diff options
Diffstat (limited to 'servers/rendering/renderer_rd/shaders')
13 files changed, 233 insertions, 72 deletions
diff --git a/servers/rendering/renderer_rd/shaders/canvas.glsl b/servers/rendering/renderer_rd/shaders/canvas.glsl index 459d798a80..8593e6b265 100644 --- a/servers/rendering/renderer_rd/shaders/canvas.glsl +++ b/servers/rendering/renderer_rd/shaders/canvas.glsl @@ -313,6 +313,14 @@ vec4 light_compute( vec2 uv, vec4 color, bool is_directional) { vec4 light = vec4(0.0); + vec3 light_direction = vec3(0.0); + + if (is_directional) { + light_direction = normalize(mix(vec3(light_position.xy, 0.0), vec3(0, 0, 1), light_position.z)); + light_position = vec3(0.0); + } else { + light_direction = normalize(light_position - light_vertex); + } #CODE : LIGHT @@ -590,14 +598,14 @@ void main() { normal = normalize((canvas_data.canvas_normal_transform * vec4(normal, 0.0)).xyz); } - vec3 base_color = color.rgb; + vec4 base_color = color; if (bool(draw_data.flags & FLAGS_USING_LIGHT_MASK)) { color = vec4(0.0); //invisible by default due to using light mask } #ifdef MODE_LIGHT_ONLY color = vec4(0.0); -#else +#elif !defined(MODE_UNSHADED) color *= canvas_data.canvas_modulation; #endif @@ -614,12 +622,14 @@ void main() { #ifdef LIGHT_CODE_USED vec4 shadow_modulate = vec4(1.0); - light_color = light_compute(light_vertex, vec3(direction, light_array.data[light_base].height), normal, light_color, light_color.a, specular_shininess, shadow_modulate, screen_uv, uv, color, true); + light_color = light_compute(light_vertex, vec3(direction, light_array.data[light_base].height), normal, light_color, light_color.a, specular_shininess, shadow_modulate, screen_uv, uv, base_color, true); #else if (normal_used) { vec3 light_vec = normalize(mix(vec3(direction, 0.0), vec3(0, 0, 1), light_array.data[light_base].height)); - light_color.rgb = light_normal_compute(light_vec, normal, base_color, light_color.rgb, specular_shininess, specular_shininess_used); + light_color.rgb = light_normal_compute(light_vec, normal, base_color.rgb, light_color.rgb, specular_shininess, specular_shininess_used); + } else { + light_color.rgb *= base_color.rgb; } #endif @@ -645,20 +655,7 @@ void main() { if (i >= light_count) { break; } - uint light_base; - if (i < 8) { - if (i < 4) { - light_base = draw_data.lights[0]; - } else { - light_base = draw_data.lights[1]; - } - } else { - if (i < 12) { - light_base = draw_data.lights[2]; - } else { - light_base = draw_data.lights[3]; - } - } + uint light_base = draw_data.lights[i >> 2]; light_base >>= (i & 3) * 8; light_base &= 0xFF; @@ -673,7 +670,7 @@ void main() { vec3 light_position = vec3(light_array.data[light_base].position, light_array.data[light_base].height); light_color.rgb *= light_base_color.rgb; - light_color = light_compute(light_vertex, light_position, normal, light_color, light_base_color.a, specular_shininess, shadow_modulate, screen_uv, uv, color, false); + light_color = light_compute(light_vertex, light_position, normal, light_color, light_base_color.a, specular_shininess, shadow_modulate, screen_uv, uv, base_color, false); #else light_color.rgb *= light_base_color.rgb * light_base_color.a; @@ -682,9 +679,10 @@ void main() { vec3 light_pos = vec3(light_array.data[light_base].position, light_array.data[light_base].height); vec3 pos = light_vertex; vec3 light_vec = normalize(light_pos - pos); - float cNdotL = max(0.0, dot(normal, light_vec)); - light_color.rgb = light_normal_compute(light_vec, normal, base_color, light_color.rgb, specular_shininess, specular_shininess_used); + light_color.rgb = light_normal_compute(light_vec, normal, base_color.rgb, light_color.rgb, specular_shininess, specular_shininess_used); + } else { + light_color.rgb *= base_color.rgb; } #endif if (any(lessThan(tex_uv, vec2(0.0, 0.0))) || any(greaterThanEqual(tex_uv, vec2(1.0, 1.0)))) { diff --git a/servers/rendering/renderer_rd/shaders/effects/SCsub b/servers/rendering/renderer_rd/shaders/effects/SCsub index 741da8fe69..f06a2d86e2 100644 --- a/servers/rendering/renderer_rd/shaders/effects/SCsub +++ b/servers/rendering/renderer_rd/shaders/effects/SCsub @@ -4,7 +4,7 @@ Import("env") if "RD_GLSL" in env["BUILDERS"]: # find all include files - gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + [str(f) for f in Glob("../*_inc.glsl")] # find all shader code(all glsl files excluding our include files) glsl_files = [str(f) for f in Glob("*.glsl") if str(f) not in gl_include_files] diff --git a/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection.glsl b/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection.glsl index 9f86643e52..9d8294a7da 100644 --- a/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection.glsl @@ -121,7 +121,7 @@ void main() { // clip z and w advance to line advance vec2 line_advance = normalize(line_dir); // down to pixel - float step_size = length(line_advance) / length(line_dir); + float step_size = 1.0 / length(line_dir); float z_advance = z_dir * step_size; // adapt z advance to line advance float w_advance = w_dir * step_size; // adapt w advance to line advance @@ -139,6 +139,14 @@ void main() { float depth; vec2 prev_pos = pos; + if (ivec2(pos + line_advance - 0.5) == ssC) { + // It is possible for rounding to cause our first pixel to check to be the pixel we're reflecting. + // Make sure we skip it + pos += line_advance; + z += z_advance; + w += w_advance; + } + bool found = false; float steps_taken = 0.0; @@ -149,8 +157,8 @@ void main() { w += w_advance; // convert to linear depth - - depth = imageLoad(source_depth, ivec2(pos - 0.5)).r; + ivec2 test_pos = ivec2(pos - 0.5); + depth = imageLoad(source_depth, test_pos).r; if (sc_multiview) { depth = depth * 2.0 - 1.0; depth = 2.0 * params.camera_z_near * params.camera_z_far / (params.camera_z_far + params.camera_z_near - depth * (params.camera_z_far - params.camera_z_near)); @@ -161,13 +169,21 @@ void main() { z_to = z / w; if (depth > z_to) { - // if depth was surpassed - if (depth <= max(z_to, z_from) + params.depth_tolerance && -depth < params.camera_z_far * 0.95) { - // check the depth tolerance and far clip - // check that normal is valid - found = true; + // Test if our ray is hitting the "right" side of the surface, if not we're likely self reflecting and should skip. + vec4 test_normal_roughness = imageLoad(source_normal_roughness, test_pos); + vec3 test_normal = test_normal_roughness.xyz * 2.0 - 1.0; + test_normal = normalize(test_normal); + test_normal.y = -test_normal.y; //because this code reads flipped + + if (dot(ray_dir, test_normal) < 0.001) { + // if depth was surpassed + if (depth <= max(z_to, z_from) + params.depth_tolerance && -depth < params.camera_z_far * 0.95) { + // check the depth tolerance and far clip + // check that normal is valid + found = true; + } + break; } - break; } steps_taken += 1.0; diff --git a/servers/rendering/renderer_rd/shaders/effects/tonemap.glsl b/servers/rendering/renderer_rd/shaders/effects/tonemap.glsl index e459756c6a..52aee8b648 100644 --- a/servers/rendering/renderer_rd/shaders/effects/tonemap.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/tonemap.glsl @@ -360,15 +360,15 @@ vec3 do_fxaa(vec3 color, float exposure, vec2 uv_interp) { const float FXAA_SPAN_MAX = 8.0; #ifdef MULTIVIEW - vec3 rgbNW = textureLod(source_color, vec3(uv_interp + vec2(-1.0, -1.0) * params.pixel_size, ViewIndex), 0.0).xyz * exposure * params.luminance_multiplier; - vec3 rgbNE = textureLod(source_color, vec3(uv_interp + vec2(1.0, -1.0) * params.pixel_size, ViewIndex), 0.0).xyz * exposure * params.luminance_multiplier; - vec3 rgbSW = textureLod(source_color, vec3(uv_interp + vec2(-1.0, 1.0) * params.pixel_size, ViewIndex), 0.0).xyz * exposure * params.luminance_multiplier; - vec3 rgbSE = textureLod(source_color, vec3(uv_interp + vec2(1.0, 1.0) * params.pixel_size, ViewIndex), 0.0).xyz * exposure * params.luminance_multiplier; + vec3 rgbNW = textureLod(source_color, vec3(uv_interp + vec2(-0.5, -0.5) * params.pixel_size, ViewIndex), 0.0).xyz * exposure * params.luminance_multiplier; + vec3 rgbNE = textureLod(source_color, vec3(uv_interp + vec2(0.5, -0.5) * params.pixel_size, ViewIndex), 0.0).xyz * exposure * params.luminance_multiplier; + vec3 rgbSW = textureLod(source_color, vec3(uv_interp + vec2(-0.5, 0.5) * params.pixel_size, ViewIndex), 0.0).xyz * exposure * params.luminance_multiplier; + vec3 rgbSE = textureLod(source_color, vec3(uv_interp + vec2(0.5, 0.5) * params.pixel_size, ViewIndex), 0.0).xyz * exposure * params.luminance_multiplier; #else - vec3 rgbNW = textureLod(source_color, uv_interp + vec2(-1.0, -1.0) * params.pixel_size, 0.0).xyz * exposure * params.luminance_multiplier; - vec3 rgbNE = textureLod(source_color, uv_interp + vec2(1.0, -1.0) * params.pixel_size, 0.0).xyz * exposure * params.luminance_multiplier; - vec3 rgbSW = textureLod(source_color, uv_interp + vec2(-1.0, 1.0) * params.pixel_size, 0.0).xyz * exposure * params.luminance_multiplier; - vec3 rgbSE = textureLod(source_color, uv_interp + vec2(1.0, 1.0) * params.pixel_size, 0.0).xyz * exposure * params.luminance_multiplier; + vec3 rgbNW = textureLod(source_color, uv_interp + vec2(-0.5, -0.5) * params.pixel_size, 0.0).xyz * exposure * params.luminance_multiplier; + vec3 rgbNE = textureLod(source_color, uv_interp + vec2(0.5, -0.5) * params.pixel_size, 0.0).xyz * exposure * params.luminance_multiplier; + vec3 rgbSW = textureLod(source_color, uv_interp + vec2(-0.5, 0.5) * params.pixel_size, 0.0).xyz * exposure * params.luminance_multiplier; + vec3 rgbSE = textureLod(source_color, uv_interp + vec2(0.5, 0.5) * params.pixel_size, 0.0).xyz * exposure * params.luminance_multiplier; #endif vec3 rgbM = color; vec3 luma = vec3(0.299, 0.587, 0.114); @@ -462,12 +462,6 @@ void main() { } #endif - if (params.use_debanding) { - // For best results, debanding should be done before tonemapping. - // Otherwise, we're adding noise to an already-quantized image. - color.rgb += screen_space_dither(gl_FragCoord.xy); - } - color.rgb = apply_tonemapping(color.rgb, params.white); color.rgb = linear_to_srgb(color.rgb); // regular linear -> SRGB conversion @@ -498,5 +492,11 @@ void main() { color.rgb = apply_color_correction(color.rgb); } + if (params.use_debanding) { + // Debanding should be done at the end of tonemapping, but before writing to the LDR buffer. + // Otherwise, we're adding noise to an already-quantized image. + color.rgb += screen_space_dither(gl_FragCoord.xy); + } + frag_color = color; } diff --git a/servers/rendering/renderer_rd/shaders/environment/SCsub b/servers/rendering/renderer_rd/shaders/environment/SCsub index 741da8fe69..f06a2d86e2 100644 --- a/servers/rendering/renderer_rd/shaders/environment/SCsub +++ b/servers/rendering/renderer_rd/shaders/environment/SCsub @@ -4,7 +4,7 @@ Import("env") if "RD_GLSL" in env["BUILDERS"]: # find all include files - gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + [str(f) for f in Glob("../*_inc.glsl")] # find all shader code(all glsl files excluding our include files) glsl_files = [str(f) for f in Glob("*.glsl") if str(f) not in gl_include_files] diff --git a/servers/rendering/renderer_rd/shaders/environment/volumetric_fog_process.glsl b/servers/rendering/renderer_rd/shaders/environment/volumetric_fog_process.glsl index eed9038502..28507e6c12 100644 --- a/servers/rendering/renderer_rd/shaders/environment/volumetric_fog_process.glsl +++ b/servers/rendering/renderer_rd/shaders/environment/volumetric_fog_process.glsl @@ -381,7 +381,7 @@ void main() { 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) { + if (total_density > 0.00005) { for (uint i = 0; i < params.directional_light_count; i++) { if (directional_lights.data[i].volumetric_fog_energy > 0.001) { vec3 shadow_attenuation = vec3(1.0); diff --git a/servers/rendering/renderer_rd/shaders/forward_clustered/SCsub b/servers/rendering/renderer_rd/shaders/forward_clustered/SCsub index 741da8fe69..f06a2d86e2 100644 --- a/servers/rendering/renderer_rd/shaders/forward_clustered/SCsub +++ b/servers/rendering/renderer_rd/shaders/forward_clustered/SCsub @@ -4,7 +4,7 @@ Import("env") if "RD_GLSL" in env["BUILDERS"]: # find all include files - gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + [str(f) for f in Glob("../*_inc.glsl")] # find all shader code(all glsl files excluding our include files) glsl_files = [str(f) for f in Glob("*.glsl") if str(f) not in gl_include_files] diff --git a/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl b/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl index a330fd5497..77ae97e621 100644 --- a/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl +++ b/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl @@ -129,12 +129,52 @@ invariant gl_Position; #GLOBALS +#ifdef USE_DOUBLE_PRECISION +// Helper functions for emulating double precision when adding floats. +vec3 quick_two_sum(vec3 a, vec3 b, out vec3 out_p) { + vec3 s = a + b; + out_p = b - (s - a); + return s; +} + +vec3 two_sum(vec3 a, vec3 b, out vec3 out_p) { + vec3 s = a + b; + vec3 v = s - a; + out_p = (a - (s - v)) + (b - v); + return s; +} + +vec3 double_add_vec3(vec3 base_a, vec3 prec_a, vec3 base_b, vec3 prec_b, out vec3 out_precision) { + vec3 s, t, se, te; + s = two_sum(base_a, base_b, se); + t = two_sum(prec_a, prec_b, te); + se += t; + s = quick_two_sum(s, se, se); + se += te; + s = quick_two_sum(s, se, out_precision); + return s; +} +#endif + void vertex_shader(in uint instance_index, in bool is_multimesh, in uint multimesh_offset, in SceneData scene_data, in mat4 model_matrix, out vec4 screen_pos) { vec4 instance_custom = vec4(0.0); #if defined(COLOR_USED) color_interp = color_attrib; #endif + mat4 inv_view_matrix = scene_data.inv_view_matrix; + +#ifdef USE_DOUBLE_PRECISION + vec3 model_precision = vec3(model_matrix[0][3], model_matrix[1][3], model_matrix[2][3]); + model_matrix[0][3] = 0.0; + model_matrix[1][3] = 0.0; + model_matrix[2][3] = 0.0; + vec3 view_precision = vec3(inv_view_matrix[0][3], inv_view_matrix[1][3], inv_view_matrix[2][3]); + inv_view_matrix[0][3] = 0.0; + inv_view_matrix[1][3] = 0.0; + inv_view_matrix[2][3] = 0.0; +#endif + mat3 model_normal_matrix; if (bool(instances.data[instance_index].flags & INSTANCE_FLAGS_NON_UNIFORM_SCALE)) { model_normal_matrix = transpose(inverse(mat3(model_matrix))); @@ -142,11 +182,12 @@ void vertex_shader(in uint instance_index, in bool is_multimesh, in uint multime model_normal_matrix = mat3(model_matrix); } + mat4 matrix; + mat4 read_model_matrix = model_matrix; + if (is_multimesh) { //multimesh, instances are for it - mat4 matrix; - #ifdef USE_PARTICLE_TRAILS uint trail_size = (instances.data[instance_index].flags >> INSTANCE_FLAGS_PARTICLE_TRAIL_SHIFT) & INSTANCE_FLAGS_PARTICLE_TRAIL_MASK; uint stride = 3 + 1 + 1; //particles always uses this format @@ -232,7 +273,14 @@ void vertex_shader(in uint instance_index, in bool is_multimesh, in uint multime #endif //transpose matrix = transpose(matrix); - model_matrix = model_matrix * matrix; +#if !defined(USE_DOUBLE_PRECISION) || defined(SKIP_TRANSFORM_USED) || defined(VERTEX_WORLD_COORDS_USED) || defined(MODEL_MATRIX_USED) + // Normally we can bake the multimesh transform into the model matrix, but when using double precision + // we avoid baking it in so we can emulate high precision. + read_model_matrix = model_matrix * matrix; +#if !defined(USE_DOUBLE_PRECISION) || defined(SKIP_TRANSFORM_USED) || defined(VERTEX_WORLD_COORDS_USED) + model_matrix = read_model_matrix; +#endif // !defined(USE_DOUBLE_PRECISION) || defined(SKIP_TRANSFORM_USED) || defined(VERTEX_WORLD_COORDS_USED) +#endif // !defined(USE_DOUBLE_PRECISION) || defined(SKIP_TRANSFORM_USED) || defined(VERTEX_WORLD_COORDS_USED) || defined(MODEL_MATRIX_USED) model_normal_matrix = model_normal_matrix * mat3(matrix); } @@ -297,7 +345,22 @@ void vertex_shader(in uint instance_index, in bool is_multimesh, in uint multime // using local coordinates (default) #if !defined(SKIP_TRANSFORM_USED) && !defined(VERTEX_WORLD_COORDS_USED) +#ifdef USE_DOUBLE_PRECISION + // We separate the basis from the origin because the basis is fine with single point precision. + // Then we combine the translations from the model matrix and the view matrix using emulated doubles. + // We add the result to the vertex and ignore the final lost precision. + vec3 model_origin = model_matrix[3].xyz; + if (is_multimesh) { + vertex = mat3(matrix) * vertex; + model_origin = double_add_vec3(model_origin, model_precision, matrix[3].xyz, vec3(0.0), model_precision); + } + vertex = mat3(model_matrix) * vertex; + vec3 temp_precision; // Will be ignored. + vertex += double_add_vec3(model_origin, model_precision, scene_data.inv_view_matrix[3].xyz, view_precision, temp_precision); + vertex = mat3(scene_data.view_matrix) * vertex; +#else vertex = (modelview * vec4(vertex, 1.0)).xyz; +#endif #ifdef NORMAL_USED normal = modelview_normal * normal; #endif @@ -490,7 +553,6 @@ layout(location = 10) in flat uint instance_index_interp; //defines to keep compatibility with vertex -#define model_matrix instances.data[draw_call.instance_index].transform #ifdef USE_MULTIVIEW #define projection_matrix scene_data.projection_matrix_view[ViewIndex] #define inv_projection_matrix scene_data.inv_projection_matrix_view[ViewIndex] @@ -737,6 +799,17 @@ void fragment_shader(in SceneData scene_data) { vec2 alpha_texture_coordinate = vec2(0.0, 0.0); #endif // ALPHA_ANTIALIASING_EDGE_USED + mat4 inv_view_matrix = scene_data.inv_view_matrix; + mat4 read_model_matrix = instances.data[instance_index].transform; +#ifdef USE_DOUBLE_PRECISION + read_model_matrix[0][3] = 0.0; + read_model_matrix[1][3] = 0.0; + read_model_matrix[2][3] = 0.0; + inv_view_matrix[0][3] = 0.0; + inv_view_matrix[1][3] = 0.0; + inv_view_matrix[2][3] = 0.0; +#endif + { #CODE : FRAGMENT } diff --git a/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered_inc.glsl b/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered_inc.glsl index e8e2dce990..0d36b98645 100644 --- a/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered_inc.glsl @@ -62,13 +62,14 @@ layout(set = 0, binding = 3) uniform sampler decal_sampler; layout(set = 0, binding = 4) uniform sampler light_projector_sampler; -#define INSTANCE_FLAGS_NON_UNIFORM_SCALE (1 << 5) -#define INSTANCE_FLAGS_USE_GI_BUFFERS (1 << 6) -#define INSTANCE_FLAGS_USE_SDFGI (1 << 7) -#define INSTANCE_FLAGS_USE_LIGHTMAP_CAPTURE (1 << 8) -#define INSTANCE_FLAGS_USE_LIGHTMAP (1 << 9) -#define INSTANCE_FLAGS_USE_SH_LIGHTMAP (1 << 10) -#define INSTANCE_FLAGS_USE_VOXEL_GI (1 << 11) +#define INSTANCE_FLAGS_NON_UNIFORM_SCALE (1 << 4) +#define INSTANCE_FLAGS_USE_GI_BUFFERS (1 << 5) +#define INSTANCE_FLAGS_USE_SDFGI (1 << 6) +#define INSTANCE_FLAGS_USE_LIGHTMAP_CAPTURE (1 << 7) +#define INSTANCE_FLAGS_USE_LIGHTMAP (1 << 8) +#define INSTANCE_FLAGS_USE_SH_LIGHTMAP (1 << 9) +#define INSTANCE_FLAGS_USE_VOXEL_GI (1 << 10) +#define INSTANCE_FLAGS_PARTICLES (1 << 11) #define INSTANCE_FLAGS_MULTIMESH (1 << 12) #define INSTANCE_FLAGS_MULTIMESH_FORMAT_2D (1 << 13) #define INSTANCE_FLAGS_MULTIMESH_HAS_COLOR (1 << 14) diff --git a/servers/rendering/renderer_rd/shaders/forward_mobile/SCsub b/servers/rendering/renderer_rd/shaders/forward_mobile/SCsub index 741da8fe69..f06a2d86e2 100644 --- a/servers/rendering/renderer_rd/shaders/forward_mobile/SCsub +++ b/servers/rendering/renderer_rd/shaders/forward_mobile/SCsub @@ -4,7 +4,7 @@ Import("env") if "RD_GLSL" in env["BUILDERS"]: # find all include files - gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + [str(f) for f in Glob("../*_inc.glsl")] # find all shader code(all glsl files excluding our include files) glsl_files = [str(f) for f in Glob("*.glsl") if str(f) not in gl_include_files] diff --git a/servers/rendering/renderer_rd/shaders/forward_mobile/scene_forward_mobile.glsl b/servers/rendering/renderer_rd/shaders/forward_mobile/scene_forward_mobile.glsl index d42fdc8746..9aeaa6d978 100644 --- a/servers/rendering/renderer_rd/shaders/forward_mobile/scene_forward_mobile.glsl +++ b/servers/rendering/renderer_rd/shaders/forward_mobile/scene_forward_mobile.glsl @@ -123,6 +123,33 @@ invariant gl_Position; #define scene_data scene_data_block.data +#ifdef USE_DOUBLE_PRECISION +// Helper functions for emulating double precision when adding floats. +vec3 quick_two_sum(vec3 a, vec3 b, out vec3 out_p) { + vec3 s = a + b; + out_p = b - (s - a); + return s; +} + +vec3 two_sum(vec3 a, vec3 b, out vec3 out_p) { + vec3 s = a + b; + vec3 v = s - a; + out_p = (a - (s - v)) + (b - v); + return s; +} + +vec3 double_add_vec3(vec3 base_a, vec3 prec_a, vec3 base_b, vec3 prec_b, out vec3 out_precision) { + vec3 s, t, se, te; + s = two_sum(base_a, base_b, se); + t = two_sum(prec_a, prec_b, te); + se += t; + s = quick_two_sum(s, se, se); + se += te; + s = quick_two_sum(s, se, out_precision); + return s; +} +#endif + void main() { vec4 instance_custom = vec4(0.0); #if defined(COLOR_USED) @@ -132,6 +159,17 @@ void main() { bool is_multimesh = bool(draw_call.flags & INSTANCE_FLAGS_MULTIMESH); mat4 model_matrix = draw_call.transform; + mat4 inv_view_matrix = scene_data.inv_view_matrix; +#ifdef USE_DOUBLE_PRECISION + vec3 model_precision = vec3(model_matrix[0][3], model_matrix[1][3], model_matrix[2][3]); + model_matrix[0][3] = 0.0; + model_matrix[1][3] = 0.0; + model_matrix[2][3] = 0.0; + vec3 view_precision = vec3(inv_view_matrix[0][3], inv_view_matrix[1][3], inv_view_matrix[2][3]); + inv_view_matrix[0][3] = 0.0; + inv_view_matrix[1][3] = 0.0; + inv_view_matrix[2][3] = 0.0; +#endif mat3 model_normal_matrix; if (bool(draw_call.flags & INSTANCE_FLAGS_NON_UNIFORM_SCALE)) { @@ -140,11 +178,12 @@ void main() { model_normal_matrix = mat3(model_matrix); } + mat4 matrix; + mat4 read_model_matrix = model_matrix; + if (is_multimesh) { //multimesh, instances are for it - mat4 matrix; - #ifdef USE_PARTICLE_TRAILS uint trail_size = (draw_call.flags >> INSTANCE_FLAGS_PARTICLE_TRAIL_SHIFT) & INSTANCE_FLAGS_PARTICLE_TRAIL_MASK; uint stride = 3 + 1 + 1; //particles always uses this format @@ -230,7 +269,15 @@ void main() { #endif //transpose matrix = transpose(matrix); - model_matrix = model_matrix * matrix; + +#if !defined(USE_DOUBLE_PRECISION) || defined(SKIP_TRANSFORM_USED) || defined(VERTEX_WORLD_COORDS_USED) || defined(MODEL_MATRIX_USED) + // Normally we can bake the multimesh transform into the model matrix, but when using double precision + // we avoid baking it in so we can emulate high precision. + read_model_matrix = model_matrix * matrix; +#if !defined(USE_DOUBLE_PRECISION) || defined(SKIP_TRANSFORM_USED) || defined(VERTEX_WORLD_COORDS_USED) + model_matrix = read_model_matrix; +#endif // !defined(USE_DOUBLE_PRECISION) || defined(SKIP_TRANSFORM_USED) || defined(VERTEX_WORLD_COORDS_USED) +#endif // !defined(USE_DOUBLE_PRECISION) || defined(SKIP_TRANSFORM_USED) || defined(VERTEX_WORLD_COORDS_USED) || defined(MODEL_MATRIX_USED) model_normal_matrix = model_normal_matrix * mat3(matrix); } @@ -297,7 +344,22 @@ void main() { // using local coordinates (default) #if !defined(SKIP_TRANSFORM_USED) && !defined(VERTEX_WORLD_COORDS_USED) +#ifdef USE_DOUBLE_PRECISION + // We separate the basis from the origin because the basis is fine with single point precision. + // Then we combine the translations from the model matrix and the view matrix using emulated doubles. + // We add the result to the vertex and ignore the final lost precision. + vec3 model_origin = model_matrix[3].xyz; + if (is_multimesh) { + vertex = mat3(matrix) * vertex; + model_origin = double_add_vec3(model_origin, model_precision, matrix[3].xyz, vec3(0.0), model_precision); + } + vertex = mat3(model_matrix) * vertex; + vec3 temp_precision; + vertex += double_add_vec3(model_origin, model_precision, scene_data.inv_view_matrix[3].xyz, view_precision, temp_precision); + vertex = mat3(scene_data.view_matrix) * vertex; +#else vertex = (modelview * vec4(vertex, 1.0)).xyz; +#endif #ifdef NORMAL_USED normal = modelview_normal * normal; #endif @@ -468,7 +530,6 @@ layout(location = 9) highp in float dp_clip; //defines to keep compatibility with vertex -#define model_matrix draw_call.transform #ifdef USE_MULTIVIEW #define projection_matrix scene_data.projection_matrix_view[ViewIndex] #define inv_projection_matrix scene_data.inv_projection_matrix_view[ViewIndex] @@ -685,6 +746,17 @@ void main() { vec2 alpha_texture_coordinate = vec2(0.0, 0.0); #endif // ALPHA_ANTIALIASING_EDGE_USED + mat4 inv_view_matrix = scene_data.inv_view_matrix; + mat4 read_model_matrix = draw_call.transform; +#ifdef USE_DOUBLE_PRECISION + read_model_matrix[0][3] = 0.0; + read_model_matrix[1][3] = 0.0; + read_model_matrix[2][3] = 0.0; + inv_view_matrix[0][3] = 0.0; + inv_view_matrix[1][3] = 0.0; + inv_view_matrix[2][3] = 0.0; +#endif + { #CODE : FRAGMENT } diff --git a/servers/rendering/renderer_rd/shaders/forward_mobile/scene_forward_mobile_inc.glsl b/servers/rendering/renderer_rd/shaders/forward_mobile/scene_forward_mobile_inc.glsl index 5e4999fa0f..631ff0575b 100644 --- a/servers/rendering/renderer_rd/shaders/forward_mobile/scene_forward_mobile_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/forward_mobile/scene_forward_mobile_inc.glsl @@ -55,13 +55,14 @@ layout(set = 0, binding = 2) uniform sampler shadow_sampler; layout(set = 0, binding = 3) uniform sampler decal_sampler; layout(set = 0, binding = 4) uniform sampler light_projector_sampler; -#define INSTANCE_FLAGS_NON_UNIFORM_SCALE (1 << 5) -#define INSTANCE_FLAGS_USE_GI_BUFFERS (1 << 6) -#define INSTANCE_FLAGS_USE_SDFGI (1 << 7) -#define INSTANCE_FLAGS_USE_LIGHTMAP_CAPTURE (1 << 8) -#define INSTANCE_FLAGS_USE_LIGHTMAP (1 << 9) -#define INSTANCE_FLAGS_USE_SH_LIGHTMAP (1 << 10) -#define INSTANCE_FLAGS_USE_VOXEL_GI (1 << 11) +#define INSTANCE_FLAGS_NON_UNIFORM_SCALE (1 << 4) +#define INSTANCE_FLAGS_USE_GI_BUFFERS (1 << 5) +#define INSTANCE_FLAGS_USE_SDFGI (1 << 6) +#define INSTANCE_FLAGS_USE_LIGHTMAP_CAPTURE (1 << 7) +#define INSTANCE_FLAGS_USE_LIGHTMAP (1 << 8) +#define INSTANCE_FLAGS_USE_SH_LIGHTMAP (1 << 9) +#define INSTANCE_FLAGS_USE_VOXEL_GI (1 << 10) +#define INSTANCE_FLAGS_PARTICLES (1 << 11) #define INSTANCE_FLAGS_MULTIMESH (1 << 12) #define INSTANCE_FLAGS_MULTIMESH_FORMAT_2D (1 << 13) #define INSTANCE_FLAGS_MULTIMESH_HAS_COLOR (1 << 14) diff --git a/servers/rendering/renderer_rd/shaders/particles.glsl b/servers/rendering/renderer_rd/shaders/particles.glsl index fb5759bc17..3a6dd579b9 100644 --- a/servers/rendering/renderer_rd/shaders/particles.glsl +++ b/servers/rendering/renderer_rd/shaders/particles.glsl @@ -475,7 +475,7 @@ void main() { float particle_size = FRAME.particle_size; -#ifdef USE_COLLISON_SCALE +#ifdef USE_COLLISION_SCALE particle_size *= dot(vec3(length(PARTICLE.xform[0].xyz), length(PARTICLE.xform[1].xyz), length(PARTICLE.xform[2].xyz)), vec3(0.33333333333)); |