diff options
Diffstat (limited to 'servers/rendering/rasterizer_rd/shaders')
13 files changed, 4135 insertions, 224 deletions
diff --git a/servers/rendering/rasterizer_rd/shaders/SCsub b/servers/rendering/rasterizer_rd/shaders/SCsub index a454d144aa..67f4edc626 100644 --- a/servers/rendering/rasterizer_rd/shaders/SCsub +++ b/servers/rendering/rasterizer_rd/shaders/SCsub @@ -28,3 +28,10 @@ if "RD_GLSL" in env["BUILDERS"]: env.RD_GLSL("screen_space_reflection_scale.glsl") env.RD_GLSL("subsurface_scattering.glsl") env.RD_GLSL("specular_merge.glsl") + env.RD_GLSL("gi.glsl") + env.RD_GLSL("resolve.glsl") + env.RD_GLSL("sdfgi_preprocess.glsl") + env.RD_GLSL("sdfgi_integrate.glsl") + env.RD_GLSL("sdfgi_direct_light.glsl") + env.RD_GLSL("sdfgi_debug.glsl") + env.RD_GLSL("sdfgi_debug_probes.glsl") diff --git a/servers/rendering/rasterizer_rd/shaders/copy_to_fb.glsl b/servers/rendering/rasterizer_rd/shaders/copy_to_fb.glsl index b1cfe1e91e..9751e13b4e 100644 --- a/servers/rendering/rasterizer_rd/shaders/copy_to_fb.glsl +++ b/servers/rendering/rasterizer_rd/shaders/copy_to_fb.glsl @@ -47,16 +47,26 @@ layout(push_constant, binding = 1, std430) uniform Params { bool force_luminance; bool alpha_to_zero; - uint pad[2]; + bool srgb; + uint pad; } params; layout(location = 0) in vec2 uv_interp; layout(set = 0, binding = 0) uniform sampler2D source_color; - +#ifdef MODE_TWO_SOURCES +layout(set = 1, binding = 0) uniform sampler2D source_color2; +#endif layout(location = 0) out vec4 frag_color; +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))); +} + void main() { vec2 uv = uv_interp; @@ -89,11 +99,17 @@ void main() { } #endif vec4 color = textureLod(source_color, uv, 0.0); +#ifdef MODE_TWO_SOURCES + color += textureLod(source_color2, uv, 0.0); +#endif if (params.force_luminance) { color.rgb = vec3(max(max(color.r, color.g), color.b)); } if (params.alpha_to_zero) { color.rgb *= color.a; } + if (params.srgb) { + color.rgb = linear_to_srgb(color.rgb); + } frag_color = color; } diff --git a/servers/rendering/rasterizer_rd/shaders/gi.glsl b/servers/rendering/rasterizer_rd/shaders/gi.glsl new file mode 100644 index 0000000000..a1939f75ad --- /dev/null +++ b/servers/rendering/rasterizer_rd/shaders/gi.glsl @@ -0,0 +1,663 @@ +#[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 giprobe_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_GI_PROBES 8 + +struct GIProbeData { + mat4 xform; + vec3 bounds; + float dynamic_range; + + float bias; + float normal_bias; + bool blend_ambient; + uint texture_slot; + + float anisotropy_strength; + float ambient_occlusion; + float ambient_occlusion_size; + uint pad2; +}; + +layout(set = 0, binding = 16, std140) uniform GIProbes { + GIProbeData data[MAX_GI_PROBES]; +} +gi_probes; + +layout(set = 0, binding = 17) uniform texture3D gi_probe_textures[MAX_GI_PROBES]; + +layout(push_constant, binding = 0, std430) uniform Params { + ivec2 screen_size; + float z_near; + float z_far; + + vec4 proj_info; + + uint max_giprobes; + bool high_quality_vct; + bool use_sdfgi; + bool orthogonal; + + vec3 ao_color; + uint pad; + + mat3x4 cam_rotation; +} +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) { + 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; +} + +void sdfgi_probe_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; + sdfgi_probe_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; + sdfgi_probe_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 1 + 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 + while (length(ray_pos) < max_distance) { + for (uint i = 0; i < sdfgi.max_cascades; i++) { + if (i >= cascade && length(ray_pos) < radius_sizes[i]) { + 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 distance = texture(sampler3D(sdf_cascades[i], linear_sampler), pos).r * 255.0 - 1.1; + + vec4 hit_light = vec4(0.0); + if (distance < softness) { + hit_light.rgb = texture(sampler3D(light_cascades[i], linear_sampler), pos).rgb; + hit_light.rgb *= 0.5; //approximation given value read is actually meant for anisotropy + hit_light.a = clamp(1.0 - (distance / softness), 0.0, 1.0); + hit_light.rgb *= hit_light.a; + } + + distance /= sdfgi.cascades[i].to_cell; + + if (i < (sdfgi.max_cascades - 1)) { + pos = ray_pos - sdfgi.cascades[i + 1].position; + pos *= sdfgi.cascades[i + 1].to_cell * pos_to_uvw; + + float distance2 = texture(sampler3D(sdf_cascades[i + 1], linear_sampler), pos).r * 255.0 - 1.1; + + vec4 hit_light2 = vec4(0.0); + if (distance2 < softness) { + hit_light2.rgb = texture(sampler3D(light_cascades[i + 1], linear_sampler), pos).rgb; + hit_light2.rgb *= 0.5; //approximation given value read is actually meant for anisotropy + hit_light2.a = clamp(1.0 - (distance2 / softness), 0.0, 1.0); + hit_light2.rgb *= hit_light2.a; + } + + float prev_radius = i == 0 ? 0.0 : radius_sizes[i - 1]; + float blend = clamp((length(ray_pos) - prev_radius) / (radius_sizes[i] - prev_radius), 0.0, 1.0); + + distance2 /= sdfgi.cascades[i + 1].to_cell; + + hit_light = mix(hit_light, hit_light2, blend); + distance = mix(distance, distance2, blend); + } + + light_accum += hit_light; + ray_pos += ray_dir * distance; + break; + } + } + + if (light_accum.a > 0.99) { + break; + } + } + + 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; + } + } + +#endif + + 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 gi_probe_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 = (gi_probes.data[index].xform * vec4(position, 1.0)).xyz; + ref_vec = normalize((gi_probes.data[index].xform * vec4(ref_vec, 0.0)).xyz); + normal = normalize((gi_probes.data[index].xform * vec4(normal, 0.0)).xyz); + + position += normal * gi_probes.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, gi_probes.data[index].bounds))))) { + return; + } + + mat3 dir_xform = mat3(gi_probes.data[index].xform) * normal_xform; + + vec3 blendv = abs(position / gi_probes.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(gi_probes.data[index].bounds); + vec3 cell_size = 1.0 / gi_probes.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(gi_probe_textures[index], cell_size, position, dir, cone_angle_tan, max_distance, gi_probes.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(gi_probe_textures[index], cell_size, position, dir, max_distance, gi_probes.data[index].bias); + } + } + + if (gi_probes.data[index].ambient_occlusion > 0.001) { + float size = 1.0 + gi_probes.data[index].ambient_occlusion_size * 7.0; + + float taps, blend; + blend = modf(size, taps); + float ao = 0.0; + for (float i = 1.0; i <= taps; i++) { + vec3 ofs = (position + normal * (i * 0.5 + 1.0)) * cell_size; + ao += textureLod(sampler3D(gi_probe_textures[index], linear_sampler_with_mipmaps), ofs, i - 1.0).a * i; + } + + if (blend > 0.001) { + vec3 ofs = (position + normal * ((taps + 1.0) * 0.5 + 1.0)) * cell_size; + ao += textureLod(sampler3D(gi_probe_textures[index], linear_sampler_with_mipmaps), ofs, taps).a * (taps + 1.0) * blend; + } + + ao = 1.0 - min(1.0, ao); + + light.rgb = mix(params.ao_color, light.rgb, mix(1.0, ao, gi_probes.data[index].ambient_occlusion)); + } + + light.rgb *= gi_probes.data[index].dynamic_range; + if (!gi_probes.data[index].blend_ambient) { + light.a = 1.0; + } + + out_diff += light * blend; + + //radiance + vec4 irr_light = voxel_cone_trace(gi_probe_textures[index], cell_size, position, ref_vec, tan(roughness * 0.5 * M_PI * 0.99), max_distance, gi_probes.data[index].bias); + irr_light.rgb *= gi_probes.data[index].dynamic_range; + if (!gi_probes.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 main() { + // Pixel being shaded + ivec2 pos = ivec2(gl_GlobalInvocationID.xy); + if (any(greaterThanEqual(pos, params.screen_size))) { //too large, do nothing + return; + } + + vec3 vertex = reconstruct_position(pos); + vertex.y = -vertex.y; + + vec4 normal_roughness = fetch_normal_and_roughness(pos); + vec3 normal = normal_roughness.xyz; + + vec4 ambient_light = vec4(0.0), reflection_light = vec4(0.0); + + if (normal.length() > 0.5) { + //valid normal, can do GI + float roughness = normal_roughness.w; + + vertex = mat3(params.cam_rotation) * vertex; + normal = normalize(mat3(params.cam_rotation) * normal); + + vec3 reflection = normalize(reflect(normalize(vertex), normal)); + + if (params.use_sdfgi) { + sdfgi_process(vertex, normal, reflection, roughness, ambient_light, reflection_light); + } + + if (params.max_giprobes > 0) { + uvec2 giprobe_tex = texelFetch(usampler2D(giprobe_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_giprobes; i++) { + if (any(equal(uvec2(i), giprobe_tex))) { + gi_probe_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; + } + + if (params.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; + } + } + } + + imageStore(ambient_buffer, pos, ambient_light); + imageStore(reflection_buffer, pos, reflection_light); +} diff --git a/servers/rendering/rasterizer_rd/shaders/resolve.glsl b/servers/rendering/rasterizer_rd/shaders/resolve.glsl new file mode 100644 index 0000000000..9429a66dc9 --- /dev/null +++ b/servers/rendering/rasterizer_rd/shaders/resolve.glsl @@ -0,0 +1,110 @@ +#[compute] + +#version 450 + +VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +#ifdef MODE_RESOLVE_GI +layout(set = 0, binding = 0) uniform sampler2DMS source_depth; +layout(set = 0, binding = 1) uniform sampler2DMS source_normal_roughness; + +layout(r32f, set = 1, binding = 0) uniform restrict writeonly image2D dest_depth; +layout(rgba8, set = 1, binding = 1) uniform restrict writeonly image2D dest_normal_roughness; + +#ifdef GIPROBE_RESOLVE +layout(set = 2, binding = 0) uniform usampler2DMS source_giprobe; +layout(rg8ui, set = 3, binding = 0) uniform restrict writeonly uimage2D dest_giprobe; +#endif + +#endif + +layout(push_constant, binding = 16, std430) uniform Params { + ivec2 screen_size; + int sample_count; + uint pad; +} +params; + +void main() { + // Pixel being shaded + ivec2 pos = ivec2(gl_GlobalInvocationID.xy); + if (any(greaterThanEqual(pos, params.screen_size))) { //too large, do nothing + return; + } + +#ifdef MODE_RESOLVE_GI + + float best_depth = 1e20; + vec4 best_normal_roughness = vec4(0.0); +#ifdef GIPROBE_RESOLVE + uvec2 best_giprobe; +#endif + +#if 0 + + for(int i=0;i<params.sample_count;i++) { + float depth = texelFetch(source_depth,pos,i).r; + if (depth < best_depth) { //use the depth closest to camera + best_depth = depth; + best_normal_roughness = texelFetch(source_normal_roughness,pos,i); + +#ifdef GIPROBE_RESOLVE + best_giprobe = texelFetch(source_giprobe,pos,i).rg; +#endif + } + } + +#else + + float depths[16]; + int depth_indices[16]; + int depth_amount[16]; + int depth_count = 0; + + for (int i = 0; i < params.sample_count; i++) { + float depth = texelFetch(source_depth, pos, i).r; + int depth_index = -1; + for (int j = 0; j < depth_count; j++) { + if (abs(depths[j] - depth) < 0.000001) { + depth_index = j; + break; + } + } + + if (depth_index == -1) { + depths[depth_count] = depth; + depth_indices[depth_count] = i; + depth_amount[depth_count] = 1; + depth_count += 1; + } else { + depth_amount[depth_index] += 1; + } + } + + int depth_least = 0xFFFF; + int best_index = 0; + for (int j = 0; j < depth_count; j++) { + if (depth_amount[j] < depth_least) { + best_index = depth_indices[j]; + depth_least = depth_amount[j]; + } + } + + best_depth = texelFetch(source_depth, pos, best_index).r; + best_normal_roughness = texelFetch(source_normal_roughness, pos, best_index); +#ifdef GIPROBE_RESOLVE + best_giprobe = texelFetch(source_giprobe, pos, best_index).rg; +#endif + +#endif + + imageStore(dest_depth, pos, vec4(best_depth)); + imageStore(dest_normal_roughness, pos, vec4(best_normal_roughness)); +#ifdef GIPROBE_RESOLVE + imageStore(dest_giprobe, pos, uvec4(best_giprobe, 0, 0)); +#endif + +#endif +} diff --git a/servers/rendering/rasterizer_rd/shaders/scene_high_end.glsl b/servers/rendering/rasterizer_rd/shaders/scene_high_end.glsl index 9f42b0f814..d6a56b2543 100644 --- a/servers/rendering/rasterizer_rd/shaders/scene_high_end.glsl +++ b/servers/rendering/rasterizer_rd/shaders/scene_high_end.glsl @@ -258,7 +258,6 @@ VERTEX_SHADER_CODE } } #endif - #ifdef MODE_RENDER_MATERIAL if (scene_data.material_uv2_mode) { gl_Position.xy = (uv2_attrib.xy + draw_call.bake_uv2_offset) * 2.0 - 1.0; @@ -341,11 +340,13 @@ layout(location = 4) out float depth_output_buffer; #endif -#ifdef MODE_RENDER_NORMAL -layout(location = 0) out vec4 normal_output_buffer; -#ifdef MODE_RENDER_ROUGHNESS -layout(location = 1) out float roughness_output_buffer; -#endif //MODE_RENDER_ROUGHNESS +#ifdef MODE_RENDER_NORMAL_ROUGHNESS +layout(location = 0) out vec4 normal_roughness_output_buffer; + +#ifdef MODE_RENDER_GIPROBE +layout(location = 1) out uvec2 giprobe_buffer; +#endif + #endif //MODE_RENDER_NORMAL #else // RENDER DEPTH @@ -1321,37 +1322,39 @@ void reflection_process(uint ref_index, vec3 vertex, vec3 normal, float roughnes reflection_accum += reflection; } -#if !defined(USE_LIGHTMAP) && !defined(USE_VOXEL_CONE_TRACING) - if (reflections.data[ref_index].ambient.a > 0.0) { //compute ambient using skybox + switch (reflections.data[ref_index].ambient_mode) { + case REFLECTION_AMBIENT_DISABLED: { + //do nothing + } break; + case REFLECTION_AMBIENT_ENVIRONMENT: { + //do nothing + vec3 local_amb_vec = (reflections.data[ref_index].local_matrix * vec4(normal, 0.0)).xyz; - vec3 local_amb_vec = (reflections.data[ref_index].local_matrix * vec4(normal, 0.0)).xyz; + vec4 ambient_out; - vec4 ambient_out; - - ambient_out.rgb = textureLod(samplerCubeArray(reflection_atlas, material_samplers[SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP]), vec4(local_amb_vec, reflections.data[ref_index].index), MAX_ROUGHNESS_LOD).rgb; - - ambient_out.a = blend; - ambient_out.rgb = mix(reflections.data[ref_index].ambient.rgb, ambient_out.rgb, reflections.data[ref_index].ambient.a); - if (reflections.data[ref_index].params.z < 0.5) { - ambient_out.rgb = mix(ambient_light, ambient_out.rgb, blend); - } + ambient_out.rgb = textureLod(samplerCubeArray(reflection_atlas, material_samplers[SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP]), vec4(local_amb_vec, reflections.data[ref_index].index), MAX_ROUGHNESS_LOD).rgb; + ambient_out.a = blend; + if (reflections.data[ref_index].params.z < 0.5) { //interior + ambient_out.rgb = mix(ambient_light, ambient_out.rgb, blend); + } - ambient_out.rgb *= ambient_out.a; - ambient_accum += ambient_out; - } else { - vec4 ambient_out; - ambient_out.a = blend; - ambient_out.rgb = reflections.data[ref_index].ambient.rgb; - if (reflections.data[ref_index].params.z < 0.5) { - ambient_out.rgb = mix(ambient_light, ambient_out.rgb, blend); - } - ambient_out.rgb *= ambient_out.a; - ambient_accum += ambient_out; + ambient_out.rgb *= ambient_out.a; + ambient_accum += ambient_out; + } break; + case REFLECTION_AMBIENT_COLOR: { + vec4 ambient_out; + ambient_out.a = blend; + ambient_out.rgb = reflections.data[ref_index].ambient; + if (reflections.data[ref_index].params.z < 0.5) { + ambient_out.rgb = mix(ambient_light, ambient_out.rgb, blend); + } + ambient_out.rgb *= ambient_out.a; + ambient_accum += ambient_out; + } break; } -#endif //USE_LIGHTMAP or VCT } -#ifdef USE_VOXEL_CONE_TRACING +#ifdef USE_FORWARD_GI //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) { @@ -1375,42 +1378,6 @@ vec4 voxel_cone_trace(texture3D probe, vec3 cell_size, vec3 pos, vec3 direction, return color; } -#ifndef GI_PROBE_HIGH_QUALITY -//faster version for 45 degrees - -#ifdef GI_PROBE_USE_ANISOTROPY - -vec4 voxel_cone_trace_anisotropic_45_degrees(texture3D probe, texture3D aniso_pos, texture3D aniso_neg, vec3 normal, 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); - float radius = max(0.5, tan_half_angle * 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, material_samplers[SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP]), uvw_pos, lod_level); - vec3 aniso_neg = textureLod(sampler3D(aniso_neg, material_samplers[SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP]), uvw_pos, lod_level).rgb; - vec3 aniso_pos = textureLod(sampler3D(aniso_pos, material_samplers[SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP]), uvw_pos, lod_level).rgb; - - scolor.rgb *= dot(max(vec3(0.0), (normal * aniso_pos)), vec3(1.0)) + dot(max(vec3(0.0), (-normal * aniso_neg)), vec3(1.0)); - lod_level += 1.0; - - float a = (1.0 - color.a); - scolor *= a; - color += scolor; - dist += radius; - radius = max(0.5, tan_half_angle * dist); - } - - return color; -} -#else - vec4 voxel_cone_trace_45_degrees(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); @@ -1437,41 +1404,6 @@ vec4 voxel_cone_trace_45_degrees(texture3D probe, vec3 cell_size, vec3 pos, vec3 return color; } -#endif - -#elif defined(GI_PROBE_USE_ANISOTROPY) - -//standard voxel cone trace -vec4 voxel_cone_trace_anisotropic(texture3D probe, texture3D aniso_pos, texture3D aniso_neg, vec3 normal, 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; - } - float log2_diameter = log2(diameter); - vec4 scolor = textureLod(sampler3D(probe, material_samplers[SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP]), uvw_pos, log2_diameter); - vec3 aniso_neg = textureLod(sampler3D(aniso_neg, material_samplers[SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP]), uvw_pos, log2_diameter).rgb; - vec3 aniso_pos = textureLod(sampler3D(aniso_pos, material_samplers[SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP]), uvw_pos, log2_diameter).rgb; - - scolor.rgb *= dot(max(vec3(0.0), (normal * aniso_pos)), vec3(1.0)) + dot(max(vec3(0.0), (-normal * aniso_neg)), vec3(1.0)); - - float a = (1.0 - color.a); - scolor *= a; - color += scolor; - dist += half_diameter; - } - - return color; -} - -#endif - void gi_probe_compute(uint index, vec3 position, vec3 normal, vec3 ref_vec, mat3 normal_xform, float roughness, vec3 ambient, vec3 environment, inout vec4 out_spec, inout vec4 out_diff) { position = (gi_probes.data[index].xform * vec4(position, 1.0)).xyz; ref_vec = normalize((gi_probes.data[index].xform * vec4(ref_vec, 0.0)).xyz); @@ -1493,31 +1425,6 @@ void gi_probe_compute(uint index, vec3 position, vec3 normal, vec3 ref_vec, mat3 //radiance -#ifdef GI_PROBE_HIGH_QUALITY - -#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 cone_angle_tan = 0.577; - -#elif defined(GI_PROBE_LOW_QUALITY) - -#define MAX_CONE_DIRS 1 - - vec3 cone_dirs[MAX_CONE_DIRS] = vec3[]( - vec3(0.0, 0.0, 1.0)); - - float cone_weights[MAX_CONE_DIRS] = float[](1.0); - float cone_angle_tan = 4; //~76 degrees -#else // MEDIUM QUALITY - #define MAX_CONE_DIRS 4 vec3 cone_dirs[MAX_CONE_DIRS] = vec3[]( @@ -1529,31 +1436,13 @@ void gi_probe_compute(uint index, vec3 position, vec3 normal, vec3 ref_vec, mat3 float cone_weights[MAX_CONE_DIRS] = float[](0.25, 0.25, 0.25, 0.25); float cone_angle_tan = 0.98269; -#endif vec3 light = vec3(0.0); for (int i = 0; i < MAX_CONE_DIRS; i++) { vec3 dir = normalize((gi_probes.data[index].xform * vec4(normal_xform * cone_dirs[i], 0.0)).xyz); -#if defined(GI_PROBE_HIGH_QUALITY) || defined(GI_PROBE_LOW_QUALITY) - -#ifdef GI_PROBE_USE_ANISOTROPY - vec4 cone_light = voxel_cone_trace_anisotropic(gi_probe_textures[gi_probes.data[index].texture_slot], gi_probe_textures[gi_probes.data[index].texture_slot + 1], gi_probe_textures[gi_probes.data[index].texture_slot + 2], normalize(mix(dir, normal, gi_probes.data[index].anisotropy_strength)), cell_size, position, dir, cone_angle_tan, max_distance, gi_probes.data[index].bias); -#else - - vec4 cone_light = voxel_cone_trace(gi_probe_textures[gi_probes.data[index].texture_slot], cell_size, position, dir, cone_angle_tan, max_distance, gi_probes.data[index].bias); - -#endif // GI_PROBE_USE_ANISOTROPY - -#else - -#ifdef GI_PROBE_USE_ANISOTROPY - vec4 cone_light = voxel_cone_trace_anisotropic_45_degrees(gi_probe_textures[gi_probes.data[index].texture_slot], gi_probe_textures[gi_probes.data[index].texture_slot + 1], gi_probe_textures[gi_probes.data[index].texture_slot + 2], normalize(mix(dir, normal, gi_probes.data[index].anisotropy_strength)), cell_size, position, dir, cone_angle_tan, max_distance, gi_probes.data[index].bias); -#else - vec4 cone_light = voxel_cone_trace_45_degrees(gi_probe_textures[gi_probes.data[index].texture_slot], cell_size, position, dir, cone_angle_tan, max_distance, gi_probes.data[index].bias); -#endif // GI_PROBE_USE_ANISOTROPY + vec4 cone_light = voxel_cone_trace_45_degrees(gi_probe_textures[index], cell_size, position, dir, cone_angle_tan, max_distance, gi_probes.data[index].bias); -#endif if (gi_probes.data[index].blend_ambient) { cone_light.rgb = mix(ambient, cone_light.rgb, min(1.0, cone_light.a / 0.95)); } @@ -1562,33 +1451,10 @@ void gi_probe_compute(uint index, vec3 position, vec3 normal, vec3 ref_vec, mat3 } light *= gi_probes.data[index].dynamic_range; - - if (gi_probes.data[index].ambient_occlusion > 0.001) { - float size = 1.0 + gi_probes.data[index].ambient_occlusion_size * 7.0; - - float taps, blend; - blend = modf(size, taps); - float ao = 0.0; - for (float i = 1.0; i <= taps; i++) { - vec3 ofs = (position + normal * (i * 0.5 + 1.0)) * cell_size; - ao += textureLod(sampler3D(gi_probe_textures[gi_probes.data[index].texture_slot], material_samplers[SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP]), ofs, i - 1.0).a * i; - } - - if (blend > 0.001) { - vec3 ofs = (position + normal * ((taps + 1.0) * 0.5 + 1.0)) * cell_size; - ao += textureLod(sampler3D(gi_probe_textures[gi_probes.data[index].texture_slot], material_samplers[SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP]), ofs, taps).a * (taps + 1.0) * blend; - } - - ao = 1.0 - min(1.0, ao); - - light = mix(scene_data.ao_color.rgb, light, mix(1.0, ao, gi_probes.data[index].ambient_occlusion)); - } - out_diff += vec4(light * blend, blend); //irradiance -#ifndef GI_PROBE_LOW_QUALITY - vec4 irr_light = voxel_cone_trace(gi_probe_textures[gi_probes.data[index].texture_slot], cell_size, position, ref_vec, tan(roughness * 0.5 * M_PI * 0.99), max_distance, gi_probes.data[index].bias); + vec4 irr_light = voxel_cone_trace(gi_probe_textures[index], cell_size, position, ref_vec, tan(roughness * 0.5 * M_PI * 0.99), max_distance, gi_probes.data[index].bias); if (gi_probes.data[index].blend_ambient) { irr_light.rgb = mix(environment, irr_light.rgb, min(1.0, irr_light.a / 0.95)); } @@ -1596,10 +1462,142 @@ void gi_probe_compute(uint index, vec3 position, vec3 normal, vec3 ref_vec, mat3 //irr_light=vec3(0.0); out_spec += vec4(irr_light.rgb * blend, blend); -#endif } -#endif //USE_VOXEL_CONE_TRACING +#endif //USE_FORWARD_GI + +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 sdfgi_process(uint cascade, vec3 cascade_pos, vec3 cam_pos, vec3 cam_normal, vec3 cam_specular_normal, bool use_specular, float roughness, out vec3 diffuse_light, out vec3 specular_light, out float blend) { + 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; + + if (use_specular) { + specular_accum = vec3(0.0); + specular_posf = (vec3(tex_pos) + vec3(octahedron_encode(cam_specular_normal) * float(SDFGI_OCT_SIZE), 0.0)) * sdfgi.lightprobe_tex_pixel_size; + } + + 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(sdfgi_occlusion_cascades, material_samplers[SAMPLER_LINEAR_CLAMP]), 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(sdfgi_lightprobe_texture, material_samplers[SAMPLER_LINEAR_CLAMP]), pos_uvw, 0.0).rgb; + + diffuse_accum += vec4(diffuse * weight, weight); + + if (use_specular) { + 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(sdfgi_lightprobe_texture, material_samplers[SAMPLER_LINEAR_CLAMP]), pos_uvw + vec3(0, 0, float(sdfgi.max_cascades)), 0.0).rgb; + } + if (roughness > 0.5) { + specular = mix(specular, textureLod(sampler2DArray(sdfgi_lightprobe_texture, material_samplers[SAMPLER_LINEAR_CLAMP]), pos_uvw, 0.0).rgb, (roughness - 0.5) * 2.0); + } + + specular_accum += specular * weight; + } + } + + if (diffuse_accum.a > 0.0) { + diffuse_accum.rgb /= diffuse_accum.a; + } + + diffuse_light = diffuse_accum.rgb; + + if (use_specular) { + if (diffuse_accum.a > 0.0) { + specular_accum /= diffuse_accum.a; + } + + specular_light = specular_accum; + } + + { + //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; + } + } +} #endif //!defined(MODE_RENDER_DEPTH) && !defined(MODE_UNSHADED) @@ -1812,6 +1810,15 @@ FRAGMENT_SHADER_CODE #endif //not render depth /////////////////////// LIGHTING ////////////////////////////// + if (scene_data.roughness_limiter_enabled) { + //http://www.jp.square-enix.com/tech/library/pdf/ImprovedGeometricSpecularAA.pdf + float roughness2 = roughness * roughness; + vec3 dndu = dFdx(normal), dndv = dFdx(normal); + float variance = scene_data.roughness_limiter_amount * (dot(dndu, dndu) + dot(dndv, dndv)); + float kernelRoughness2 = min(2.0 * variance, scene_data.roughness_limiter_limit); //limit effect + float filteredRoughness2 = min(1.0, roughness2 + kernelRoughness2); + roughness = sqrt(filteredRoughness2); + } //apply energy conservation vec3 specular_light = vec3(0.0, 0.0, 0.0); @@ -1820,11 +1827,6 @@ FRAGMENT_SHADER_CODE #if !defined(MODE_RENDER_DEPTH) && !defined(MODE_UNSHADED) - if (scene_data.roughness_limiter_enabled) { - float limit = texelFetch(sampler2D(roughness_buffer, material_samplers[SAMPLER_NEAREST_CLAMP]), ivec2(gl_FragCoord.xy), 0).r; - roughness = max(roughness, limit); - } - if (scene_data.use_reflection_cubemap) { vec3 ref_vec = reflect(-view, normal); ref_vec = scene_data.radiance_inverse_xform * ref_vec; @@ -1871,7 +1873,6 @@ FRAGMENT_SHADER_CODE #endif #if !defined(MODE_RENDER_DEPTH) && !defined(MODE_UNSHADED) - //gi probes #ifdef USE_LIGHTMAP @@ -1928,10 +1929,80 @@ FRAGMENT_SHADER_CODE ambient_light += textureLod(sampler2DArray(lightmap_textures[ofs], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw, 0.0).rgb; } } -#endif - //lightmap capture +#elif defined(USE_FORWARD_GI) + + if (bool(instances.data[instance_index].flags & INSTANCE_FLAGS_USE_SDFGI)) { //has lightmap capture + + //make vertex orientation the world one, but still align to camera + vec3 cam_pos = mat3(scene_data.camera_matrix) * vertex; + vec3 cam_normal = mat3(scene_data.camera_matrix) * normal; + vec3 cam_reflection = mat3(scene_data.camera_matrix) * reflect(-view, normal); + + //apply y-mult + cam_pos.y *= sdfgi.y_mult; + cam_normal.y *= sdfgi.y_mult; + cam_normal = normalize(cam_normal); + cam_reflection.y *= sdfgi.y_mult; + cam_normal = normalize(cam_normal); + cam_reflection = normalize(cam_reflection); + + 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) { + bool use_specular = true; + float blend; + vec3 diffuse, specular; + sdfgi_process(cascade, cascade_pos, cam_pos, cam_normal, cam_reflection, use_specular, roughness, diffuse, specular, blend); + + if (blend > 0.0) { + //blend + if (cascade == sdfgi.max_cascades - 1) { + diffuse = mix(diffuse, ambient_light, blend); + if (use_specular) { + specular = mix(specular, specular_light, blend); + } + } else { + vec3 diffuse2, specular2; + float blend2; + cascade_pos = (cam_pos - sdfgi.cascades[cascade + 1].position) * sdfgi.cascades[cascade + 1].to_probe; + sdfgi_process(cascade + 1, cascade_pos, cam_pos, cam_normal, cam_reflection, use_specular, roughness, diffuse2, specular2, blend2); + diffuse = mix(diffuse, diffuse2, blend); + if (use_specular) { + specular = mix(specular, specular2, blend); + } + } + } + + ambient_light = diffuse; + if (use_specular) { + specular_light = specular; + } + } + } -#ifdef USE_VOXEL_CONE_TRACING if (bool(instances.data[instance_index].flags & INSTANCE_FLAGS_USE_GIPROBE)) { // process giprobes uint index1 = instances.data[instance_index].gi_offset & 0xFFFF; @@ -1963,6 +2034,56 @@ FRAGMENT_SHADER_CODE specular_light = spec_accum.rgb; ambient_light = amb_accum.rgb; } +#else + if (bool(instances.data[instance_index].flags & INSTANCE_FLAGS_USE_GI_BUFFERS)) { //use GI buffers + + ivec2 coord; + + if (scene_data.gi_upscale_for_msaa) { + /* + //find the closest depth to upscale from, based on neighbours + ivec2 base_coord = ivec2(gl_FragCoord.xy); + float z_dist = gl_FragCoord.z; + ivec2 closest_coord = base_coord; + float closest_z_dist = abs(texelFetch(sampler2D(depth_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), base_coord,0).r-z_dist); + + for(int i=0;i<4;i++) { + const ivec2 neighbours[4]=ivec2[](ivec2(-1,0),ivec2(1,0),ivec2(0,-1),ivec2(0,1)); + ivec2 neighbour_coord = base_coord + neighbours[i]; + float neighbour_z_dist = abs(texelFetch(sampler2D(depth_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), neighbour_coord,0).r-z_dist); + if (neighbour_z_dist < closest_z_dist) { + closest_z_dist = neighbour_z_dist; + closest_coord = neighbour_coord; + } + } + +*/ + ivec2 base_coord = ivec2(gl_FragCoord.xy); + ivec2 closest_coord = base_coord; + float closest_ang = dot(normal, texelFetch(sampler2D(normal_roughness_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), base_coord, 0).xyz * 2.0 - 1.0); + + for (int i = 0; i < 4; i++) { + const ivec2 neighbours[4] = ivec2[](ivec2(-1, 0), ivec2(1, 0), ivec2(0, -1), ivec2(0, 1)); + ivec2 neighbour_coord = base_coord + neighbours[i]; + float neighbour_ang = dot(normal, texelFetch(sampler2D(normal_roughness_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), neighbour_coord, 0).xyz * 2.0 - 1.0); + if (neighbour_ang > closest_ang) { + closest_ang = neighbour_ang; + closest_coord = neighbour_coord; + } + } + + coord = closest_coord; + + } else { + coord = ivec2(gl_FragCoord.xy); + } + + vec4 buffer_ambient = texelFetch(sampler2D(ambient_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), coord, 0); + vec4 buffer_reflection = texelFetch(sampler2D(reflection_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), coord, 0); + + ambient_light = mix(ambient_light, buffer_ambient.rgb, buffer_ambient.a); + specular_light = mix(specular_light, buffer_reflection.rgb, buffer_reflection.a); + } #endif { // process reflections @@ -2376,6 +2497,97 @@ FRAGMENT_SHADER_CODE #ifdef MODE_RENDER_DEPTH +#ifdef MODE_RENDER_SDF + + { + vec3 local_pos = (scene_data.sdf_to_bounds * vec4(vertex, 1.0)).xyz; + ivec3 grid_pos = scene_data.sdf_offset + ivec3(local_pos * vec3(scene_data.sdf_size)); + + uint albedo16 = 0x1; //solid flag + albedo16 |= clamp(uint(albedo.r * 31.0), 0, 31) << 11; + albedo16 |= clamp(uint(albedo.g * 31.0), 0, 31) << 6; + albedo16 |= clamp(uint(albedo.b * 31.0), 0, 31) << 1; + + imageStore(albedo_volume_grid, grid_pos, uvec4(albedo16)); + + uint facing_bits = 0; + 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)); + + vec3 cam_normal = mat3(scene_data.camera_matrix) * normalize(normal_interp); + + float closest_dist = -1e20; + + for (uint i = 0; i < 6; i++) { + float d = dot(cam_normal, aniso_dir[i]); + if (d > closest_dist) { + closest_dist = d; + facing_bits = (1 << i); + } + } + + imageAtomicOr(geom_facing_grid, grid_pos, facing_bits); //store facing bits + + if (length(emission) > 0.001) { + float lumas[6]; + vec3 light_total = vec3(0); + + for (int i = 0; i < 6; i++) { + float strength = max(0.0, dot(cam_normal, aniso_dir[i])); + vec3 light = emission * strength; + light_total += light; + lumas[i] = max(light.r, max(light.g, light.b)); + } + + float luma_total = max(light_total.r, max(light_total.g, light_total.b)); + + uint light_aniso = 0; + + for (int i = 0; i < 6; i++) { + light_aniso |= min(31, uint((lumas[i] / luma_total) * 31.0)) << (i * 5); + } + + //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); + //store as 8985 to have 2 extra neighbour bits + uint light_rgbe = ((uint(sRed) & 0x1FF) >> 1) | ((uint(sGreen) & 0x1FF) << 8) | (((uint(sBlue) & 0x1FF) >> 1) << 17) | ((uint(exps) & 0x1F) << 25); + + imageStore(emission_grid, grid_pos, uvec4(light_rgbe)); + imageStore(emission_aniso_grid, grid_pos, uvec4(light_aniso)); + } + } + +#endif + #ifdef MODE_RENDER_MATERIAL albedo_output_buffer.rgb = albedo; @@ -2398,11 +2610,21 @@ FRAGMENT_SHADER_CODE emission_output_buffer.a = 0.0; #endif -#ifdef MODE_RENDER_NORMAL - normal_output_buffer = vec4(normal * 0.5 + 0.5, 0.0); -#ifdef MODE_RENDER_ROUGHNESS - roughness_output_buffer = roughness; -#endif //MODE_RENDER_ROUGHNESS +#ifdef MODE_RENDER_NORMAL_ROUGHNESS + normal_roughness_output_buffer = vec4(normal * 0.5 + 0.5, roughness); + +#ifdef MODE_RENDER_GIPROBE + if (bool(instances.data[instance_index].flags & INSTANCE_FLAGS_USE_GIPROBE)) { // process giprobes + uint index1 = instances.data[instance_index].gi_offset & 0xFFFF; + uint index2 = instances.data[instance_index].gi_offset >> 16; + giprobe_buffer.x = index1 & 0xFF; + giprobe_buffer.y = index2 & 0xFF; + } else { + giprobe_buffer.x = 0xFF; + giprobe_buffer.y = 0xFF; + } +#endif + #endif //MODE_RENDER_NORMAL //nothing happens, so a tree-ssa optimizer will result in no fragment shader :) @@ -2455,7 +2677,6 @@ FRAGMENT_SHADER_CODE #endif diffuse_buffer = vec4(emission + diffuse_light + ambient_light, sss_strength); specular_buffer = vec4(specular_light, metallic); - #endif #else //MODE_MULTIPLE_RENDER_TARGETS diff --git a/servers/rendering/rasterizer_rd/shaders/scene_high_end_inc.glsl b/servers/rendering/rasterizer_rd/shaders/scene_high_end_inc.glsl index 1cac12406a..1244599097 100644 --- a/servers/rendering/rasterizer_rd/shaders/scene_high_end_inc.glsl +++ b/servers/rendering/rasterizer_rd/shaders/scene_high_end_inc.glsl @@ -1,6 +1,8 @@ #define M_PI 3.14159265359 #define ROUGHNESS_MAX_LOD 5 +#define MAX_GI_PROBES 8 + layout(push_constant, binding = 0, std430) uniform DrawCall { uint instance_index; uint pad; //16 bits minimum size @@ -27,6 +29,8 @@ layout(set = 0, binding = 1) uniform sampler material_samplers[12]; layout(set = 0, binding = 2) uniform sampler shadow_sampler; +#define SDFGI_MAX_CASCADES 8 + layout(set = 0, binding = 3, std140) uniform SceneData { mat4 projection_matrix; mat4 inv_projection_matrix; @@ -76,11 +80,19 @@ layout(set = 0, binding = 3, std140) uniform SceneData { float ssao_ao_affect; bool roughness_limiter_enabled; + float roughness_limiter_amount; + float roughness_limiter_limit; + uvec2 roughness_limiter_pad; + vec4 ao_color; + + mat4 sdf_to_bounds; + + ivec3 sdf_offset; bool material_uv2_mode; - uint pad_material0; - uint pad_material1; - uint pad_material2; + + ivec3 sdf_size; + bool gi_upscale_for_msaa; #if 0 vec4 ambient_light_color; @@ -120,6 +132,8 @@ layout(set = 0, binding = 3, std140) uniform SceneData { scene_data; +#define INSTANCE_FLAGS_USE_GI_BUFFERS (1 << 6) +#define INSTANCE_FLAGS_USE_SDFGI (1 << 7) #define INSTANCE_FLAGS_USE_LIGHTMAP_CAPTURE (1 << 8) #define INSTANCE_FLAGS_USE_LIGHTMAP (1 << 9) #define INSTANCE_FLAGS_USE_SH_LIGHTMAP (1 << 10) @@ -175,13 +189,18 @@ layout(set = 0, binding = 5, std430) restrict readonly buffer Lights { } lights; +#define REFLECTION_AMBIENT_DISABLED 0 +#define REFLECTION_AMBIENT_ENVIRONMENT 1 +#define REFLECTION_AMBIENT_COLOR 2 + struct ReflectionData { vec3 box_extents; float index; vec3 box_offset; uint mask; vec4 params; // intensity, 0, interior , boxproject - vec4 ambient; // ambient color, energy + vec3 ambient; // ambient color + uint ambient_mode; mat4 local_matrix; // up to here for spot and omni, rest is for directional // notes: for ambientblend, use distance to edge to blend between already existing global environment }; @@ -229,29 +248,6 @@ layout(set = 0, binding = 7, std140) uniform DirectionalLights { } directional_lights; -struct GIProbeData { - mat4 xform; - vec3 bounds; - float dynamic_range; - - float bias; - float normal_bias; - bool blend_ambient; - uint texture_slot; - - float anisotropy_strength; - float ambient_occlusion; - float ambient_occlusion_size; - uint pad2; -}; - -layout(set = 0, binding = 8, std140) uniform GIProbes { - GIProbeData data[MAX_GI_PROBES]; -} -gi_probes; - -layout(set = 0, binding = 9) uniform texture3D gi_probe_textures[MAX_GI_PROBE_TEXTURES]; - #define LIGHTMAP_FLAG_USE_DIRECTION 1 #define LIGHTMAP_FLAG_USE_SPECULAR_DIRECTION 2 @@ -319,6 +315,41 @@ layout(set = 0, binding = 19, std430) restrict readonly buffer GlobalVariableDat } global_variables; +struct SDFGIProbeCascadeData { + vec3 position; + float to_probe; + ivec3 probe_world_offset; + float to_cell; // 1/bounds * grid_size +}; + +layout(set = 0, binding = 20, 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; + + SDFGIProbeCascadeData cascades[SDFGI_MAX_CASCADES]; +} +sdfgi; + // decal atlas /* Set 1, Radiance */ @@ -339,13 +370,57 @@ layout(set = 2, binding = 0) uniform textureCubeArray reflection_atlas; layout(set = 2, binding = 1) uniform texture2D shadow_atlas; +layout(set = 2, binding = 2) uniform texture3D gi_probe_textures[MAX_GI_PROBES]; + /* Set 3, Render Buffers */ +#ifdef MODE_RENDER_SDF + +layout(r16ui, set = 3, binding = 0) uniform restrict writeonly uimage3D albedo_volume_grid; +layout(r32ui, set = 3, binding = 1) uniform restrict writeonly uimage3D emission_grid; +layout(r32ui, set = 3, binding = 2) uniform restrict writeonly uimage3D emission_aniso_grid; +layout(r32ui, set = 3, binding = 3) uniform restrict uimage3D geom_facing_grid; + +//still need to be present for shaders that use it, so remap them to something +#define depth_buffer shadow_atlas +#define color_buffer shadow_atlas +#define normal_roughness_buffer shadow_atlas + +#else + layout(set = 3, binding = 0) uniform texture2D depth_buffer; layout(set = 3, binding = 1) uniform texture2D color_buffer; -layout(set = 3, binding = 2) uniform texture2D normal_buffer; -layout(set = 3, binding = 3) uniform texture2D roughness_buffer; +layout(set = 3, binding = 2) uniform texture2D normal_roughness_buffer; layout(set = 3, binding = 4) uniform texture2D ao_buffer; +layout(set = 3, binding = 5) uniform texture2D ambient_buffer; +layout(set = 3, binding = 6) uniform texture2D reflection_buffer; + +layout(set = 3, binding = 7) uniform texture2DArray sdfgi_lightprobe_texture; + +layout(set = 3, binding = 8) uniform texture3D sdfgi_occlusion_cascades; + +struct GIProbeData { + mat4 xform; + vec3 bounds; + float dynamic_range; + + float bias; + float normal_bias; + bool blend_ambient; + uint texture_slot; + + float anisotropy_strength; + float ambient_occlusion; + float ambient_occlusion_size; + uint pad2; +}; + +layout(set = 3, binding = 9, std140) uniform GIProbes { + GIProbeData data[MAX_GI_PROBES]; +} +gi_probes; + +#endif /* Set 4 Skeleton & Instancing (Multimesh) */ diff --git a/servers/rendering/rasterizer_rd/shaders/screen_space_reflection.glsl b/servers/rendering/rasterizer_rd/shaders/screen_space_reflection.glsl index 084f28d932..a8ee33a664 100644 --- a/servers/rendering/rasterizer_rd/shaders/screen_space_reflection.glsl +++ b/servers/rendering/rasterizer_rd/shaders/screen_space_reflection.glsl @@ -12,11 +12,8 @@ layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly image2D ssr_ima #ifdef MODE_ROUGH layout(r8, set = 1, binding = 1) uniform restrict writeonly image2D blur_radius_image; #endif -layout(rgba8, set = 2, binding = 0) uniform restrict readonly image2D source_normal; +layout(rgba8, set = 2, binding = 0) uniform restrict readonly image2D source_normal_roughness; layout(set = 3, binding = 0) uniform sampler2D source_metallic; -#ifdef MODE_ROUGH -layout(set = 3, binding = 1) uniform sampler2D source_roughness; -#endif layout(push_constant, binding = 2, std430) uniform Params { vec4 proj_info; @@ -75,7 +72,8 @@ void main() { // World space point being shaded vec3 vertex = reconstructCSPosition(uv * vec2(params.screen_size), base_depth); - vec3 normal = imageLoad(source_normal, ssC).xyz * 2.0 - 1.0; + vec4 normal_roughness = imageLoad(source_normal_roughness, ssC); + vec3 normal = normal_roughness.xyz * 2.0 - 1.0; normal = normalize(normal); normal.y = -normal.y; //because this code reads flipped @@ -208,7 +206,7 @@ void main() { // if roughness is enabled, do screen space cone tracing float blur_radius = 0.0; - float roughness = texelFetch(source_roughness, ssC << 1, 0).r; + float roughness = normal_roughness.w; if (roughness > 0.001) { float cone_angle = min(roughness, 0.999) * M_PI * 0.5; diff --git a/servers/rendering/rasterizer_rd/shaders/sdfgi_debug.glsl b/servers/rendering/rasterizer_rd/shaders/sdfgi_debug.glsl new file mode 100644 index 0000000000..813ea29fa1 --- /dev/null +++ b/servers/rendering/rasterizer_rd/shaders/sdfgi_debug.glsl @@ -0,0 +1,275 @@ +#[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, binding = 0, std430) uniform Params { + vec3 grid_size; + uint max_cascades; + + ivec2 screen_size; + bool use_occlusion; + float y_mult; + + vec3 cam_extent; + int probe_axis_size; + + mat4 cam_transform; +} +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 = params.cam_extent.xy * ((vec2(screen_pos) / vec2(params.screen_size)) * 2.0 - 1.0); + ray_dir.z = params.cam_extent.z; + + 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 + 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))); + + if (blend > 0.0) { + light = mix(light, hit_light, blend); + blend = 0.0; + } else { + light = hit_light; + + //process blend + float blend_from = (float(params.probe_axis_size - 1) / 2.0) - 2.5; + float blend_to = blend_from + 2.0; + + vec3 cam_pos = params.cam_transform[3].xyz - cascades.data[i].offset; + cam_pos *= cascades.data[i].to_cell; + + pos += ray_dir * min(advance, max_advance); + vec3 inner_pos = pos - cam_pos; + + inner_pos = inner_pos * float(params.probe_axis_size - 1) / params.grid_size.x; + + 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); + + pos /= cascades.data[i].to_cell; + pos += cascades.data[i].offset; + ray_pos = pos; + hit = false; //continue trace for blend + + continue; + } + } + + break; + } + + light = mix(light, vec3(0.0), blend); + +#else + + vec3 inv_dir = 1.0 / ray_dir; + + bool hit = false; + vec4 light_accum = vec4(0.0); + + float blend_size = (params.grid_size.x / float(params.probe_axis_size - 1)) * 0.5; + + float radius_sizes[MAX_CASCADES]; + for (uint i = 0; i < params.max_cascades; i++) { + radius_sizes[i] = (1.0 / cascades.data[i].to_cell) * (params.grid_size.x * 0.5 - blend_size); + } + + float max_distance = radius_sizes[params.max_cascades - 1]; + float advance = 0; + while (advance < max_distance) { + for (uint i = 0; i < params.max_cascades; i++) { + if (advance < radius_sizes[i]) { + vec3 pos = (ray_pos + ray_dir * advance) - cascades.data[i].offset; + pos *= cascades.data[i].to_cell * pos_to_uvw; + + float distance = texture(sampler3D(sdf_cascades[i], linear_sampler), pos).r * 255.0 - 1.0; + + vec4 hit_light = vec4(0.0); + if (distance < 1.0) { + hit_light.a = max(0.0, 1.0 - distance); + hit_light.rgb = texture(sampler3D(light_cascades[i], linear_sampler), pos).rgb; + hit_light.rgb *= hit_light.a; + } + + distance /= cascades.data[i].to_cell; + + if (i < (params.max_cascades - 1)) { + pos = (ray_pos + ray_dir * advance) - cascades.data[i + 1].offset; + pos *= cascades.data[i + 1].to_cell * pos_to_uvw; + + float distance2 = texture(sampler3D(sdf_cascades[i + 1], linear_sampler), pos).r * 255.0 - 1.0; + + vec4 hit_light2 = vec4(0.0); + if (distance2 < 1.0) { + hit_light2.a = max(0.0, 1.0 - distance2); + hit_light2.rgb = texture(sampler3D(light_cascades[i + 1], linear_sampler), pos).rgb; + hit_light2.rgb *= hit_light2.a; + } + + float prev_radius = i == 0 ? 0.0 : radius_sizes[i - 1]; + float blend = (advance - prev_radius) / (radius_sizes[i] - prev_radius); + + distance2 /= cascades.data[i + 1].to_cell; + + hit_light = mix(hit_light, hit_light2, blend); + distance = mix(distance, distance2, blend); + } + + light_accum += hit_light; + advance += distance; + break; + } + } + + if (light_accum.a > 0.98) { + break; + } + } + + light = light_accum.rgb / light_accum.a; + +#endif + + imageStore(screen_buffer, screen_pos, vec4(linear_to_srgb(light), 1.0)); +} diff --git a/servers/rendering/rasterizer_rd/shaders/sdfgi_debug_probes.glsl b/servers/rendering/rasterizer_rd/shaders/sdfgi_debug_probes.glsl new file mode 100644 index 0000000000..08da283dad --- /dev/null +++ b/servers/rendering/rasterizer_rd/shaders/sdfgi_debug_probes.glsl @@ -0,0 +1,231 @@ +#[vertex] + +#version 450 + +VERSION_DEFINES + +#define MAX_CASCADES 8 + +layout(push_constant, binding = 0, std430) uniform Params { + mat4 projection; + + 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; + +// http://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; + +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 = params.projection * 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 = params.projection * vec4(vertex, 1.0); + +#endif +} + +#[fragment] + +#version 450 + +VERSION_DEFINES + +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, binding = 0, std430) uniform Params { + mat4 projection; + + 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/rasterizer_rd/shaders/sdfgi_direct_light.glsl b/servers/rendering/rasterizer_rd/shaders/sdfgi_direct_light.glsl new file mode 100644 index 0000000000..c4b29216d5 --- /dev/null +++ b/servers/rendering/rasterizer_rd/shaders/sdfgi_direct_light.glsl @@ -0,0 +1,472 @@ +#[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 neigbours + uint albedo; //rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neibhbours + uint light; //rgbe8985 encoded total saved light, extra 2 bits for neighbous + uint light_aniso; //55555 light anisotropy, extra 2 bits for neighbours + //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 spot_angle; + float spot_attenuation; + float radius; + + vec4 shadow_color; +}; + +layout(set = 0, binding = 9, std140) buffer restrict readonly Lights { + Light data[]; +} +lights; + +layout(set = 0, binding = 10) uniform texture2DArray lightprobe_texture; + +layout(push_constant, binding = 0, std430) uniform Params { + vec3 grid_size; + uint max_cascades; + + uint cascade; + uint light_count; + uint process_offset; + uint process_increment; + + int probe_axis_size; + bool multibounce; + float y_mult; + uint pad; +} +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; +} + +void main() { + uint voxel_index = uint(gl_GlobalInvocationID.x); + + //used for skipping voxels every N frames + voxel_index = params.process_offset + voxel_index * params.process_increment; + + 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]; + + uint valid_aniso = (voxel_albedo >> 15) & 0x3F; + + { + 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; + } + } + + 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)); + + // 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 = pow(clamp(1.0 - length(rel_vec) / lights.data[i].radius, 0.0, 1.0), 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 = pow(clamp(1.0 - length(rel_vec) / lights.data[i].radius, 0.0, 1.0), lights.data[i].attenuation); + + float angle = acos(dot(normalize(rel_vec), -lights.data[i].direction)); + if (angle > lights.data[i].spot_angle) { + attenuation = 0.0; + } else { + float d = clamp(angle / lights.data[i].spot_angle, 0, 1); + attenuation *= pow(1.0 - d, lights.data[i].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; + } + } + } + } + + // Add indirect light + + if (params.multibounce) { + vec3 pos = (vec3(positioni) + vec3(0.5)) * float(params.probe_axis_size - 1) / params.grid_size; + ivec3 probe_base_pos = ivec3(pos); + + vec4 probe_accum[6] = vec4[](vec4(0.0), vec4(0.0), vec4(0.0), vec4(0.0), vec4(0.0), vec4(0.0)); + 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 = (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(0.005, dot(n, probe_dir)); + + 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; + vec4 indirect_light = textureLod(sampler2DArray(lightprobe_texture, linear_sampler), pos_uvw, 0.0); + + probe_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] += probe_accum[k].rgb * albedo / weight_accum[k]; + } + } + } + + // 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/rasterizer_rd/shaders/sdfgi_fields.glsl b/servers/rendering/rasterizer_rd/shaders/sdfgi_fields.glsl new file mode 100644 index 0000000000..eec0a90c0d --- /dev/null +++ b/servers/rendering/rasterizer_rd/shaders/sdfgi_fields.glsl @@ -0,0 +1,182 @@ +/* clang-format off */ +[compute] + +#version 450 + +VERSION_DEFINES + +layout(local_size_x = OCT_RES, local_size_y = OCT_RES, local_size_z = 1) in; + +/* clang-format on */ + +#define MAX_CASCADES 8 + +layout(rgba16f, set = 0, binding = 1) uniform restrict image2DArray irradiance_texture; +layout(rg16f, set = 0, binding = 2) uniform restrict image2DArray depth_texture; + +ayout(rgba32ui, set = 0, binding = 3) uniform restrict uimage2DArray irradiance_history_texture; +layout(rg32ui, set = 0, binding = 4) uniform restrict uimage2DArray depth_history_texture; + +struct CascadeData { + vec3 offset; //offset of (0,0,0) in world coordinates + float to_cell; // 1/bounds * grid_size +}; + +layout(set = 0, binding = 5, std140) uniform Cascades { + CascadeData data[MAX_CASCADES]; +} +cascades; + +#define DEPTH_HISTORY_BITS 24 +#define IRRADIANCE_HISTORY_BITS 16 + +layout(push_constant, binding = 0, std430) uniform Params { + vec3 grid_size; + uint max_cascades; + + uint probe_axis_size; + uint cascade; + uint history_size; + uint pad0; + + ivec3 scroll; //scroll in probes + uint pad1; +} +params; + +void main() { + ivec2 local = ivec2(gl_LocalInvocationID.xy); + ivec2 probe = ivec2(gl_WorkGroupID.xy); + + ivec3 probe_cell; + probe_cell.x = probe.x % int(params.probe_axis_size); + probe_cell.y = probe.y; + probe_cell.z = probe.x / int(params.probe_axis_size); + +#ifdef MODE_SCROLL_BEGIN + + ivec3 read_cell = probe_cell - params.scroll; + + uint src_layer = (params.history_size + 1) * params.cascade; + uint dst_layer = (params.history_size + 1) * params.max_cascades; + + for (uint i = 0; i <= params.history_size; i++) { + ivec3 write_pos = ivec3(probe * OCT_RES + local, int(i)); + + if (any(lessThan(read_pos, ivec3(0))) || any(greaterThanEqual(read_pos, ivec3(params.probe_axis_size)))) { + // nowhere to read from for scrolling, try finding the value from upper probes + +#ifdef MODE_IRRADIANCE + imageStore(irradiance_history_texture, write_pos, uvec4(0)); +#endif +#ifdef MODE_DEPTH + imageStore(depth_history_texture, write_pos, uvec4(0)); +#endif + } else { + ivec3 read_pos; + read_pos.xy = read_cell.xy; + read_pos.x += read_cell.z * params.probe_axis_size; + read_pos.xy = read_pos.xy * OCT_RES + local; + read_pos.z = int(i); + +#ifdef MODE_IRRADIANCE + uvec4 value = imageLoad(irradiance_history_texture, read_pos); + imageStore(irradiance_history_texture, write_pos, value); +#endif +#ifdef MODE_DEPTH + uvec2 value = imageLoad(depth_history_texture, read_pos); + imageStore(depth_history_texture, write_pos, value); +#endif + } + } + +#endif // MODE_SCROLL_BEGIN + +#ifdef MODE_SCROLL_END + + uint src_layer = (params.history_size + 1) * params.max_cascades; + uint dst_layer = (params.history_size + 1) * params.cascade; + + for (uint i = 0; i <= params.history_size; i++) { + ivec3 pos = ivec3(probe * OCT_RES + local, int(i)); + +#ifdef MODE_IRRADIANCE + uvec4 value = imageLoad(irradiance_history_texture, read_pos); + imageStore(irradiance_history_texture, write_pos, value); +#endif +#ifdef MODE_DEPTH + uvec2 value = imageLoad(depth_history_texture, read_pos); + imageStore(depth_history_texture, write_pos, value); +#endif + } + +#endif //MODE_SCROLL_END + +#ifdef MODE_STORE + + uint src_layer = (params.history_size + 1) * params.cascade + params.history_size; + ivec3 read_pos = ivec3(probe * OCT_RES + local, int(src_layer)); + + ivec3 write_pos = ivec3(probe * (OCT_RES + 2) + ivec2(1), int(params.cascade)); + + ivec3 copy_to[4] = ivec3[](write_pos, ivec3(-2, -2, -2), ivec3(-2, -2, -2), ivec3(-2, -2, -2)); + +#ifdef MODE_IRRADIANCE + uvec4 average = imageLoad(irradiance_history_texture, read_pos); + vec4 light_accum = vec4(average / params.history_size) / float(1 << IRRADIANCE_HISTORY_BITS); + +#endif +#ifdef MODE_DEPTH + uvec2 value = imageLoad(depth_history_texture, read_pos); + vec2 depth_accum = vec4(average / params.history_size) / float(1 << IRRADIANCE_HISTORY_BITS); + + float probe_cell_size = float(params.grid_size / float(params.probe_axis_size - 1)) / cascades.data[params.cascade].to_cell; + float max_depth = length(params.grid_size / cascades.data[params.max_cascades - 1].to_cell); + max_depth /= probe_cell_size; + + depth_value = (vec2(average / params.history_size) / float(1 << DEPTH_HISTORY_BITS)) * vec2(max_depth, max_depth * max_depth); + +#endif + + /* Fill the border if required */ + + if (local == ivec2(0, 0)) { + copy_to[1] = texture_pos + ivec3(OCT_RES - 1, -1, 0); + copy_to[2] = texture_pos + ivec3(-1, OCT_RES - 1, 0); + copy_to[3] = texture_pos + ivec3(OCT_RES, OCT_RES, 0); + } else if (local == ivec2(OCT_RES - 1, 0)) { + copy_to[1] = texture_pos + ivec3(0, -1, 0); + copy_to[2] = texture_pos + ivec3(OCT_RES, OCT_RES - 1, 0); + copy_to[3] = texture_pos + ivec3(-1, OCT_RES, 0); + } else if (local == ivec2(0, OCT_RES - 1)) { + copy_to[1] = texture_pos + ivec3(-1, 0, 0); + copy_to[2] = texture_pos + ivec3(OCT_RES - 1, OCT_RES, 0); + copy_to[3] = texture_pos + ivec3(OCT_RES, -1, 0); + } else if (local == ivec2(OCT_RES - 1, OCT_RES - 1)) { + copy_to[1] = texture_pos + ivec3(0, OCT_RES, 0); + copy_to[2] = texture_pos + ivec3(OCT_RES, 0, 0); + copy_to[3] = texture_pos + ivec3(-1, -1, 0); + } else if (local.y == 0) { + copy_to[1] = texture_pos + ivec3(OCT_RES - local.x - 1, local.y - 1, 0); + } else if (local.x == 0) { + copy_to[1] = texture_pos + ivec3(local.x - 1, OCT_RES - local.y - 1, 0); + } else if (local.y == OCT_RES - 1) { + copy_to[1] = texture_pos + ivec3(OCT_RES - local.x - 1, local.y + 1, 0); + } else if (local.x == OCT_RES - 1) { + copy_to[1] = texture_pos + ivec3(local.x + 1, OCT_RES - local.y - 1, 0); + } + + for (int i = 0; i < 4; i++) { + if (copy_to[i] == ivec3(-2, -2, -2)) { + continue; + } +#ifdef MODE_IRRADIANCE + imageStore(irradiance_texture, copy_to[i], light_accum); +#endif +#ifdef MODE_DEPTH + imageStore(depth_texture, copy_to[i], vec4(depth_value, 0.0, 0.0)); +#endif + } + +#endif // MODE_STORE +} diff --git a/servers/rendering/rasterizer_rd/shaders/sdfgi_integrate.glsl b/servers/rendering/rasterizer_rd/shaders/sdfgi_integrate.glsl new file mode 100644 index 0000000000..e4779aafaf --- /dev/null +++ b/servers/rendering/rasterizer_rd/shaders/sdfgi_integrate.glsl @@ -0,0 +1,605 @@ +#[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(set = 1, binding = 0) uniform textureCube sky_irradiance; + +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, binding = 0, 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; +} +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); +} + +void main() { + ivec2 pos = ivec2(gl_GlobalInvocationID.xy); + if (any(greaterThanEqual(pos, params.image_size))) { //too large, do nothing + return; + } + +#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; + + vec4 probe_sh_accum[SH_SIZE] = vec4[]( + vec4(0.0), + vec4(0.0), + vec4(0.0), + vec4(0.0), + vec4(0.0), + vec4(0.0), + vec4(0.0), + vec4(0.0), + vec4(0.0) +#if (SH_SIZE == 16) + , + vec4(0.0), + vec4(0.0), + vec4(0.0), + vec4(0.0), + vec4(0.0), + vec4(0.0), + vec4(0.0) +#endif + ); + + // 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; + vec3 hit_normal; + vec3 hit_light; + vec3 hit_aniso0; + vec3 hit_aniso1; + + 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; + + 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; + + vec3 uvw; + + 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 < 0.001) { + //consider hit + hit = true; + break; + } + + advance += distance; + } + + if (hit) { + const float EPSILON = 0.001; + hit_normal = normalize(vec3( + texture(sampler3D(sdf_cascades[j], linear_sampler), uvw + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_cascades[j], linear_sampler), uvw - vec3(EPSILON, 0.0, 0.0)).r, + texture(sampler3D(sdf_cascades[j], linear_sampler), uvw + vec3(0.0, EPSILON, 0.0)).r - texture(sampler3D(sdf_cascades[j], linear_sampler), uvw - vec3(0.0, EPSILON, 0.0)).r, + texture(sampler3D(sdf_cascades[j], linear_sampler), uvw + vec3(0.0, 0.0, EPSILON)).r - texture(sampler3D(sdf_cascades[j], linear_sampler), uvw - vec3(0.0, 0.0, EPSILON)).r)); + + hit_light = texture(sampler3D(light_cascades[j], linear_sampler), uvw).rgb; + vec4 aniso0 = texture(sampler3D(aniso0_cascades[j], linear_sampler), uvw); + hit_aniso0 = aniso0.rgb; + hit_aniso1 = vec3(aniso0.a, texture(sampler3D(aniso1_cascades[j], linear_sampler), uvw).rg); + + 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) { + //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) { + light.rgb = textureLod(samplerCube(sky_irradiance, linear_sampler_mipmaps), ray_dir, 2.0).rgb; //use second mipmap because we dont usually throw a lot of rays, so this compensates + 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; + float c[SH_SIZE] = float[]( + + 0.282095, //l0 + 0.488603 * ray_dir.y, //l1n1 + 0.488603 * ray_dir.z, //l1n0 + 0.488603 * ray_dir.x, //l1p1 + 1.092548 * ray_dir.x * ray_dir.y, //l2n2 + 1.092548 * ray_dir.y * ray_dir.z, //l2n1 + 0.315392 * (3.0 * ray_dir2.z - 1.0), //l20 + 1.092548 * ray_dir.x * ray_dir.z, //l2p1 + 0.546274 * (ray_dir2.x - ray_dir2.y) //l2p2 +#if (SH_SIZE == 16) + , + 0.590043 * ray_dir.y * (3.0f * ray_dir2.x - ray_dir2.y), + 2.890611 * ray_dir.y * ray_dir.x * ray_dir.z, + 0.646360 * ray_dir.y * (-1.0f + 5.0f * ray_dir2.z), + 0.373176 * (5.0f * ray_dir2.z * ray_dir.z - 3.0f * ray_dir.z), + 0.457045 * ray_dir.x * (-1.0f + 5.0f * ray_dir2.z), + 1.445305 * (ray_dir2.x - ray_dir2.y) * ray_dir.z, + 0.590043 * ray_dir.x * (ray_dir2.x - 3.0f * ray_dir2.y) + +#endif + ); + + for (uint j = 0; j < SH_SIZE; j++) { + probe_sh_accum[j] += light * c[j]; + } + } + + 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 = probe_sh_accum[i] * 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); + } +#endif // MODE PROCESS + +#ifdef MODE_STORE + + // converting to octahedral in this step is requiered 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; + + //fill the spherical harmonic + vec4 sh[SH_SIZE]; + + 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); + + sh[i] = (vec4(average) / float(params.history_size)) / float(1 << HISTORY_BITS); + } + + //compute the octahedral normal for this texel + vec3 normal = octahedron_encode(vec2(local_pos) / float(OCT_SIZE)); + /* + // read the spherical harmonic + const float c1 = 0.429043; + const float c2 = 0.511664; + const float c3 = 0.743125; + const float c4 = 0.886227; + const float c5 = 0.247708; + vec4 light = (c1 * sh[8] * (normal.x * normal.x - normal.y * normal.y) + + c3 * sh[6] * normal.z * normal.z + + c4 * sh[0] - + c5 * sh[6] + + 2.0 * c1 * sh[4] * normal.x * normal.y + + 2.0 * c1 * sh[7] * normal.x * normal.z + + 2.0 * c1 * sh[5] * normal.y * normal.z + + 2.0 * c2 * sh[3] * normal.x + + 2.0 * c2 * sh[1] * normal.y + + 2.0 * c2 * sh[2] * normal.z); +*/ + 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++) { + vec3 m = sh[i].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) { + //cant scroll, must look for position in parent cascade + + //to global coords + float probe_cell_size = float(params.grid_size.x / float(params.probe_axis_size - 1)) / cascades.data[params.cascade].to_cell; + vec3 probe_pos = cascades.data[params.cascade].offset + vec3(probe_cell) * probe_cell_size; + + //to parent local coords + probe_pos -= cascades.data[params.cascade + 1].offset; + probe_pos *= cascades.data[params.cascade + 1].to_cell; + probe_pos = probe_pos * float(params.probe_axis_size - 1) / float(params.grid_size.x); + + 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 { + // clear and let it re-raytrace, only for the last cascade, which happens very un-often + //scroll + for (uint j = 0; j < params.history_size; j++) { + for (int i = 0; i < SH_SIZE; i++) { + // copy from history texture + ivec3 dst_pos = ivec3(pos.x, pos.y * SH_SIZE + i, int(j)); + imageStore(lightprobe_history_scroll_texture, dst_pos, ivec4(0)); + } + } + + for (int i = 0; i < SH_SIZE; i++) { + // copy from average texture + ivec2 dst_pos = ivec2(pos.x, pos.y * SH_SIZE + i); + imageStore(lightprobe_average_scroll_texture, dst_pos, ivec4(0)); + } + } + +#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/rasterizer_rd/shaders/sdfgi_preprocess.glsl b/servers/rendering/rasterizer_rd/shaders/sdfgi_preprocess.glsl new file mode 100644 index 0000000000..d7d19897e3 --- /dev/null +++ b/servers/rendering/rasterizer_rd/shaders/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 neigbours + uint albedo; //rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neibhbours + uint light; //rgbe8985 encoded total saved light, extra 2 bits for neighbous + 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 neigbours + uint albedo; //rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neibhbours + uint light; //rgbe8985 encoded total saved light, extra 2 bits for neighbous + 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, binding = 0, 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, dont 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, efficent 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(p.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 +} |