diff options
Diffstat (limited to 'servers/rendering/renderer_rd/shaders/environment')
10 files changed, 4272 insertions, 0 deletions
diff --git a/servers/rendering/renderer_rd/shaders/environment/SCsub b/servers/rendering/renderer_rd/shaders/environment/SCsub new file mode 100644 index 0000000000..fc513d3fb9 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/environment/SCsub @@ -0,0 +1,17 @@ +#!/usr/bin/env python + +Import("env") + +if "RD_GLSL" in env["BUILDERS"]: + # find all include files + gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + + # find all shader code(all glsl files excluding our include files) + glsl_files = [str(f) for f in Glob("*.glsl") if str(f) not in gl_include_files] + + # make sure we recompile shaders if include files change + env.Depends([f + ".gen.h" for f in glsl_files], gl_include_files) + + # compile shaders + for glsl_file in glsl_files: + env.RD_GLSL(glsl_file) diff --git a/servers/rendering/renderer_rd/shaders/environment/gi.glsl b/servers/rendering/renderer_rd/shaders/environment/gi.glsl new file mode 100644 index 0000000000..f687d50a2d --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/environment/gi.glsl @@ -0,0 +1,672 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +#define M_PI 3.141592 + +#define SDFGI_MAX_CASCADES 8 + +//set 0 for SDFGI and render buffers + +layout(set = 0, binding = 1) uniform texture3D sdf_cascades[SDFGI_MAX_CASCADES]; +layout(set = 0, binding = 2) uniform texture3D light_cascades[SDFGI_MAX_CASCADES]; +layout(set = 0, binding = 3) uniform texture3D aniso0_cascades[SDFGI_MAX_CASCADES]; +layout(set = 0, binding = 4) uniform texture3D aniso1_cascades[SDFGI_MAX_CASCADES]; +layout(set = 0, binding = 5) uniform texture3D occlusion_texture; + +layout(set = 0, binding = 6) uniform sampler linear_sampler; +layout(set = 0, binding = 7) uniform sampler linear_sampler_with_mipmaps; + +struct ProbeCascadeData { + vec3 position; + float to_probe; + ivec3 probe_world_offset; + float to_cell; // 1/bounds * grid_size +}; + +layout(rgba16f, set = 0, binding = 9) uniform restrict writeonly image2D ambient_buffer; +layout(rgba16f, set = 0, binding = 10) uniform restrict writeonly image2D reflection_buffer; + +layout(set = 0, binding = 11) uniform texture2DArray lightprobe_texture; + +layout(set = 0, binding = 12) uniform texture2D depth_buffer; +layout(set = 0, binding = 13) uniform texture2D normal_roughness_buffer; +layout(set = 0, binding = 14) uniform utexture2D voxel_gi_buffer; + +layout(set = 0, binding = 15, std140) uniform SDFGI { + vec3 grid_size; + uint max_cascades; + + bool use_occlusion; + int probe_axis_size; + float probe_to_uvw; + float normal_bias; + + vec3 lightprobe_tex_pixel_size; + float energy; + + vec3 lightprobe_uv_offset; + float y_mult; + + vec3 occlusion_clamp; + uint pad3; + + vec3 occlusion_renormalize; + uint pad4; + + vec3 cascade_probe_size; + uint pad5; + + ProbeCascadeData cascades[SDFGI_MAX_CASCADES]; +} +sdfgi; + +#define MAX_VOXEL_GI_INSTANCES 8 + +struct VoxelGIData { + mat4 xform; // 64 - 64 + + vec3 bounds; // 12 - 76 + float dynamic_range; // 4 - 80 + + float bias; // 4 - 84 + float normal_bias; // 4 - 88 + bool blend_ambient; // 4 - 92 + uint mipmaps; // 4 - 96 +}; + +layout(set = 0, binding = 16, std140) uniform VoxelGIs { + VoxelGIData data[MAX_VOXEL_GI_INSTANCES]; +} +voxel_gi_instances; + +layout(set = 0, binding = 17) uniform texture3D voxel_gi_textures[MAX_VOXEL_GI_INSTANCES]; + +layout(set = 0, binding = 18, std140) uniform SceneData { + mat4x4 inv_projection[2]; + mat4x4 cam_transform; + vec4 eye_offset[2]; + + ivec2 screen_size; + float pad1; + float pad2; +} +scene_data; + +layout(push_constant, std430) uniform Params { + uint view_index; + uint max_voxel_gi_instances; + bool high_quality_vct; + bool orthogonal; + + vec4 proj_info; + + float z_near; + float z_far; + float pad1; + float pad2; +} +params; + +vec2 octahedron_wrap(vec2 v) { + vec2 signVal; + signVal.x = v.x >= 0.0 ? 1.0 : -1.0; + signVal.y = v.y >= 0.0 ? 1.0 : -1.0; + return (1.0 - abs(v.yx)) * signVal; +} + +vec2 octahedron_encode(vec3 n) { + // https://twitter.com/Stubbesaurus/status/937994790553227264 + n /= (abs(n.x) + abs(n.y) + abs(n.z)); + n.xy = n.z >= 0.0 ? n.xy : octahedron_wrap(n.xy); + n.xy = n.xy * 0.5 + 0.5; + return n.xy; +} + +vec4 blend_color(vec4 src, vec4 dst) { + vec4 res; + float sa = 1.0 - src.a; + res.a = dst.a * sa + src.a; + if (res.a == 0.0) { + res.rgb = vec3(0); + } else { + res.rgb = (dst.rgb * dst.a * sa + src.rgb * src.a) / res.a; + } + return res; +} + +vec3 reconstruct_position(ivec2 screen_pos) { +#ifdef USE_MULTIVIEW + vec4 pos; + pos.xy = (2.0 * vec2(screen_pos) / vec2(scene_data.screen_size)) - 1.0; + pos.z = texelFetch(sampler2D(depth_buffer, linear_sampler), screen_pos, 0).r * 2.0 - 1.0; + pos.w = 1.0; + + pos = scene_data.inv_projection[params.view_index] * pos; + + return pos.xyz / pos.w; +#else + vec3 pos; + pos.z = texelFetch(sampler2D(depth_buffer, linear_sampler), screen_pos, 0).r; + + pos.z = pos.z * 2.0 - 1.0; + if (params.orthogonal) { + pos.z = ((pos.z + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0; + } else { + pos.z = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near - pos.z * (params.z_far - params.z_near)); + } + pos.z = -pos.z; + + pos.xy = vec2(screen_pos) * params.proj_info.xy + params.proj_info.zw; + if (!params.orthogonal) { + pos.xy *= pos.z; + } + + return pos; +#endif +} + +void sdfvoxel_gi_process(uint cascade, vec3 cascade_pos, vec3 cam_pos, vec3 cam_normal, vec3 cam_specular_normal, float roughness, out vec3 diffuse_light, out vec3 specular_light) { + cascade_pos += cam_normal * sdfgi.normal_bias; + + vec3 base_pos = floor(cascade_pos); + //cascade_pos += mix(vec3(0.0),vec3(0.01),lessThan(abs(cascade_pos-base_pos),vec3(0.01))) * cam_normal; + ivec3 probe_base_pos = ivec3(base_pos); + + vec4 diffuse_accum = vec4(0.0); + vec3 specular_accum; + + ivec3 tex_pos = ivec3(probe_base_pos.xy, int(cascade)); + tex_pos.x += probe_base_pos.z * sdfgi.probe_axis_size; + tex_pos.xy = tex_pos.xy * (SDFGI_OCT_SIZE + 2) + ivec2(1); + + vec3 diffuse_posf = (vec3(tex_pos) + vec3(octahedron_encode(cam_normal) * float(SDFGI_OCT_SIZE), 0.0)) * sdfgi.lightprobe_tex_pixel_size; + + vec3 specular_posf = (vec3(tex_pos) + vec3(octahedron_encode(cam_specular_normal) * float(SDFGI_OCT_SIZE), 0.0)) * sdfgi.lightprobe_tex_pixel_size; + + specular_accum = vec3(0.0); + + vec4 light_accum = vec4(0.0); + float weight_accum = 0.0; + + for (uint j = 0; j < 8; j++) { + ivec3 offset = (ivec3(j) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1); + ivec3 probe_posi = probe_base_pos; + probe_posi += offset; + + // Compute weight + + vec3 probe_pos = vec3(probe_posi); + vec3 probe_to_pos = cascade_pos - probe_pos; + vec3 probe_dir = normalize(-probe_to_pos); + + vec3 trilinear = vec3(1.0) - abs(probe_to_pos); + float weight = trilinear.x * trilinear.y * trilinear.z * max(0.005, dot(cam_normal, probe_dir)); + + // Compute lightprobe occlusion + + if (sdfgi.use_occlusion) { + ivec3 occ_indexv = abs((sdfgi.cascades[cascade].probe_world_offset + probe_posi) & ivec3(1, 1, 1)) * ivec3(1, 2, 4); + vec4 occ_mask = mix(vec4(0.0), vec4(1.0), equal(ivec4(occ_indexv.x | occ_indexv.y), ivec4(0, 1, 2, 3))); + + vec3 occ_pos = clamp(cascade_pos, probe_pos - sdfgi.occlusion_clamp, probe_pos + sdfgi.occlusion_clamp) * sdfgi.probe_to_uvw; + occ_pos.z += float(cascade); + if (occ_indexv.z != 0) { //z bit is on, means index is >=4, so make it switch to the other half of textures + occ_pos.x += 1.0; + } + + occ_pos *= sdfgi.occlusion_renormalize; + float occlusion = dot(textureLod(sampler3D(occlusion_texture, linear_sampler), occ_pos, 0.0), occ_mask); + + weight *= max(occlusion, 0.01); + } + + // Compute lightprobe texture position + + vec3 diffuse; + vec3 pos_uvw = diffuse_posf; + pos_uvw.xy += vec2(offset.xy) * sdfgi.lightprobe_uv_offset.xy; + pos_uvw.x += float(offset.z) * sdfgi.lightprobe_uv_offset.z; + diffuse = textureLod(sampler2DArray(lightprobe_texture, linear_sampler), pos_uvw, 0.0).rgb; + + diffuse_accum += vec4(diffuse * weight, weight); + + { + vec3 specular = vec3(0.0); + vec3 pos_uvw = specular_posf; + pos_uvw.xy += vec2(offset.xy) * sdfgi.lightprobe_uv_offset.xy; + pos_uvw.x += float(offset.z) * sdfgi.lightprobe_uv_offset.z; + if (roughness < 0.99) { + specular = textureLod(sampler2DArray(lightprobe_texture, linear_sampler), pos_uvw + vec3(0, 0, float(sdfgi.max_cascades)), 0.0).rgb; + } + if (roughness > 0.2) { + specular = mix(specular, textureLod(sampler2DArray(lightprobe_texture, linear_sampler), pos_uvw, 0.0).rgb, (roughness - 0.2) * 1.25); + } + + specular_accum += specular * weight; + } + } + + if (diffuse_accum.a > 0.0) { + diffuse_accum.rgb /= diffuse_accum.a; + } + + diffuse_light = diffuse_accum.rgb; + + if (diffuse_accum.a > 0.0) { + specular_accum /= diffuse_accum.a; + } + + specular_light = specular_accum; +} + +void sdfgi_process(vec3 vertex, vec3 normal, vec3 reflection, float roughness, out vec4 ambient_light, out vec4 reflection_light) { + //make vertex orientation the world one, but still align to camera + vertex.y *= sdfgi.y_mult; + normal.y *= sdfgi.y_mult; + reflection.y *= sdfgi.y_mult; + + //renormalize + normal = normalize(normal); + reflection = normalize(reflection); + + vec3 cam_pos = vertex; + vec3 cam_normal = normal; + + vec4 light_accum = vec4(0.0); + float weight_accum = 0.0; + + vec4 light_blend_accum = vec4(0.0); + float weight_blend_accum = 0.0; + + float blend = -1.0; + + // helper constants, compute once + + uint cascade = 0xFFFFFFFF; + vec3 cascade_pos; + vec3 cascade_normal; + + for (uint i = 0; i < sdfgi.max_cascades; i++) { + cascade_pos = (cam_pos - sdfgi.cascades[i].position) * sdfgi.cascades[i].to_probe; + + if (any(lessThan(cascade_pos, vec3(0.0))) || any(greaterThanEqual(cascade_pos, sdfgi.cascade_probe_size))) { + continue; //skip cascade + } + + cascade = i; + break; + } + + if (cascade < SDFGI_MAX_CASCADES) { + ambient_light = vec4(0, 0, 0, 1); + reflection_light = vec4(0, 0, 0, 1); + + float blend; + vec3 diffuse, specular; + sdfvoxel_gi_process(cascade, cascade_pos, cam_pos, cam_normal, reflection, roughness, diffuse, specular); + + { + //process blend + float blend_from = (float(sdfgi.probe_axis_size - 1) / 2.0) - 2.5; + float blend_to = blend_from + 2.0; + + vec3 inner_pos = cam_pos * sdfgi.cascades[cascade].to_probe; + + float len = length(inner_pos); + + inner_pos = abs(normalize(inner_pos)); + len *= max(inner_pos.x, max(inner_pos.y, inner_pos.z)); + + if (len >= blend_from) { + blend = smoothstep(blend_from, blend_to, len); + } else { + blend = 0.0; + } + } + + if (blend > 0.0) { + //blend + if (cascade == sdfgi.max_cascades - 1) { + ambient_light.a = 1.0 - blend; + reflection_light.a = 1.0 - blend; + + } else { + vec3 diffuse2, specular2; + cascade_pos = (cam_pos - sdfgi.cascades[cascade + 1].position) * sdfgi.cascades[cascade + 1].to_probe; + sdfvoxel_gi_process(cascade + 1, cascade_pos, cam_pos, cam_normal, reflection, roughness, diffuse2, specular2); + diffuse = mix(diffuse, diffuse2, blend); + specular = mix(specular, specular2, blend); + } + } + + ambient_light.rgb = diffuse; + + if (roughness < 0.2) { + vec3 pos_to_uvw = 1.0 / sdfgi.grid_size; + vec4 light_accum = vec4(0.0); + + float blend_size = (sdfgi.grid_size.x / float(sdfgi.probe_axis_size - 1)) * 0.5; + + float radius_sizes[SDFGI_MAX_CASCADES]; + cascade = 0xFFFF; + + float base_distance = length(cam_pos); + for (uint i = 0; i < sdfgi.max_cascades; i++) { + radius_sizes[i] = (1.0 / sdfgi.cascades[i].to_cell) * (sdfgi.grid_size.x * 0.5 - blend_size); + if (cascade == 0xFFFF && base_distance < radius_sizes[i]) { + cascade = i; + } + } + + cascade = min(cascade, sdfgi.max_cascades - 1); + + float max_distance = radius_sizes[sdfgi.max_cascades - 1]; + vec3 ray_pos = cam_pos; + vec3 ray_dir = reflection; + + { + float prev_radius = cascade > 0 ? radius_sizes[cascade - 1] : 0.0; + float base_blend = (base_distance - prev_radius) / (radius_sizes[cascade] - prev_radius); + float bias = (1.0 + base_blend) * 1.1; + vec3 abs_ray_dir = abs(ray_dir); + //ray_pos += ray_dir * (bias / sdfgi.cascades[cascade].to_cell); //bias to avoid self occlusion + ray_pos += (ray_dir * 1.0 / max(abs_ray_dir.x, max(abs_ray_dir.y, abs_ray_dir.z)) + cam_normal * 1.4) * bias / sdfgi.cascades[cascade].to_cell; + } + float softness = 0.2 + min(1.0, roughness * 5.0) * 4.0; //approximation to roughness so it does not seem like a hard fade + uint i = 0; + bool found = false; + while (true) { + if (length(ray_pos) >= max_distance || light_accum.a > 0.99) { + break; + } + if (!found && i >= cascade && length(ray_pos) < radius_sizes[i]) { + uint next_i = min(i + 1, sdfgi.max_cascades - 1); + cascade = max(i, cascade); //never go down + + vec3 pos = ray_pos - sdfgi.cascades[i].position; + pos *= sdfgi.cascades[i].to_cell * pos_to_uvw; + + float fdistance = textureLod(sampler3D(sdf_cascades[i], linear_sampler), pos, 0.0).r * 255.0 - 1.1; + + vec4 hit_light = vec4(0.0); + if (fdistance < softness) { + hit_light.rgb = textureLod(sampler3D(light_cascades[i], linear_sampler), pos, 0.0).rgb; + hit_light.rgb *= 0.5; //approximation given value read is actually meant for anisotropy + hit_light.a = clamp(1.0 - (fdistance / softness), 0.0, 1.0); + hit_light.rgb *= hit_light.a; + } + + fdistance /= sdfgi.cascades[i].to_cell; + + if (i < (sdfgi.max_cascades - 1)) { + pos = ray_pos - sdfgi.cascades[next_i].position; + pos *= sdfgi.cascades[next_i].to_cell * pos_to_uvw; + + float fdistance2 = textureLod(sampler3D(sdf_cascades[next_i], linear_sampler), pos, 0.0).r * 255.0 - 1.1; + + vec4 hit_light2 = vec4(0.0); + if (fdistance2 < softness) { + hit_light2.rgb = textureLod(sampler3D(light_cascades[next_i], linear_sampler), pos, 0.0).rgb; + hit_light2.rgb *= 0.5; //approximation given value read is actually meant for anisotropy + hit_light2.a = clamp(1.0 - (fdistance2 / softness), 0.0, 1.0); + hit_light2.rgb *= hit_light2.a; + } + + float prev_radius = i == 0 ? 0.0 : radius_sizes[max(0, i - 1)]; + float blend = clamp((length(ray_pos) - prev_radius) / (radius_sizes[i] - prev_radius), 0.0, 1.0); + + fdistance2 /= sdfgi.cascades[next_i].to_cell; + + hit_light = mix(hit_light, hit_light2, blend); + fdistance = mix(fdistance, fdistance2, blend); + } + + light_accum += hit_light; + ray_pos += ray_dir * fdistance; + found = true; + } + i++; + if (i == sdfgi.max_cascades) { + i = 0; + found = false; + } + } + + vec3 light = light_accum.rgb / max(light_accum.a, 0.00001); + float alpha = min(1.0, light_accum.a); + + float b = min(1.0, roughness * 5.0); + + float sa = 1.0 - b; + + reflection_light.a = alpha * sa + b; + if (reflection_light.a == 0) { + specular = vec3(0.0); + } else { + specular = (light * alpha * sa + specular * b) / reflection_light.a; + } + } + + reflection_light.rgb = specular; + + ambient_light.rgb *= sdfgi.energy; + reflection_light.rgb *= sdfgi.energy; + } else { + ambient_light = vec4(0); + reflection_light = vec4(0); + } +} + +//standard voxel cone trace +vec4 voxel_cone_trace(texture3D probe, vec3 cell_size, vec3 pos, vec3 direction, float tan_half_angle, float max_distance, float p_bias) { + float dist = p_bias; + vec4 color = vec4(0.0); + + while (dist < max_distance && color.a < 0.95) { + float diameter = max(1.0, 2.0 * tan_half_angle * dist); + vec3 uvw_pos = (pos + dist * direction) * cell_size; + float half_diameter = diameter * 0.5; + //check if outside, then break + if (any(greaterThan(abs(uvw_pos - 0.5), vec3(0.5f + half_diameter * cell_size)))) { + break; + } + vec4 scolor = textureLod(sampler3D(probe, linear_sampler_with_mipmaps), uvw_pos, log2(diameter)); + float a = (1.0 - color.a); + color += a * scolor; + dist += half_diameter; + } + + return color; +} + +vec4 voxel_cone_trace_45_degrees(texture3D probe, vec3 cell_size, vec3 pos, vec3 direction, float max_distance, float p_bias) { + float dist = p_bias; + vec4 color = vec4(0.0); + float radius = max(0.5, dist); + float lod_level = log2(radius * 2.0); + + while (dist < max_distance && color.a < 0.95) { + vec3 uvw_pos = (pos + dist * direction) * cell_size; + + //check if outside, then break + if (any(greaterThan(abs(uvw_pos - 0.5), vec3(0.5f + radius * cell_size)))) { + break; + } + vec4 scolor = textureLod(sampler3D(probe, linear_sampler_with_mipmaps), uvw_pos, lod_level); + lod_level += 1.0; + + float a = (1.0 - color.a); + scolor *= a; + color += scolor; + dist += radius; + radius = max(0.5, dist); + } + return color; +} + +void voxel_gi_compute(uint index, vec3 position, vec3 normal, vec3 ref_vec, mat3 normal_xform, float roughness, inout vec4 out_spec, inout vec4 out_diff, inout float out_blend) { + position = (voxel_gi_instances.data[index].xform * vec4(position, 1.0)).xyz; + ref_vec = normalize((voxel_gi_instances.data[index].xform * vec4(ref_vec, 0.0)).xyz); + normal = normalize((voxel_gi_instances.data[index].xform * vec4(normal, 0.0)).xyz); + + position += normal * voxel_gi_instances.data[index].normal_bias; + + //this causes corrupted pixels, i have no idea why.. + if (any(bvec2(any(lessThan(position, vec3(0.0))), any(greaterThan(position, voxel_gi_instances.data[index].bounds))))) { + return; + } + + mat3 dir_xform = mat3(voxel_gi_instances.data[index].xform) * normal_xform; + + vec3 blendv = abs(position / voxel_gi_instances.data[index].bounds * 2.0 - 1.0); + float blend = clamp(1.0 - max(blendv.x, max(blendv.y, blendv.z)), 0.0, 1.0); + //float blend=1.0; + + float max_distance = length(voxel_gi_instances.data[index].bounds); + vec3 cell_size = 1.0 / voxel_gi_instances.data[index].bounds; + + //irradiance + + vec4 light = vec4(0.0); + + if (params.high_quality_vct) { + const uint cone_dir_count = 6; + vec3 cone_dirs[cone_dir_count] = vec3[]( + vec3(0.0, 0.0, 1.0), + vec3(0.866025, 0.0, 0.5), + vec3(0.267617, 0.823639, 0.5), + vec3(-0.700629, 0.509037, 0.5), + vec3(-0.700629, -0.509037, 0.5), + vec3(0.267617, -0.823639, 0.5)); + + float cone_weights[cone_dir_count] = float[](0.25, 0.15, 0.15, 0.15, 0.15, 0.15); + float cone_angle_tan = 0.577; + + for (uint i = 0; i < cone_dir_count; i++) { + vec3 dir = normalize(dir_xform * cone_dirs[i]); + light += cone_weights[i] * voxel_cone_trace(voxel_gi_textures[index], cell_size, position, dir, cone_angle_tan, max_distance, voxel_gi_instances.data[index].bias); + } + } else { + const uint cone_dir_count = 4; + vec3 cone_dirs[cone_dir_count] = vec3[]( + vec3(0.707107, 0.0, 0.707107), + vec3(0.0, 0.707107, 0.707107), + vec3(-0.707107, 0.0, 0.707107), + vec3(0.0, -0.707107, 0.707107)); + + float cone_weights[cone_dir_count] = float[](0.25, 0.25, 0.25, 0.25); + for (int i = 0; i < cone_dir_count; i++) { + vec3 dir = normalize(dir_xform * cone_dirs[i]); + light += cone_weights[i] * voxel_cone_trace_45_degrees(voxel_gi_textures[index], cell_size, position, dir, max_distance, voxel_gi_instances.data[index].bias); + } + } + + light.rgb *= voxel_gi_instances.data[index].dynamic_range; + if (!voxel_gi_instances.data[index].blend_ambient) { + light.a = 1.0; + } + + out_diff += light * blend; + + //radiance + vec4 irr_light = voxel_cone_trace(voxel_gi_textures[index], cell_size, position, ref_vec, tan(roughness * 0.5 * M_PI * 0.99), max_distance, voxel_gi_instances.data[index].bias); + irr_light.rgb *= voxel_gi_instances.data[index].dynamic_range; + if (!voxel_gi_instances.data[index].blend_ambient) { + irr_light.a = 1.0; + } + + out_spec += irr_light * blend; + + out_blend += blend; +} + +vec4 fetch_normal_and_roughness(ivec2 pos) { + vec4 normal_roughness = texelFetch(sampler2D(normal_roughness_buffer, linear_sampler), pos, 0); + + normal_roughness.xyz = normalize(normal_roughness.xyz * 2.0 - 1.0); + return normal_roughness; +} + +void process_gi(ivec2 pos, vec3 vertex, inout vec4 ambient_light, inout vec4 reflection_light) { + vec4 normal_roughness = fetch_normal_and_roughness(pos); + + vec3 normal = normal_roughness.xyz; + + if (normal.length() > 0.5) { + //valid normal, can do GI + float roughness = normal_roughness.w; + vec3 view = -normalize(mat3(scene_data.cam_transform) * (vertex - scene_data.eye_offset[params.view_index].xyz)); + vertex = mat3(scene_data.cam_transform) * vertex; + normal = normalize(mat3(scene_data.cam_transform) * normal); + vec3 reflection = normalize(reflect(-view, normal)); + +#ifdef USE_SDFGI + sdfgi_process(vertex, normal, reflection, roughness, ambient_light, reflection_light); +#endif + +#ifdef USE_VOXEL_GI_INSTANCES + { + uvec2 voxel_gi_tex = texelFetch(usampler2D(voxel_gi_buffer, linear_sampler), pos, 0).rg; + roughness *= roughness; + //find arbitrary tangent and bitangent, then build a matrix + 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); + + vec4 amb_accum = vec4(0.0); + vec4 spec_accum = vec4(0.0); + float blend_accum = 0.0; + + for (uint i = 0; i < params.max_voxel_gi_instances; i++) { + if (any(equal(uvec2(i), voxel_gi_tex))) { + voxel_gi_compute(i, vertex, normal, reflection, normal_mat, roughness, spec_accum, amb_accum, blend_accum); + } + } + if (blend_accum > 0.0) { + amb_accum /= blend_accum; + spec_accum /= blend_accum; + } + +#ifdef USE_SDFGI + reflection_light = blend_color(spec_accum, reflection_light); + ambient_light = blend_color(amb_accum, ambient_light); +#else + reflection_light = spec_accum; + ambient_light = amb_accum; +#endif + } +#endif + } +} + +void main() { + ivec2 pos = ivec2(gl_GlobalInvocationID.xy); + +#ifdef MODE_HALF_RES + pos <<= 1; +#endif + if (any(greaterThanEqual(pos, scene_data.screen_size))) { //too large, do nothing + return; + } + + vec4 ambient_light = vec4(0.0); + vec4 reflection_light = vec4(0.0); + + vec3 vertex = reconstruct_position(pos); + vertex.y = -vertex.y; + + process_gi(pos, vertex, ambient_light, reflection_light); + +#ifdef MODE_HALF_RES + pos >>= 1; +#endif + + imageStore(ambient_buffer, pos, ambient_light); + imageStore(reflection_buffer, pos, reflection_light); +} diff --git a/servers/rendering/renderer_rd/shaders/environment/sdfgi_debug.glsl b/servers/rendering/renderer_rd/shaders/environment/sdfgi_debug.glsl new file mode 100644 index 0000000000..af5f7d0a58 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/environment/sdfgi_debug.glsl @@ -0,0 +1,178 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +#define MAX_CASCADES 8 + +layout(set = 0, binding = 1) uniform texture3D sdf_cascades[MAX_CASCADES]; +layout(set = 0, binding = 2) uniform texture3D light_cascades[MAX_CASCADES]; +layout(set = 0, binding = 3) uniform texture3D aniso0_cascades[MAX_CASCADES]; +layout(set = 0, binding = 4) uniform texture3D aniso1_cascades[MAX_CASCADES]; +layout(set = 0, binding = 5) uniform texture3D occlusion_texture; + +layout(set = 0, binding = 8) uniform sampler linear_sampler; + +struct CascadeData { + vec3 offset; //offset of (0,0,0) in world coordinates + float to_cell; // 1/bounds * grid_size + ivec3 probe_world_offset; + uint pad; +}; + +layout(set = 0, binding = 9, std140) uniform Cascades { + CascadeData data[MAX_CASCADES]; +} +cascades; + +layout(rgba16f, set = 0, binding = 10) uniform restrict writeonly image2D screen_buffer; + +layout(set = 0, binding = 11) uniform texture2DArray lightprobe_texture; + +layout(push_constant, std430) uniform Params { + vec3 grid_size; + uint max_cascades; + + ivec2 screen_size; + bool use_occlusion; + float y_mult; + + int probe_axis_size; + float z_near; + float reserved1; + float reserved2; + + mat4 cam_transform; + mat4 inv_projection; +} +params; + +vec3 linear_to_srgb(vec3 color) { + //if going to srgb, clamp from 0 to 1. + color = clamp(color, vec3(0.0), vec3(1.0)); + const vec3 a = vec3(0.055f); + return mix((vec3(1.0f) + a) * pow(color.rgb, vec3(1.0f / 2.4f)) - a, 12.92f * color.rgb, lessThan(color.rgb, vec3(0.0031308f))); +} + +vec2 octahedron_wrap(vec2 v) { + vec2 signVal; + signVal.x = v.x >= 0.0 ? 1.0 : -1.0; + signVal.y = v.y >= 0.0 ? 1.0 : -1.0; + return (1.0 - abs(v.yx)) * signVal; +} + +vec2 octahedron_encode(vec3 n) { + // https://twitter.com/Stubbesaurus/status/937994790553227264 + n /= (abs(n.x) + abs(n.y) + abs(n.z)); + n.xy = n.z >= 0.0 ? n.xy : octahedron_wrap(n.xy); + n.xy = n.xy * 0.5 + 0.5; + return n.xy; +} + +void main() { + // Pixel being shaded + ivec2 screen_pos = ivec2(gl_GlobalInvocationID.xy); + if (any(greaterThanEqual(screen_pos, params.screen_size))) { //too large, do nothing + return; + } + + vec3 ray_pos; + vec3 ray_dir; + { + ray_pos = params.cam_transform[3].xyz; + + ray_dir.xy = ((vec2(screen_pos) / vec2(params.screen_size)) * 2.0 - 1.0); + ray_dir.z = params.z_near; + ray_dir = (params.inv_projection * vec4(ray_dir, 1.0)).xyz; + + ray_dir = normalize(mat3(params.cam_transform) * ray_dir); + } + + ray_pos.y *= params.y_mult; + ray_dir.y *= params.y_mult; + ray_dir = normalize(ray_dir); + + vec3 pos_to_uvw = 1.0 / params.grid_size; + + vec3 light = vec3(0.0); + float blend = 0.0; + +#if 1 + // No interpolation + + vec3 inv_dir = 1.0 / ray_dir; + + float rough = 0.5; + bool hit = false; + + for (uint i = 0; i < params.max_cascades; i++) { + //convert to local bounds + vec3 pos = ray_pos - cascades.data[i].offset; + pos *= cascades.data[i].to_cell; + + // Should never happen for debug, since we start mostly at the bounds center, + // but add anyway. + //if (any(lessThan(pos,vec3(0.0))) || any(greaterThanEqual(pos,params.grid_size))) { + // continue; //already past bounds for this cascade, goto next + //} + + //find maximum advance distance (until reaching bounds) + vec3 t0 = -pos * inv_dir; + vec3 t1 = (params.grid_size - pos) * inv_dir; + vec3 tmax = max(t0, t1); + float max_advance = min(tmax.x, min(tmax.y, tmax.z)); + + float advance = 0.0; + vec3 uvw; + hit = false; + + while (advance < max_advance) { + //read how much to advance from SDF + uvw = (pos + ray_dir * advance) * pos_to_uvw; + + float distance = texture(sampler3D(sdf_cascades[i], linear_sampler), uvw).r * 255.0 - 1.7; + + if (distance < 0.001) { + //consider hit + hit = true; + break; + } + + advance += distance; + } + + if (!hit) { + pos += ray_dir * min(advance, max_advance); + pos /= cascades.data[i].to_cell; + pos += cascades.data[i].offset; + ray_pos = pos; + continue; + } + + //compute albedo, emission and normal at hit point + + const float EPSILON = 0.001; + vec3 hit_normal = normalize(vec3( + texture(sampler3D(sdf_cascades[i], linear_sampler), uvw + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_cascades[i], linear_sampler), uvw - vec3(EPSILON, 0.0, 0.0)).r, + texture(sampler3D(sdf_cascades[i], linear_sampler), uvw + vec3(0.0, EPSILON, 0.0)).r - texture(sampler3D(sdf_cascades[i], linear_sampler), uvw - vec3(0.0, EPSILON, 0.0)).r, + texture(sampler3D(sdf_cascades[i], linear_sampler), uvw + vec3(0.0, 0.0, EPSILON)).r - texture(sampler3D(sdf_cascades[i], linear_sampler), uvw - vec3(0.0, 0.0, EPSILON)).r)); + + vec3 hit_light = texture(sampler3D(light_cascades[i], linear_sampler), uvw).rgb; + vec4 aniso0 = texture(sampler3D(aniso0_cascades[i], linear_sampler), uvw); + vec3 hit_aniso0 = aniso0.rgb; + vec3 hit_aniso1 = vec3(aniso0.a, texture(sampler3D(aniso1_cascades[i], linear_sampler), uvw).rg); + + hit_light *= (dot(max(vec3(0.0), (hit_normal * hit_aniso0)), vec3(1.0)) + dot(max(vec3(0.0), (-hit_normal * hit_aniso1)), vec3(1.0))); + + light = hit_light; + + break; + } + +#endif + + imageStore(screen_buffer, screen_pos, vec4(linear_to_srgb(light), 1.0)); +} diff --git a/servers/rendering/renderer_rd/shaders/environment/sdfgi_debug_probes.glsl b/servers/rendering/renderer_rd/shaders/environment/sdfgi_debug_probes.glsl new file mode 100644 index 0000000000..75b1ad2130 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/environment/sdfgi_debug_probes.glsl @@ -0,0 +1,267 @@ +#[vertex] + +#version 450 + +#if defined(USE_MULTIVIEW) && defined(has_VK_KHR_multiview) +#extension GL_EXT_multiview : enable +#endif + +#ifdef USE_MULTIVIEW +#ifdef has_VK_KHR_multiview +#define ViewIndex gl_ViewIndex +#else // has_VK_KHR_multiview +// !BAS! This needs to become an input once we implement our fallback! +#define ViewIndex 0 +#endif // has_VK_KHR_multiview +#else // USE_MULTIVIEW +// Set to zero, not supported in non stereo +#define ViewIndex 0 +#endif //USE_MULTIVIEW + +#VERSION_DEFINES + +#define MAX_CASCADES 8 +#define MAX_VIEWS 2 + +layout(push_constant, std430) uniform Params { + uint band_power; + uint sections_in_band; + uint band_mask; + float section_arc; + + vec3 grid_size; + uint cascade; + + uint pad; + float y_mult; + uint probe_debug_index; + int probe_axis_size; +} +params; + +// https://in4k.untergrund.net/html_articles/hugi_27_-_coding_corner_polaris_sphere_tessellation_101.htm + +vec3 get_sphere_vertex(uint p_vertex_id) { + float x_angle = float(p_vertex_id & 1u) + (p_vertex_id >> params.band_power); + + float y_angle = + float((p_vertex_id & params.band_mask) >> 1) + ((p_vertex_id >> params.band_power) * params.sections_in_band); + + x_angle *= params.section_arc * 0.5f; // remember - 180AA x rot not 360 + y_angle *= -params.section_arc; + + vec3 point = vec3(sin(x_angle) * sin(y_angle), cos(x_angle), sin(x_angle) * cos(y_angle)); + + return point; +} + +#ifdef MODE_PROBES + +layout(location = 0) out vec3 normal_interp; +layout(location = 1) out flat uint probe_index; + +#endif + +#ifdef MODE_VISIBILITY + +layout(location = 0) out float visibility; + +#endif + +struct CascadeData { + vec3 offset; //offset of (0,0,0) in world coordinates + float to_cell; // 1/bounds * grid_size + ivec3 probe_world_offset; + uint pad; +}; + +layout(set = 0, binding = 1, std140) uniform Cascades { + CascadeData data[MAX_CASCADES]; +} +cascades; + +layout(set = 0, binding = 4) uniform texture3D occlusion_texture; +layout(set = 0, binding = 3) uniform sampler linear_sampler; + +layout(set = 0, binding = 5, std140) uniform SceneData { + mat4 projection[MAX_VIEWS]; +} +scene_data; + +void main() { +#ifdef MODE_PROBES + probe_index = gl_InstanceIndex; + + normal_interp = get_sphere_vertex(gl_VertexIndex); + + vec3 vertex = normal_interp * 0.2; + + float probe_cell_size = float(params.grid_size / float(params.probe_axis_size - 1)) / cascades.data[params.cascade].to_cell; + + ivec3 probe_cell; + probe_cell.x = int(probe_index % params.probe_axis_size); + probe_cell.y = int(probe_index / (params.probe_axis_size * params.probe_axis_size)); + probe_cell.z = int((probe_index / params.probe_axis_size) % params.probe_axis_size); + + vertex += (cascades.data[params.cascade].offset + vec3(probe_cell) * probe_cell_size) / vec3(1.0, params.y_mult, 1.0); + + gl_Position = scene_data.projection[ViewIndex] * vec4(vertex, 1.0); +#endif + +#ifdef MODE_VISIBILITY + + int probe_index = int(params.probe_debug_index); + + vec3 vertex = get_sphere_vertex(gl_VertexIndex) * 0.01; + + float probe_cell_size = float(params.grid_size / float(params.probe_axis_size - 1)) / cascades.data[params.cascade].to_cell; + + ivec3 probe_cell; + probe_cell.x = int(probe_index % params.probe_axis_size); + probe_cell.y = int((probe_index % (params.probe_axis_size * params.probe_axis_size)) / params.probe_axis_size); + probe_cell.z = int(probe_index / (params.probe_axis_size * params.probe_axis_size)); + + vertex += (cascades.data[params.cascade].offset + vec3(probe_cell) * probe_cell_size) / vec3(1.0, params.y_mult, 1.0); + + int probe_voxels = int(params.grid_size.x) / int(params.probe_axis_size - 1); + int occluder_index = int(gl_InstanceIndex); + + int diameter = probe_voxels * 2; + ivec3 occluder_pos; + occluder_pos.x = int(occluder_index % diameter); + occluder_pos.y = int(occluder_index / (diameter * diameter)); + occluder_pos.z = int((occluder_index / diameter) % diameter); + + float cell_size = 1.0 / cascades.data[params.cascade].to_cell; + + ivec3 occluder_offset = occluder_pos - ivec3(diameter / 2); + vertex += ((vec3(occluder_offset) + vec3(0.5)) * cell_size) / vec3(1.0, params.y_mult, 1.0); + + ivec3 global_cell = probe_cell + cascades.data[params.cascade].probe_world_offset; + uint occlusion_layer = 0; + if ((global_cell.x & 1) != 0) { + occlusion_layer |= 1; + } + if ((global_cell.y & 1) != 0) { + occlusion_layer |= 2; + } + if ((global_cell.z & 1) != 0) { + occlusion_layer |= 4; + } + ivec3 tex_pos = probe_cell * probe_voxels + occluder_offset; + + const vec4 layer_axis[4] = vec4[]( + vec4(1, 0, 0, 0), + vec4(0, 1, 0, 0), + vec4(0, 0, 1, 0), + vec4(0, 0, 0, 1)); + + tex_pos.z += int(params.cascade) * int(params.grid_size); + if (occlusion_layer >= 4) { + tex_pos.x += int(params.grid_size.x); + occlusion_layer &= 3; + } + + visibility = dot(texelFetch(sampler3D(occlusion_texture, linear_sampler), tex_pos, 0), layer_axis[occlusion_layer]); + + gl_Position = scene_data.projection[ViewIndex] * vec4(vertex, 1.0); + +#endif +} + +#[fragment] + +#version 450 + +#if defined(USE_MULTIVIEW) && defined(has_VK_KHR_multiview) +#extension GL_EXT_multiview : enable +#endif + +#ifdef USE_MULTIVIEW +#ifdef has_VK_KHR_multiview +#define ViewIndex gl_ViewIndex +#else // has_VK_KHR_multiview +// !BAS! This needs to become an input once we implement our fallback! +#define ViewIndex 0 +#endif // has_VK_KHR_multiview +#else // USE_MULTIVIEW +// Set to zero, not supported in non stereo +#define ViewIndex 0 +#endif //USE_MULTIVIEW + +#VERSION_DEFINES + +#define MAX_VIEWS 2 + +layout(location = 0) out vec4 frag_color; + +layout(set = 0, binding = 2) uniform texture2DArray lightprobe_texture; +layout(set = 0, binding = 3) uniform sampler linear_sampler; + +layout(push_constant, std430) uniform Params { + uint band_power; + uint sections_in_band; + uint band_mask; + float section_arc; + + vec3 grid_size; + uint cascade; + + uint pad; + float y_mult; + uint probe_debug_index; + int probe_axis_size; +} +params; + +#ifdef MODE_PROBES + +layout(location = 0) in vec3 normal_interp; +layout(location = 1) in flat uint probe_index; + +#endif + +#ifdef MODE_VISIBILITY +layout(location = 0) in float visibility; +#endif + +vec2 octahedron_wrap(vec2 v) { + vec2 signVal; + signVal.x = v.x >= 0.0 ? 1.0 : -1.0; + signVal.y = v.y >= 0.0 ? 1.0 : -1.0; + return (1.0 - abs(v.yx)) * signVal; +} + +vec2 octahedron_encode(vec3 n) { + // https://twitter.com/Stubbesaurus/status/937994790553227264 + n /= (abs(n.x) + abs(n.y) + abs(n.z)); + n.xy = n.z >= 0.0 ? n.xy : octahedron_wrap(n.xy); + n.xy = n.xy * 0.5 + 0.5; + return n.xy; +} + +void main() { +#ifdef MODE_PROBES + + ivec3 tex_pos; + tex_pos.x = int(probe_index) % params.probe_axis_size; //x + tex_pos.y = int(probe_index) / (params.probe_axis_size * params.probe_axis_size); + tex_pos.x += params.probe_axis_size * ((int(probe_index) / params.probe_axis_size) % params.probe_axis_size); //z + tex_pos.z = int(params.cascade); + + vec3 tex_pos_ofs = vec3(octahedron_encode(normal_interp) * float(OCT_SIZE), 0.0); + vec3 tex_posf = vec3(vec2(tex_pos.xy * (OCT_SIZE + 2) + ivec2(1)), float(tex_pos.z)) + tex_pos_ofs; + + tex_posf.xy /= vec2(ivec2(params.probe_axis_size * params.probe_axis_size * (OCT_SIZE + 2), params.probe_axis_size * (OCT_SIZE + 2))); + + vec4 indirect_light = textureLod(sampler2DArray(lightprobe_texture, linear_sampler), tex_posf, 0.0); + + frag_color = indirect_light; + +#endif + +#ifdef MODE_VISIBILITY + + frag_color = vec4(vec3(1, visibility, visibility), 1.0); +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/environment/sdfgi_direct_light.glsl b/servers/rendering/renderer_rd/shaders/environment/sdfgi_direct_light.glsl new file mode 100644 index 0000000000..b95fad650e --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/environment/sdfgi_direct_light.glsl @@ -0,0 +1,506 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in; + +#define MAX_CASCADES 8 + +layout(set = 0, binding = 1) uniform texture3D sdf_cascades[MAX_CASCADES]; +layout(set = 0, binding = 2) uniform sampler linear_sampler; + +layout(set = 0, binding = 3, std430) restrict readonly buffer DispatchData { + uint x; + uint y; + uint z; + uint total_count; +} +dispatch_data; + +struct ProcessVoxel { + uint position; // xyz 7 bit packed, extra 11 bits for neighbors. + uint albedo; // rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neighbors. + uint light; // rgbe8985 encoded total saved light, extra 2 bits for neighbors. + uint light_aniso; // 55555 light anisotropy, extra 2 bits for neighbors. + //total neighbours: 26 +}; + +#ifdef MODE_PROCESS_STATIC +layout(set = 0, binding = 4, std430) restrict buffer ProcessVoxels { +#else +layout(set = 0, binding = 4, std430) restrict buffer readonly ProcessVoxels { +#endif + ProcessVoxel data[]; +} +process_voxels; + +layout(r32ui, set = 0, binding = 5) uniform restrict uimage3D dst_light; +layout(rgba8, set = 0, binding = 6) uniform restrict image3D dst_aniso0; +layout(rg8, set = 0, binding = 7) uniform restrict image3D dst_aniso1; + +struct CascadeData { + vec3 offset; //offset of (0,0,0) in world coordinates + float to_cell; // 1/bounds * grid_size + ivec3 probe_world_offset; + uint pad; +}; + +layout(set = 0, binding = 8, std140) uniform Cascades { + CascadeData data[MAX_CASCADES]; +} +cascades; + +#define LIGHT_TYPE_DIRECTIONAL 0 +#define LIGHT_TYPE_OMNI 1 +#define LIGHT_TYPE_SPOT 2 + +struct Light { + vec3 color; + float energy; + + vec3 direction; + bool has_shadow; + + vec3 position; + float attenuation; + + uint type; + float cos_spot_angle; + float inv_spot_attenuation; + float radius; +}; + +layout(set = 0, binding = 9, std140) buffer restrict readonly Lights { + Light data[]; +} +lights; + +layout(set = 0, binding = 10) uniform texture2DArray lightprobe_texture; +layout(set = 0, binding = 11) uniform texture3D occlusion_texture; + +layout(push_constant, std430) uniform Params { + vec3 grid_size; + uint max_cascades; + + uint cascade; + uint light_count; + uint process_offset; + uint process_increment; + + int probe_axis_size; + float bounce_feedback; + float y_mult; + bool use_occlusion; +} +params; + +vec2 octahedron_wrap(vec2 v) { + vec2 signVal; + signVal.x = v.x >= 0.0 ? 1.0 : -1.0; + signVal.y = v.y >= 0.0 ? 1.0 : -1.0; + return (1.0 - abs(v.yx)) * signVal; +} + +vec2 octahedron_encode(vec3 n) { + // https://twitter.com/Stubbesaurus/status/937994790553227264 + n /= (abs(n.x) + abs(n.y) + abs(n.z)); + n.xy = n.z >= 0.0 ? n.xy : octahedron_wrap(n.xy); + n.xy = n.xy * 0.5 + 0.5; + return n.xy; +} + +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); +} + +void main() { + uint voxel_index = uint(gl_GlobalInvocationID.x); + + //used for skipping voxels every N frames + if (params.process_increment > 1) { + voxel_index *= params.process_increment; + voxel_index += params.process_offset; + } + + if (voxel_index >= dispatch_data.total_count) { + return; + } + + uint voxel_position = process_voxels.data[voxel_index].position; + + //keep for storing to texture + ivec3 positioni = ivec3((uvec3(voxel_position, voxel_position, voxel_position) >> uvec3(0, 7, 14)) & uvec3(0x7F)); + + vec3 position = vec3(positioni) + vec3(0.5); + position /= cascades.data[params.cascade].to_cell; + position += cascades.data[params.cascade].offset; + + uint voxel_albedo = process_voxels.data[voxel_index].albedo; + + vec3 albedo = vec3(uvec3(voxel_albedo >> 10, voxel_albedo >> 5, voxel_albedo) & uvec3(0x1F)) / float(0x1F); + vec3 light_accum[6] = vec3[](vec3(0.0), vec3(0.0), vec3(0.0), vec3(0.0), vec3(0.0), vec3(0.0)); + uint valid_aniso = (voxel_albedo >> 15) & 0x3F; + + const vec3 aniso_dir[6] = vec3[]( + vec3(1, 0, 0), + vec3(0, 1, 0), + vec3(0, 0, 1), + vec3(-1, 0, 0), + vec3(0, -1, 0), + vec3(0, 0, -1)); + + // Add indirect light first, in order to save computation resources +#ifdef MODE_PROCESS_DYNAMIC + if (params.bounce_feedback > 0.001) { + vec3 feedback = (params.bounce_feedback < 1.0) ? (albedo * params.bounce_feedback) : mix(albedo, vec3(1.0), params.bounce_feedback - 1.0); + vec3 pos = (vec3(positioni) + vec3(0.5)) * float(params.probe_axis_size - 1) / params.grid_size; + ivec3 probe_base_pos = ivec3(pos); + + float weight_accum[6] = float[](0, 0, 0, 0, 0, 0); + + ivec3 tex_pos = ivec3(probe_base_pos.xy, int(params.cascade)); + tex_pos.x += probe_base_pos.z * int(params.probe_axis_size); + + tex_pos.xy = tex_pos.xy * (OCT_SIZE + 2) + ivec2(1); + + vec3 base_tex_posf = vec3(tex_pos); + vec2 tex_pixel_size = 1.0 / vec2(ivec2((OCT_SIZE + 2) * params.probe_axis_size * params.probe_axis_size, (OCT_SIZE + 2) * params.probe_axis_size)); + vec3 probe_uv_offset = vec3(ivec3(OCT_SIZE + 2, OCT_SIZE + 2, (OCT_SIZE + 2) * params.probe_axis_size)) * tex_pixel_size.xyx; + + for (uint j = 0; j < 8; j++) { + ivec3 offset = (ivec3(j) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1); + ivec3 probe_posi = probe_base_pos; + probe_posi += offset; + + // Compute weight + + vec3 probe_pos = vec3(probe_posi); + vec3 probe_to_pos = pos - probe_pos; + vec3 probe_dir = normalize(-probe_to_pos); + + // Compute lightprobe texture position + + vec3 trilinear = vec3(1.0) - abs(probe_to_pos); + + for (uint k = 0; k < 6; k++) { + if (bool(valid_aniso & (1 << k))) { + vec3 n = aniso_dir[k]; + float weight = trilinear.x * trilinear.y * trilinear.z * max(0, dot(n, probe_dir)); + + if (weight > 0.0 && params.use_occlusion) { + ivec3 occ_indexv = abs((cascades.data[params.cascade].probe_world_offset + probe_posi) & ivec3(1, 1, 1)) * ivec3(1, 2, 4); + vec4 occ_mask = mix(vec4(0.0), vec4(1.0), equal(ivec4(occ_indexv.x | occ_indexv.y), ivec4(0, 1, 2, 3))); + + vec3 occ_pos = (vec3(positioni) + aniso_dir[k] + vec3(0.5)) / params.grid_size; + occ_pos.z += float(params.cascade); + if (occ_indexv.z != 0) { //z bit is on, means index is >=4, so make it switch to the other half of textures + occ_pos.x += 1.0; + } + occ_pos *= vec3(0.5, 1.0, 1.0 / float(params.max_cascades)); //renormalize + float occlusion = dot(textureLod(sampler3D(occlusion_texture, linear_sampler), occ_pos, 0.0), occ_mask); + + weight *= occlusion; + } + + if (weight > 0.0) { + vec3 tex_posf = base_tex_posf + vec3(octahedron_encode(n) * float(OCT_SIZE), 0.0); + tex_posf.xy *= tex_pixel_size; + + vec3 pos_uvw = tex_posf; + pos_uvw.xy += vec2(offset.xy) * probe_uv_offset.xy; + pos_uvw.x += float(offset.z) * probe_uv_offset.z; + vec3 indirect_light = textureLod(sampler2DArray(lightprobe_texture, linear_sampler), pos_uvw, 0.0).rgb; + + light_accum[k] += indirect_light * weight; + weight_accum[k] += weight; + } + } + } + } + + for (uint k = 0; k < 6; k++) { + if (weight_accum[k] > 0.0) { + light_accum[k] /= weight_accum[k]; + light_accum[k] *= feedback; + } + } + } + +#endif + + { + uint rgbe = process_voxels.data[voxel_index].light; + + //read rgbe8985 + float r = float((rgbe & 0xff) << 1); + float g = float((rgbe >> 8) & 0x1ff); + float b = float(((rgbe >> 17) & 0xff) << 1); + float e = float((rgbe >> 25) & 0x1F); + float m = pow(2.0, e - 15.0 - 9.0); + + vec3 l = vec3(r, g, b) * m; + + uint aniso = process_voxels.data[voxel_index].light_aniso; + for (uint i = 0; i < 6; i++) { + float strength = ((aniso >> (i * 5)) & 0x1F) / float(0x1F); + light_accum[i] += l * strength; + } + } + + // Raytrace light + + vec3 pos_to_uvw = 1.0 / params.grid_size; + vec3 uvw_ofs = pos_to_uvw * 0.5; + + for (uint i = 0; i < params.light_count; i++) { + float attenuation = 1.0; + vec3 direction; + float light_distance = 1e20; + + switch (lights.data[i].type) { + case LIGHT_TYPE_DIRECTIONAL: { + direction = -lights.data[i].direction; + } break; + case LIGHT_TYPE_OMNI: { + vec3 rel_vec = lights.data[i].position - position; + direction = normalize(rel_vec); + light_distance = length(rel_vec); + rel_vec.y /= params.y_mult; + attenuation = get_omni_attenuation(light_distance, 1.0 / lights.data[i].radius, lights.data[i].attenuation); + + } break; + case LIGHT_TYPE_SPOT: { + vec3 rel_vec = lights.data[i].position - position; + direction = normalize(rel_vec); + light_distance = length(rel_vec); + rel_vec.y /= params.y_mult; + attenuation = get_omni_attenuation(light_distance, 1.0 / lights.data[i].radius, lights.data[i].attenuation); + + float cos_spot_angle = lights.data[i].cos_spot_angle; + float cos_angle = dot(-direction, lights.data[i].direction); + + if (cos_angle < cos_spot_angle) { + continue; + } + + 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, lights.data[i].inv_spot_attenuation); + } break; + } + + if (attenuation < 0.001) { + continue; + } + + bool hit = false; + + vec3 ray_pos = position; + vec3 ray_dir = direction; + vec3 inv_dir = 1.0 / ray_dir; + + //this is how to properly bias outgoing rays + float cell_size = 1.0 / cascades.data[params.cascade].to_cell; + ray_pos += sign(direction) * cell_size * 0.48; // go almost to the box edge but remain inside + ray_pos += ray_dir * 0.4 * cell_size; //apply a small bias from there + + for (uint j = params.cascade; j < params.max_cascades; j++) { + //convert to local bounds + vec3 pos = ray_pos - cascades.data[j].offset; + pos *= cascades.data[j].to_cell; + float local_distance = light_distance * cascades.data[j].to_cell; + + if (any(lessThan(pos, vec3(0.0))) || any(greaterThanEqual(pos, params.grid_size))) { + continue; //already past bounds for this cascade, goto next + } + + //find maximum advance distance (until reaching bounds) + vec3 t0 = -pos * inv_dir; + vec3 t1 = (params.grid_size - pos) * inv_dir; + vec3 tmax = max(t0, t1); + float max_advance = min(tmax.x, min(tmax.y, tmax.z)); + + max_advance = min(local_distance, max_advance); + + float advance = 0.0; + float occlusion = 1.0; + + while (advance < max_advance) { + //read how much to advance from SDF + vec3 uvw = (pos + ray_dir * advance) * pos_to_uvw; + + float distance = texture(sampler3D(sdf_cascades[j], linear_sampler), uvw).r * 255.0 - 1.0; + if (distance < 0.001) { + //consider hit + hit = true; + break; + } + + occlusion = min(occlusion, distance); + + advance += distance; + } + + if (hit) { + attenuation *= occlusion; + break; + } + + if (advance >= local_distance) { + break; //past light distance, abandon search + } + //change ray origin to collision with bounds + pos += ray_dir * max_advance; + pos /= cascades.data[j].to_cell; + pos += cascades.data[j].offset; + light_distance -= max_advance / cascades.data[j].to_cell; + ray_pos = pos; + } + + if (!hit) { + vec3 light = albedo * lights.data[i].color.rgb * lights.data[i].energy * attenuation; + + for (int j = 0; j < 6; j++) { + if (bool(valid_aniso & (1 << j))) { + light_accum[j] += max(0.0, dot(aniso_dir[j], direction)) * light; + } + } + } + } + + // Store the light in the light texture + + float lumas[6]; + vec3 light_total = vec3(0); + + for (int i = 0; i < 6; i++) { + light_total += light_accum[i]; + lumas[i] = max(light_accum[i].r, max(light_accum[i].g, light_accum[i].b)); + } + + float luma_total = max(light_total.r, max(light_total.g, light_total.b)); + + uint light_total_rgbe; + + { + //compress to RGBE9995 to save space + + const float pow2to9 = 512.0f; + const float B = 15.0f; + const float N = 9.0f; + const float LN2 = 0.6931471805599453094172321215; + + float cRed = clamp(light_total.r, 0.0, 65408.0); + float cGreen = clamp(light_total.g, 0.0, 65408.0); + float cBlue = clamp(light_total.b, 0.0, 65408.0); + + float cMax = max(cRed, max(cGreen, cBlue)); + + float expp = max(-B - 1.0f, floor(log(cMax) / LN2)) + 1.0f + B; + + float sMax = floor((cMax / pow(2.0f, expp - B - N)) + 0.5f); + + float exps = expp + 1.0f; + + if (0.0 <= sMax && sMax < pow2to9) { + exps = expp; + } + + float sRed = floor((cRed / pow(2.0f, exps - B - N)) + 0.5f); + float sGreen = floor((cGreen / pow(2.0f, exps - B - N)) + 0.5f); + float sBlue = floor((cBlue / pow(2.0f, exps - B - N)) + 0.5f); +#ifdef MODE_PROCESS_STATIC + //since its self-save, use RGBE8985 + light_total_rgbe = ((uint(sRed) & 0x1FF) >> 1) | ((uint(sGreen) & 0x1FF) << 8) | (((uint(sBlue) & 0x1FF) >> 1) << 17) | ((uint(exps) & 0x1F) << 25); + +#else + light_total_rgbe = (uint(sRed) & 0x1FF) | ((uint(sGreen) & 0x1FF) << 9) | ((uint(sBlue) & 0x1FF) << 18) | ((uint(exps) & 0x1F) << 27); +#endif + } + +#ifdef MODE_PROCESS_DYNAMIC + + vec4 aniso0; + aniso0.r = lumas[0] / luma_total; + aniso0.g = lumas[1] / luma_total; + aniso0.b = lumas[2] / luma_total; + aniso0.a = lumas[3] / luma_total; + + vec2 aniso1; + aniso1.r = lumas[4] / luma_total; + aniso1.g = lumas[5] / luma_total; + + //save to 3D textures + imageStore(dst_aniso0, positioni, aniso0); + imageStore(dst_aniso1, positioni, vec4(aniso1, 0.0, 0.0)); + imageStore(dst_light, positioni, uvec4(light_total_rgbe)); + + //also fill neighbours, so light interpolation during the indirect pass works + + //recover the neighbour list from the leftover bits + uint neighbours = (voxel_albedo >> 21) | ((voxel_position >> 21) << 11) | ((process_voxels.data[voxel_index].light >> 30) << 22) | ((process_voxels.data[voxel_index].light_aniso >> 30) << 24); + + const uint max_neighbours = 26; + const ivec3 neighbour_positions[max_neighbours] = ivec3[]( + ivec3(-1, -1, -1), + ivec3(-1, -1, 0), + ivec3(-1, -1, 1), + ivec3(-1, 0, -1), + ivec3(-1, 0, 0), + ivec3(-1, 0, 1), + ivec3(-1, 1, -1), + ivec3(-1, 1, 0), + ivec3(-1, 1, 1), + ivec3(0, -1, -1), + ivec3(0, -1, 0), + ivec3(0, -1, 1), + ivec3(0, 0, -1), + ivec3(0, 0, 1), + ivec3(0, 1, -1), + ivec3(0, 1, 0), + ivec3(0, 1, 1), + ivec3(1, -1, -1), + ivec3(1, -1, 0), + ivec3(1, -1, 1), + ivec3(1, 0, -1), + ivec3(1, 0, 0), + ivec3(1, 0, 1), + ivec3(1, 1, -1), + ivec3(1, 1, 0), + ivec3(1, 1, 1)); + + for (uint i = 0; i < max_neighbours; i++) { + if (bool(neighbours & (1 << i))) { + ivec3 neighbour_pos = positioni + neighbour_positions[i]; + imageStore(dst_light, neighbour_pos, uvec4(light_total_rgbe)); + imageStore(dst_aniso0, neighbour_pos, aniso0); + imageStore(dst_aniso1, neighbour_pos, vec4(aniso1, 0.0, 0.0)); + } + } + +#endif + +#ifdef MODE_PROCESS_STATIC + + //save back the anisotropic + + uint light = process_voxels.data[voxel_index].light & (3 << 30); + light |= light_total_rgbe; + process_voxels.data[voxel_index].light = light; //replace + + uint light_aniso = process_voxels.data[voxel_index].light_aniso & (3 << 30); + for (int i = 0; i < 6; i++) { + light_aniso |= min(31, uint((lumas[i] / luma_total) * 31.0)) << (i * 5); + } + + process_voxels.data[voxel_index].light_aniso = light_aniso; + +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/environment/sdfgi_integrate.glsl b/servers/rendering/renderer_rd/shaders/environment/sdfgi_integrate.glsl new file mode 100644 index 0000000000..9c03297f5c --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/environment/sdfgi_integrate.glsl @@ -0,0 +1,612 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +#define MAX_CASCADES 8 + +layout(set = 0, binding = 1) uniform texture3D sdf_cascades[MAX_CASCADES]; +layout(set = 0, binding = 2) uniform texture3D light_cascades[MAX_CASCADES]; +layout(set = 0, binding = 3) uniform texture3D aniso0_cascades[MAX_CASCADES]; +layout(set = 0, binding = 4) uniform texture3D aniso1_cascades[MAX_CASCADES]; + +layout(set = 0, binding = 6) uniform sampler linear_sampler; + +struct CascadeData { + vec3 offset; //offset of (0,0,0) in world coordinates + float to_cell; // 1/bounds * grid_size + ivec3 probe_world_offset; + uint pad; +}; + +layout(set = 0, binding = 7, std140) uniform Cascades { + CascadeData data[MAX_CASCADES]; +} +cascades; + +layout(r32ui, set = 0, binding = 8) uniform restrict uimage2DArray lightprobe_texture_data; +layout(rgba16i, set = 0, binding = 9) uniform restrict iimage2DArray lightprobe_history_texture; +layout(rgba32i, set = 0, binding = 10) uniform restrict iimage2D lightprobe_average_texture; + +//used for scrolling +layout(rgba16i, set = 0, binding = 11) uniform restrict iimage2DArray lightprobe_history_scroll_texture; +layout(rgba32i, set = 0, binding = 12) uniform restrict iimage2D lightprobe_average_scroll_texture; + +layout(rgba32i, set = 0, binding = 13) uniform restrict iimage2D lightprobe_average_parent_texture; + +layout(rgba16f, set = 0, binding = 14) uniform restrict writeonly image2DArray lightprobe_ambient_texture; + +#ifdef USE_CUBEMAP_ARRAY +layout(set = 1, binding = 0) uniform textureCubeArray sky_irradiance; +#else +layout(set = 1, binding = 0) uniform textureCube sky_irradiance; +#endif +layout(set = 1, binding = 1) uniform sampler linear_sampler_mipmaps; + +#define HISTORY_BITS 10 + +#define SKY_MODE_DISABLED 0 +#define SKY_MODE_COLOR 1 +#define SKY_MODE_SKY 2 + +layout(push_constant, std430) uniform Params { + vec3 grid_size; + uint max_cascades; + + uint probe_axis_size; + uint cascade; + uint history_index; + uint history_size; + + uint ray_count; + float ray_bias; + ivec2 image_size; + + ivec3 world_offset; + uint sky_mode; + + ivec3 scroll; + float sky_energy; + + vec3 sky_color; + float y_mult; + + bool store_ambient_texture; + uint pad[3]; +} +params; + +const float PI = 3.14159265f; +const float GOLDEN_ANGLE = PI * (3.0 - sqrt(5.0)); + +vec3 vogel_hemisphere(uint p_index, uint p_count, float p_offset) { + float r = sqrt(float(p_index) + 0.5f) / sqrt(float(p_count)); + float theta = float(p_index) * GOLDEN_ANGLE + p_offset; + float y = cos(r * PI * 0.5); + float l = sin(r * PI * 0.5); + return vec3(l * cos(theta), l * sin(theta), y * (float(p_index & 1) * 2.0 - 1.0)); +} + +uvec3 hash3(uvec3 x) { + x = ((x >> 16) ^ x) * 0x45d9f3b; + x = ((x >> 16) ^ x) * 0x45d9f3b; + x = (x >> 16) ^ x; + return x; +} + +float hashf3(vec3 co) { + return fract(sin(dot(co, vec3(12.9898, 78.233, 137.13451))) * 43758.5453); +} + +vec3 octahedron_encode(vec2 f) { + // https://twitter.com/Stubbesaurus/status/937994790553227264 + f = f * 2.0 - 1.0; + vec3 n = vec3(f.x, f.y, 1.0f - abs(f.x) - abs(f.y)); + float t = clamp(-n.z, 0.0, 1.0); + n.x += n.x >= 0 ? -t : t; + n.y += n.y >= 0 ? -t : t; + return normalize(n); +} + +uint rgbe_encode(vec3 color) { + const float pow2to9 = 512.0f; + const float B = 15.0f; + const float N = 9.0f; + const float LN2 = 0.6931471805599453094172321215; + + float cRed = clamp(color.r, 0.0, 65408.0); + float cGreen = clamp(color.g, 0.0, 65408.0); + float cBlue = clamp(color.b, 0.0, 65408.0); + + float cMax = max(cRed, max(cGreen, cBlue)); + + float expp = max(-B - 1.0f, floor(log(cMax) / LN2)) + 1.0f + B; + + float sMax = floor((cMax / pow(2.0f, expp - B - N)) + 0.5f); + + float exps = expp + 1.0f; + + if (0.0 <= sMax && sMax < pow2to9) { + exps = expp; + } + + float sRed = floor((cRed / pow(2.0f, exps - B - N)) + 0.5f); + float sGreen = floor((cGreen / pow(2.0f, exps - B - N)) + 0.5f); + float sBlue = floor((cBlue / pow(2.0f, exps - B - N)) + 0.5f); + return (uint(sRed) & 0x1FF) | ((uint(sGreen) & 0x1FF) << 9) | ((uint(sBlue) & 0x1FF) << 18) | ((uint(exps) & 0x1F) << 27); +} + +struct SH { +#if (SH_SIZE == 16) + float c[48]; +#else + float c[28]; +#endif +}; + +shared SH sh_accum[64]; //8x8 + +void main() { + ivec2 pos = ivec2(gl_GlobalInvocationID.xy); + if (any(greaterThanEqual(pos, params.image_size))) { //too large, do nothing + return; + } + + uint probe_index = gl_LocalInvocationID.x + gl_LocalInvocationID.y * 8; + +#ifdef MODE_PROCESS + + float probe_cell_size = float(params.grid_size.x / float(params.probe_axis_size - 1)) / cascades.data[params.cascade].to_cell; + + ivec3 probe_cell; + probe_cell.x = pos.x % int(params.probe_axis_size); + probe_cell.y = pos.y; + probe_cell.z = pos.x / int(params.probe_axis_size); + + vec3 probe_pos = cascades.data[params.cascade].offset + vec3(probe_cell) * probe_cell_size; + vec3 pos_to_uvw = 1.0 / params.grid_size; + + for (uint i = 0; i < SH_SIZE * 3; i++) { + sh_accum[probe_index].c[i] = 0.0; + } + + // quickly ensure each probe has a different "offset" for the vogel function, based on integer world position + uvec3 h3 = hash3(uvec3(params.world_offset + probe_cell)); + float offset = hashf3(vec3(h3 & uvec3(0xFFFFF))); + + //for a more homogeneous hemisphere, alternate based on history frames + uint ray_offset = params.history_index; + uint ray_mult = params.history_size; + uint ray_total = ray_mult * params.ray_count; + + for (uint i = 0; i < params.ray_count; i++) { + vec3 ray_dir = vogel_hemisphere(ray_offset + i * ray_mult, ray_total, offset); + ray_dir.y *= params.y_mult; + ray_dir = normalize(ray_dir); + + //needs to be visible + vec3 ray_pos = probe_pos; + vec3 inv_dir = 1.0 / ray_dir; + + bool hit = false; + uint hit_cascade; + + float bias = params.ray_bias; + vec3 abs_ray_dir = abs(ray_dir); + ray_pos += ray_dir * 1.0 / max(abs_ray_dir.x, max(abs_ray_dir.y, abs_ray_dir.z)) * bias / cascades.data[params.cascade].to_cell; + vec3 uvw; + + for (uint j = params.cascade; j < params.max_cascades; j++) { + //convert to local bounds + vec3 pos = ray_pos - cascades.data[j].offset; + pos *= cascades.data[j].to_cell; + + if (any(lessThan(pos, vec3(0.0))) || any(greaterThanEqual(pos, params.grid_size))) { + continue; //already past bounds for this cascade, goto next + } + + //find maximum advance distance (until reaching bounds) + vec3 t0 = -pos * inv_dir; + vec3 t1 = (params.grid_size - pos) * inv_dir; + vec3 tmax = max(t0, t1); + float max_advance = min(tmax.x, min(tmax.y, tmax.z)); + + float advance = 0.0; + + while (advance < max_advance) { + //read how much to advance from SDF + uvw = (pos + ray_dir * advance) * pos_to_uvw; + + float distance = texture(sampler3D(sdf_cascades[j], linear_sampler), uvw).r * 255.0 - 1.0; + if (distance < 0.05) { + //consider hit + hit = true; + break; + } + + advance += distance; + } + + if (hit) { + hit_cascade = j; + break; + } + + //change ray origin to collision with bounds + pos += ray_dir * max_advance; + pos /= cascades.data[j].to_cell; + pos += cascades.data[j].offset; + ray_pos = pos; + } + + vec4 light; + if (hit) { + //avoid reading different texture from different threads + for (uint j = params.cascade; j < params.max_cascades; j++) { + if (j == hit_cascade) { + const float EPSILON = 0.001; + vec3 hit_normal = normalize(vec3( + texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw - vec3(EPSILON, 0.0, 0.0)).r, + texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw + vec3(0.0, EPSILON, 0.0)).r - texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw - vec3(0.0, EPSILON, 0.0)).r, + texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw + vec3(0.0, 0.0, EPSILON)).r - texture(sampler3D(sdf_cascades[hit_cascade], linear_sampler), uvw - vec3(0.0, 0.0, EPSILON)).r)); + + vec3 hit_light = texture(sampler3D(light_cascades[hit_cascade], linear_sampler), uvw).rgb; + vec4 aniso0 = texture(sampler3D(aniso0_cascades[hit_cascade], linear_sampler), uvw); + vec3 hit_aniso0 = aniso0.rgb; + vec3 hit_aniso1 = vec3(aniso0.a, texture(sampler3D(aniso1_cascades[hit_cascade], linear_sampler), uvw).rg); + + //one liner magic + light.rgb = hit_light * (dot(max(vec3(0.0), (hit_normal * hit_aniso0)), vec3(1.0)) + dot(max(vec3(0.0), (-hit_normal * hit_aniso1)), vec3(1.0))); + light.a = 1.0; + } + } + + } else if (params.sky_mode == SKY_MODE_SKY) { +#ifdef USE_CUBEMAP_ARRAY + light.rgb = textureLod(samplerCubeArray(sky_irradiance, linear_sampler_mipmaps), vec4(ray_dir, 0.0), 2.0).rgb; // Use second mipmap because we don't usually throw a lot of rays, so this compensates. +#else + light.rgb = textureLod(samplerCube(sky_irradiance, linear_sampler_mipmaps), ray_dir, 2.0).rgb; // Use second mipmap because we don't usually throw a lot of rays, so this compensates. +#endif + light.rgb *= params.sky_energy; + light.a = 0.0; + + } else if (params.sky_mode == SKY_MODE_COLOR) { + light.rgb = params.sky_color; + light.rgb *= params.sky_energy; + light.a = 0.0; + } else { + light = vec4(0, 0, 0, 0); + } + + vec3 ray_dir2 = ray_dir * ray_dir; + +#define SH_ACCUM(m_idx, m_value) \ + { \ + vec3 l = light.rgb * (m_value); \ + sh_accum[probe_index].c[m_idx * 3 + 0] += l.r; \ + sh_accum[probe_index].c[m_idx * 3 + 1] += l.g; \ + sh_accum[probe_index].c[m_idx * 3 + 2] += l.b; \ + } + SH_ACCUM(0, 0.282095); //l0 + SH_ACCUM(1, 0.488603 * ray_dir.y); //l1n1 + SH_ACCUM(2, 0.488603 * ray_dir.z); //l1n0 + SH_ACCUM(3, 0.488603 * ray_dir.x); //l1p1 + SH_ACCUM(4, 1.092548 * ray_dir.x * ray_dir.y); //l2n2 + SH_ACCUM(5, 1.092548 * ray_dir.y * ray_dir.z); //l2n1 + SH_ACCUM(6, 0.315392 * (3.0 * ray_dir2.z - 1.0)); //l20 + SH_ACCUM(7, 1.092548 * ray_dir.x * ray_dir.z); //l2p1 + SH_ACCUM(8, 0.546274 * (ray_dir2.x - ray_dir2.y)); //l2p2 +#if (SH_SIZE == 16) + SH_ACCUM(9, 0.590043 * ray_dir.y * (3.0f * ray_dir2.x - ray_dir2.y)); + SH_ACCUM(10, 2.890611 * ray_dir.y * ray_dir.x * ray_dir.z); + SH_ACCUM(11, 0.646360 * ray_dir.y * (-1.0f + 5.0f * ray_dir2.z)); + SH_ACCUM(12, 0.373176 * (5.0f * ray_dir2.z * ray_dir.z - 3.0f * ray_dir.z)); + SH_ACCUM(13, 0.457045 * ray_dir.x * (-1.0f + 5.0f * ray_dir2.z)); + SH_ACCUM(14, 1.445305 * (ray_dir2.x - ray_dir2.y) * ray_dir.z); + SH_ACCUM(15, 0.590043 * ray_dir.x * (ray_dir2.x - 3.0f * ray_dir2.y)); + +#endif + } + + for (uint i = 0; i < SH_SIZE; i++) { + // store in history texture + ivec3 prev_pos = ivec3(pos.x, pos.y * SH_SIZE + i, int(params.history_index)); + ivec2 average_pos = prev_pos.xy; + + vec4 value = vec4(sh_accum[probe_index].c[i * 3 + 0], sh_accum[probe_index].c[i * 3 + 1], sh_accum[probe_index].c[i * 3 + 2], 1.0) * 4.0 / float(params.ray_count); + + ivec4 ivalue = clamp(ivec4(value * float(1 << HISTORY_BITS)), -32768, 32767); //clamp to 16 bits, so higher values don't break average + + ivec4 prev_value = imageLoad(lightprobe_history_texture, prev_pos); + ivec4 average = imageLoad(lightprobe_average_texture, average_pos); + + average -= prev_value; + average += ivalue; + + imageStore(lightprobe_history_texture, prev_pos, ivalue); + imageStore(lightprobe_average_texture, average_pos, average); + + if (params.store_ambient_texture && i == 0) { + ivec3 ambient_pos = ivec3(pos, int(params.cascade)); + vec4 ambient_light = (vec4(average) / float(params.history_size)) / float(1 << HISTORY_BITS); + ambient_light *= 0.88622; // SHL0 + imageStore(lightprobe_ambient_texture, ambient_pos, ambient_light); + } + } +#endif // MODE PROCESS + +#ifdef MODE_STORE + + // converting to octahedral in this step is required because + // octahedral is much faster to read from the screen than spherical harmonics, + // despite the very slight quality loss + + ivec2 sh_pos = (pos / OCT_SIZE) * ivec2(1, SH_SIZE); + ivec2 oct_pos = (pos / OCT_SIZE) * (OCT_SIZE + 2) + ivec2(1); + ivec2 local_pos = pos % OCT_SIZE; + + //compute the octahedral normal for this texel + vec3 normal = octahedron_encode(vec2(local_pos) / float(OCT_SIZE)); + + // read the spherical harmonic + + vec3 normal2 = normal * normal; + float c[SH_SIZE] = float[]( + + 0.282095, //l0 + 0.488603 * normal.y, //l1n1 + 0.488603 * normal.z, //l1n0 + 0.488603 * normal.x, //l1p1 + 1.092548 * normal.x * normal.y, //l2n2 + 1.092548 * normal.y * normal.z, //l2n1 + 0.315392 * (3.0 * normal2.z - 1.0), //l20 + 1.092548 * normal.x * normal.z, //l2p1 + 0.546274 * (normal2.x - normal2.y) //l2p2 +#if (SH_SIZE == 16) + , + 0.590043 * normal.y * (3.0f * normal2.x - normal2.y), + 2.890611 * normal.y * normal.x * normal.z, + 0.646360 * normal.y * (-1.0f + 5.0f * normal2.z), + 0.373176 * (5.0f * normal2.z * normal.z - 3.0f * normal.z), + 0.457045 * normal.x * (-1.0f + 5.0f * normal2.z), + 1.445305 * (normal2.x - normal2.y) * normal.z, + 0.590043 * normal.x * (normal2.x - 3.0f * normal2.y) + +#endif + ); + + const float l_mult[SH_SIZE] = float[]( + 1.0, + 2.0 / 3.0, + 2.0 / 3.0, + 2.0 / 3.0, + 1.0 / 4.0, + 1.0 / 4.0, + 1.0 / 4.0, + 1.0 / 4.0, + 1.0 / 4.0 +#if (SH_SIZE == 16) + , // l4 does not contribute to irradiance + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0, + 0.0 +#endif + ); + + vec3 irradiance = vec3(0.0); + vec3 radiance = vec3(0.0); + + for (uint i = 0; i < SH_SIZE; i++) { + // store in history texture + ivec2 average_pos = sh_pos + ivec2(0, i); + ivec4 average = imageLoad(lightprobe_average_texture, average_pos); + + vec4 sh = (vec4(average) / float(params.history_size)) / float(1 << HISTORY_BITS); + + vec3 m = sh.rgb * c[i] * 4.0; + + irradiance += m * l_mult[i]; + radiance += m; + } + + //encode RGBE9995 for the final texture + + uint irradiance_rgbe = rgbe_encode(irradiance); + uint radiance_rgbe = rgbe_encode(radiance); + + //store in octahedral map + + ivec3 texture_pos = ivec3(oct_pos, int(params.cascade)); + ivec3 copy_to[4] = ivec3[](ivec3(-2, -2, -2), ivec3(-2, -2, -2), ivec3(-2, -2, -2), ivec3(-2, -2, -2)); + copy_to[0] = texture_pos + ivec3(local_pos, 0); + + if (local_pos == ivec2(0, 0)) { + copy_to[1] = texture_pos + ivec3(OCT_SIZE - 1, -1, 0); + copy_to[2] = texture_pos + ivec3(-1, OCT_SIZE - 1, 0); + copy_to[3] = texture_pos + ivec3(OCT_SIZE, OCT_SIZE, 0); + } else if (local_pos == ivec2(OCT_SIZE - 1, 0)) { + copy_to[1] = texture_pos + ivec3(0, -1, 0); + copy_to[2] = texture_pos + ivec3(OCT_SIZE, OCT_SIZE - 1, 0); + copy_to[3] = texture_pos + ivec3(-1, OCT_SIZE, 0); + } else if (local_pos == ivec2(0, OCT_SIZE - 1)) { + copy_to[1] = texture_pos + ivec3(-1, 0, 0); + copy_to[2] = texture_pos + ivec3(OCT_SIZE - 1, OCT_SIZE, 0); + copy_to[3] = texture_pos + ivec3(OCT_SIZE, -1, 0); + } else if (local_pos == ivec2(OCT_SIZE - 1, OCT_SIZE - 1)) { + copy_to[1] = texture_pos + ivec3(0, OCT_SIZE, 0); + copy_to[2] = texture_pos + ivec3(OCT_SIZE, 0, 0); + copy_to[3] = texture_pos + ivec3(-1, -1, 0); + } else if (local_pos.y == 0) { + copy_to[1] = texture_pos + ivec3(OCT_SIZE - local_pos.x - 1, local_pos.y - 1, 0); + } else if (local_pos.x == 0) { + copy_to[1] = texture_pos + ivec3(local_pos.x - 1, OCT_SIZE - local_pos.y - 1, 0); + } else if (local_pos.y == OCT_SIZE - 1) { + copy_to[1] = texture_pos + ivec3(OCT_SIZE - local_pos.x - 1, local_pos.y + 1, 0); + } else if (local_pos.x == OCT_SIZE - 1) { + copy_to[1] = texture_pos + ivec3(local_pos.x + 1, OCT_SIZE - local_pos.y - 1, 0); + } + + for (int i = 0; i < 4; i++) { + if (copy_to[i] == ivec3(-2, -2, -2)) { + continue; + } + imageStore(lightprobe_texture_data, copy_to[i], uvec4(irradiance_rgbe)); + imageStore(lightprobe_texture_data, copy_to[i] + ivec3(0, 0, int(params.max_cascades)), uvec4(radiance_rgbe)); + } + +#endif + +#ifdef MODE_SCROLL + + ivec3 probe_cell; + probe_cell.x = pos.x % int(params.probe_axis_size); + probe_cell.y = pos.y; + probe_cell.z = pos.x / int(params.probe_axis_size); + + ivec3 read_probe = probe_cell - params.scroll; + + if (all(greaterThanEqual(read_probe, ivec3(0))) && all(lessThan(read_probe, ivec3(params.probe_axis_size)))) { + // can scroll + ivec2 tex_pos; + tex_pos = read_probe.xy; + tex_pos.x += read_probe.z * int(params.probe_axis_size); + + //scroll + for (uint j = 0; j < params.history_size; j++) { + for (int i = 0; i < SH_SIZE; i++) { + // copy from history texture + ivec3 src_pos = ivec3(tex_pos.x, tex_pos.y * SH_SIZE + i, int(j)); + ivec3 dst_pos = ivec3(pos.x, pos.y * SH_SIZE + i, int(j)); + ivec4 value = imageLoad(lightprobe_history_texture, src_pos); + imageStore(lightprobe_history_scroll_texture, dst_pos, value); + } + } + + for (int i = 0; i < SH_SIZE; i++) { + // copy from average texture + ivec2 src_pos = ivec2(tex_pos.x, tex_pos.y * SH_SIZE + i); + ivec2 dst_pos = ivec2(pos.x, pos.y * SH_SIZE + i); + ivec4 value = imageLoad(lightprobe_average_texture, src_pos); + imageStore(lightprobe_average_scroll_texture, dst_pos, value); + } + } else if (params.cascade < params.max_cascades - 1) { + //can't scroll, must look for position in parent cascade + + //to global coords + float cell_to_probe = float(params.grid_size.x / float(params.probe_axis_size - 1)); + + float probe_cell_size = cell_to_probe / cascades.data[params.cascade].to_cell; + vec3 probe_pos = cascades.data[params.cascade].offset + vec3(probe_cell) * probe_cell_size; + + //to parent local coords + float probe_cell_size_next = cell_to_probe / cascades.data[params.cascade + 1].to_cell; + probe_pos -= cascades.data[params.cascade + 1].offset; + probe_pos /= probe_cell_size_next; + + ivec3 probe_posi = ivec3(probe_pos); + //add up all light, no need to use occlusion here, since occlusion will do its work afterwards + + vec4 average_light[SH_SIZE] = vec4[](vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0) +#if (SH_SIZE == 16) + , + vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0), vec4(0) +#endif + ); + float total_weight = 0.0; + + for (int i = 0; i < 8; i++) { + ivec3 offset = probe_posi + ((ivec3(i) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1)); + + vec3 trilinear = vec3(1.0) - abs(probe_pos - vec3(offset)); + float weight = trilinear.x * trilinear.y * trilinear.z; + + ivec2 tex_pos; + tex_pos = offset.xy; + tex_pos.x += offset.z * int(params.probe_axis_size); + + for (int j = 0; j < SH_SIZE; j++) { + // copy from history texture + ivec2 src_pos = ivec2(tex_pos.x, tex_pos.y * SH_SIZE + j); + ivec4 average = imageLoad(lightprobe_average_parent_texture, src_pos); + vec4 value = (vec4(average) / float(params.history_size)) / float(1 << HISTORY_BITS); + average_light[j] += value * weight; + } + + total_weight += weight; + } + + if (total_weight > 0.0) { + total_weight = 1.0 / total_weight; + } + //store the averaged values everywhere + + for (int i = 0; i < SH_SIZE; i++) { + ivec4 ivalue = clamp(ivec4(average_light[i] * total_weight * float(1 << HISTORY_BITS)), ivec4(-32768), ivec4(32767)); //clamp to 16 bits, so higher values don't break average + // copy from history texture + ivec3 dst_pos = ivec3(pos.x, pos.y * SH_SIZE + i, 0); + for (uint j = 0; j < params.history_size; j++) { + dst_pos.z = int(j); + imageStore(lightprobe_history_scroll_texture, dst_pos, ivalue); + } + + ivalue *= int(params.history_size); //average needs to have all history added up + imageStore(lightprobe_average_scroll_texture, dst_pos.xy, ivalue); + } + + } else { + //scroll at the edge of the highest cascade, just copy what is there, + //since its the closest we have anyway + + for (uint j = 0; j < params.history_size; j++) { + ivec2 tex_pos; + tex_pos = probe_cell.xy; + tex_pos.x += probe_cell.z * int(params.probe_axis_size); + + for (int i = 0; i < SH_SIZE; i++) { + // copy from history texture + ivec3 src_pos = ivec3(tex_pos.x, tex_pos.y * SH_SIZE + i, int(j)); + ivec3 dst_pos = ivec3(pos.x, pos.y * SH_SIZE + i, int(j)); + ivec4 value = imageLoad(lightprobe_history_texture, dst_pos); + imageStore(lightprobe_history_scroll_texture, dst_pos, value); + } + } + + for (int i = 0; i < SH_SIZE; i++) { + // copy from average texture + ivec2 spos = ivec2(pos.x, pos.y * SH_SIZE + i); + ivec4 average = imageLoad(lightprobe_average_texture, spos); + imageStore(lightprobe_average_scroll_texture, spos, average); + } + } + +#endif + +#ifdef MODE_SCROLL_STORE + + //do not update probe texture, as these will be updated later + + for (uint j = 0; j < params.history_size; j++) { + for (int i = 0; i < SH_SIZE; i++) { + // copy from history texture + ivec3 spos = ivec3(pos.x, pos.y * SH_SIZE + i, int(j)); + ivec4 value = imageLoad(lightprobe_history_scroll_texture, spos); + imageStore(lightprobe_history_texture, spos, value); + } + } + + for (int i = 0; i < SH_SIZE; i++) { + // copy from average texture + ivec2 spos = ivec2(pos.x, pos.y * SH_SIZE + i); + ivec4 average = imageLoad(lightprobe_average_scroll_texture, spos); + imageStore(lightprobe_average_texture, spos, average); + } + +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/environment/sdfgi_preprocess.glsl b/servers/rendering/renderer_rd/shaders/environment/sdfgi_preprocess.glsl new file mode 100644 index 0000000000..bce98f4054 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/environment/sdfgi_preprocess.glsl @@ -0,0 +1,1056 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#ifdef MODE_JUMPFLOOD_OPTIMIZED +#define GROUP_SIZE 8 + +layout(local_size_x = GROUP_SIZE, local_size_y = GROUP_SIZE, local_size_z = GROUP_SIZE) in; + +#elif defined(MODE_OCCLUSION) || defined(MODE_SCROLL) +//buffer layout +layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in; + +#else +//grid layout +layout(local_size_x = 4, local_size_y = 4, local_size_z = 4) in; + +#endif + +#if defined(MODE_INITIALIZE_JUMP_FLOOD) || defined(MODE_INITIALIZE_JUMP_FLOOD_HALF) +layout(r16ui, set = 0, binding = 1) uniform restrict readonly uimage3D src_color; +layout(rgba8ui, set = 0, binding = 2) uniform restrict writeonly uimage3D dst_positions; +#endif + +#ifdef MODE_UPSCALE_JUMP_FLOOD +layout(r16ui, set = 0, binding = 1) uniform restrict readonly uimage3D src_color; +layout(rgba8ui, set = 0, binding = 2) uniform restrict readonly uimage3D src_positions_half; +layout(rgba8ui, set = 0, binding = 3) uniform restrict writeonly uimage3D dst_positions; +#endif + +#if defined(MODE_JUMPFLOOD) || defined(MODE_JUMPFLOOD_OPTIMIZED) +layout(rgba8ui, set = 0, binding = 1) uniform restrict readonly uimage3D src_positions; +layout(rgba8ui, set = 0, binding = 2) uniform restrict writeonly uimage3D dst_positions; +#endif + +#ifdef MODE_JUMPFLOOD_OPTIMIZED + +shared uvec4 group_positions[(GROUP_SIZE + 2) * (GROUP_SIZE + 2) * (GROUP_SIZE + 2)]; //4x4x4 with margins + +void group_store(ivec3 p_pos, uvec4 p_value) { + uint offset = uint(p_pos.z * (GROUP_SIZE + 2) * (GROUP_SIZE + 2) + p_pos.y * (GROUP_SIZE + 2) + p_pos.x); + group_positions[offset] = p_value; +} + +uvec4 group_load(ivec3 p_pos) { + uint offset = uint(p_pos.z * (GROUP_SIZE + 2) * (GROUP_SIZE + 2) + p_pos.y * (GROUP_SIZE + 2) + p_pos.x); + return group_positions[offset]; +} + +#endif + +#ifdef MODE_OCCLUSION + +layout(r16ui, set = 0, binding = 1) uniform restrict readonly uimage3D src_color; +layout(r8, set = 0, binding = 2) uniform restrict image3D dst_occlusion[8]; +layout(r32ui, set = 0, binding = 3) uniform restrict readonly uimage3D src_facing; + +const uvec2 group_size_offset[11] = uvec2[](uvec2(1, 0), uvec2(3, 1), uvec2(6, 4), uvec2(10, 10), uvec2(15, 20), uvec2(21, 35), uvec2(28, 56), uvec2(36, 84), uvec2(42, 120), uvec2(46, 162), uvec2(48, 208)); +const uint group_pos[256] = uint[](0, + 65536, 256, 1, + 131072, 65792, 512, 65537, 257, 2, + 196608, 131328, 66048, 768, 131073, 65793, 513, 65538, 258, 3, + 262144, 196864, 131584, 66304, 1024, 196609, 131329, 66049, 769, 131074, 65794, 514, 65539, 259, 4, + 327680, 262400, 197120, 131840, 66560, 1280, 262145, 196865, 131585, 66305, 1025, 196610, 131330, 66050, 770, 131075, 65795, 515, 65540, 260, 5, + 393216, 327936, 262656, 197376, 132096, 66816, 1536, 327681, 262401, 197121, 131841, 66561, 1281, 262146, 196866, 131586, 66306, 1026, 196611, 131331, 66051, 771, 131076, 65796, 516, 65541, 261, 6, + 458752, 393472, 328192, 262912, 197632, 132352, 67072, 1792, 393217, 327937, 262657, 197377, 132097, 66817, 1537, 327682, 262402, 197122, 131842, 66562, 1282, 262147, 196867, 131587, 66307, 1027, 196612, 131332, 66052, 772, 131077, 65797, 517, 65542, 262, 7, + 459008, 393728, 328448, 263168, 197888, 132608, 67328, 458753, 393473, 328193, 262913, 197633, 132353, 67073, 1793, 393218, 327938, 262658, 197378, 132098, 66818, 1538, 327683, 262403, 197123, 131843, 66563, 1283, 262148, 196868, 131588, 66308, 1028, 196613, 131333, 66053, 773, 131078, 65798, 518, 65543, 263, + 459264, 393984, 328704, 263424, 198144, 132864, 459009, 393729, 328449, 263169, 197889, 132609, 67329, 458754, 393474, 328194, 262914, 197634, 132354, 67074, 1794, 393219, 327939, 262659, 197379, 132099, 66819, 1539, 327684, 262404, 197124, 131844, 66564, 1284, 262149, 196869, 131589, 66309, 1029, 196614, 131334, 66054, 774, 131079, 65799, 519, + 459520, 394240, 328960, 263680, 198400, 459265, 393985, 328705, 263425, 198145, 132865, 459010, 393730, 328450, 263170, 197890, 132610, 67330, 458755, 393475, 328195, 262915, 197635, 132355, 67075, 1795, 393220, 327940, 262660, 197380, 132100, 66820, 1540, 327685, 262405, 197125, 131845, 66565, 1285, 262150, 196870, 131590, 66310, 1030, 196615, 131335, 66055, 775); + +shared uint occlusion_facing[((OCCLUSION_SIZE * 2) * (OCCLUSION_SIZE * 2) * (OCCLUSION_SIZE * 2)) / 4]; + +uint get_facing(ivec3 p_pos) { + uint ofs = uint(p_pos.z * OCCLUSION_SIZE * 2 * OCCLUSION_SIZE * 2 + p_pos.y * OCCLUSION_SIZE * 2 + p_pos.x); + uint v = occlusion_facing[ofs / 4]; + return (v >> ((ofs % 4) * 8)) & 0xFF; +} + +#endif + +#ifdef MODE_STORE + +layout(rgba8ui, set = 0, binding = 1) uniform restrict readonly uimage3D src_positions; +layout(r16ui, set = 0, binding = 2) uniform restrict readonly uimage3D src_albedo; +layout(r8, set = 0, binding = 3) uniform restrict readonly image3D src_occlusion[8]; +layout(r32ui, set = 0, binding = 4) uniform restrict readonly uimage3D src_light; +layout(r32ui, set = 0, binding = 5) uniform restrict readonly uimage3D src_light_aniso; +layout(r32ui, set = 0, binding = 6) uniform restrict readonly uimage3D src_facing; + +layout(r8, set = 0, binding = 7) uniform restrict writeonly image3D dst_sdf; +layout(r16ui, set = 0, binding = 8) uniform restrict writeonly uimage3D dst_occlusion; + +layout(set = 0, binding = 10, std430) restrict buffer DispatchData { + uint x; + uint y; + uint z; + uint total_count; +} +dispatch_data; + +struct ProcessVoxel { + uint position; // xyz 7 bit packed, extra 11 bits for neighbors. + uint albedo; //rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neighbours + uint light; //rgbe8985 encoded total saved light, extra 2 bits for neighbours + uint light_aniso; //55555 light anisotropy, extra 2 bits for neighbours + //total neighbours: 26 +}; + +layout(set = 0, binding = 11, std430) restrict buffer writeonly ProcessVoxels { + ProcessVoxel data[]; +} +dst_process_voxels; + +shared ProcessVoxel store_positions[4 * 4 * 4]; +shared uint store_position_count; +shared uint store_from_index; +#endif + +#ifdef MODE_SCROLL + +layout(r16ui, set = 0, binding = 1) uniform restrict writeonly uimage3D dst_albedo; +layout(r32ui, set = 0, binding = 2) uniform restrict writeonly uimage3D dst_facing; +layout(r32ui, set = 0, binding = 3) uniform restrict writeonly uimage3D dst_light; +layout(r32ui, set = 0, binding = 4) uniform restrict writeonly uimage3D dst_light_aniso; + +layout(set = 0, binding = 5, std430) restrict buffer readonly DispatchData { + uint x; + uint y; + uint z; + uint total_count; +} +dispatch_data; + +struct ProcessVoxel { + uint position; // xyz 7 bit packed, extra 11 bits for neighbors. + uint albedo; //rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neighbours + uint light; //rgbe8985 encoded total saved light, extra 2 bits for neighbours + uint light_aniso; //55555 light anisotropy, extra 2 bits for neighbours + //total neighbours: 26 +}; + +layout(set = 0, binding = 6, std430) restrict buffer readonly ProcessVoxels { + ProcessVoxel data[]; +} +src_process_voxels; + +#endif + +#ifdef MODE_SCROLL_OCCLUSION + +layout(r8, set = 0, binding = 1) uniform restrict image3D dst_occlusion[8]; +layout(r16ui, set = 0, binding = 2) uniform restrict readonly uimage3D src_occlusion; + +#endif + +layout(push_constant, std430) uniform Params { + ivec3 scroll; + + int grid_size; + + ivec3 probe_offset; + int step_size; + + bool half_size; + uint occlusion_index; + int cascade; + uint pad; +} +params; + +void main() { +#ifdef MODE_SCROLL + + // Pixel being shaded + int index = int(gl_GlobalInvocationID.x); + if (index >= dispatch_data.total_count) { //too big + return; + } + + ivec3 read_pos = (ivec3(src_process_voxels.data[index].position) >> ivec3(0, 7, 14)) & ivec3(0x7F); + ivec3 write_pos = read_pos + params.scroll; + + if (any(lessThan(write_pos, ivec3(0))) || any(greaterThanEqual(write_pos, ivec3(params.grid_size)))) { + return; // Fits outside the 3D texture, don't do anything. + } + + uint albedo = ((src_process_voxels.data[index].albedo & 0x7FFF) << 1) | 1; //add solid bit + imageStore(dst_albedo, write_pos, uvec4(albedo)); + + uint facing = (src_process_voxels.data[index].albedo >> 15) & 0x3F; //6 anisotropic facing bits + imageStore(dst_facing, write_pos, uvec4(facing)); + + uint light = src_process_voxels.data[index].light & 0x3fffffff; //30 bits of RGBE8985 + imageStore(dst_light, write_pos, uvec4(light)); + + uint light_aniso = src_process_voxels.data[index].light_aniso & 0x3fffffff; //30 bits of 6 anisotropic 5 bits values + imageStore(dst_light_aniso, write_pos, uvec4(light_aniso)); + +#endif + +#ifdef MODE_SCROLL_OCCLUSION + + ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); + if (any(greaterThanEqual(pos, ivec3(params.grid_size) - abs(params.scroll)))) { //too large, do nothing + return; + } + + ivec3 read_pos = pos + max(ivec3(0), -params.scroll); + ivec3 write_pos = pos + max(ivec3(0), params.scroll); + + read_pos.z += params.cascade * params.grid_size; + uint occlusion = imageLoad(src_occlusion, read_pos).r; + read_pos.x += params.grid_size; + occlusion |= imageLoad(src_occlusion, read_pos).r << 16; + + const uint occlusion_shift[8] = uint[](12, 8, 4, 0, 28, 24, 20, 16); + + for (uint i = 0; i < 8; i++) { + float o = float((occlusion >> occlusion_shift[i]) & 0xF) / 15.0; + imageStore(dst_occlusion[i], write_pos, vec4(o)); + } + +#endif + +#ifdef MODE_INITIALIZE_JUMP_FLOOD + + ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); + + uint c = imageLoad(src_color, pos).r; + uvec4 v; + if (bool(c & 0x1)) { + //bit set means this is solid + v.xyz = uvec3(pos); + v.w = 255; //not zero means used + } else { + v.xyz = uvec3(0); + v.w = 0; // zero means unused + } + + imageStore(dst_positions, pos, v); +#endif + +#ifdef MODE_INITIALIZE_JUMP_FLOOD_HALF + + ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); + ivec3 base_pos = pos * 2; + + //since we store in half size, lets kind of randomize what we store, so + //the half size jump flood has a bit better chance to find something + uvec4 closest[8]; + int closest_count = 0; + + for (uint i = 0; i < 8; i++) { + ivec3 src_pos = base_pos + ((ivec3(i) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1)); + uint c = imageLoad(src_color, src_pos).r; + if (bool(c & 1)) { + uvec4 v = uvec4(uvec3(src_pos), 255); + closest[closest_count] = v; + closest_count++; + } + } + + if (closest_count == 0) { + imageStore(dst_positions, pos, uvec4(0)); + } else { + ivec3 indexv = (pos & ivec3(1, 1, 1)) * ivec3(1, 2, 4); + int index = (indexv.x | indexv.y | indexv.z) % closest_count; + imageStore(dst_positions, pos, closest[index]); + } + +#endif + +#ifdef MODE_JUMPFLOOD + + //regular jumpflood, efficient for large steps, inefficient for small steps + ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); + + vec3 posf = vec3(pos); + + if (params.half_size) { + posf = posf * 2.0 + 0.5; + } + + uvec4 p = imageLoad(src_positions, pos); + + if (!params.half_size && p == uvec4(uvec3(pos), 255)) { + imageStore(dst_positions, pos, p); + return; //points to itself and valid, nothing better can be done, just pass + } + + float p_dist; + + if (p.w != 0) { + p_dist = distance(posf, vec3(p.xyz)); + } else { + p_dist = 0.0; //should not matter + } + + const uint offset_count = 26; + const ivec3 offsets[offset_count] = ivec3[]( + ivec3(-1, -1, -1), + ivec3(-1, -1, 0), + ivec3(-1, -1, 1), + ivec3(-1, 0, -1), + ivec3(-1, 0, 0), + ivec3(-1, 0, 1), + ivec3(-1, 1, -1), + ivec3(-1, 1, 0), + ivec3(-1, 1, 1), + ivec3(0, -1, -1), + ivec3(0, -1, 0), + ivec3(0, -1, 1), + ivec3(0, 0, -1), + ivec3(0, 0, 1), + ivec3(0, 1, -1), + ivec3(0, 1, 0), + ivec3(0, 1, 1), + ivec3(1, -1, -1), + ivec3(1, -1, 0), + ivec3(1, -1, 1), + ivec3(1, 0, -1), + ivec3(1, 0, 0), + ivec3(1, 0, 1), + ivec3(1, 1, -1), + ivec3(1, 1, 0), + ivec3(1, 1, 1)); + + for (uint i = 0; i < offset_count; i++) { + ivec3 ofs = pos + offsets[i] * params.step_size; + if (any(lessThan(ofs, ivec3(0))) || any(greaterThanEqual(ofs, ivec3(params.grid_size)))) { + continue; + } + uvec4 q = imageLoad(src_positions, ofs); + + if (q.w == 0) { + continue; //was not initialized yet, ignore + } + + float q_dist = distance(posf, vec3(q.xyz)); + if (p.w == 0 || q_dist < p_dist) { + p = q; //just replace because current is unused + p_dist = q_dist; + } + } + + imageStore(dst_positions, pos, p); +#endif + +#ifdef MODE_JUMPFLOOD_OPTIMIZED + //optimized version using shared compute memory + + ivec3 group_offset = ivec3(gl_WorkGroupID.xyz) % params.step_size; + ivec3 group_pos = group_offset + (ivec3(gl_WorkGroupID.xyz) / params.step_size) * ivec3(GROUP_SIZE * params.step_size); + + //load data into local group memory + + if (all(lessThan(ivec3(gl_LocalInvocationID.xyz), ivec3((GROUP_SIZE + 2) / 2)))) { + //use this thread for loading, this method uses less threads for this but its simpler and less divergent + ivec3 base_pos = ivec3(gl_LocalInvocationID.xyz) * 2; + for (uint i = 0; i < 8; i++) { + ivec3 load_pos = base_pos + ((ivec3(i) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1)); + ivec3 load_global_pos = group_pos + (load_pos - ivec3(1)) * params.step_size; + uvec4 q; + if (all(greaterThanEqual(load_global_pos, ivec3(0))) && all(lessThan(load_global_pos, ivec3(params.grid_size)))) { + q = imageLoad(src_positions, load_global_pos); + } else { + q = uvec4(0); //unused + } + + group_store(load_pos, q); + } + } + + ivec3 global_pos = group_pos + ivec3(gl_LocalInvocationID.xyz) * params.step_size; + + if (any(lessThan(global_pos, ivec3(0))) || any(greaterThanEqual(global_pos, ivec3(params.grid_size)))) { + return; //do nothing else, end here because outside range + } + + //sync + groupMemoryBarrier(); + barrier(); + + ivec3 local_pos = ivec3(gl_LocalInvocationID.xyz) + ivec3(1); + + const uint offset_count = 27; + const ivec3 offsets[offset_count] = ivec3[]( + ivec3(-1, -1, -1), + ivec3(-1, -1, 0), + ivec3(-1, -1, 1), + ivec3(-1, 0, -1), + ivec3(-1, 0, 0), + ivec3(-1, 0, 1), + ivec3(-1, 1, -1), + ivec3(-1, 1, 0), + ivec3(-1, 1, 1), + ivec3(0, -1, -1), + ivec3(0, -1, 0), + ivec3(0, -1, 1), + ivec3(0, 0, -1), + ivec3(0, 0, 0), + ivec3(0, 0, 1), + ivec3(0, 1, -1), + ivec3(0, 1, 0), + ivec3(0, 1, 1), + ivec3(1, -1, -1), + ivec3(1, -1, 0), + ivec3(1, -1, 1), + ivec3(1, 0, -1), + ivec3(1, 0, 0), + ivec3(1, 0, 1), + ivec3(1, 1, -1), + ivec3(1, 1, 0), + ivec3(1, 1, 1)); + + //only makes sense if point is inside screen + uvec4 closest = uvec4(0); + float closest_dist = 0.0; + + vec3 posf = vec3(global_pos); + + if (params.half_size) { + posf = posf * 2.0 + 0.5; + } + + for (uint i = 0; i < offset_count; i++) { + uvec4 point = group_load(local_pos + offsets[i]); + + if (point.w == 0) { + continue; //was not initialized yet, ignore + } + + float dist = distance(posf, vec3(point.xyz)); + if (closest.w == 0 || dist < closest_dist) { + closest = point; + closest_dist = dist; + } + } + + imageStore(dst_positions, global_pos, closest); + +#endif + +#ifdef MODE_UPSCALE_JUMP_FLOOD + + ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); + + uint c = imageLoad(src_color, pos).r; + uvec4 v; + if (bool(c & 1)) { + //bit set means this is solid + v.xyz = uvec3(pos); + v.w = 255; //not zero means used + } else { + v = imageLoad(src_positions_half, pos >> 1); + float d = length(vec3(ivec3(v.xyz) - pos)); + + ivec3 vbase = ivec3(v.xyz - (v.xyz & uvec3(1))); + + //search around if there is a better candidate from the same block + for (int i = 0; i < 8; i++) { + ivec3 bits = ((ivec3(i) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1)); + ivec3 p = vbase + bits; + + float d2 = length(vec3(p - pos)); + if (d2 < d) { //check valid distance before test so we avoid a read + uint c2 = imageLoad(src_color, p).r; + if (bool(c2 & 1)) { + v.xyz = uvec3(p); + d = d2; + } + } + } + + //could validate better position.. + } + + imageStore(dst_positions, pos, v); + +#endif + +#ifdef MODE_OCCLUSION + + uint invocation_idx = uint(gl_LocalInvocationID.x); + ivec3 region = ivec3(gl_WorkGroupID); + + ivec3 region_offset = -ivec3(OCCLUSION_SIZE); + region_offset += region * OCCLUSION_SIZE * 2; + region_offset += params.probe_offset * OCCLUSION_SIZE; + + if (params.scroll != ivec3(0)) { + //validate scroll region + ivec3 region_offset_to = region_offset + ivec3(OCCLUSION_SIZE * 2); + uvec3 scroll_mask = uvec3(notEqual(params.scroll, ivec3(0))); //save which axes acre scrolling + ivec3 scroll_from = mix(ivec3(0), ivec3(params.grid_size) + params.scroll, lessThan(params.scroll, ivec3(0))); + ivec3 scroll_to = mix(ivec3(params.grid_size), params.scroll, greaterThan(params.scroll, ivec3(0))); + + if ((uvec3(lessThanEqual(region_offset_to, scroll_from)) | uvec3(greaterThanEqual(region_offset, scroll_to))) * scroll_mask == scroll_mask) { //all axes that scroll are out, exit + return; //region outside scroll bounds, quit + } + } + +#define OCC_HALF_SIZE (OCCLUSION_SIZE / 2) + + ivec3 local_ofs = ivec3(uvec3(invocation_idx % OCC_HALF_SIZE, (invocation_idx % (OCC_HALF_SIZE * OCC_HALF_SIZE)) / OCC_HALF_SIZE, invocation_idx / (OCC_HALF_SIZE * OCC_HALF_SIZE))) * 4; + + /* for(int i=0;i<64;i++) { + ivec3 offset = region_offset + local_ofs + ((ivec3(i) >> ivec3(0,2,4)) & ivec3(3,3,3)); + uint facig = + if (all(greaterThanEqual(offset,ivec3(0))) && all(lessThan(offset,ivec3(params.grid_size)))) {*/ + + for (int i = 0; i < 16; i++) { //skip x, so it can be packed + + ivec3 offset = local_ofs + ((ivec3(i * 4) >> ivec3(0, 2, 4)) & ivec3(3, 3, 3)); + + uint facing_pack = 0; + for (int j = 0; j < 4; j++) { + ivec3 foffset = region_offset + offset + ivec3(j, 0, 0); + if (all(greaterThanEqual(foffset, ivec3(0))) && all(lessThan(foffset, ivec3(params.grid_size)))) { + uint f = imageLoad(src_facing, foffset).r; + facing_pack |= f << (j * 8); + } + } + + occlusion_facing[(offset.z * (OCCLUSION_SIZE * 2 * OCCLUSION_SIZE * 2) + offset.y * (OCCLUSION_SIZE * 2) + offset.x) / 4] = facing_pack; + } + + //sync occlusion saved + groupMemoryBarrier(); + barrier(); + + //process occlusion + +#define OCC_STEPS (OCCLUSION_SIZE * 3 - 2) +#define OCC_HALF_STEPS (OCC_STEPS / 2) + + for (int step = 0; step < OCC_STEPS; step++) { + bool shrink = step >= OCC_HALF_STEPS; + int occ_step = shrink ? OCC_HALF_STEPS - (step - OCC_HALF_STEPS) - 1 : step; + + if (invocation_idx < group_size_offset[occ_step].x) { + uint pv = group_pos[group_size_offset[occ_step].y + invocation_idx]; + ivec3 proc_abs = (ivec3(int(pv)) >> ivec3(0, 8, 16)) & ivec3(0xFF); + + if (shrink) { + proc_abs = ivec3(OCCLUSION_SIZE) - proc_abs - ivec3(1); + } + + for (int i = 0; i < 8; i++) { + ivec3 bits = ((ivec3(i) >> ivec3(0, 1, 2)) & ivec3(1, 1, 1)); + ivec3 proc_sign = bits * 2 - 1; + ivec3 local_offset = ivec3(OCCLUSION_SIZE) + proc_abs * proc_sign - (ivec3(1) - bits); + ivec3 offset = local_offset + region_offset; + if (all(greaterThanEqual(offset, ivec3(0))) && all(lessThan(offset, ivec3(params.grid_size)))) { + float occ; + + uint facing = get_facing(local_offset); + + if (facing != 0) { //solid + occ = 0.0; + } else if (step == 0) { +#if 0 + occ = 0.0; + if (get_facing(local_offset - ivec3(proc_sign.x,0,0))==0) { + occ+=1.0; + } + if (get_facing(local_offset - ivec3(0,proc_sign.y,0))==0) { + occ+=1.0; + } + if (get_facing(local_offset - ivec3(0,0,proc_sign.z))==0) { + occ+=1.0; + } + /* + if (get_facing(local_offset - proc_sign)==0) { + occ+=1.0; + }*/ + + occ/=3.0; +#endif + occ = 1.0; + + } else { + ivec3 read_dir = -proc_sign; + + ivec3 major_axis; + if (proc_abs.x < proc_abs.y) { + if (proc_abs.z < proc_abs.y) { + major_axis = ivec3(0, 1, 0); + } else { + major_axis = ivec3(0, 0, 1); + } + } else { + if (proc_abs.z < proc_abs.x) { + major_axis = ivec3(1, 0, 0); + } else { + major_axis = ivec3(0, 0, 1); + } + } + + float avg = 0.0; + occ = 0.0; + + ivec3 read_x = offset + ivec3(read_dir.x, 0, 0) + (proc_abs.x == 0 ? major_axis * read_dir : ivec3(0)); + ivec3 read_y = offset + ivec3(0, read_dir.y, 0) + (proc_abs.y == 0 ? major_axis * read_dir : ivec3(0)); + ivec3 read_z = offset + ivec3(0, 0, read_dir.z) + (proc_abs.z == 0 ? major_axis * read_dir : ivec3(0)); + + uint facing_x = get_facing(read_x - region_offset); + if (facing_x == 0) { + if (all(greaterThanEqual(read_x, ivec3(0))) && all(lessThan(read_x, ivec3(params.grid_size)))) { + occ += imageLoad(dst_occlusion[params.occlusion_index], read_x).r; + avg += 1.0; + } + } else { + if (proc_abs.x != 0) { //do not occlude from voxels in the opposite octant + avg += 1.0; + } + } + + uint facing_y = get_facing(read_y - region_offset); + if (facing_y == 0) { + if (all(greaterThanEqual(read_y, ivec3(0))) && all(lessThan(read_y, ivec3(params.grid_size)))) { + occ += imageLoad(dst_occlusion[params.occlusion_index], read_y).r; + avg += 1.0; + } + } else { + if (proc_abs.y != 0) { + avg += 1.0; + } + } + + uint facing_z = get_facing(read_z - region_offset); + if (facing_z == 0) { + if (all(greaterThanEqual(read_z, ivec3(0))) && all(lessThan(read_z, ivec3(params.grid_size)))) { + occ += imageLoad(dst_occlusion[params.occlusion_index], read_z).r; + avg += 1.0; + } + } else { + if (proc_abs.z != 0) { + avg += 1.0; + } + } + + if (avg > 0.0) { + occ /= avg; + } + } + + imageStore(dst_occlusion[params.occlusion_index], offset, vec4(occ)); + } + } + } + + groupMemoryBarrier(); + barrier(); + } +#if 1 + //bias solid voxels away + + for (int i = 0; i < 64; i++) { + ivec3 local_offset = local_ofs + ((ivec3(i) >> ivec3(0, 2, 4)) & ivec3(3, 3, 3)); + ivec3 offset = region_offset + local_offset; + + if (all(greaterThanEqual(offset, ivec3(0))) && all(lessThan(offset, ivec3(params.grid_size)))) { + uint facing = get_facing(local_offset); + + if (facing != 0) { + //only work on solids + + ivec3 proc_pos = local_offset - ivec3(OCCLUSION_SIZE); + proc_pos += mix(ivec3(0), ivec3(1), greaterThanEqual(proc_pos, ivec3(0))); + + float avg = 0.0; + float occ = 0.0; + + ivec3 read_dir = -sign(proc_pos); + ivec3 read_dir_x = ivec3(read_dir.x, 0, 0); + ivec3 read_dir_y = ivec3(0, read_dir.y, 0); + ivec3 read_dir_z = ivec3(0, 0, read_dir.z); + //solid +#if 0 + + uvec3 facing_pos_base = (uvec3(facing) >> uvec3(0,1,2)) & uvec3(1,1,1); + uvec3 facing_neg_base = (uvec3(facing) >> uvec3(3,4,5)) & uvec3(1,1,1); + uvec3 facing_pos= facing_pos_base &((~facing_neg_base)&uvec3(1,1,1)); + uvec3 facing_neg= facing_neg_base &((~facing_pos_base)&uvec3(1,1,1)); +#else + uvec3 facing_pos = (uvec3(facing) >> uvec3(0, 1, 2)) & uvec3(1, 1, 1); + uvec3 facing_neg = (uvec3(facing) >> uvec3(3, 4, 5)) & uvec3(1, 1, 1); +#endif + bvec3 read_valid = bvec3(mix(facing_neg, facing_pos, greaterThan(read_dir, ivec3(0)))); + + //sides + if (read_valid.x) { + ivec3 read_offset = local_offset + read_dir_x; + uint f = get_facing(read_offset); + if (f == 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occ += imageLoad(dst_occlusion[params.occlusion_index], read_offset).r; + avg += 1.0; + } + } + } + + if (read_valid.y) { + ivec3 read_offset = local_offset + read_dir_y; + uint f = get_facing(read_offset); + if (f == 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occ += imageLoad(dst_occlusion[params.occlusion_index], read_offset).r; + avg += 1.0; + } + } + } + + if (read_valid.z) { + ivec3 read_offset = local_offset + read_dir_z; + uint f = get_facing(read_offset); + if (f == 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occ += imageLoad(dst_occlusion[params.occlusion_index], read_offset).r; + avg += 1.0; + } + } + } + + //adjacents + + if (all(read_valid.yz)) { + ivec3 read_offset = local_offset + read_dir_y + read_dir_z; + uint f = get_facing(read_offset); + if (f == 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occ += imageLoad(dst_occlusion[params.occlusion_index], read_offset).r; + avg += 1.0; + } + } + } + + if (all(read_valid.xz)) { + ivec3 read_offset = local_offset + read_dir_x + read_dir_z; + uint f = get_facing(read_offset); + if (f == 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occ += imageLoad(dst_occlusion[params.occlusion_index], read_offset).r; + avg += 1.0; + } + } + } + + if (all(read_valid.xy)) { + ivec3 read_offset = local_offset + read_dir_x + read_dir_y; + uint f = get_facing(read_offset); + if (f == 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occ += imageLoad(dst_occlusion[params.occlusion_index], read_offset).r; + avg += 1.0; + } + } + } + + //diagonal + + if (all(read_valid)) { + ivec3 read_offset = local_offset + read_dir; + uint f = get_facing(read_offset); + if (f == 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occ += imageLoad(dst_occlusion[params.occlusion_index], read_offset).r; + avg += 1.0; + } + } + } + + if (avg > 0.0) { + occ /= avg; + } + + imageStore(dst_occlusion[params.occlusion_index], offset, vec4(occ)); + } + } + } + +#endif + +#if 1 + groupMemoryBarrier(); + barrier(); + + for (int i = 0; i < 64; i++) { + ivec3 local_offset = local_ofs + ((ivec3(i) >> ivec3(0, 2, 4)) & ivec3(3, 3, 3)); + ivec3 offset = region_offset + local_offset; + + if (all(greaterThanEqual(offset, ivec3(0))) && all(lessThan(offset, ivec3(params.grid_size)))) { + uint facing = get_facing(local_offset); + + if (facing == 0) { + ivec3 proc_pos = local_offset - ivec3(OCCLUSION_SIZE); + proc_pos += mix(ivec3(0), ivec3(1), greaterThanEqual(proc_pos, ivec3(0))); + + ivec3 proc_abs = abs(proc_pos); + + ivec3 read_dir = sign(proc_pos); //opposite direction + ivec3 read_dir_x = ivec3(read_dir.x, 0, 0); + ivec3 read_dir_y = ivec3(0, read_dir.y, 0); + ivec3 read_dir_z = ivec3(0, 0, read_dir.z); + //solid + uvec3 read_mask = mix(uvec3(1, 2, 4), uvec3(8, 16, 32), greaterThan(read_dir, ivec3(0))); //match positive with negative normals + uvec3 block_mask = mix(uvec3(1, 2, 4), uvec3(8, 16, 32), lessThan(read_dir, ivec3(0))); //match positive with negative normals + + block_mask = uvec3(0); + + float visible = 0.0; + float occlude_total = 0.0; + + if (proc_abs.x < OCCLUSION_SIZE) { + ivec3 read_offset = local_offset + read_dir_x; + uint x_mask = get_facing(read_offset); + if (x_mask != 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occlude_total += 1.0; + if (bool(x_mask & read_mask.x) && !bool(x_mask & block_mask.x)) { + visible += 1.0; + } + } + } + } + + if (proc_abs.y < OCCLUSION_SIZE) { + ivec3 read_offset = local_offset + read_dir_y; + uint y_mask = get_facing(read_offset); + if (y_mask != 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occlude_total += 1.0; + if (bool(y_mask & read_mask.y) && !bool(y_mask & block_mask.y)) { + visible += 1.0; + } + } + } + } + + if (proc_abs.z < OCCLUSION_SIZE) { + ivec3 read_offset = local_offset + read_dir_z; + uint z_mask = get_facing(read_offset); + if (z_mask != 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occlude_total += 1.0; + if (bool(z_mask & read_mask.z) && !bool(z_mask & block_mask.z)) { + visible += 1.0; + } + } + } + } + + //if near the cartesian plane, test in opposite direction too + + read_mask = mix(uvec3(1, 2, 4), uvec3(8, 16, 32), lessThan(read_dir, ivec3(0))); //match negative with positive normals + block_mask = mix(uvec3(1, 2, 4), uvec3(8, 16, 32), greaterThan(read_dir, ivec3(0))); //match negative with positive normals + block_mask = uvec3(0); + + if (proc_abs.x == 1) { + ivec3 read_offset = local_offset - read_dir_x; + uint x_mask = get_facing(read_offset); + if (x_mask != 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occlude_total += 1.0; + if (bool(x_mask & read_mask.x) && !bool(x_mask & block_mask.x)) { + visible += 1.0; + } + } + } + } + + if (proc_abs.y == 1) { + ivec3 read_offset = local_offset - read_dir_y; + uint y_mask = get_facing(read_offset); + if (y_mask != 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occlude_total += 1.0; + if (bool(y_mask & read_mask.y) && !bool(y_mask & block_mask.y)) { + visible += 1.0; + } + } + } + } + + if (proc_abs.z == 1) { + ivec3 read_offset = local_offset - read_dir_z; + uint z_mask = get_facing(read_offset); + if (z_mask != 0) { + read_offset += region_offset; + if (all(greaterThanEqual(read_offset, ivec3(0))) && all(lessThan(read_offset, ivec3(params.grid_size)))) { + occlude_total += 1.0; + if (bool(z_mask & read_mask.z) && !bool(z_mask & block_mask.z)) { + visible += 1.0; + } + } + } + } + + if (occlude_total > 0.0) { + float occ = imageLoad(dst_occlusion[params.occlusion_index], offset).r; + occ *= visible / occlude_total; + imageStore(dst_occlusion[params.occlusion_index], offset, vec4(occ)); + } + } + } + } + +#endif + + /* + for(int i=0;i<8;i++) { + ivec3 local_offset = local_pos + ((ivec3(i) >> ivec3(2,1,0)) & ivec3(1,1,1)) * OCCLUSION_SIZE; + ivec3 offset = local_offset - ivec3(OCCLUSION_SIZE); //looking around probe, so starts negative + offset += region * OCCLUSION_SIZE * 2; //offset by region + offset += params.probe_offset * OCCLUSION_SIZE; // offset by probe offset + if (all(greaterThanEqual(offset,ivec3(0))) && all(lessThan(offset,ivec3(params.grid_size)))) { + imageStore(dst_occlusion[params.occlusion_index],offset,vec4( occlusion_data[ to_linear(local_offset) ] )); + //imageStore(dst_occlusion[params.occlusion_index],offset,vec4( occlusion_solid[ to_linear(local_offset) ] )); + } + } +*/ + +#endif + +#ifdef MODE_STORE + + ivec3 local = ivec3(gl_LocalInvocationID.xyz); + ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); + // store SDF + uvec4 p = imageLoad(src_positions, pos); + + bool solid = false; + float d; + if (ivec3(p.xyz) == pos) { + //solid block + d = 0; + solid = true; + } else { + //distance block + d = 1.0 + length(vec3(p.xyz) - vec3(pos)); + } + + d /= 255.0; + + imageStore(dst_sdf, pos, vec4(d)); + + // STORE OCCLUSION + + uint occlusion = 0; + const uint occlusion_shift[8] = uint[](12, 8, 4, 0, 28, 24, 20, 16); + for (int i = 0; i < 8; i++) { + float occ = imageLoad(src_occlusion[i], pos).r; + occlusion |= uint(clamp(occ * 15.0, 0.0, 15.0)) << occlusion_shift[i]; + } + { + ivec3 occ_pos = pos; + occ_pos.z += params.cascade * params.grid_size; + imageStore(dst_occlusion, occ_pos, uvec4(occlusion & 0xFFFF)); + occ_pos.x += params.grid_size; + imageStore(dst_occlusion, occ_pos, uvec4(occlusion >> 16)); + } + + // STORE POSITIONS + + if (local == ivec3(0)) { + store_position_count = 0; //base one stores as zero, the others wait + } + + groupMemoryBarrier(); + barrier(); + + if (solid) { + uint index = atomicAdd(store_position_count, 1); + // At least do the conversion work in parallel + store_positions[index].position = uint(pos.x | (pos.y << 7) | (pos.z << 14)); + + //see around which voxels point to this one, add them to the list + uint bit_index = 0; + uint neighbour_bits = 0; + for (int i = -1; i <= 1; i++) { + for (int j = -1; j <= 1; j++) { + for (int k = -1; k <= 1; k++) { + if (i == 0 && j == 0 && k == 0) { + continue; + } + ivec3 npos = pos + ivec3(i, j, k); + if (all(greaterThanEqual(npos, ivec3(0))) && all(lessThan(npos, ivec3(params.grid_size)))) { + p = imageLoad(src_positions, npos); + if (ivec3(p.xyz) == pos) { + neighbour_bits |= (1 << bit_index); + } + } + bit_index++; + } + } + } + + uint rgb = imageLoad(src_albedo, pos).r; + uint facing = imageLoad(src_facing, pos).r; + + store_positions[index].albedo = rgb >> 1; //store as it comes (555) to avoid precision loss (and move away the alpha bit) + store_positions[index].albedo |= (facing & 0x3F) << 15; // store facing in bits 15-21 + + store_positions[index].albedo |= neighbour_bits << 21; //store lower 11 bits of neighbours with remaining albedo + store_positions[index].position |= (neighbour_bits >> 11) << 21; //store 11 bits more of neighbours with position + + store_positions[index].light = imageLoad(src_light, pos).r; + store_positions[index].light_aniso = imageLoad(src_light_aniso, pos).r; + //add neighbours + store_positions[index].light |= (neighbour_bits >> 22) << 30; //store 2 bits more of neighbours with light + store_positions[index].light_aniso |= (neighbour_bits >> 24) << 30; //store 2 bits more of neighbours with aniso + } + + groupMemoryBarrier(); + barrier(); + + // global increment only once per group, to reduce pressure + + if (local == ivec3(0) && store_position_count > 0) { + store_from_index = atomicAdd(dispatch_data.total_count, store_position_count); + uint group_count = (store_from_index + store_position_count - 1) / 64 + 1; + atomicMax(dispatch_data.x, group_count); + } + + groupMemoryBarrier(); + barrier(); + + uint read_index = uint(local.z * 4 * 4 + local.y * 4 + local.x); + uint write_index = store_from_index + read_index; + + if (read_index < store_position_count) { + dst_process_voxels.data[write_index] = store_positions[read_index]; + } + + if (pos == ivec3(0)) { + //this thread clears y and z + dispatch_data.y = 1; + dispatch_data.z = 1; + } +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/environment/voxel_gi.glsl b/servers/rendering/renderer_rd/shaders/environment/voxel_gi.glsl new file mode 100644 index 0000000000..577c6d0cd0 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/environment/voxel_gi.glsl @@ -0,0 +1,616 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#ifdef MODE_DYNAMIC +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; +#else +layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in; +#endif + +#ifndef MODE_DYNAMIC + +#define NO_CHILDREN 0xFFFFFFFF + +struct CellChildren { + uint children[8]; +}; + +layout(set = 0, binding = 1, std430) buffer CellChildrenBuffer { + CellChildren data[]; +} +cell_children; + +struct CellData { + uint position; // xyz 10 bits + uint albedo; //rgb albedo + uint emission; //rgb normalized with e as multiplier + uint normal; //RGB normal encoded +}; + +layout(set = 0, binding = 2, std430) buffer CellDataBuffer { + CellData data[]; +} +cell_data; + +#endif // MODE DYNAMIC + +#define LIGHT_TYPE_DIRECTIONAL 0 +#define LIGHT_TYPE_OMNI 1 +#define LIGHT_TYPE_SPOT 2 + +#if defined(MODE_COMPUTE_LIGHT) || defined(MODE_DYNAMIC_LIGHTING) + +struct Light { + uint type; + float energy; + float radius; + float attenuation; + + vec3 color; + float cos_spot_angle; + + vec3 position; + float inv_spot_attenuation; + + vec3 direction; + bool has_shadow; +}; + +layout(set = 0, binding = 3, std140) uniform Lights { + Light data[MAX_LIGHTS]; +} +lights; + +#endif // MODE COMPUTE LIGHT + +#ifdef MODE_SECOND_BOUNCE + +layout(set = 0, binding = 5) uniform texture3D color_texture; + +#endif // MODE_SECOND_BOUNCE + +#ifndef MODE_DYNAMIC + +layout(push_constant, std430) uniform Params { + ivec3 limits; + uint stack_size; + + float emission_scale; + float propagation; + float dynamic_range; + + uint light_count; + uint cell_offset; + uint cell_count; + float aniso_strength; + uint pad; +} +params; + +layout(set = 0, binding = 4, std430) buffer Outputs { + vec4 data[]; +} +outputs; + +#endif // MODE DYNAMIC + +layout(set = 0, binding = 9) uniform texture3D texture_sdf; +layout(set = 0, binding = 10) uniform sampler texture_sampler; + +#ifdef MODE_WRITE_TEXTURE + +layout(rgba8, set = 0, binding = 5) uniform restrict writeonly image3D color_tex; + +#endif + +#ifdef MODE_DYNAMIC + +layout(push_constant, std430) uniform Params { + ivec3 limits; + uint light_count; //when not lighting + ivec3 x_dir; + float z_base; + ivec3 y_dir; + float z_sign; + ivec3 z_dir; + float pos_multiplier; + ivec2 rect_pos; + ivec2 rect_size; + ivec2 prev_rect_ofs; + ivec2 prev_rect_size; + bool flip_x; + bool flip_y; + float dynamic_range; + bool on_mipmap; + float propagation; + float pad[3]; +} +params; + +#ifdef MODE_DYNAMIC_LIGHTING + +layout(rgba8, set = 0, binding = 5) uniform restrict readonly image2D source_albedo; +layout(rgba8, set = 0, binding = 6) uniform restrict readonly image2D source_normal; +layout(rgba8, set = 0, binding = 7) uniform restrict readonly image2D source_orm; +//layout (set=0,binding=8) uniform texture2D source_depth; +layout(rgba16f, set = 0, binding = 11) uniform restrict image2D emission; +layout(r32f, set = 0, binding = 12) uniform restrict image2D depth; + +#endif + +#ifdef MODE_DYNAMIC_SHRINK + +layout(rgba16f, set = 0, binding = 5) uniform restrict readonly image2D source_light; +layout(r32f, set = 0, binding = 6) uniform restrict readonly image2D source_depth; + +#ifdef MODE_DYNAMIC_SHRINK_WRITE + +layout(rgba16f, set = 0, binding = 7) uniform restrict writeonly image2D light; +layout(r32f, set = 0, binding = 8) uniform restrict writeonly image2D depth; + +#endif // MODE_DYNAMIC_SHRINK_WRITE + +#ifdef MODE_DYNAMIC_SHRINK_PLOT + +layout(rgba8, set = 0, binding = 11) uniform restrict image3D color_texture; + +#endif //MODE_DYNAMIC_SHRINK_PLOT + +#endif // MODE_DYNAMIC_SHRINK + +//layout (rgba8,set=0,binding=5) uniform restrict writeonly image3D color_tex; + +#endif // MODE DYNAMIC + +#if defined(MODE_COMPUTE_LIGHT) || defined(MODE_DYNAMIC_LIGHTING) + +float raymarch(float distance, float distance_adv, vec3 from, vec3 direction) { + vec3 cell_size = 1.0 / vec3(params.limits); + float occlusion = 1.0; + while (distance > 0.5) { //use this to avoid precision errors + float advance = texture(sampler3D(texture_sdf, texture_sampler), from * cell_size).r * 255.0 - 1.0; + if (advance < 0.0) { + occlusion = 0.0; + break; + } + + occlusion = min(advance, occlusion); + + advance = max(distance_adv, advance - mod(advance, distance_adv)); //should always advance in multiples of distance_adv + + from += direction * advance; + distance -= advance; + } + + return occlusion; //max(0.0,distance); +} + +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); +} + +bool compute_light_vector(uint light, vec3 pos, out float attenuation, out vec3 light_pos) { + if (lights.data[light].type == LIGHT_TYPE_DIRECTIONAL) { + light_pos = pos - lights.data[light].direction * length(vec3(params.limits)); + attenuation = 1.0; + + } else { + light_pos = lights.data[light].position; + float distance = length(pos - light_pos); + if (distance >= lights.data[light].radius) { + return false; + } + + attenuation = get_omni_attenuation(distance, 1.0 / lights.data[light].radius, lights.data[light].attenuation); + + if (lights.data[light].type == LIGHT_TYPE_SPOT) { + vec3 rel = normalize(pos - light_pos); + float cos_spot_angle = lights.data[light].cos_spot_angle; + float cos_angle = dot(rel, lights.data[light].direction); + if (cos_angle < cos_spot_angle) { + return false; + } + + 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, lights.data[light].inv_spot_attenuation); + } + } + + return true; +} + +float get_normal_advance(vec3 p_normal) { + vec3 normal = p_normal; + vec3 unorm = abs(normal); + + if ((unorm.x >= unorm.y) && (unorm.x >= unorm.z)) { + // x code + unorm = normal.x > 0.0 ? vec3(1.0, 0.0, 0.0) : vec3(-1.0, 0.0, 0.0); + } else if ((unorm.y > unorm.x) && (unorm.y >= unorm.z)) { + // y code + unorm = normal.y > 0.0 ? vec3(0.0, 1.0, 0.0) : vec3(0.0, -1.0, 0.0); + } else if ((unorm.z > unorm.x) && (unorm.z > unorm.y)) { + // z code + unorm = normal.z > 0.0 ? vec3(0.0, 0.0, 1.0) : vec3(0.0, 0.0, -1.0); + } else { + // oh-no we messed up code + // has to be + unorm = vec3(1.0, 0.0, 0.0); + } + + return 1.0 / dot(normal, unorm); +} + +void clip_segment(vec4 plane, vec3 begin, inout vec3 end) { + vec3 segment = begin - end; + float den = dot(plane.xyz, segment); + + //printf("den is %i\n",den); + if (den < 0.0001) { + return; + } + + float dist = (dot(plane.xyz, begin) - plane.w) / den; + + if (dist < 0.0001 || dist > 1.0001) { + return; + } + + end = begin + segment * -dist; +} + +bool compute_light_at_pos(uint index, vec3 pos, vec3 normal, inout vec3 light, inout vec3 light_dir) { + float attenuation; + vec3 light_pos; + + if (!compute_light_vector(index, pos, attenuation, light_pos)) { + return false; + } + + light_dir = normalize(pos - light_pos); + + if (attenuation < 0.01 || (length(normal) > 0.2 && dot(normal, light_dir) >= 0)) { + return false; //not facing the light, or attenuation is near zero + } + + if (lights.data[index].has_shadow) { + float distance_adv = get_normal_advance(light_dir); + + vec3 to = pos; + if (length(normal) > 0.2) { + to += normal * distance_adv * 0.51; + } else { + to -= sign(light_dir) * 0.45; //go near the edge towards the light direction to avoid self occlusion + } + + //clip + clip_segment(mix(vec4(-1.0, 0.0, 0.0, 0.0), vec4(1.0, 0.0, 0.0, float(params.limits.x - 1)), bvec4(light_dir.x < 0.0)), to, light_pos); + clip_segment(mix(vec4(0.0, -1.0, 0.0, 0.0), vec4(0.0, 1.0, 0.0, float(params.limits.y - 1)), bvec4(light_dir.y < 0.0)), to, light_pos); + clip_segment(mix(vec4(0.0, 0.0, -1.0, 0.0), vec4(0.0, 0.0, 1.0, float(params.limits.z - 1)), bvec4(light_dir.z < 0.0)), to, light_pos); + + float distance = length(to - light_pos); + if (distance < 0.1) { + return false; // hit + } + + distance += distance_adv - mod(distance, distance_adv); //make it reach the center of the box always + light_pos = to - light_dir * distance; + + //from -= sign(light_dir)*0.45; //go near the edge towards the light direction to avoid self occlusion + + /*float dist = raymarch(distance,distance_adv,light_pos,light_dir); + + if (dist > distance_adv) { + return false; + } + + attenuation *= 1.0 - smoothstep(0.1*distance_adv,distance_adv,dist); + */ + + float occlusion = raymarch(distance, distance_adv, light_pos, light_dir); + + if (occlusion == 0.0) { + return false; + } + + attenuation *= occlusion; //1.0 - smoothstep(0.1*distance_adv,distance_adv,dist); + } + + light = lights.data[index].color * attenuation * lights.data[index].energy; + return true; +} + +#endif // MODE COMPUTE LIGHT + +void main() { +#ifndef MODE_DYNAMIC + + uint cell_index = gl_GlobalInvocationID.x; + if (cell_index >= params.cell_count) { + return; + } + cell_index += params.cell_offset; + + uvec3 posu = uvec3(cell_data.data[cell_index].position & 0x7FF, (cell_data.data[cell_index].position >> 11) & 0x3FF, cell_data.data[cell_index].position >> 21); + vec4 albedo = unpackUnorm4x8(cell_data.data[cell_index].albedo); + +#endif + + /////////////////COMPUTE LIGHT/////////////////////////////// + +#ifdef MODE_COMPUTE_LIGHT + + vec3 pos = vec3(posu) + vec3(0.5); + + vec3 emission = vec3(uvec3(cell_data.data[cell_index].emission & 0x1ff, (cell_data.data[cell_index].emission >> 9) & 0x1ff, (cell_data.data[cell_index].emission >> 18) & 0x1ff)) * pow(2.0, float(cell_data.data[cell_index].emission >> 27) - 15.0 - 9.0); + vec3 normal = unpackSnorm4x8(cell_data.data[cell_index].normal).xyz; + + vec3 accum = vec3(0.0); + + for (uint i = 0; i < params.light_count; i++) { + vec3 light; + vec3 light_dir; + if (!compute_light_at_pos(i, pos, normal.xyz, light, light_dir)) { + continue; + } + + light *= albedo.rgb; + + if (length(normal) > 0.2) { + accum += max(0.0, dot(normal, -light_dir)) * light; + } else { + //all directions + accum += light; + } + } + + outputs.data[cell_index] = vec4(accum + emission, 0.0); + +#endif //MODE_COMPUTE_LIGHT + + /////////////////SECOND BOUNCE/////////////////////////////// + +#ifdef MODE_SECOND_BOUNCE + vec3 pos = vec3(posu) + vec3(0.5); + ivec3 ipos = ivec3(posu); + vec4 normal = unpackSnorm4x8(cell_data.data[cell_index].normal); + + vec3 accum = outputs.data[cell_index].rgb; + + if (length(normal.xyz) > 0.2) { + 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.xyz)); + vec3 bitangent = normalize(cross(tangent, normal.xyz)); + mat3 normal_mat = mat3(tangent, bitangent, normal.xyz); + +#define MAX_CONE_DIRS 6 + + vec3 cone_dirs[MAX_CONE_DIRS] = vec3[]( + vec3(0.0, 0.0, 1.0), + vec3(0.866025, 0.0, 0.5), + vec3(0.267617, 0.823639, 0.5), + vec3(-0.700629, 0.509037, 0.5), + vec3(-0.700629, -0.509037, 0.5), + vec3(0.267617, -0.823639, 0.5)); + + float cone_weights[MAX_CONE_DIRS] = float[](0.25, 0.15, 0.15, 0.15, 0.15, 0.15); + float tan_half_angle = 0.577; + + for (int i = 0; i < MAX_CONE_DIRS; i++) { + vec3 direction = normal_mat * cone_dirs[i]; + vec4 color = vec4(0.0); + { + float dist = 1.5; + float max_distance = length(vec3(params.limits)); + vec3 cell_size = 1.0 / vec3(params.limits); + + while (dist < max_distance && color.a < 0.95) { + float diameter = max(1.0, 2.0 * tan_half_angle * dist); + vec3 uvw_pos = (pos + dist * direction) * cell_size; + float half_diameter = diameter * 0.5; + //check if outside, then break + //if ( any(greaterThan(abs(uvw_pos - 0.5),vec3(0.5f + half_diameter * cell_size)) ) ) { + // break; + //} + + float log2_diameter = log2(diameter); + vec4 scolor = textureLod(sampler3D(color_texture, texture_sampler), uvw_pos, log2_diameter); + float a = (1.0 - color.a); + color += a * scolor; + dist += half_diameter; + } + } + color *= cone_weights[i] * vec4(albedo.rgb, 1.0) * params.dynamic_range; //restore range + accum += color.rgb; + } + } + + outputs.data[cell_index] = vec4(accum, 0.0); + +#endif // MODE_SECOND_BOUNCE + + /////////////////UPDATE MIPMAPS/////////////////////////////// + +#ifdef MODE_UPDATE_MIPMAPS + + { + vec3 light_accum = vec3(0.0); + float count = 0.0; + for (uint i = 0; i < 8; i++) { + uint child_index = cell_children.data[cell_index].children[i]; + if (child_index == NO_CHILDREN) { + continue; + } + light_accum += outputs.data[child_index].rgb; + + count += 1.0; + } + + float divisor = mix(8.0, count, params.propagation); + outputs.data[cell_index] = vec4(light_accum / divisor, 0.0); + } +#endif + + ///////////////////WRITE TEXTURE///////////////////////////// + +#ifdef MODE_WRITE_TEXTURE + { + imageStore(color_tex, ivec3(posu), vec4(outputs.data[cell_index].rgb / params.dynamic_range, albedo.a)); + } +#endif + + ///////////////////DYNAMIC LIGHTING///////////////////////////// + +#ifdef MODE_DYNAMIC + + ivec2 pos_xy = ivec2(gl_GlobalInvocationID.xy); + if (any(greaterThanEqual(pos_xy, params.rect_size))) { + return; //out of bounds + } + + ivec2 uv_xy = pos_xy; + if (params.flip_x) { + uv_xy.x = params.rect_size.x - pos_xy.x - 1; + } + if (params.flip_y) { + uv_xy.y = params.rect_size.y - pos_xy.y - 1; + } + +#ifdef MODE_DYNAMIC_LIGHTING + + { + float z = params.z_base + imageLoad(depth, uv_xy).x * params.z_sign; + + ivec3 pos = params.x_dir * (params.rect_pos.x + pos_xy.x) + params.y_dir * (params.rect_pos.y + pos_xy.y) + abs(params.z_dir) * int(z); + + vec3 normal = imageLoad(source_normal, uv_xy).xyz * 2.0 - 1.0; + normal = vec3(params.x_dir) * normal.x * mix(1.0, -1.0, params.flip_x) + vec3(params.y_dir) * normal.y * mix(1.0, -1.0, params.flip_y) - vec3(params.z_dir) * normal.z; + + vec4 albedo = imageLoad(source_albedo, uv_xy); + + //determine the position in space + + vec3 accum = vec3(0.0); + for (uint i = 0; i < params.light_count; i++) { + vec3 light; + vec3 light_dir; + if (!compute_light_at_pos(i, vec3(pos) * params.pos_multiplier, normal, light, light_dir)) { + continue; + } + + light *= albedo.rgb; + + accum += max(0.0, dot(normal, -light_dir)) * light; + } + + accum += imageLoad(emission, uv_xy).xyz; + + imageStore(emission, uv_xy, vec4(accum, albedo.a)); + imageStore(depth, uv_xy, vec4(z)); + } + +#endif // MODE DYNAMIC LIGHTING + +#ifdef MODE_DYNAMIC_SHRINK + + { + vec4 accum = vec4(0.0); + float accum_z = 0.0; + float count = 0.0; + + for (int i = 0; i < 4; i++) { + ivec2 ofs = pos_xy * 2 + ivec2(i & 1, i >> 1) - params.prev_rect_ofs; + if (any(lessThan(ofs, ivec2(0))) || any(greaterThanEqual(ofs, params.prev_rect_size))) { + continue; + } + if (params.flip_x) { + ofs.x = params.prev_rect_size.x - ofs.x - 1; + } + if (params.flip_y) { + ofs.y = params.prev_rect_size.y - ofs.y - 1; + } + + vec4 light = imageLoad(source_light, ofs); + if (light.a == 0.0) { //ignore empty + continue; + } + accum += light; + float z = imageLoad(source_depth, ofs).x; + accum_z += z * 0.5; //shrink half too + count += 1.0; + } + + if (params.on_mipmap) { + accum.rgb /= mix(8.0, count, params.propagation); + accum.a /= 8.0; + } else { + accum /= 4.0; + } + + if (count == 0.0) { + accum_z = 0.0; //avoid nan + } else { + accum_z /= count; + } + +#ifdef MODE_DYNAMIC_SHRINK_WRITE + + imageStore(light, uv_xy, accum); + imageStore(depth, uv_xy, vec4(accum_z)); +#endif + +#ifdef MODE_DYNAMIC_SHRINK_PLOT + + if (accum.a < 0.001) { + return; //do not blit if alpha is too low + } + + ivec3 pos = params.x_dir * (params.rect_pos.x + pos_xy.x) + params.y_dir * (params.rect_pos.y + pos_xy.y) + abs(params.z_dir) * int(accum_z); + + float z_frac = fract(accum_z); + + for (int i = 0; i < 2; i++) { + ivec3 pos3d = pos + abs(params.z_dir) * i; + if (any(lessThan(pos3d, ivec3(0))) || any(greaterThanEqual(pos3d, params.limits))) { + //skip if offlimits + continue; + } + vec4 color_blit = accum * (i == 0 ? 1.0 - z_frac : z_frac); + vec4 color = imageLoad(color_texture, pos3d); + color.rgb *= params.dynamic_range; + +#if 0 + color.rgb = mix(color.rgb,color_blit.rgb,color_blit.a); + color.a+=color_blit.a; +#else + + float sa = 1.0 - color_blit.a; + vec4 result; + result.a = color.a * sa + color_blit.a; + if (result.a == 0.0) { + result = vec4(0.0); + } else { + result.rgb = (color.rgb * color.a * sa + color_blit.rgb * color_blit.a) / result.a; + color = result; + } + +#endif + color.rgb /= params.dynamic_range; + imageStore(color_texture, pos3d, color); + //imageStore(color_texture,pos3d,vec4(1,1,1,1)); + } +#endif // MODE_DYNAMIC_SHRINK_PLOT + } +#endif + +#endif // MODE DYNAMIC +} diff --git a/servers/rendering/renderer_rd/shaders/environment/voxel_gi_debug.glsl b/servers/rendering/renderer_rd/shaders/environment/voxel_gi_debug.glsl new file mode 100644 index 0000000000..fd7a2bf8ad --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/environment/voxel_gi_debug.glsl @@ -0,0 +1,168 @@ +#[vertex] + +#version 450 + +#VERSION_DEFINES + +struct CellData { + uint position; // xyz 10 bits + uint albedo; //rgb albedo + uint emission; //rgb normalized with e as multiplier + uint normal; //RGB normal encoded +}; + +layout(set = 0, binding = 1, std140) buffer CellDataBuffer { + CellData data[]; +} +cell_data; + +layout(set = 0, binding = 2) uniform texture3D color_tex; + +layout(set = 0, binding = 3) uniform sampler tex_sampler; + +layout(push_constant, std430) uniform Params { + mat4 projection; + uint cell_offset; + float dynamic_range; + float alpha; + uint level; + ivec3 bounds; + uint pad; +} +params; + +layout(location = 0) out vec4 color_interp; + +void main() { + const vec3 cube_triangles[36] = vec3[]( + vec3(-1.0f, -1.0f, -1.0f), + vec3(-1.0f, -1.0f, 1.0f), + vec3(-1.0f, 1.0f, 1.0f), + vec3(1.0f, 1.0f, -1.0f), + vec3(-1.0f, -1.0f, -1.0f), + vec3(-1.0f, 1.0f, -1.0f), + vec3(1.0f, -1.0f, 1.0f), + vec3(-1.0f, -1.0f, -1.0f), + vec3(1.0f, -1.0f, -1.0f), + vec3(1.0f, 1.0f, -1.0f), + vec3(1.0f, -1.0f, -1.0f), + vec3(-1.0f, -1.0f, -1.0f), + vec3(-1.0f, -1.0f, -1.0f), + vec3(-1.0f, 1.0f, 1.0f), + vec3(-1.0f, 1.0f, -1.0f), + vec3(1.0f, -1.0f, 1.0f), + vec3(-1.0f, -1.0f, 1.0f), + vec3(-1.0f, -1.0f, -1.0f), + vec3(-1.0f, 1.0f, 1.0f), + vec3(-1.0f, -1.0f, 1.0f), + vec3(1.0f, -1.0f, 1.0f), + vec3(1.0f, 1.0f, 1.0f), + vec3(1.0f, -1.0f, -1.0f), + vec3(1.0f, 1.0f, -1.0f), + vec3(1.0f, -1.0f, -1.0f), + vec3(1.0f, 1.0f, 1.0f), + vec3(1.0f, -1.0f, 1.0f), + vec3(1.0f, 1.0f, 1.0f), + vec3(1.0f, 1.0f, -1.0f), + vec3(-1.0f, 1.0f, -1.0f), + vec3(1.0f, 1.0f, 1.0f), + vec3(-1.0f, 1.0f, -1.0f), + vec3(-1.0f, 1.0f, 1.0f), + vec3(1.0f, 1.0f, 1.0f), + vec3(-1.0f, 1.0f, 1.0f), + vec3(1.0f, -1.0f, 1.0f)); + + vec3 vertex = cube_triangles[gl_VertexIndex] * 0.5 + 0.5; +#ifdef MODE_DEBUG_LIGHT_FULL + uvec3 posu = uvec3(gl_InstanceIndex % params.bounds.x, (gl_InstanceIndex / params.bounds.x) % params.bounds.y, gl_InstanceIndex / (params.bounds.y * params.bounds.x)); +#else + uint cell_index = gl_InstanceIndex + params.cell_offset; + + uvec3 posu = uvec3(cell_data.data[cell_index].position & 0x7FF, (cell_data.data[cell_index].position >> 11) & 0x3FF, cell_data.data[cell_index].position >> 21); +#endif + +#ifdef MODE_DEBUG_EMISSION + color_interp.xyz = vec3(uvec3(cell_data.data[cell_index].emission & 0x1ff, (cell_data.data[cell_index].emission >> 9) & 0x1ff, (cell_data.data[cell_index].emission >> 18) & 0x1ff)) * pow(2.0, float(cell_data.data[cell_index].emission >> 27) - 15.0 - 9.0); +#endif + +#ifdef MODE_DEBUG_COLOR + color_interp.xyz = unpackUnorm4x8(cell_data.data[cell_index].albedo).xyz; +#endif + +#ifdef MODE_DEBUG_LIGHT + color_interp = texelFetch(sampler3D(color_tex, tex_sampler), ivec3(posu), int(params.level)); + color_interp.xyz *params.dynamic_range; +#endif + + float scale = (1 << params.level); + + gl_Position = params.projection * vec4((vec3(posu) + vertex) * scale, 1.0); + +#ifdef MODE_DEBUG_LIGHT_FULL + if (color_interp.a == 0.0) { + gl_Position = vec4(0.0); //force clip and not draw + } +#else + color_interp.a = params.alpha; +#endif +} + +#[fragment] + +#version 450 + +#VERSION_DEFINES + +layout(location = 0) in vec4 color_interp; +layout(location = 0) out vec4 frag_color; + +void main() { + frag_color = color_interp; + +#ifdef MODE_DEBUG_LIGHT_FULL + + //there really is no alpha, so use dither + + int x = int(gl_FragCoord.x) % 4; + int y = int(gl_FragCoord.y) % 4; + int index = x + y * 4; + float limit = 0.0; + if (x < 8) { + if (index == 0) + limit = 0.0625; + if (index == 1) + limit = 0.5625; + if (index == 2) + limit = 0.1875; + if (index == 3) + limit = 0.6875; + if (index == 4) + limit = 0.8125; + if (index == 5) + limit = 0.3125; + if (index == 6) + limit = 0.9375; + if (index == 7) + limit = 0.4375; + if (index == 8) + limit = 0.25; + if (index == 9) + limit = 0.75; + if (index == 10) + limit = 0.125; + if (index == 11) + limit = 0.625; + if (index == 12) + limit = 1.0; + if (index == 13) + limit = 0.5; + if (index == 14) + limit = 0.875; + if (index == 15) + limit = 0.375; + } + if (frag_color.a < limit) { + discard; + } +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/environment/voxel_gi_sdf.glsl b/servers/rendering/renderer_rd/shaders/environment/voxel_gi_sdf.glsl new file mode 100644 index 0000000000..47a611a543 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/environment/voxel_gi_sdf.glsl @@ -0,0 +1,180 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 4, local_size_y = 4, local_size_z = 4) in; + +#define MAX_DISTANCE 100000.0 + +#define NO_CHILDREN 0xFFFFFFFF + +struct CellChildren { + uint children[8]; +}; + +layout(set = 0, binding = 1, std430) buffer CellChildrenBuffer { + CellChildren data[]; +} +cell_children; + +struct CellData { + uint position; // xyz 10 bits + uint albedo; //rgb albedo + uint emission; //rgb normalized with e as multiplier + uint normal; //RGB normal encoded +}; + +layout(set = 0, binding = 2, std430) buffer CellDataBuffer { + CellData data[]; +} +cell_data; + +layout(r8ui, set = 0, binding = 3) uniform restrict writeonly uimage3D sdf_tex; + +layout(push_constant, std430) uniform Params { + uint offset; + uint end; + uint pad0; + uint pad1; +} +params; + +void main() { + vec3 pos = vec3(gl_GlobalInvocationID); + float closest_dist = MAX_DISTANCE; + + for (uint i = params.offset; i < params.end; i++) { + vec3 posu = vec3(uvec3(cell_data.data[i].position & 0x7FF, (cell_data.data[i].position >> 11) & 0x3FF, cell_data.data[i].position >> 21)); + float dist = length(pos - posu); + if (dist < closest_dist) { + closest_dist = dist; + } + } + + uint dist_8; + + if (closest_dist < 0.0001) { // same cell + dist_8 = 0; //equals to -1 + } else { + dist_8 = clamp(uint(closest_dist), 0, 254) + 1; //conservative, 0 is 1, so <1 is considered solid + } + + imageStore(sdf_tex, ivec3(gl_GlobalInvocationID), uvec4(dist_8)); + //imageStore(sdf_tex,pos,uvec4(pos*2,0)); +} + +#if 0 +layout(push_constant, std430) uniform Params { + ivec3 limits; + uint stack_size; +} +params; + +float distance_to_aabb(ivec3 pos, ivec3 aabb_pos, ivec3 aabb_size) { + vec3 delta = vec3(max(ivec3(0), max(aabb_pos - pos, pos - (aabb_pos + aabb_size - ivec3(1))))); + return length(delta); +} + +void main() { + ivec3 pos = ivec3(gl_GlobalInvocationID); + + uint stack[10] = uint[](0, 0, 0, 0, 0, 0, 0, 0, 0, 0); + uint stack_indices[10] = uint[](0, 0, 0, 0, 0, 0, 0, 0, 0, 0); + ivec3 stack_positions[10] = ivec3[](ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0), ivec3(0)); + + const uint cell_orders[8] = uint[]( + 0x11f58d1, + 0xe2e70a, + 0xd47463, + 0xbb829c, + 0x8d11f5, + 0x70ae2e, + 0x463d47, + 0x29cbb8); + + bool cell_found = false; + bool cell_found_exact = false; + ivec3 closest_cell_pos; + float closest_distance = MAX_DISTANCE; + int stack_pos = 0; + + while (true) { + uint index = stack_indices[stack_pos] >> 24; + + if (index == 8) { + //go up + if (stack_pos == 0) { + break; //done going through octree + } + stack_pos--; + continue; + } + + stack_indices[stack_pos] = (stack_indices[stack_pos] & ((1 << 24) - 1)) | ((index + 1) << 24); + + uint cell_index = (stack_indices[stack_pos] >> (index * 3)) & 0x7; + uint child_cell = cell_children.data[stack[stack_pos]].children[cell_index]; + + if (child_cell == NO_CHILDREN) { + continue; + } + + ivec3 child_cell_size = params.limits >> (stack_pos + 1); + ivec3 child_cell_pos = stack_positions[stack_pos]; + + child_cell_pos += mix(ivec3(0), child_cell_size, bvec3(uvec3(index & 1, index & 2, index & 4) != uvec3(0))); + + bool is_leaf = stack_pos == (params.stack_size - 2); + + if (child_cell_pos == pos && is_leaf) { + //we may actually end up in the exact cell. + //if this happens, just abort + cell_found_exact = true; + break; + } + + if (cell_found) { + //discard by distance + float distance = distance_to_aabb(pos, child_cell_pos, child_cell_size); + if (distance >= closest_distance) { + continue; //pointless, just test next child + } else if (is_leaf) { + //closer than what we have AND end of stack, save and continue + closest_cell_pos = child_cell_pos; + closest_distance = distance; + continue; + } + } else if (is_leaf) { + //first solid cell we find, save and continue + closest_distance = distance_to_aabb(pos, child_cell_pos, child_cell_size); + closest_cell_pos = child_cell_pos; + cell_found = true; + continue; + } + + bvec3 direction = greaterThan((pos - (child_cell_pos + (child_cell_size >> 1))), ivec3(0)); + uint cell_order = 0; + cell_order |= mix(0, 1, direction.x); + cell_order |= mix(0, 2, direction.y); + cell_order |= mix(0, 4, direction.z); + + stack[stack_pos + 1] = child_cell; + stack_indices[stack_pos + 1] = cell_orders[cell_order]; //start counting + stack_positions[stack_pos + 1] = child_cell_pos; + stack_pos++; //go up stack + } + + uint dist_8; + + if (cell_found_exact) { + dist_8 = 0; //equals to -1 + } else { + float closest_distance = length(vec3(pos - closest_cell_pos)); + dist_8 = clamp(uint(closest_distance), 0, 254) + 1; //conservative, 0 is 1, so <1 is considered solid + } + + imageStore(sdf_tex, pos, uvec4(dist_8)); +} +#endif |