)) * r;
}

void trace_direct_light(vec3 p_position, vec3 p_normal, uint p_light_index, bool p_soft_shadowing, out vec3 r_light, out vec3 r_light_dir, inout uint r_noise, float p_texel_size, out float r_shadow) {
	const float EPSILON = 0.00001;

	r_light = vec3(0.0f);
	r_shadow = 0.0f;

	vec3 light_pos;
	float dist;
	float attenuation;
	float soft_shadowing_disk_size;
	Light light_data = lights.data[p_light_index];
	if (light_data.type == LIGHT_TYPE_DIRECTIONAL) {
		vec3 light_vec = light_data.direction;
		light_pos = p_position - light_vec * length(bake_params.world_size);
		r_light_dir = normalize(light_pos - p_position);
		dist = length(bake_params.world_size);
		attenuation = 1.0;
		soft_shadowing_disk_size = light_data.size;
	} else {
		light_pos = light_data.position;
		r_light_dir = normalize(light_pos - p_position);
		dist = distance(p_position, light_pos);
		if (dist > light_data.range) {
			return;
		}

		soft_shadowing_disk_size = light_data.size / dist;

		attenuation = get_omni_attenuation(dist, 1.0 / light_data.range, light_data.attenuation);

		if (light_data.type == LIGHT_TYPE_SPOT) {
			vec3 rel = normalize(p_position - light_pos);
			float cos_spot_angle = light_data.cos_spot_angle;
			float cos_angle = dot(rel, light_data.direction);

			if (cos_angle < cos_spot_angle) {
				return;
			}

			float scos = max(cos_angle, cos_spot_angle);
			float spot_rim = max(0.0001, (1.0 - scos) / (1.0 - cos_spot_angle));
			attenuation *= 1.0 - pow(spot_rim, light_data.inv_spot_attenuation);
		}
	}

	attenuation *= max(0.0, dot(p_normal, r_light_dir));
	if (attenuation <= 0.0001) {
		return;
	}

	float penumbra = 0.0;
	if (p_soft_shadowing) {
		const bool use_soft_shadows = (light_data.size > 0.0);
		const uint ray_count = AA_SAMPLES;
		const uint total_ray_count = use_soft_shadows ? params.ray_count : ray_count;
		const uint shadowing_rays_check_penumbra_denom = 2;
		const uint shadowing_ray_count = max(1, params.ray_count / ray_count);
		const float shadowing_ray_count_sqrt = sqrt(float(total_ray_count));

		// Setup tangent pass to calculate AA samples over the current texel.
		vec3 aux = p_normal.y < 0.777 ? vec3(0.0, 1.0, 0.0) : vec3(1.0, 0.0, 0.0);
		vec3 tangent = normalize(cross(p_normal, aux));
		vec3 bitan = normalize(cross(p_normal, tangent));

		// Setup light tangent pass to calculate samples over disk aligned towards the light
		vec3 light_to_point = -r_light_dir;
		vec3 light_aux = light_to_point.y < 0.777 ? vec3(0.0, 1.0, 0.0) : vec3(1.0, 0.0, 0.0);
		vec3 light_to_point_tan = normalize(cross(light_to_point, light_aux));
		vec3 light_to_point_bitan = normalize(cross(light_to_point, light_to_point_tan));

		uint hits = 0;
		float aa_power = 0.0;
		for (uint i = 0; i < ray_count; i++) {
			// Create a random sample within the texel.
			vec2 disk_sample = (halton_map[i] - vec2(0.5)) * p_texel_size * light_data.shadow_blur;
			// Align the sample to world space.
			vec3 disk_aligned = (disk_sample.x * tangent + disk_sample.y * bitan);
			vec3 origin = p_position - disk_aligned;
			vec3 light_dir = normalize(light_pos - origin);

			float power = 0.0;
			uint power_accm = 0;
			vec3 prev_pos = origin;
			if (use_soft_shadows) {
				uint soft_shadow_hits = 0;
				for (uint j = 0; j < shadowing_ray_count; j++) {
					origin = prev_pos;
					// Optimization:
					// Once already traced an important proportion of rays, if all are hits or misses,
					// assume we're not in the penumbra so we can infer the rest would have the same result.
					if (j == shadowing_ray_count / shadowing_rays_check_penumbra_denom) {
						if (soft_shadow_hits == j) {
							// Assume totally lit
							soft_shadow_hits = shadowing_ray_count;
							break;
						} else if (soft_shadow_hits == 0) {
							// Assume totally dark
							soft_shadow_hits = 0;
							break;
						}
					}

					float a = randomize(r_noise) * 2.0 * PI;
					float vogel_index = float(total_ray_count - 1 - (i * shadowing_ray_count + j)); // Start from (total_ray_count - 1) so we check the outer points first.
					vec2 light_disk_sample = get_vogel_disk(vogel_index, a, shadowing_ray_count_sqrt) * soft_shadowing_disk_size * light_data.shadow_blur;
					vec3 light_disk_to_point = normalize(light_to_point + light_disk_sample.x * light_to_point_tan + light_disk_sample.y * light_to_point_bitan);
					float sample_penumbra = 0.0;
					bool sample_did_hit = false;

					for (uint iter = 0; iter < bake_params.transparency_rays; iter++) {
						vec4 hit_albedo = vec4(1.0);
						vec3 hit_position;
						// Offset the ray origin for AA, offset the light position for soft shadows.
						uint ret = trace_ray_closest_hit_triangle_albedo_alpha(origin - light_disk_to_point * (bake_params.bias + length(disk_sample)), p_position - light_disk_to_point * dist, hit_albedo, hit_position);
						if (ret == RAY_MISS) {
							if (!sample_did_hit) {
								sample_penumbra = 1.0;
							}
							soft_shadow_hits += 1;
							break;
						} else if (ret == RAY_FRONT || ret == RAY_BACK) {
							bool contribute = ret == RAY_FRONT || !sample_did_hit;
							if (!sample_did_hit) {
								sample_penumbra = 1.0;
								sample_did_hit = true;
							}

							soft_shadow_hits += 1;

							if (contribute) {
								sample_penumbra = max(sample_penumbra - hit_albedo.a - EPSILON, 0.0);
							}
							origin = hit_position + r_light_dir * bake_params.bias;

							if (sample_penumbra - EPSILON <= 0) {
								break;
							}
						}
					}

					power += sample_penumbra;
					power_accm++;
				}

				hits += soft_shadow_hits;
			} else { // No soft shadows.
				float sample_penumbra = 0.0;
				bool sample_did_hit = false;
				for (uint iter = 0; iter < bake_params.transparency_rays; iter++) {
					vec4 hit_albedo = vec4(1.0);
					vec3 hit_position;
					// Offset the ray origin for AA, offset the light position for soft shadows.
					uint ret = trace_ray_closest_hit_triangle_albedo_alpha(origin + light_dir * (bake_params.bias + length(disk_sample)), light_pos, hit_albedo, hit_position);
					if (ret == RAY_MISS) {
						if (!sample_did_hit) {
							sample_penumbra = 1.0;
						}
						hits++;
						break;
					} else if (ret == RAY_FRONT || ret == RAY_BACK) {
						bool contribute = ret == RAY_FRONT || !sample_did_hit;
						if (!sample_did_hit) {
							sample_penumbra = 1.0;
							sample_did_hit = true;
						}

						hits++;

						if (contribute) {
							sample_penumbra = max(sample_penumbra - hit_albedo.a - EPSILON, 0.0);
						}
						origin = hit_position + r_light_dir * bake_params.bias;

						if (sample_penumbra - EPSILON <= 0) {
							break;
						}
					}
				}
				power += sample_penumbra;
				power_accm = 1;
			}
			aa_power = power / float(power_accm);
		}
		penumbra = aa_power;
	} else { // No soft shadows.
		bool did_hit = false;
		penumbra = 0.0;
		for (uint iter = 0; iter < bake_params.transparency_rays; iter++) {
			vec4 hit_albedo = vec4(1.0);
			vec3 hit_position;
			uint ret = trace_ray_closest_hit_triangle_albedo_alpha(p_position + r_light_dir * bake_params.bias, light_pos, hit_albedo, hit_position);
			if (ret == RAY_MISS) {
				if (!did_hit) {
					penumbra = 1.0;
				}
				break;
			} else if (ret == RAY_FRONT || ret == RAY_BACK) {
				bool contribute = (ret == RAY_FRONT || !did_hit);
				if (!did_hit) {
					penumbra = 1.0;
					did_hit = true;
				}

				if (contribute) {
					penumbra = max(penumbra - hit_albedo.a - EPSILON, 0.0);
				}

				p_position = hit_position + r_light_dir * bake_params.bias;

				if (penumbra - EPSILON <= 0) {
					break;
				}
			}
		}

		penumbra = clamp(penumbra, 0.0, 1.0);
	}

	r_shadow = penumbra;
	r_light = light_data.color * light_data.energy * attenuation * penumbra;
}

#endif

#if defined(MODE_BOUNCE_LIGHT) || defined(MODE_LIGHT_PROBES)

vec3 trace_environment_color(vec3 ray_dir) {
	vec3 sky_dir = normalize(mat3(bake_params.env_transform) * ray_dir);
	vec2 st = vec2(atan(sky_dir.x, sky_dir.z), acos(sky_dir.y));
	if (st.x < 0.0) {
		st.x += PI * 2.0;
	}

	return textureLod(sampler2D(environment, linear_sampler), st / vec2(PI * 2.0, PI), 0.0).rgb;
}

vec3 trace_indirect_light(vec3 p_position, vec3 p_ray_dir, inout uint r_noise, float p_texel_size) {
	// The lower limit considers the case where the lightmapper might have bounces disabled but light probes are requested.
	vec3 position = p_position;
	vec3 ray_dir = p_ray_dir;
	uint max_depth = max(bake_params.bounces, 1);
	uint transparency_rays_left = bake_params.transparency_rays;
	vec3 throughput = vec3(1.0);
	vec3 light = vec3(0.0);
	for (uint depth = 0; depth < max_depth; depth++) {
		uint tidx;
		vec3 barycentric;
		uint trace_result = trace_ray_closest_hit_triangle(position + ray_dir * bake_params.bias, position + ray_dir * length(bake_params.world_size), tidx, barycentric);
		if (trace_result == RAY_FRONT) {
			Vertex vert0 = vertices.data[triangles.data[tidx].indices.x];
			Vertex vert1 = vertices.data[triangles.data[tidx].indices.y];
			Vertex vert2 = vertices.data[triangles.data[tidx].indices.z];
			vec3 uvw = vec3(barycentric.x * vert0.uv + barycentric.y * vert1.uv + barycentric.z * vert2.uv, float(triangles.data[tidx].slice));
			position = barycentric.x * vert0.position + barycentric.y * vert1.position + barycentric.z * vert2.position;

			vec3 prev_normal = ray_dir;

			vec3 norm0 = vec3(vert0.normal_xy, vert0.normal_z);
			vec3 norm1 = vec3(vert1.normal_xy, vert1.normal_z);
			vec3 norm2 = vec3(vert2.normal_xy, vert2.normal_z);
			vec3 normal = barycentric.x * norm0 + barycentric.y * norm1 + barycentric.z * norm2;

			vec3 direct_light = vec3(0.0f);
#ifdef USE_LIGHT_TEXTURE_FOR_BOUNCES
			direct_light += textureLod(sampler2DArray(source_light, linear_sampler), uvw, 0.0).rgb;
#else
			// Trace the lights directly. Significantly more expensive but more accurate in scenarios
			// where the lightmap texture isn't reliable.
			for (uint i = 0; i < bake_params.light_count; i++) {
				vec3 light;
				vec3 light_dir;
				float shadow;
				trace_direct_light(position, normal, i, false, light, light_dir, r_noise, p_texel_size, shadow);
				direct_light += light * lights.data[i].indirect_energy;
			}

			direct_light *= bake_params.exposure_normalization;
#endif

			vec4 albedo_alpha = textureLod(sampler2DArray(albedo_tex, linear_sampler), uvw, 0).rgba;
			vec3 emissive = textureLod(sampler2DArray(emission_tex, linear_sampler), uvw, 0).rgb;
			emissive *= bake_params.exposure_normalization;

			light += throughput * emissive * albedo_alpha.a;
			throughput = mix(throughput, throughput * albedo_alpha.rgb, albedo_alpha.a);
			light += throughput * direct_light * bake_params.bounce_indirect_energy * albedo_alpha.a;

			if (albedo_alpha.a < 1.0) {
				transparency_rays_left -= 1;
				depth -= 1;
				if (transparency_rays_left <= 0) {
					break;
				}

				// Either bounce off the transparent surface or keep going forward.
				float pa = albedo_alpha.a * albedo_alpha.a;
				if (randomize(r_noise) > pa) {
					normal = prev_normal;
				}

				position += normal * bake_params.bias;
			}

			// Use Russian Roulette to determine a probability to terminate the bounce earlier as an optimization.
			// <https://computergraphics.stackexchange.com/questions/2316/is-russian-roulette-really-the-answer>
			float p = max(max(throughput.x, throughput.y), throughput.z);
			if (randomize(r_noise) > p) {
				break;
			}

			// Boost the throughput from the probability of the ray being terminated early.
			throughput *= 1.0 / p;

			// Generate a new ray direction for the next bounce from this surface's normal.
			ray_dir = generate_ray_dir_from_normal(normal, r_noise);
		} else if (trace_result == RAY_MISS) {
			// Look for the environment color and stop bouncing.
			light += throughput * trace_environment_color(ray_dir);
			break;
		} else if (trace_result == RAY_BACK) {
			Vertex vert0 = vertices.data[triangles.data[tidx].indices.x];
			Vertex vert1 = vertices.data[triangles.data[tidx].indices.y];
			Vertex vert2 = vertices.data[triangles.data[tidx].indices.z];
			vec3 uvw = vec3(barycentric.x * vert0.uv + barycentric.y * vert1.uv + barycentric.z * vert2.uv, float(triangles.data[tidx].slice));
			position = barycentric.x * vert0.position + barycentric.y * vert1.position + barycentric.z * vert2.position;

			vec4 albedo_alpha = textureLod(sampler2DArray(albedo_tex, linear_sampler), uvw, 0).rgba;

			if (albedo_alpha.a > 1.0) {
				break;
			}

			transparency_rays_left -= 1;
			depth -= 1;
			if (transparency_rays_left <= 0) {
				break;
			}

			vec3 norm0 = vec3(vert0.normal_xy, vert0.normal_z);
			vec3 norm1 = vec3(vert1.normal_xy, vert1.normal_z);
			vec3 norm2 = vec3(vert2.normal_xy, vert2.normal_z);
			vec3 normal = barycentric.x * norm0 + barycentric.y * norm1 + barycentric.z * norm2;

			vec3 direct_light = vec3(0.0f);
#ifdef USE_LIGHT_TEXTURE_FOR_BOUNCES
			direct_light += textureLod(sampler2DArray(source_light, linear_sampler), uvw, 0.0).rgb;
#else
			// Trace the lights directly. Significantly more expensive but more accurate in scenarios
			// where the lightmap texture isn't reliable.
			for (uint i = 0; i < bake_params.light_count; i++) {
				vec3 light;
				vec3 light_dir;
				float shadow;
				trace_direct_light(position, normal, i, false, light, light_dir, r_noise, p_texel_size, shadow);
				direct_light += light * lights.data[i].indirect_energy;
			}

			direct_light *= bake_params.exposure_normalization;
#endif

			vec3 emissive = textureLod(sampler2DArray(emission_tex, linear_sampler), uvw, 0).rgb;
			emissive *= bake_params.exposure_normalization;

			light += throughput * emissive * albedo_alpha.a;
			throughput = mix(mix(throughput, throughput * albedo_alpha.rgb, albedo_alpha.a), vec3(0.0), albedo_alpha.a);
			light += throughput * direct_light * bake_params.bounce_indirect_energy * albedo_alpha.a;

			position += ray_dir * bake_params.bias;
		}
	}

	return light;
}

#endif

void main() {
	// Check if invocation is out of bounds.
#ifdef MODE_LIGHT_PROBES
	int probe_index = int(gl_GlobalInvocationID.x);
	if (probe_index >= params.probe_count) {
		return;
	}

#else
	ivec2 atlas_pos = ivec2(gl_GlobalInvocationID.xy) + params.region_ofs;
	if (any(greaterThanEqual(atlas_pos, bake_params.atlas_size))) {
		return;
	}
#endif

#ifdef MODE_DIRECT_LIGHT
	vec3 normal = texelFetch(sampler2DArray(source_normal, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).xyz;
	if (length(normal) < 0.5) {
		return; //empty texel, no process
	}
	vec3 position = texelFetch(sampler2DArray(source_position, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).xyz;
	vec4 neighbor_position = texelFetch(sampler2DArray(source_position, linear_sampler), ivec3(atlas_pos + ivec2(1, 0), params.atlas_slice), 0).xyzw;

	if (neighbor_position.w < 0.001) {
		// Empty texel, try again.
		neighbor_position.xyz = texelFetch(sampler2DArray(source_position, linear_sampler), ivec3(atlas_pos + ivec2(-1, 0), params.atlas_slice), 0).xyz;
	}
	float texel_size_world_space = distance(position, neighbor_position.xyz);

	vec3 light_for_texture = vec3(0.0);
	vec3 light_for_bounces = vec3(0.0);

#ifdef USE_SHADOWMASK
	float shadowmask_value = 0.0f;
#endif

#ifdef USE_SH_LIGHTMAPS
	vec4 sh_accum[4] = vec4[](
			vec4(0.0, 0.0, 0.0, 1.0),
			vec4(0.0, 0.0, 0.0, 1.0),
			vec4(0.0, 0.0, 0.0, 1.0),
			vec4(0.0, 0.0, 0.0, 1.0));
#endif

	// Use atlas position and a prime number as the seed.
	uint noise = random_seed(ivec3(atlas_pos, 43573547));
	for (uint i = 0; i < bake_params.light_count; i++) {
		vec3 light;
		vec3 light_dir;
		float shadow;
		trace_direct_light(position, normal, i, true, light, light_dir, noise, texel_size_world_space, shadow);

		if (lights.data[i].static_bake) {
			light_for_texture += light;

#ifdef USE_SH_LIGHTMAPS
			// These coefficients include the factored out SH evaluation, diffuse convolution, and final application, as well as the BRDF 1/PI and the spherical monte carlo factor.
			// LO: 1/(2*sqrtPI) * 1/(2*sqrtPI) * PI * PI * 1/PI = 0.25
			// L1: sqrt(3/(4*pi)) * sqrt(3/(4*pi)) * (PI*2/3) * (2 * PI) * 1/PI = 1.0
			// Note: This only works because we aren't scaling, rotating, or combing harmonics, we are just directing applying them in the shader.

			float c[4] = float[](
					0.25, //l0
					light_dir.y, //l1n1
					light_dir.z, //l1n0
					light_dir.x //l1p1
			);

			for (uint j = 0; j < 4; j++) {
				sh_accum[j].rgb += light * c[j] * bake_params.exposure_normalization;
			}
#endif
		}

		light_for_bounces += light * lights.data[i].indirect_energy;

#ifdef USE_SHADOWMASK
		if (lights.data[i].type == LIGHT_TYPE_DIRECTIONAL && i == bake_params.shadowmask_light_idx) {
			shadowmask_value = max(shadowmask_value, shadow);
		}
#endif
	}

	light_for_bounces *= bake_params.exposure_normalization;
	imageStore(dest_light, ivec3(atlas_pos, params.atlas_slice), vec4(light_for_bounces, 1.0));

#ifdef USE_SH_LIGHTMAPS
	// Keep for adding at the end.
	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + 0), sh_accum[0]);
	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + 1), sh_accum[1]);
	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + 2), sh_accum[2]);
	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + 3), sh_accum[3]);
#else
	light_for_texture *= bake_params.exposure_normalization;
	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice), vec4(light_for_texture, 1.0));
#endif

#ifdef USE_SHADOWMASK
	imageStore(shadowmask, ivec3(atlas_pos, params.atlas_slice), vec4(shadowmask_value, shadowmask_value, shadowmask_value, 1.0));
#endif

#endif

#ifdef MODE_BOUNCE_LIGHT

#ifdef USE_SH_LIGHTMAPS
	vec4 sh_accum[4] = vec4[](
			vec4(0.0, 0.0, 0.0, 1.0),
			vec4(0.0, 0.0, 0.0, 1.0),
			vec4(0.0, 0.0, 0.0, 1.0),
			vec4(0.0, 0.0, 0.0, 1.0));
#else
	vec3 light_accum = vec3(0.0);
#endif

	// Retrieve starting normal and position.
	vec3 normal = texelFetch(sampler2DArray(source_normal, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).xyz;
	if (length(normal) < 0.5) {
		// The pixel is empty, skip processing it.
		return;
	}

	vec3 position = texelFetch(sampler2DArray(source_position, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).xyz;
	int neighbor_offset = atlas_pos.x < bake_params.atlas_size.x - 1 ? 1 : -1;
	vec3 neighbor_position = texelFetch(sampler2DArray(source_position, linear_sampler), ivec3(atlas_pos + ivec2(neighbor_offset, 0), params.atlas_slice), 0).xyz;
	float texel_size_world_space = distance(position, neighbor_position);
	uint noise = random_seed(ivec3(params.ray_from, atlas_pos));
	for (uint i = params.ray_from; i < params.ray_to; i++) {
		vec3 ray_dir = generate_ray_dir_from_normal(normal, noise);
		vec3 light = trace_indirect_light(position, ray_dir, noise, texel_size_world_space);

#ifdef USE_SH_LIGHTMAPS
		// These coefficients include the factored out SH evaluation, diffuse convolution, and final application, as well as the BRDF 1/PI and the spherical monte carlo factor.
		// LO: 1/(2*sqrtPI) * 1/(2*sqrtPI) * PI * PI * 1/PI = 0.25
		// L1: sqrt(3/(4*pi)) * sqrt(3/(4*pi)) * (PI*2/3) * (2 * PI) * 1/PI = 1.0
		// Note: This only works because we aren't scaling, rotating, or combing harmonics, we are just directing applying them in the shader.

		float c[4] = float[](
				0.25, //l0
				ray_dir.y, //l1n1
				ray_dir.z, //l1n0
				ray_dir.x //l1p1
		);

		for (uint j = 0; j < 4; j++) {
			sh_accum[j].rgb += light * c[j];
		}
#else
		light_accum += light;
#endif
	}

	// Add the averaged result to the accumulated light texture.
#ifdef USE_SH_LIGHTMAPS
	for (int i = 0; i < 4; i++) {
		vec4 accum = imageLoad(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + i));
		accum.rgb += sh_accum[i].rgb / float(params.ray_count);
		imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + i), accum);
	}
#else
	vec4 accum = imageLoad(accum_light, ivec3(atlas_pos, params.atlas_slice));
	accum.rgb += light_accum / float(params.ray_count);
	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice), accum);
#endif

#endif

#ifdef MODE_UNOCCLUDE

	//texel_size = 0.5;
	//compute tangents

	vec4 position_alpha = imageLoad(position, ivec3(atlas_pos, params.atlas_slice));
	if (position_alpha.a < 0.5) {
		return;
	}

	vec3 vertex_pos = position_alpha.xyz;
	vec4 normal_tsize = imageLoad(unocclude, ivec3(atlas_pos, params.atlas_slice));

	vec3 face_normal = normal_tsize.xyz;
	float texel_size = normal_tsize.w;

	vec3 v0 = abs(face_normal.z) < 0.999 ? vec3(0.0, 0.0, 1.0) : vec3(0.0, 1.0, 0.0);
	vec3 tangent = normalize(cross(v0, face_normal));
	vec3 bitangent = normalize(cross(tangent, face_normal));
	vec3 base_pos = vertex_pos + face_normal * bake_params.bias; // Raise a bit.

	vec3 rays[4] = vec3[](tangent, bitangent, -tangent, -bitangent);
	float min_d = 1e20;
	for (int i = 0; i < 4; i++) {
		vec3 ray_to = base_pos + rays[i] * texel_size;
		float d;
		vec3 norm;

		if (trace_ray_closest_hit_distance(base_pos, ray_to, d, norm) == RAY_BACK) {
			if (d < min_d) {
				// This bias needs to be greater than the regular bias, because otherwise later, rays will go the other side when pointing back.
				vertex_pos = base_pos + rays[i] * d + norm * bake_params.bias * 10.0;
				min_d = d;
			}
		}
	}

	position_alpha.xyz = vertex_pos;

	imageStore(position, ivec3(atlas_pos, params.atlas_slice), position_alpha);

#endif

#ifdef MODE_LIGHT_PROBES

	vec3 position = probe_positions.data[probe_index].xyz;

	vec4 probe_sh_accum[9] = vec4[](
			vec4(0.0),
			vec4(0.0),
			vec4(0.0),
			vec4(0.0),
			vec4(0.0),
			vec4(0.0),
			vec4(0.0),
			vec4(0.0),
			vec4(0.0));

	uint noise = random_seed(ivec3(params.ray_from, probe_index, 49502741 /* some prime */));
	for (uint i = params.ray_from; i < params.ray_to; i++) {
		vec3 ray_dir = generate_sphere_uniform_direction(noise);
		vec3 light = trace_indirect_light(position, ray_dir, noise, 0.0);

		float c[9] = float[](
				0.282095, //l0
				0.488603 * ray_dir.y, //l1n1
				0.488603 * ray_dir.z, //l1n0
				0.488603 * ray_dir.x, //l1p1
				1.092548 * ray_dir.x * ray_dir.y, //l2n2
				1.092548 * ray_dir.y * ray_dir.z, //l2n1
				//0.315392 * (ray_dir.x * ray_dir.x + ray_dir.y * ray_dir.y + 2.0 * ray_dir.z * ray_dir.z), //l20
				0.315392 * (3.0 * ray_dir.z * ray_dir.z - 1.0), //l20
				1.092548 * ray_dir.x * ray_dir.z, //l2p1
				0.546274 * (ray_dir.x * ray_dir.x - ray_dir.y * ray_dir.y) //l2p2
		);

		for (uint j = 0; j < 9; j++) {
			probe_sh_accum[j].rgb += light * c[j];
		}
	}

	if (params.ray_from > 0) {
		for (uint j = 0; j < 9; j++) { //accum from existing
			probe_sh_accum[j] += light_probes.data[probe_index * 9 + j];
		}
	}

	if (params.ray_to == params.ray_count) {
		for (uint j = 0; j < 9; j++) { //accum from existing
			probe_sh_accum[j] *= 4.0 / float(params.ray_count);
		}
	}

	for (uint j = 0; j < 9; j++) { //accum from existing
		light_probes.data[probe_index * 9 + j] = probe_sh_accum[j];
	}

#endif

#if defined(MODE_DILATE)

	vec4 c = texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0);
	//sides first, as they are closer
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-1, 0), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(0, 1), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(1, 0), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(0, -1), params.atlas_slice), 0);
	//endpoints second
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-1, -1), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-1, 1), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(1, -1), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(1, 1), params.atlas_slice), 0);

	//far sides third
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-2, 0), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(0, 2), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(2, 0), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(0, -2), params.atlas_slice), 0);

	//far-mid endpoints
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-2, -1), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-2, 1), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(2, -1), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(2, 1), params.atlas_slice), 0);

	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-1, -2), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-1, 2), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(1, -2), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(1, 2), params.atlas_slice), 0);
	//far endpoints
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-2, -2), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-2, 2), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(2, -2), params.atlas_slice), 0);
	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(2, 2), params.atlas_slice), 0);

	imageStore(dest_light, ivec3(atlas_pos, params.atlas_slice), c);

#endif

#ifdef MODE_DENOISE
	// Joint Non-local means (JNLM) denoiser.
	//
	// Based on YoctoImageDenoiser's JNLM implementation with corrections from "Nonlinearly Weighted First-order Regression for Denoising Monte Carlo Renderings".
	//
	// <https://github.com/ManuelPrandini/YoctoImageDenoiser/blob/06e19489dd64e47792acffde536393802ba48607/libs/yocto_extension/yocto_extension.cpp#L207>
	// <https://benedikt-bitterli.me/nfor/nfor.pdf>
	//
	// MIT License
	//
	// Copyright (c) 2020 ManuelPrandini
	//
	// Permission is hereby granted, free of charge, to any person obtaining a copy
	// of this software and associated documentation files (the "Software"), to deal
	// in the Software without restriction, including without limitation the rights
	// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
	// copies of the Software, and to permit persons to whom the Software is
	// furnished to do so, subject to the following conditions:
	//
	// The above copyright notice and this permission notice shall be included in all
	// copies or substantial portions of the Software.
	//
	// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
	// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
	// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
	// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
	// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
	// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
	// SOFTWARE.
	//
	// Most of the constants below have been hand-picked to fit the common scenarios lightmaps
	// are generated with, but they can be altered freely to experiment and achieve better results.

	// Half the size of the patch window around each pixel that is weighted to compute the denoised pixel.
	// A value of 1 represents a 3x3 window, a value of 2 a 5x5 window, etc.
	const int HALF_PATCH_WINDOW = 3;

	// Half the size of the search window around each pixel that is denoised and weighted to compute the denoised pixel.
	const int HALF_SEARCH_WINDOW = denoise_params.half_search_window;

	// For all of the following sigma values, smaller values will give less weight to pixels that have a bigger distance
	// in the feature being evaluated. Therefore, smaller values are likely to cause more noise to appear, but will also
	// cause less features to be erased in the process.

	// Controls how much the spatial distance of the pixels influences the denoising weight.
	const float SIGMA_SPATIAL = denoise_params.spatial_bandwidth;

	// Controls how much the light color distance of the pixels influences the denoising weight.
	const float SIGMA_LIGHT = denoise_params.light_bandwidth;

	// Controls how much the albedo color distance of the pixels influences the denoising weight.
	const float SIGMA_ALBEDO = denoise_params.albedo_bandwidth;

	// Controls how much the normal vector distance of the pixels influences the denoising weight.
	const float SIGMA_NORMAL = denoise_params.normal_bandwidth;

	// Strength of the filter. The original paper recommends values around 10 to 15 times the Sigma parameter.
	const float FILTER_VALUE = denoise_params.filter_strength * SIGMA_LIGHT;

	// Formula constants.
	const int PATCH_WINDOW_DIMENSION = (HALF_PATCH_WINDOW * 2 + 1);
	const int PATCH_WINDOW_DIMENSION_SQUARE = (PATCH_WINDOW_DIMENSION * PATCH_WINDOW_DIMENSION);
	const float TWO_SIGMA_SPATIAL_SQUARE = 2.0f * SIGMA_SPATIAL * SIGMA_SPATIAL;
	const float TWO_SIGMA_LIGHT_SQUARE = 2.0f * SIGMA_LIGHT * SIGMA_LIGHT;
	const float TWO_SIGMA_ALBEDO_SQUARE = 2.0f * SIGMA_ALBEDO * SIGMA_ALBEDO;
	const float TWO_SIGMA_NORMAL_SQUARE = 2.0f * SIGMA_NORMAL * SIGMA_NORMAL;
	const float FILTER_SQUARE_TWO_SIGMA_LIGHT_SQUARE = FILTER_VALUE * FILTER_VALUE * TWO_SIGMA_LIGHT_SQUARE;
	const float EPSILON = 1e-6f;

#ifdef USE_SH_LIGHTMAPS
	const uint slice_count = 4;
	const uint slice_base = params.atlas_slice * slice_count;
#else
	const uint slice_count = 1;
	const uint slice_base = params.atlas_slice;
#endif

	for (uint i = 0; i < slice_count; i++) {
		uint lightmap_slice = slice_base + i;
		vec3 denoised_rgb = vec3(0.0f);
		vec4 input_light = texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos, lightmap_slice), 0);
		vec3 input_albedo = texelFetch(sampler2DArray(albedo_tex, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).rgb;
		vec3 input_normal = texelFetch(sampler2DArray(source_normal, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).xyz;
		if (length(input_normal) > EPSILON) {
			// Compute the denoised pixel if the normal is valid.
			float sum_weights = 0.0f;
			vec3 input_rgb = input_light.rgb;
			for (int search_y = -HALF_SEARCH_WINDOW; search_y <= HALF_SEARCH_WINDOW; search_y++) {
				for (int search_x = -HALF_SEARCH_WINDOW; search_x <= HALF_SEARCH_WINDOW; search_x++) {
					ivec2 search_pos = atlas_pos + ivec2(search_x, search_y);
					vec3 search_rgb = texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(search_pos, lightmap_slice), 0).rgb;
					vec3 search_albedo = texelFetch(sampler2DArray(albedo_tex, linear_sampler), ivec3(search_pos, params.atlas_slice), 0).rgb;
					vec3 search_normal = texelFetch(sampler2DArray(source_normal, linear_sampler), ivec3(search_pos, params.atlas_slice), 0).xyz;
					float patch_square_dist = 0.0f;
					for (int offset_y = -HALF_PATCH_WINDOW; offset_y <= HALF_PATCH_WINDOW; offset_y++) {
						for (int offset_x = -HALF_PATCH_WINDOW; offset_x <= HALF_PATCH_WINDOW; offset_x++) {
							ivec2 offset_input_pos = atlas_pos + ivec2(offset_x, offset_y);
							ivec2 offset_search_pos = search_pos + ivec2(offset_x, offset_y);
							vec3 offset_input_rgb = texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(offset_input_pos, lightmap_slice), 0).rgb;
							vec3 offset_search_rgb = texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(offset_search_pos, lightmap_slice), 0).rgb;
							vec3 offset_delta_rgb = offset_input_rgb - offset_search_rgb;
							patch_square_dist += dot(offset_delta_rgb, offset_delta_rgb) - TWO_SIGMA_LIGHT_SQUARE;
						}
					}

					patch_square_dist = max(0.0f, patch_square_dist / (3.0f * PATCH_WINDOW_DIMENSION_SQUARE));

					float weight = 1.0f;

					// Ignore weight if search position is out of bounds.
					weight *= step(0, search_pos.x) * step(search_pos.x, bake_params.atlas_size.x - 1);
					weight *= step(0, search_pos.y) * step(search_pos.y, bake_params.atlas_size.y - 1);

					// Ignore weight if normal is zero length.
					weight *= step(EPSILON, length(search_normal));

					// Weight with pixel distance.
					vec2 pixel_delta = vec2(search_x, search_y);
					float pixel_square_dist = dot(pixel_delta, pixel_delta);
					weight *= exp(-pixel_square_dist / TWO_SIGMA_SPATIAL_SQUARE);

					// Weight with patch.
					weight *= exp(-patch_square_dist / FILTER_SQUARE_TWO_SIGMA_LIGHT_SQUARE);

					// Weight with albedo.
					vec3 albedo_delta = input_albedo - search_albedo;
					float albedo_square_dist = dot(albedo_delta, albedo_delta);
					weight *= exp(-albedo_square_dist / TWO_SIGMA_ALBEDO_SQUARE);

					// Weight with normal.
					vec3 normal_delta = input_normal - search_normal;
					float normal_square_dist = dot(normal_delta, normal_delta);
					weight *= exp(-normal_square_dist / TWO_SIGMA_NORMAL_SQUARE);

					denoised_rgb += weight * search_rgb;
					sum_weights += weight;
				}
			}

			denoised_rgb /= sum_weights;
		} else {
			// Ignore pixels where the normal is empty, just copy the light color.
			denoised_rgb = input_light.rgb;
		}

		imageStore(dest_light, ivec3(atlas_pos, lightmap_slice), vec4(denoised_rgb, input_light.a));
	}
#endif

#ifdef MODE_PACK_L1_COEFFS
	vec4 base_coeff = texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos, params.atlas_slice * 4), 0);

	for (int i = 1; i < 4; i++) {
		vec4 c = texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos, params.atlas_slice * 4 + i), 0);

		if (abs(base_coeff.r) > 0.0) {
			c.r /= (base_coeff.r * 8);
		}

		if (abs(base_coeff.g) > 0.0) {
			c.g /= (base_coeff.g * 8);
		}

		if (abs(base_coeff.b) > 0.0) {
			c.b /= (base_coeff.b * 8);
		}

		c.rgb += vec3(0.5);
		c.rgb = clamp(c.rgb, vec3(0.0), vec3(1.0));
		imageStore(dest_light, ivec3(atlas_pos, params.atlas_slice * 4 + i), c);
	}
#endif
}