diff options
Diffstat (limited to 'servers/rendering/renderer_rd/shaders/effects')
31 files changed, 3778 insertions, 115 deletions
diff --git a/servers/rendering/renderer_rd/shaders/effects/SCsub b/servers/rendering/renderer_rd/shaders/effects/SCsub index fc513d3fb9..f06a2d86e2 100644 --- a/servers/rendering/renderer_rd/shaders/effects/SCsub +++ b/servers/rendering/renderer_rd/shaders/effects/SCsub @@ -4,13 +4,13 @@ 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] # make sure we recompile shaders if include files change - env.Depends([f + ".gen.h" for f in glsl_files], gl_include_files) + env.Depends([f + ".gen.h" for f in glsl_files], gl_include_files + ["#glsl_builders.py"]) # compile shaders for glsl_file in glsl_files: diff --git a/servers/rendering/renderer_rd/shaders/effects/blur_raster.glsl b/servers/rendering/renderer_rd/shaders/effects/blur_raster.glsl index 96f5c3e9f2..31aabbe9d2 100644 --- a/servers/rendering/renderer_rd/shaders/effects/blur_raster.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/blur_raster.glsl @@ -53,30 +53,31 @@ void main() { #ifdef MODE_GAUSSIAN_BLUR - // Simpler blur uses SIGMA2 for the gaussian kernel for a stronger effect - - // note, for blur blur.luminance_multiplier is irrelavant, we would be multiplying and then dividing by this amount. - - if (bool(blur.flags & FLAG_HORIZONTAL)) { - vec2 pix_size = blur.pixel_size; - pix_size *= 0.5; //reading from larger buffer, so use more samples - vec4 color = texture(source_color, uv_interp + vec2(0.0, 0.0) * pix_size) * 0.214607; - color += texture(source_color, uv_interp + vec2(1.0, 0.0) * pix_size) * 0.189879; - color += texture(source_color, uv_interp + vec2(2.0, 0.0) * pix_size) * 0.131514; - color += texture(source_color, uv_interp + vec2(3.0, 0.0) * pix_size) * 0.071303; - color += texture(source_color, uv_interp + vec2(-1.0, 0.0) * pix_size) * 0.189879; - color += texture(source_color, uv_interp + vec2(-2.0, 0.0) * pix_size) * 0.131514; - color += texture(source_color, uv_interp + vec2(-3.0, 0.0) * pix_size) * 0.071303; - frag_color = color; - } else { - vec2 pix_size = blur.pixel_size; - vec4 color = texture(source_color, uv_interp + vec2(0.0, 0.0) * pix_size) * 0.38774; - color += texture(source_color, uv_interp + vec2(0.0, 1.0) * pix_size) * 0.24477; - color += texture(source_color, uv_interp + vec2(0.0, 2.0) * pix_size) * 0.06136; - color += texture(source_color, uv_interp + vec2(0.0, -1.0) * pix_size) * 0.24477; - color += texture(source_color, uv_interp + vec2(0.0, -2.0) * pix_size) * 0.06136; - frag_color = color; - } + // For Gaussian Blur we use 13 taps in a single pass instead of 12 taps over 2 passes. + // This minimizes the number of times we change framebuffers which is very important for mobile. + // Source: http://www.iryoku.com/next-generation-post-processing-in-call-of-duty-advanced-warfare + vec4 A = texture(source_color, uv_interp + blur.pixel_size * vec2(-1.0, -1.0)); + vec4 B = texture(source_color, uv_interp + blur.pixel_size * vec2(0.0, -1.0)); + vec4 C = texture(source_color, uv_interp + blur.pixel_size * vec2(1.0, -1.0)); + vec4 D = texture(source_color, uv_interp + blur.pixel_size * vec2(-0.5, -0.5)); + vec4 E = texture(source_color, uv_interp + blur.pixel_size * vec2(0.5, -0.5)); + vec4 F = texture(source_color, uv_interp + blur.pixel_size * vec2(-1.0, 0.0)); + vec4 G = texture(source_color, uv_interp); + vec4 H = texture(source_color, uv_interp + blur.pixel_size * vec2(1.0, 0.0)); + vec4 I = texture(source_color, uv_interp + blur.pixel_size * vec2(-0.5, 0.5)); + vec4 J = texture(source_color, uv_interp + blur.pixel_size * vec2(0.5, 0.5)); + vec4 K = texture(source_color, uv_interp + blur.pixel_size * vec2(-1.0, 1.0)); + vec4 L = texture(source_color, uv_interp + blur.pixel_size * vec2(0.0, 1.0)); + vec4 M = texture(source_color, uv_interp + blur.pixel_size * vec2(1.0, 1.0)); + + float base_weight = 0.5 / 4.0; + float lesser_weight = 0.125 / 4.0; + + frag_color = (D + E + I + J) * base_weight; + frag_color += (A + B + G + F) * lesser_weight; + frag_color += (B + C + H + G) * lesser_weight; + frag_color += (F + G + L + K) * lesser_weight; + frag_color += (G + H + M + L) * lesser_weight; #endif #ifdef MODE_GAUSSIAN_GLOW @@ -129,7 +130,7 @@ void main() { #ifdef GLOW_USE_AUTO_EXPOSURE - frag_color /= texelFetch(source_auto_exposure, ivec2(0, 0), 0).r / blur.glow_auto_exposure_grey; + frag_color /= texelFetch(source_auto_exposure, ivec2(0, 0), 0).r / blur.glow_auto_exposure_scale; #endif frag_color *= blur.glow_exposure; diff --git a/servers/rendering/renderer_rd/shaders/effects/blur_raster_inc.glsl b/servers/rendering/renderer_rd/shaders/effects/blur_raster_inc.glsl index 730504571a..06ca198f37 100644 --- a/servers/rendering/renderer_rd/shaders/effects/blur_raster_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/blur_raster_inc.glsl @@ -16,7 +16,7 @@ layout(push_constant, std430) uniform Blur { float glow_exposure; // 04 - 36 float glow_white; // 04 - 40 float glow_luminance_cap; // 04 - 44 - float glow_auto_exposure_grey; // 04 - 48 + float glow_auto_exposure_scale; // 04 - 48 float luminance_multiplier; // 04 - 52 float res1; // 04 - 56 diff --git a/servers/rendering/renderer_rd/shaders/effects/bokeh_dof.glsl b/servers/rendering/renderer_rd/shaders/effects/bokeh_dof.glsl index 0438671dd2..fe770ac065 100644 --- a/servers/rendering/renderer_rd/shaders/effects/bokeh_dof.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/bokeh_dof.glsl @@ -30,7 +30,7 @@ layout(set = 1, binding = 0) uniform sampler2D source_bokeh; #ifdef MODE_GEN_BLUR_SIZE float get_depth_at_pos(vec2 uv) { - float depth = textureLod(source_depth, uv, 0.0).x; + float depth = textureLod(source_depth, uv, 0.0).x * 2.0 - 1.0; if (params.orthogonal) { depth = ((depth + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0; } else { @@ -41,11 +41,25 @@ float get_depth_at_pos(vec2 uv) { float get_blur_size(float depth) { if (params.blur_near_active && depth < params.blur_near_begin) { - return -(1.0 - smoothstep(params.blur_near_end, params.blur_near_begin, depth)) * params.blur_size - DEPTH_GAP; //near blur is negative + if (params.use_physical_near) { + // Physically-based. + float d = abs(params.blur_near_begin - depth); + return -(d / (params.blur_near_begin - d)) * params.blur_size_near - DEPTH_GAP; // Near blur is negative. + } else { + // Non-physically-based. + return -(1.0 - smoothstep(params.blur_near_end, params.blur_near_begin, depth)) * params.blur_size - DEPTH_GAP; // Near blur is negative. + } } if (params.blur_far_active && depth > params.blur_far_begin) { - return smoothstep(params.blur_far_begin, params.blur_far_end, depth) * params.blur_size + DEPTH_GAP; + if (params.use_physical_far) { + // Physically-based. + float d = abs(params.blur_far_begin - depth); + return (d / (params.blur_far_begin + d)) * params.blur_size_far + DEPTH_GAP; + } else { + // Non-physically-based. + return smoothstep(params.blur_far_begin, params.blur_far_end, depth) * params.blur_size + DEPTH_GAP; + } } return 0.0; @@ -172,6 +186,7 @@ void main() { uv += pixel_size * 0.5; //half pixel to read centers vec4 color = texture(color_texture, uv); + float initial_blur = color.a; float accum = 1.0; float radius = params.blur_scale; @@ -179,8 +194,8 @@ void main() { vec2 suv = uv + vec2(cos(ang), sin(ang)) * pixel_size * radius; vec4 sample_color = texture(color_texture, suv); float sample_size = abs(sample_color.a); - if (sample_color.a > color.a) { - sample_size = clamp(sample_size, 0.0, abs(color.a) * 2.0); + if (sample_color.a > initial_blur) { + sample_size = clamp(sample_size, 0.0, abs(initial_blur) * 2.0); } float m = smoothstep(radius - 0.5, radius + 0.5, sample_size); diff --git a/servers/rendering/renderer_rd/shaders/effects/bokeh_dof_inc.glsl b/servers/rendering/renderer_rd/shaders/effects/bokeh_dof_inc.glsl index b90a527554..4a2b0edc18 100644 --- a/servers/rendering/renderer_rd/shaders/effects/bokeh_dof_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/bokeh_dof_inc.glsl @@ -20,6 +20,11 @@ layout(push_constant, std430) uniform Params { bool use_jitter; float jitter_seed; + bool use_physical_near; + bool use_physical_far; + + float blur_size_near; + float blur_size_far; uint pad[2]; } params; diff --git a/servers/rendering/renderer_rd/shaders/effects/bokeh_dof_raster.glsl b/servers/rendering/renderer_rd/shaders/effects/bokeh_dof_raster.glsl index a3b3938ee9..1b487835d2 100644 --- a/servers/rendering/renderer_rd/shaders/effects/bokeh_dof_raster.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/bokeh_dof_raster.glsl @@ -52,7 +52,7 @@ layout(set = 2, binding = 0) uniform sampler2D original_weight; #ifdef MODE_GEN_BLUR_SIZE float get_depth_at_pos(vec2 uv) { - float depth = textureLod(source_depth, uv, 0.0).x; + float depth = textureLod(source_depth, uv, 0.0).x * 2.0 - 1.0; if (params.orthogonal) { depth = ((depth + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0; } else { @@ -63,11 +63,25 @@ float get_depth_at_pos(vec2 uv) { float get_blur_size(float depth) { if (params.blur_near_active && depth < params.blur_near_begin) { - return -(1.0 - smoothstep(params.blur_near_end, params.blur_near_begin, depth)) * params.blur_size - DEPTH_GAP; //near blur is negative + if (params.use_physical_near) { + // Physically-based. + float d = abs(params.blur_near_begin - depth); + return -(d / (params.blur_near_begin - d)) * params.blur_size_near - DEPTH_GAP; // Near blur is negative. + } else { + // Non-physically-based. + return -(1.0 - smoothstep(params.blur_near_end, params.blur_near_begin, depth)) * params.blur_size - DEPTH_GAP; // Near blur is negative. + } } if (params.blur_far_active && depth > params.blur_far_begin) { - return smoothstep(params.blur_far_begin, params.blur_far_end, depth) * params.blur_size + DEPTH_GAP; + if (params.use_physical_far) { + // Physically-based. + float d = abs(params.blur_far_begin - depth); + return (d / (params.blur_far_begin + d)) * params.blur_size_far + DEPTH_GAP; + } else { + // Non-physically-based. + return smoothstep(params.blur_far_begin, params.blur_far_end, depth) * params.blur_size + DEPTH_GAP; + } } return 0.0; @@ -207,12 +221,9 @@ void main() { vec4 sample_color = texture(source_color, uv_adj); sample_color.a = texture(source_weight, uv_adj).r; - float limit; - - if (sample_color.a < color.a) { - limit = abs(sample_color.a); - } else { - limit = abs(color.a); + float limit = abs(sample_color.a); + if (sample_color.a > color.a) { + limit = clamp(limit, 0.0, abs(color.a) * 2.0); } limit -= DEPTH_GAP; diff --git a/servers/rendering/renderer_rd/shaders/effects/copy.glsl b/servers/rendering/renderer_rd/shaders/effects/copy.glsl index 3a4ef86ef0..3a82861057 100644 --- a/servers/rendering/renderer_rd/shaders/effects/copy.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/copy.glsl @@ -14,8 +14,7 @@ layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; #define FLAG_FLIP_Y (1 << 5) #define FLAG_FORCE_LUMINANCE (1 << 6) #define FLAG_COPY_ALL_SOURCE (1 << 7) -#define FLAG_HIGH_QUALITY_GLOW (1 << 8) -#define FLAG_ALPHA_TO_ONE (1 << 9) +#define FLAG_ALPHA_TO_ONE (1 << 8) layout(push_constant, std430) uniform Params { ivec4 section; @@ -31,7 +30,7 @@ layout(push_constant, std430) uniform Params { float glow_exposure; float glow_white; float glow_luminance_cap; - float glow_auto_exposure_grey; + float glow_auto_exposure_scale; // DOF. float camera_z_far; float camera_z_near; @@ -93,25 +92,14 @@ void main() { #ifdef MODE_GAUSSIAN_BLUR // First pass copy texture into 16x16 local memory for every 8x8 thread block - vec2 quad_center_uv = clamp(vec2(gl_GlobalInvocationID.xy + gl_LocalInvocationID.xy - 3.5) / params.section.zw, vec2(0.5 / params.section.zw), vec2(1.0 - 1.5 / params.section.zw)); + vec2 quad_center_uv = clamp(vec2(params.section.xy + gl_GlobalInvocationID.xy + gl_LocalInvocationID.xy - 3.5) / params.section.zw, vec2(0.5 / params.section.zw), vec2(1.0 - 1.5 / params.section.zw)); uint dest_index = gl_LocalInvocationID.x * 2 + gl_LocalInvocationID.y * 2 * 16; -#ifdef MODE_GLOW - if (bool(params.flags & FLAG_HIGH_QUALITY_GLOW)) { - vec2 quad_offset_uv = clamp((vec2(gl_GlobalInvocationID.xy + gl_LocalInvocationID.xy - 3.0)) / params.section.zw, vec2(0.5 / params.section.zw), vec2(1.0 - 1.5 / params.section.zw)); - - local_cache[dest_index] = (textureLod(source_color, quad_center_uv, 0) + textureLod(source_color, quad_offset_uv, 0)) * 0.5; - local_cache[dest_index + 1] = (textureLod(source_color, quad_center_uv + vec2(1.0 / params.section.z, 0.0), 0) + textureLod(source_color, quad_offset_uv + vec2(1.0 / params.section.z, 0.0), 0)) * 0.5; - local_cache[dest_index + 16] = (textureLod(source_color, quad_center_uv + vec2(0.0, 1.0 / params.section.w), 0) + textureLod(source_color, quad_offset_uv + vec2(0.0, 1.0 / params.section.w), 0)) * 0.5; - local_cache[dest_index + 16 + 1] = (textureLod(source_color, quad_center_uv + vec2(1.0 / params.section.zw), 0) + textureLod(source_color, quad_offset_uv + vec2(1.0 / params.section.zw), 0)) * 0.5; - } else -#endif - { - local_cache[dest_index] = textureLod(source_color, quad_center_uv, 0); - local_cache[dest_index + 1] = textureLod(source_color, quad_center_uv + vec2(1.0 / params.section.z, 0.0), 0); - 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); - } + local_cache[dest_index] = textureLod(source_color, quad_center_uv, 0); + local_cache[dest_index + 1] = textureLod(source_color, quad_center_uv + vec2(1.0 / params.section.z, 0.0), 0); + 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); + #ifdef MODE_GLOW 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 @@ -185,7 +173,7 @@ void main() { if (bool(params.flags & FLAG_GLOW_FIRST_PASS)) { #ifdef GLOW_USE_AUTO_EXPOSURE - color /= texelFetch(source_auto_exposure, ivec2(0, 0), 0).r / params.glow_auto_exposure_grey; + color /= texelFetch(source_auto_exposure, ivec2(0, 0), 0).r / params.glow_auto_exposure_scale; #endif color *= params.glow_exposure; @@ -194,10 +182,10 @@ void main() { color = min(color * feedback, vec4(params.glow_luminance_cap)); } -#endif +#endif // MODE_GLOW imageStore(dest_buffer, pos + params.target, color); -#endif +#endif // MODE_GAUSSIAN_BLUR #ifdef MODE_SIMPLE_COPY @@ -227,7 +215,7 @@ void main() { imageStore(dest_buffer, pos + params.target, color); -#endif +#endif // MODE_SIMPLE_COPY #ifdef MODE_SIMPLE_COPY_DEPTH @@ -239,7 +227,7 @@ void main() { imageStore(dest_buffer, pos + params.target, vec4(color.r)); -#endif +#endif // MODE_SIMPLE_COPY_DEPTH #ifdef MODE_LINEARIZE_DEPTH_COPY @@ -253,7 +241,7 @@ void main() { } imageStore(dest_buffer, pos + params.target, color); -#endif +#endif // MODE_LINEARIZE_DEPTH_COPY #if defined(MODE_CUBEMAP_TO_PANORAMA) || defined(MODE_CUBEMAP_ARRAY_TO_PANORAMA) @@ -276,7 +264,7 @@ void main() { vec4 color = textureLod(source_color, vec4(normal, params.camera_z_far), 0.0); //the biggest the lod the least the acne #endif imageStore(dest_buffer, pos + params.target, color); -#endif +#endif // defined(MODE_CUBEMAP_TO_PANORAMA) || defined(MODE_CUBEMAP_ARRAY_TO_PANORAMA) #ifdef MODE_SET_COLOR imageStore(dest_buffer, pos + params.target, params.set_color); diff --git a/servers/rendering/renderer_rd/shaders/effects/copy_to_fb.glsl b/servers/rendering/renderer_rd/shaders/effects/copy_to_fb.glsl index 9787c9879d..6137224162 100644 --- a/servers/rendering/renderer_rd/shaders/effects/copy_to_fb.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/copy_to_fb.glsl @@ -13,6 +13,14 @@ #endif // has_VK_KHR_multiview #endif //MULTIVIEW +#define FLAG_FLIP_Y (1 << 0) +#define FLAG_USE_SECTION (1 << 1) +#define FLAG_FORCE_LUMINANCE (1 << 2) +#define FLAG_ALPHA_TO_ZERO (1 << 3) +#define FLAG_SRGB (1 << 4) +#define FLAG_ALPHA_TO_ONE (1 << 5) +#define FLAG_LINEAR (1 << 6) + #ifdef MULTIVIEW layout(location = 0) out vec3 uv_interp; #else @@ -22,11 +30,10 @@ layout(location = 0) out vec2 uv_interp; layout(push_constant, std430) uniform Params { vec4 section; vec2 pixel_size; - bool flip_y; - bool use_section; + float luminance_multiplier; + uint flags; - bool force_luminance; - uint pad[3]; + vec4 color; } params; @@ -37,13 +44,13 @@ void main() { uv_interp.z = ViewIndex; #endif vec2 vpos = uv_interp.xy; - if (params.use_section) { + if (bool(params.flags & FLAG_USE_SECTION)) { vpos = params.section.xy + vpos * params.section.zw; } gl_Position = vec4(vpos * 2.0 - 1.0, 0.0, 1.0); - if (params.flip_y) { + if (bool(params.flags & FLAG_FLIP_Y)) { uv_interp.y = 1.0 - uv_interp.y; } } @@ -63,19 +70,25 @@ void main() { #endif // has_VK_KHR_multiview #endif //MULTIVIEW +#define FLAG_FLIP_Y (1 << 0) +#define FLAG_USE_SECTION (1 << 1) +#define FLAG_FORCE_LUMINANCE (1 << 2) +#define FLAG_ALPHA_TO_ZERO (1 << 3) +#define FLAG_SRGB (1 << 4) +#define FLAG_ALPHA_TO_ONE (1 << 5) +#define FLAG_LINEAR (1 << 6) + layout(push_constant, std430) uniform Params { vec4 section; vec2 pixel_size; - bool flip_y; - bool use_section; + float luminance_multiplier; + uint flags; - bool force_luminance; - bool alpha_to_zero; - bool srgb; - uint pad; + vec4 color; } params; +#ifndef MODE_SET_COLOR #ifdef MULTIVIEW layout(location = 0) in vec3 uv_interp; #else @@ -88,12 +101,13 @@ layout(set = 0, binding = 0) uniform sampler2DArray source_color; layout(set = 1, binding = 0) uniform sampler2DArray source_depth; layout(location = 1) out float depth; #endif /* MODE_TWO_SOURCES */ -#else +#else /* MULTIVIEW */ layout(set = 0, binding = 0) uniform sampler2D source_color; #ifdef MODE_TWO_SOURCES layout(set = 1, binding = 0) uniform sampler2D source_color2; #endif /* MODE_TWO_SOURCES */ #endif /* MULTIVIEW */ +#endif /* !SET_COLOR */ layout(location = 0) out vec4 frag_color; @@ -104,7 +118,15 @@ vec3 linear_to_srgb(vec3 color) { return mix((vec3(1.0f) + a) * pow(color.rgb, vec3(1.0f / 2.4f)) - a, 12.92f * color.rgb, lessThan(color.rgb, vec3(0.0031308f))); } +vec3 srgb_to_linear(vec3 color) { + return mix(pow((color.rgb + vec3(0.055)) * (1.0 / (1.0 + 0.055)), vec3(2.4)), color.rgb * (1.0 / 12.92), lessThan(color.rgb, vec3(0.04045))); +} + void main() { +#ifdef MODE_SET_COLOR + frag_color = params.color; +#else + #ifdef MULTIVIEW vec3 uv = uv_interp; #else @@ -139,7 +161,7 @@ void main() { //uv.y = 1.0 - uv.y; uv = 1.0 - uv; } -#endif +#endif /* MODE_PANORAMA_TO_DP */ #ifdef MULTIVIEW vec4 color = textureLod(source_color, uv, 0.0); @@ -148,20 +170,29 @@ void main() { depth = textureLod(source_depth, uv, 0.0).r; #endif /* MODE_TWO_SOURCES */ -#else +#else /* MULTIVIEW */ vec4 color = textureLod(source_color, uv, 0.0); #ifdef MODE_TWO_SOURCES color += textureLod(source_color2, uv, 0.0); #endif /* MODE_TWO_SOURCES */ #endif /* MULTIVIEW */ - if (params.force_luminance) { + + if (bool(params.flags & FLAG_FORCE_LUMINANCE)) { color.rgb = vec3(max(max(color.r, color.g), color.b)); } - if (params.alpha_to_zero) { + if (bool(params.flags & FLAG_ALPHA_TO_ZERO)) { color.rgb *= color.a; } - if (params.srgb) { + if (bool(params.flags & FLAG_SRGB)) { color.rgb = linear_to_srgb(color.rgb); } - frag_color = color; + if (bool(params.flags & FLAG_ALPHA_TO_ONE)) { + color.a = 1.0; + } + if (bool(params.flags & FLAG_LINEAR)) { + color.rgb = srgb_to_linear(color.rgb); + } + + frag_color = color / params.luminance_multiplier; +#endif // MODE_SET_COLOR } diff --git a/servers/rendering/renderer_rd/shaders/effects/cubemap_roughness_inc.glsl b/servers/rendering/renderer_rd/shaders/effects/cubemap_roughness_inc.glsl index 1bee428a6f..c0597fe3f3 100644 --- a/servers/rendering/renderer_rd/shaders/effects/cubemap_roughness_inc.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/cubemap_roughness_inc.glsl @@ -70,17 +70,6 @@ float DistributionGGX(float NdotH, float roughness4) { return roughness4 / denom; } -// https://graphicrants.blogspot.com.au/2013/08/specular-brdf-reference.html -float GGX(float NdotV, float a) { - float k = a / 2.0; - return NdotV / (NdotV * (1.0 - k) + k); -} - -// https://graphicrants.blogspot.com.au/2013/08/specular-brdf-reference.html -float G_Smith(float a, float nDotV, float nDotL) { - return GGX(nDotL, a * a) * GGX(nDotV, a * a); -} - float radicalInverse_VdC(uint bits) { bits = (bits << 16u) | (bits >> 16u); bits = ((bits & 0x55555555u) << 1u) | ((bits & 0xAAAAAAAAu) >> 1u); diff --git a/servers/rendering/renderer_rd/shaders/effects/fsr_upscale.glsl b/servers/rendering/renderer_rd/shaders/effects/fsr_upscale.glsl new file mode 100644 index 0000000000..221e97bece --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/fsr_upscale.glsl @@ -0,0 +1,173 @@ +/**************************************************************************/ +/* fsr_upscale.glsl */ +/**************************************************************************/ +/* This file is part of: */ +/* GODOT ENGINE */ +/* https://godotengine.org */ +/**************************************************************************/ +/* Copyright (c) 2014-present Godot Engine contributors (see AUTHORS.md). */ +/* Copyright (c) 2007-2014 Juan Linietsky, Ariel Manzur. */ +/* */ +/* 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, 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/effects/luminance_reduce.glsl b/servers/rendering/renderer_rd/shaders/effects/luminance_reduce.glsl new file mode 100644 index 0000000000..0ee4cf6e31 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/luminance_reduce.glsl @@ -0,0 +1,82 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +#define BLOCK_SIZE 8 + +layout(local_size_x = BLOCK_SIZE, local_size_y = BLOCK_SIZE, local_size_z = 1) in; + +shared float tmp_data[BLOCK_SIZE * BLOCK_SIZE]; + +#ifdef READ_TEXTURE + +//use for main texture +layout(set = 0, binding = 0) uniform sampler2D source_texture; + +#else + +//use for intermediate textures +layout(r32f, set = 0, binding = 0) uniform restrict readonly image2D source_luminance; + +#endif + +layout(r32f, set = 1, binding = 0) uniform restrict writeonly image2D dest_luminance; + +#ifdef WRITE_LUMINANCE +layout(set = 2, binding = 0) uniform sampler2D prev_luminance; +#endif + +layout(push_constant, std430) uniform Params { + ivec2 source_size; + float max_luminance; + float min_luminance; + float exposure_adjust; + float pad[3]; +} +params; + +void main() { + uint t = gl_LocalInvocationID.y * BLOCK_SIZE + gl_LocalInvocationID.x; + ivec2 pos = ivec2(gl_GlobalInvocationID.xy); + + if (any(lessThan(pos, params.source_size))) { +#ifdef READ_TEXTURE + vec3 v = texelFetch(source_texture, pos, 0).rgb; + tmp_data[t] = max(v.r, max(v.g, v.b)); +#else + tmp_data[t] = imageLoad(source_luminance, pos).r; +#endif + } else { + tmp_data[t] = 0.0; + } + + groupMemoryBarrier(); + barrier(); + + uint size = (BLOCK_SIZE * BLOCK_SIZE) >> 1; + + do { + if (t < size) { + tmp_data[t] += tmp_data[t + size]; + } + groupMemoryBarrier(); + barrier(); + + size >>= 1; + } while (size >= 1); + + if (t == 0) { + //compute rect size + ivec2 rect_size = min(params.source_size - pos, ivec2(BLOCK_SIZE)); + float avg = tmp_data[0] / float(rect_size.x * rect_size.y); + //float avg = tmp_data[0] / float(BLOCK_SIZE*BLOCK_SIZE); + pos /= ivec2(BLOCK_SIZE); +#ifdef WRITE_LUMINANCE + float prev_lum = texelFetch(prev_luminance, ivec2(0, 0), 0).r; //1 pixel previous exposure + avg = clamp(prev_lum + (avg - prev_lum) * params.exposure_adjust, params.min_luminance, params.max_luminance); +#endif + imageStore(dest_luminance, pos, vec4(avg)); + } +} diff --git a/servers/rendering/renderer_rd/shaders/effects/luminance_reduce_raster.glsl b/servers/rendering/renderer_rd/shaders/effects/luminance_reduce_raster.glsl new file mode 100644 index 0000000000..29ebd74a90 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/luminance_reduce_raster.glsl @@ -0,0 +1,74 @@ +/* clang-format off */ +#[vertex] + +#version 450 + +#VERSION_DEFINES + +#include "luminance_reduce_raster_inc.glsl" + +layout(location = 0) out vec2 uv_interp; +/* clang-format on */ + +void main() { + vec2 base_arr[4] = vec2[](vec2(0.0, 0.0), vec2(0.0, 1.0), vec2(1.0, 1.0), vec2(1.0, 0.0)); + uv_interp = base_arr[gl_VertexIndex]; + + gl_Position = vec4(uv_interp * 2.0 - 1.0, 0.0, 1.0); +} + +/* clang-format off */ +#[fragment] + +#version 450 + +#VERSION_DEFINES + +#include "luminance_reduce_raster_inc.glsl" + +layout(location = 0) in vec2 uv_interp; +/* clang-format on */ + +layout(set = 0, binding = 0) uniform sampler2D source_exposure; + +#ifdef FINAL_PASS +layout(set = 1, binding = 0) uniform sampler2D prev_luminance; +#endif + +layout(location = 0) out highp float luminance; + +void main() { + ivec2 dest_pos = ivec2(uv_interp * settings.dest_size); + ivec2 src_pos = ivec2(uv_interp * settings.source_size); + + ivec2 next_pos = (dest_pos + ivec2(1)) * settings.source_size / settings.dest_size; + next_pos = max(next_pos, src_pos + ivec2(1)); //so it at least reads one pixel + + highp vec3 source_color = vec3(0.0); + for (int i = src_pos.x; i < next_pos.x; i++) { + for (int j = src_pos.y; j < next_pos.y; j++) { + source_color += texelFetch(source_exposure, ivec2(i, j), 0).rgb; + } + } + + source_color /= float((next_pos.x - src_pos.x) * (next_pos.y - src_pos.y)); + +#ifdef FIRST_PASS + luminance = max(source_color.r, max(source_color.g, source_color.b)); + + // This formula should be more "accurate" but gave an overexposed result when testing. + // Leaving it here so we can revisit it if we want. + // luminance = source_color.r * 0.21 + source_color.g * 0.71 + source_color.b * 0.07; +#else + luminance = source_color.r; +#endif + +#ifdef FINAL_PASS + // Obtain our target luminance + luminance = clamp(luminance, settings.min_luminance, settings.max_luminance); + + // Now smooth to our transition + highp float prev_lum = texelFetch(prev_luminance, ivec2(0, 0), 0).r; //1 pixel previous luminance + luminance = prev_lum + (luminance - prev_lum) * clamp(settings.exposure_adjust, 0.0, 1.0); +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/effects/luminance_reduce_raster_inc.glsl b/servers/rendering/renderer_rd/shaders/effects/luminance_reduce_raster_inc.glsl new file mode 100644 index 0000000000..b8860f6518 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/luminance_reduce_raster_inc.glsl @@ -0,0 +1,11 @@ + +layout(push_constant, std430) uniform PushConstant { + ivec2 source_size; + ivec2 dest_size; + + float exposure_adjust; + float min_luminance; + float max_luminance; + uint pad1; +} +settings; diff --git a/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection.glsl b/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection.glsl new file mode 100644 index 0000000000..631d1968b0 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection.glsl @@ -0,0 +1,281 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +layout(rgba16f, set = 0, binding = 0) uniform restrict readonly image2D source_diffuse; +layout(r32f, set = 0, binding = 1) uniform restrict readonly image2D source_depth; +layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly image2D ssr_image; +#ifdef MODE_ROUGH +layout(r8, set = 1, binding = 1) uniform restrict writeonly image2D blur_radius_image; +#endif +layout(rgba8, set = 2, binding = 0) uniform restrict readonly image2D source_normal_roughness; +layout(set = 3, binding = 0) uniform sampler2D source_metallic; + +layout(push_constant, std430) uniform Params { + vec4 proj_info; + + ivec2 screen_size; + float camera_z_near; + float camera_z_far; + + int num_steps; + float depth_tolerance; + float distance_fade; + float curve_fade_in; + + bool orthogonal; + float filter_mipmap_levels; + bool use_half_res; + uint view_index; +} +params; + +#include "screen_space_reflection_inc.glsl" + +vec2 view_to_screen(vec3 view_pos, out float w) { + vec4 projected = scene_data.projection[params.view_index] * vec4(view_pos, 1.0); + projected.xyz /= projected.w; + projected.xy = projected.xy * 0.5 + 0.5; + w = projected.w; + return projected.xy; +} + +#define M_PI 3.14159265359 + +void main() { + // Pixel being shaded + ivec2 ssC = ivec2(gl_GlobalInvocationID.xy); + + if (any(greaterThanEqual(ssC.xy, params.screen_size))) { //too large, do nothing + return; + } + + vec2 pixel_size = 1.0 / vec2(params.screen_size); + vec2 uv = vec2(ssC.xy) * pixel_size; + + uv += pixel_size * 0.5; + + float base_depth = imageLoad(source_depth, ssC).r; + + // World space point being shaded + vec3 vertex = reconstructCSPosition(uv * vec2(params.screen_size), base_depth); + + vec4 normal_roughness = imageLoad(source_normal_roughness, ssC); + vec3 normal = normal_roughness.xyz * 2.0 - 1.0; + float roughness = normal_roughness.w; + + // The roughness cutoff of 0.6 is chosen to match the roughness fadeout from GH-69828. + if (roughness > 0.6) { + // Do not compute SSR for rough materials to improve performance at the cost of + // subtle artifacting. +#ifdef MODE_ROUGH + imageStore(blur_radius_image, ssC, vec4(0.0)); +#endif + imageStore(ssr_image, ssC, vec4(0.0)); + return; + } + + normal = normalize(normal); + normal.y = -normal.y; //because this code reads flipped + + vec3 view_dir; + if (sc_multiview) { + view_dir = normalize(vertex + scene_data.eye_offset[params.view_index].xyz); + } else { + view_dir = normalize(vertex); + } + vec3 ray_dir = normalize(reflect(view_dir, normal)); + + if (dot(ray_dir, normal) < 0.001) { + imageStore(ssr_image, ssC, vec4(0.0)); + return; + } + + //////////////// + + // make ray length and clip it against the near plane (don't want to trace beyond visible) + float ray_len = (vertex.z + ray_dir.z * params.camera_z_far) > -params.camera_z_near ? (-params.camera_z_near - vertex.z) / ray_dir.z : params.camera_z_far; + vec3 ray_end = vertex + ray_dir * ray_len; + + float w_begin; + vec2 vp_line_begin = view_to_screen(vertex, w_begin); + float w_end; + vec2 vp_line_end = view_to_screen(ray_end, w_end); + vec2 vp_line_dir = vp_line_end - vp_line_begin; + + // we need to interpolate w along the ray, to generate perspective correct reflections + w_begin = 1.0 / w_begin; + w_end = 1.0 / w_end; + + float z_begin = vertex.z * w_begin; + float z_end = ray_end.z * w_end; + + vec2 line_begin = vp_line_begin / pixel_size; + vec2 line_dir = vp_line_dir / pixel_size; + float z_dir = z_end - z_begin; + float w_dir = w_end - w_begin; + + // clip the line to the viewport edges + + float scale_max_x = min(1.0, 0.99 * (1.0 - vp_line_begin.x) / max(1e-5, vp_line_dir.x)); + float scale_max_y = min(1.0, 0.99 * (1.0 - vp_line_begin.y) / max(1e-5, vp_line_dir.y)); + float scale_min_x = min(1.0, 0.99 * vp_line_begin.x / max(1e-5, -vp_line_dir.x)); + float scale_min_y = min(1.0, 0.99 * vp_line_begin.y / max(1e-5, -vp_line_dir.y)); + float line_clip = min(scale_max_x, scale_max_y) * min(scale_min_x, scale_min_y); + line_dir *= line_clip; + z_dir *= line_clip; + w_dir *= line_clip; + + // clip z and w advance to line advance + vec2 line_advance = normalize(line_dir); // down to pixel + 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 + + // make line advance faster if direction is closer to pixel edges (this avoids sampling the same pixel twice) + float advance_angle_adj = 1.0 / max(abs(line_advance.x), abs(line_advance.y)); + line_advance *= advance_angle_adj; // adapt z advance to line advance + z_advance *= advance_angle_adj; + w_advance *= advance_angle_adj; + + vec2 pos = line_begin; + float z = z_begin; + float w = w_begin; + float z_from = z / w; + float z_to = z_from; + 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; + + for (int i = 0; i < params.num_steps; i++) { + pos += line_advance; + z += z_advance; + w += w_advance; + + // convert to linear depth + 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)); + depth = -depth; + } + + z_from = z_to; + z_to = z / w; + + if (depth > z_to) { + // 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; + } + } + + steps_taken += 1.0; + prev_pos = pos; + } + + if (found) { + float margin_blend = 1.0; + + vec2 margin = vec2((params.screen_size.x + params.screen_size.y) * 0.05); // make a uniform margin + if (any(bvec4(lessThan(pos, vec2(0.0, 0.0)), greaterThan(pos, params.screen_size)))) { + // clip at the screen edges + imageStore(ssr_image, ssC, vec4(0.0)); + return; + } + + { + //blend fading out towards inner margin + // 0.5 = midpoint of reflection + vec2 margin_grad = mix(params.screen_size - pos, pos, lessThan(pos, params.screen_size * 0.5)); + margin_blend = smoothstep(0.0, margin.x * margin.y, margin_grad.x * margin_grad.y); + //margin_blend = 1.0; + } + + vec2 final_pos; + float grad = (steps_taken + 1.0) / float(params.num_steps); + float initial_fade = params.curve_fade_in == 0.0 ? 1.0 : pow(clamp(grad, 0.0, 1.0), params.curve_fade_in); + float fade = pow(clamp(1.0 - grad, 0.0, 1.0), params.distance_fade) * initial_fade; + // This is an ad-hoc term to fade out the SSR as roughness increases. Values used + // are meant to match the visual appearance of a ReflectionProbe. + float roughness_fade = smoothstep(0.4, 0.7, 1.0 - normal_roughness.w); + final_pos = pos; + + vec4 final_color; + +#ifdef MODE_ROUGH + + // if roughness is enabled, do screen space cone tracing + float blur_radius = 0.0; + + if (roughness > 0.001) { + float cone_angle = min(roughness, 0.999) * M_PI * 0.5; + float cone_len = length(final_pos - line_begin); + float op_len = 2.0 * tan(cone_angle) * cone_len; // opposite side of iso triangle + { + // fit to sphere inside cone (sphere ends at end of cone), something like this: + // ___ + // \O/ + // V + // + // as it avoids bleeding from beyond the reflection as much as possible. As a plus + // it also makes the rough reflection more elongated. + float a = op_len; + float h = cone_len; + float a2 = a * a; + float fh2 = 4.0f * h * h; + blur_radius = (a * (sqrt(a2 + fh2) - a)) / (4.0f * h); + } + } + + imageStore(blur_radius_image, ssC, vec4(blur_radius / 255.0)); //stored in r8 + +#endif // MODE_ROUGH + + final_color = vec4(imageLoad(source_diffuse, ivec2(final_pos - 0.5)).rgb, fade * margin_blend * roughness_fade); + + // Schlick term. + float metallic = texelFetch(source_metallic, ssC << 1, 0).w; + float f0 = mix(0.04, 1.0, metallic); // Assume a "specular" amount of 0.5 + normal.y = -normal.y; + float m = clamp(1.0 - dot(normalize(normal), -view_dir), 0.0, 1.0); + float m2 = m * m; + m = m2 * m2 * m; // pow(m,5) + final_color.a *= f0 + (1.0 - f0) * m; // Fresnel Schlick term. + + imageStore(ssr_image, ssC, final_color); + + } else { +#ifdef MODE_ROUGH + imageStore(blur_radius_image, ssC, vec4(0.0)); +#endif + imageStore(ssr_image, ssC, vec4(0.0)); + } +} diff --git a/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection_filter.glsl b/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection_filter.glsl new file mode 100644 index 0000000000..a63d60e0b2 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection_filter.glsl @@ -0,0 +1,148 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +layout(rgba16f, set = 0, binding = 0) uniform restrict readonly image2D source_ssr; +layout(r8, set = 0, binding = 1) uniform restrict readonly image2D source_radius; +layout(rgba8, set = 1, binding = 0) uniform restrict readonly image2D source_normal; + +layout(rgba16f, set = 2, binding = 0) uniform restrict writeonly image2D dest_ssr; +#ifndef VERTICAL_PASS +layout(r8, set = 2, binding = 1) uniform restrict writeonly image2D dest_radius; +#endif +layout(r32f, set = 3, binding = 0) uniform restrict readonly image2D source_depth; + +layout(push_constant, std430) uniform Params { + vec4 proj_info; + + bool orthogonal; + float edge_tolerance; + int increment; + uint view_index; + + ivec2 screen_size; + bool vertical; + uint steps; +} +params; + +#include "screen_space_reflection_inc.glsl" + +#define GAUSS_TABLE_SIZE 15 + +const float gauss_table[GAUSS_TABLE_SIZE + 1] = float[]( + 0.1847392078702266, + 0.16595854345772326, + 0.12031364177766891, + 0.07038755277896766, + 0.03322925565155569, + 0.012657819729901945, + 0.0038903040680094217, + 0.0009646503390864025, + 0.00019297087402915717, + 0.000031139936308099136, + 0.000004053309048174758, + 4.255228059965837e-7, + 3.602517634249573e-8, + 2.4592560765896795e-9, + 1.3534945386863618e-10, + 0.0 //one more for interpolation +); + +float gauss_weight(float p_val) { + float idxf; + float c = modf(max(0.0, p_val * float(GAUSS_TABLE_SIZE)), idxf); + int idx = int(idxf); + if (idx >= GAUSS_TABLE_SIZE + 1) { + return 0.0; + } + + return mix(gauss_table[idx], gauss_table[idx + 1], c); +} + +#define M_PI 3.14159265359 + +void do_filter(inout vec4 accum, inout float accum_radius, inout float divisor, ivec2 texcoord, ivec2 increment, vec3 p_pos, vec3 normal, float p_limit_radius) { + for (int i = 1; i < params.steps; i++) { + float d = float(i * params.increment); + ivec2 tc = texcoord + increment * i; + float depth = imageLoad(source_depth, tc).r; + vec3 view_pos = reconstructCSPosition(vec2(tc) + 0.5, depth); + vec3 view_normal = normalize(imageLoad(source_normal, tc).rgb * 2.0 - 1.0); + view_normal.y = -view_normal.y; + + float r = imageLoad(source_radius, tc).r; + float radius = round(r * 255.0); + + float angle_n = 1.0 - abs(dot(normal, view_normal)); + if (angle_n > params.edge_tolerance) { + break; + } + + float angle = abs(dot(normal, normalize(view_pos - p_pos))); + + if (angle > params.edge_tolerance) { + break; + } + + if (d < radius) { + float w = gauss_weight(d / radius); + accum += imageLoad(source_ssr, tc) * w; +#ifndef VERTICAL_PASS + accum_radius += r * w; +#endif + divisor += w; + } + } +} + +void main() { + // Pixel being shaded + ivec2 ssC = ivec2(gl_GlobalInvocationID.xy); + + if (any(greaterThanEqual(ssC.xy, params.screen_size))) { //too large, do nothing + return; + } + + float base_contrib = gauss_table[0]; + + vec4 accum = imageLoad(source_ssr, ssC); + + float accum_radius = imageLoad(source_radius, ssC).r; + float radius = accum_radius * 255.0; + + float divisor = gauss_table[0]; + accum *= divisor; + accum_radius *= divisor; +#ifdef VERTICAL_PASS + ivec2 direction = ivec2(0, params.increment); +#else + ivec2 direction = ivec2(params.increment, 0); +#endif + float depth = imageLoad(source_depth, ssC).r; + vec3 pos = reconstructCSPosition(vec2(ssC.xy) + 0.5, depth); + vec3 normal = imageLoad(source_normal, ssC).xyz * 2.0 - 1.0; + normal = normalize(normal); + normal.y = -normal.y; + + do_filter(accum, accum_radius, divisor, ssC.xy, direction, pos, normal, radius); + do_filter(accum, accum_radius, divisor, ssC.xy, -direction, pos, normal, radius); + + if (divisor > 0.0) { + accum /= divisor; + accum_radius /= divisor; + } else { + accum = vec4(0.0); + accum_radius = 0.0; + } + + imageStore(dest_ssr, ssC, accum); + +#ifndef VERTICAL_PASS + imageStore(dest_radius, ssC, vec4(accum_radius)); +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection_inc.glsl b/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection_inc.glsl new file mode 100644 index 0000000000..26405ab040 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection_inc.glsl @@ -0,0 +1,28 @@ +layout(constant_id = 0) const bool sc_multiview = false; + +layout(set = 4, binding = 0, std140) uniform SceneData { + mat4x4 projection[2]; + mat4x4 inv_projection[2]; + vec4 eye_offset[2]; +} +scene_data; + +vec3 reconstructCSPosition(vec2 screen_pos, float z) { + if (sc_multiview) { + vec4 pos; + pos.xy = (2.0 * vec2(screen_pos) / vec2(params.screen_size)) - 1.0; + pos.z = z * 2.0 - 1.0; + pos.w = 1.0; + + pos = scene_data.inv_projection[params.view_index] * pos; + pos.xyz /= pos.w; + + return pos.xyz; + } else { + if (params.orthogonal) { + return vec3((screen_pos.xy * params.proj_info.xy + params.proj_info.zw), z); + } else { + return vec3((screen_pos.xy * params.proj_info.xy + params.proj_info.zw) * z, z); + } + } +} diff --git a/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection_scale.glsl b/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection_scale.glsl new file mode 100644 index 0000000000..a7da0812df --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/screen_space_reflection_scale.glsl @@ -0,0 +1,106 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +/* Specialization Constants (Toggles) */ + +layout(constant_id = 0) const bool sc_multiview = false; + +/* inputs */ +layout(set = 0, binding = 0) uniform sampler2D source_ssr; +layout(set = 1, binding = 0) uniform sampler2D source_depth; +layout(set = 1, binding = 1) uniform sampler2D source_normal; +layout(rgba16f, set = 2, binding = 0) uniform restrict writeonly image2D dest_ssr; +layout(r32f, set = 3, binding = 0) uniform restrict writeonly image2D dest_depth; +layout(rgba8, set = 3, binding = 1) uniform restrict writeonly image2D dest_normal; + +layout(push_constant, std430) uniform Params { + ivec2 screen_size; + float camera_z_near; + float camera_z_far; + + bool orthogonal; + bool filtered; + uint pad[2]; +} +params; + +void main() { + // Pixel being shaded + ivec2 ssC = ivec2(gl_GlobalInvocationID.xy); + + if (any(greaterThanEqual(ssC.xy, params.screen_size))) { //too large, do nothing + return; + } + //do not filter, SSR will generate arctifacts if this is done + + float divisor = 0.0; + vec4 color; + float depth; + vec4 normal; + + if (params.filtered) { + color = vec4(0.0); + depth = 0.0; + normal = vec4(0.0); + + for (int i = 0; i < 4; i++) { + ivec2 ofs = ssC << 1; + if (bool(i & 1)) { + ofs.x += 1; + } + if (bool(i & 2)) { + ofs.y += 1; + } + color += texelFetch(source_ssr, ofs, 0); + float d = texelFetch(source_depth, ofs, 0).r; + vec4 nr = texelFetch(source_normal, ofs, 0); + normal.xyz += nr.xyz * 2.0 - 1.0; + normal.w += nr.w; + + if (sc_multiview) { + // we're doing a full unproject so we need the value as is. + depth += d; + } else { + // unproject our Z value so we can use it directly. + d = d * 2.0 - 1.0; + if (params.orthogonal) { + d = ((d + (params.camera_z_far + params.camera_z_near) / (params.camera_z_far - params.camera_z_near)) * (params.camera_z_far - params.camera_z_near)) / 2.0; + } else { + d = 2.0 * params.camera_z_near * params.camera_z_far / (params.camera_z_far + params.camera_z_near - d * (params.camera_z_far - params.camera_z_near)); + } + depth += -d; + } + } + + color /= 4.0; + depth /= 4.0; + normal.xyz = normalize(normal.xyz / 4.0) * 0.5 + 0.5; + normal.w /= 4.0; + } else { + ivec2 ofs = ssC << 1; + + color = texelFetch(source_ssr, ofs, 0); + depth = texelFetch(source_depth, ofs, 0).r; + normal = texelFetch(source_normal, ofs, 0); + + if (!sc_multiview) { + // unproject our Z value so we can use it directly. + depth = depth * 2.0 - 1.0; + if (params.orthogonal) { + depth = ((depth + (params.camera_z_far + params.camera_z_near) / (params.camera_z_far - params.camera_z_near)) * (params.camera_z_far - params.camera_z_near)) / 2.0; + } else { + 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)); + } + depth = -depth; + } + } + + imageStore(dest_ssr, ssC, color); + imageStore(dest_depth, ssC, vec4(depth)); + imageStore(dest_normal, ssC, normal); +} diff --git a/servers/rendering/renderer_rd/shaders/effects/specular_merge.glsl b/servers/rendering/renderer_rd/shaders/effects/specular_merge.glsl new file mode 100644 index 0000000000..c62144fdaf --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/specular_merge.glsl @@ -0,0 +1,112 @@ +#[vertex] + +#version 450 + +#VERSION_DEFINES + +#if defined(USE_MULTIVIEW) && defined(has_VK_KHR_multiview) +#extension GL_EXT_multiview : enable +#endif + +#ifdef USE_MULTIVIEW +#ifdef has_VK_KHR_multiview +#define ViewIndex gl_ViewIndex +#else // has_VK_KHR_multiview +// !BAS! This needs to become an input once we implement our fallback! +#define ViewIndex 0 +#endif // has_VK_KHR_multiview +#else // USE_MULTIVIEW +// Set to zero, not supported in non stereo +#define ViewIndex 0 +#endif //USE_MULTIVIEW + +#ifdef USE_MULTIVIEW +layout(location = 0) out vec3 uv_interp; +#else // USE_MULTIVIEW +layout(location = 0) out vec2 uv_interp; +#endif //USE_MULTIVIEW + +void main() { + vec2 base_arr[4] = vec2[](vec2(0.0, 0.0), vec2(0.0, 1.0), vec2(1.0, 1.0), vec2(1.0, 0.0)); + +#ifdef USE_MULTIVIEW + uv_interp = vec3(base_arr[gl_VertexIndex], ViewIndex); + + gl_Position = vec4(uv_interp.xy * 2.0 - 1.0, 0.0, 1.0); +#else + uv_interp = base_arr[gl_VertexIndex]; + + gl_Position = vec4(uv_interp * 2.0 - 1.0, 0.0, 1.0); +#endif +} + +#[fragment] + +#version 450 + +#VERSION_DEFINES + +#if defined(USE_MULTIVIEW) && defined(has_VK_KHR_multiview) +#extension GL_EXT_multiview : enable +#endif + +#ifdef USE_MULTIVIEW +#ifdef has_VK_KHR_multiview +#define ViewIndex gl_ViewIndex +#else // has_VK_KHR_multiview +// !BAS! This needs to become an input once we implement our fallback! +#define ViewIndex 0 +#endif // has_VK_KHR_multiview +#else // USE_MULTIVIEW +// Set to zero, not supported in non stereo +#define ViewIndex 0 +#endif //USE_MULTIVIEW + +#ifdef USE_MULTIVIEW +layout(location = 0) in vec3 uv_interp; +#else // USE_MULTIVIEW +layout(location = 0) in vec2 uv_interp; +#endif //USE_MULTIVIEW + +#ifdef USE_MULTIVIEW +layout(set = 0, binding = 0) uniform sampler2DArray specular; +#else // USE_MULTIVIEW +layout(set = 0, binding = 0) uniform sampler2D specular; +#endif //USE_MULTIVIEW + +#ifdef MODE_SSR + +#ifdef USE_MULTIVIEW +layout(set = 1, binding = 0) uniform sampler2DArray ssr; +#else // USE_MULTIVIEW +layout(set = 1, binding = 0) uniform sampler2D ssr; +#endif //USE_MULTIVIEW + +#endif + +#ifdef MODE_MERGE + +#ifdef USE_MULTIVIEW +layout(set = 2, binding = 0) uniform sampler2DArray diffuse; +#else // USE_MULTIVIEW +layout(set = 2, binding = 0) uniform sampler2D diffuse; +#endif //USE_MULTIVIEW + +#endif + +layout(location = 0) out vec4 frag_color; + +void main() { + frag_color.rgb = texture(specular, uv_interp).rgb; + frag_color.a = 0.0; +#ifdef MODE_SSR + + vec4 ssr_color = texture(ssr, uv_interp); + frag_color.rgb = mix(frag_color.rgb, ssr_color.rgb, ssr_color.a); +#endif + +#ifdef MODE_MERGE + frag_color += texture(diffuse, uv_interp); +#endif + //added using additive blend +} diff --git a/servers/rendering/renderer_rd/shaders/effects/ss_effects_downsample.glsl b/servers/rendering/renderer_rd/shaders/effects/ss_effects_downsample.glsl new file mode 100644 index 0000000000..b1ff46dd3b --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/ss_effects_downsample.glsl @@ -0,0 +1,229 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// 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; + +layout(push_constant, std430) uniform Params { + vec2 pixel_size; + float z_far; + float z_near; + bool orthogonal; + float radius_sq; + uvec2 pad; +} +params; + +layout(set = 0, binding = 0) uniform sampler2D source_depth; + +layout(r16f, set = 1, binding = 0) uniform restrict writeonly image2DArray dest_image0; //rename +#ifdef GENERATE_MIPS +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) { + if (params.orthogonal) { + vec4 depth = p_depth * 2.0 - 1.0; + return ((depth + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0; + } + + float depth_linearize_mul = params.z_near; + float depth_linearize_add = params.z_far; + + // Optimized version of "-cameraClipNear / (cameraClipFar - projDepth * (cameraClipFar - cameraClipNear)) * cameraClipFar" + + // Set your depth_linearize_mul and depth_linearize_add to: + // depth_linearize_mul = ( cameraClipFar * cameraClipNear) / ( cameraClipFar - cameraClipNear ); + // depth_linearize_add = cameraClipFar / ( cameraClipFar - cameraClipNear ); + + return depth_linearize_mul / (depth_linearize_add - p_depth); +} + +float screen_space_to_view_space_depth(float p_depth) { + if (params.orthogonal) { + float depth = p_depth * 2.0 - 1.0; + return ((depth + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / (2.0 * params.z_far); + } + + float depth_linearize_mul = params.z_near; + float depth_linearize_add = params.z_far; + + return depth_linearize_mul / (depth_linearize_add - p_depth); +} + +#ifdef GENERATE_MIPS + +shared float depth_buffer[4][8][8]; + +float mip_smart_average(vec4 p_depths) { + float closest = min(min(p_depths.x, p_depths.y), min(p_depths.z, p_depths.w)); + float fallof_sq = -1.0f / params.radius_sq; + vec4 dists = p_depths - closest.xxxx; + vec4 weights = clamp(dists * dists * fallof_sq + 1.0, 0.0, 1.0); + return dot(weights, p_depths) / dot(weights, vec4(1.0, 1.0, 1.0, 1.0)); +} + +void prepare_depths_and_mips(vec4 p_samples, uvec2 p_output_coord, uvec2 p_gtid) { + p_samples = screen_space_to_view_space_depth(p_samples); + + depth_buffer[0][p_gtid.x][p_gtid.y] = p_samples.w; + depth_buffer[1][p_gtid.x][p_gtid.y] = p_samples.z; + depth_buffer[2][p_gtid.x][p_gtid.y] = p_samples.x; + depth_buffer[3][p_gtid.x][p_gtid.y] = p_samples.y; + + imageStore(dest_image0, ivec3(p_output_coord.x, p_output_coord.y, 0), vec4(p_samples.w)); + imageStore(dest_image0, ivec3(p_output_coord.x, p_output_coord.y, 1), vec4(p_samples.z)); + imageStore(dest_image0, ivec3(p_output_coord.x, p_output_coord.y, 2), vec4(p_samples.x)); + imageStore(dest_image0, ivec3(p_output_coord.x, p_output_coord.y, 3), vec4(p_samples.y)); + + uint depth_array_index = 2 * (p_gtid.y % 2) + (p_gtid.x % 2); + uvec2 depth_array_offset = ivec2(p_gtid.x % 2, p_gtid.y % 2); + ivec2 buffer_coord = ivec2(p_gtid) - ivec2(depth_array_offset); + + p_output_coord /= 2; + groupMemoryBarrier(); + barrier(); + + // if (still_alive) <-- all threads alive here + { + 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 + 1]; + float sample_10 = depth_buffer[depth_array_index][buffer_coord.x + 1][buffer_coord.y + 0]; + float sample_11 = depth_buffer[depth_array_index][buffer_coord.x + 1][buffer_coord.y + 1]; + + float avg = mip_smart_average(vec4(sample_00, sample_01, sample_10, sample_11)); + imageStore(dest_image1, ivec3(p_output_coord.x, p_output_coord.y, depth_array_index), vec4(avg)); + depth_buffer[depth_array_index][buffer_coord.x][buffer_coord.y] = avg; + } + + bool still_alive = p_gtid.x % 4 == depth_array_offset.x && p_gtid.y % 4 == 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 + 2]; + float sample_10 = depth_buffer[depth_array_index][buffer_coord.x + 2][buffer_coord.y + 0]; + float sample_11 = depth_buffer[depth_array_index][buffer_coord.x + 2][buffer_coord.y + 2]; + + float avg = mip_smart_average(vec4(sample_00, sample_01, sample_10, sample_11)); + imageStore(dest_image2, ivec3(p_output_coord.x, p_output_coord.y, depth_array_index), vec4(avg)); + depth_buffer[depth_array_index][buffer_coord.x][buffer_coord.y] = avg; + } + + still_alive = p_gtid.x % 8 == depth_array_offset.x && depth_array_offset.y % 8 == 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 + 4]; + float sample_10 = depth_buffer[depth_array_index][buffer_coord.x + 4][buffer_coord.y + 0]; + float sample_11 = depth_buffer[depth_array_index][buffer_coord.x + 4][buffer_coord.y + 4]; + + 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 +void prepare_depths(vec4 p_samples, uvec2 p_tid) { + p_samples = screen_space_to_view_space_depth(p_samples); + + imageStore(dest_image0, ivec3(p_tid, 0), vec4(p_samples.w)); + imageStore(dest_image0, ivec3(p_tid, 1), vec4(p_samples.z)); + imageStore(dest_image0, ivec3(p_tid, 2), vec4(p_samples.x)); + imageStore(dest_image0, ivec3(p_tid, 3), vec4(p_samples.y)); +} +#endif +#endif + +void main() { +#ifdef USE_HALF_BUFFERS +#ifdef USE_HALF_SIZE + float sample_00 = texelFetch(source_depth, ivec2(4 * gl_GlobalInvocationID.x + 0, 4 * gl_GlobalInvocationID.y + 0), 0).x; + float sample_11 = texelFetch(source_depth, ivec2(4 * gl_GlobalInvocationID.x + 2, 4 * gl_GlobalInvocationID.y + 2), 0).x; +#else + float sample_00 = texelFetch(source_depth, ivec2(2 * gl_GlobalInvocationID.x + 0, 2 * gl_GlobalInvocationID.y + 0), 0).x; + float sample_11 = texelFetch(source_depth, ivec2(2 * gl_GlobalInvocationID.x + 1, 2 * gl_GlobalInvocationID.y + 1), 0).x; +#endif + sample_00 = screen_space_to_view_space_depth(sample_00); + sample_11 = screen_space_to_view_space_depth(sample_11); + + imageStore(dest_image0, ivec3(gl_GlobalInvocationID.xy, 0), vec4(sample_00)); + imageStore(dest_image0, ivec3(gl_GlobalInvocationID.xy, 3), vec4(sample_11)); +#else //!USE_HALF_BUFFERS +#ifdef USE_HALF_SIZE + ivec2 depth_buffer_coord = 4 * ivec2(gl_GlobalInvocationID.xy); + ivec2 output_coord = ivec2(gl_GlobalInvocationID); + + vec2 uv = (vec2(depth_buffer_coord) + 0.5f) * params.pixel_size; + vec4 samples; + samples.x = textureLodOffset(source_depth, uv, 0, ivec2(0, 2)).x; + samples.y = textureLodOffset(source_depth, uv, 0, ivec2(2, 2)).x; + samples.z = textureLodOffset(source_depth, uv, 0, ivec2(2, 0)).x; + samples.w = textureLodOffset(source_depth, uv, 0, ivec2(0, 0)).x; +#else + ivec2 depth_buffer_coord = 2 * ivec2(gl_GlobalInvocationID.xy); + ivec2 output_coord = ivec2(gl_GlobalInvocationID); + + vec2 uv = (vec2(depth_buffer_coord) + 0.5f) * params.pixel_size; + vec4 samples = textureGather(source_depth, uv); +#endif +#ifdef GENERATE_MIPS + prepare_depths_and_mips(samples, output_coord, gl_LocalInvocationID.xy); +#else + prepare_depths(samples, gl_GlobalInvocationID.xy); +#endif +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/effects/ssao.glsl b/servers/rendering/renderer_rd/shaders/effects/ssao.glsl new file mode 100644 index 0000000000..ffaa6872c9 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/ssao.glsl @@ -0,0 +1,483 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// 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 + +#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), + 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 SSAO_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 SSAO_TILT_SAMPLES_ENABLE_AT_QUALITY_PRESET (99) // to disable simply set to 99 or similar +#define SSAO_TILT_SAMPLES_AMOUNT (0.4) +// +#define SSAO_HALOING_REDUCTION_ENABLE_AT_QUALITY_PRESET (1) // to disable simply set to 99 or similar +#define SSAO_HALOING_REDUCTION_AMOUNT (0.6) // values from 0.0 - 1.0, 1.0 means max weighting (will cause artifacts, 0.8 is more reasonable) +// +#define SSAO_NORMAL_BASED_EDGES_ENABLE_AT_QUALITY_PRESET (2) // to disable simply set to 99 or similar +#define SSAO_NORMAL_BASED_EDGES_DOT_THRESHOLD (0.5) // use 0-0.1 for super-sharp normal-based edges +// +#define SSAO_DETAIL_AO_ENABLE_AT_QUALITY_PRESET (1) // whether to use detail; to disable simply set to 99 or similar +// +#define SSAO_DEPTH_MIPS_ENABLE_AT_QUALITY_PRESET (2) // !!warning!! the MIP generation on the C++ side will be enabled on quality preset 2 regardless of this value, so if changing here, change the C++ side too +#define SSAO_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 SSAO_DEPTH_BASED_EDGES_ENABLE_AT_QUALITY_PRESET (1) +// +#define SSAO_REDUCE_RADIUS_NEAR_SCREEN_BORDER_ENABLE_AT_QUALITY_PRESET (1) + +#define SSAO_MAX_TAPS 32 +#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 + +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(rg8, set = 1, binding = 0) uniform restrict readonly image2DArray source_ssao; +layout(set = 1, binding = 1) uniform sampler2D source_importance; +layout(set = 1, binding = 2, std430) buffer Counter { + uint sum; +} +counter; +#endif + +layout(rg8, set = 2, binding = 0) uniform restrict writeonly image2D dest_image; + +// This push_constant is full - 128 bytes - if you need to add more data, consider adding to the uniform buffer instead +layout(push_constant, std430) uniform Params { + ivec2 screen_size; + int pass; + int quality; + + vec2 half_screen_pixel_size; + int size_multiplier; + float detail_intensity; + + vec2 NDC_to_view_mul; + vec2 NDC_to_view_add; + + vec2 pad2; + vec2 half_screen_pixel_size_x025; + + float radius; + float intensity; + float shadow_power; + float shadow_clamp; + + float fade_out_mul; + float fade_out_add; + float horizon_angle_threshold; + 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; + +// packing/unpacking for edges; 2 bits per edge mean 4 gradient values (0, 0.33, 0.66, 1) for smoother transitions! +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 - params.horizon_angle_threshold) * falloff_mult; +} + +void SSAO_tap_inner(const int p_quality_level, 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; + + // convert to viewspace + vec3 hit_pos = NDC_to_view_space(p_sampling_uv.xy, viewspace_sample_z).xyz; + 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 >= SSAO_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 = SSAO_HALOING_REDUCTION_AMOUNT * reduct + (1.0 - SSAO_HALOING_REDUCTION_AMOUNT); + } + weight *= p_weight_mod; + r_obscurance_sum += obscurance * weight; + r_weight_sum += weight; +} + +void SSAOTap(const int p_quality_level, 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 center, similar to as described + // in http://graphics.cs.williams.edu/papers/SAOHPG12/. + float mip_level = (p_quality_level < SSAO_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; + + SSAO_tap_inner(p_quality_level, 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 >= SSAO_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; + + SSAO_tap_inner(p_quality_level, 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_SSAO_shadows_internal(out float r_shadow_term, out vec4 r_edges, 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) ? (SSAO_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 neighboring 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 >= SSAO_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 + float obscurance_sum = 0.0; + float weight_sum = 0.0; + + // edge mask for between this and left/right/top/bottom neighbor 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 >= SSAO_DEPTH_BASED_EDGES_ENABLE_AT_QUALITY_PRESET)) { + edgesLRTB = calculate_edges(pix_z, pix_left_z, pix_right_z, pix_top_z, pix_bottom_z); + } + + // adds a more high definition sharp effect, which gets blurred out (reuses left/right/top/bottom samples that we used for edge detection) + if (!p_adaptive_base && (p_quality_level >= SSAO_DETAIL_AO_ENABLE_AT_QUALITY_PRESET)) { + // disable in case of quality level 4 (reference) + if (p_quality_level != 4) { + //approximate neighboring pixels positions (actually just deltas or "positions - pix_center_pos" ) + vec3 normalized_viewspace_dir = vec3(pix_center_pos.xy / pix_center_pos.zz, 1.0); + vec3 pixel_left_delta = vec3(-pixel_size_at_center.x, 0.0, 0.0) + normalized_viewspace_dir * (pix_left_z - pix_center_pos.z); + vec3 pixel_right_delta = vec3(+pixel_size_at_center.x, 0.0, 0.0) + normalized_viewspace_dir * (pix_right_z - pix_center_pos.z); + vec3 pixel_top_delta = vec3(0.0, -pixel_size_at_center.y, 0.0) + normalized_viewspace_dir * (pix_top_z - pix_center_pos.z); + vec3 pixel_bottom_delta = vec3(0.0, +pixel_size_at_center.y, 0.0) + normalized_viewspace_dir * (pix_bottom_z - pix_center_pos.z); + + const float range_reduction = 4.0f; // this is to avoid various artifacts + const float modified_fallof_sq = range_reduction * fallof_sq; + + vec4 additional_obscurance; + additional_obscurance.x = calculate_pixel_obscurance(pixel_normal, pixel_left_delta, modified_fallof_sq); + additional_obscurance.y = calculate_pixel_obscurance(pixel_normal, pixel_right_delta, modified_fallof_sq); + additional_obscurance.z = calculate_pixel_obscurance(pixel_normal, pixel_top_delta, modified_fallof_sq); + additional_obscurance.w = calculate_pixel_obscurance(pixel_normal, pixel_bottom_delta, modified_fallof_sq); + + obscurance_sum += params.detail_intensity * dot(additional_obscurance, edgesLRTB); + } + } + + // Sharp normals also create edges - but this adds to the cost as well + if (!p_adaptive_base && (p_quality_level >= SSAO_NORMAL_BASED_EDGES_ENABLE_AT_QUALITY_PRESET)) { + vec3 neighbour_normal_left = load_normal(ivec2(full_res_coord), ivec2(-2, 0)); + vec3 neighbour_normal_right = load_normal(ivec2(full_res_coord), ivec2(2, 0)); + vec3 neighbour_normal_top = load_normal(ivec2(full_res_coord), ivec2(0, -2)); + vec3 neighbour_normal_bottom = load_normal(ivec2(full_res_coord), ivec2(0, 2)); + + const float dot_threshold = SSAO_NORMAL_BASED_EDGES_DOT_THRESHOLD; + + vec4 normal_edgesLRTB; + normal_edgesLRTB.x = clamp((dot(pixel_normal, neighbour_normal_left) + dot_threshold), 0.0, 1.0); + normal_edgesLRTB.y = clamp((dot(pixel_normal, neighbour_normal_right) + dot_threshold), 0.0, 1.0); + normal_edgesLRTB.z = clamp((dot(pixel_normal, neighbour_normal_top) + dot_threshold), 0.0, 1.0); + normal_edgesLRTB.w = clamp((dot(pixel_normal, neighbour_normal_bottom) + dot_threshold), 0.0, 1.0); + + edgesLRTB *= normal_edgesLRTB; + } + + const float global_mip_offset = SSAO_DEPTH_MIPS_GLOBAL_OFFSET; + float mip_offset = (p_quality_level < SSAO_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 *= SSAO_TILT_SAMPLES_AMOUNT; + + // standard, non-adaptive approach + if ((p_quality_level != 3) || p_adaptive_base) { + for (int i = 0; i < number_of_taps; i++) { + SSAOTap(p_quality_level, 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; + + // this is to normalize SSAO_DETAIL_AO_AMOUNT across all pixel regardless of importance + obscurance_sum *= (SSAO_ADAPTIVE_TAP_BASE_COUNT / float(SSAO_MAX_TAPS)) + (importance * SSAO_ADAPTIVE_TAP_FLEXIBLE_COUNT / float(SSAO_MAX_TAPS)); + + // load existing base values + vec2 base_values = imageLoad(source_ssao, ivec3(upos, params.pass)).xy; + weight_sum += base_values.y * float(SSAO_ADAPTIVE_TAP_BASE_COUNT * 4.0); + obscurance_sum += (base_values.x) * 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 = SSAO_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(SSAO_MAX_TAPS, additional_samples + SSAO_ADAPTIVE_TAP_BASE_COUNT); + + for (uint i = SSAO_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); + SSAOTap(p_quality_level, 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 - just output weight (used for the next pass) + if (p_adaptive_base) { + float obscurance = obscurance_sum / weight_sum; + + r_shadow_term = obscurance; + r_edges = vec4(0.0); + r_weight = weight_sum; + return; + } + + // calculate weighted average + float obscurance = obscurance_sum / weight_sum; + + // 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 SSAO shadowing 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 >= SSAO_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); + } + + // strength + obscurance = params.intensity * obscurance; + + // clamp + obscurance = min(obscurance, params.shadow_clamp); + + // fadeout + obscurance *= fade_out; + + // conceptually switch to occlusion with the meaning being visibility (grows with visibility, occlusion == 1 implies full visibility), + // to be in line with what is more commonly used. + float occlusion = 1.0 - obscurance; + + // modify the gradient + // note: this cannot be moved to a later pass because of loss of precision after storing in the render target + occlusion = pow(clamp(occlusion, 0.0, 1.0), params.shadow_power); + + // outputs! + r_shadow_term = occlusion; // Our final 'occlusion' term (0 means fully occluded, 1 means fully lit) + 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_weight = weight_sum; +} + +void main() { + float out_shadow_term; + 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 SSAO_BASE + generate_SSAO_shadows_internal(out_shadow_term, out_edges, out_weight, uv, params.quality, true); + + imageStore(dest_image, ivec2(gl_GlobalInvocationID.xy), vec4(out_shadow_term, out_weight / (float(SSAO_ADAPTIVE_TAP_BASE_COUNT) * 4.0), 0.0, 0.0)); +#else + generate_SSAO_shadows_internal(out_shadow_term, out_edges, out_weight, uv, params.quality, false); // pass in quality levels + if (params.quality == 0) { + out_edges = vec4(1.0); + } + + imageStore(dest_image, ivec2(gl_GlobalInvocationID.xy), vec4(out_shadow_term, pack_edges(out_edges), 0.0, 0.0)); +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/effects/ssao_blur.glsl b/servers/rendering/renderer_rd/shaders/effects/ssao_blur.glsl new file mode 100644 index 0000000000..f42734c46d --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/ssao_blur.glsl @@ -0,0 +1,154 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// 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; + +layout(set = 0, binding = 0) uniform sampler2D source_ssao; + +layout(rg8, set = 1, binding = 0) uniform restrict writeonly image2D dest_image; + +layout(push_constant, 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(float p_ssao_value, float p_edge_value, inout float r_sum, inout float r_sum_weight) { + float weight = p_edge_value; + + r_sum += (weight * p_ssao_value); + r_sum_weight += weight; +} + +#ifdef MODE_WIDE +vec2 sample_blurred_wide(vec2 p_coord) { + vec2 vC = textureLodOffset(source_ssao, vec2(p_coord), 0.0, ivec2(0, 0)).xy; + vec2 vL = textureLodOffset(source_ssao, vec2(p_coord), 0.0, ivec2(-2, 0)).xy; + vec2 vT = textureLodOffset(source_ssao, vec2(p_coord), 0.0, ivec2(0, -2)).xy; + vec2 vR = textureLodOffset(source_ssao, vec2(p_coord), 0.0, ivec2(2, 0)).xy; + vec2 vB = textureLodOffset(source_ssao, vec2(p_coord), 0.0, ivec2(0, 2)).xy; + + float packed_edges = vC.y; + vec4 edgesLRTB = unpack_edges(packed_edges); + edgesLRTB.x *= unpack_edges(vL.y).y; + edgesLRTB.z *= unpack_edges(vT.y).w; + edgesLRTB.y *= unpack_edges(vR.y).x; + edgesLRTB.w *= unpack_edges(vB.y).z; + + float ssao_value = vC.x; + float ssao_valueL = vL.x; + float ssao_valueT = vT.x; + float ssao_valueR = vR.x; + float ssao_valueB = vB.x; + + float sum_weight = 0.8f; + float sum = ssao_value * sum_weight; + + add_sample(ssao_valueL, edgesLRTB.x, sum, sum_weight); + add_sample(ssao_valueR, edgesLRTB.y, sum, sum_weight); + add_sample(ssao_valueT, edgesLRTB.z, sum, sum_weight); + add_sample(ssao_valueB, edgesLRTB.w, sum, sum_weight); + + float ssao_avg = sum / sum_weight; + + ssao_value = ssao_avg; + + return vec2(ssao_value, packed_edges); +} +#endif + +#ifdef MODE_SMART +vec2 sample_blurred(vec3 p_pos, vec2 p_coord) { + float packed_edges = texelFetch(source_ssao, ivec2(p_pos.xy), 0).y; + vec4 edgesLRTB = unpack_edges(packed_edges); + + vec4 valuesUL = textureGather(source_ssao, vec2(p_coord - params.half_screen_pixel_size * 0.5)); + vec4 valuesBR = textureGather(source_ssao, vec2(p_coord + params.half_screen_pixel_size * 0.5)); + + float ssao_value = valuesUL.y; + float ssao_valueL = valuesUL.x; + float ssao_valueT = valuesUL.z; + float ssao_valueR = valuesBR.z; + float ssao_valueB = valuesBR.x; + + float sum_weight = 0.5; + float sum = ssao_value * sum_weight; + + add_sample(ssao_valueL, edgesLRTB.x, sum, sum_weight); + add_sample(ssao_valueR, edgesLRTB.y, sum, sum_weight); + + add_sample(ssao_valueT, edgesLRTB.z, sum, sum_weight); + add_sample(ssao_valueB, edgesLRTB.w, sum, sum_weight); + + float ssao_avg = sum / sum_weight; + + ssao_value = ssao_avg; + + return vec2(ssao_value, packed_edges); +} +#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; + + vec2 center = textureLod(source_ssao, vec2(uv), 0.0).xy; + + vec4 vals; + 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)) + center.x * 0.2, center.y); + +#else +#ifdef MODE_SMART + vec2 sampled = sample_blurred(vec3(gl_GlobalInvocationID), (vec2(gl_GlobalInvocationID.xy) + vec2(0.5, 0.5)) * params.half_screen_pixel_size); +#else // MODE_WIDE + vec2 sampled = sample_blurred_wide((vec2(gl_GlobalInvocationID.xy) + vec2(0.5, 0.5)) * params.half_screen_pixel_size); +#endif + +#endif + imageStore(dest_image, ivec2(ssC), vec4(sampled, 0.0, 0.0)); +} diff --git a/servers/rendering/renderer_rd/shaders/effects/ssao_importance_map.glsl b/servers/rendering/renderer_rd/shaders/effects/ssao_importance_map.glsl new file mode 100644 index 0000000000..d234ab4417 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/ssao_importance_map.glsl @@ -0,0 +1,123 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// 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, std430) uniform Params { + vec2 half_screen_pixel_size; + float intensity; + float power; +} +params; + +void main() { + // Pixel being shaded + ivec2 ssC = ivec2(gl_GlobalInvocationID.xy); + +#ifdef GENERATE_MAP + // importance map stuff + uvec2 base_position = ssC * 2; + + vec2 base_uv = (vec2(base_position) + vec2(0.5f, 0.5f)) * params.half_screen_pixel_size; + + float minV = 1.0; + float maxV = 0.0; + for (int i = 0; i < 4; 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; + + vals = 1 - vals; + + vals = pow(clamp(vals, 0.0, 1.0), vec4(params.power)); + + 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))); + } + + float min_max_diff = maxV - minV; + + imageStore(dest_image, ssC, vec4(pow(clamp(min_max_diff * 2.0, 0.0, 1.0), 0.8))); +#endif + +#ifdef PROCESS_MAPA + vec2 uv = (vec2(ssC) + 0.5f) * params.half_screen_pixel_size * 2.0; + + float center = 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 center = 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/effects/ssao_interleave.glsl b/servers/rendering/renderer_rd/shaders/effects/ssao_interleave.glsl new file mode 100644 index 0000000000..45cc62d361 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/ssao_interleave.glsl @@ -0,0 +1,119 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// 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; + +layout(rgba8, set = 0, binding = 0) uniform restrict writeonly image2D dest_image; +layout(set = 1, binding = 0) uniform sampler2DArray source_texture; + +layout(push_constant, 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 + float ao; + 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; // neighboring, horizontal + int index_vertical = mx + (1 - my) * 2; // neighboring, vertical + int index_diagonal = (1 - mx) + (1 - my) * 2; // diagonal + + vec2 center_val = texelFetch(source_texture, ivec3(pix_pos / uvec2(params.size_modifier), index_center), 0).xy; + + ao = center_val.x; + + vec4 edgesLRTB = unpack_edges(center_val.y); + + // 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; + float ao_horizontal = textureLod(source_texture, vec3(uv_horizontal, index_horizontal), 0.0).x; + vec2 uv_vertical = (gl_GlobalInvocationID.xy + vec2(0.5) + vec2(0.5 - fmx, fmy - 0.5 + fmye)) * params.pixel_size; + float ao_vertical = textureLod(source_texture, vec3(uv_vertical, index_vertical), 0.0).x; + vec2 uv_diagonal = (gl_GlobalInvocationID.xy + vec2(0.5) + vec2(fmx - 0.5 + fmxe, fmy - 0.5 + fmye)) * params.pixel_size; + float ao_diagonal = textureLod(source_texture, vec3(uv_diagonal, index_diagonal), 0.0).x; + + // 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)); + ao = dot(vec4(ao, ao_horizontal, ao_vertical, ao_diagonal), blendWeights) / blendWeightsSum; + + imageStore(dest_image, ivec2(gl_GlobalInvocationID.xy), vec4(ao)); +#else // !MODE_SMART + + vec2 uv = (gl_GlobalInvocationID.xy + vec2(0.5)) * params.pixel_size; +#ifdef MODE_HALF + float a = textureLod(source_texture, vec3(uv, 0), 0.0).x; + float d = textureLod(source_texture, vec3(uv, 3), 0.0).x; + float avg = (a + d) * 0.5; + +#else + float a = textureLod(source_texture, vec3(uv, 0), 0.0).x; + float b = textureLod(source_texture, vec3(uv, 1), 0.0).x; + float c = textureLod(source_texture, vec3(uv, 2), 0.0).x; + float d = textureLod(source_texture, vec3(uv, 3), 0.0).x; + float avg = (a + b + c + d) * 0.25; + +#endif + imageStore(dest_image, ivec2(gl_GlobalInvocationID.xy), vec4(avg)); +#endif +} diff --git a/servers/rendering/renderer_rd/shaders/effects/ssil.glsl b/servers/rendering/renderer_rd/shaders/effects/ssil.glsl new file mode 100644 index 0000000000..de7b97953f --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/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, 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 center, 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 neighboring 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 neighbor 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/effects/ssil_blur.glsl b/servers/rendering/renderer_rd/shaders/effects/ssil_blur.glsl new file mode 100644 index 0000000000..f48e6c4341 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/ssil_blur.glsl @@ -0,0 +1,144 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// 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 + +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, 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 center = 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 + center * 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/effects/ssil_importance_map.glsl b/servers/rendering/renderer_rd/shaders/effects/ssil_importance_map.glsl new file mode 100644 index 0000000000..193e3458ab --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/ssil_importance_map.glsl @@ -0,0 +1,125 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// 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 + +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, 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 center = 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 center = 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/effects/ssil_interleave.glsl b/servers/rendering/renderer_rd/shaders/effects/ssil_interleave.glsl new file mode 100644 index 0000000000..ed85b8ee4c --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/ssil_interleave.glsl @@ -0,0 +1,122 @@ +/////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// 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 + +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, 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; // neighboring, horizontal + int index_vertical = mx + (1 - my) * 2; // neighboring, 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/effects/subsurface_scattering.glsl b/servers/rendering/renderer_rd/shaders/effects/subsurface_scattering.glsl new file mode 100644 index 0000000000..fb35d3cde6 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/subsurface_scattering.glsl @@ -0,0 +1,189 @@ +#[compute] + +#version 450 + +#VERSION_DEFINES + +layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in; + +#ifdef USE_25_SAMPLES +const int kernel_size = 13; + +const vec2 kernel[kernel_size] = vec2[]( + vec2(0.530605, 0.0), + vec2(0.0211412, 0.0208333), + vec2(0.0402784, 0.0833333), + vec2(0.0493588, 0.1875), + vec2(0.0410172, 0.333333), + vec2(0.0263642, 0.520833), + vec2(0.017924, 0.75), + vec2(0.0128496, 1.02083), + vec2(0.0094389, 1.33333), + vec2(0.00700976, 1.6875), + vec2(0.00500364, 2.08333), + vec2(0.00333804, 2.52083), + vec2(0.000973794, 3.0)); + +const vec4 skin_kernel[kernel_size] = vec4[]( + vec4(0.530605, 0.613514, 0.739601, 0), + vec4(0.0211412, 0.0459286, 0.0378196, 0.0208333), + vec4(0.0402784, 0.0657244, 0.04631, 0.0833333), + vec4(0.0493588, 0.0367726, 0.0219485, 0.1875), + vec4(0.0410172, 0.0199899, 0.0118481, 0.333333), + vec4(0.0263642, 0.0119715, 0.00684598, 0.520833), + vec4(0.017924, 0.00711691, 0.00347194, 0.75), + vec4(0.0128496, 0.00356329, 0.00132016, 1.02083), + vec4(0.0094389, 0.00139119, 0.000416598, 1.33333), + vec4(0.00700976, 0.00049366, 0.000151938, 1.6875), + vec4(0.00500364, 0.00020094, 5.28848e-005, 2.08333), + vec4(0.00333804, 7.85443e-005, 1.2945e-005, 2.52083), + vec4(0.000973794, 1.11862e-005, 9.43437e-007, 3)); + +#endif //USE_25_SAMPLES + +#ifdef USE_17_SAMPLES +const int kernel_size = 9; +const vec2 kernel[kernel_size] = vec2[]( + vec2(0.536343, 0.0), + vec2(0.0324462, 0.03125), + vec2(0.0582416, 0.125), + vec2(0.0571056, 0.28125), + vec2(0.0347317, 0.5), + vec2(0.0216301, 0.78125), + vec2(0.0144609, 1.125), + vec2(0.0100386, 1.53125), + vec2(0.00317394, 2.0)); + +const vec4 skin_kernel[kernel_size] = vec4[]( + vec4(0.536343, 0.624624, 0.748867, 0), + vec4(0.0324462, 0.0656718, 0.0532821, 0.03125), + vec4(0.0582416, 0.0659959, 0.0411329, 0.125), + vec4(0.0571056, 0.0287432, 0.0172844, 0.28125), + vec4(0.0347317, 0.0151085, 0.00871983, 0.5), + vec4(0.0216301, 0.00794618, 0.00376991, 0.78125), + vec4(0.0144609, 0.00317269, 0.00106399, 1.125), + vec4(0.0100386, 0.000914679, 0.000275702, 1.53125), + vec4(0.00317394, 0.000134823, 3.77269e-005, 2)); +#endif //USE_17_SAMPLES + +#ifdef USE_11_SAMPLES +const int kernel_size = 6; +const vec2 kernel[kernel_size] = vec2[]( + vec2(0.560479, 0.0), + vec2(0.0771802, 0.08), + vec2(0.0821904, 0.32), + vec2(0.03639, 0.72), + vec2(0.0192831, 1.28), + vec2(0.00471691, 2.0)); + +const vec4 skin_kernel[kernel_size] = vec4[]( + + vec4(0.560479, 0.669086, 0.784728, 0), + vec4(0.0771802, 0.113491, 0.0793803, 0.08), + vec4(0.0821904, 0.0358608, 0.0209261, 0.32), + vec4(0.03639, 0.0130999, 0.00643685, 0.72), + vec4(0.0192831, 0.00282018, 0.00084214, 1.28), + vec4(0.00471691, 0.000184771, 5.07565e-005, 2)); + +#endif //USE_11_SAMPLES + +layout(push_constant, std430) uniform Params { + ivec2 screen_size; + float camera_z_far; + float camera_z_near; + + bool vertical; + bool orthogonal; + float unit_size; + float scale; + + float depth_scale; + uint pad[3]; +} +params; + +layout(set = 0, binding = 0) uniform sampler2D source_image; +layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly image2D dest_image; +layout(set = 2, binding = 0) uniform sampler2D source_depth; + +void do_filter(inout vec3 color_accum, inout vec3 divisor, vec2 uv, vec2 step, bool p_skin) { + // Accumulate the other samples: + for (int i = 1; i < kernel_size; i++) { + // Fetch color and depth for current sample: + vec2 offset = uv + kernel[i].y * step; + vec4 color = texture(source_image, offset); + + if (abs(color.a) < 0.001) { + break; //mix no more + } + + vec3 w; + if (p_skin) { + //skin + w = skin_kernel[i].rgb; + } else { + w = vec3(kernel[i].x); + } + + color_accum += color.rgb * w; + divisor += w; + } +} + +void main() { + // Pixel being shaded + ivec2 ssC = ivec2(gl_GlobalInvocationID.xy); + + if (any(greaterThanEqual(ssC, params.screen_size))) { //too large, do nothing + return; + } + + vec2 uv = (vec2(ssC) + 0.5) / vec2(params.screen_size); + + // Fetch color of current pixel: + vec4 base_color = texture(source_image, uv); + float strength = abs(base_color.a); + + if (strength > 0.0) { + vec2 dir = params.vertical ? vec2(0.0, 1.0) : vec2(1.0, 0.0); + + // Fetch linear depth of current pixel: + float depth = texture(source_depth, uv).r * 2.0 - 1.0; + float depth_scale; + + if (params.orthogonal) { + depth = ((depth + (params.camera_z_far + params.camera_z_near) / (params.camera_z_far - params.camera_z_near)) * (params.camera_z_far - params.camera_z_near)) / 2.0; + depth_scale = params.unit_size; //remember depth is negative by default in OpenGL + } else { + 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)); + depth_scale = params.unit_size / depth; //remember depth is negative by default in OpenGL + } + + float scale = mix(params.scale, depth_scale, params.depth_scale); + + // Calculate the final step to fetch the surrounding pixels: + vec2 step = scale * dir; + step *= strength; + step /= 3.0; + // Accumulate the center sample: + + vec3 divisor; + bool skin = bool(base_color.a < 0.0); + + if (skin) { + //skin + divisor = skin_kernel[0].rgb; + } else { + divisor = vec3(kernel[0].x); + } + + vec3 color = base_color.rgb * divisor; + + do_filter(color, divisor, uv, step, skin); + do_filter(color, divisor, uv, -step, skin); + + base_color.rgb = color / divisor; + } + + imageStore(dest_image, ssC, base_color); +} diff --git a/servers/rendering/renderer_rd/shaders/effects/taa_resolve.glsl b/servers/rendering/renderer_rd/shaders/effects/taa_resolve.glsl new file mode 100644 index 0000000000..02566d8e35 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/taa_resolve.glsl @@ -0,0 +1,396 @@ +/////////////////////////////////////////////////////////////////////////////////// +// Copyright(c) 2016-2022 Panos Karabelas +// +// 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) +// 2022-05-06: Panos Karabelas: first commit +// 2020-12-05: Joan Fons: convert to Vulkan and Godot +/////////////////////////////////////////////////////////////////////////////////// + +#[compute] + +#version 450 + +#VERSION_DEFINES + +// Based on Spartan Engine's TAA implementation (without TAA upscale). +// <https://github.com/PanosK92/SpartanEngine/blob/a8338d0609b85dc32f3732a5c27fb4463816a3b9/Data/shaders/temporal_antialiasing.hlsl> + +#ifndef MOLTENVK_USED +#define USE_SUBGROUPS +#endif // MOLTENVK_USED + +#define GROUP_SIZE 8 +#define FLT_MIN 0.00000001 +#define FLT_MAX 32767.0 +#define RPC_9 0.11111111111 +#define RPC_16 0.0625 + +#ifdef USE_SUBGROUPS +layout(local_size_x = GROUP_SIZE, local_size_y = GROUP_SIZE, local_size_z = 1) in; +#endif + +layout(rgba16f, set = 0, binding = 0) uniform restrict readonly image2D color_buffer; +layout(set = 0, binding = 1) uniform sampler2D depth_buffer; +layout(rg16f, set = 0, binding = 2) uniform restrict readonly image2D velocity_buffer; +layout(rg16f, set = 0, binding = 3) uniform restrict readonly image2D last_velocity_buffer; +layout(set = 0, binding = 4) uniform sampler2D history_buffer; +layout(rgba16f, set = 0, binding = 5) uniform restrict writeonly image2D output_buffer; + +layout(push_constant, std430) uniform Params { + vec2 resolution; + float disocclusion_threshold; // 0.1 / max(params.resolution.x, params.resolution.y + float disocclusion_scale; +} +params; + +const ivec2 kOffsets3x3[9] = { + ivec2(-1, -1), + ivec2(0, -1), + ivec2(1, -1), + ivec2(-1, 0), + ivec2(0, 0), + ivec2(1, 0), + ivec2(-1, 1), + ivec2(0, 1), + ivec2(1, 1), +}; + +/*------------------------------------------------------------------------------ + THREAD GROUP SHARED MEMORY (LDS) +------------------------------------------------------------------------------*/ + +const int kBorderSize = 1; +const int kGroupSize = GROUP_SIZE; +const int kTileDimension = kGroupSize + kBorderSize * 2; +const int kTileDimension2 = kTileDimension * kTileDimension; + +vec3 reinhard(vec3 hdr) { + return hdr / (hdr + 1.0); +} +vec3 reinhard_inverse(vec3 sdr) { + return sdr / (1.0 - sdr); +} + +float get_depth(ivec2 thread_id) { + return texelFetch(depth_buffer, thread_id, 0).r; +} + +#ifdef USE_SUBGROUPS +shared vec3 tile_color[kTileDimension][kTileDimension]; +shared float tile_depth[kTileDimension][kTileDimension]; + +vec3 load_color(uvec2 group_thread_id) { + group_thread_id += kBorderSize; + return tile_color[group_thread_id.x][group_thread_id.y]; +} + +void store_color(uvec2 group_thread_id, vec3 color) { + tile_color[group_thread_id.x][group_thread_id.y] = color; +} + +float load_depth(uvec2 group_thread_id) { + group_thread_id += kBorderSize; + return tile_depth[group_thread_id.x][group_thread_id.y]; +} + +void store_depth(uvec2 group_thread_id, float depth) { + tile_depth[group_thread_id.x][group_thread_id.y] = depth; +} + +void store_color_depth(uvec2 group_thread_id, ivec2 thread_id) { + // out of bounds clamp + thread_id = clamp(thread_id, ivec2(0, 0), ivec2(params.resolution) - ivec2(1, 1)); + + store_color(group_thread_id, imageLoad(color_buffer, thread_id).rgb); + store_depth(group_thread_id, get_depth(thread_id)); +} + +void populate_group_shared_memory(uvec2 group_id, uint group_index) { + // Populate group shared memory + ivec2 group_top_left = ivec2(group_id) * kGroupSize - kBorderSize; + if (group_index < (kTileDimension2 >> 2)) { + ivec2 group_thread_id_1 = ivec2(group_index % kTileDimension, group_index / kTileDimension); + ivec2 group_thread_id_2 = ivec2((group_index + (kTileDimension2 >> 2)) % kTileDimension, (group_index + (kTileDimension2 >> 2)) / kTileDimension); + ivec2 group_thread_id_3 = ivec2((group_index + (kTileDimension2 >> 1)) % kTileDimension, (group_index + (kTileDimension2 >> 1)) / kTileDimension); + ivec2 group_thread_id_4 = ivec2((group_index + kTileDimension2 * 3 / 4) % kTileDimension, (group_index + kTileDimension2 * 3 / 4) / kTileDimension); + + store_color_depth(group_thread_id_1, group_top_left + group_thread_id_1); + store_color_depth(group_thread_id_2, group_top_left + group_thread_id_2); + store_color_depth(group_thread_id_3, group_top_left + group_thread_id_3); + store_color_depth(group_thread_id_4, group_top_left + group_thread_id_4); + } + + // Wait for group threads to load store data. + groupMemoryBarrier(); + barrier(); +} +#else +vec3 load_color(uvec2 screen_pos) { + return imageLoad(color_buffer, ivec2(screen_pos)).rgb; +} + +float load_depth(uvec2 screen_pos) { + return get_depth(ivec2(screen_pos)); +} +#endif + +/*------------------------------------------------------------------------------ + VELOCITY +------------------------------------------------------------------------------*/ + +void depth_test_min(uvec2 pos, inout float min_depth, inout uvec2 min_pos) { + float depth = load_depth(pos); + + if (depth < min_depth) { + min_depth = depth; + min_pos = pos; + } +} + +// Returns velocity with closest depth (3x3 neighborhood) +void get_closest_pixel_velocity_3x3(in uvec2 group_pos, uvec2 group_top_left, out vec2 velocity) { + float min_depth = 1.0; + uvec2 min_pos = group_pos; + + depth_test_min(group_pos + kOffsets3x3[0], min_depth, min_pos); + depth_test_min(group_pos + kOffsets3x3[1], min_depth, min_pos); + depth_test_min(group_pos + kOffsets3x3[2], min_depth, min_pos); + depth_test_min(group_pos + kOffsets3x3[3], min_depth, min_pos); + depth_test_min(group_pos + kOffsets3x3[4], min_depth, min_pos); + depth_test_min(group_pos + kOffsets3x3[5], min_depth, min_pos); + depth_test_min(group_pos + kOffsets3x3[6], min_depth, min_pos); + depth_test_min(group_pos + kOffsets3x3[7], min_depth, min_pos); + depth_test_min(group_pos + kOffsets3x3[8], min_depth, min_pos); + + // Velocity out + velocity = imageLoad(velocity_buffer, ivec2(group_top_left + min_pos)).xy; +} + +/*------------------------------------------------------------------------------ + HISTORY SAMPLING +------------------------------------------------------------------------------*/ + +vec3 sample_catmull_rom_9(sampler2D stex, vec2 uv, vec2 resolution) { + // Source: https://gist.github.com/TheRealMJP/c83b8c0f46b63f3a88a5986f4fa982b1 + // License: https://gist.github.com/TheRealMJP/bc503b0b87b643d3505d41eab8b332ae + + // We're going to sample a 4x4 grid of texels surrounding the target UV coordinate. We'll do this by rounding + // down the sample location to get the exact center of our "starting" texel. The starting texel will be at + // location [1, 1] in the grid, where [0, 0] is the top left corner. + vec2 sample_pos = uv * resolution; + vec2 texPos1 = floor(sample_pos - 0.5f) + 0.5f; + + // Compute the fractional offset from our starting texel to our original sample location, which we'll + // feed into the Catmull-Rom spline function to get our filter weights. + vec2 f = sample_pos - texPos1; + + // Compute the Catmull-Rom weights using the fractional offset that we calculated earlier. + // These equations are pre-expanded based on our knowledge of where the texels will be located, + // which lets us avoid having to evaluate a piece-wise function. + vec2 w0 = f * (-0.5f + f * (1.0f - 0.5f * f)); + vec2 w1 = 1.0f + f * f * (-2.5f + 1.5f * f); + vec2 w2 = f * (0.5f + f * (2.0f - 1.5f * f)); + vec2 w3 = f * f * (-0.5f + 0.5f * f); + + // Work out weighting factors and sampling offsets that will let us use bilinear filtering to + // simultaneously evaluate the middle 2 samples from the 4x4 grid. + vec2 w12 = w1 + w2; + vec2 offset12 = w2 / (w1 + w2); + + // Compute the final UV coordinates we'll use for sampling the texture + vec2 texPos0 = texPos1 - 1.0f; + vec2 texPos3 = texPos1 + 2.0f; + vec2 texPos12 = texPos1 + offset12; + + texPos0 /= resolution; + texPos3 /= resolution; + texPos12 /= resolution; + + vec3 result = vec3(0.0f, 0.0f, 0.0f); + + result += textureLod(stex, vec2(texPos0.x, texPos0.y), 0.0).xyz * w0.x * w0.y; + result += textureLod(stex, vec2(texPos12.x, texPos0.y), 0.0).xyz * w12.x * w0.y; + result += textureLod(stex, vec2(texPos3.x, texPos0.y), 0.0).xyz * w3.x * w0.y; + + result += textureLod(stex, vec2(texPos0.x, texPos12.y), 0.0).xyz * w0.x * w12.y; + result += textureLod(stex, vec2(texPos12.x, texPos12.y), 0.0).xyz * w12.x * w12.y; + result += textureLod(stex, vec2(texPos3.x, texPos12.y), 0.0).xyz * w3.x * w12.y; + + result += textureLod(stex, vec2(texPos0.x, texPos3.y), 0.0).xyz * w0.x * w3.y; + result += textureLod(stex, vec2(texPos12.x, texPos3.y), 0.0).xyz * w12.x * w3.y; + result += textureLod(stex, vec2(texPos3.x, texPos3.y), 0.0).xyz * w3.x * w3.y; + + return max(result, 0.0f); +} + +/*------------------------------------------------------------------------------ + HISTORY CLIPPING +------------------------------------------------------------------------------*/ + +// Based on "Temporal Reprojection Anti-Aliasing" - https://github.com/playdeadgames/temporal +vec3 clip_aabb(vec3 aabb_min, vec3 aabb_max, vec3 p, vec3 q) { + vec3 r = q - p; + vec3 rmax = (aabb_max - p.xyz); + vec3 rmin = (aabb_min - p.xyz); + + if (r.x > rmax.x + FLT_MIN) + r *= (rmax.x / r.x); + if (r.y > rmax.y + FLT_MIN) + r *= (rmax.y / r.y); + if (r.z > rmax.z + FLT_MIN) + r *= (rmax.z / r.z); + + if (r.x < rmin.x - FLT_MIN) + r *= (rmin.x / r.x); + if (r.y < rmin.y - FLT_MIN) + r *= (rmin.y / r.y); + if (r.z < rmin.z - FLT_MIN) + r *= (rmin.z / r.z); + + return p + r; +} + +// Clip history to the neighbourhood of the current sample +vec3 clip_history_3x3(uvec2 group_pos, vec3 color_history, vec2 velocity_closest) { + // Sample a 3x3 neighbourhood + vec3 s1 = load_color(group_pos + kOffsets3x3[0]); + vec3 s2 = load_color(group_pos + kOffsets3x3[1]); + vec3 s3 = load_color(group_pos + kOffsets3x3[2]); + vec3 s4 = load_color(group_pos + kOffsets3x3[3]); + vec3 s5 = load_color(group_pos + kOffsets3x3[4]); + vec3 s6 = load_color(group_pos + kOffsets3x3[5]); + vec3 s7 = load_color(group_pos + kOffsets3x3[6]); + vec3 s8 = load_color(group_pos + kOffsets3x3[7]); + vec3 s9 = load_color(group_pos + kOffsets3x3[8]); + + // Compute min and max (with an adaptive box size, which greatly reduces ghosting) + vec3 color_avg = (s1 + s2 + s3 + s4 + s5 + s6 + s7 + s8 + s9) * RPC_9; + vec3 color_avg2 = ((s1 * s1) + (s2 * s2) + (s3 * s3) + (s4 * s4) + (s5 * s5) + (s6 * s6) + (s7 * s7) + (s8 * s8) + (s9 * s9)) * RPC_9; + float box_size = mix(0.0f, 2.5f, smoothstep(0.02f, 0.0f, length(velocity_closest))); + vec3 dev = sqrt(abs(color_avg2 - (color_avg * color_avg))) * box_size; + vec3 color_min = color_avg - dev; + vec3 color_max = color_avg + dev; + + // Variance clipping + vec3 color = clip_aabb(color_min, color_max, clamp(color_avg, color_min, color_max), color_history); + + // Clamp to prevent NaNs + color = clamp(color, FLT_MIN, FLT_MAX); + + return color; +} + +/*------------------------------------------------------------------------------ + TAA +------------------------------------------------------------------------------*/ + +const vec3 lumCoeff = vec3(0.299f, 0.587f, 0.114f); + +float luminance(vec3 color) { + return max(dot(color, lumCoeff), 0.0001f); +} + +float get_factor_disocclusion(vec2 uv_reprojected, vec2 velocity) { + vec2 velocity_previous = imageLoad(last_velocity_buffer, ivec2(uv_reprojected * params.resolution)).xy; + vec2 velocity_texels = velocity * params.resolution; + vec2 prev_velocity_texels = velocity_previous * params.resolution; + float disocclusion = length(prev_velocity_texels - velocity_texels) - params.disocclusion_threshold; + return clamp(disocclusion * params.disocclusion_scale, 0.0, 1.0); +} + +vec3 temporal_antialiasing(uvec2 pos_group_top_left, uvec2 pos_group, uvec2 pos_screen, vec2 uv, sampler2D tex_history) { + // Get the velocity of the current pixel + vec2 velocity = imageLoad(velocity_buffer, ivec2(pos_screen)).xy; + + // Get reprojected uv + vec2 uv_reprojected = uv - velocity; + + // Get input color + vec3 color_input = load_color(pos_group); + + // Get history color (catmull-rom reduces a lot of the blurring that you get under motion) + vec3 color_history = sample_catmull_rom_9(tex_history, uv_reprojected, params.resolution).rgb; + + // Clip history to the neighbourhood of the current sample (fixes a lot of the ghosting). + vec2 velocity_closest = vec2(0.0); // This is best done by using the velocity with the closest depth. + get_closest_pixel_velocity_3x3(pos_group, pos_group_top_left, velocity_closest); + color_history = clip_history_3x3(pos_group, color_history, velocity_closest); + + // Compute blend factor + float blend_factor = RPC_16; // We want to be able to accumulate as many jitter samples as we generated, that is, 16. + { + // If re-projected UV is out of screen, converge to current color immediatel + float factor_screen = any(lessThan(uv_reprojected, vec2(0.0))) || any(greaterThan(uv_reprojected, vec2(1.0))) ? 1.0 : 0.0; + + // Increase blend factor when there is disocclusion (fixes a lot of the remaining ghosting). + float factor_disocclusion = get_factor_disocclusion(uv_reprojected, velocity); + + // Add to the blend factor + blend_factor = clamp(blend_factor + factor_screen + factor_disocclusion, 0.0, 1.0); + } + + // Resolve + vec3 color_resolved = vec3(0.0); + { + // Tonemap + color_history = reinhard(color_history); + color_input = reinhard(color_input); + + // Reduce flickering + float lum_color = luminance(color_input); + float lum_history = luminance(color_history); + float diff = abs(lum_color - lum_history) / max(lum_color, max(lum_history, 1.001)); + diff = 1.0 - diff; + diff = diff * diff; + blend_factor = mix(0.0, blend_factor, diff); + + // Lerp/blend + color_resolved = mix(color_history, color_input, blend_factor); + + // Inverse tonemap + color_resolved = reinhard_inverse(color_resolved); + } + + return color_resolved; +} + +void main() { +#ifdef USE_SUBGROUPS + populate_group_shared_memory(gl_WorkGroupID.xy, gl_LocalInvocationIndex); +#endif + + // Out of bounds check + if (any(greaterThanEqual(vec2(gl_GlobalInvocationID.xy), params.resolution))) { + return; + } + +#ifdef USE_SUBGROUPS + const uvec2 pos_group = gl_LocalInvocationID.xy; + const uvec2 pos_group_top_left = gl_WorkGroupID.xy * kGroupSize - kBorderSize; +#else + const uvec2 pos_group = gl_GlobalInvocationID.xy; + const uvec2 pos_group_top_left = uvec2(0, 0); +#endif + const uvec2 pos_screen = gl_GlobalInvocationID.xy; + const vec2 uv = (gl_GlobalInvocationID.xy + 0.5f) / params.resolution; + + vec3 result = temporal_antialiasing(pos_group_top_left, pos_group, pos_screen, uv, history_buffer); + imageStore(output_buffer, ivec2(gl_GlobalInvocationID.xy), vec4(result, 1.0)); +} diff --git a/servers/rendering/renderer_rd/shaders/effects/tonemap.glsl b/servers/rendering/renderer_rd/shaders/effects/tonemap.glsl index 62a7b0e7d7..52aee8b648 100644 --- a/servers/rendering/renderer_rd/shaders/effects/tonemap.glsl +++ b/servers/rendering/renderer_rd/shaders/effects/tonemap.glsl @@ -75,7 +75,7 @@ layout(push_constant, std430) uniform Params { float exposure; float white; - float auto_exposure_grey; + float auto_exposure_scale; float luminance_multiplier; vec2 pixel_size; @@ -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); @@ -440,7 +440,7 @@ void main() { #ifndef SUBPASS if (params.use_auto_exposure) { - exposure *= 1.0 / (texelFetch(source_auto_exposure, ivec2(0, 0), 0).r * params.luminance_multiplier / params.auto_exposure_grey); + exposure *= 1.0 / (texelFetch(source_auto_exposure, ivec2(0, 0), 0).r * params.luminance_multiplier / params.auto_exposure_scale); } #endif @@ -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/effects/vrs.glsl b/servers/rendering/renderer_rd/shaders/effects/vrs.glsl new file mode 100644 index 0000000000..b450bb9fe9 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/effects/vrs.glsl @@ -0,0 +1,80 @@ +#[vertex] + +#version 450 + +#VERSION_DEFINES + +#ifdef MULTIVIEW +#ifdef has_VK_KHR_multiview +#extension GL_EXT_multiview : enable +#define ViewIndex gl_ViewIndex +#else // has_VK_KHR_multiview +#define ViewIndex 0 +#endif // has_VK_KHR_multiview +#endif //MULTIVIEW + +#ifdef MULTIVIEW +layout(location = 0) out vec3 uv_interp; +#else +layout(location = 0) out vec2 uv_interp; +#endif + +void main() { + vec2 base_arr[4] = vec2[](vec2(0.0, 0.0), vec2(0.0, 1.0), vec2(1.0, 1.0), vec2(1.0, 0.0)); + uv_interp.xy = base_arr[gl_VertexIndex]; +#ifdef MULTIVIEW + uv_interp.z = ViewIndex; +#endif + + gl_Position = vec4(uv_interp.xy * 2.0 - 1.0, 0.0, 1.0); +} + +#[fragment] + +#version 450 + +#VERSION_DEFINES + +#ifdef MULTIVIEW +#ifdef has_VK_KHR_multiview +#extension GL_EXT_multiview : enable +#define ViewIndex gl_ViewIndex +#else // has_VK_KHR_multiview +#define ViewIndex 0 +#endif // has_VK_KHR_multiview +#endif //MULTIVIEW + +#ifdef MULTIVIEW +layout(location = 0) in vec3 uv_interp; +layout(set = 0, binding = 0) uniform sampler2DArray source_color; +#else /* MULTIVIEW */ +layout(location = 0) in vec2 uv_interp; +layout(set = 0, binding = 0) uniform sampler2D source_color; +#endif /* MULTIVIEW */ + +layout(location = 0) out uint frag_color; + +void main() { +#ifdef MULTIVIEW + vec3 uv = uv_interp; +#else + vec2 uv = uv_interp; +#endif + +#ifdef MULTIVIEW + vec4 color = textureLod(source_color, uv, 0.0); + frag_color = uint(color.r * 255.0); +#else /* MULTIVIEW */ + vec4 color = textureLod(source_color, uv, 0.0); + + // for user supplied VRS map we do a color mapping + color.r *= 3.0; + frag_color = int(color.r) << 2; + + color.g *= 3.0; + frag_color += int(color.g); + + // note 1x4, 4x1, 1x8, 8x1, 2x8 and 8x2 are not supported + // 4x8, 8x4 and 8x8 are only available on some GPUs +#endif /* MULTIVIEW */ +} |