diff options
Diffstat (limited to 'servers/rendering/renderer_rd/shaders')
29 files changed, 2182 insertions, 893 deletions
diff --git a/servers/rendering/renderer_rd/shaders/canvas.glsl b/servers/rendering/renderer_rd/shaders/canvas.glsl index 2911e8b731..65a621b203 100644 --- a/servers/rendering/renderer_rd/shaders/canvas.glsl +++ b/servers/rendering/renderer_rd/shaders/canvas.glsl @@ -91,7 +91,6 @@ void main() { uint instancing = draw_data.flags & FLAGS_INSTANCING_MASK; #ifdef USE_ATTRIBUTES - if (instancing > 1) { // trails @@ -128,37 +127,37 @@ void main() { vertex = new_vertex; color *= pcolor; - } else #endif // USE_ATTRIBUTES + { + if (instancing == 1) { + uint stride = 2; + { + if (bool(draw_data.flags & FLAGS_INSTANCING_HAS_COLORS)) { + stride += 1; + } + if (bool(draw_data.flags & FLAGS_INSTANCING_HAS_CUSTOM_DATA)) { + stride += 1; + } + } + + uint offset = stride * gl_InstanceIndex; + + mat4 matrix = mat4(transforms.data[offset + 0], transforms.data[offset + 1], vec4(0.0, 0.0, 1.0, 0.0), vec4(0.0, 0.0, 0.0, 1.0)); + offset += 2; - if (instancing == 1) { - uint stride = 2; - { if (bool(draw_data.flags & FLAGS_INSTANCING_HAS_COLORS)) { - stride += 1; + color *= transforms.data[offset]; + offset += 1; } + if (bool(draw_data.flags & FLAGS_INSTANCING_HAS_CUSTOM_DATA)) { - stride += 1; + instance_custom = transforms.data[offset]; } - } - - uint offset = stride * gl_InstanceIndex; - mat4 matrix = mat4(transforms.data[offset + 0], transforms.data[offset + 1], vec4(0.0, 0.0, 1.0, 0.0), vec4(0.0, 0.0, 0.0, 1.0)); - offset += 2; - - if (bool(draw_data.flags & FLAGS_INSTANCING_HAS_COLORS)) { - color *= transforms.data[offset]; - offset += 1; - } - - if (bool(draw_data.flags & FLAGS_INSTANCING_HAS_CUSTOM_DATA)) { - instance_custom = transforms.data[offset]; + matrix = transpose(matrix); + world_matrix = world_matrix * matrix; } - - matrix = transpose(matrix); - world_matrix = world_matrix * matrix; } #if !defined(USE_ATTRIBUTES) && !defined(USE_PRIMITIVE) diff --git a/servers/rendering/renderer_rd/shaders/copy.glsl b/servers/rendering/renderer_rd/shaders/copy.glsl index 4110a95ddb..d4d0ed0f56 100644 --- a/servers/rendering/renderer_rd/shaders/copy.glsl +++ b/servers/rendering/renderer_rd/shaders/copy.glsl @@ -84,6 +84,8 @@ void main() { color += texelFetch(source_color, base_pos + ivec2(1, 0), 0); color += texelFetch(source_color, base_pos + ivec2(1, 1), 0); color /= 4.0; + color = mix(color, vec4(100.0, 100.0, 100.0, 1.0), isinf(color)); + color = mix(color, vec4(100.0, 100.0, 100.0, 1.0), isnan(color)); imageStore(dest_buffer, pos + params.target, color); #endif @@ -132,6 +134,13 @@ void main() { local_cache[dest_index + 16] = textureLod(source_color, quad_center_uv + vec2(0.0, 1.0 / params.section.w), 0); local_cache[dest_index + 16 + 1] = textureLod(source_color, quad_center_uv + vec2(1.0 / params.section.zw), 0); } + if (bool(params.flags & FLAG_GLOW_FIRST_PASS)) { + // Tonemap initial samples to reduce weight of fireflies: https://graphicrants.blogspot.com/2013/12/tone-mapping.html + local_cache[dest_index] /= 1.0 + dot(local_cache[dest_index].rgb, vec3(0.299, 0.587, 0.114)); + local_cache[dest_index + 1] /= 1.0 + dot(local_cache[dest_index + 1].rgb, vec3(0.299, 0.587, 0.114)); + local_cache[dest_index + 16] /= 1.0 + dot(local_cache[dest_index + 16].rgb, vec3(0.299, 0.587, 0.114)); + local_cache[dest_index + 16 + 1] /= 1.0 + dot(local_cache[dest_index + 16 + 1].rgb, vec3(0.299, 0.587, 0.114)); + } memoryBarrierShared(); barrier(); @@ -177,6 +186,11 @@ void main() { color += temp_cache[index - 2] * 0.140367; color += temp_cache[index - 3] * 0.106595; + if (bool(params.flags & FLAG_GLOW_FIRST_PASS)) { + // Undo tonemap to restore range: https://graphicrants.blogspot.com/2013/12/tone-mapping.html + color /= 1.0 - dot(color.rgb, vec3(0.299, 0.587, 0.114)); + } + color *= params.glow_strength; if (bool(params.flags & FLAG_GLOW_FIRST_PASS)) { @@ -186,7 +200,7 @@ void main() { #endif color *= params.glow_exposure; - float luminance = max(color.r, max(color.g, color.b)); + float luminance = dot(color.rgb, vec3(0.299, 0.587, 0.114)); float feedback = max(smoothstep(params.glow_hdr_threshold, params.glow_hdr_threshold + params.glow_hdr_scale, luminance), params.glow_bloom); color = min(color * feedback, vec4(params.glow_luminance_cap)); @@ -256,7 +270,9 @@ void main() { const float PI = 3.14159265359; vec2 uv = vec2(pos) / vec2(params.section.zw); - uv.y = 1.0 - uv.y; + if (bool(params.flags & FLAG_FLIP_Y)) { + uv.y = 1.0 - uv.y; + } float phi = uv.x * 2.0 * PI; float theta = uv.y * PI; diff --git a/servers/rendering/renderer_rd/shaders/fsr_upscale.glsl b/servers/rendering/renderer_rd/shaders/fsr_upscale.glsl new file mode 100644 index 0000000000..54a7790f77 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/fsr_upscale.glsl @@ -0,0 +1,173 @@ +/*************************************************************************/ +/* fsr_upscale.glsl */ +/*************************************************************************/ +/* This file is part of: */ +/* GODOT ENGINE */ +/* https://godotengine.org */ +/*************************************************************************/ +/* Copyright (c) 2007-2022 Juan Linietsky, Ariel Manzur. */ +/* Copyright (c) 2014-2022 Godot Engine contributors (cf. AUTHORS.md). */ +/* */ +/* Permission is hereby granted, free of charge, to any person obtaining */ +/* a copy of this software and associated documentation files (the */ +/* "Software"), to deal in the Software without restriction, including */ +/* without limitation the rights to use, copy, modify, merge, publish, */ +/* distribute, sublicense, and/or sell copies of the Software, and to */ +/* permit persons to whom the Software is furnished to do so, subject to */ +/* the following conditions: */ +/* */ +/* The above copyright notice and this permission notice shall be */ +/* included in all copies or substantial portions of the Software. */ +/* */ +/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, */ +/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF */ +/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.*/ +/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY */ +/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, */ +/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE */ +/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ +/*************************************************************************/ + +#[compute] + +#version 450 + +#VERSION_DEFINES + +#define A_GPU +#define A_GLSL + +#ifdef MODE_FSR_UPSCALE_NORMAL + +#define A_HALF + +#endif + +#include "thirdparty/amd-fsr/ffx_a.h" + +layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in; + +layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly image2D fsr_image; +layout(set = 0, binding = 0) uniform sampler2D source_image; + +#define FSR_UPSCALE_PASS_TYPE_EASU 0 +#define FSR_UPSCALE_PASS_TYPE_RCAS 1 + +layout(push_constant, binding = 1, std430) uniform Params { + float resolution_width; + float resolution_height; + float upscaled_width; + float upscaled_height; + float sharpness; + int pass; +} +params; + +AU4 Const0, Const1, Const2, Const3; + +#ifdef MODE_FSR_UPSCALE_FALLBACK + +#define FSR_EASU_F +AF4 FsrEasuRF(AF2 p) { + AF4 res = textureGather(source_image, p, 0); + return res; +} +AF4 FsrEasuGF(AF2 p) { + AF4 res = textureGather(source_image, p, 1); + return res; +} +AF4 FsrEasuBF(AF2 p) { + AF4 res = textureGather(source_image, p, 2); + return res; +} + +#define FSR_RCAS_F +AF4 FsrRcasLoadF(ASU2 p) { + return AF4(texelFetch(source_image, ASU2(p), 0)); +} +void FsrRcasInputF(inout AF1 r, inout AF1 g, inout AF1 b) {} + +#else + +#define FSR_EASU_H +AH4 FsrEasuRH(AF2 p) { + AH4 res = AH4(textureGather(source_image, p, 0)); + return res; +} +AH4 FsrEasuGH(AF2 p) { + AH4 res = AH4(textureGather(source_image, p, 1)); + return res; +} +AH4 FsrEasuBH(AF2 p) { + AH4 res = AH4(textureGather(source_image, p, 2)); + return res; +} + +#define FSR_RCAS_H +AH4 FsrRcasLoadH(ASW2 p) { + return AH4(texelFetch(source_image, ASU2(p), 0)); +} +void FsrRcasInputH(inout AH1 r, inout AH1 g, inout AH1 b) {} + +#endif + +#include "thirdparty/amd-fsr/ffx_fsr1.h" + +void fsr_easu_pass(AU2 pos) { +#ifdef MODE_FSR_UPSCALE_NORMAL + + AH3 Gamma2Color = AH3(0, 0, 0); + FsrEasuH(Gamma2Color, pos, Const0, Const1, Const2, Const3); + imageStore(fsr_image, ASU2(pos), AH4(Gamma2Color, 1)); + +#else + + AF3 Gamma2Color = AF3(0, 0, 0); + FsrEasuF(Gamma2Color, pos, Const0, Const1, Const2, Const3); + imageStore(fsr_image, ASU2(pos), AF4(Gamma2Color, 1)); + +#endif +} + +void fsr_rcas_pass(AU2 pos) { +#ifdef MODE_FSR_UPSCALE_NORMAL + + AH3 Gamma2Color = AH3(0, 0, 0); + FsrRcasH(Gamma2Color.r, Gamma2Color.g, Gamma2Color.b, pos, Const0); + imageStore(fsr_image, ASU2(pos), AH4(Gamma2Color, 1)); + +#else + + AF3 Gamma2Color = AF3(0, 0, 0); + FsrRcasF(Gamma2Color.r, Gamma2Color.g, Gamma2Color.b, pos, Const0); + imageStore(fsr_image, ASU2(pos), AF4(Gamma2Color, 1)); + +#endif +} + +void fsr_pass(AU2 pos) { + if (params.pass == FSR_UPSCALE_PASS_TYPE_EASU) { + fsr_easu_pass(pos); + } else if (params.pass == FSR_UPSCALE_PASS_TYPE_RCAS) { + fsr_rcas_pass(pos); + } +} + +void main() { + // Clang does not like unused functions. If ffx_a.h is included in the binary, clang will throw a fit and not compile so we must configure FSR in this shader + if (params.pass == FSR_UPSCALE_PASS_TYPE_EASU) { + FsrEasuCon(Const0, Const1, Const2, Const3, params.resolution_width, params.resolution_height, params.resolution_width, params.resolution_height, params.upscaled_width, params.upscaled_height); + } else if (params.pass == FSR_UPSCALE_PASS_TYPE_RCAS) { + FsrRcasCon(Const0, params.sharpness); + } + + AU2 gxy = ARmp8x8(gl_LocalInvocationID.x) + AU2(gl_WorkGroupID.x << 4u, gl_WorkGroupID.y << 4u); + + fsr_pass(gxy); + gxy.x += 8u; + fsr_pass(gxy); + gxy.y += 8u; + fsr_pass(gxy); + gxy.x -= 8u; + fsr_pass(gxy); +} diff --git a/servers/rendering/renderer_rd/shaders/gi.glsl b/servers/rendering/renderer_rd/shaders/gi.glsl index 60c881881d..9854f124d7 100644 --- a/servers/rendering/renderer_rd/shaders/gi.glsl +++ b/servers/rendering/renderer_rd/shaders/gi.glsl @@ -68,19 +68,15 @@ sdfgi; #define MAX_VOXEL_GI_INSTANCES 8 struct VoxelGIData { - mat4 xform; - vec3 bounds; - float dynamic_range; + mat4 xform; // 64 - 64 - float bias; - float normal_bias; - bool blend_ambient; - uint texture_slot; + vec3 bounds; // 12 - 76 + float dynamic_range; // 4 - 80 - uint pad0; - uint pad1; - uint pad2; - uint mipmaps; + float bias; // 4 - 84 + float normal_bias; // 4 - 88 + bool blend_ambient; // 4 - 92 + uint mipmaps; // 4 - 96 }; layout(set = 0, binding = 16, std140) uniform VoxelGIs { @@ -97,12 +93,10 @@ layout(push_constant, binding = 0, std430) uniform Params { vec4 proj_info; - vec3 ao_color; uint max_voxel_gi_instances; - bool high_quality_vct; bool orthogonal; - uint pad[2]; + uint pad; mat3x4 cam_rotation; } diff --git a/servers/rendering/renderer_rd/shaders/giprobe_write.glsl b/servers/rendering/renderer_rd/shaders/giprobe_write.glsl index 25d87ca45d..a6d65bffeb 100644 --- a/servers/rendering/renderer_rd/shaders/giprobe_write.glsl +++ b/servers/rendering/renderer_rd/shaders/giprobe_write.glsl @@ -7,7 +7,6 @@ layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in; #define NO_CHILDREN 0xFFFFFFFF -#define GREY_VEC vec3(0.33333, 0.33333, 0.33333) struct CellChildren { uint children[8]; diff --git a/servers/rendering/renderer_rd/shaders/light_data_inc.glsl b/servers/rendering/renderer_rd/shaders/light_data_inc.glsl index fdc7729338..52787bb204 100644 --- a/servers/rendering/renderer_rd/shaders/light_data_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/light_data_inc.glsl @@ -1,6 +1,6 @@ #define LIGHT_BAKE_DISABLED 0 -#define LIGHT_BAKE_DYNAMIC 1 -#define LIGHT_BAKE_STATIC 2 +#define LIGHT_BAKE_STATIC 1 +#define LIGHT_BAKE_DYNAMIC 2 struct LightData { //this structure needs to be as packed as possible highp vec3 position; diff --git a/servers/rendering/renderer_rd/shaders/particles.glsl b/servers/rendering/renderer_rd/shaders/particles.glsl index 9f8410fd8a..328becbc20 100644 --- a/servers/rendering/renderer_rd/shaders/particles.glsl +++ b/servers/rendering/renderer_rd/shaders/particles.glsl @@ -567,11 +567,11 @@ void main() { depth = particle_size - s; const float EPSILON = 0.001; normal = mat3(FRAME.colliders[i].transform) * - normalize( - vec3( - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos - vec3(EPSILON, 0.0, 0.0)).r, - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos + vec3(0.0, EPSILON, 0.0)).r - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos - vec3(0.0, EPSILON, 0.0)).r, - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos + vec3(0.0, 0.0, EPSILON)).r - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos - vec3(0.0, 0.0, EPSILON)).r)); + normalize( + vec3( + texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos + vec3(EPSILON, 0.0, 0.0)).r - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos - vec3(EPSILON, 0.0, 0.0)).r, + texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos + vec3(0.0, EPSILON, 0.0)).r - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos - vec3(0.0, EPSILON, 0.0)).r, + texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos + vec3(0.0, 0.0, EPSILON)).r - texture(sampler3D(sdf_vec_textures[FRAME.colliders[i].texture_index], material_samplers[SAMPLER_LINEAR_CLAMP]), uvw_pos - vec3(0.0, 0.0, EPSILON)).r)); } } break; diff --git a/servers/rendering/renderer_rd/shaders/scene_forward_aa_inc.glsl b/servers/rendering/renderer_rd/shaders/scene_forward_aa_inc.glsl index 99714b4504..97c913d489 100644 --- a/servers/rendering/renderer_rd/shaders/scene_forward_aa_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/scene_forward_aa_inc.glsl @@ -2,7 +2,7 @@ float hash_2d(vec2 p) { return fract(1.0e4 * sin(17.0 * p.x + 0.1 * p.y) * - (0.1 + abs(sin(13.0 * p.y + p.x)))); + (0.1 + abs(sin(13.0 * p.y + p.x)))); } float hash_3d(vec3 p) { @@ -29,8 +29,7 @@ float compute_alpha_hash_threshold(vec3 pos, float hash_scale) { vec3 cases = vec3(a_interp * a_interp / (2.0 * min_lerp * (1.0 - min_lerp)), (a_interp - 0.5 * min_lerp) / (1.0 - min_lerp), - 1.0 - ((1.0 - a_interp) * (1.0 - a_interp) / - (2.0 * min_lerp * (1.0 - min_lerp)))); + 1.0 - ((1.0 - a_interp) * (1.0 - a_interp) / (2.0 * min_lerp * (1.0 - min_lerp)))); float alpha_hash_threshold = (lerp_factor < (1.0 - min_lerp)) ? ((lerp_factor < min_lerp) ? cases.x : cases.y) : cases.z; diff --git a/servers/rendering/renderer_rd/shaders/scene_forward_clustered.glsl b/servers/rendering/renderer_rd/shaders/scene_forward_clustered.glsl index 8cb56fbc83..97f7e0a6e6 100644 --- a/servers/rendering/renderer_rd/shaders/scene_forward_clustered.glsl +++ b/servers/rendering/renderer_rd/shaders/scene_forward_clustered.glsl @@ -6,6 +6,8 @@ #include "scene_forward_clustered_inc.glsl" +#define SHADER_IS_SRGB false + /* INPUT ATTRIBS */ layout(location = 0) in vec3 vertex_attrib; @@ -95,7 +97,7 @@ layout(location = 8) out float dp_clip; #endif -layout(location = 9) out flat uint instance_index; +layout(location = 9) out flat uint instance_index_interp; invariant gl_Position; @@ -107,13 +109,15 @@ void main() { color_interp = color_attrib; #endif - instance_index = draw_call.instance_index; + uint instance_index = draw_call.instance_index; bool is_multimesh = bool(instances.data[instance_index].flags & INSTANCE_FLAGS_MULTIMESH); if (!is_multimesh) { instance_index += gl_InstanceIndex; } + instance_index_interp = instance_index; + mat4 world_matrix = instances.data[instance_index].transform; mat3 world_normal_matrix; @@ -247,7 +251,9 @@ void main() { vertex = (world_matrix * vec4(vertex, 1.0)).xyz; +#ifdef NORMAL_USED normal = world_normal_matrix * normal; +#endif #if defined(TANGENT_USED) || defined(NORMAL_MAP_USED) || defined(LIGHT_ANISOTROPY_USED) @@ -286,12 +292,13 @@ void main() { #if !defined(SKIP_TRANSFORM_USED) && defined(VERTEX_WORLD_COORDS_USED) vertex = (scene_data.inv_camera_matrix * vec4(vertex, 1.0)).xyz; - normal = mat3(scene_data.inverse_normal_matrix) * normal; +#ifdef NORMAL_USED + normal = (scene_data.inv_camera_matrix * vec4(normal, 0.0)).xyz; +#endif #if defined(TANGENT_USED) || defined(NORMAL_MAP_USED) || defined(LIGHT_ANISOTROPY_USED) - - binormal = mat3(scene_data.camera_inverse_binormal_matrix) * binormal; - tangent = mat3(scene_data.camera_inverse_tangent_matrix) * tangent; + binormal = (scene_data.inv_camera_matrix * vec4(binormal, 0.0)).xyz; + tangent = (scene_data.inv_camera_matrix * vec4(tangent, 0.0)).xyz; #endif #endif @@ -356,6 +363,8 @@ void main() { #VERSION_DEFINES +#define SHADER_IS_SRGB false + /* Specialization Constants (Toggles) */ layout(constant_id = 0) const bool sc_use_forward_gi = false; @@ -410,7 +419,7 @@ layout(location = 8) in float dp_clip; #endif -layout(location = 9) in flat uint instance_index; +layout(location = 9) in flat uint instance_index_interp; //defines to keep compatibility with vertex @@ -524,14 +533,14 @@ vec4 fog_process(vec3 vertex) { } } - float fog_amount = 1.0 - exp(min(0.0, vertex.z * scene_data.fog_density)); + float fog_amount = 1.0 - exp(min(0.0, -length(vertex) * scene_data.fog_density)); - if (abs(scene_data.fog_height_density) > 0.001) { + if (abs(scene_data.fog_height_density) >= 0.0001) { float y = (scene_data.camera_matrix * vec4(vertex, 1.0)).y; - float y_dist = scene_data.fog_height - y; + float y_dist = y - scene_data.fog_height; - float vfog_amount = clamp(exp(y_dist * scene_data.fog_height_density), 0.0, 1.0); + float vfog_amount = 1.0 - exp(min(0.0, y_dist * scene_data.fog_height_density)); fog_amount = max(vfog_amount, fog_amount); } @@ -564,7 +573,9 @@ void main() { discard; #endif - //lay out everything, whathever is unused is optimized away anyway + uint instance_index = instance_index_interp; + + //lay out everything, whatever is unused is optimized away anyway vec3 vertex = vertex_interp; vec3 view = -normalize(vertex_interp); vec3 albedo = vec3(1.0); @@ -593,7 +604,7 @@ void main() { float ao = 1.0; float ao_light_affect = 0.0; - float alpha = 1.0; + float alpha = float(instances.data[instance_index].flags >> INSTANCE_FLAGS_FADE_SHIFT) / float(255.0); #if defined(TANGENT_USED) || defined(NORMAL_MAP_USED) || defined(LIGHT_ANISOTROPY_USED) vec3 binormal = normalize(binormal_interp); @@ -687,7 +698,7 @@ void main() { #endif // ALPHA_ANTIALIASING_EDGE_USED #ifdef USE_OPAQUE_PREPASS - if (alpha < opaque_prepass_threshold) { + if (alpha < scene_data.opaque_prepass_threshold) { discard; } #endif // USE_OPAQUE_PREPASS @@ -964,15 +975,15 @@ void main() { const float c4 = 0.886227; const float c5 = 0.247708; ambient_light += (c1 * lightmap_captures.data[index].sh[8].rgb * (wnormal.x * wnormal.x - wnormal.y * wnormal.y) + - c3 * lightmap_captures.data[index].sh[6].rgb * wnormal.z * wnormal.z + - c4 * lightmap_captures.data[index].sh[0].rgb - - c5 * lightmap_captures.data[index].sh[6].rgb + - 2.0 * c1 * lightmap_captures.data[index].sh[4].rgb * wnormal.x * wnormal.y + - 2.0 * c1 * lightmap_captures.data[index].sh[7].rgb * wnormal.x * wnormal.z + - 2.0 * c1 * lightmap_captures.data[index].sh[5].rgb * wnormal.y * wnormal.z + - 2.0 * c2 * lightmap_captures.data[index].sh[3].rgb * wnormal.x + - 2.0 * c2 * lightmap_captures.data[index].sh[1].rgb * wnormal.y + - 2.0 * c2 * lightmap_captures.data[index].sh[2].rgb * wnormal.z); + c3 * lightmap_captures.data[index].sh[6].rgb * wnormal.z * wnormal.z + + c4 * lightmap_captures.data[index].sh[0].rgb - + c5 * lightmap_captures.data[index].sh[6].rgb + + 2.0 * c1 * lightmap_captures.data[index].sh[4].rgb * wnormal.x * wnormal.y + + 2.0 * c1 * lightmap_captures.data[index].sh[7].rgb * wnormal.x * wnormal.z + + 2.0 * c1 * lightmap_captures.data[index].sh[5].rgb * wnormal.y * wnormal.z + + 2.0 * c2 * lightmap_captures.data[index].sh[3].rgb * wnormal.x + + 2.0 * c2 * lightmap_captures.data[index].sh[1].rgb * wnormal.y + + 2.0 * c2 * lightmap_captures.data[index].sh[2].rgb * wnormal.z); } else if (bool(instances.data[instance_index].flags & INSTANCE_FLAGS_USE_LIGHTMAP)) { // has actual lightmap bool uses_sh = bool(instances.data[instance_index].flags & INSTANCE_FLAGS_USE_SH_LIGHTMAP); @@ -1145,7 +1156,7 @@ void main() { } #endif // !USE_LIGHTMAP - if (scene_data.ssao_enabled) { + if (bool(scene_data.ss_effects_flags & SCREEN_SPACE_EFFECTS_FLAGS_USE_SSAO)) { float ssao = texture(sampler2D(ao_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), screen_uv).r; ao = min(ao, ssao); ao_light_affect = mix(ao_light_affect, max(ao_light_affect, scene_data.ssao_light_affect), scene_data.ssao_ao_affect); @@ -1215,6 +1226,12 @@ void main() { // convert ao to direct light ao ao = mix(1.0, ao, ao_light_affect); + if (bool(scene_data.ss_effects_flags & SCREEN_SPACE_EFFECTS_FLAGS_USE_SSIL)) { + vec4 ssil = textureLod(sampler2D(ssil_buffer, material_samplers[SAMPLER_LINEAR_CLAMP]), screen_uv, 0.0); + ambient_light *= 1.0 - ssil.a; + ambient_light += ssil.rgb * albedo.rgb; + } + //this saves some VGPRs vec3 f0 = F0(metallic, specular, albedo); @@ -1249,9 +1266,10 @@ void main() { // LIGHTING #if !defined(MODE_RENDER_DEPTH) && !defined(MODE_UNSHADED) - { //directional light + { // Directional light. - // Do shadow and lighting in two passes to reduce register pressure + // Do shadow and lighting in two passes to reduce register pressure. +#ifndef SHADOWS_DISABLED uint shadow0 = 0; uint shadow1 = 0; @@ -1270,21 +1288,21 @@ void main() { float shadow = 1.0; - //version with soft shadows, more expensive if (directional_lights.data[i].shadow_enabled) { - if (sc_use_directional_soft_shadows && directional_lights.data[i].softshadow_angle > 0) { - float depth_z = -vertex.z; - - vec3 shadow_color = vec3(0.0); - vec3 light_dir = directional_lights.data[i].direction; + float depth_z = -vertex.z; + vec3 light_dir = directional_lights.data[i].direction; + vec3 base_normal_bias = normalize(normal_interp) * (1.0 - max(0.0, dot(light_dir, -normalize(normal_interp)))); -#define BIAS_FUNC(m_var, m_idx) \ - m_var.xyz += light_dir * directional_lights.data[i].shadow_bias[m_idx]; \ - vec3 normal_bias = normalize(normal_interp) * (1.0 - max(0.0, dot(light_dir, -normalize(normal_interp)))) * directional_lights.data[i].shadow_normal_bias[m_idx]; \ - normal_bias -= light_dir * dot(light_dir, normal_bias); \ +#define BIAS_FUNC(m_var, m_idx) \ + m_var.xyz += light_dir * directional_lights.data[i].shadow_bias[m_idx]; \ + vec3 normal_bias = base_normal_bias * directional_lights.data[i].shadow_normal_bias[m_idx]; \ + normal_bias -= light_dir * dot(light_dir, normal_bias); \ m_var.xyz += normal_bias; - uint blend_index = 0; + //version with soft shadows, more expensive + if (sc_use_directional_soft_shadows && directional_lights.data[i].softshadow_angle > 0) { + uint blend_count = 0; + const uint blend_max = directional_lights.data[i].blend_splits ? 2 : 1; if (depth_z < directional_lights.data[i].shadow_split_offsets.x) { vec4 v = vec4(vertex, 1.0); @@ -1299,10 +1317,10 @@ void main() { float test_radius = (range_pos - range_begin) * directional_lights.data[i].softshadow_angle; vec2 tex_scale = directional_lights.data[i].uv_scale1 * test_radius; shadow = sample_directional_soft_shadow(directional_shadow_atlas, pssm_coord.xyz, tex_scale * directional_lights.data[i].soft_shadow_scale); - blend_index++; + blend_count++; } - if (blend_index < 2 && depth_z < directional_lights.data[i].shadow_split_offsets.y) { + if (blend_count < blend_max && depth_z < directional_lights.data[i].shadow_split_offsets.y) { vec4 v = vec4(vertex, 1.0); BIAS_FUNC(v, 1) @@ -1316,7 +1334,7 @@ void main() { vec2 tex_scale = directional_lights.data[i].uv_scale2 * test_radius; float s = sample_directional_soft_shadow(directional_shadow_atlas, pssm_coord.xyz, tex_scale * directional_lights.data[i].soft_shadow_scale); - if (blend_index == 0) { + if (blend_count == 0) { shadow = s; } else { //blend @@ -1324,10 +1342,10 @@ void main() { shadow = mix(shadow, s, blend); } - blend_index++; + blend_count++; } - if (blend_index < 2 && depth_z < directional_lights.data[i].shadow_split_offsets.z) { + if (blend_count < blend_max && depth_z < directional_lights.data[i].shadow_split_offsets.z) { vec4 v = vec4(vertex, 1.0); BIAS_FUNC(v, 2) @@ -1341,7 +1359,7 @@ void main() { vec2 tex_scale = directional_lights.data[i].uv_scale3 * test_radius; float s = sample_directional_soft_shadow(directional_shadow_atlas, pssm_coord.xyz, tex_scale * directional_lights.data[i].soft_shadow_scale); - if (blend_index == 0) { + if (blend_count == 0) { shadow = s; } else { //blend @@ -1349,10 +1367,10 @@ void main() { shadow = mix(shadow, s, blend); } - blend_index++; + blend_count++; } - if (blend_index < 2) { + if (blend_count < blend_max) { vec4 v = vec4(vertex, 1.0); BIAS_FUNC(v, 3) @@ -1366,7 +1384,7 @@ void main() { vec2 tex_scale = directional_lights.data[i].uv_scale4 * test_radius; float s = sample_directional_soft_shadow(directional_shadow_atlas, pssm_coord.xyz, tex_scale * directional_lights.data[i].soft_shadow_scale); - if (blend_index == 0) { + if (blend_count == 0) { shadow = s; } else { //blend @@ -1375,21 +1393,9 @@ void main() { } } -#undef BIAS_FUNC } else { //no soft shadows - float depth_z = -vertex.z; - vec4 pssm_coord; - vec3 light_dir = directional_lights.data[i].direction; - vec3 base_normal_bias = normalize(normal_interp) * (1.0 - max(0.0, dot(light_dir, -normalize(normal_interp)))); - -#define BIAS_FUNC(m_var, m_idx) \ - m_var.xyz += light_dir * directional_lights.data[i].shadow_bias[m_idx]; \ - vec3 normal_bias = base_normal_bias * directional_lights.data[i].shadow_normal_bias[m_idx]; \ - normal_bias -= light_dir * dot(light_dir, normal_bias); \ - m_var.xyz += normal_bias; - if (depth_z < directional_lights.data[i].shadow_split_offsets.x) { vec4 v = vec4(vertex, 1.0); @@ -1448,11 +1454,11 @@ void main() { float shadow2 = sample_directional_pcf_shadow(directional_shadow_atlas, scene_data.directional_shadow_pixel_size * directional_lights.data[i].soft_shadow_scale, pssm_coord); shadow = mix(shadow, shadow2, pssm_blend); } + } - shadow = mix(shadow, 1.0, smoothstep(directional_lights.data[i].fade_from, directional_lights.data[i].fade_to, vertex.z)); //done with negative values for performance + shadow = mix(shadow, 1.0, smoothstep(directional_lights.data[i].fade_from, directional_lights.data[i].fade_to, vertex.z)); //done with negative values for performance #undef BIAS_FUNC - } } // shadows if (i < 4) { @@ -1461,6 +1467,7 @@ void main() { shadow1 |= uint(clamp(shadow * 255.0, 0.0, 255.0)) << ((i - 4) * 8); } } +#endif // SHADOWS_DISABLED for (uint i = 0; i < 8; i++) { if (i >= scene_data.directional_light_count) { @@ -1523,18 +1530,19 @@ void main() { #endif float shadow = 1.0; - +#ifndef SHADOWS_DISABLED if (i < 4) { shadow = float(shadow0 >> (i * 8) & 0xFF) / 255.0; } else { shadow = float(shadow1 >> ((i - 4) * 8) & 0xFF) / 255.0; } +#endif blur_shadow(shadow); float size_A = sc_use_light_soft_shadows ? directional_lights.data[i].size : 0.0; - light_compute(normal, directional_lights.data[i].direction, normalize(view), size_A, directional_lights.data[i].color * directional_lights.data[i].energy, shadow, f0, orms, 1.0, + light_compute(normal, directional_lights.data[i].direction, normalize(view), size_A, directional_lights.data[i].color * directional_lights.data[i].energy, shadow, f0, orms, 1.0, albedo, alpha, #ifdef LIGHT_BACKLIGHT_USED backlight, #endif @@ -1545,7 +1553,7 @@ void main() { transmittance_z, #endif #ifdef LIGHT_RIM_USED - rim, rim_tint, albedo, + rim, rim_tint, #endif #ifdef LIGHT_CLEARCOAT_USED clearcoat, clearcoat_gloss, @@ -1553,9 +1561,6 @@ void main() { #ifdef LIGHT_ANISOTROPY_USED binormal, tangent, anisotropy, #endif -#ifdef USE_SHADOW_TO_OPACITY - alpha, -#endif diffuse_light, specular_light); } @@ -1608,7 +1613,7 @@ void main() { shadow = blur_shadow(shadow); - light_process_omni(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, shadow, + light_process_omni(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, shadow, albedo, alpha, #ifdef LIGHT_BACKLIGHT_USED backlight, #endif @@ -1620,7 +1625,6 @@ void main() { #ifdef LIGHT_RIM_USED rim, rim_tint, - albedo, #endif #ifdef LIGHT_CLEARCOAT_USED clearcoat, clearcoat_gloss, @@ -1628,9 +1632,6 @@ void main() { #ifdef LIGHT_ANISOTROPY_USED tangent, binormal, anisotropy, #endif -#ifdef USE_SHADOW_TO_OPACITY - alpha, -#endif diffuse_light, specular_light); } } @@ -1684,7 +1685,7 @@ void main() { shadow = blur_shadow(shadow); - light_process_spot(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, shadow, + light_process_spot(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, shadow, albedo, alpha, #ifdef LIGHT_BACKLIGHT_USED backlight, #endif @@ -1696,7 +1697,6 @@ void main() { #ifdef LIGHT_RIM_USED rim, rim_tint, - albedo, #endif #ifdef LIGHT_CLEARCOAT_USED clearcoat, clearcoat_gloss, @@ -1704,9 +1704,6 @@ void main() { #ifdef LIGHT_ANISOTROPY_USED tangent, binormal, anisotropy, #endif -#ifdef USE_SHADOW_TO_OPACITY - alpha, -#endif diffuse_light, specular_light); } } @@ -1723,7 +1720,7 @@ void main() { #ifdef USE_OPAQUE_PREPASS - if (alpha < opaque_prepass_threshold) { + if (alpha < scene_data.opaque_prepass_threshold) { discard; } @@ -1769,7 +1766,11 @@ void main() { } } +#ifdef MOLTENVK_USED + imageStore(geom_facing_grid, grid_pos, uvec4(imageLoad(geom_facing_grid, grid_pos).r | facing_bits)); //store facing bits +#else imageAtomicOr(geom_facing_grid, grid_pos, facing_bits); //store facing bits +#endif if (length(emission) > 0.001) { float lumas[6]; @@ -1909,7 +1910,6 @@ void main() { // Draw "fixed" fog before volumetric fog to ensure volumetric fog can appear in front of the sky. frag_color.rgb = mix(frag_color.rgb, fog.rgb, fog.a); - ; #endif //MODE_MULTIPLE_RENDER_TARGETS diff --git a/servers/rendering/renderer_rd/shaders/scene_forward_clustered_inc.glsl b/servers/rendering/renderer_rd/shaders/scene_forward_clustered_inc.glsl index b53bf6a6d4..0f7cd18534 100644 --- a/servers/rendering/renderer_rd/shaders/scene_forward_clustered_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/scene_forward_clustered_inc.glsl @@ -68,9 +68,13 @@ layout(set = 0, binding = 4) uniform sampler light_projector_sampler; #define INSTANCE_FLAGS_MULTIMESH_HAS_COLOR (1 << 14) #define INSTANCE_FLAGS_MULTIMESH_HAS_CUSTOM_DATA (1 << 15) #define INSTANCE_FLAGS_PARTICLE_TRAIL_SHIFT 16 +#define INSTANCE_FLAGS_FADE_SHIFT 24 //3 bits of stride #define INSTANCE_FLAGS_PARTICLE_TRAIL_MASK 0xFF +#define SCREEN_SPACE_EFFECTS_FLAGS_USE_SSAO 1 +#define SCREEN_SPACE_EFFECTS_FLAGS_USE_SSIL 2 + layout(set = 0, binding = 5, std430) restrict readonly buffer OmniLights { LightData data[]; } @@ -200,16 +204,15 @@ layout(set = 1, binding = 0, std140) uniform SceneData { float z_far; float z_near; - bool ssao_enabled; + uint ss_effects_flags; float ssao_light_affect; float ssao_ao_affect; bool roughness_limiter_enabled; float roughness_limiter_amount; float roughness_limiter_limit; - uvec2 roughness_limiter_pad; - - vec4 ao_color; + float opaque_prepass_threshold; + uint roughness_limiter_pad; mat4 sdf_to_bounds; @@ -305,19 +308,15 @@ layout(set = 1, binding = 15) uniform texture2DArray sdfgi_lightprobe_texture; layout(set = 1, binding = 16) uniform texture3D sdfgi_occlusion_cascades; struct VoxelGIData { - mat4 xform; - vec3 bounds; - float dynamic_range; + mat4 xform; // 64 - 64 - float bias; - float normal_bias; - bool blend_ambient; - uint texture_slot; + vec3 bounds; // 12 - 76 + float dynamic_range; // 4 - 80 - float anisotropy_strength; - float ambient_occlusion; - float ambient_occlusion_size; - uint mipmaps; + float bias; // 4 - 84 + float normal_bias; // 4 - 88 + bool blend_ambient; // 4 - 92 + uint mipmaps; // 4 - 96 }; layout(set = 1, binding = 17, std140) uniform VoxelGIs { @@ -327,6 +326,8 @@ voxel_gi_instances; layout(set = 1, binding = 18) uniform texture3D volumetric_fog_texture; +layout(set = 1, binding = 19) uniform texture2D ssil_buffer; + #endif /* Set 2 Skeleton & Instancing (can change per item) */ diff --git a/servers/rendering/renderer_rd/shaders/scene_forward_lights_inc.glsl b/servers/rendering/renderer_rd/shaders/scene_forward_lights_inc.glsl index f3db4abe3b..d22f936a35 100644 --- a/servers/rendering/renderer_rd/shaders/scene_forward_lights_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/scene_forward_lights_inc.glsl @@ -73,7 +73,7 @@ vec3 F0(float metallic, float specular, vec3 albedo) { return mix(vec3(dielectric), albedo, vec3(metallic)); } -void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float attenuation, vec3 f0, uint orms, float specular_amount, +void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float attenuation, vec3 f0, uint orms, float specular_amount, vec3 albedo, inout float alpha, #ifdef LIGHT_BACKLIGHT_USED vec3 backlight, #endif @@ -84,7 +84,7 @@ void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float atte float transmittance_z, #endif #ifdef LIGHT_RIM_USED - float rim, float rim_tint, vec3 rim_color, + float rim, float rim_tint, #endif #ifdef LIGHT_CLEARCOAT_USED float clearcoat, float clearcoat_gloss, @@ -92,11 +92,13 @@ void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float atte #ifdef LIGHT_ANISOTROPY_USED vec3 B, vec3 T, float anisotropy, #endif -#ifdef USE_SHADOW_TO_OPACITY - inout float alpha, -#endif inout vec3 diffuse_light, inout vec3 specular_light) { + vec4 orms_unpacked = unpackUnorm4x8(orms); + + float roughness = orms_unpacked.y; + float metallic = orms_unpacked.z; + #if defined(LIGHT_CODE_USED) // light is written by the light shader @@ -125,9 +127,7 @@ void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float atte float cLdotH = clamp(A + dot(L, H), 0.0, 1.0); #endif - float metallic = unpackUnorm4x8(orms).z; if (metallic < 1.0) { - float roughness = unpackUnorm4x8(orms).y; float diffuse_brdf_NL; // BRDF times N.L for calculating diffuse radiance #if defined(DIFFUSE_LAMBERT_WRAP) @@ -168,7 +168,7 @@ void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float atte #if defined(LIGHT_RIM_USED) float rim_light = pow(max(0.0, 1.0 - cNdotV), max(0.0, (1.0 - roughness) * 16.0)); - diffuse_light += rim_light * rim * mix(vec3(1.0), rim_color, rim_tint) * light_color; + diffuse_light += rim_light * rim * mix(vec3(1.0), albedo, rim_tint) * light_color; #endif #ifdef LIGHT_TRANSMITTANCE_USED @@ -179,11 +179,11 @@ void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float atte float d = scale * abs(transmittance_z); float dd = -d * d; vec3 profile = vec3(0.233, 0.455, 0.649) * exp(dd / 0.0064) + - vec3(0.1, 0.336, 0.344) * exp(dd / 0.0484) + - vec3(0.118, 0.198, 0.0) * exp(dd / 0.187) + - vec3(0.113, 0.007, 0.007) * exp(dd / 0.567) + - vec3(0.358, 0.004, 0.0) * exp(dd / 1.99) + - vec3(0.078, 0.0, 0.0) * exp(dd / 7.41); + vec3(0.1, 0.336, 0.344) * exp(dd / 0.0484) + + vec3(0.118, 0.198, 0.0) * exp(dd / 0.187) + + vec3(0.113, 0.007, 0.007) * exp(dd / 0.567) + + vec3(0.358, 0.004, 0.0) * exp(dd / 1.99) + + vec3(0.078, 0.0, 0.0) * exp(dd / 7.41); diffuse_light += profile * transmittance_color.a * light_color * clamp(transmittance_boost - NdotL, 0.0, 1.0) * (1.0 / M_PI); #else @@ -199,7 +199,6 @@ void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float atte #endif //LIGHT_TRANSMITTANCE_USED } - float roughness = unpackUnorm4x8(orms).y; if (roughness > 0.0) { // FIXME: roughness == 0 should not disable specular light entirely // D @@ -211,7 +210,7 @@ void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float atte float blinn = pow(cNdotH, shininess); blinn *= (shininess + 2.0) * (1.0 / (8.0 * M_PI)); - specular_light += light_color * attenuation * specular_amount * blinn * f0 * unpackUnorm4x8(orms).w; + specular_light += light_color * attenuation * specular_amount * blinn * f0 * orms_unpacked.w; #elif defined(SPECULAR_PHONG) @@ -221,7 +220,7 @@ void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float atte float phong = pow(cRdotV, shininess); phong *= (shininess + 1.0) * (1.0 / (8.0 * M_PI)); - specular_light += light_color * attenuation * specular_amount * phong * f0 * unpackUnorm4x8(orms).w; + specular_light += light_color * attenuation * specular_amount * phong * f0 * orms_unpacked.w; #elif defined(SPECULAR_TOON) @@ -285,7 +284,7 @@ void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, float atte #endif //defined(LIGHT_CODE_USED) } -#ifndef USE_NO_SHADOWS +#ifndef SHADOWS_DISABLED // Interleaved Gradient Noise // https://www.iryoku.com/next-generation-post-processing-in-call-of-duty-advanced-warfare @@ -299,7 +298,7 @@ float sample_directional_pcf_shadow(texture2D shadow, vec2 shadow_pixel_size, ve float depth = coord.z; //if only one sample is taken, take it from the center - if (sc_directional_soft_shadow_samples == 1) { + if (sc_directional_soft_shadow_samples == 0) { return textureProj(sampler2DShadow(shadow, shadow_sampler), vec4(pos, depth, 1.0)); } @@ -325,7 +324,7 @@ float sample_pcf_shadow(texture2D shadow, vec2 shadow_pixel_size, vec3 coord) { float depth = coord.z; //if only one sample is taken, take it from the center - if (sc_soft_shadow_samples == 1) { + if (sc_soft_shadow_samples == 0) { return textureProj(sampler2DShadow(shadow, shadow_sampler), vec4(pos, depth, 1.0)); } @@ -348,7 +347,7 @@ float sample_pcf_shadow(texture2D shadow, vec2 shadow_pixel_size, vec3 coord) { float sample_omni_pcf_shadow(texture2D shadow, float blur_scale, vec2 coord, vec4 uv_rect, vec2 flip_offset, float depth) { //if only one sample is taken, take it from the center - if (sc_soft_shadow_samples == 1) { + if (sc_soft_shadow_samples == 0) { vec2 pos = coord * 0.5 + 0.5; pos = uv_rect.xy + pos * uv_rect.zw; return textureProj(sampler2DShadow(shadow, shadow_sampler), vec4(pos, depth, 1.0)); @@ -431,7 +430,7 @@ float sample_directional_soft_shadow(texture2D shadow, vec3 pssm_coord, vec2 tex } } -#endif //USE_NO_SHADOWS +#endif // SHADOWS_DISABLED float get_omni_attenuation(float distance, float inv_range, float decay) { float nd = distance * inv_range; @@ -443,7 +442,7 @@ float get_omni_attenuation(float distance, float inv_range, float decay) { } float light_process_omni_shadow(uint idx, vec3 vertex, vec3 normal) { -#ifndef USE_NO_SHADOWS +#ifndef SHADOWS_DISABLED if (omni_lights.data[idx].shadow_enabled) { // there is a shadowmap vec2 texel_size = scene_data.shadow_atlas_pixel_size; @@ -575,7 +574,7 @@ float light_process_omni_shadow(uint idx, vec3 vertex, vec3 normal) { return 1.0; } -void light_process_omni(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 vertex_ddx, vec3 vertex_ddy, vec3 f0, uint orms, float shadow, +void light_process_omni(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 vertex_ddx, vec3 vertex_ddy, vec3 f0, uint orms, float shadow, vec3 albedo, inout float alpha, #ifdef LIGHT_BACKLIGHT_USED vec3 backlight, #endif @@ -585,7 +584,7 @@ void light_process_omni(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v float transmittance_boost, #endif #ifdef LIGHT_RIM_USED - float rim, float rim_tint, vec3 rim_color, + float rim, float rim_tint, #endif #ifdef LIGHT_CLEARCOAT_USED float clearcoat, float clearcoat_gloss, @@ -593,9 +592,6 @@ void light_process_omni(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v #ifdef LIGHT_ANISOTROPY_USED vec3 binormal, vec3 tangent, float anisotropy, #endif -#ifdef USE_SHADOW_TO_OPACITY - inout float alpha, -#endif inout vec3 diffuse_light, inout vec3 specular_light) { vec3 light_rel_vec = omni_lights.data[idx].position - vertex; float light_length = length(light_rel_vec); @@ -701,7 +697,7 @@ void light_process_omni(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v light_attenuation *= shadow; - light_compute(normal, normalize(light_rel_vec), eye_vec, size_A, color, light_attenuation, f0, orms, omni_lights.data[idx].specular_amount, + light_compute(normal, normalize(light_rel_vec), eye_vec, size_A, color, light_attenuation, f0, orms, omni_lights.data[idx].specular_amount, albedo, alpha, #ifdef LIGHT_BACKLIGHT_USED backlight, #endif @@ -712,7 +708,7 @@ void light_process_omni(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v transmittance_z, #endif #ifdef LIGHT_RIM_USED - rim * omni_attenuation, rim_tint, rim_color, + rim * omni_attenuation, rim_tint, #endif #ifdef LIGHT_CLEARCOAT_USED clearcoat, clearcoat_gloss, @@ -720,15 +716,12 @@ void light_process_omni(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v #ifdef LIGHT_ANISOTROPY_USED binormal, tangent, anisotropy, #endif -#ifdef USE_SHADOW_TO_OPACITY - alpha, -#endif diffuse_light, specular_light); } float light_process_spot_shadow(uint idx, vec3 vertex, vec3 normal) { -#ifndef USE_NO_SHADOWS +#ifndef SHADOWS_DISABLED if (spot_lights.data[idx].shadow_enabled) { vec3 light_rel_vec = spot_lights.data[idx].position - vertex; float light_length = length(light_rel_vec); @@ -804,7 +797,7 @@ float light_process_spot_shadow(uint idx, vec3 vertex, vec3 normal) { return shadow; } -#endif //USE_NO_SHADOWS +#endif // SHADOWS_DISABLED return 1.0; } @@ -821,7 +814,7 @@ vec2 normal_to_panorama(vec3 n) { return panorama_coords; } -void light_process_spot(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 vertex_ddx, vec3 vertex_ddy, vec3 f0, uint orms, float shadow, +void light_process_spot(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 vertex_ddx, vec3 vertex_ddy, vec3 f0, uint orms, float shadow, vec3 albedo, inout float alpha, #ifdef LIGHT_BACKLIGHT_USED vec3 backlight, #endif @@ -831,7 +824,7 @@ void light_process_spot(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v float transmittance_boost, #endif #ifdef LIGHT_RIM_USED - float rim, float rim_tint, vec3 rim_color, + float rim, float rim_tint, #endif #ifdef LIGHT_CLEARCOAT_USED float clearcoat, float clearcoat_gloss, @@ -839,9 +832,6 @@ void light_process_spot(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v #ifdef LIGHT_ANISOTROPY_USED vec3 binormal, vec3 tangent, float anisotropy, #endif -#ifdef USE_SHADOW_TO_OPACITY - inout float alpha, -#endif inout vec3 diffuse_light, inout vec3 specular_light) { vec3 light_rel_vec = spot_lights.data[idx].position - vertex; @@ -908,7 +898,7 @@ void light_process_spot(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v } light_attenuation *= shadow; - light_compute(normal, normalize(light_rel_vec), eye_vec, size_A, color, light_attenuation, f0, orms, spot_lights.data[idx].specular_amount, + light_compute(normal, normalize(light_rel_vec), eye_vec, size_A, color, light_attenuation, f0, orms, spot_lights.data[idx].specular_amount, albedo, alpha, #ifdef LIGHT_BACKLIGHT_USED backlight, #endif @@ -919,7 +909,7 @@ void light_process_spot(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v transmittance_z, #endif #ifdef LIGHT_RIM_USED - rim * spot_attenuation, rim_tint, rim_color, + rim * spot_attenuation, rim_tint, #endif #ifdef LIGHT_CLEARCOAT_USED clearcoat, clearcoat_gloss, @@ -927,9 +917,6 @@ void light_process_spot(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v #ifdef LIGHT_ANISOTROPY_USED binormal, tangent, anisotropy, #endif -#ifdef USE_SHADOW_TO_OPACITY - alpha, -#endif diffuse_light, specular_light); } diff --git a/servers/rendering/renderer_rd/shaders/scene_forward_mobile.glsl b/servers/rendering/renderer_rd/shaders/scene_forward_mobile.glsl index c3c4139450..4d6a3b5864 100644 --- a/servers/rendering/renderer_rd/shaders/scene_forward_mobile.glsl +++ b/servers/rendering/renderer_rd/shaders/scene_forward_mobile.glsl @@ -7,6 +7,8 @@ /* Include our forward mobile UBOs definitions etc. */ #include "scene_forward_mobile_inc.glsl" +#define SHADER_IS_SRGB false + /* INPUT ATTRIBS */ layout(location = 0) in vec3 vertex_attrib; @@ -259,7 +261,9 @@ void main() { vertex = (world_matrix * vec4(vertex, 1.0)).xyz; +#ifdef NORMAL_USED normal = world_normal_matrix * normal; +#endif #if defined(TANGENT_USED) || defined(NORMAL_MAP_USED) || defined(LIGHT_ANISOTROPY_USED) @@ -300,12 +304,13 @@ void main() { #if !defined(SKIP_TRANSFORM_USED) && defined(VERTEX_WORLD_COORDS_USED) vertex = (scene_data.inv_camera_matrix * vec4(vertex, 1.0)).xyz; - normal = mat3(scene_data.inverse_normal_matrix) * normal; +#ifdef NORMAL_USED + normal = (scene_data.inv_camera_matrix * vec4(normal, 0.0)).xyz; +#endif #if defined(TANGENT_USED) || defined(NORMAL_MAP_USED) || defined(LIGHT_ANISOTROPY_USED) - - binormal = mat3(scene_data.camera_inverse_binormal_matrix) * binormal; - tangent = mat3(scene_data.camera_inverse_tangent_matrix) * tangent; + binormal = (scene_data.inv_camera_matrix * vec4(binormal, 0.0)).xyz; + tangent = (scene_data.inv_camera_matrix * vec4(tangent, 0.0)).xyz; #endif #endif @@ -370,6 +375,8 @@ void main() { #VERSION_DEFINES +#define SHADER_IS_SRGB false + /* Specialization Constants */ #if !defined(MODE_RENDER_DEPTH) @@ -550,14 +557,14 @@ vec4 fog_process(vec3 vertex) { } } - float fog_amount = 1.0 - exp(min(0.0, vertex.z * scene_data.fog_density)); + float fog_amount = 1.0 - exp(min(0.0, -length(vertex) * scene_data.fog_density)); - if (abs(scene_data.fog_height_density) > 0.001) { + if (abs(scene_data.fog_height_density) >= 0.0001) { float y = (scene_data.camera_matrix * vec4(vertex, 1.0)).y; - float y_dist = scene_data.fog_height - y; + float y_dist = y - scene_data.fog_height; - float vfog_amount = clamp(exp(y_dist * scene_data.fog_height_density), 0.0, 1.0); + float vfog_amount = 1.0 - exp(min(0.0, y_dist * scene_data.fog_height_density)); fog_amount = max(vfog_amount, fog_amount); } @@ -574,7 +581,7 @@ void main() { discard; #endif - //lay out everything, whathever is unused is optimized away anyway + //lay out everything, whatever is unused is optimized away anyway vec3 vertex = vertex_interp; vec3 view = -normalize(vertex_interp); vec3 albedo = vec3(1.0); @@ -701,7 +708,7 @@ void main() { #endif // ALPHA_ANTIALIASING_EDGE_USED #ifdef USE_OPAQUE_PREPASS - if (alpha < opaque_prepass_threshold) { + if (alpha < scene_data.opaque_prepass_threshold) { discard; } #endif // USE_OPAQUE_PREPASS @@ -930,15 +937,15 @@ void main() { const float c4 = 0.886227; const float c5 = 0.247708; ambient_light += (c1 * lightmap_captures.data[index].sh[8].rgb * (wnormal.x * wnormal.x - wnormal.y * wnormal.y) + - c3 * lightmap_captures.data[index].sh[6].rgb * wnormal.z * wnormal.z + - c4 * lightmap_captures.data[index].sh[0].rgb - - c5 * lightmap_captures.data[index].sh[6].rgb + - 2.0 * c1 * lightmap_captures.data[index].sh[4].rgb * wnormal.x * wnormal.y + - 2.0 * c1 * lightmap_captures.data[index].sh[7].rgb * wnormal.x * wnormal.z + - 2.0 * c1 * lightmap_captures.data[index].sh[5].rgb * wnormal.y * wnormal.z + - 2.0 * c2 * lightmap_captures.data[index].sh[3].rgb * wnormal.x + - 2.0 * c2 * lightmap_captures.data[index].sh[1].rgb * wnormal.y + - 2.0 * c2 * lightmap_captures.data[index].sh[2].rgb * wnormal.z); + c3 * lightmap_captures.data[index].sh[6].rgb * wnormal.z * wnormal.z + + c4 * lightmap_captures.data[index].sh[0].rgb - + c5 * lightmap_captures.data[index].sh[6].rgb + + 2.0 * c1 * lightmap_captures.data[index].sh[4].rgb * wnormal.x * wnormal.y + + 2.0 * c1 * lightmap_captures.data[index].sh[7].rgb * wnormal.x * wnormal.z + + 2.0 * c1 * lightmap_captures.data[index].sh[5].rgb * wnormal.y * wnormal.z + + 2.0 * c2 * lightmap_captures.data[index].sh[3].rgb * wnormal.x + + 2.0 * c2 * lightmap_captures.data[index].sh[1].rgb * wnormal.y + + 2.0 * c2 * lightmap_captures.data[index].sh[2].rgb * wnormal.z); } else if (bool(draw_call.flags & INSTANCE_FLAGS_USE_LIGHTMAP)) { // has actual lightmap bool uses_sh = bool(draw_call.flags & INSTANCE_FLAGS_USE_SH_LIGHTMAP); @@ -1046,7 +1053,7 @@ void main() { #if !defined(MODE_RENDER_DEPTH) && !defined(MODE_UNSHADED) if (!sc_disable_directional_lights) { //directional light - +#ifndef SHADOWS_DISABLED // Do shadow and lighting in two passes to reduce register pressure uint shadow0 = 0; uint shadow1 = 0; @@ -1322,6 +1329,8 @@ void main() { } } +#endif // SHADOWS_DISABLED + for (uint i = 0; i < 8; i++) { if (i >= scene_data.directional_light_count) { break; @@ -1334,16 +1343,16 @@ void main() { // We're not doing light transmittence float shadow = 1.0; - +#ifndef SHADOWS_DISABLED if (i < 4) { shadow = float(shadow0 >> (i * 8) & 0xFF) / 255.0; } else { shadow = float(shadow1 >> ((i - 4) * 8) & 0xFF) / 255.0; } - +#endif blur_shadow(shadow); - light_compute(normal, directional_lights.data[i].direction, normalize(view), 0.0, directional_lights.data[i].color * directional_lights.data[i].energy, shadow, f0, orms, 1.0, + light_compute(normal, directional_lights.data[i].direction, normalize(view), 0.0, directional_lights.data[i].color * directional_lights.data[i].energy, shadow, f0, orms, 1.0, albedo, alpha, #ifdef LIGHT_BACKLIGHT_USED backlight, #endif @@ -1356,7 +1365,7 @@ void main() { #endif */ #ifdef LIGHT_RIM_USED - rim, rim_tint, albedo, + rim, rim_tint, #endif #ifdef LIGHT_CLEARCOAT_USED clearcoat, clearcoat_gloss, @@ -1367,9 +1376,6 @@ void main() { #ifdef USE_SOFT_SHADOW directional_lights.data[i].size, #endif -#ifdef USE_SHADOW_TO_OPACITY - alpha, -#endif diffuse_light, specular_light); } @@ -1393,7 +1399,7 @@ void main() { shadow = blur_shadow(shadow); - light_process_omni(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, shadow, + light_process_omni(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, shadow, albedo, alpha, #ifdef LIGHT_BACKLIGHT_USED backlight, #endif @@ -1407,7 +1413,6 @@ void main() { #ifdef LIGHT_RIM_USED rim, rim_tint, - albedo, #endif #ifdef LIGHT_CLEARCOAT_USED clearcoat, clearcoat_gloss, @@ -1415,9 +1420,6 @@ void main() { #ifdef LIGHT_ANISOTROPY_USED tangent, binormal, anisotropy, #endif -#ifdef USE_SHADOW_TO_OPACITY - alpha, -#endif diffuse_light, specular_light); } } //omni lights @@ -1441,7 +1443,7 @@ void main() { shadow = blur_shadow(shadow); - light_process_spot(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, shadow, + light_process_spot(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, shadow, albedo, alpha, #ifdef LIGHT_BACKLIGHT_USED backlight, #endif @@ -1455,7 +1457,6 @@ void main() { #ifdef LIGHT_RIM_USED rim, rim_tint, - albedo, #endif #ifdef LIGHT_CLEARCOAT_USED clearcoat, clearcoat_gloss, @@ -1463,9 +1464,6 @@ void main() { #ifdef LIGHT_ANISOTROPY_USED tangent, binormal, anisotropy, #endif -#ifdef USE_SHADOW_TO_OPACITY - alpha, -#endif diffuse_light, specular_light); } } //spot lights @@ -1481,7 +1479,7 @@ void main() { #ifdef USE_OPAQUE_PREPASS - if (alpha < opaque_prepass_threshold) { + if (alpha < scene_data.opaque_prepass_threshold) { discard; } diff --git a/servers/rendering/renderer_rd/shaders/scene_forward_mobile_inc.glsl b/servers/rendering/renderer_rd/shaders/scene_forward_mobile_inc.glsl index dd8879acb4..a9a4fce82a 100644 --- a/servers/rendering/renderer_rd/shaders/scene_forward_mobile_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/scene_forward_mobile_inc.glsl @@ -168,9 +168,8 @@ layout(set = 1, binding = 0, std140) uniform SceneData { mediump float roughness_limiter_amount; mediump float roughness_limiter_limit; - uvec2 roughness_limiter_pad; - - mediump vec4 ao_color; + mediump float opaque_prepass_threshold; + uint roughness_limiter_pad; bool fog_enabled; highp float fog_density; diff --git a/servers/rendering/renderer_rd/shaders/screen_space_reflection.glsl b/servers/rendering/renderer_rd/shaders/screen_space_reflection.glsl index 78e0a85341..fde5688301 100644 --- a/servers/rendering/renderer_rd/shaders/screen_space_reflection.glsl +++ b/servers/rendering/renderer_rd/shaders/screen_space_reflection.glsl @@ -223,7 +223,6 @@ void main() { blur_radius = (a * (sqrt(a2 + fh2) - a)) / (4.0f * h); } } - final_color = imageLoad(source_diffuse, ivec2((final_pos - 0.5) * pixel_size)); imageStore(blur_radius_image, ssC, vec4(blur_radius / 255.0)); //stored in r8 diff --git a/servers/rendering/renderer_rd/shaders/sdfgi_preprocess.glsl b/servers/rendering/renderer_rd/shaders/sdfgi_preprocess.glsl index 4d9fa85a74..f6ec249b5e 100644 --- a/servers/rendering/renderer_rd/shaders/sdfgi_preprocess.glsl +++ b/servers/rendering/renderer_rd/shaders/sdfgi_preprocess.glsl @@ -102,7 +102,7 @@ dispatch_data; struct ProcessVoxel { uint position; // xyz 7 bit packed, extra 11 bits for neighbors. - uint albedo; //rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neibhbours + uint albedo; //rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neighbours uint light; //rgbe8985 encoded total saved light, extra 2 bits for neighbours uint light_aniso; //55555 light anisotropy, extra 2 bits for neighbours //total neighbours: 26 @@ -135,7 +135,7 @@ dispatch_data; struct ProcessVoxel { uint position; // xyz 7 bit packed, extra 11 bits for neighbors. - uint albedo; //rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neibhbours + uint albedo; //rgb bits 0-15 albedo, bits 16-21 are normal bits (set if geometry exists toward that side), extra 11 bits for neighbours uint light; //rgbe8985 encoded total saved light, extra 2 bits for neighbours uint light_aniso; //55555 light anisotropy, extra 2 bits for neighbours //total neighbours: 26 diff --git a/servers/rendering/renderer_rd/shaders/ssao_downsample.glsl b/servers/rendering/renderer_rd/shaders/ss_effects_downsample.glsl index ee0db6a6f0..bdabc146d8 100644 --- a/servers/rendering/renderer_rd/shaders/ssao_downsample.glsl +++ b/servers/rendering/renderer_rd/shaders/ss_effects_downsample.glsl @@ -42,6 +42,9 @@ layout(r16f, set = 1, binding = 0) uniform restrict writeonly image2DArray dest_ layout(r16f, set = 2, binding = 0) uniform restrict writeonly image2DArray dest_image1; layout(r16f, set = 2, binding = 1) uniform restrict writeonly image2DArray dest_image2; layout(r16f, set = 2, binding = 2) uniform restrict writeonly image2DArray dest_image3; +#ifdef GENERATE_FULL_MIPS +layout(r16f, set = 2, binding = 3) uniform restrict writeonly image2DArray dest_image4; +#endif #endif vec4 screen_space_to_view_space_depth(vec4 p_depth) { @@ -150,7 +153,27 @@ void prepare_depths_and_mips(vec4 p_samples, uvec2 p_output_coord, uvec2 p_gtid) float avg = mip_smart_average(vec4(sample_00, sample_01, sample_10, sample_11)); imageStore(dest_image3, ivec3(p_output_coord.x, p_output_coord.y, depth_array_index), vec4(avg)); +#ifndef GENERATE_FULL_MIPS + } +#else + depth_buffer[depth_array_index][buffer_coord.x][buffer_coord.y] = avg; } + still_alive = p_gtid.x % 16 == depth_array_offset.x && depth_array_offset.y % 16 == depth_array_offset.y; + + p_output_coord /= 2; + groupMemoryBarrier(); + barrier(); + + if (still_alive) { + float sample_00 = depth_buffer[depth_array_index][buffer_coord.x + 0][buffer_coord.y + 0]; + float sample_01 = depth_buffer[depth_array_index][buffer_coord.x + 0][buffer_coord.y + 8]; + float sample_10 = depth_buffer[depth_array_index][buffer_coord.x + 8][buffer_coord.y + 0]; + float sample_11 = depth_buffer[depth_array_index][buffer_coord.x + 8][buffer_coord.y + 8]; + + float avg = mip_smart_average(vec4(sample_00, sample_01, sample_10, sample_11)); + imageStore(dest_image4, ivec3(p_output_coord.x, p_output_coord.y, depth_array_index), vec4(avg)); + } +#endif } #else #ifndef USE_HALF_BUFFERS diff --git a/servers/rendering/renderer_rd/shaders/ssao.glsl b/servers/rendering/renderer_rd/shaders/ssao.glsl index 6e945edfcd..18cab75c3b 100644 --- a/servers/rendering/renderer_rd/shaders/ssao.glsl +++ b/servers/rendering/renderer_rd/shaders/ssao.glsl @@ -23,8 +23,6 @@ #VERSION_DEFINES -#define SSAO_ADAPTIVE_TAP_BASE_COUNT 5 - #define INTELSSAO_MAIN_DISK_SAMPLE_COUNT (32) const vec4 sample_pattern[INTELSSAO_MAIN_DISK_SAMPLE_COUNT] = { vec4(0.78488064, 0.56661671, 1.500000, -0.126083), vec4(0.26022232, -0.29575172, 1.500000, -1.064030), vec4(0.10459357, 0.08372527, 1.110000, -2.730563), vec4(-0.68286800, 0.04963045, 1.090000, -0.498827), @@ -62,7 +60,6 @@ const int num_taps[5] = { 3, 5, 12, 0, 0 }; #define SSAO_REDUCE_RADIUS_NEAR_SCREEN_BORDER_ENABLE_AT_QUALITY_PRESET (1) #define SSAO_MAX_TAPS 32 -#define SSAO_MAX_REF_TAPS 512 #define SSAO_ADAPTIVE_TAP_BASE_COUNT 5 #define SSAO_ADAPTIVE_TAP_FLEXIBLE_COUNT (SSAO_MAX_TAPS - SSAO_ADAPTIVE_TAP_BASE_COUNT) #define SSAO_DEPTH_MIP_LEVELS 4 diff --git a/servers/rendering/renderer_rd/shaders/ssao_blur.glsl b/servers/rendering/renderer_rd/shaders/ssao_blur.glsl index d9cd2b4e85..b154f5e527 100644 --- a/servers/rendering/renderer_rd/shaders/ssao_blur.glsl +++ b/servers/rendering/renderer_rd/shaders/ssao_blur.glsl @@ -128,19 +128,19 @@ void main() { #ifdef MODE_NON_SMART - vec2 halfPixel = params.half_screen_pixel_size * 0.5f; + vec2 half_pixel = params.half_screen_pixel_size * 0.5; vec2 uv = (vec2(gl_GlobalInvocationID.xy) + vec2(0.5, 0.5)) * params.half_screen_pixel_size; - vec2 centre = textureLod(source_ssao, vec2(uv), 0.0).xy; + vec2 center = textureLod(source_ssao, vec2(uv), 0.0).xy; vec4 vals; - vals.x = textureLod(source_ssao, vec2(uv + vec2(-halfPixel.x * 3, -halfPixel.y)), 0.0).x; - vals.y = textureLod(source_ssao, vec2(uv + vec2(+halfPixel.x, -halfPixel.y * 3)), 0.0).x; - vals.z = textureLod(source_ssao, vec2(uv + vec2(-halfPixel.x, +halfPixel.y * 3)), 0.0).x; - vals.w = textureLod(source_ssao, vec2(uv + vec2(+halfPixel.x * 3, +halfPixel.y)), 0.0).x; + vals.x = textureLod(source_ssao, vec2(uv + vec2(-half_pixel.x * 3, -half_pixel.y)), 0.0).x; + vals.y = textureLod(source_ssao, vec2(uv + vec2(+half_pixel.x, -half_pixel.y * 3)), 0.0).x; + vals.z = textureLod(source_ssao, vec2(uv + vec2(-half_pixel.x, +half_pixel.y * 3)), 0.0).x; + vals.w = textureLod(source_ssao, vec2(uv + vec2(+half_pixel.x * 3, +half_pixel.y)), 0.0).x; - vec2 sampled = vec2(dot(vals, vec4(0.2)) + centre.x * 0.2, centre.y); + vec2 sampled = vec2(dot(vals, vec4(0.2)) + center.x * 0.2, center.y); #else #ifdef MODE_SMART diff --git a/servers/rendering/renderer_rd/shaders/ssao_importance_map.glsl b/servers/rendering/renderer_rd/shaders/ssao_importance_map.glsl index 687fe1e6e2..23eba34d63 100644 --- a/servers/rendering/renderer_rd/shaders/ssao_importance_map.glsl +++ b/servers/rendering/renderer_rd/shaders/ssao_importance_map.glsl @@ -26,7 +26,7 @@ layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; #ifdef GENERATE_MAP -layout(set = 0, binding = 0) uniform sampler2DArray source_ssao; +layout(set = 0, binding = 0) uniform sampler2DArray source_texture; #else layout(set = 0, binding = 0) uniform sampler2D source_importance; #endif @@ -56,11 +56,10 @@ void main() { vec2 base_uv = (vec2(base_position) + vec2(0.5f, 0.5f)) * params.half_screen_pixel_size; - float avg = 0.0; float minV = 1.0; float maxV = 0.0; for (int i = 0; i < 4; i++) { - vec4 vals = textureGather(source_ssao, vec3(base_uv, i)); + vec4 vals = textureGather(source_texture, vec3(base_uv, i)); // apply the same modifications that would have been applied in the main shader vals = params.intensity * vals; @@ -69,8 +68,6 @@ void main() { vals = pow(clamp(vals, 0.0, 1.0), vec4(params.power)); - avg += dot(vec4(vals.x, vals.y, vals.z, vals.w), vec4(1.0 / 16.0, 1.0 / 16.0, 1.0 / 16.0, 1.0 / 16.0)); - maxV = max(maxV, max(max(vals.x, vals.y), max(vals.z, vals.w))); minV = min(minV, min(min(vals.x, vals.y), min(vals.z, vals.w))); } diff --git a/servers/rendering/renderer_rd/shaders/ssil.glsl b/servers/rendering/renderer_rd/shaders/ssil.glsl new file mode 100644 index 0000000000..d1f8f42790 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/ssil.glsl @@ -0,0 +1,444 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// Copyright (c) 2016, Intel Corporation +// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated +// documentation files (the "Software"), to deal in the Software without restriction, including without limitation +// the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to +// permit persons to whom the Software is furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in all copies or substantial portions of +// the Software. +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO +// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, +// TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +// SOFTWARE. +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// File changes (yyyy-mm-dd) +// 2016-09-07: filip.strugar@intel.com: first commit +// 2020-12-05: clayjohn: convert to Vulkan and Godot +// 2021-05-27: clayjohn: convert SSAO to SSIL +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// + +#[compute] + +#version 450 + +#VERSION_DEFINES + +#define SSIL_MAIN_DISK_SAMPLE_COUNT (32) +const vec4 sample_pattern[SSIL_MAIN_DISK_SAMPLE_COUNT] = { + vec4(0.78488064, 0.56661671, 1.500000, -0.126083), vec4(0.26022232, -0.29575172, 1.500000, -1.064030), vec4(0.10459357, 0.08372527, 1.110000, -2.730563), vec4(-0.68286800, 0.04963045, 1.090000, -0.498827), + vec4(-0.13570161, -0.64190155, 1.250000, -0.532765), vec4(-0.26193795, -0.08205118, 0.670000, -1.783245), vec4(-0.61177456, 0.66664219, 0.710000, -0.044234), vec4(0.43675563, 0.25119025, 0.610000, -1.167283), + vec4(0.07884444, 0.86618668, 0.640000, -0.459002), vec4(-0.12790935, -0.29869005, 0.600000, -1.729424), vec4(-0.04031125, 0.02413622, 0.600000, -4.792042), vec4(0.16201244, -0.52851415, 0.790000, -1.067055), + vec4(-0.70991218, 0.47301072, 0.640000, -0.335236), vec4(0.03277707, -0.22349690, 0.600000, -1.982384), vec4(0.68921727, 0.36800742, 0.630000, -0.266718), vec4(0.29251814, 0.37775412, 0.610000, -1.422520), + vec4(-0.12224089, 0.96582592, 0.600000, -0.426142), vec4(0.11071457, -0.16131058, 0.600000, -2.165947), vec4(0.46562141, -0.59747696, 0.600000, -0.189760), vec4(-0.51548797, 0.11804193, 0.600000, -1.246800), + vec4(0.89141309, -0.42090443, 0.600000, 0.028192), vec4(-0.32402530, -0.01591529, 0.600000, -1.543018), vec4(0.60771245, 0.41635221, 0.600000, -0.605411), vec4(0.02379565, -0.08239821, 0.600000, -3.809046), + vec4(0.48951152, -0.23657045, 0.600000, -1.189011), vec4(-0.17611565, -0.81696892, 0.600000, -0.513724), vec4(-0.33930185, -0.20732205, 0.600000, -1.698047), vec4(-0.91974425, 0.05403209, 0.600000, 0.062246), + vec4(-0.15064627, -0.14949332, 0.600000, -1.896062), vec4(0.53180975, -0.35210401, 0.600000, -0.758838), vec4(0.41487166, 0.81442589, 0.600000, -0.505648), vec4(-0.24106961, -0.32721516, 0.600000, -1.665244) +}; + +// these values can be changed (up to SSIL_MAX_TAPS) with no changes required elsewhere; values for 4th and 5th preset are ignored but array needed to avoid compilation errors +// the actual number of texture samples is two times this value (each "tap" has two symmetrical depth texture samples) +const int num_taps[5] = { 3, 5, 12, 0, 0 }; + +#define SSIL_TILT_SAMPLES_ENABLE_AT_QUALITY_PRESET (99) // to disable simply set to 99 or similar +#define SSIL_TILT_SAMPLES_AMOUNT (0.4) +// +#define SSIL_HALOING_REDUCTION_ENABLE_AT_QUALITY_PRESET (1) // to disable simply set to 99 or similar +#define SSIL_HALOING_REDUCTION_AMOUNT (0.8) // values from 0.0 - 1.0, 1.0 means max weighting (will cause artifacts, 0.8 is more reasonable) +// +#define SSIL_DEPTH_MIPS_ENABLE_AT_QUALITY_PRESET (2) +#define SSIL_DEPTH_MIPS_GLOBAL_OFFSET (-4.3) // best noise/quality/performance tradeoff, found empirically +// +// !!warning!! the edge handling is hard-coded to 'disabled' on quality level 0, and enabled above, on the C++ side; while toggling it here will work for +// testing purposes, it will not yield performance gains (or correct results) +#define SSIL_DEPTH_BASED_EDGES_ENABLE_AT_QUALITY_PRESET (1) +// +#define SSIL_REDUCE_RADIUS_NEAR_SCREEN_BORDER_ENABLE_AT_QUALITY_PRESET (1) + +#define SSIL_MAX_TAPS 32 +#define SSIL_ADAPTIVE_TAP_BASE_COUNT 5 +#define SSIL_ADAPTIVE_TAP_FLEXIBLE_COUNT (SSIL_MAX_TAPS - SSIL_ADAPTIVE_TAP_BASE_COUNT) +#define SSIL_DEPTH_MIP_LEVELS 4 + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +layout(set = 0, binding = 0) uniform sampler2DArray source_depth_mipmaps; +layout(rgba8, set = 0, binding = 1) uniform restrict readonly image2D source_normal; +layout(set = 0, binding = 2) uniform Constants { //get into a lower set + vec4 rotation_matrices[20]; +} +constants; + +#ifdef ADAPTIVE +layout(rgba16, set = 1, binding = 0) uniform restrict readonly image2DArray source_ssil; +layout(set = 1, binding = 1) uniform sampler2D source_importance; +layout(set = 1, binding = 2, std430) buffer Counter { + uint sum; +} +counter; +#endif + +layout(rgba16, set = 2, binding = 0) uniform restrict writeonly image2D dest_image; +layout(r8, set = 2, binding = 1) uniform image2D edges_weights_image; + +layout(set = 3, binding = 0) uniform sampler2D last_frame; +layout(set = 3, binding = 1) uniform ProjectionConstants { + mat4 reprojection; +} +projection_constants; + +layout(push_constant, binding = 3, std430) uniform Params { + ivec2 screen_size; + int pass; + int quality; + + vec2 half_screen_pixel_size; + vec2 half_screen_pixel_size_x025; + + vec2 NDC_to_view_mul; + vec2 NDC_to_view_add; + + vec2 pad2; + float z_near; + float z_far; + + float radius; + float intensity; + int size_multiplier; + int pad; + + float fade_out_mul; + float fade_out_add; + float normal_rejection_amount; + float inv_radius_near_limit; + + bool is_orthogonal; + float neg_inv_radius; + float load_counter_avg_div; + float adaptive_sample_limit; + + ivec2 pass_coord_offset; + vec2 pass_uv_offset; +} +params; + +float pack_edges(vec4 p_edgesLRTB) { + p_edgesLRTB = round(clamp(p_edgesLRTB, 0.0, 1.0) * 3.05); + return dot(p_edgesLRTB, vec4(64.0 / 255.0, 16.0 / 255.0, 4.0 / 255.0, 1.0 / 255.0)); +} + +vec3 NDC_to_view_space(vec2 p_pos, float p_viewspace_depth) { + if (params.is_orthogonal) { + return vec3((params.NDC_to_view_mul * p_pos.xy + params.NDC_to_view_add), p_viewspace_depth); + } else { + return vec3((params.NDC_to_view_mul * p_pos.xy + params.NDC_to_view_add) * p_viewspace_depth, p_viewspace_depth); + } +} + +// calculate effect radius and fit our screen sampling pattern inside it +void calculate_radius_parameters(const float p_pix_center_length, const vec2 p_pixel_size_at_center, out float r_lookup_radius, out float r_radius, out float r_fallof_sq) { + r_radius = params.radius; + + // when too close, on-screen sampling disk will grow beyond screen size; limit this to avoid closeup temporal artifacts + const float too_close_limit = clamp(p_pix_center_length * params.inv_radius_near_limit, 0.0, 1.0) * 0.8 + 0.2; + + r_radius *= too_close_limit; + + // 0.85 is to reduce the radius to allow for more samples on a slope to still stay within influence + r_lookup_radius = (0.85 * r_radius) / p_pixel_size_at_center.x; + + // used to calculate falloff (both for AO samples and per-sample weights) + r_fallof_sq = -1.0 / (r_radius * r_radius); +} + +vec4 calculate_edges(const float p_center_z, const float p_left_z, const float p_right_z, const float p_top_z, const float p_bottom_z) { + // slope-sensitive depth-based edge detection + vec4 edgesLRTB = vec4(p_left_z, p_right_z, p_top_z, p_bottom_z) - p_center_z; + vec4 edgesLRTB_slope_adjusted = edgesLRTB + edgesLRTB.yxwz; + edgesLRTB = min(abs(edgesLRTB), abs(edgesLRTB_slope_adjusted)); + return clamp((1.3 - edgesLRTB / (p_center_z * 0.040)), 0.0, 1.0); +} + +vec3 decode_normal(vec3 p_encoded_normal) { + vec3 normal = p_encoded_normal * 2.0 - 1.0; + return normal; +} + +vec3 load_normal(ivec2 p_pos) { + vec3 encoded_normal = imageLoad(source_normal, p_pos).xyz; + encoded_normal.z = 1.0 - encoded_normal.z; + return decode_normal(encoded_normal); +} + +vec3 load_normal(ivec2 p_pos, ivec2 p_offset) { + vec3 encoded_normal = imageLoad(source_normal, p_pos + p_offset).xyz; + encoded_normal.z = 1.0 - encoded_normal.z; + return decode_normal(encoded_normal); +} + +// all vectors in viewspace +float calculate_pixel_obscurance(vec3 p_pixel_normal, vec3 p_hit_delta, float p_fallof_sq) { + float length_sq = dot(p_hit_delta, p_hit_delta); + float NdotD = dot(p_pixel_normal, p_hit_delta) / sqrt(length_sq); + + float falloff_mult = max(0.0, length_sq * p_fallof_sq + 1.0); + + return max(0, NdotD - 0.05) * falloff_mult; +} + +void SSIL_tap_inner(const int p_quality_level, inout vec3 r_color_sum, inout float r_obscurance_sum, inout float r_weight_sum, const vec2 p_sampling_uv, const float p_mip_level, const vec3 p_pix_center_pos, vec3 p_pixel_normal, const float p_fallof_sq, const float p_weight_mod) { + // get depth at sample + float viewspace_sample_z = textureLod(source_depth_mipmaps, vec3(p_sampling_uv, params.pass), p_mip_level).x; + vec3 sample_normal = load_normal(ivec2(p_sampling_uv * vec2(params.screen_size))); + + // convert to viewspace + vec3 hit_pos = NDC_to_view_space(p_sampling_uv.xy, viewspace_sample_z); + vec3 hit_delta = hit_pos - p_pix_center_pos; + + float obscurance = calculate_pixel_obscurance(p_pixel_normal, hit_delta, p_fallof_sq); + float weight = 1.0; + + if (p_quality_level >= SSIL_HALOING_REDUCTION_ENABLE_AT_QUALITY_PRESET) { + float reduct = max(0, -hit_delta.z); + reduct = clamp(reduct * params.neg_inv_radius + 2.0, 0.0, 1.0); + weight = SSIL_HALOING_REDUCTION_AMOUNT * reduct + (1.0 - SSIL_HALOING_REDUCTION_AMOUNT); + } + + // Translate sampling_uv to last screen's coordinates + const vec4 sample_pos = projection_constants.reprojection * vec4(p_sampling_uv * 2.0 - 1.0, (viewspace_sample_z - params.z_near) / (params.z_far - params.z_near) * 2.0 - 1.0, 1.0); + vec2 reprojected_sampling_uv = (sample_pos.xy / sample_pos.w) * 0.5 + 0.5; + + weight *= p_weight_mod; + + r_obscurance_sum += obscurance * weight; + + vec3 sample_color = textureLod(last_frame, reprojected_sampling_uv, 5.0).rgb; + // Reduce impact of fireflies by tonemapping before averaging: http://graphicrants.blogspot.com/2013/12/tone-mapping.html + sample_color /= (1.0 + dot(sample_color, vec3(0.299, 0.587, 0.114))); + r_color_sum += sample_color * obscurance * weight * mix(1.0, smoothstep(0.0, 0.1, -dot(sample_normal, normalize(hit_delta))), params.normal_rejection_amount); + r_weight_sum += weight; +} + +void SSILTap(const int p_quality_level, inout vec3 r_color_sum, inout float r_obscurance_sum, inout float r_weight_sum, const int p_tap_index, const mat2 p_rot_scale, const vec3 p_pix_center_pos, vec3 p_pixel_normal, const vec2 p_normalized_screen_pos, const float p_mip_offset, const float p_fallof_sq, float p_weight_mod, vec2 p_norm_xy, float p_norm_xy_length) { + vec2 sample_offset; + float sample_pow_2_len; + + // patterns + { + vec4 new_sample = sample_pattern[p_tap_index]; + sample_offset = new_sample.xy * p_rot_scale; + sample_pow_2_len = new_sample.w; // precalculated, same as: sample_pow_2_len = log2( length( new_sample.xy ) ); + p_weight_mod *= new_sample.z; + } + + // snap to pixel center (more correct obscurance math, avoids artifacts) + sample_offset = round(sample_offset); + + // calculate MIP based on the sample distance from the centre, similar to as described + // in http://graphics.cs.williams.edu/papers/SAOHPG12/. + float mip_level = (p_quality_level < SSIL_DEPTH_MIPS_ENABLE_AT_QUALITY_PRESET) ? (0) : (sample_pow_2_len + p_mip_offset); + + vec2 sampling_uv = sample_offset * params.half_screen_pixel_size + p_normalized_screen_pos; + + SSIL_tap_inner(p_quality_level, r_color_sum, r_obscurance_sum, r_weight_sum, sampling_uv, mip_level, p_pix_center_pos, p_pixel_normal, p_fallof_sq, p_weight_mod); + + // for the second tap, just use the mirrored offset + vec2 sample_offset_mirrored_uv = -sample_offset; + + // tilt the second set of samples so that the disk is effectively rotated by the normal + // effective at removing one set of artifacts, but too expensive for lower quality settings + if (p_quality_level >= SSIL_TILT_SAMPLES_ENABLE_AT_QUALITY_PRESET) { + float dot_norm = dot(sample_offset_mirrored_uv, p_norm_xy); + sample_offset_mirrored_uv -= dot_norm * p_norm_xy_length * p_norm_xy; + sample_offset_mirrored_uv = round(sample_offset_mirrored_uv); + } + + // snap to pixel center (more correct obscurance math, avoids artifacts) + vec2 sampling_mirrored_uv = sample_offset_mirrored_uv * params.half_screen_pixel_size + p_normalized_screen_pos; + + SSIL_tap_inner(p_quality_level, r_color_sum, r_obscurance_sum, r_weight_sum, sampling_mirrored_uv, mip_level, p_pix_center_pos, p_pixel_normal, p_fallof_sq, p_weight_mod); +} + +void generate_SSIL(out vec3 r_color, out vec4 r_edges, out float r_obscurance, out float r_weight, const vec2 p_pos, int p_quality_level, bool p_adaptive_base) { + vec2 pos_rounded = trunc(p_pos); + uvec2 upos = uvec2(pos_rounded); + + const int number_of_taps = (p_adaptive_base) ? (SSIL_ADAPTIVE_TAP_BASE_COUNT) : (num_taps[p_quality_level]); + float pix_z, pix_left_z, pix_top_z, pix_right_z, pix_bottom_z; + + vec4 valuesUL = textureGather(source_depth_mipmaps, vec3(pos_rounded * params.half_screen_pixel_size, params.pass)); + vec4 valuesBR = textureGather(source_depth_mipmaps, vec3((pos_rounded + vec2(1.0)) * params.half_screen_pixel_size, params.pass)); + + // get this pixel's viewspace depth + pix_z = valuesUL.y; + + // get left right top bottom neighbouring pixels for edge detection (gets compiled out on quality_level == 0) + pix_left_z = valuesUL.x; + pix_top_z = valuesUL.z; + pix_right_z = valuesBR.z; + pix_bottom_z = valuesBR.x; + + vec2 normalized_screen_pos = pos_rounded * params.half_screen_pixel_size + params.half_screen_pixel_size_x025; + vec3 pix_center_pos = NDC_to_view_space(normalized_screen_pos, pix_z); + + // Load this pixel's viewspace normal + uvec2 full_res_coord = upos * 2 * params.size_multiplier + params.pass_coord_offset.xy; + vec3 pixel_normal = load_normal(ivec2(full_res_coord)); + + const vec2 pixel_size_at_center = NDC_to_view_space(normalized_screen_pos.xy + params.half_screen_pixel_size, pix_center_pos.z).xy - pix_center_pos.xy; + + float pixel_lookup_radius; + float fallof_sq; + + // calculate effect radius and fit our screen sampling pattern inside it + float viewspace_radius; + calculate_radius_parameters(length(pix_center_pos), pixel_size_at_center, pixel_lookup_radius, viewspace_radius, fallof_sq); + + // calculate samples rotation/scaling + mat2 rot_scale_matrix; + uint pseudo_random_index; + + { + vec4 rotation_scale; + // reduce effect radius near the screen edges slightly; ideally, one would render a larger depth buffer (5% on each side) instead + if (!p_adaptive_base && (p_quality_level >= SSIL_REDUCE_RADIUS_NEAR_SCREEN_BORDER_ENABLE_AT_QUALITY_PRESET)) { + float near_screen_border = min(min(normalized_screen_pos.x, 1.0 - normalized_screen_pos.x), min(normalized_screen_pos.y, 1.0 - normalized_screen_pos.y)); + near_screen_border = clamp(10.0 * near_screen_border + 0.6, 0.0, 1.0); + pixel_lookup_radius *= near_screen_border; + } + + // load & update pseudo-random rotation matrix + pseudo_random_index = uint(pos_rounded.y * 2 + pos_rounded.x) % 5; + rotation_scale = constants.rotation_matrices[params.pass * 5 + pseudo_random_index]; + rot_scale_matrix = mat2(rotation_scale.x * pixel_lookup_radius, rotation_scale.y * pixel_lookup_radius, rotation_scale.z * pixel_lookup_radius, rotation_scale.w * pixel_lookup_radius); + } + + // the main obscurance & sample weight storage + vec3 color_sum = vec3(0.0); + float obscurance_sum = 0.0; + float weight_sum = 0.0; + + // edge mask for between this and left/right/top/bottom neighbour pixels - not used in quality level 0 so initialize to "no edge" (1 is no edge, 0 is edge) + vec4 edgesLRTB = vec4(1.0, 1.0, 1.0, 1.0); + + // Move center pixel slightly towards camera to avoid imprecision artifacts due to using of 16bit depth buffer; a lot smaller offsets needed when using 32bit floats + pix_center_pos *= 0.9992; + + if (!p_adaptive_base && (p_quality_level >= SSIL_DEPTH_BASED_EDGES_ENABLE_AT_QUALITY_PRESET)) { + edgesLRTB = calculate_edges(pix_z, pix_left_z, pix_right_z, pix_top_z, pix_bottom_z); + } + + const float global_mip_offset = SSIL_DEPTH_MIPS_GLOBAL_OFFSET; + float mip_offset = (p_quality_level < SSIL_DEPTH_MIPS_ENABLE_AT_QUALITY_PRESET) ? (0) : (log2(pixel_lookup_radius) + global_mip_offset); + + // Used to tilt the second set of samples so that the disk is effectively rotated by the normal + // effective at removing one set of artifacts, but too expensive for lower quality settings + vec2 norm_xy = vec2(pixel_normal.x, pixel_normal.y); + float norm_xy_length = length(norm_xy); + norm_xy /= vec2(norm_xy_length, -norm_xy_length); + norm_xy_length *= SSIL_TILT_SAMPLES_AMOUNT; + + // standard, non-adaptive approach + if ((p_quality_level != 3) || p_adaptive_base) { + for (int i = 0; i < number_of_taps; i++) { + SSILTap(p_quality_level, color_sum, obscurance_sum, weight_sum, i, rot_scale_matrix, pix_center_pos, pixel_normal, normalized_screen_pos, mip_offset, fallof_sq, 1.0, norm_xy, norm_xy_length); + } + } +#ifdef ADAPTIVE + else { + // add new ones if needed + vec2 full_res_uv = normalized_screen_pos + params.pass_uv_offset.xy; + float importance = textureLod(source_importance, full_res_uv, 0.0).x; + + //Need to store obscurance from base pass + // load existing base values + vec4 base_values = imageLoad(source_ssil, ivec3(upos, params.pass)); + weight_sum += imageLoad(edges_weights_image, ivec2(upos)).r * float(SSIL_ADAPTIVE_TAP_BASE_COUNT * 4.0); + color_sum += (base_values.rgb) * weight_sum; + obscurance_sum += (base_values.a) * weight_sum; + + // increase importance around edges + float edge_count = dot(1.0 - edgesLRTB, vec4(1.0, 1.0, 1.0, 1.0)); + + float avg_total_importance = float(counter.sum) * params.load_counter_avg_div; + + float importance_limiter = clamp(params.adaptive_sample_limit / avg_total_importance, 0.0, 1.0); + importance *= importance_limiter; + + float additional_sample_count = SSIL_ADAPTIVE_TAP_FLEXIBLE_COUNT * importance; + + const float blend_range = 3.0; + const float blend_range_inv = 1.0 / blend_range; + + additional_sample_count += 0.5; + uint additional_samples = uint(additional_sample_count); + uint additional_samples_to = min(SSIL_MAX_TAPS, additional_samples + SSIL_ADAPTIVE_TAP_BASE_COUNT); + + for (uint i = SSIL_ADAPTIVE_TAP_BASE_COUNT; i < additional_samples_to; i++) { + additional_sample_count -= 1.0f; + float weight_mod = clamp(additional_sample_count * blend_range_inv, 0.0, 1.0); + SSILTap(p_quality_level, color_sum, obscurance_sum, weight_sum, int(i), rot_scale_matrix, pix_center_pos, pixel_normal, normalized_screen_pos, mip_offset, fallof_sq, weight_mod, norm_xy, norm_xy_length); + } + } +#endif + + // Early out for adaptive base + if (p_adaptive_base) { + vec3 color = color_sum / weight_sum; + + r_color = color; + r_edges = vec4(0.0); + r_obscurance = obscurance_sum / weight_sum; + r_weight = weight_sum; + return; + } + + // Calculate weighted average + vec3 color = color_sum / weight_sum; + color /= 1.0 - dot(color, vec3(0.299, 0.587, 0.114)); + + // Calculate fadeout (1 close, gradient, 0 far) + float fade_out = clamp(pix_center_pos.z * params.fade_out_mul + params.fade_out_add, 0.0, 1.0); + + // Reduce the SSIL if we're on the edge to remove artifacts on edges (we don't care for the lower quality one) + if (!p_adaptive_base && (p_quality_level >= SSIL_DEPTH_BASED_EDGES_ENABLE_AT_QUALITY_PRESET)) { + // when there's more than 2 opposite edges, start fading out the occlusion to reduce aliasing artifacts + float edge_fadeout_factor = clamp((1.0 - edgesLRTB.x - edgesLRTB.y) * 0.35, 0.0, 1.0) + clamp((1.0 - edgesLRTB.z - edgesLRTB.w) * 0.35, 0.0, 1.0); + + fade_out *= clamp(1.0 - edge_fadeout_factor, 0.0, 1.0); + } + + color = params.intensity * color; + + color *= fade_out; + + // outputs! + r_color = color; + r_edges = edgesLRTB; // These are used to prevent blurring across edges, 1 means no edge, 0 means edge, 0.5 means half way there, etc. + r_obscurance = clamp((obscurance_sum / weight_sum) * params.intensity, 0.0, 1.0); + r_weight = weight_sum; +} + +void main() { + vec3 out_color; + float out_obscurance; + float out_weight; + vec4 out_edges; + ivec2 ssC = ivec2(gl_GlobalInvocationID.xy); + if (any(greaterThanEqual(ssC, params.screen_size))) { //too large, do nothing + return; + } + + vec2 uv = vec2(gl_GlobalInvocationID) + vec2(0.5); +#ifdef SSIL_BASE + generate_SSIL(out_color, out_edges, out_obscurance, out_weight, uv, params.quality, true); + + imageStore(dest_image, ssC, vec4(out_color, out_obscurance)); + imageStore(edges_weights_image, ssC, vec4(out_weight / (float(SSIL_ADAPTIVE_TAP_BASE_COUNT) * 4.0))); +#else + generate_SSIL(out_color, out_edges, out_obscurance, out_weight, uv, params.quality, false); // pass in quality levels + + imageStore(dest_image, ssC, vec4(out_color, out_obscurance)); + imageStore(edges_weights_image, ssC, vec4(pack_edges(out_edges))); +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/ssil_blur.glsl b/servers/rendering/renderer_rd/shaders/ssil_blur.glsl new file mode 100644 index 0000000000..11861e261f --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/ssil_blur.glsl @@ -0,0 +1,124 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +layout(set = 0, binding = 0) uniform sampler2D source_ssil; + +layout(rgba16, set = 1, binding = 0) uniform restrict writeonly image2D dest_image; + +layout(r8, set = 2, binding = 0) uniform restrict readonly image2D source_edges; + +layout(push_constant, binding = 1, std430) uniform Params { + float edge_sharpness; + float pad; + vec2 half_screen_pixel_size; +} +params; + +vec4 unpack_edges(float p_packed_val) { + uint packed_val = uint(p_packed_val * 255.5); + vec4 edgesLRTB; + edgesLRTB.x = float((packed_val >> 6) & 0x03) / 3.0; + edgesLRTB.y = float((packed_val >> 4) & 0x03) / 3.0; + edgesLRTB.z = float((packed_val >> 2) & 0x03) / 3.0; + edgesLRTB.w = float((packed_val >> 0) & 0x03) / 3.0; + + return clamp(edgesLRTB + params.edge_sharpness, 0.0, 1.0); +} + +void add_sample(vec4 p_ssil_value, float p_edge_value, inout vec4 r_sum, inout float r_sum_weight) { + float weight = p_edge_value; + + r_sum += (weight * p_ssil_value); + r_sum_weight += weight; +} + +#ifdef MODE_WIDE +vec4 sample_blurred_wide(ivec2 p_pos, vec2 p_coord) { + vec4 ssil_value = textureLodOffset(source_ssil, vec2(p_coord), 0.0, ivec2(0, 0)); + vec4 ssil_valueL = textureLodOffset(source_ssil, vec2(p_coord), 0.0, ivec2(-2, 0)); + vec4 ssil_valueT = textureLodOffset(source_ssil, vec2(p_coord), 0.0, ivec2(0, -2)); + vec4 ssil_valueR = textureLodOffset(source_ssil, vec2(p_coord), 0.0, ivec2(2, 0)); + vec4 ssil_valueB = textureLodOffset(source_ssil, vec2(p_coord), 0.0, ivec2(0, 2)); + + vec4 edgesLRTB = unpack_edges(imageLoad(source_edges, p_pos).r); + edgesLRTB.x *= unpack_edges(imageLoad(source_edges, p_pos + ivec2(-2, 0)).r).y; + edgesLRTB.z *= unpack_edges(imageLoad(source_edges, p_pos + ivec2(0, -2)).r).w; + edgesLRTB.y *= unpack_edges(imageLoad(source_edges, p_pos + ivec2(2, 0)).r).x; + edgesLRTB.w *= unpack_edges(imageLoad(source_edges, p_pos + ivec2(0, 2)).r).z; + + float sum_weight = 0.8; + vec4 sum = ssil_value * sum_weight; + + add_sample(ssil_valueL, edgesLRTB.x, sum, sum_weight); + add_sample(ssil_valueR, edgesLRTB.y, sum, sum_weight); + add_sample(ssil_valueT, edgesLRTB.z, sum, sum_weight); + add_sample(ssil_valueB, edgesLRTB.w, sum, sum_weight); + + vec4 ssil_avg = sum / sum_weight; + + ssil_value = ssil_avg; + + return ssil_value; +} +#endif + +#ifdef MODE_SMART +vec4 sample_blurred(ivec2 p_pos, vec2 p_coord) { + vec4 vC = textureLodOffset(source_ssil, vec2(p_coord), 0.0, ivec2(0, 0)); + vec4 vL = textureLodOffset(source_ssil, vec2(p_coord), 0.0, ivec2(-1, 0)); + vec4 vT = textureLodOffset(source_ssil, vec2(p_coord), 0.0, ivec2(0, -1)); + vec4 vR = textureLodOffset(source_ssil, vec2(p_coord), 0.0, ivec2(1, 0)); + vec4 vB = textureLodOffset(source_ssil, vec2(p_coord), 0.0, ivec2(0, 1)); + + float packed_edges = imageLoad(source_edges, p_pos).r; + vec4 edgesLRTB = unpack_edges(packed_edges); + + float sum_weight = 0.5; + vec4 sum = vC * sum_weight; + + add_sample(vL, edgesLRTB.x, sum, sum_weight); + add_sample(vR, edgesLRTB.y, sum, sum_weight); + add_sample(vT, edgesLRTB.z, sum, sum_weight); + add_sample(vB, edgesLRTB.w, sum, sum_weight); + + vec4 ssil_avg = sum / sum_weight; + + vec4 ssil_value = ssil_avg; + + return ssil_value; +} +#endif + +void main() { + // Pixel being shaded + ivec2 ssC = ivec2(gl_GlobalInvocationID.xy); + +#ifdef MODE_NON_SMART + + vec2 half_pixel = params.half_screen_pixel_size * 0.5; + + vec2 uv = (vec2(gl_GlobalInvocationID.xy) + vec2(0.5, 0.5)) * params.half_screen_pixel_size; + + vec4 centre = textureLod(source_ssil, uv, 0.0); + + vec4 value = textureLod(source_ssil, vec2(uv + vec2(-half_pixel.x * 3, -half_pixel.y)), 0.0) * 0.2; + value += textureLod(source_ssil, vec2(uv + vec2(+half_pixel.x, -half_pixel.y * 3)), 0.0) * 0.2; + value += textureLod(source_ssil, vec2(uv + vec2(-half_pixel.x, +half_pixel.y * 3)), 0.0) * 0.2; + value += textureLod(source_ssil, vec2(uv + vec2(+half_pixel.x * 3, +half_pixel.y)), 0.0) * 0.2; + + vec4 sampled = value + centre * 0.2; + +#else +#ifdef MODE_SMART + vec4 sampled = sample_blurred(ssC, (vec2(gl_GlobalInvocationID.xy) + vec2(0.5, 0.5)) * params.half_screen_pixel_size); +#else // MODE_WIDE + vec4 sampled = sample_blurred_wide(ssC, (vec2(gl_GlobalInvocationID.xy) + vec2(0.5, 0.5)) * params.half_screen_pixel_size); +#endif +#endif // MODE_NON_SMART + imageStore(dest_image, ssC, sampled); +} diff --git a/servers/rendering/renderer_rd/shaders/ssil_importance_map.glsl b/servers/rendering/renderer_rd/shaders/ssil_importance_map.glsl new file mode 100644 index 0000000000..815aa55fd4 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/ssil_importance_map.glsl @@ -0,0 +1,124 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// Copyright (c) 2016, Intel Corporation +// Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated +// documentation files (the "Software"), to deal in the Software without restriction, including without limitation +// the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to +// permit persons to whom the Software is furnished to do so, subject to the following conditions: +// The above copyright notice and this permission notice shall be included in all copies or substantial portions of +// the Software. +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO +// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, +// TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE +// SOFTWARE. +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// File changes (yyyy-mm-dd) +// 2016-09-07: filip.strugar@intel.com: first commit +// 2020-12-05: clayjohn: convert to Vulkan and Godot +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// + +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +#ifdef GENERATE_MAP +layout(set = 0, binding = 0) uniform sampler2DArray source_texture; +#else +layout(set = 0, binding = 0) uniform sampler2D source_importance; +#endif +layout(r8, set = 1, binding = 0) uniform restrict writeonly image2D dest_image; + +#ifdef PROCESS_MAPB +layout(set = 2, binding = 0, std430) buffer Counter { + uint sum; +} +counter; +#endif + +layout(push_constant, binding = 1, std430) uniform Params { + vec2 half_screen_pixel_size; + float intensity; + float pad; +} +params; + +void main() { + // Pixel being shaded + ivec2 ssC = ivec2(gl_GlobalInvocationID.xy); + +#ifdef GENERATE_MAP + // importance map stuff + uvec2 base_position = ssC * 2; + + float avg = 0.0; + float minV = 1.0; + float maxV = 0.0; + for (int i = 0; i < 4; i++) { + vec3 value_a = texelFetch(source_texture, ivec3(base_position, i), 0).rgb * params.intensity; + vec3 value_b = texelFetch(source_texture, ivec3(base_position, i) + ivec3(0, 1, 0), 0).rgb * params.intensity; + vec3 value_c = texelFetch(source_texture, ivec3(base_position, i) + ivec3(1, 0, 0), 0).rgb * params.intensity; + vec3 value_d = texelFetch(source_texture, ivec3(base_position, i) + ivec3(1, 1, 0), 0).rgb * params.intensity; + + // Calculate luminance (black and white value) + float a = dot(value_a, vec3(0.2125, 0.7154, 0.0721)); + float b = dot(value_b, vec3(0.2125, 0.7154, 0.0721)); + float c = dot(value_c, vec3(0.2125, 0.7154, 0.0721)); + float d = dot(value_d, vec3(0.2125, 0.7154, 0.0721)); + + maxV = max(maxV, max(max(a, b), max(c, d))); + minV = min(minV, min(min(a, b), min(c, d))); + } + + float min_max_diff = maxV - minV; + + imageStore(dest_image, ssC, vec4(pow(clamp(min_max_diff * 2.0, 0.0, 1.0), 0.6))); +#endif + +#ifdef PROCESS_MAPA + vec2 uv = (vec2(ssC) + 0.5) * params.half_screen_pixel_size * 2.0; + + float centre = textureLod(source_importance, uv, 0.0).x; + + vec2 half_pixel = params.half_screen_pixel_size; + + vec4 vals; + vals.x = textureLod(source_importance, uv + vec2(-half_pixel.x * 3, -half_pixel.y), 0.0).x; + vals.y = textureLod(source_importance, uv + vec2(+half_pixel.x, -half_pixel.y * 3), 0.0).x; + vals.z = textureLod(source_importance, uv + vec2(+half_pixel.x * 3, +half_pixel.y), 0.0).x; + vals.w = textureLod(source_importance, uv + vec2(-half_pixel.x, +half_pixel.y * 3), 0.0).x; + + float avg = dot(vals, vec4(0.25, 0.25, 0.25, 0.25)); + + imageStore(dest_image, ssC, vec4(avg)); +#endif + +#ifdef PROCESS_MAPB + vec2 uv = (vec2(ssC) + 0.5f) * params.half_screen_pixel_size * 2.0; + + float centre = textureLod(source_importance, uv, 0.0).x; + + vec2 half_pixel = params.half_screen_pixel_size; + + vec4 vals; + vals.x = textureLod(source_importance, uv + vec2(-half_pixel.x, -half_pixel.y * 3), 0.0).x; + vals.y = textureLod(source_importance, uv + vec2(+half_pixel.x * 3, -half_pixel.y), 0.0).x; + vals.z = textureLod(source_importance, uv + vec2(+half_pixel.x, +half_pixel.y * 3), 0.0).x; + vals.w = textureLod(source_importance, uv + vec2(-half_pixel.x * 3, +half_pixel.y), 0.0).x; + + float avg = dot(vals, vec4(0.25, 0.25, 0.25, 0.25)); + + imageStore(dest_image, ssC, vec4(avg)); + + // sum the average; to avoid overflowing we assume max AO resolution is not bigger than 16384x16384; so quarter res (used here) will be 4096x4096, which leaves us with 8 bits per pixel + uint sum = uint(clamp(avg, 0.0, 1.0) * 255.0 + 0.5); + + // save every 9th to avoid InterlockedAdd congestion - since we're blurring, this is good enough; compensated by multiplying load_counter_avg_div by 9 + if (((ssC.x % 3) + (ssC.y % 3)) == 0) { + atomicAdd(counter.sum, sum); + } +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/ssil_interleave.glsl b/servers/rendering/renderer_rd/shaders/ssil_interleave.glsl new file mode 100644 index 0000000000..8a7a5ae4fd --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/ssil_interleave.glsl @@ -0,0 +1,102 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +layout(rgba16, set = 0, binding = 0) uniform restrict writeonly image2D dest_image; +layout(set = 1, binding = 0) uniform sampler2DArray source_texture; +layout(r8, set = 2, binding = 0) uniform restrict readonly image2DArray source_edges; + +layout(push_constant, binding = 1, std430) uniform Params { + float inv_sharpness; + uint size_modifier; + vec2 pixel_size; +} +params; + +vec4 unpack_edges(float p_packed_val) { + uint packed_val = uint(p_packed_val * 255.5); + vec4 edgesLRTB; + edgesLRTB.x = float((packed_val >> 6) & 0x03) / 3.0; + edgesLRTB.y = float((packed_val >> 4) & 0x03) / 3.0; + edgesLRTB.z = float((packed_val >> 2) & 0x03) / 3.0; + edgesLRTB.w = float((packed_val >> 0) & 0x03) / 3.0; + + return clamp(edgesLRTB + params.inv_sharpness, 0.0, 1.0); +} + +void main() { + ivec2 ssC = ivec2(gl_GlobalInvocationID.xy); + if (any(greaterThanEqual(ssC, ivec2(1.0 / params.pixel_size)))) { //too large, do nothing + return; + } + +#ifdef MODE_SMART + uvec2 pix_pos = uvec2(gl_GlobalInvocationID.xy); + vec2 uv = (gl_GlobalInvocationID.xy + vec2(0.5)) * params.pixel_size; + + // calculate index in the four deinterleaved source array texture + int mx = int(pix_pos.x % 2); + int my = int(pix_pos.y % 2); + int index_center = mx + my * 2; // center index + int index_horizontal = (1 - mx) + my * 2; // neighbouring, horizontal + int index_vertical = mx + (1 - my) * 2; // neighbouring, vertical + int index_diagonal = (1 - mx) + (1 - my) * 2; // diagonal + + vec4 color = texelFetch(source_texture, ivec3(pix_pos / uvec2(params.size_modifier), index_center), 0); + + vec4 edgesLRTB = unpack_edges(imageLoad(source_edges, ivec3(pix_pos / uvec2(params.size_modifier), index_center)).r); + + // convert index shifts to sampling offsets + float fmx = float(mx); + float fmy = float(my); + + // in case of an edge, push sampling offsets away from the edge (towards pixel center) + float fmxe = (edgesLRTB.y - edgesLRTB.x); + float fmye = (edgesLRTB.w - edgesLRTB.z); + + // calculate final sampling offsets and sample using bilinear filter + vec2 uv_horizontal = (gl_GlobalInvocationID.xy + vec2(0.5) + vec2(fmx + fmxe - 0.5, 0.5 - fmy)) * params.pixel_size; + vec4 color_horizontal = textureLod(source_texture, vec3(uv_horizontal, index_horizontal), 0.0); + vec2 uv_vertical = (gl_GlobalInvocationID.xy + vec2(0.5) + vec2(0.5 - fmx, fmy - 0.5 + fmye)) * params.pixel_size; + vec4 color_vertical = textureLod(source_texture, vec3(uv_vertical, index_vertical), 0.0); + vec2 uv_diagonal = (gl_GlobalInvocationID.xy + vec2(0.5) + vec2(fmx - 0.5 + fmxe, fmy - 0.5 + fmye)) * params.pixel_size; + vec4 color_diagonal = textureLod(source_texture, vec3(uv_diagonal, index_diagonal), 0.0); + + // reduce weight for samples near edge - if the edge is on both sides, weight goes to 0 + vec4 blendWeights; + blendWeights.x = 1.0; + blendWeights.y = (edgesLRTB.x + edgesLRTB.y) * 0.5; + blendWeights.z = (edgesLRTB.z + edgesLRTB.w) * 0.5; + blendWeights.w = (blendWeights.y + blendWeights.z) * 0.5; + + // calculate weighted average + float blendWeightsSum = dot(blendWeights, vec4(1.0, 1.0, 1.0, 1.0)); + color += color_horizontal * blendWeights.y; + color += color_vertical * blendWeights.z; + color += color_diagonal * blendWeights.w; + color /= blendWeightsSum; + + imageStore(dest_image, ivec2(gl_GlobalInvocationID.xy), color); +#else // !MODE_SMART + + vec2 uv = (gl_GlobalInvocationID.xy + vec2(0.5)) * params.pixel_size; +#ifdef MODE_HALF + vec4 a = textureLod(source_texture, vec3(uv, 0), 0.0); + vec4 d = textureLod(source_texture, vec3(uv, 3), 0.0); + vec4 avg = (a + d) * 0.5; + +#else + vec4 a = textureLod(source_texture, vec3(uv, 0), 0.0); + vec4 b = textureLod(source_texture, vec3(uv, 1), 0.0); + vec4 c = textureLod(source_texture, vec3(uv, 2), 0.0); + vec4 d = textureLod(source_texture, vec3(uv, 3), 0.0); + vec4 avg = (a + b + c + d) * 0.25; + +#endif + imageStore(dest_image, ivec2(gl_GlobalInvocationID.xy), avg); +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/tonemap.glsl b/servers/rendering/renderer_rd/shaders/tonemap.glsl index 1ce3e04421..948c6e1e39 100644 --- a/servers/rendering/renderer_rd/shaders/tonemap.glsl +++ b/servers/rendering/renderer_rd/shaders/tonemap.glsl @@ -140,7 +140,7 @@ vec4 texture2D_bicubic(sampler2D tex, vec2 uv, int p_lod) { vec2 p3 = (vec2(iuv.x + h1x, iuv.y + h1y) - vec2(0.5f)) * pixel_size; return (g0(fuv.y) * (g0x * textureLod(tex, p0, lod) + g1x * textureLod(tex, p1, lod))) + - (g1(fuv.y) * (g0x * textureLod(tex, p2, lod) + g1x * textureLod(tex, p3, lod))); + (g1(fuv.y) * (g0x * textureLod(tex, p2, lod) + g1x * textureLod(tex, p3, lod))); } #define GLOW_TEXTURE_SAMPLE(m_tex, m_uv, m_lod) texture2D_bicubic(m_tex, m_uv, m_lod) @@ -341,14 +341,14 @@ vec3 do_fxaa(vec3 color, float exposure, vec2 uv_interp) { dir.y = ((lumaNW + lumaSW) - (lumaNE + lumaSE)); float dirReduce = max((lumaNW + lumaNE + lumaSW + lumaSE) * - (0.25 * FXAA_REDUCE_MUL), + (0.25 * FXAA_REDUCE_MUL), FXAA_REDUCE_MIN); float rcpDirMin = 1.0 / (min(abs(dir.x), abs(dir.y)) + dirReduce); dir = min(vec2(FXAA_SPAN_MAX, FXAA_SPAN_MAX), max(vec2(-FXAA_SPAN_MAX, -FXAA_SPAN_MAX), dir * rcpDirMin)) * - params.pixel_size; + params.pixel_size; #ifdef MULTIVIEW vec3 rgbA = 0.5 * exposure * (textureLod(source_color, vec3(uv_interp + dir * (1.0 / 3.0 - 0.5), ViewIndex), 0.0).xyz + textureLod(source_color, vec3(uv_interp + dir * (2.0 / 3.0 - 0.5), ViewIndex), 0.0).xyz) * params.luminance_multiplier; diff --git a/servers/rendering/renderer_rd/shaders/volumetric_fog.glsl b/servers/rendering/renderer_rd/shaders/volumetric_fog.glsl index f2010222e5..181d3b272f 100644 --- a/servers/rendering/renderer_rd/shaders/volumetric_fog.glsl +++ b/servers/rendering/renderer_rd/shaders/volumetric_fog.glsl @@ -4,219 +4,103 @@ #VERSION_DEFINES -/* Do not use subgroups here, seems there is not much advantage and causes glitches -#if defined(has_GL_KHR_shader_subgroup_ballot) && defined(has_GL_KHR_shader_subgroup_arithmetic) -#extension GL_KHR_shader_subgroup_ballot: enable -#extension GL_KHR_shader_subgroup_arithmetic: enable - -#define USE_SUBGROUPS -#endif -*/ - -#if defined(MODE_FOG) || defined(MODE_FILTER) - -layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; - -#endif - -#if defined(MODE_DENSITY) - layout(local_size_x = 4, local_size_y = 4, local_size_z = 4) in; -#endif +#define SAMPLER_NEAREST_CLAMP 0 +#define SAMPLER_LINEAR_CLAMP 1 +#define SAMPLER_NEAREST_WITH_MIPMAPS_CLAMP 2 +#define SAMPLER_LINEAR_WITH_MIPMAPS_CLAMP 3 +#define SAMPLER_NEAREST_WITH_MIPMAPS_ANISOTROPIC_CLAMP 4 +#define SAMPLER_LINEAR_WITH_MIPMAPS_ANISOTROPIC_CLAMP 5 +#define SAMPLER_NEAREST_REPEAT 6 +#define SAMPLER_LINEAR_REPEAT 7 +#define SAMPLER_NEAREST_WITH_MIPMAPS_REPEAT 8 +#define SAMPLER_LINEAR_WITH_MIPMAPS_REPEAT 9 +#define SAMPLER_NEAREST_WITH_MIPMAPS_ANISOTROPIC_REPEAT 10 +#define SAMPLER_LINEAR_WITH_MIPMAPS_ANISOTROPIC_REPEAT 11 + +#define DENSITY_SCALE 1024.0 #include "cluster_data_inc.glsl" #include "light_data_inc.glsl" #define M_PI 3.14159265359 -layout(set = 0, binding = 1) uniform texture2D shadow_atlas; -layout(set = 0, binding = 2) uniform texture2D directional_shadow_atlas; - -layout(set = 0, binding = 3, std430) restrict readonly buffer OmniLights { - LightData data[]; -} -omni_lights; - -layout(set = 0, binding = 4, std430) restrict readonly buffer SpotLights { - LightData data[]; -} -spot_lights; +layout(set = 0, binding = 1) uniform sampler material_samplers[12]; -layout(set = 0, binding = 5, std140) uniform DirectionalLights { - DirectionalLightData data[MAX_DIRECTIONAL_LIGHT_DATA_STRUCTS]; +layout(set = 0, binding = 2, std430) restrict readonly buffer GlobalVariableData { + vec4 data[]; } -directional_lights; - -layout(set = 0, binding = 6, std430) buffer restrict readonly ClusterBuffer { - uint data[]; -} -cluster_buffer; - -layout(set = 0, binding = 7) uniform sampler linear_sampler; - -#ifdef MODE_DENSITY -layout(rgba16f, set = 0, binding = 8) uniform restrict writeonly image3D density_map; -layout(rgba16f, set = 0, binding = 9) uniform restrict readonly image3D fog_map; //unused -#endif - -#ifdef MODE_FOG -layout(rgba16f, set = 0, binding = 8) uniform restrict readonly image3D density_map; -layout(rgba16f, set = 0, binding = 9) uniform restrict writeonly image3D fog_map; -#endif - -#ifdef MODE_FILTER -layout(rgba16f, set = 0, binding = 8) uniform restrict readonly image3D source_map; -layout(rgba16f, set = 0, binding = 9) uniform restrict writeonly image3D dest_map; -#endif - -layout(set = 0, binding = 10) uniform sampler shadow_sampler; - -#define MAX_VOXEL_GI_INSTANCES 8 - -struct VoxelGIData { - mat4 xform; - vec3 bounds; - float dynamic_range; +global_variables; - float bias; - float normal_bias; - bool blend_ambient; - uint texture_slot; - - float anisotropy_strength; - float ambient_occlusion; - float ambient_occlusion_size; - uint mipmaps; -}; - -layout(set = 0, binding = 11, std140) uniform VoxelGIs { - VoxelGIData data[MAX_VOXEL_GI_INSTANCES]; -} -voxel_gi_instances; - -layout(set = 0, binding = 12) uniform texture3D voxel_gi_textures[MAX_VOXEL_GI_INSTANCES]; - -layout(set = 0, binding = 13) uniform sampler linear_sampler_with_mipmaps; - -#ifdef ENABLE_SDFGI - -// SDFGI Integration on set 1 -#define SDFGI_MAX_CASCADES 8 - -struct SDFVoxelGICascadeData { +layout(push_constant, binding = 0, std430) uniform Params { vec3 position; - float to_probe; - ivec3 probe_world_offset; - float to_cell; // 1/bounds * grid_size -}; - -layout(set = 1, binding = 0, std140) uniform SDFGI { - vec3 grid_size; - uint max_cascades; - - bool use_occlusion; - int probe_axis_size; - float probe_to_uvw; - float normal_bias; + float pad; - vec3 lightprobe_tex_pixel_size; - float energy; + vec3 extents; + float pad2; - vec3 lightprobe_uv_offset; - float y_mult; + ivec3 corner; + uint shape; - vec3 occlusion_clamp; - uint pad3; - - vec3 occlusion_renormalize; - uint pad4; - - vec3 cascade_probe_size; - uint pad5; - - SDFVoxelGICascadeData cascades[SDFGI_MAX_CASCADES]; + mat4 transform; } -sdfgi; - -layout(set = 1, binding = 1) uniform texture2DArray sdfgi_ambient_texture; - -layout(set = 1, binding = 2) uniform texture3D sdfgi_occlusion_texture; +params; -#endif //SDFGI +#ifdef MOLTENVK_USED +layout(set = 1, binding = 1) volatile buffer emissive_only_map_buffer { + uint emissive_only_map[]; +}; +#else +layout(r32ui, set = 1, binding = 1) uniform volatile uimage3D emissive_only_map; +#endif -layout(set = 0, binding = 14, std140) uniform Params { +layout(set = 1, binding = 2, std140) uniform SceneParams { vec2 fog_frustum_size_begin; vec2 fog_frustum_size_end; float fog_frustum_end; - float z_near; - float z_far; - int filter_axis; + float z_near; // + float z_far; // + float time; ivec3 fog_volume_size; - uint directional_light_count; - - vec3 light_color; - float base_density; - - float detail_spread; - float gi_inject; - uint max_voxel_gi_instances; - uint cluster_type_size; + uint directional_light_count; // - vec2 screen_size; - uint cluster_shift; - uint cluster_width; - - uint max_cluster_element_count_div_32; bool use_temporal_reprojection; uint temporal_frame; + float detail_spread; float temporal_blend; - mat3x4 cam_rotation; mat4 to_prev_view; + mat4 transform; } -params; - -layout(set = 0, binding = 15) uniform texture3D prev_density_texture; +scene_params; -float get_depth_at_pos(float cell_depth_size, int z) { - float d = float(z) * cell_depth_size + cell_depth_size * 0.5; //center of voxels - d = pow(d, params.detail_spread); - return params.fog_frustum_end * d; -} - -vec3 hash3f(uvec3 x) { - x = ((x >> 16) ^ x) * 0x45d9f3b; - x = ((x >> 16) ^ x) * 0x45d9f3b; - x = (x >> 16) ^ x; - return vec3(x & 0xFFFFF) / vec3(float(0xFFFFF)); -} - -float get_omni_attenuation(float distance, float inv_range, float decay) { - float nd = distance * inv_range; - nd *= nd; - nd *= nd; // nd^4 - nd = max(1.0 - nd, 0.0); - nd *= nd; // nd^2 - return nd * pow(max(distance, 0.0001), -decay); -} +#ifdef MOLTENVK_USED +layout(set = 1, binding = 3) volatile buffer density_only_map_buffer { + uint density_only_map[]; +}; +layout(set = 1, binding = 4) volatile buffer light_only_map_buffer { + uint light_only_map[]; +}; +#else +layout(r32ui, set = 1, binding = 3) uniform volatile uimage3D density_only_map; +layout(r32ui, set = 1, binding = 4) uniform volatile uimage3D light_only_map; +#endif -void cluster_get_item_range(uint p_offset, out uint item_min, out uint item_max, out uint item_from, out uint item_to) { - uint item_min_max = cluster_buffer.data[p_offset]; - item_min = item_min_max & 0xFFFF; - item_max = item_min_max >> 16; - ; +#ifdef MATERIAL_UNIFORMS_USED +layout(set = 2, binding = 0, std140) uniform MaterialUniforms{ +#MATERIAL_UNIFORMS +} material; +#endif - item_from = item_min >> 5; - item_to = (item_max == 0) ? 0 : ((item_max - 1) >> 5) + 1; //side effect of how it is stored, as item_max 0 means no elements -} +#GLOBALS -uint cluster_get_range_clip_mask(uint i, uint z_min, uint z_max) { - int local_min = clamp(int(z_min) - int(i) * 32, 0, 31); - int mask_width = min(int(z_max) - int(z_min), 32 - local_min); - return bitfieldInsert(uint(0), uint(0xFFFFFFFF), local_min, mask_width); +float get_depth_at_pos(float cell_depth_size, int z) { + float d = float(z) * cell_depth_size + cell_depth_size * 0.5; //center of voxels + d = pow(d, scene_params.detail_spread); + return scene_params.fog_frustum_end * d; } #define TEMPORAL_FRAMES 16 @@ -240,464 +124,167 @@ const vec3 halton_map[TEMPORAL_FRAMES] = vec3[]( vec3(0.03125, 0.59259259, 0.32)); void main() { - vec3 fog_cell_size = 1.0 / vec3(params.fog_volume_size); + vec3 fog_cell_size = 1.0 / vec3(scene_params.fog_volume_size); -#ifdef MODE_DENSITY - - ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); - if (any(greaterThanEqual(pos, params.fog_volume_size))) { + ivec3 pos = ivec3(gl_GlobalInvocationID.xyz) + params.corner; + if (any(greaterThanEqual(pos, scene_params.fog_volume_size))) { return; //do not compute } +#ifdef MOLTENVK_USED + uint lpos = pos.z * scene_params.fog_volume_size.x * scene_params.fog_volume_size.y + pos.y * scene_params.fog_volume_size.x + pos.x; +#endif vec3 posf = vec3(pos); - //posf += mix(vec3(0.0),vec3(1.0),0.3) * hash3f(uvec3(pos)) * 2.0 - 1.0; - vec3 fog_unit_pos = posf * fog_cell_size + fog_cell_size * 0.5; //center of voxels - - uvec2 screen_pos = uvec2(fog_unit_pos.xy * params.screen_size); - uvec2 cluster_pos = screen_pos >> params.cluster_shift; - uint cluster_offset = (params.cluster_width * cluster_pos.y + cluster_pos.x) * (params.max_cluster_element_count_div_32 + 32); - //positions in screen are too spread apart, no hopes for optimizing with subgroups - - fog_unit_pos.z = pow(fog_unit_pos.z, params.detail_spread); + fog_unit_pos.z = pow(fog_unit_pos.z, scene_params.detail_spread); vec3 view_pos; - view_pos.xy = (fog_unit_pos.xy * 2.0 - 1.0) * mix(params.fog_frustum_size_begin, params.fog_frustum_size_end, vec2(fog_unit_pos.z)); - view_pos.z = -params.fog_frustum_end * fog_unit_pos.z; + view_pos.xy = (fog_unit_pos.xy * 2.0 - 1.0) * mix(scene_params.fog_frustum_size_begin, scene_params.fog_frustum_size_end, vec2(fog_unit_pos.z)); + view_pos.z = -scene_params.fog_frustum_end * fog_unit_pos.z; view_pos.y = -view_pos.y; - vec4 reprojected_density = vec4(0.0); - float reproject_amount = 0.0; - - if (params.use_temporal_reprojection) { - vec3 prev_view = (params.to_prev_view * vec4(view_pos, 1.0)).xyz; + if (scene_params.use_temporal_reprojection) { + vec3 prev_view = (scene_params.to_prev_view * vec4(view_pos, 1.0)).xyz; //undo transform into prev view prev_view.y = -prev_view.y; //z back to unit size - prev_view.z /= -params.fog_frustum_end; + prev_view.z /= -scene_params.fog_frustum_end; //xy back to unit size - prev_view.xy /= mix(params.fog_frustum_size_begin, params.fog_frustum_size_end, vec2(prev_view.z)); + prev_view.xy /= mix(scene_params.fog_frustum_size_begin, scene_params.fog_frustum_size_end, vec2(prev_view.z)); prev_view.xy = prev_view.xy * 0.5 + 0.5; //z back to unspread value - prev_view.z = pow(prev_view.z, 1.0 / params.detail_spread); + prev_view.z = pow(prev_view.z, 1.0 / scene_params.detail_spread); if (all(greaterThan(prev_view, vec3(0.0))) && all(lessThan(prev_view, vec3(1.0)))) { //reprojectinon fits - - reprojected_density = textureLod(sampler3D(prev_density_texture, linear_sampler), prev_view, 0.0); - reproject_amount = params.temporal_blend; - // Since we can reproject, now we must jitter the current view pos. // This is done here because cells that can't reproject should not jitter. - fog_unit_pos = posf * fog_cell_size + fog_cell_size * halton_map[params.temporal_frame]; //center of voxels, offset by halton table - - screen_pos = uvec2(fog_unit_pos.xy * params.screen_size); - cluster_pos = screen_pos >> params.cluster_shift; - cluster_offset = (params.cluster_width * cluster_pos.y + cluster_pos.x) * (params.max_cluster_element_count_div_32 + 32); - //positions in screen are too spread apart, no hopes for optimizing with subgroups + fog_unit_pos = posf * fog_cell_size + fog_cell_size * halton_map[scene_params.temporal_frame]; //center of voxels, offset by halton table + fog_unit_pos.z = pow(fog_unit_pos.z, scene_params.detail_spread); - fog_unit_pos.z = pow(fog_unit_pos.z, params.detail_spread); - - view_pos.xy = (fog_unit_pos.xy * 2.0 - 1.0) * mix(params.fog_frustum_size_begin, params.fog_frustum_size_end, vec2(fog_unit_pos.z)); - view_pos.z = -params.fog_frustum_end * fog_unit_pos.z; + view_pos.xy = (fog_unit_pos.xy * 2.0 - 1.0) * mix(scene_params.fog_frustum_size_begin, scene_params.fog_frustum_size_end, vec2(fog_unit_pos.z)); + view_pos.z = -scene_params.fog_frustum_end * fog_unit_pos.z; view_pos.y = -view_pos.y; } } - uint cluster_z = uint(clamp((abs(view_pos.z) / params.z_far) * 32.0, 0.0, 31.0)); - - vec3 total_light = params.light_color; + float density = 0.0; + vec3 emission = vec3(0.0); + vec3 albedo = vec3(0.0); - float total_density = params.base_density; float cell_depth_size = abs(view_pos.z - get_depth_at_pos(fog_cell_size.z, pos.z + 1)); - //compute directional lights - - for (uint i = 0; i < params.directional_light_count; i++) { - vec3 shadow_attenuation = vec3(1.0); - - if (directional_lights.data[i].shadow_enabled) { - float depth_z = -view_pos.z; - - vec4 pssm_coord; - vec3 shadow_color = directional_lights.data[i].shadow_color1.rgb; - vec3 light_dir = directional_lights.data[i].direction; - vec4 v = vec4(view_pos, 1.0); - float z_range; - - if (depth_z < directional_lights.data[i].shadow_split_offsets.x) { - pssm_coord = (directional_lights.data[i].shadow_matrix1 * v); - pssm_coord /= pssm_coord.w; - z_range = directional_lights.data[i].shadow_z_range.x; - - } else if (depth_z < directional_lights.data[i].shadow_split_offsets.y) { - pssm_coord = (directional_lights.data[i].shadow_matrix2 * v); - pssm_coord /= pssm_coord.w; - z_range = directional_lights.data[i].shadow_z_range.y; - - } else if (depth_z < directional_lights.data[i].shadow_split_offsets.z) { - pssm_coord = (directional_lights.data[i].shadow_matrix3 * v); - pssm_coord /= pssm_coord.w; - z_range = directional_lights.data[i].shadow_z_range.z; - - } else { - pssm_coord = (directional_lights.data[i].shadow_matrix4 * v); - pssm_coord /= pssm_coord.w; - z_range = directional_lights.data[i].shadow_z_range.w; - } - - float depth = texture(sampler2D(directional_shadow_atlas, linear_sampler), pssm_coord.xy).r; - float shadow = exp(min(0.0, (depth - pssm_coord.z)) * z_range * directional_lights.data[i].shadow_volumetric_fog_fade); - - /* - //float shadow = textureProj(sampler2DShadow(directional_shadow_atlas,shadow_sampler),pssm_coord); - float shadow = 0.0; - for(float xi=-1;xi<=1;xi++) { - for(float yi=-1;yi<=1;yi++) { - vec2 ofs = vec2(xi,yi) * 1.5 * params.directional_shadow_pixel_size; - shadow += textureProj(sampler2DShadow(directional_shadow_atlas,shadow_sampler),pssm_coord + vec4(ofs,0.0,0.0)); - } - - } - - shadow /= 3.0 * 3.0; -*/ - shadow = mix(shadow, 1.0, smoothstep(directional_lights.data[i].fade_from, directional_lights.data[i].fade_to, view_pos.z)); //done with negative values for performance - - shadow_attenuation = mix(shadow_color, vec3(1.0), shadow); - } - - total_light += shadow_attenuation * directional_lights.data[i].color * directional_lights.data[i].energy / M_PI; + vec4 world = scene_params.transform * vec4(view_pos, 1.0); + world.xyz /= world.w; + + vec3 uvw = fog_unit_pos; + + vec4 local_pos = params.transform * world; + local_pos.xyz /= local_pos.w; + + float sdf = -1.0; + if (params.shape == 0) { + //Ellipsoid + // https://www.shadertoy.com/view/tdS3DG + float k0 = length(local_pos.xyz / params.extents); + float k1 = length(local_pos.xyz / (params.extents * params.extents)); + sdf = k0 * (k0 - 1.0) / k1; + } else if (params.shape == 1) { + // Box + // https://iquilezles.org/www/articles/distfunctions/distfunctions.htm + vec3 q = abs(local_pos.xyz) - params.extents; + sdf = length(max(q, 0.0)) + min(max(q.x, max(q.y, q.z)), 0.0); } - //compute lights from cluster - - { //omni lights - - uint cluster_omni_offset = cluster_offset; - - uint item_min; - uint item_max; - uint item_from; - uint item_to; - - cluster_get_item_range(cluster_omni_offset + params.max_cluster_element_count_div_32 + cluster_z, item_min, item_max, item_from, item_to); - -#ifdef USE_SUBGROUPS - item_from = subgroupBroadcastFirst(subgroupMin(item_from)); - item_to = subgroupBroadcastFirst(subgroupMax(item_to)); + float cull_mask = 1.0; //used to cull cells that do not contribute + if (params.shape <= 1) { +#ifndef SDF_USED + cull_mask = 1.0 - smoothstep(-0.1, 0.0, sdf); #endif + uvw = clamp((local_pos.xyz + params.extents) / (2.0 * params.extents), 0.0, 1.0); + } - for (uint i = item_from; i < item_to; i++) { - uint mask = cluster_buffer.data[cluster_omni_offset + i]; - mask &= cluster_get_range_clip_mask(i, item_min, item_max); -#ifdef USE_SUBGROUPS - uint merged_mask = subgroupBroadcastFirst(subgroupOr(mask)); + if (cull_mask > 0.0) { + { +#CODE : FOG + } + +#ifdef DENSITY_USED + density *= cull_mask; + if (abs(density) > 0.001) { + int final_density = int(density * DENSITY_SCALE); +#ifdef MOLTENVK_USED + atomicAdd(density_only_map[lpos], uint(final_density)); #else - uint merged_mask = mask; + imageAtomicAdd(density_only_map, pos, uint(final_density)); #endif - while (merged_mask != 0) { - uint bit = findMSB(merged_mask); - merged_mask &= ~(1 << bit); -#ifdef USE_SUBGROUPS - if (((1 << bit) & mask) == 0) { //do not process if not originally here - continue; - } +#ifdef EMISSION_USED + { + emission *= clamp(density, 0.0, 1.0); + emission = clamp(emission, vec3(0.0), vec3(4.0)); + // Scale to fit into R11G11B10 with a range of 0-4 + uvec3 emission_u = uvec3(emission.r * 511.0, emission.g * 511.0, emission.b * 255.0); + // R and G have 11 bits each and B has 10. Then pack them into a 32 bit uint + uint final_emission = emission_u.r << 21 | emission_u.g << 10 | emission_u.b; +#ifdef MOLTENVK_USED + uint prev_emission = atomicAdd(emissive_only_map[lpos], final_emission); +#else + uint prev_emission = imageAtomicAdd(emissive_only_map, pos, final_emission); #endif - uint light_index = 32 * i + bit; - - //if (!bool(omni_omni_lights.data[light_index].mask & draw_call.layer_mask)) { - // continue; //not masked - //} - - vec3 light_pos = omni_lights.data[light_index].position; - float d = distance(omni_lights.data[light_index].position, view_pos); - float shadow_attenuation = 1.0; - - if (d * omni_lights.data[light_index].inv_radius < 1.0) { - float attenuation = get_omni_attenuation(d, omni_lights.data[light_index].inv_radius, omni_lights.data[light_index].attenuation); - - vec3 light = omni_lights.data[light_index].color / M_PI; - - if (omni_lights.data[light_index].shadow_enabled) { - //has shadow - vec4 v = vec4(view_pos, 1.0); - - vec4 splane = (omni_lights.data[light_index].shadow_matrix * v); - float shadow_len = length(splane.xyz); //need to remember shadow len from here - splane.xyz = normalize(splane.xyz); - vec4 clamp_rect = omni_lights.data[light_index].atlas_rect; + // Adding can lead to colors overflowing, so validate + uvec3 prev_emission_u = uvec3(prev_emission >> 21, (prev_emission << 11) >> 21, prev_emission % 1024); + uint add_emission = final_emission + prev_emission; + uvec3 add_emission_u = uvec3(add_emission >> 21, (add_emission << 11) >> 21, add_emission % 1024); - if (splane.z >= 0.0) { - splane.z += 1.0; + bvec3 overflowing = lessThan(add_emission_u, prev_emission_u + emission_u); - clamp_rect.y += clamp_rect.w; - - } else { - splane.z = 1.0 - splane.z; - } - - splane.xy /= splane.z; - - splane.xy = splane.xy * 0.5 + 0.5; - splane.z = shadow_len * omni_lights.data[light_index].inv_radius; - splane.xy = clamp_rect.xy + splane.xy * clamp_rect.zw; - splane.w = 1.0; //needed? i think it should be 1 already - - float depth = texture(sampler2D(shadow_atlas, linear_sampler), splane.xy).r; - - shadow_attenuation = exp(min(0.0, (depth - splane.z)) / omni_lights.data[light_index].inv_radius * omni_lights.data[light_index].shadow_volumetric_fog_fade); - } - total_light += light * attenuation * shadow_attenuation; + if (any(overflowing)) { + uvec3 overflow_factor = mix(uvec3(0), uvec3(2047 << 21, 2047 << 10, 1023), overflowing); + uint force_max = overflow_factor.r | overflow_factor.g | overflow_factor.b; +#ifdef MOLTENVK_USED + atomicOr(emissive_only_map[lpos], force_max); +#else + imageAtomicOr(emissive_only_map, pos, force_max); +#endif } } - } - } - - { //spot lights - - uint cluster_spot_offset = cluster_offset + params.cluster_type_size; - - uint item_min; - uint item_max; - uint item_from; - uint item_to; - - cluster_get_item_range(cluster_spot_offset + params.max_cluster_element_count_div_32 + cluster_z, item_min, item_max, item_from, item_to); - -#ifdef USE_SUBGROUPS - item_from = subgroupBroadcastFirst(subgroupMin(item_from)); - item_to = subgroupBroadcastFirst(subgroupMax(item_to)); #endif - - for (uint i = item_from; i < item_to; i++) { - uint mask = cluster_buffer.data[cluster_spot_offset + i]; - mask &= cluster_get_range_clip_mask(i, item_min, item_max); -#ifdef USE_SUBGROUPS - uint merged_mask = subgroupBroadcastFirst(subgroupOr(mask)); +#ifdef ALBEDO_USED + { + vec3 scattering = albedo * clamp(density, 0.0, 1.0); + scattering = clamp(scattering, vec3(0.0), vec3(1.0)); + uvec3 scattering_u = uvec3(scattering.r * 2047.0, scattering.g * 2047.0, scattering.b * 1023.0); + // R and G have 11 bits each and B has 10. Then pack them into a 32 bit uint + uint final_scattering = scattering_u.r << 21 | scattering_u.g << 10 | scattering_u.b; +#ifdef MOLTENVK_USED + uint prev_scattering = atomicAdd(light_only_map[lpos], final_scattering); #else - uint merged_mask = mask; -#endif - - while (merged_mask != 0) { - uint bit = findMSB(merged_mask); - merged_mask &= ~(1 << bit); -#ifdef USE_SUBGROUPS - if (((1 << bit) & mask) == 0) { //do not process if not originally here - continue; - } + uint prev_scattering = imageAtomicAdd(light_only_map, pos, final_scattering); #endif - //if (!bool(omni_lights.data[light_index].mask & draw_call.layer_mask)) { - // continue; //not masked - //} - - uint light_index = 32 * i + bit; - - vec3 light_pos = spot_lights.data[light_index].position; - vec3 light_rel_vec = spot_lights.data[light_index].position - view_pos; - float d = length(light_rel_vec); - float shadow_attenuation = 1.0; - - if (d * spot_lights.data[light_index].inv_radius < 1.0) { - float attenuation = get_omni_attenuation(d, spot_lights.data[light_index].inv_radius, spot_lights.data[light_index].attenuation); - - vec3 spot_dir = spot_lights.data[light_index].direction; - float scos = max(dot(-normalize(light_rel_vec), spot_dir), spot_lights.data[light_index].cone_angle); - float spot_rim = max(0.0001, (1.0 - scos) / (1.0 - spot_lights.data[light_index].cone_angle)); - attenuation *= 1.0 - pow(spot_rim, spot_lights.data[light_index].cone_attenuation); - - vec3 light = spot_lights.data[light_index].color / M_PI; - - if (spot_lights.data[light_index].shadow_enabled) { - //has shadow - vec4 v = vec4(view_pos, 1.0); - - vec4 splane = (spot_lights.data[light_index].shadow_matrix * v); - splane /= splane.w; - - float depth = texture(sampler2D(shadow_atlas, linear_sampler), splane.xy).r; - - shadow_attenuation = exp(min(0.0, (depth - splane.z)) / spot_lights.data[light_index].inv_radius * spot_lights.data[light_index].shadow_volumetric_fog_fade); - } - - total_light += light * attenuation * shadow_attenuation; - } - } - } - } - - vec3 world_pos = mat3(params.cam_rotation) * view_pos; - - for (uint i = 0; i < params.max_voxel_gi_instances; i++) { - vec3 position = (voxel_gi_instances.data[i].xform * vec4(world_pos, 1.0)).xyz; - - //this causes corrupted pixels, i have no idea why.. - if (all(bvec2(all(greaterThanEqual(position, vec3(0.0))), all(lessThan(position, voxel_gi_instances.data[i].bounds))))) { - position /= voxel_gi_instances.data[i].bounds; - - vec4 light = vec4(0.0); - for (uint j = 0; j < voxel_gi_instances.data[i].mipmaps; j++) { - vec4 slight = textureLod(sampler3D(voxel_gi_textures[i], linear_sampler_with_mipmaps), position, float(j)); - float a = (1.0 - light.a); - light += a * slight; - } - - light.rgb *= voxel_gi_instances.data[i].dynamic_range * params.gi_inject; - - total_light += light.rgb; - } - } - - //sdfgi -#ifdef ENABLE_SDFGI + // Adding can lead to colors overflowing, so validate + uvec3 prev_scattering_u = uvec3(prev_scattering >> 21, (prev_scattering << 11) >> 21, prev_scattering % 1024); + uint add_scattering = final_scattering + prev_scattering; + uvec3 add_scattering_u = uvec3(add_scattering >> 21, (add_scattering << 11) >> 21, add_scattering % 1024); - { - float blend = -1.0; - vec3 ambient_total = vec3(0.0); + bvec3 overflowing = lessThan(add_scattering_u, prev_scattering_u + scattering_u); - for (uint i = 0; i < sdfgi.max_cascades; i++) { - vec3 cascade_pos = (world_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 - } - - vec3 base_pos = floor(cascade_pos); - ivec3 probe_base_pos = ivec3(base_pos); - - vec4 ambient_accum = vec4(0.0); - - ivec3 tex_pos = ivec3(probe_base_pos.xy, int(i)); - tex_pos.x += probe_base_pos.z * sdfgi.probe_axis_size; - - 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 trilinear = vec3(1.0) - abs(probe_to_pos); - float weight = trilinear.x * trilinear.y * trilinear.z; - - // Compute lightprobe occlusion - - if (sdfgi.use_occlusion) { - ivec3 occ_indexv = abs((sdfgi.cascades[i].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(i); - 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_texture, linear_sampler), occ_pos, 0.0), occ_mask); - - weight *= max(occlusion, 0.01); + if (any(overflowing)) { + uvec3 overflow_factor = mix(uvec3(0), uvec3(2047 << 21, 2047 << 10, 1023), overflowing); + uint force_max = overflow_factor.r | overflow_factor.g | overflow_factor.b; +#ifdef MOLTENVK_USED + atomicOr(light_only_map[lpos], force_max); +#else + imageAtomicOr(light_only_map, pos, force_max); +#endif } - - // Compute ambient texture position - - ivec3 uvw = tex_pos; - uvw.xy += offset.xy; - uvw.x += offset.z * sdfgi.probe_axis_size; - - vec3 ambient = texelFetch(sampler2DArray(sdfgi_ambient_texture, linear_sampler), uvw, 0).rgb; - - ambient_accum.rgb += ambient * weight; - ambient_accum.a += weight; - } - - if (ambient_accum.a > 0) { - ambient_accum.rgb /= ambient_accum.a; } - ambient_total = ambient_accum.rgb; - break; +#endif // ALBEDO_USED } - - total_light += ambient_total * params.gi_inject; +#endif // DENSITY_USED } - -#endif - - vec4 final_density = vec4(total_light, total_density); - - final_density = mix(final_density, reprojected_density, reproject_amount); - - imageStore(density_map, pos, final_density); -#endif - -#ifdef MODE_FOG - - ivec3 pos = ivec3(gl_GlobalInvocationID.xy, 0); - - if (any(greaterThanEqual(pos, params.fog_volume_size))) { - return; //do not compute - } - - vec4 fog_accum = vec4(0.0); - float prev_z = 0.0; - - float t = 1.0; - - for (int i = 0; i < params.fog_volume_size.z; i++) { - //compute fog position - ivec3 fog_pos = pos + ivec3(0, 0, i); - //get fog value - vec4 fog = imageLoad(density_map, fog_pos); - - //get depth at cell pos - float z = get_depth_at_pos(fog_cell_size.z, i); - //get distance from previous pos - float d = abs(prev_z - z); - //compute exinction based on beer's - float extinction = t * exp(-d * fog.a); - //compute alpha based on different of extinctions - float alpha = t - extinction; - //update extinction - t = extinction; - - fog_accum += vec4(fog.rgb * alpha, alpha); - prev_z = z; - - vec4 fog_value; - - if (fog_accum.a > 0.0) { - fog_value = vec4(fog_accum.rgb / fog_accum.a, 1.0 - t); - } else { - fog_value = vec4(0.0); - } - - imageStore(fog_map, fog_pos, fog_value); - } - -#endif - -#ifdef MODE_FILTER - - ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); - - const float gauss[7] = float[](0.071303, 0.131514, 0.189879, 0.214607, 0.189879, 0.131514, 0.071303); - - const ivec3 filter_dir[3] = ivec3[](ivec3(1, 0, 0), ivec3(0, 1, 0), ivec3(0, 0, 1)); - ivec3 offset = filter_dir[params.filter_axis]; - - vec4 accum = vec4(0.0); - for (int i = -3; i <= 3; i++) { - accum += imageLoad(source_map, clamp(pos + offset * i, ivec3(0), params.fog_volume_size - ivec3(1))) * gauss[i + 3]; - } - - imageStore(dest_map, pos, accum); - -#endif } diff --git a/servers/rendering/renderer_rd/shaders/volumetric_fog_process.glsl b/servers/rendering/renderer_rd/shaders/volumetric_fog_process.glsl new file mode 100644 index 0000000000..7c8d4f7f99 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/volumetric_fog_process.glsl @@ -0,0 +1,785 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +/* Do not use subgroups here, seems there is not much advantage and causes glitches +#if defined(has_GL_KHR_shader_subgroup_ballot) && defined(has_GL_KHR_shader_subgroup_arithmetic) +#extension GL_KHR_shader_subgroup_ballot: enable +#extension GL_KHR_shader_subgroup_arithmetic: enable + +#define USE_SUBGROUPS +#endif +*/ + +#ifdef MODE_DENSITY +layout(local_size_x = 4, local_size_y = 4, local_size_z = 4) in; +#else +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; +#endif + +#include "cluster_data_inc.glsl" +#include "light_data_inc.glsl" + +#define M_PI 3.14159265359 + +#define DENSITY_SCALE 1024.0 + +layout(set = 0, binding = 1) uniform texture2D shadow_atlas; +layout(set = 0, binding = 2) uniform texture2D directional_shadow_atlas; + +layout(set = 0, binding = 3, std430) restrict readonly buffer OmniLights { + LightData data[]; +} +omni_lights; + +layout(set = 0, binding = 4, std430) restrict readonly buffer SpotLights { + LightData data[]; +} +spot_lights; + +layout(set = 0, binding = 5, std140) uniform DirectionalLights { + DirectionalLightData data[MAX_DIRECTIONAL_LIGHT_DATA_STRUCTS]; +} +directional_lights; + +layout(set = 0, binding = 6, std430) buffer restrict readonly ClusterBuffer { + uint data[]; +} +cluster_buffer; + +layout(set = 0, binding = 7) uniform sampler linear_sampler; + +#ifdef MODE_DENSITY +layout(rgba16f, set = 0, binding = 8) uniform restrict writeonly image3D density_map; +layout(rgba16f, set = 0, binding = 9) uniform restrict readonly image3D fog_map; //unused +#endif + +#ifdef MODE_FOG +layout(rgba16f, set = 0, binding = 8) uniform restrict readonly image3D density_map; +layout(rgba16f, set = 0, binding = 9) uniform restrict writeonly image3D fog_map; +#endif + +#ifdef MODE_COPY +layout(rgba16f, set = 0, binding = 8) uniform restrict readonly image3D source_map; +layout(rgba16f, set = 0, binding = 9) uniform restrict writeonly image3D dest_map; +#endif + +#ifdef MODE_FILTER +layout(rgba16f, set = 0, binding = 8) uniform restrict readonly image3D source_map; +layout(rgba16f, set = 0, binding = 9) uniform restrict writeonly image3D dest_map; +#endif + +layout(set = 0, binding = 10) uniform sampler shadow_sampler; + +#define MAX_VOXEL_GI_INSTANCES 8 + +struct VoxelGIData { + mat4 xform; // 64 - 64 + + vec3 bounds; // 12 - 76 + float dynamic_range; // 4 - 80 + + float bias; // 4 - 84 + float normal_bias; // 4 - 88 + bool blend_ambient; // 4 - 92 + uint mipmaps; // 4 - 96 +}; + +layout(set = 0, binding = 11, std140) uniform VoxelGIs { + VoxelGIData data[MAX_VOXEL_GI_INSTANCES]; +} +voxel_gi_instances; + +layout(set = 0, binding = 12) uniform texture3D voxel_gi_textures[MAX_VOXEL_GI_INSTANCES]; + +layout(set = 0, binding = 13) uniform sampler linear_sampler_with_mipmaps; + +#ifdef ENABLE_SDFGI + +// SDFGI Integration on set 1 +#define SDFGI_MAX_CASCADES 8 + +struct SDFVoxelGICascadeData { + vec3 position; + float to_probe; + ivec3 probe_world_offset; + float to_cell; // 1/bounds * grid_size +}; + +layout(set = 1, binding = 0, 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; + + SDFVoxelGICascadeData cascades[SDFGI_MAX_CASCADES]; +} +sdfgi; + +layout(set = 1, binding = 1) uniform texture2DArray sdfgi_ambient_texture; + +layout(set = 1, binding = 2) uniform texture3D sdfgi_occlusion_texture; + +#endif //SDFGI + +layout(set = 0, binding = 14, std140) uniform Params { + vec2 fog_frustum_size_begin; + vec2 fog_frustum_size_end; + + float fog_frustum_end; + float ambient_inject; + float z_far; + int filter_axis; + + vec3 ambient_color; + float sky_contribution; + + ivec3 fog_volume_size; + uint directional_light_count; + + vec3 base_emission; + float base_density; + + vec3 base_scattering; + float phase_g; + + float detail_spread; + float gi_inject; + uint max_voxel_gi_instances; + uint cluster_type_size; + + vec2 screen_size; + uint cluster_shift; + uint cluster_width; + + uint max_cluster_element_count_div_32; + bool use_temporal_reprojection; + uint temporal_frame; + float temporal_blend; + + mat3x4 cam_rotation; + mat4 to_prev_view; + + mat3 radiance_inverse_xform; +} +params; +#ifndef MODE_COPY +layout(set = 0, binding = 15) uniform texture3D prev_density_texture; + +#ifdef MOLTENVK_USED +layout(set = 0, binding = 16) buffer density_only_map_buffer { + uint density_only_map[]; +}; +layout(set = 0, binding = 17) buffer light_only_map_buffer { + uint light_only_map[]; +}; +layout(set = 0, binding = 18) buffer emissive_only_map_buffer { + uint emissive_only_map[]; +}; +#else +layout(r32ui, set = 0, binding = 16) uniform uimage3D density_only_map; +layout(r32ui, set = 0, binding = 17) uniform uimage3D light_only_map; +layout(r32ui, set = 0, binding = 18) uniform uimage3D emissive_only_map; +#endif + +#ifdef USE_RADIANCE_CUBEMAP_ARRAY +layout(set = 0, binding = 19) uniform textureCubeArray sky_texture; +#else +layout(set = 0, binding = 19) uniform textureCube sky_texture; +#endif +#endif // MODE_COPY + +float get_depth_at_pos(float cell_depth_size, int z) { + float d = float(z) * cell_depth_size + cell_depth_size * 0.5; //center of voxels + d = pow(d, params.detail_spread); + return params.fog_frustum_end * d; +} + +vec3 hash3f(uvec3 x) { + x = ((x >> 16) ^ x) * 0x45d9f3b; + x = ((x >> 16) ^ x) * 0x45d9f3b; + x = (x >> 16) ^ x; + return vec3(x & 0xFFFFF) / vec3(float(0xFFFFF)); +} + +float get_omni_attenuation(float dist, float inv_range, float decay) { + float nd = dist * inv_range; + nd *= nd; + nd *= nd; // nd^4 + nd = max(1.0 - nd, 0.0); + nd *= nd; // nd^2 + return nd * pow(max(dist, 0.0001), -decay); +} + +void cluster_get_item_range(uint p_offset, out uint item_min, out uint item_max, out uint item_from, out uint item_to) { + uint item_min_max = cluster_buffer.data[p_offset]; + item_min = item_min_max & 0xFFFF; + item_max = item_min_max >> 16; + ; + + item_from = item_min >> 5; + item_to = (item_max == 0) ? 0 : ((item_max - 1) >> 5) + 1; //side effect of how it is stored, as item_max 0 means no elements +} + +uint cluster_get_range_clip_mask(uint i, uint z_min, uint z_max) { + int local_min = clamp(int(z_min) - int(i) * 32, 0, 31); + int mask_width = min(int(z_max) - int(z_min), 32 - local_min); + return bitfieldInsert(uint(0), uint(0xFFFFFFFF), local_min, mask_width); +} + +float henyey_greenstein(float cos_theta, float g) { + const float k = 0.0795774715459; // 1 / (4 * PI) + return k * (1.0 - g * g) / (pow(1.0 + g * g - 2.0 * g * cos_theta, 1.5)); +} + +#define TEMPORAL_FRAMES 16 + +const vec3 halton_map[TEMPORAL_FRAMES] = vec3[]( + vec3(0.5, 0.33333333, 0.2), + vec3(0.25, 0.66666667, 0.4), + vec3(0.75, 0.11111111, 0.6), + vec3(0.125, 0.44444444, 0.8), + vec3(0.625, 0.77777778, 0.04), + vec3(0.375, 0.22222222, 0.24), + vec3(0.875, 0.55555556, 0.44), + vec3(0.0625, 0.88888889, 0.64), + vec3(0.5625, 0.03703704, 0.84), + vec3(0.3125, 0.37037037, 0.08), + vec3(0.8125, 0.7037037, 0.28), + vec3(0.1875, 0.14814815, 0.48), + vec3(0.6875, 0.48148148, 0.68), + vec3(0.4375, 0.81481481, 0.88), + vec3(0.9375, 0.25925926, 0.12), + vec3(0.03125, 0.59259259, 0.32)); + +void main() { + vec3 fog_cell_size = 1.0 / vec3(params.fog_volume_size); + +#ifdef MODE_DENSITY + + ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); + if (any(greaterThanEqual(pos, params.fog_volume_size))) { + return; //do not compute + } +#ifdef MOLTENVK_USED + uint lpos = pos.z * params.fog_volume_size.x * params.fog_volume_size.y + pos.y * params.fog_volume_size.x + pos.x; +#endif + + vec3 posf = vec3(pos); + + //posf += mix(vec3(0.0),vec3(1.0),0.3) * hash3f(uvec3(pos)) * 2.0 - 1.0; + + vec3 fog_unit_pos = posf * fog_cell_size + fog_cell_size * 0.5; //center of voxels + + uvec2 screen_pos = uvec2(fog_unit_pos.xy * params.screen_size); + uvec2 cluster_pos = screen_pos >> params.cluster_shift; + uint cluster_offset = (params.cluster_width * cluster_pos.y + cluster_pos.x) * (params.max_cluster_element_count_div_32 + 32); + //positions in screen are too spread apart, no hopes for optimizing with subgroups + + fog_unit_pos.z = pow(fog_unit_pos.z, params.detail_spread); + + vec3 view_pos; + view_pos.xy = (fog_unit_pos.xy * 2.0 - 1.0) * mix(params.fog_frustum_size_begin, params.fog_frustum_size_end, vec2(fog_unit_pos.z)); + view_pos.z = -params.fog_frustum_end * fog_unit_pos.z; + view_pos.y = -view_pos.y; + + vec4 reprojected_density = vec4(0.0); + float reproject_amount = 0.0; + + if (params.use_temporal_reprojection) { + vec3 prev_view = (params.to_prev_view * vec4(view_pos, 1.0)).xyz; + //undo transform into prev view + prev_view.y = -prev_view.y; + //z back to unit size + prev_view.z /= -params.fog_frustum_end; + //xy back to unit size + prev_view.xy /= mix(params.fog_frustum_size_begin, params.fog_frustum_size_end, vec2(prev_view.z)); + prev_view.xy = prev_view.xy * 0.5 + 0.5; + //z back to unspread value + prev_view.z = pow(prev_view.z, 1.0 / params.detail_spread); + + if (all(greaterThan(prev_view, vec3(0.0))) && all(lessThan(prev_view, vec3(1.0)))) { + //reprojectinon fits + + reprojected_density = textureLod(sampler3D(prev_density_texture, linear_sampler), prev_view, 0.0); + reproject_amount = params.temporal_blend; + + // Since we can reproject, now we must jitter the current view pos. + // This is done here because cells that can't reproject should not jitter. + + fog_unit_pos = posf * fog_cell_size + fog_cell_size * halton_map[params.temporal_frame]; //center of voxels, offset by halton table + + screen_pos = uvec2(fog_unit_pos.xy * params.screen_size); + cluster_pos = screen_pos >> params.cluster_shift; + cluster_offset = (params.cluster_width * cluster_pos.y + cluster_pos.x) * (params.max_cluster_element_count_div_32 + 32); + //positions in screen are too spread apart, no hopes for optimizing with subgroups + + fog_unit_pos.z = pow(fog_unit_pos.z, params.detail_spread); + + view_pos.xy = (fog_unit_pos.xy * 2.0 - 1.0) * mix(params.fog_frustum_size_begin, params.fog_frustum_size_end, vec2(fog_unit_pos.z)); + view_pos.z = -params.fog_frustum_end * fog_unit_pos.z; + view_pos.y = -view_pos.y; + } + } + + uint cluster_z = uint(clamp((abs(view_pos.z) / params.z_far) * 32.0, 0.0, 31.0)); + + vec3 total_light = vec3(0.0); + + float total_density = params.base_density; +#ifdef MOLTENVK_USED + uint local_density = density_only_map[lpos]; +#else + uint local_density = imageLoad(density_only_map, pos).x; +#endif + + total_density += float(int(local_density)) / DENSITY_SCALE; + total_density = max(0.0, total_density); + +#ifdef MOLTENVK_USED + uint scattering_u = light_only_map[lpos]; +#else + uint scattering_u = imageLoad(light_only_map, pos).x; +#endif + vec3 scattering = vec3(scattering_u >> 21, (scattering_u << 11) >> 21, scattering_u % 1024) / vec3(2047.0, 2047.0, 1023.0); + scattering += params.base_scattering * params.base_density; + +#ifdef MOLTENVK_USED + uint emission_u = emissive_only_map[lpos]; +#else + uint emission_u = imageLoad(emissive_only_map, pos).x; +#endif + vec3 emission = vec3(emission_u >> 21, (emission_u << 11) >> 21, emission_u % 1024) / vec3(511.0, 511.0, 255.0); + emission += params.base_emission * params.base_density; + + float cell_depth_size = abs(view_pos.z - get_depth_at_pos(fog_cell_size.z, pos.z + 1)); + //compute directional lights + + if (total_density > 0.001) { + for (uint i = 0; i < params.directional_light_count; i++) { + vec3 shadow_attenuation = vec3(1.0); + + if (directional_lights.data[i].shadow_enabled) { + float depth_z = -view_pos.z; + + vec4 pssm_coord; + vec3 shadow_color = directional_lights.data[i].shadow_color1.rgb; + vec3 light_dir = directional_lights.data[i].direction; + vec4 v = vec4(view_pos, 1.0); + float z_range; + + if (depth_z < directional_lights.data[i].shadow_split_offsets.x) { + pssm_coord = (directional_lights.data[i].shadow_matrix1 * v); + pssm_coord /= pssm_coord.w; + z_range = directional_lights.data[i].shadow_z_range.x; + + } else if (depth_z < directional_lights.data[i].shadow_split_offsets.y) { + pssm_coord = (directional_lights.data[i].shadow_matrix2 * v); + pssm_coord /= pssm_coord.w; + z_range = directional_lights.data[i].shadow_z_range.y; + + } else if (depth_z < directional_lights.data[i].shadow_split_offsets.z) { + pssm_coord = (directional_lights.data[i].shadow_matrix3 * v); + pssm_coord /= pssm_coord.w; + z_range = directional_lights.data[i].shadow_z_range.z; + + } else { + pssm_coord = (directional_lights.data[i].shadow_matrix4 * v); + pssm_coord /= pssm_coord.w; + z_range = directional_lights.data[i].shadow_z_range.w; + } + + float depth = texture(sampler2D(directional_shadow_atlas, linear_sampler), pssm_coord.xy).r; + float shadow = exp(min(0.0, (depth - pssm_coord.z)) * z_range * directional_lights.data[i].shadow_volumetric_fog_fade); + + shadow = mix(shadow, 1.0, smoothstep(directional_lights.data[i].fade_from, directional_lights.data[i].fade_to, view_pos.z)); //done with negative values for performance + + shadow_attenuation = mix(shadow_color, vec3(1.0), shadow); + } + + total_light += shadow_attenuation * directional_lights.data[i].color * directional_lights.data[i].energy * henyey_greenstein(dot(normalize(view_pos), normalize(directional_lights.data[i].direction)), params.phase_g); + } + + // Compute light from sky + if (params.ambient_inject > 0.0) { + vec3 isotropic = vec3(0.0); + vec3 anisotropic = vec3(0.0); + if (params.sky_contribution > 0.0) { + float mip_bias = 2.0 + total_density * (MAX_SKY_LOD - 2.0); // Not physically based, but looks nice + vec3 scatter_direction = (params.radiance_inverse_xform * normalize(view_pos)) * sign(params.phase_g); +#ifdef USE_RADIANCE_CUBEMAP_ARRAY + isotropic = texture(samplerCubeArray(sky_texture, linear_sampler_with_mipmaps), vec4(0.0, 1.0, 0.0, mip_bias)).rgb; + anisotropic = texture(samplerCubeArray(sky_texture, linear_sampler_with_mipmaps), vec4(scatter_direction, mip_bias)).rgb; +#else + isotropic = textureLod(samplerCube(sky_texture, linear_sampler_with_mipmaps), vec3(0.0, 1.0, 0.0), mip_bias).rgb; + anisotropic = textureLod(samplerCube(sky_texture, linear_sampler_with_mipmaps), vec3(scatter_direction), mip_bias).rgb; +#endif //USE_RADIANCE_CUBEMAP_ARRAY + } + + total_light += mix(params.ambient_color, mix(isotropic, anisotropic, abs(params.phase_g)), params.sky_contribution) * params.ambient_inject; + } + + //compute lights from cluster + + { //omni lights + + uint cluster_omni_offset = cluster_offset; + + uint item_min; + uint item_max; + uint item_from; + uint item_to; + + cluster_get_item_range(cluster_omni_offset + params.max_cluster_element_count_div_32 + cluster_z, item_min, item_max, item_from, item_to); + +#ifdef USE_SUBGROUPS + item_from = subgroupBroadcastFirst(subgroupMin(item_from)); + item_to = subgroupBroadcastFirst(subgroupMax(item_to)); +#endif + + for (uint i = item_from; i < item_to; i++) { + uint mask = cluster_buffer.data[cluster_omni_offset + i]; + mask &= cluster_get_range_clip_mask(i, item_min, item_max); +#ifdef USE_SUBGROUPS + uint merged_mask = subgroupBroadcastFirst(subgroupOr(mask)); +#else + uint merged_mask = mask; +#endif + + while (merged_mask != 0) { + uint bit = findMSB(merged_mask); + merged_mask &= ~(1 << bit); +#ifdef USE_SUBGROUPS + if (((1 << bit) & mask) == 0) { //do not process if not originally here + continue; + } +#endif + uint light_index = 32 * i + bit; + + //if (!bool(omni_omni_lights.data[light_index].mask & draw_call.layer_mask)) { + // continue; //not masked + //} + + vec3 light_pos = omni_lights.data[light_index].position; + float d = distance(omni_lights.data[light_index].position, view_pos); + float shadow_attenuation = 1.0; + + if (d * omni_lights.data[light_index].inv_radius < 1.0) { + float attenuation = get_omni_attenuation(d, omni_lights.data[light_index].inv_radius, omni_lights.data[light_index].attenuation); + + vec3 light = omni_lights.data[light_index].color; + + if (omni_lights.data[light_index].shadow_enabled) { + //has shadow + vec4 uv_rect = omni_lights.data[light_index].atlas_rect; + vec2 flip_offset = omni_lights.data[light_index].direction.xy; + + vec3 local_vert = (omni_lights.data[light_index].shadow_matrix * vec4(view_pos, 1.0)).xyz; + + float shadow_len = length(local_vert); //need to remember shadow len from here + vec3 shadow_sample = normalize(local_vert); + + if (shadow_sample.z >= 0.0) { + uv_rect.xy += flip_offset; + } + + shadow_sample.z = 1.0 + abs(shadow_sample.z); + vec3 pos = vec3(shadow_sample.xy / shadow_sample.z, shadow_len - omni_lights.data[light_index].shadow_bias); + pos.z *= omni_lights.data[light_index].inv_radius; + + pos.xy = pos.xy * 0.5 + 0.5; + pos.xy = uv_rect.xy + pos.xy * uv_rect.zw; + + float depth = texture(sampler2D(shadow_atlas, linear_sampler), pos.xy).r; + + shadow_attenuation = exp(min(0.0, (depth - pos.z)) / omni_lights.data[light_index].inv_radius * omni_lights.data[light_index].shadow_volumetric_fog_fade); + } + total_light += light * attenuation * shadow_attenuation * henyey_greenstein(dot(normalize(light_pos - view_pos), normalize(view_pos)), params.phase_g); + } + } + } + } + + { //spot lights + + uint cluster_spot_offset = cluster_offset + params.cluster_type_size; + + uint item_min; + uint item_max; + uint item_from; + uint item_to; + + cluster_get_item_range(cluster_spot_offset + params.max_cluster_element_count_div_32 + cluster_z, item_min, item_max, item_from, item_to); + +#ifdef USE_SUBGROUPS + item_from = subgroupBroadcastFirst(subgroupMin(item_from)); + item_to = subgroupBroadcastFirst(subgroupMax(item_to)); +#endif + + for (uint i = item_from; i < item_to; i++) { + uint mask = cluster_buffer.data[cluster_spot_offset + i]; + mask &= cluster_get_range_clip_mask(i, item_min, item_max); +#ifdef USE_SUBGROUPS + uint merged_mask = subgroupBroadcastFirst(subgroupOr(mask)); +#else + uint merged_mask = mask; +#endif + + while (merged_mask != 0) { + uint bit = findMSB(merged_mask); + merged_mask &= ~(1 << bit); +#ifdef USE_SUBGROUPS + if (((1 << bit) & mask) == 0) { //do not process if not originally here + continue; + } +#endif + + //if (!bool(omni_lights.data[light_index].mask & draw_call.layer_mask)) { + // continue; //not masked + //} + + uint light_index = 32 * i + bit; + + vec3 light_pos = spot_lights.data[light_index].position; + vec3 light_rel_vec = spot_lights.data[light_index].position - view_pos; + float d = length(light_rel_vec); + float shadow_attenuation = 1.0; + + if (d * spot_lights.data[light_index].inv_radius < 1.0) { + float attenuation = get_omni_attenuation(d, spot_lights.data[light_index].inv_radius, spot_lights.data[light_index].attenuation); + + vec3 spot_dir = spot_lights.data[light_index].direction; + float scos = max(dot(-normalize(light_rel_vec), spot_dir), spot_lights.data[light_index].cone_angle); + float spot_rim = max(0.0001, (1.0 - scos) / (1.0 - spot_lights.data[light_index].cone_angle)); + attenuation *= 1.0 - pow(spot_rim, spot_lights.data[light_index].cone_attenuation); + + vec3 light = spot_lights.data[light_index].color; + + if (spot_lights.data[light_index].shadow_enabled) { + //has shadow + vec4 uv_rect = spot_lights.data[light_index].atlas_rect; + vec2 flip_offset = spot_lights.data[light_index].direction.xy; + + vec3 local_vert = (spot_lights.data[light_index].shadow_matrix * vec4(view_pos, 1.0)).xyz; + + float shadow_len = length(local_vert); //need to remember shadow len from here + vec3 shadow_sample = normalize(local_vert); + + if (shadow_sample.z >= 0.0) { + uv_rect.xy += flip_offset; + } + + shadow_sample.z = 1.0 + abs(shadow_sample.z); + vec3 pos = vec3(shadow_sample.xy / shadow_sample.z, shadow_len - spot_lights.data[light_index].shadow_bias); + pos.z *= spot_lights.data[light_index].inv_radius; + + pos.xy = pos.xy * 0.5 + 0.5; + pos.xy = uv_rect.xy + pos.xy * uv_rect.zw; + + float depth = texture(sampler2D(shadow_atlas, linear_sampler), pos.xy).r; + + shadow_attenuation = exp(min(0.0, (depth - pos.z)) / spot_lights.data[light_index].inv_radius * spot_lights.data[light_index].shadow_volumetric_fog_fade); + } + total_light += light * attenuation * shadow_attenuation * henyey_greenstein(dot(normalize(light_rel_vec), normalize(view_pos)), params.phase_g); + } + } + } + } + + vec3 world_pos = mat3(params.cam_rotation) * view_pos; + + for (uint i = 0; i < params.max_voxel_gi_instances; i++) { + vec3 position = (voxel_gi_instances.data[i].xform * vec4(world_pos, 1.0)).xyz; + + //this causes corrupted pixels, i have no idea why.. + if (all(bvec2(all(greaterThanEqual(position, vec3(0.0))), all(lessThan(position, voxel_gi_instances.data[i].bounds))))) { + position /= voxel_gi_instances.data[i].bounds; + + vec4 light = vec4(0.0); + for (uint j = 0; j < voxel_gi_instances.data[i].mipmaps; j++) { + vec4 slight = textureLod(sampler3D(voxel_gi_textures[i], linear_sampler_with_mipmaps), position, float(j)); + float a = (1.0 - light.a); + light += a * slight; + } + + light.rgb *= voxel_gi_instances.data[i].dynamic_range * params.gi_inject; + + total_light += light.rgb; + } + } + + //sdfgi +#ifdef ENABLE_SDFGI + + { + float blend = -1.0; + vec3 ambient_total = vec3(0.0); + + for (uint i = 0; i < sdfgi.max_cascades; i++) { + vec3 cascade_pos = (world_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 + } + + vec3 base_pos = floor(cascade_pos); + ivec3 probe_base_pos = ivec3(base_pos); + + vec4 ambient_accum = vec4(0.0); + + ivec3 tex_pos = ivec3(probe_base_pos.xy, int(i)); + tex_pos.x += probe_base_pos.z * sdfgi.probe_axis_size; + + 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 trilinear = vec3(1.0) - abs(probe_to_pos); + float weight = trilinear.x * trilinear.y * trilinear.z; + + // Compute lightprobe occlusion + + if (sdfgi.use_occlusion) { + ivec3 occ_indexv = abs((sdfgi.cascades[i].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(i); + 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_texture, linear_sampler), occ_pos, 0.0), occ_mask); + + weight *= max(occlusion, 0.01); + } + + // Compute ambient texture position + + ivec3 uvw = tex_pos; + uvw.xy += offset.xy; + uvw.x += offset.z * sdfgi.probe_axis_size; + + vec3 ambient = texelFetch(sampler2DArray(sdfgi_ambient_texture, linear_sampler), uvw, 0).rgb; + + ambient_accum.rgb += ambient * weight; + ambient_accum.a += weight; + } + + if (ambient_accum.a > 0) { + ambient_accum.rgb /= ambient_accum.a; + } + ambient_total = ambient_accum.rgb; + break; + } + + total_light += ambient_total * params.gi_inject; + } + +#endif + } + + vec4 final_density = vec4(total_light * scattering + emission, total_density); + + final_density = mix(final_density, reprojected_density, reproject_amount); + + imageStore(density_map, pos, final_density); +#ifdef MOLTENVK_USED + density_only_map[lpos] = 0; + light_only_map[lpos] = 0; + emissive_only_map[lpos] = 0; +#else + imageStore(density_only_map, pos, uvec4(0)); + imageStore(light_only_map, pos, uvec4(0)); + imageStore(emissive_only_map, pos, uvec4(0)); +#endif +#endif + +#ifdef MODE_FOG + + ivec3 pos = ivec3(gl_GlobalInvocationID.xy, 0); + + if (any(greaterThanEqual(pos, params.fog_volume_size))) { + return; //do not compute + } + + vec4 fog_accum = vec4(0.0, 0.0, 0.0, 1.0); + float prev_z = 0.0; + + for (int i = 0; i < params.fog_volume_size.z; i++) { + //compute fog position + ivec3 fog_pos = pos + ivec3(0, 0, i); + //get fog value + vec4 fog = imageLoad(density_map, fog_pos); + + //get depth at cell pos + float z = get_depth_at_pos(fog_cell_size.z, i); + //get distance from previous pos + float d = abs(prev_z - z); + //compute transmittance using beer's law + float transmittance = exp(-d * fog.a); + + fog_accum.rgb += ((fog.rgb - fog.rgb * transmittance) / max(fog.a, 0.00001)) * fog_accum.a; + fog_accum.a *= transmittance; + + prev_z = z; + + imageStore(fog_map, fog_pos, vec4(fog_accum.rgb, 1.0 - fog_accum.a)); + } + +#endif + +#ifdef MODE_FILTER + + ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); + + const float gauss[7] = float[](0.071303, 0.131514, 0.189879, 0.214607, 0.189879, 0.131514, 0.071303); + + const ivec3 filter_dir[3] = ivec3[](ivec3(1, 0, 0), ivec3(0, 1, 0), ivec3(0, 0, 1)); + ivec3 offset = filter_dir[params.filter_axis]; + + vec4 accum = vec4(0.0); + for (int i = -3; i <= 3; i++) { + accum += imageLoad(source_map, clamp(pos + offset * i, ivec3(0), params.fog_volume_size - ivec3(1))) * gauss[i + 3]; + } + + imageStore(dest_map, pos, accum); + +#endif +#ifdef MODE_COPY + ivec3 pos = ivec3(gl_GlobalInvocationID.xyz); + if (any(greaterThanEqual(pos, params.fog_volume_size))) { + return; //do not compute + } + + imageStore(dest_map, pos, imageLoad(source_map, pos)); + +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/voxel_gi.glsl b/servers/rendering/renderer_rd/shaders/voxel_gi.glsl index 779f04ed35..73a97d9df1 100644 --- a/servers/rendering/renderer_rd/shaders/voxel_gi.glsl +++ b/servers/rendering/renderer_rd/shaders/voxel_gi.glsl @@ -13,7 +13,6 @@ layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in; #ifndef MODE_DYNAMIC #define NO_CHILDREN 0xFFFFFFFF -#define GREY_VEC vec3(0.33333, 0.33333, 0.33333) struct CellChildren { uint children[8]; diff --git a/servers/rendering/renderer_rd/shaders/voxel_gi_debug.glsl b/servers/rendering/renderer_rd/shaders/voxel_gi_debug.glsl index 281c496df3..3f3437f527 100644 --- a/servers/rendering/renderer_rd/shaders/voxel_gi_debug.glsl +++ b/servers/rendering/renderer_rd/shaders/voxel_gi_debug.glsl @@ -90,66 +90,10 @@ void main() { #endif #ifdef MODE_DEBUG_LIGHT - -#ifdef USE_ANISOTROPY - -#define POS_X 0 -#define POS_Y 1 -#define POS_Z 2 -#define NEG_X 3 -#define NEG_Y 4 -#define NEG_Z 5 - - const uint triangle_aniso[12] = uint[]( - NEG_X, - NEG_Z, - NEG_Y, - NEG_Z, - NEG_X, - NEG_Y, - POS_Z, - POS_X, - POS_X, - POS_Y, - POS_Y, - POS_Z); - - color_interp.xyz = texelFetch(sampler3D(color_tex, tex_sampler), ivec3(posu), int(params.level)).xyz * params.dynamic_range; - vec3 aniso_pos = texelFetch(sampler3D(aniso_pos_tex, tex_sampler), ivec3(posu), int(params.level)).xyz; - vec3 aniso_neg = texelFetch(sampler3D(aniso_neg_tex, tex_sampler), ivec3(posu), int(params.level)).xyz; - uint side = triangle_aniso[gl_VertexIndex / 3]; - - float strength = 0.0; - switch (side) { - case POS_X: - strength = aniso_pos.x; - break; - case POS_Y: - strength = aniso_pos.y; - break; - case POS_Z: - strength = aniso_pos.z; - break; - case NEG_X: - strength = aniso_neg.x; - break; - case NEG_Y: - strength = aniso_neg.y; - break; - case NEG_Z: - strength = aniso_neg.z; - break; - } - - color_interp.xyz *= strength; - -#else color_interp = texelFetch(sampler3D(color_tex, tex_sampler), ivec3(posu), int(params.level)); color_interp.xyz *params.dynamic_range; - #endif -#endif float scale = (1 << params.level); gl_Position = params.projection * vec4((vec3(posu) + vertex) * scale, 1.0); diff --git a/servers/rendering/renderer_rd/shaders/voxel_gi_sdf.glsl b/servers/rendering/renderer_rd/shaders/voxel_gi_sdf.glsl index e20b3f680d..3bb4421646 100644 --- a/servers/rendering/renderer_rd/shaders/voxel_gi_sdf.glsl +++ b/servers/rendering/renderer_rd/shaders/voxel_gi_sdf.glsl @@ -6,10 +6,9 @@ layout(local_size_x = 4, local_size_y = 4, local_size_z = 4) in; -#define MAX_DISTANCE 100000 +#define MAX_DISTANCE 100000.0 #define NO_CHILDREN 0xFFFFFFFF -#define GREY_VEC vec3(0.33333, 0.33333, 0.33333) struct CellChildren { uint children[8]; @@ -44,7 +43,7 @@ params; void main() { vec3 pos = vec3(gl_GlobalInvocationID); - float closest_dist = 100000.0; + float closest_dist = MAX_DISTANCE; for (uint i = params.offset; i < params.end; i++) { vec3 posu = vec3(uvec3(cell_data.data[i].position & 0x7FF, (cell_data.data[i].position >> 11) & 0x3FF, cell_data.data[i].position >> 21)); |