initial commit, 4.5 stable

servers/rendering/renderer_rd/shaders/effects/SCsub

@@ -0,0 +1,20 @@
+#!/usr/bin/env python
+from misc.utility.scons_hints import *
+
+Import("env")
+
+if "RD_GLSL" in env["BUILDERS"]:
+    # find all include files
+    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 + ["#glsl_builders.py"])
+
+    # compile shaders
+    for glsl_file in glsl_files:
+        env.RD_GLSL(glsl_file)
+
+SConscript("fsr2/SCsub")

servers/rendering/renderer_rd/shaders/effects/blur_raster.glsl

@@ -0,0 +1,161 @@
+/* clang-format off */
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "blur_raster_inc.glsl"
+
+layout(location = 0) out vec2 uv_interp;
+/* clang-format on */
+
+void main() {
+	// old code, ARM driver bug on Mali-GXXx GPUs and Vulkan API 1.3.xxx
+	// https://github.com/godotengine/godot/pull/92817#issuecomment-2168625982
+	//vec2 base_arr[3] = vec2[](vec2(-1.0, -1.0), vec2(-1.0, 3.0), vec2(3.0, -1.0));
+	//gl_Position = vec4(base_arr[gl_VertexIndex], 0.0, 1.0);
+	//uv_interp = clamp(gl_Position.xy, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+
+	vec2 vertex_base;
+	if (gl_VertexIndex == 0) {
+		vertex_base = vec2(-1.0, -1.0);
+	} else if (gl_VertexIndex == 1) {
+		vertex_base = vec2(-1.0, 3.0);
+	} else {
+		vertex_base = vec2(3.0, -1.0);
+	}
+	gl_Position = vec4(vertex_base, 0.0, 1.0);
+	uv_interp = clamp(vertex_base, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+}
+
+/* clang-format off */
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "blur_raster_inc.glsl"
+
+layout(location = 0) in vec2 uv_interp;
+/* clang-format on */
+
+layout(set = 0, binding = 0) uniform sampler2D source_color;
+
+#ifdef GLOW_USE_AUTO_EXPOSURE
+layout(set = 1, binding = 0) uniform sampler2D source_auto_exposure;
+#endif
+
+layout(location = 0) out vec4 frag_color;
+
+void main() {
+	// We do not apply our color scale for our mobile renderer here, we'll leave our colors at half brightness and apply scale in the tonemap raster.
+
+#ifdef MODE_MIPMAP
+
+	vec2 pix_size = blur.pixel_size;
+	vec4 color = texture(source_color, uv_interp + vec2(-0.5, -0.5) * pix_size);
+	color += texture(source_color, uv_interp + vec2(0.5, -0.5) * pix_size);
+	color += texture(source_color, uv_interp + vec2(0.5, 0.5) * pix_size);
+	color += texture(source_color, uv_interp + vec2(-0.5, 0.5) * pix_size);
+	frag_color = color / 4.0;
+
+#endif
+
+#ifdef MODE_GAUSSIAN_BLUR
+
+	// 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
+
+	//Glow uses larger sigma 1 for a more rounded blur effect
+
+#define GLOW_ADD(m_ofs, m_mult)                                                  \
+	{                                                                            \
+		vec2 ofs = uv_interp + m_ofs * pix_size;                                 \
+		vec4 c = texture(source_color, ofs) * m_mult;                            \
+		if (any(lessThan(ofs, vec2(0.0))) || any(greaterThan(ofs, vec2(1.0)))) { \
+			c *= 0.0;                                                            \
+		}                                                                        \
+		color += c;                                                              \
+	}
+
+	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) * 0.174938;
+		GLOW_ADD(vec2(1.0, 0.0), 0.165569);
+		GLOW_ADD(vec2(2.0, 0.0), 0.140367);
+		GLOW_ADD(vec2(3.0, 0.0), 0.106595);
+		GLOW_ADD(vec2(-1.0, 0.0), 0.165569);
+		GLOW_ADD(vec2(-2.0, 0.0), 0.140367);
+		GLOW_ADD(vec2(-3.0, 0.0), 0.106595);
+
+		// only do this in the horizontal pass, if we also do this in the vertical pass we're doubling up.
+		color *= blur.glow_strength;
+
+		frag_color = color;
+	} else {
+		vec2 pix_size = blur.pixel_size;
+		vec4 color = texture(source_color, uv_interp + vec2(0.0, 0.0) * pix_size) * 0.288713;
+		GLOW_ADD(vec2(0.0, 1.0), 0.233062);
+		GLOW_ADD(vec2(0.0, 2.0), 0.122581);
+		GLOW_ADD(vec2(0.0, -1.0), 0.233062);
+		GLOW_ADD(vec2(0.0, -2.0), 0.122581);
+
+		frag_color = color;
+	}
+
+#undef GLOW_ADD
+
+	if (bool(blur.flags & FLAG_GLOW_FIRST_PASS)) {
+		// In the first pass bring back to correct color range else we're applying the wrong threshold
+		// in subsequent passes we can use it as is as we'd just be undoing it right after.
+		frag_color *= blur.luminance_multiplier;
+
+#ifdef GLOW_USE_AUTO_EXPOSURE
+
+		frag_color /= texelFetch(source_auto_exposure, ivec2(0, 0), 0).r / blur.glow_auto_exposure_scale;
+#endif
+		frag_color *= blur.glow_exposure;
+
+		float luminance = max(frag_color.r, max(frag_color.g, frag_color.b));
+		float feedback = max(smoothstep(blur.glow_hdr_threshold, blur.glow_hdr_threshold + blur.glow_hdr_scale, luminance), blur.glow_bloom);
+
+		frag_color = min(frag_color * feedback, vec4(blur.glow_luminance_cap)) / blur.luminance_multiplier;
+	}
+
+#endif // MODE_GAUSSIAN_GLOW
+
+#ifdef MODE_COPY
+	vec4 color = textureLod(source_color, uv_interp, 0.0);
+	frag_color = color;
+#endif
+}

servers/rendering/renderer_rd/shaders/effects/blur_raster_inc.glsl

@@ -0,0 +1,26 @@
+#define FLAG_HORIZONTAL (1 << 0)
+#define FLAG_USE_ORTHOGONAL_PROJECTION (1 << 1)
+#define FLAG_GLOW_FIRST_PASS (1 << 2)
+
+layout(push_constant, std430) uniform Blur {
+	vec2 pixel_size; // 08 - 08
+	uint flags; // 04 - 12
+	uint pad; // 04 - 16
+
+	// Glow.
+	float glow_strength; // 04 - 20
+	float glow_bloom; // 04 - 24
+	float glow_hdr_threshold; // 04 - 28
+	float glow_hdr_scale; // 04 - 32
+
+	float glow_exposure; // 04 - 36
+	float glow_white; // 04 - 40
+	float glow_luminance_cap; // 04 - 44
+	float glow_auto_exposure_scale; // 04 - 48
+
+	float luminance_multiplier; // 04 - 52
+	float res1; // 04 - 56
+	float res2; // 04 - 60
+	float res3; // 04 - 64
+}
+blur;

servers/rendering/renderer_rd/shaders/effects/bokeh_dof.glsl

@@ -0,0 +1,230 @@
+#[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;
+
+#ifdef MODE_GEN_BLUR_SIZE
+layout(rgba16f, set = 0, binding = 0) uniform restrict image2D color_image;
+layout(set = 1, binding = 0) uniform sampler2D source_depth;
+#endif
+
+#if defined(MODE_BOKEH_BOX) || defined(MODE_BOKEH_HEXAGONAL) || defined(MODE_BOKEH_CIRCULAR)
+layout(set = 1, binding = 0) uniform sampler2D color_texture;
+layout(rgba16f, set = 0, binding = 0) uniform restrict writeonly image2D bokeh_image;
+#endif
+
+#ifdef MODE_COMPOSITE_BOKEH
+layout(rgba16f, set = 0, binding = 0) uniform restrict image2D color_image;
+layout(set = 1, binding = 0) uniform sampler2D source_bokeh;
+#endif
+
+// based on https://www.shadertoy.com/view/Xd3GDl
+
+#include "bokeh_dof_inc.glsl"
+
+#ifdef MODE_GEN_BLUR_SIZE
+
+float get_depth_at_pos(vec2 uv) {
+	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)) / 2.0;
+	} else {
+		depth = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near + depth * (params.z_far - params.z_near));
+	}
+	return depth;
+}
+
+float get_blur_size(float depth) {
+	if (params.blur_near_active && depth < params.blur_near_begin) {
+		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) {
+		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;
+}
+
+#endif
+
+#if defined(MODE_BOKEH_BOX) || defined(MODE_BOKEH_HEXAGONAL)
+
+vec4 weighted_filter_dir(vec2 dir, vec2 uv, vec2 pixel_size) {
+	dir *= pixel_size;
+	vec4 color = texture(color_texture, uv);
+
+	vec4 accum = color;
+	float total = 1.0;
+
+	float blur_scale = params.blur_size / float(params.blur_steps);
+
+	if (params.use_jitter) {
+		uv += dir * (hash12n(uv + params.jitter_seed) - 0.5);
+	}
+
+	for (int i = -params.blur_steps; i <= params.blur_steps; i++) {
+		if (i == 0) {
+			continue;
+		}
+		float radius = float(i) * blur_scale;
+		vec2 suv = uv + dir * radius;
+		radius = abs(radius);
+
+		vec4 sample_color = texture(color_texture, suv);
+		float limit;
+
+		if (sample_color.a < color.a) {
+			limit = abs(sample_color.a);
+		} else {
+			limit = abs(color.a);
+		}
+
+		limit -= DEPTH_GAP;
+
+		float m = smoothstep(radius - 0.5, radius + 0.5, limit);
+
+		accum += mix(color, sample_color, m);
+
+		total += 1.0;
+	}
+
+	return accum / total;
+}
+
+#endif
+
+void main() {
+	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
+
+	if (any(greaterThan(pos, params.size))) { //too large, do nothing
+		return;
+	}
+
+	vec2 pixel_size = 1.0 / vec2(params.size);
+	vec2 uv = vec2(pos) / vec2(params.size);
+
+#ifdef MODE_GEN_BLUR_SIZE
+	uv += pixel_size * 0.5;
+	//precompute size in alpha channel
+	float depth = get_depth_at_pos(uv);
+	float size = get_blur_size(depth);
+
+	vec4 color = imageLoad(color_image, pos);
+	color.a = size;
+	imageStore(color_image, pos, color);
+#endif
+
+#ifdef MODE_BOKEH_BOX
+
+	//pixel_size*=0.5; //resolution is doubled
+	if (params.second_pass || !params.half_size) {
+		uv += pixel_size * 0.5; //half pixel to read centers
+	} else {
+		uv += pixel_size * 0.25; //half pixel to read centers from full res
+	}
+
+	vec2 dir = (params.second_pass ? vec2(0.0, 1.0) : vec2(1.0, 0.0));
+
+	vec4 color = weighted_filter_dir(dir, uv, pixel_size);
+
+	imageStore(bokeh_image, pos, color);
+
+#endif
+
+#ifdef MODE_BOKEH_HEXAGONAL
+
+	//pixel_size*=0.5; //resolution is doubled
+	if (params.second_pass || !params.half_size) {
+		uv += pixel_size * 0.5; //half pixel to read centers
+	} else {
+		uv += pixel_size * 0.25; //half pixel to read centers from full res
+	}
+
+	vec2 dir = (params.second_pass ? normalize(vec2(1.0, 0.577350269189626)) : vec2(0.0, 1.0));
+
+	vec4 color = weighted_filter_dir(dir, uv, pixel_size);
+
+	if (params.second_pass) {
+		dir = normalize(vec2(-1.0, 0.577350269189626));
+
+		vec4 color2 = weighted_filter_dir(dir, uv, pixel_size);
+
+		color.rgb = min(color.rgb, color2.rgb);
+		color.a = (color.a + color2.a) * 0.5;
+	}
+
+	imageStore(bokeh_image, pos, color);
+
+#endif
+
+#ifdef MODE_BOKEH_CIRCULAR
+
+	if (params.half_size) {
+		pixel_size *= 0.5; //resolution is doubled
+	}
+
+	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;
+
+	for (float ang = 0.0; radius < params.blur_size; ang += GOLDEN_ANGLE) {
+		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 > 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);
+		color += mix(color / accum, sample_color, m);
+		accum += 1.0;
+		radius += params.blur_scale / radius;
+	}
+
+	color /= accum;
+
+	imageStore(bokeh_image, pos, color);
+#endif
+
+#ifdef MODE_COMPOSITE_BOKEH
+
+	uv += pixel_size * 0.5;
+	vec4 color = imageLoad(color_image, pos);
+	vec4 bokeh = texture(source_bokeh, uv);
+
+	float mix_amount;
+	if (bokeh.a < color.a) {
+		mix_amount = min(1.0, max(0.0, max(abs(color.a), abs(bokeh.a)) - DEPTH_GAP));
+	} else {
+		mix_amount = min(1.0, max(0.0, abs(color.a) - DEPTH_GAP));
+	}
+
+	color.rgb = mix(color.rgb, bokeh.rgb, mix_amount); //blend between hires and lowres
+
+	color.a = 0; //reset alpha
+	imageStore(color_image, pos, color);
+#endif
+}

servers/rendering/renderer_rd/shaders/effects/bokeh_dof_inc.glsl

@@ -0,0 +1,42 @@
+layout(push_constant, std430) uniform Params {
+	ivec2 size;
+	float z_far;
+	float z_near;
+
+	bool orthogonal;
+	float blur_size;
+	float blur_scale;
+	int blur_steps;
+
+	bool blur_near_active;
+	float blur_near_begin;
+	float blur_near_end;
+	bool blur_far_active;
+
+	float blur_far_begin;
+	float blur_far_end;
+	bool second_pass;
+	bool half_size;
+
+	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;
+
+//used to work around downsampling filter
+#define DEPTH_GAP 0.0
+
+const float GOLDEN_ANGLE = 2.39996323;
+
+//note: uniform pdf rand [0;1[
+float hash12n(vec2 p) {
+	p = fract(p * vec2(5.3987, 5.4421));
+	p += dot(p.yx, p.xy + vec2(21.5351, 14.3137));
+	return fract(p.x * p.y * 95.4307);
+}

servers/rendering/renderer_rd/shaders/effects/bokeh_dof_raster.glsl

@@ -0,0 +1,276 @@
+/* clang-format off */
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "bokeh_dof_inc.glsl"
+
+layout(location = 0) out vec2 uv_interp;
+/* clang-format on */
+
+void main() {
+	// old code, ARM driver bug on Mali-GXXx GPUs and Vulkan API 1.3.xxx
+	// https://github.com/godotengine/godot/pull/92817#issuecomment-2168625982
+	//vec2 base_arr[3] = vec2[](vec2(-1.0, -1.0), vec2(-1.0, 3.0), vec2(3.0, -1.0));
+	//gl_Position = vec4(base_arr[gl_VertexIndex], 0.0, 1.0);
+	//uv_interp = clamp(gl_Position.xy, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+
+	vec2 vertex_base;
+	if (gl_VertexIndex == 0) {
+		vertex_base = vec2(-1.0, -1.0);
+	} else if (gl_VertexIndex == 1) {
+		vertex_base = vec2(-1.0, 3.0);
+	} else {
+		vertex_base = vec2(3.0, -1.0);
+	}
+	gl_Position = vec4(vertex_base, 0.0, 1.0);
+	uv_interp = clamp(vertex_base, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+}
+
+/* clang-format off */
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "bokeh_dof_inc.glsl"
+
+layout(location = 0) in vec2 uv_interp;
+/* clang-format on */
+
+#ifdef MODE_GEN_BLUR_SIZE
+layout(location = 0) out float weight;
+
+layout(set = 0, binding = 0) uniform sampler2D source_depth;
+#else
+layout(location = 0) out vec4 frag_color;
+#ifdef OUTPUT_WEIGHT
+layout(location = 1) out float weight;
+#endif
+
+layout(set = 0, binding = 0) uniform sampler2D source_color;
+layout(set = 1, binding = 0) uniform sampler2D source_weight;
+#ifdef MODE_COMPOSITE_BOKEH
+layout(set = 2, binding = 0) uniform sampler2D original_weight;
+#endif
+#endif
+
+//DOF
+// Bokeh single pass implementation based on https://tuxedolabs.blogspot.com/2018/05/bokeh-depth-of-field-in-single-pass.html
+
+#ifdef MODE_GEN_BLUR_SIZE
+
+float get_depth_at_pos(vec2 uv) {
+	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)) / 2.0;
+	} else {
+		depth = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near + depth * (params.z_far - params.z_near));
+	}
+	return depth;
+}
+
+float get_blur_size(float depth) {
+	if (params.blur_near_active && depth < params.blur_near_begin) {
+		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) {
+		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;
+}
+
+#endif
+
+#if defined(MODE_BOKEH_BOX) || defined(MODE_BOKEH_HEXAGONAL)
+
+vec4 weighted_filter_dir(vec2 dir, vec2 uv, vec2 pixel_size) {
+	dir *= pixel_size;
+	vec4 color = texture(source_color, uv);
+	color.a = texture(source_weight, uv).r;
+
+	vec4 accum = color;
+	float total = 1.0;
+
+	float blur_scale = params.blur_size / float(params.blur_steps);
+
+	if (params.use_jitter) {
+		uv += dir * (hash12n(uv + params.jitter_seed) - 0.5);
+	}
+
+	for (int i = -params.blur_steps; i <= params.blur_steps; i++) {
+		if (i == 0) {
+			continue;
+		}
+		float radius = float(i) * blur_scale;
+		vec2 suv = uv + dir * radius;
+		radius = abs(radius);
+
+		vec4 sample_color = texture(source_color, suv);
+		sample_color.a = texture(source_weight, suv).r;
+		float limit;
+
+		if (sample_color.a < color.a) {
+			limit = abs(sample_color.a);
+		} else {
+			limit = abs(color.a);
+		}
+
+		limit -= DEPTH_GAP;
+
+		float m = smoothstep(radius - 0.5, radius + 0.5, limit);
+
+		accum += mix(color, sample_color, m);
+
+		total += 1.0;
+	}
+
+	return accum / total;
+}
+
+#endif
+
+void main() {
+	vec2 pixel_size = 1.0 / vec2(params.size);
+	vec2 uv = uv_interp;
+
+#ifdef MODE_GEN_BLUR_SIZE
+	uv += pixel_size * 0.5;
+	float center_depth = get_depth_at_pos(uv);
+	weight = get_blur_size(center_depth);
+#endif
+
+#ifdef MODE_BOKEH_BOX
+	//pixel_size*=0.5; //resolution is doubled
+	if (params.second_pass || !params.half_size) {
+		uv += pixel_size * 0.5; //half pixel to read centers
+	} else {
+		uv += pixel_size * 0.25; //half pixel to read centers from full res
+	}
+
+	float alpha = texture(source_color, uv).a; // retain this
+	vec2 dir = (params.second_pass ? vec2(0.0, 1.0) : vec2(1.0, 0.0));
+
+	vec4 color = weighted_filter_dir(dir, uv, pixel_size);
+
+	frag_color = color;
+	frag_color.a = alpha; // attempt to retain this in case we have a transparent background, ignored if half_size
+#ifdef OUTPUT_WEIGHT
+	weight = color.a;
+#endif
+
+#endif
+
+#ifdef MODE_BOKEH_HEXAGONAL
+
+	//pixel_size*=0.5; //resolution is doubled
+	if (params.second_pass || !params.half_size) {
+		uv += pixel_size * 0.5; //half pixel to read centers
+	} else {
+		uv += pixel_size * 0.25; //half pixel to read centers from full res
+	}
+
+	float alpha = texture(source_color, uv).a; // retain this
+
+	vec2 dir = (params.second_pass ? normalize(vec2(1.0, 0.577350269189626)) : vec2(0.0, 1.0));
+
+	vec4 color = weighted_filter_dir(dir, uv, pixel_size);
+
+	if (params.second_pass) {
+		dir = normalize(vec2(-1.0, 0.577350269189626));
+
+		vec4 color2 = weighted_filter_dir(dir, uv, pixel_size);
+
+		color.rgb = min(color.rgb, color2.rgb);
+		color.a = (color.a + color2.a) * 0.5;
+	}
+
+	frag_color = color;
+	frag_color.a = alpha; // attempt to retain this in case we have a transparent background, ignored if half_size
+#ifdef OUTPUT_WEIGHT
+	weight = color.a;
+#endif
+
+#endif
+
+#ifdef MODE_BOKEH_CIRCULAR
+	if (params.half_size) {
+		pixel_size *= 0.5; //resolution is doubled
+	}
+
+	uv += pixel_size * 0.5; //half pixel to read centers
+
+	vec4 color = texture(source_color, uv);
+	float alpha = color.a; // retain this
+	color.a = texture(source_weight, uv).r;
+
+	vec4 color_accum = color;
+	float accum = 1.0;
+
+	float radius = params.blur_scale;
+	for (float ang = 0.0; radius < params.blur_size; ang += GOLDEN_ANGLE) {
+		vec2 uv_adj = uv + vec2(cos(ang), sin(ang)) * pixel_size * radius;
+
+		vec4 sample_color = texture(source_color, uv_adj);
+		sample_color.a = texture(source_weight, uv_adj).r;
+
+		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;
+
+		float m = smoothstep(radius - 0.5, radius + 0.5, limit);
+		color_accum += mix(color_accum / accum, sample_color, m);
+		accum += 1.0;
+
+		radius += params.blur_scale / radius;
+	}
+
+	color_accum = color_accum / accum;
+
+	frag_color.rgb = color_accum.rgb;
+	frag_color.a = alpha; // attempt to retain this in case we have a transparent background, ignored if half_size
+#ifdef OUTPUT_WEIGHT
+	weight = color_accum.a;
+#endif
+
+#endif
+
+#ifdef MODE_COMPOSITE_BOKEH
+	frag_color.rgb = texture(source_color, uv).rgb;
+
+	float center_weight = texture(source_weight, uv).r;
+	float sample_weight = texture(original_weight, uv).r;
+
+	float mix_amount;
+	if (sample_weight < center_weight) {
+		mix_amount = min(1.0, max(0.0, max(abs(center_weight), abs(sample_weight)) - DEPTH_GAP));
+	} else {
+		mix_amount = min(1.0, max(0.0, abs(center_weight) - DEPTH_GAP));
+	}
+
+	// let alpha blending take care of mixing
+	frag_color.a = mix_amount;
+#endif
+}

servers/rendering/renderer_rd/shaders/effects/copy.glsl

@@ -0,0 +1,285 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+#define FLAG_HORIZONTAL (1 << 0)
+#define FLAG_USE_BLUR_SECTION (1 << 1)
+#define FLAG_USE_ORTHOGONAL_PROJECTION (1 << 2)
+#define FLAG_DOF_NEAR_FIRST_TAP (1 << 3)
+#define FLAG_GLOW_FIRST_PASS (1 << 4)
+#define FLAG_FLIP_Y (1 << 5)
+#define FLAG_FORCE_LUMINANCE (1 << 6)
+#define FLAG_COPY_ALL_SOURCE (1 << 7)
+#define FLAG_ALPHA_TO_ONE (1 << 8)
+
+layout(push_constant, std430) uniform Params {
+	ivec4 section;
+	ivec2 target;
+	uint flags;
+	uint pad;
+	// Glow.
+	float glow_strength;
+	float glow_bloom;
+	float glow_hdr_threshold;
+	float glow_hdr_scale;
+
+	float glow_exposure;
+	float glow_white;
+	float glow_luminance_cap;
+	float glow_auto_exposure_scale;
+	// DOF.
+	float camera_z_far;
+	float camera_z_near;
+	uint pad2[2];
+
+	vec4 set_color;
+}
+params;
+
+#ifdef MODE_CUBEMAP_ARRAY_TO_PANORAMA
+layout(set = 0, binding = 0) uniform samplerCubeArray source_color;
+#elif defined(MODE_CUBEMAP_TO_PANORAMA)
+layout(set = 0, binding = 0) uniform samplerCube source_color;
+#elif !defined(MODE_SET_COLOR)
+layout(set = 0, binding = 0) uniform sampler2D source_color;
+#endif
+
+#ifdef GLOW_USE_AUTO_EXPOSURE
+layout(set = 1, binding = 0) uniform sampler2D source_auto_exposure;
+#endif
+
+#if defined(MODE_LINEARIZE_DEPTH_COPY) || defined(MODE_SIMPLE_COPY_DEPTH)
+layout(r32f, set = 3, binding = 0) uniform restrict writeonly image2D dest_buffer;
+#elif defined(DST_IMAGE_8BIT)
+layout(rgba8, set = 3, binding = 0) uniform restrict writeonly image2D dest_buffer;
+#else
+layout(rgba16f, set = 3, binding = 0) uniform restrict writeonly image2D dest_buffer;
+#endif
+
+#ifdef MODE_GAUSSIAN_BLUR
+shared vec4 local_cache[256];
+shared vec4 temp_cache[128];
+#endif
+
+void main() {
+	// Pixel being shaded
+	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
+
+#ifndef MODE_GAUSSIAN_BLUR // Gaussian blur needs the extra threads
+	if (any(greaterThanEqual(pos, params.section.zw))) { //too large, do nothing
+		return;
+	}
+#endif
+
+#ifdef MODE_MIPMAP
+
+	ivec2 base_pos = (pos + params.section.xy) << 1;
+	vec4 color = texelFetch(source_color, base_pos, 0);
+	color += texelFetch(source_color, base_pos + ivec2(0, 1), 0);
+	color += texelFetch(source_color, base_pos + ivec2(1, 0), 0);
+	color += texelFetch(source_color, base_pos + ivec2(1, 1), 0);
+	color /= 4.0;
+	color = mix(color, vec4(100.0, 100.0, 100.0, 1.0), isinf(color));
+	color = mix(color, vec4(100.0, 100.0, 100.0, 1.0), isnan(color));
+
+	imageStore(dest_buffer, pos + params.target, color);
+#endif
+
+#ifdef MODE_GAUSSIAN_BLUR
+
+	// First pass copy texture into 16x16 local memory for every 8x8 thread block
+	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;
+
+	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
+		vec3 tonemap_col = vec3(0.299, 0.587, 0.114) / max(params.glow_luminance_cap, 6.0);
+		local_cache[dest_index] /= 1.0 + dot(local_cache[dest_index].rgb, tonemap_col);
+		local_cache[dest_index + 1] /= 1.0 + dot(local_cache[dest_index + 1].rgb, tonemap_col);
+		local_cache[dest_index + 16] /= 1.0 + dot(local_cache[dest_index + 16].rgb, tonemap_col);
+		local_cache[dest_index + 16 + 1] /= 1.0 + dot(local_cache[dest_index + 16 + 1].rgb, tonemap_col);
+	}
+	const float kernel[5] = { 0.2024, 0.1790, 0.1240, 0.0672, 0.0285 };
+#else
+	// Simpler blur uses SIGMA2 for the gaussian kernel for a stronger effect.
+	const float kernel[4] = { 0.214607, 0.189879, 0.131514, 0.071303 };
+#endif
+	memoryBarrierShared();
+	barrier();
+
+	// Horizontal pass. Needs to copy into 8x16 chunk of local memory so vertical pass has full resolution
+	uint read_index = gl_LocalInvocationID.x + gl_LocalInvocationID.y * 32 + 4;
+	vec4 color_top = vec4(0.0);
+	color_top += local_cache[read_index] * kernel[0];
+	color_top += local_cache[read_index + 1] * kernel[1];
+	color_top += local_cache[read_index + 2] * kernel[2];
+	color_top += local_cache[read_index + 3] * kernel[3];
+	color_top += local_cache[read_index - 1] * kernel[1];
+	color_top += local_cache[read_index - 2] * kernel[2];
+	color_top += local_cache[read_index - 3] * kernel[3];
+#ifdef MODE_GLOW
+	color_top += local_cache[read_index + 4] * kernel[4];
+	color_top += local_cache[read_index - 4] * kernel[4];
+#endif // MODE_GLOW
+
+	vec4 color_bottom = vec4(0.0);
+	color_bottom += local_cache[read_index + 16] * kernel[0];
+	color_bottom += local_cache[read_index + 1 + 16] * kernel[1];
+	color_bottom += local_cache[read_index + 2 + 16] * kernel[2];
+	color_bottom += local_cache[read_index + 3 + 16] * kernel[3];
+	color_bottom += local_cache[read_index - 1 + 16] * kernel[1];
+	color_bottom += local_cache[read_index - 2 + 16] * kernel[2];
+	color_bottom += local_cache[read_index - 3 + 16] * kernel[3];
+#ifdef MODE_GLOW
+	color_bottom += local_cache[read_index + 4 + 16] * kernel[4];
+	color_bottom += local_cache[read_index - 4 + 16] * kernel[4];
+#endif // MODE_GLOW
+
+	// rotate samples to take advantage of cache coherency
+	uint write_index = gl_LocalInvocationID.y * 2 + gl_LocalInvocationID.x * 16;
+
+	temp_cache[write_index] = color_top;
+	temp_cache[write_index + 1] = color_bottom;
+
+	memoryBarrierShared();
+	barrier();
+
+	// If destination outside of texture, can stop doing work now
+	if (any(greaterThanEqual(pos, params.section.zw))) {
+		return;
+	}
+
+	// Vertical pass
+	uint index = gl_LocalInvocationID.y + gl_LocalInvocationID.x * 16 + 4;
+	vec4 color = vec4(0.0);
+
+	color += temp_cache[index] * kernel[0];
+	color += temp_cache[index + 1] * kernel[1];
+	color += temp_cache[index + 2] * kernel[2];
+	color += temp_cache[index + 3] * kernel[3];
+	color += temp_cache[index - 1] * kernel[1];
+	color += temp_cache[index - 2] * kernel[2];
+	color += temp_cache[index - 3] * kernel[3];
+#ifdef MODE_GLOW
+	color += temp_cache[index + 4] * kernel[4];
+	color += temp_cache[index - 4] * kernel[4];
+#endif // MODE_GLOW
+
+#ifdef MODE_GLOW
+	if (bool(params.flags & FLAG_GLOW_FIRST_PASS)) {
+		// Undo tonemap to restore range: https://graphicrants.blogspot.com/2013/12/tone-mapping.html
+		color /= 1.0 - dot(color.rgb, vec3(0.299, 0.587, 0.114) / max(params.glow_luminance_cap, 6.0));
+	}
+
+	color *= params.glow_strength;
+
+	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_scale;
+#endif
+		color *= params.glow_exposure;
+
+		float luminance = max(color.r, max(color.g, color.b));
+		float feedback = max(smoothstep(params.glow_hdr_threshold, params.glow_hdr_threshold + params.glow_hdr_scale, luminance), params.glow_bloom);
+
+		color = min(color * feedback, vec4(params.glow_luminance_cap));
+	}
+#endif // MODE_GLOW
+	imageStore(dest_buffer, pos + params.target, color);
+
+#endif // MODE_GAUSSIAN_BLUR
+
+#ifdef MODE_SIMPLE_COPY
+
+	vec4 color;
+	if (bool(params.flags & FLAG_COPY_ALL_SOURCE)) {
+		vec2 uv = vec2(pos) / vec2(params.section.zw);
+		if (bool(params.flags & FLAG_FLIP_Y)) {
+			uv.y = 1.0 - uv.y;
+		}
+		color = textureLod(source_color, uv, 0.0);
+
+	} else {
+		color = texelFetch(source_color, pos + params.section.xy, 0);
+
+		if (bool(params.flags & FLAG_FLIP_Y)) {
+			pos.y = params.section.w - pos.y - 1;
+		}
+	}
+
+	if (bool(params.flags & FLAG_FORCE_LUMINANCE)) {
+		color.rgb = vec3(max(max(color.r, color.g), color.b));
+	}
+
+	if (bool(params.flags & FLAG_ALPHA_TO_ONE)) {
+		color.a = 1.0;
+	}
+
+	imageStore(dest_buffer, pos + params.target, color);
+
+#endif // MODE_SIMPLE_COPY
+
+#ifdef MODE_SIMPLE_COPY_DEPTH
+
+	vec4 color = texelFetch(source_color, pos + params.section.xy, 0);
+
+	if (bool(params.flags & FLAG_FLIP_Y)) {
+		pos.y = params.section.w - pos.y - 1;
+	}
+
+	imageStore(dest_buffer, pos + params.target, vec4(color.r));
+
+#endif // MODE_SIMPLE_COPY_DEPTH
+
+#ifdef MODE_LINEARIZE_DEPTH_COPY
+
+	float depth = texelFetch(source_color, pos + params.section.xy, 0).r;
+	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));
+	vec4 color = vec4(depth / params.camera_z_far);
+
+	if (bool(params.flags & FLAG_FLIP_Y)) {
+		pos.y = params.section.w - pos.y - 1;
+	}
+
+	imageStore(dest_buffer, pos + params.target, color);
+#endif // MODE_LINEARIZE_DEPTH_COPY
+
+#if defined(MODE_CUBEMAP_TO_PANORAMA) || defined(MODE_CUBEMAP_ARRAY_TO_PANORAMA)
+
+	const float PI = 3.14159265359;
+	vec2 uv = vec2(pos) / vec2(params.section.zw);
+	if (bool(params.flags & FLAG_FLIP_Y)) {
+		uv.y = 1.0 - uv.y;
+	}
+	float phi = uv.x * 2.0 * PI;
+	float theta = uv.y * PI;
+
+	vec3 normal;
+	normal.x = sin(phi) * sin(theta) * -1.0;
+	normal.y = cos(theta);
+	normal.z = cos(phi) * sin(theta) * -1.0;
+
+#ifdef MODE_CUBEMAP_TO_PANORAMA
+	vec4 color = textureLod(source_color, normal, params.camera_z_far); //the biggest the lod the least the acne
+#else
+	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 // defined(MODE_CUBEMAP_TO_PANORAMA) || defined(MODE_CUBEMAP_ARRAY_TO_PANORAMA)
+
+#ifdef MODE_SET_COLOR
+	imageStore(dest_buffer, pos + params.target, params.set_color);
+#endif
+}

servers/rendering/renderer_rd/shaders/effects/copy_to_fb.glsl

@@ -0,0 +1,195 @@
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef USE_MULTIVIEW
+#extension GL_EXT_multiview : enable
+#define ViewIndex gl_ViewIndex
+#endif // USE_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)
+#define FLAG_NORMAL (1 << 7)
+#define FLAG_USE_SRC_SECTION (1 << 8)
+
+#ifdef USE_MULTIVIEW
+layout(location = 0) out vec3 uv_interp;
+#else
+layout(location = 0) out vec2 uv_interp;
+#endif
+
+layout(push_constant, std430) uniform Params {
+	vec4 section;
+	vec2 pixel_size;
+	float luminance_multiplier;
+	uint flags;
+
+	vec4 color;
+}
+params;
+
+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 USE_MULTIVIEW
+	uv_interp.z = ViewIndex;
+#endif
+	vec2 vpos = uv_interp.xy;
+	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 (bool(params.flags & FLAG_FLIP_Y)) {
+		uv_interp.y = 1.0 - uv_interp.y;
+	}
+
+	if (bool(params.flags & FLAG_USE_SRC_SECTION)) {
+		uv_interp.xy = params.section.xy + uv_interp.xy * params.section.zw;
+	}
+}
+
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#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)
+#define FLAG_NORMAL (1 << 7)
+
+layout(push_constant, std430) uniform Params {
+	vec4 section;
+	vec2 pixel_size;
+	float luminance_multiplier;
+	uint flags;
+
+	vec4 color;
+}
+params;
+
+#ifndef MODE_SET_COLOR
+#ifdef USE_MULTIVIEW
+layout(location = 0) in vec3 uv_interp;
+#else
+layout(location = 0) in vec2 uv_interp;
+#endif
+
+#ifdef USE_MULTIVIEW
+layout(set = 0, binding = 0) uniform sampler2DArray source_color;
+#ifdef MODE_TWO_SOURCES
+layout(set = 1, binding = 0) uniform sampler2DArray source_depth;
+layout(location = 1) out float depth;
+#endif /* MODE_TWO_SOURCES */
+#else /* USE_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 /* USE_MULTIVIEW */
+#endif /* !SET_COLOR */
+
+layout(location = 0) out vec4 frag_color;
+
+vec3 linear_to_srgb(vec3 color) {
+	//if going to srgb, clamp from 0 to 1.
+	color = clamp(color, vec3(0.0), vec3(1.0));
+	const vec3 a = vec3(0.055f);
+	return mix((vec3(1.0f) + a) * pow(color.rgb, vec3(1.0f / 2.4f)) - a, 12.92f * color.rgb, lessThan(color.rgb, vec3(0.0031308f)));
+}
+
+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 USE_MULTIVIEW
+	vec3 uv = uv_interp;
+#else
+	vec2 uv = uv_interp;
+#endif
+
+#ifdef MODE_PANORAMA_TO_DP
+	// Note, multiview and panorama should not be mixed at this time
+
+	//obtain normal from dual paraboloid uv
+#define M_PI 3.14159265359
+
+	float side;
+	uv.y = modf(uv.y * 2.0, side);
+	side = side * 2.0 - 1.0;
+	vec3 normal = vec3(uv * 2.0 - 1.0, 0.0);
+	normal.z = 0.5 - 0.5 * ((normal.x * normal.x) + (normal.y * normal.y));
+	normal *= -side;
+	normal = normalize(normal);
+
+	//now convert normal to panorama uv
+
+	vec2 st = vec2(atan(normal.x, normal.z), acos(normal.y));
+
+	if (st.x < 0.0) {
+		st.x += M_PI * 2.0;
+	}
+
+	uv = st / vec2(M_PI * 2.0, M_PI);
+
+	if (side < 0.0) {
+		//uv.y = 1.0 - uv.y;
+		uv = 1.0 - uv;
+	}
+#endif /* MODE_PANORAMA_TO_DP */
+
+#ifdef USE_MULTIVIEW
+	vec4 color = textureLod(source_color, uv, 0.0);
+#ifdef MODE_TWO_SOURCES
+	// In multiview our 2nd input will be our depth map
+	depth = textureLod(source_depth, uv, 0.0).r;
+#endif /* MODE_TWO_SOURCES */
+
+#else /* USE_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 /* USE_MULTIVIEW */
+
+	if (bool(params.flags & FLAG_FORCE_LUMINANCE)) {
+		color.rgb = vec3(max(max(color.r, color.g), color.b));
+	}
+	if (bool(params.flags & FLAG_ALPHA_TO_ZERO)) {
+		color.rgb *= color.a;
+	}
+	if (bool(params.flags & FLAG_SRGB)) {
+		color.rgb = linear_to_srgb(color.rgb);
+	}
+	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);
+	}
+	if (bool(params.flags & FLAG_NORMAL)) {
+		color.rgb = normalize(color.rgb * 2.0 - 1.0) * 0.5 + 0.5;
+	}
+
+	frag_color = color / params.luminance_multiplier;
+#endif // MODE_SET_COLOR
+}

servers/rendering/renderer_rd/shaders/effects/cube_to_dp.glsl

@@ -0,0 +1,81 @@
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(push_constant, std430) uniform Params {
+	float z_far;
+	float z_near;
+	vec2 texel_size;
+}
+params;
+
+layout(location = 0) out vec2 uv_interp;
+
+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);
+}
+
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(location = 0) in vec2 uv_interp;
+
+layout(set = 0, binding = 0) uniform samplerCube source_cube;
+
+layout(push_constant, std430) uniform Params {
+	float z_far;
+	float z_near;
+	vec2 texel_size;
+}
+params;
+
+void main() {
+	vec2 uv = uv_interp;
+	vec2 texel_size = abs(params.texel_size);
+
+	uv = clamp(uv * (1.0 + 2.0 * texel_size) - texel_size, vec2(0.0), vec2(1.0));
+
+	vec3 normal = vec3(uv * 2.0 - 1.0, 0.0);
+	normal.z = 0.5 * (1.0 - dot(normal.xy, normal.xy)); // z = 1/2 - 1/2 * (x^2 + y^2)
+	normal = normalize(normal);
+
+	normal.y = -normal.y; //needs to be flipped to match projection matrix
+	if (params.texel_size.x >= 0.0) { // Sign is used to encode Z flip
+		normal.z = -normal.z;
+	}
+
+	float depth = texture(source_cube, normal).r;
+
+	// absolute values for direction cosines, bigger value equals closer to basis axis
+	vec3 unorm = abs(normal);
+
+	if ((unorm.x >= unorm.y) && (unorm.x >= unorm.z)) {
+		// x code
+		unorm = normal.x > 0.0 ? vec3(1.0, 0.0, 0.0) : vec3(-1.0, 0.0, 0.0);
+	} else if ((unorm.y > unorm.x) && (unorm.y >= unorm.z)) {
+		// y code
+		unorm = normal.y > 0.0 ? vec3(0.0, 1.0, 0.0) : vec3(0.0, -1.0, 0.0);
+	} else if ((unorm.z > unorm.x) && (unorm.z > unorm.y)) {
+		// z code
+		unorm = normal.z > 0.0 ? vec3(0.0, 0.0, 1.0) : vec3(0.0, 0.0, -1.0);
+	} else {
+		// oh-no we messed up code
+		// has to be
+		unorm = vec3(1.0, 0.0, 0.0);
+	}
+
+	float depth_fix = 1.0 / dot(normal, unorm);
+
+	depth = 2.0 * depth - 1.0;
+	float linear_depth = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near + depth * (params.z_far - params.z_near));
+	// linear_depth equal to view space depth
+	depth = (params.z_far - linear_depth * depth_fix) / params.z_far;
+	gl_FragDepth = depth;
+}

servers/rendering/renderer_rd/shaders/effects/cubemap_downsampler.glsl

@@ -0,0 +1,145 @@
+// Copyright 2016 Activision Publishing, Inc.
+//
+// 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 BLOCK_SIZE 8
+
+layout(local_size_x = BLOCK_SIZE, local_size_y = BLOCK_SIZE, local_size_z = 1) in;
+
+layout(set = 0, binding = 0) uniform samplerCube source_cubemap;
+
+layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly imageCube dest_cubemap;
+
+#include "cubemap_downsampler_inc.glsl"
+
+void main() {
+	uvec3 id = gl_GlobalInvocationID;
+	uint face_size = params.face_size;
+
+	if (id.x < face_size && id.y < face_size) {
+		float inv_face_size = 1.0 / float(face_size);
+
+		float u0 = (float(id.x) * 2.0 + 1.0 - 0.75) * inv_face_size - 1.0;
+		float u1 = (float(id.x) * 2.0 + 1.0 + 0.75) * inv_face_size - 1.0;
+
+		float v0 = (float(id.y) * 2.0 + 1.0 - 0.75) * -inv_face_size + 1.0;
+		float v1 = (float(id.y) * 2.0 + 1.0 + 0.75) * -inv_face_size + 1.0;
+
+		float weights[4];
+		weights[0] = calcWeight(u0, v0);
+		weights[1] = calcWeight(u1, v0);
+		weights[2] = calcWeight(u0, v1);
+		weights[3] = calcWeight(u1, v1);
+
+		const float wsum = 0.5 / (weights[0] + weights[1] + weights[2] + weights[3]);
+		for (int i = 0; i < 4; i++) {
+			weights[i] = weights[i] * wsum + .125;
+		}
+
+		vec3 dir;
+		vec4 color;
+		switch (id.z) {
+			case 0:
+				get_dir_0(dir, u0, v0);
+				color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+				get_dir_0(dir, u1, v0);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+				get_dir_0(dir, u0, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+				get_dir_0(dir, u1, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+				break;
+			case 1:
+				get_dir_1(dir, u0, v0);
+				color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+				get_dir_1(dir, u1, v0);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+				get_dir_1(dir, u0, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+				get_dir_1(dir, u1, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+				break;
+			case 2:
+				get_dir_2(dir, u0, v0);
+				color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+				get_dir_2(dir, u1, v0);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+				get_dir_2(dir, u0, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+				get_dir_2(dir, u1, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+				break;
+			case 3:
+				get_dir_3(dir, u0, v0);
+				color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+				get_dir_3(dir, u1, v0);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+				get_dir_3(dir, u0, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+				get_dir_3(dir, u1, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+				break;
+			case 4:
+				get_dir_4(dir, u0, v0);
+				color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+				get_dir_4(dir, u1, v0);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+				get_dir_4(dir, u0, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+				get_dir_4(dir, u1, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+				break;
+			default:
+				get_dir_5(dir, u0, v0);
+				color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+				get_dir_5(dir, u1, v0);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+				get_dir_5(dir, u0, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+				get_dir_5(dir, u1, v1);
+				color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+				break;
+		}
+		imageStore(dest_cubemap, ivec3(id), color);
+	}
+}

servers/rendering/renderer_rd/shaders/effects/cubemap_downsampler_inc.glsl

@@ -0,0 +1,48 @@
+layout(push_constant, std430) uniform Params {
+	uint face_size;
+	uint face_id; // only used in raster shader
+}
+params;
+
+#define M_PI 3.14159265359
+
+void get_dir_0(out vec3 dir, in float u, in float v) {
+	dir[0] = 1.0;
+	dir[1] = v;
+	dir[2] = -u;
+}
+
+void get_dir_1(out vec3 dir, in float u, in float v) {
+	dir[0] = -1.0;
+	dir[1] = v;
+	dir[2] = u;
+}
+
+void get_dir_2(out vec3 dir, in float u, in float v) {
+	dir[0] = u;
+	dir[1] = 1.0;
+	dir[2] = -v;
+}
+
+void get_dir_3(out vec3 dir, in float u, in float v) {
+	dir[0] = u;
+	dir[1] = -1.0;
+	dir[2] = v;
+}
+
+void get_dir_4(out vec3 dir, in float u, in float v) {
+	dir[0] = u;
+	dir[1] = v;
+	dir[2] = 1.0;
+}
+
+void get_dir_5(out vec3 dir, in float u, in float v) {
+	dir[0] = -u;
+	dir[1] = v;
+	dir[2] = -1.0;
+}
+
+float calcWeight(float u, float v) {
+	float val = u * u + v * v + 1.0;
+	return val * sqrt(val);
+}

servers/rendering/renderer_rd/shaders/effects/cubemap_downsampler_raster.glsl

@@ -0,0 +1,161 @@
+// Copyright 2016 Activision Publishing, Inc.
+//
+// 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.
+
+/* clang-format off */
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "cubemap_downsampler_inc.glsl"
+
+layout(location = 0) out vec2 uv_interp;
+/* clang-format on */
+
+void main() {
+	vec2 base_arr[3] = vec2[](vec2(-1.0, -1.0), vec2(-1.0, 3.0), vec2(3.0, -1.0));
+	gl_Position = vec4(base_arr[gl_VertexIndex], 0.0, 1.0);
+	uv_interp = clamp(gl_Position.xy, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0 * float(params.face_size); // saturate(x) * 2.0
+}
+
+/* clang-format off */
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "cubemap_downsampler_inc.glsl"
+
+layout(set = 0, binding = 0) uniform samplerCube source_cubemap;
+
+layout(location = 0) in vec2 uv_interp;
+layout(location = 0) out vec4 frag_color;
+/* clang-format on */
+
+void main() {
+	// Converted from compute shader which uses absolute coordinates.
+	// Could possibly simplify this
+	float face_size = float(params.face_size);
+	float inv_face_size = 1.0 / face_size;
+	vec2 id = floor(uv_interp);
+
+	float u1 = (id.x * 2.0 + 1.0 + 0.75) * inv_face_size - 1.0;
+	float u0 = (id.x * 2.0 + 1.0 - 0.75) * inv_face_size - 1.0;
+
+	float v0 = (id.y * 2.0 + 1.0 - 0.75) * -inv_face_size + 1.0;
+	float v1 = (id.y * 2.0 + 1.0 + 0.75) * -inv_face_size + 1.0;
+
+	float weights[4];
+	weights[0] = calcWeight(u0, v0);
+	weights[1] = calcWeight(u1, v0);
+	weights[2] = calcWeight(u0, v1);
+	weights[3] = calcWeight(u1, v1);
+
+	const float wsum = 0.5 / (weights[0] + weights[1] + weights[2] + weights[3]);
+	for (int i = 0; i < 4; i++) {
+		weights[i] = weights[i] * wsum + .125;
+	}
+
+	vec3 dir;
+	vec4 color;
+	switch (params.face_id) {
+		case 0:
+			get_dir_0(dir, u0, v0);
+			color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+			get_dir_0(dir, u1, v0);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+			get_dir_0(dir, u0, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+			get_dir_0(dir, u1, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+			break;
+		case 1:
+			get_dir_1(dir, u0, v0);
+			color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+			get_dir_1(dir, u1, v0);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+			get_dir_1(dir, u0, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+			get_dir_1(dir, u1, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+			break;
+		case 2:
+			get_dir_2(dir, u0, v0);
+			color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+			get_dir_2(dir, u1, v0);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+			get_dir_2(dir, u0, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+			get_dir_2(dir, u1, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+			break;
+		case 3:
+			get_dir_3(dir, u0, v0);
+			color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+			get_dir_3(dir, u1, v0);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+			get_dir_3(dir, u0, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+			get_dir_3(dir, u1, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+			break;
+		case 4:
+			get_dir_4(dir, u0, v0);
+			color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+			get_dir_4(dir, u1, v0);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+			get_dir_4(dir, u0, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+			get_dir_4(dir, u1, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+			break;
+		default:
+			get_dir_5(dir, u0, v0);
+			color = textureLod(source_cubemap, normalize(dir), 0.0) * weights[0];
+
+			get_dir_5(dir, u1, v0);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[1];
+
+			get_dir_5(dir, u0, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[2];
+
+			get_dir_5(dir, u1, v1);
+			color += textureLod(source_cubemap, normalize(dir), 0.0) * weights[3];
+			break;
+	}
+	frag_color = color;
+}

servers/rendering/renderer_rd/shaders/effects/cubemap_filter.glsl

@@ -0,0 +1,329 @@
+// Copyright 2016 Activision Publishing, Inc.
+//
+// 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 GROUP_SIZE 64
+
+layout(local_size_x = GROUP_SIZE, local_size_y = 1, local_size_z = 1) in;
+
+layout(set = 0, binding = 0) uniform samplerCube source_cubemap;
+layout(rgba16f, set = 2, binding = 0) uniform restrict writeonly imageCube dest_cubemap0;
+layout(rgba16f, set = 2, binding = 1) uniform restrict writeonly imageCube dest_cubemap1;
+layout(rgba16f, set = 2, binding = 2) uniform restrict writeonly imageCube dest_cubemap2;
+layout(rgba16f, set = 2, binding = 3) uniform restrict writeonly imageCube dest_cubemap3;
+layout(rgba16f, set = 2, binding = 4) uniform restrict writeonly imageCube dest_cubemap4;
+layout(rgba16f, set = 2, binding = 5) uniform restrict writeonly imageCube dest_cubemap5;
+layout(rgba16f, set = 2, binding = 6) uniform restrict writeonly imageCube dest_cubemap6;
+
+#ifdef USE_HIGH_QUALITY
+#define NUM_TAPS 32
+#else
+#define NUM_TAPS 8
+#endif
+
+#define BASE_RESOLUTION 128
+
+#ifdef USE_HIGH_QUALITY
+layout(set = 1, binding = 0, std430) buffer restrict readonly Data {
+	vec4[7][5][3][24] coeffs;
+}
+data;
+#else
+layout(set = 1, binding = 0, std430) buffer restrict readonly Data {
+	vec4[7][5][6] coeffs;
+}
+data;
+#endif
+
+void get_dir(out vec3 dir, in vec2 uv, in uint face) {
+	switch (face) {
+		case 0:
+			dir = vec3(1.0, uv[1], -uv[0]);
+			break;
+		case 1:
+			dir = vec3(-1.0, uv[1], uv[0]);
+			break;
+		case 2:
+			dir = vec3(uv[0], 1.0, -uv[1]);
+			break;
+		case 3:
+			dir = vec3(uv[0], -1.0, uv[1]);
+			break;
+		case 4:
+			dir = vec3(uv[0], uv[1], 1.0);
+			break;
+		default:
+			dir = vec3(-uv[0], uv[1], -1.0);
+			break;
+	}
+}
+
+void main() {
+	// INPUT:
+	// id.x = the linear address of the texel (ignoring face)
+	// id.y = the face
+	// -> use to index output texture
+	// id.x = texel x
+	// id.y = texel y
+	// id.z = face
+	uvec3 id = gl_GlobalInvocationID;
+
+	// determine which texel this is
+#ifndef USE_TEXTURE_ARRAY
+	// NOTE (macOS/MoltenVK): Do not rename, "level" variable name conflicts with the Metal "level(float lod)" mipmap sampling function name.
+	int mip_level = 0;
+	if (id.x < (128 * 128)) {
+		mip_level = 0;
+	} else if (id.x < (128 * 128 + 64 * 64)) {
+		mip_level = 1;
+		id.x -= (128 * 128);
+	} else if (id.x < (128 * 128 + 64 * 64 + 32 * 32)) {
+		mip_level = 2;
+		id.x -= (128 * 128 + 64 * 64);
+	} else if (id.x < (128 * 128 + 64 * 64 + 32 * 32 + 16 * 16)) {
+		mip_level = 3;
+		id.x -= (128 * 128 + 64 * 64 + 32 * 32);
+	} else if (id.x < (128 * 128 + 64 * 64 + 32 * 32 + 16 * 16 + 8 * 8)) {
+		mip_level = 4;
+		id.x -= (128 * 128 + 64 * 64 + 32 * 32 + 16 * 16);
+	} else if (id.x < (128 * 128 + 64 * 64 + 32 * 32 + 16 * 16 + 8 * 8 + 4 * 4)) {
+		mip_level = 5;
+		id.x -= (128 * 128 + 64 * 64 + 32 * 32 + 16 * 16 + 8 * 8);
+	} else if (id.x < (128 * 128 + 64 * 64 + 32 * 32 + 16 * 16 + 8 * 8 + 4 * 4 + 2 * 2)) {
+		mip_level = 6;
+		id.x -= (128 * 128 + 64 * 64 + 32 * 32 + 16 * 16 + 8 * 8 + 4 * 4);
+	} else {
+		return;
+	}
+	int res = BASE_RESOLUTION >> mip_level;
+#else // Using Texture Arrays so all levels are the same resolution
+	int res = BASE_RESOLUTION;
+	int mip_level = int(id.x / (BASE_RESOLUTION * BASE_RESOLUTION));
+	id.x -= mip_level * BASE_RESOLUTION * BASE_RESOLUTION;
+#endif
+
+	// determine dir / pos for the texel
+	vec3 dir, adir, frameZ;
+	{
+		id.z = id.y;
+		id.y = id.x / res;
+		id.x -= id.y * res;
+
+		vec2 uv;
+		uv.x = (float(id.x) * 2.0 + 1.0) / float(res) - 1.0;
+		uv.y = -(float(id.y) * 2.0 + 1.0) / float(res) + 1.0;
+
+		get_dir(dir, uv, id.z);
+		frameZ = normalize(dir);
+
+		adir = abs(dir);
+	}
+
+	// GGX gather colors
+	vec4 color = vec4(0.0);
+	for (int axis = 0; axis < 3; axis++) {
+		const int otherAxis0 = 1 - (axis & 1) - (axis >> 1);
+		const int otherAxis1 = 2 - (axis >> 1);
+
+		float frameweight = (max(adir[otherAxis0], adir[otherAxis1]) - .75) / .25;
+		if (frameweight > 0.0) {
+			// determine frame
+			vec3 UpVector;
+			switch (axis) {
+				case 0:
+					UpVector = vec3(1, 0, 0);
+					break;
+				case 1:
+					UpVector = vec3(0, 1, 0);
+					break;
+				default:
+					UpVector = vec3(0, 0, 1);
+					break;
+			}
+
+			vec3 frameX = normalize(cross(UpVector, frameZ));
+			vec3 frameY = cross(frameZ, frameX);
+
+			// calculate parametrization for polynomial
+			float Nx = dir[otherAxis0];
+			float Ny = dir[otherAxis1];
+			float Nz = adir[axis];
+
+			float NmaxXY = max(abs(Ny), abs(Nx));
+			Nx /= NmaxXY;
+			Ny /= NmaxXY;
+
+			float theta;
+			if (Ny < Nx) {
+				if (Ny <= -0.999) {
+					theta = Nx;
+				} else {
+					theta = Ny;
+				}
+			} else {
+				if (Ny >= 0.999) {
+					theta = -Nx;
+				} else {
+					theta = -Ny;
+				}
+			}
+
+			float phi;
+			if (Nz <= -0.999) {
+				phi = -NmaxXY;
+			} else if (Nz >= 0.999) {
+				phi = NmaxXY;
+			} else {
+				phi = Nz;
+			}
+
+			float theta2 = theta * theta;
+			float phi2 = phi * phi;
+
+			// sample
+			for (int iSuperTap = 0; iSuperTap < NUM_TAPS / 4; iSuperTap++) {
+				const int index = (NUM_TAPS / 4) * axis + iSuperTap;
+
+#ifdef USE_HIGH_QUALITY
+				vec4 coeffsDir0[3];
+				vec4 coeffsDir1[3];
+				vec4 coeffsDir2[3];
+				vec4 coeffsLevel[3];
+				vec4 coeffsWeight[3];
+
+				for (int iCoeff = 0; iCoeff < 3; iCoeff++) {
+					coeffsDir0[iCoeff] = data.coeffs[mip_level][0][iCoeff][index];
+					coeffsDir1[iCoeff] = data.coeffs[mip_level][1][iCoeff][index];
+					coeffsDir2[iCoeff] = data.coeffs[mip_level][2][iCoeff][index];
+					coeffsLevel[iCoeff] = data.coeffs[mip_level][3][iCoeff][index];
+					coeffsWeight[iCoeff] = data.coeffs[mip_level][4][iCoeff][index];
+				}
+
+				for (int iSubTap = 0; iSubTap < 4; iSubTap++) {
+					// determine sample attributes (dir, weight, mip_level)
+					vec3 sample_dir = frameX * (coeffsDir0[0][iSubTap] + coeffsDir0[1][iSubTap] * theta2 + coeffsDir0[2][iSubTap] * phi2) + frameY * (coeffsDir1[0][iSubTap] + coeffsDir1[1][iSubTap] * theta2 + coeffsDir1[2][iSubTap] * phi2) + frameZ * (coeffsDir2[0][iSubTap] + coeffsDir2[1][iSubTap] * theta2 + coeffsDir2[2][iSubTap] * phi2);
+
+					float sample_level = coeffsLevel[0][iSubTap] + coeffsLevel[1][iSubTap] * theta2 + coeffsLevel[2][iSubTap] * phi2;
+
+					float sample_weight = coeffsWeight[0][iSubTap] + coeffsWeight[1][iSubTap] * theta2 + coeffsWeight[2][iSubTap] * phi2;
+#else
+				vec4 coeffsDir0 = data.coeffs[mip_level][0][index];
+				vec4 coeffsDir1 = data.coeffs[mip_level][1][index];
+				vec4 coeffsDir2 = data.coeffs[mip_level][2][index];
+				vec4 coeffsLevel = data.coeffs[mip_level][3][index];
+				vec4 coeffsWeight = data.coeffs[mip_level][4][index];
+
+				for (int iSubTap = 0; iSubTap < 4; iSubTap++) {
+					// determine sample attributes (dir, weight, mip_level)
+					vec3 sample_dir = frameX * coeffsDir0[iSubTap] + frameY * coeffsDir1[iSubTap] + frameZ * coeffsDir2[iSubTap];
+
+					float sample_level = coeffsLevel[iSubTap];
+
+					float sample_weight = coeffsWeight[iSubTap];
+#endif
+
+					sample_weight *= frameweight;
+
+					// adjust for jacobian
+					sample_dir /= max(abs(sample_dir[0]), max(abs(sample_dir[1]), abs(sample_dir[2])));
+					sample_level += 0.75 * log2(dot(sample_dir, sample_dir));
+#ifndef USE_TEXTURE_ARRAY
+					sample_level += float(mip_level) / 6.0; // Hack to increase the perceived roughness and reduce upscaling artifacts
+#endif
+					// sample cubemap
+					color.xyz += textureLod(source_cubemap, normalize(sample_dir), sample_level).xyz * sample_weight;
+					color.w += sample_weight;
+				}
+			}
+		}
+	}
+	color /= color.w;
+
+	// write color
+	color.xyz = max(vec3(0.0), color.xyz);
+	color.w = 1.0;
+#ifdef USE_TEXTURE_ARRAY
+	id.xy *= uvec2(2, 2);
+#endif
+
+	switch (mip_level) {
+		case 0:
+			imageStore(dest_cubemap0, ivec3(id), color);
+#ifdef USE_TEXTURE_ARRAY
+			imageStore(dest_cubemap0, ivec3(id) + ivec3(1.0, 0.0, 0.0), color);
+			imageStore(dest_cubemap0, ivec3(id) + ivec3(0.0, 1.0, 0.0), color);
+			imageStore(dest_cubemap0, ivec3(id) + ivec3(1.0, 1.0, 0.0), color);
+#endif
+			break;
+		case 1:
+			imageStore(dest_cubemap1, ivec3(id), color);
+#ifdef USE_TEXTURE_ARRAY
+			imageStore(dest_cubemap1, ivec3(id) + ivec3(1.0, 0.0, 0.0), color);
+			imageStore(dest_cubemap1, ivec3(id) + ivec3(0.0, 1.0, 0.0), color);
+			imageStore(dest_cubemap1, ivec3(id) + ivec3(1.0, 1.0, 0.0), color);
+#endif
+			break;
+		case 2:
+			imageStore(dest_cubemap2, ivec3(id), color);
+#ifdef USE_TEXTURE_ARRAY
+			imageStore(dest_cubemap2, ivec3(id) + ivec3(1.0, 0.0, 0.0), color);
+			imageStore(dest_cubemap2, ivec3(id) + ivec3(0.0, 1.0, 0.0), color);
+			imageStore(dest_cubemap2, ivec3(id) + ivec3(1.0, 1.0, 0.0), color);
+#endif
+			break;
+		case 3:
+			imageStore(dest_cubemap3, ivec3(id), color);
+#ifdef USE_TEXTURE_ARRAY
+			imageStore(dest_cubemap3, ivec3(id) + ivec3(1.0, 0.0, 0.0), color);
+			imageStore(dest_cubemap3, ivec3(id) + ivec3(0.0, 1.0, 0.0), color);
+			imageStore(dest_cubemap3, ivec3(id) + ivec3(1.0, 1.0, 0.0), color);
+#endif
+			break;
+		case 4:
+			imageStore(dest_cubemap4, ivec3(id), color);
+#ifdef USE_TEXTURE_ARRAY
+			imageStore(dest_cubemap4, ivec3(id) + ivec3(1.0, 0.0, 0.0), color);
+			imageStore(dest_cubemap4, ivec3(id) + ivec3(0.0, 1.0, 0.0), color);
+			imageStore(dest_cubemap4, ivec3(id) + ivec3(1.0, 1.0, 0.0), color);
+#endif
+			break;
+		case 5:
+			imageStore(dest_cubemap5, ivec3(id), color);
+#ifdef USE_TEXTURE_ARRAY
+			imageStore(dest_cubemap5, ivec3(id) + ivec3(1.0, 0.0, 0.0), color);
+			imageStore(dest_cubemap5, ivec3(id) + ivec3(0.0, 1.0, 0.0), color);
+			imageStore(dest_cubemap5, ivec3(id) + ivec3(1.0, 1.0, 0.0), color);
+#endif
+			break;
+		default:
+			imageStore(dest_cubemap6, ivec3(id), color);
+#ifdef USE_TEXTURE_ARRAY
+			imageStore(dest_cubemap6, ivec3(id) + ivec3(1.0, 0.0, 0.0), color);
+			imageStore(dest_cubemap6, ivec3(id) + ivec3(0.0, 1.0, 0.0), color);
+			imageStore(dest_cubemap6, ivec3(id) + ivec3(1.0, 1.0, 0.0), color);
+#endif
+			break;
+	}
+}

servers/rendering/renderer_rd/shaders/effects/cubemap_filter_raster.glsl

@@ -0,0 +1,259 @@
+// Copyright 2016 Activision Publishing, Inc.
+//
+// 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.
+
+/* clang-format off */
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(push_constant, std430) uniform Params {
+	int mip_level;
+	uint face_id;
+}
+params;
+
+layout(location = 0) out vec2 uv_interp;
+/* clang-format on */
+
+void main() {
+	vec2 base_arr[3] = vec2[](vec2(-1.0, -1.0), vec2(-1.0, 3.0), vec2(3.0, -1.0));
+	gl_Position = vec4(base_arr[gl_VertexIndex], 0.0, 1.0);
+	uv_interp = clamp(gl_Position.xy, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+}
+
+/* clang-format off */
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(push_constant, std430) uniform Params {
+	int mip_level;
+	uint face_id;
+}
+params;
+
+layout(set = 0, binding = 0) uniform samplerCube source_cubemap;
+
+layout(location = 0) in vec2 uv_interp;
+layout(location = 0) out vec4 frag_color;
+
+/* clang-format on */
+
+#ifdef USE_HIGH_QUALITY
+#define NUM_TAPS 32
+#else
+#define NUM_TAPS 8
+#endif
+
+#define BASE_RESOLUTION 128
+
+#ifdef USE_HIGH_QUALITY
+layout(set = 1, binding = 0, std430) buffer restrict readonly Data {
+	vec4[7][5][3][24] coeffs;
+}
+data;
+#else
+layout(set = 1, binding = 0, std430) buffer restrict readonly Data {
+	vec4[7][5][6] coeffs;
+}
+data;
+#endif
+
+void get_dir(out vec3 dir, in vec2 uv, in uint face) {
+	switch (face) {
+		case 0:
+			dir = vec3(1.0, uv[1], -uv[0]);
+			break;
+		case 1:
+			dir = vec3(-1.0, uv[1], uv[0]);
+			break;
+		case 2:
+			dir = vec3(uv[0], 1.0, -uv[1]);
+			break;
+		case 3:
+			dir = vec3(uv[0], -1.0, uv[1]);
+			break;
+		case 4:
+			dir = vec3(uv[0], uv[1], 1.0);
+			break;
+		default:
+			dir = vec3(-uv[0], uv[1], -1.0);
+			break;
+	}
+}
+
+void main() {
+	// determine dir / pos for the texel
+	vec3 dir, adir, frameZ;
+	{
+		vec2 uv;
+		uv.x = uv_interp.x;
+		uv.y = 1.0 - uv_interp.y;
+		uv = uv * 2.0 - 1.0;
+
+		get_dir(dir, uv, params.face_id);
+		frameZ = normalize(dir);
+
+		adir = abs(dir);
+	}
+
+	// determine which texel this is
+	// NOTE (macOS/MoltenVK): Do not rename, "level" variable name conflicts with the Metal "level(float lod)" mipmap sampling function name.
+	int mip_level = 0;
+
+	if (params.mip_level < 0) {
+		// return as is
+		frag_color.rgb = textureLod(source_cubemap, frameZ, 0.0).rgb;
+		frag_color.a = 1.0;
+		return;
+	} else if (params.mip_level > 6) {
+		// maximum level
+		mip_level = 6;
+	} else {
+		mip_level = params.mip_level;
+	}
+
+	// GGX gather colors
+	vec4 color = vec4(0.0);
+	for (int axis = 0; axis < 3; axis++) {
+		const int otherAxis0 = 1 - (axis & 1) - (axis >> 1);
+		const int otherAxis1 = 2 - (axis >> 1);
+
+		float frameweight = (max(adir[otherAxis0], adir[otherAxis1]) - .75) / .25;
+		if (frameweight > 0.0) {
+			// determine frame
+			vec3 UpVector;
+			switch (axis) {
+				case 0:
+					UpVector = vec3(1, 0, 0);
+					break;
+				case 1:
+					UpVector = vec3(0, 1, 0);
+					break;
+				default:
+					UpVector = vec3(0, 0, 1);
+					break;
+			}
+
+			vec3 frameX = normalize(cross(UpVector, frameZ));
+			vec3 frameY = cross(frameZ, frameX);
+
+			// calculate parametrization for polynomial
+			float Nx = dir[otherAxis0];
+			float Ny = dir[otherAxis1];
+			float Nz = adir[axis];
+
+			float NmaxXY = max(abs(Ny), abs(Nx));
+			Nx /= NmaxXY;
+			Ny /= NmaxXY;
+
+			float theta;
+			if (Ny < Nx) {
+				if (Ny <= -0.999) {
+					theta = Nx;
+				} else {
+					theta = Ny;
+				}
+			} else {
+				if (Ny >= 0.999) {
+					theta = -Nx;
+				} else {
+					theta = -Ny;
+				}
+			}
+
+			float phi;
+			if (Nz <= -0.999) {
+				phi = -NmaxXY;
+			} else if (Nz >= 0.999) {
+				phi = NmaxXY;
+			} else {
+				phi = Nz;
+			}
+
+			float theta2 = theta * theta;
+			float phi2 = phi * phi;
+
+			// sample
+			for (int iSuperTap = 0; iSuperTap < NUM_TAPS / 4; iSuperTap++) {
+				const int index = (NUM_TAPS / 4) * axis + iSuperTap;
+
+#ifdef USE_HIGH_QUALITY
+				vec4 coeffsDir0[3];
+				vec4 coeffsDir1[3];
+				vec4 coeffsDir2[3];
+				vec4 coeffsLevel[3];
+				vec4 coeffsWeight[3];
+
+				for (int iCoeff = 0; iCoeff < 3; iCoeff++) {
+					coeffsDir0[iCoeff] = data.coeffs[mip_level][0][iCoeff][index];
+					coeffsDir1[iCoeff] = data.coeffs[mip_level][1][iCoeff][index];
+					coeffsDir2[iCoeff] = data.coeffs[mip_level][2][iCoeff][index];
+					coeffsLevel[iCoeff] = data.coeffs[mip_level][3][iCoeff][index];
+					coeffsWeight[iCoeff] = data.coeffs[mip_level][4][iCoeff][index];
+				}
+
+				for (int iSubTap = 0; iSubTap < 4; iSubTap++) {
+					// determine sample attributes (dir, weight, mip_level)
+					vec3 sample_dir = frameX * (coeffsDir0[0][iSubTap] + coeffsDir0[1][iSubTap] * theta2 + coeffsDir0[2][iSubTap] * phi2) + frameY * (coeffsDir1[0][iSubTap] + coeffsDir1[1][iSubTap] * theta2 + coeffsDir1[2][iSubTap] * phi2) + frameZ * (coeffsDir2[0][iSubTap] + coeffsDir2[1][iSubTap] * theta2 + coeffsDir2[2][iSubTap] * phi2);
+
+					float sample_level = coeffsLevel[0][iSubTap] + coeffsLevel[1][iSubTap] * theta2 + coeffsLevel[2][iSubTap] * phi2;
+
+					float sample_weight = coeffsWeight[0][iSubTap] + coeffsWeight[1][iSubTap] * theta2 + coeffsWeight[2][iSubTap] * phi2;
+#else
+				vec4 coeffsDir0 = data.coeffs[mip_level][0][index];
+				vec4 coeffsDir1 = data.coeffs[mip_level][1][index];
+				vec4 coeffsDir2 = data.coeffs[mip_level][2][index];
+				vec4 coeffsLevel = data.coeffs[mip_level][3][index];
+				vec4 coeffsWeight = data.coeffs[mip_level][4][index];
+
+				for (int iSubTap = 0; iSubTap < 4; iSubTap++) {
+					// determine sample attributes (dir, weight, mip_level)
+					vec3 sample_dir = frameX * coeffsDir0[iSubTap] + frameY * coeffsDir1[iSubTap] + frameZ * coeffsDir2[iSubTap];
+
+					float sample_level = coeffsLevel[iSubTap];
+
+					float sample_weight = coeffsWeight[iSubTap];
+#endif
+
+					sample_weight *= frameweight;
+
+					// adjust for jacobian
+					sample_dir /= max(abs(sample_dir[0]), max(abs(sample_dir[1]), abs(sample_dir[2])));
+					sample_level += 0.75 * log2(dot(sample_dir, sample_dir));
+					// sample cubemap
+					color.xyz += textureLod(source_cubemap, normalize(sample_dir), sample_level).xyz * sample_weight;
+					color.w += sample_weight;
+				}
+			}
+		}
+	}
+	color /= color.w;
+
+	// write color
+	color.xyz = max(vec3(0.0), color.xyz);
+	color.w = 1.0;
+
+	frag_color = color;
+}

servers/rendering/renderer_rd/shaders/effects/cubemap_roughness.glsl

@@ -0,0 +1,63 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#define GROUP_SIZE 8
+
+layout(local_size_x = GROUP_SIZE, local_size_y = GROUP_SIZE, local_size_z = 1) in;
+
+layout(set = 0, binding = 0) uniform samplerCube source_cube;
+
+layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly imageCube dest_cubemap;
+
+#include "cubemap_roughness_inc.glsl"
+
+void main() {
+	uvec3 id = gl_GlobalInvocationID;
+	id.z += params.face_id;
+
+	vec2 uv = ((vec2(id.xy) * 2.0 + 1.0) / (params.face_size) - 1.0);
+	vec3 N = texelCoordToVec(uv, id.z);
+
+	if (params.use_direct_write) {
+		imageStore(dest_cubemap, ivec3(id), vec4(texture(source_cube, N).rgb, 1.0));
+	} else {
+		vec4 sum = vec4(0.0, 0.0, 0.0, 0.0);
+
+		float solid_angle_texel = 4.0 * M_PI / (6.0 * params.face_size * params.face_size);
+		float roughness2 = params.roughness * params.roughness;
+		float roughness4 = roughness2 * roughness2;
+		vec3 UpVector = abs(N.z) < 0.999 ? vec3(0.0, 0.0, 1.0) : vec3(1.0, 0.0, 0.0);
+		mat3 T;
+		T[0] = normalize(cross(UpVector, N));
+		T[1] = cross(N, T[0]);
+		T[2] = N;
+
+		for (uint sampleNum = 0u; sampleNum < params.sample_count; sampleNum++) {
+			vec2 xi = Hammersley(sampleNum, params.sample_count);
+
+			vec3 H = T * ImportanceSampleGGX(xi, roughness4);
+			float NdotH = dot(N, H);
+			vec3 L = (2.0 * NdotH * H - N);
+
+			float ndotl = clamp(dot(N, L), 0.0, 1.0);
+
+			if (ndotl > 0.0) {
+				float D = DistributionGGX(NdotH, roughness4);
+				float pdf = D * NdotH / (4.0 * NdotH) + 0.0001;
+
+				float solid_angle_sample = 1.0 / (float(params.sample_count) * pdf + 0.0001);
+
+				float mipLevel = params.roughness == 0.0 ? 0.0 : 0.5 * log2(solid_angle_sample / solid_angle_texel);
+
+				sum.rgb += textureLod(source_cube, L, mipLevel).rgb * ndotl;
+				sum.a += ndotl;
+			}
+		}
+		sum /= sum.a;
+
+		imageStore(dest_cubemap, ivec3(id), vec4(sum.rgb, 1.0));
+	}
+}

servers/rendering/renderer_rd/shaders/effects/cubemap_roughness_inc.glsl

@@ -0,0 +1,84 @@
+#define M_PI 3.14159265359
+
+layout(push_constant, std430) uniform Params {
+	uint face_id;
+	uint sample_count;
+	float roughness;
+	bool use_direct_write;
+	float face_size;
+}
+params;
+
+vec3 texelCoordToVec(vec2 uv, uint faceID) {
+	mat3 faceUvVectors[6];
+
+	// -x
+	faceUvVectors[1][0] = vec3(0.0, 0.0, 1.0); // u -> +z
+	faceUvVectors[1][1] = vec3(0.0, -1.0, 0.0); // v -> -y
+	faceUvVectors[1][2] = vec3(-1.0, 0.0, 0.0); // -x face
+
+	// +x
+	faceUvVectors[0][0] = vec3(0.0, 0.0, -1.0); // u -> -z
+	faceUvVectors[0][1] = vec3(0.0, -1.0, 0.0); // v -> -y
+	faceUvVectors[0][2] = vec3(1.0, 0.0, 0.0); // +x face
+
+	// -y
+	faceUvVectors[3][0] = vec3(1.0, 0.0, 0.0); // u -> +x
+	faceUvVectors[3][1] = vec3(0.0, 0.0, -1.0); // v -> -z
+	faceUvVectors[3][2] = vec3(0.0, -1.0, 0.0); // -y face
+
+	// +y
+	faceUvVectors[2][0] = vec3(1.0, 0.0, 0.0); // u -> +x
+	faceUvVectors[2][1] = vec3(0.0, 0.0, 1.0); // v -> +z
+	faceUvVectors[2][2] = vec3(0.0, 1.0, 0.0); // +y face
+
+	// -z
+	faceUvVectors[5][0] = vec3(-1.0, 0.0, 0.0); // u -> -x
+	faceUvVectors[5][1] = vec3(0.0, -1.0, 0.0); // v -> -y
+	faceUvVectors[5][2] = vec3(0.0, 0.0, -1.0); // -z face
+
+	// +z
+	faceUvVectors[4][0] = vec3(1.0, 0.0, 0.0); // u -> +x
+	faceUvVectors[4][1] = vec3(0.0, -1.0, 0.0); // v -> -y
+	faceUvVectors[4][2] = vec3(0.0, 0.0, 1.0); // +z face
+
+	// out = u * s_faceUv[0] + v * s_faceUv[1] + s_faceUv[2].
+	vec3 result = (faceUvVectors[faceID][0] * uv.x) + (faceUvVectors[faceID][1] * uv.y) + faceUvVectors[faceID][2];
+	return normalize(result);
+}
+
+vec3 ImportanceSampleGGX(vec2 xi, float roughness4) {
+	// Compute distribution direction
+	float Phi = 2.0 * M_PI * xi.x;
+	float CosTheta = sqrt((1.0 - xi.y) / (1.0 + (roughness4 - 1.0) * xi.y));
+	float SinTheta = sqrt(1.0 - CosTheta * CosTheta);
+
+	// Convert to spherical direction
+	vec3 H;
+	H.x = SinTheta * cos(Phi);
+	H.y = SinTheta * sin(Phi);
+	H.z = CosTheta;
+
+	return H;
+}
+
+float DistributionGGX(float NdotH, float roughness4) {
+	float NdotH2 = NdotH * NdotH;
+	float denom = (NdotH2 * (roughness4 - 1.0) + 1.0);
+	denom = M_PI * denom * denom;
+
+	return roughness4 / denom;
+}
+
+float radicalInverse_VdC(uint bits) {
+	bits = (bits << 16u) | (bits >> 16u);
+	bits = ((bits & 0x55555555u) << 1u) | ((bits & 0xAAAAAAAAu) >> 1u);
+	bits = ((bits & 0x33333333u) << 2u) | ((bits & 0xCCCCCCCCu) >> 2u);
+	bits = ((bits & 0x0F0F0F0Fu) << 4u) | ((bits & 0xF0F0F0F0u) >> 4u);
+	bits = ((bits & 0x00FF00FFu) << 8u) | ((bits & 0xFF00FF00u) >> 8u);
+	return float(bits) * 2.3283064365386963e-10; // / 0x100000000
+}
+
+vec2 Hammersley(uint i, uint N) {
+	return vec2(float(i) / float(N), radicalInverse_VdC(i));
+}

servers/rendering/renderer_rd/shaders/effects/cubemap_roughness_raster.glsl

@@ -0,0 +1,79 @@
+/* clang-format off */
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "cubemap_roughness_inc.glsl"
+
+layout(location = 0) out vec2 uv_interp;
+/* clang-format on */
+
+void main() {
+	vec2 base_arr[3] = vec2[](vec2(-1.0, -1.0), vec2(-1.0, 3.0), vec2(3.0, -1.0));
+	gl_Position = vec4(base_arr[gl_VertexIndex], 0.0, 1.0);
+	uv_interp = clamp(gl_Position.xy, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+}
+
+/* clang-format off */
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "cubemap_roughness_inc.glsl"
+
+layout(location = 0) in vec2 uv_interp;
+
+layout(set = 0, binding = 0) uniform samplerCube source_cube;
+
+layout(location = 0) out vec4 frag_color;
+/* clang-format on */
+
+void main() {
+	vec3 N = texelCoordToVec(uv_interp * 2.0 - 1.0, params.face_id);
+
+	//vec4 color = color_interp;
+
+	if (params.use_direct_write) {
+		frag_color = vec4(texture(source_cube, N).rgb, 1.0);
+	} else {
+		vec4 sum = vec4(0.0, 0.0, 0.0, 0.0);
+
+		float solid_angle_texel = 4.0 * M_PI / (6.0 * params.face_size * params.face_size);
+		float roughness2 = params.roughness * params.roughness;
+		float roughness4 = roughness2 * roughness2;
+		vec3 UpVector = abs(N.z) < 0.999 ? vec3(0.0, 0.0, 1.0) : vec3(1.0, 0.0, 0.0);
+		mat3 T;
+		T[0] = normalize(cross(UpVector, N));
+		T[1] = cross(N, T[0]);
+		T[2] = N;
+
+		for (uint sampleNum = 0u; sampleNum < params.sample_count; sampleNum++) {
+			vec2 xi = Hammersley(sampleNum, params.sample_count);
+
+			vec3 H = T * ImportanceSampleGGX(xi, roughness4);
+			float NdotH = dot(N, H);
+			vec3 L = (2.0 * NdotH * H - N);
+
+			float ndotl = clamp(dot(N, L), 0.0, 1.0);
+
+			if (ndotl > 0.0) {
+				float D = DistributionGGX(NdotH, roughness4);
+				float pdf = D * NdotH / (4.0 * NdotH) + 0.0001;
+
+				float solid_angle_sample = 1.0 / (float(params.sample_count) * pdf + 0.0001);
+
+				float mipLevel = params.roughness == 0.0 ? 0.0 : 0.5 * log2(solid_angle_sample / solid_angle_texel);
+
+				sum.rgb += textureLod(source_cube, L, mipLevel).rgb * ndotl;
+				sum.a += ndotl;
+			}
+		}
+		sum /= sum.a;
+
+		frag_color = vec4(sum.rgb, 1.0);
+	}
+}

servers/rendering/renderer_rd/shaders/effects/fsr2/SCsub

@@ -0,0 +1,23 @@
+#!/usr/bin/env python
+from misc.utility.scons_hints import *
+
+Import("env")
+
+if "RD_GLSL" in env["BUILDERS"]:
+    # find all include files
+    gl_include_files = [str(f) for f in Glob("*_inc.glsl")] + [str(f) for f in Glob("../*_inc.glsl")]
+
+    # Add all FSR2 shader and header files.
+    fsr2_dir = "#thirdparty/amd-fsr2/shaders"
+    gl_include_files += [str(f) for f in Glob(fsr2_dir + "/*.h")]
+    gl_include_files += [str(f) for f in Glob(fsr2_dir + "/*.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 + ["#glsl_builders.py"])
+
+    # compile shaders
+    for glsl_file in glsl_files:
+        env.RD_GLSL(glsl_file)

servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_accumulate_pass.glsl

@@ -0,0 +1,8 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "../motion_vector_inc.glsl"
+#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_accumulate_pass.glsl"

servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_autogen_reactive_pass.glsl

@@ -0,0 +1,8 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "../motion_vector_inc.glsl"
+#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_autogen_reactive_pass.glsl"

servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_compute_luminance_pyramid_pass.glsl

@@ -0,0 +1,7 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_compute_luminance_pyramid_pass.glsl"

servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_depth_clip_pass.glsl

@@ -0,0 +1,8 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "../motion_vector_inc.glsl"
+#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_depth_clip_pass.glsl"

servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_lock_pass.glsl

@@ -0,0 +1,7 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_lock_pass.glsl"

servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_rcas_pass.glsl

@@ -0,0 +1,7 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_rcas_pass.glsl"

servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_reconstruct_previous_depth_pass.glsl

@@ -0,0 +1,8 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "../motion_vector_inc.glsl"
+#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_reconstruct_previous_depth_pass.glsl"

servers/rendering/renderer_rd/shaders/effects/fsr2/fsr2_tcr_autogen_pass.glsl

@@ -0,0 +1,8 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "../motion_vector_inc.glsl"
+#include "thirdparty/amd-fsr2/shaders/ffx_fsr2_tcr_autogen_pass.glsl"

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);
+}

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));
+	}
+}

servers/rendering/renderer_rd/shaders/effects/luminance_reduce_raster.glsl

@@ -0,0 +1,73 @@
+/* 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[3] = vec2[](vec2(-1.0, -1.0), vec2(-1.0, 3.0), vec2(3.0, -1.0));
+	gl_Position = vec4(base_arr[gl_VertexIndex], 0.0, 1.0);
+	uv_interp = clamp(gl_Position.xy, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+}
+
+/* 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
+}

servers/rendering/renderer_rd/shaders/effects/luminance_reduce_raster_inc.glsl

@@ -0,0 +1,10 @@
+layout(push_constant, std430) uniform PushConstant {
+	ivec2 source_size;
+	ivec2 dest_size;
+
+	float exposure_adjust;
+	float min_luminance;
+	float max_luminance;
+	uint pad1;
+}
+settings;

servers/rendering/renderer_rd/shaders/effects/motion_vector_inc.glsl

@@ -0,0 +1,6 @@
+vec2 derive_motion_vector(vec2 uv, float depth, mat4 reprojection_matrix) {
+	vec4 previous_pos_ndc = reprojection_matrix * vec4(uv * 2.0f - 1.0f, depth * 2.0f - 1.0f, 1.0f);
+	return 0.5f + (previous_pos_ndc.xy / previous_pos_ndc.w) * 0.5f - uv;
+}
+
+#define FFX_FSR2_OPTION_GODOT_DERIVE_INVALID_MOTION_VECTORS_FUNCTION(i, j, k) derive_motion_vector(i, j, k)

servers/rendering/renderer_rd/shaders/effects/motion_vectors.glsl

@@ -0,0 +1,97 @@
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(location = 0) out vec2 uv_interp;
+
+void main() {
+	vec2 base_arr[3] = vec2[](vec2(-1.0, -1.0), vec2(-1.0, 3.0), vec2(3.0, -1.0));
+	gl_Position = vec4(base_arr[gl_VertexIndex], 0.0, 1.0);
+	uv_interp = clamp(gl_Position.xy, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+}
+
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "motion_vector_inc.glsl"
+
+layout(location = 0) in vec2 uv_interp;
+
+layout(set = 0, binding = 0) uniform sampler2D source_velocity;
+layout(set = 0, binding = 1) uniform sampler2D source_depth;
+
+layout(location = 0) out vec4 frag_color;
+
+layout(push_constant, std430) uniform Params {
+	highp mat4 reprojection_matrix;
+	vec2 resolution;
+	bool force_derive_from_depth;
+}
+params;
+
+// Based on distance to line segment from https://www.shadertoy.com/view/3tdSDj
+
+float line_segment(in vec2 p, in vec2 a, in vec2 b) {
+	vec2 aspect = vec2(params.resolution.x / params.resolution.y, 1.0f);
+	vec2 ba = (b - a) * aspect;
+	vec2 pa = (p - a) * aspect;
+	float h = clamp(dot(pa, ba) / dot(ba, ba), 0.0f, 1.0f);
+	return length(pa - h * ba) * (params.resolution.y / 2.0f);
+}
+
+void main() {
+	// Retrieve motion vector data.
+	float cell_size = 32.0f;
+	float circle_radius = 2.0f;
+	vec3 nan_color = vec3(1.0f, 0.0f, 0.0f);
+	vec3 active_color = vec3(1.0f, 0.8f, 0.1f);
+	vec3 inactive_color = vec3(0.5f, 0.5f, 0.5f);
+	vec2 pos_pixel = uv_interp * params.resolution;
+	vec2 cell_pos_pixel = floor(pos_pixel / cell_size) * cell_size + (cell_size * 0.5f);
+	vec2 cell_pos_uv = cell_pos_pixel / params.resolution;
+	vec2 cell_pos_velocity = textureLod(source_velocity, cell_pos_uv, 0.0f).xy;
+	bool derive_velocity = params.force_derive_from_depth || all(lessThanEqual(cell_pos_velocity, vec2(-1.0f, -1.0f)));
+	if (derive_velocity) {
+		float depth = textureLod(source_depth, cell_pos_uv, 0.0f).x;
+		cell_pos_velocity = derive_motion_vector(cell_pos_uv, depth, params.reprojection_matrix);
+	}
+
+	vec2 cell_pos_previous_uv = cell_pos_uv + cell_pos_velocity;
+
+	// Draw the shapes.
+	float epsilon = 1e-6f;
+	vec2 cell_pos_delta_uv = cell_pos_uv - cell_pos_previous_uv;
+	bool motion_active = length(cell_pos_delta_uv) > epsilon;
+	vec3 color;
+	if (any(isnan(cell_pos_delta_uv))) {
+		color = nan_color;
+	} else if (motion_active) {
+		color = active_color;
+	} else {
+		color = inactive_color;
+	}
+
+	float alpha;
+	if (length(cell_pos_pixel - pos_pixel) <= circle_radius) {
+		// Circle center.
+		alpha = 1.0f;
+	} else if (motion_active) {
+		// Motion vector line.
+		alpha = 1.0f - line_segment(uv_interp, cell_pos_uv, cell_pos_previous_uv);
+	} else {
+		// Ignore pixel.
+		alpha = 0.0f;
+	}
+
+	if (derive_velocity) {
+		color = vec3(1.0f, 1.0f, 1.0f) - color;
+		alpha *= 0.5f;
+	}
+
+	frag_color = vec4(color, alpha);
+}

servers/rendering/renderer_rd/shaders/effects/motion_vectors_store.glsl

@@ -0,0 +1,32 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+#include "motion_vector_inc.glsl"
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+layout(set = 0, binding = 0) uniform sampler2D depth_buffer;
+layout(rg16f, set = 0, binding = 1) uniform restrict writeonly image2D velocity_buffer;
+
+layout(push_constant, std430) uniform Params {
+	highp mat4 reprojection_matrix;
+	vec2 resolution;
+	uint pad[2];
+}
+params;
+
+void main() {
+	// Out of bounds check.
+	if (any(greaterThanEqual(vec2(gl_GlobalInvocationID.xy), params.resolution))) {
+		return;
+	}
+
+	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
+
+	float depth = texelFetch(depth_buffer, pos, 0).x;
+	vec2 uv = (vec2(pos) + 0.5f) / params.resolution;
+	vec2 velocity = derive_motion_vector(uv, depth, params.reprojection_matrix);
+	imageStore(velocity_buffer, pos, vec4(velocity, 0.0f, 0.0f));
+}

servers/rendering/renderer_rd/shaders/effects/resolve.glsl

@@ -0,0 +1,236 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+#ifdef MODE_RESOLVE_DEPTH
+layout(set = 0, binding = 0) uniform sampler2DMS source_depth;
+layout(r32f, set = 1, binding = 0) uniform restrict writeonly image2D dest_depth;
+#endif
+
+#ifdef MODE_RESOLVE_GI
+layout(set = 0, binding = 0) uniform sampler2DMS source_depth;
+layout(set = 0, binding = 1) uniform sampler2DMS source_normal_roughness;
+
+layout(r32f, set = 1, binding = 0) uniform restrict writeonly image2D dest_depth;
+layout(rgba8, set = 1, binding = 1) uniform restrict writeonly image2D dest_normal_roughness;
+
+#ifdef VOXEL_GI_RESOLVE
+layout(set = 2, binding = 0) uniform usampler2DMS source_voxel_gi;
+layout(rg8ui, set = 3, binding = 0) uniform restrict writeonly uimage2D dest_voxel_gi;
+#endif
+
+#endif
+
+layout(push_constant, std430) uniform Params {
+	ivec2 screen_size;
+	int sample_count;
+	uint pad;
+}
+params;
+
+void main() {
+	// Pixel being shaded
+	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
+	if (any(greaterThanEqual(pos, params.screen_size))) { //too large, do nothing
+		return;
+	}
+
+#ifdef MODE_RESOLVE_DEPTH
+
+	float depth_avg = 0.0;
+	for (int i = 0; i < params.sample_count; i++) {
+		depth_avg += texelFetch(source_depth, pos, i).r;
+	}
+	depth_avg /= float(params.sample_count);
+	imageStore(dest_depth, pos, vec4(depth_avg));
+
+#endif
+
+#ifdef MODE_RESOLVE_GI
+
+	float best_depth = 1e20;
+	vec4 best_normal_roughness = vec4(0.0);
+#ifdef VOXEL_GI_RESOLVE
+	uvec2 best_voxel_gi;
+#endif
+
+#if 0
+
+	for(int i=0;i<params.sample_count;i++) {
+		float depth = texelFetch(source_depth,pos,i).r;
+		if (depth < best_depth) { //use the depth closest to camera
+			best_depth = depth;
+			best_normal_roughness = texelFetch(source_normal_roughness,pos,i);
+
+#ifdef VOXEL_GI_RESOLVE
+			best_voxel_gi = texelFetch(source_voxel_gi,pos,i).rg;
+#endif
+		}
+	}
+
+#else
+
+#if 1
+
+	vec4 group1;
+	vec4 group2;
+	vec4 group3;
+	vec4 group4;
+	int best_index = 0;
+
+	//2X
+	group1.x = texelFetch(source_depth, pos, 0).r;
+	group1.y = texelFetch(source_depth, pos, 1).r;
+
+	//4X
+	if (params.sample_count >= 4) {
+		group1.z = texelFetch(source_depth, pos, 2).r;
+		group1.w = texelFetch(source_depth, pos, 3).r;
+	}
+	//8X
+	if (params.sample_count >= 8) {
+		group2.x = texelFetch(source_depth, pos, 4).r;
+		group2.y = texelFetch(source_depth, pos, 5).r;
+		group2.z = texelFetch(source_depth, pos, 6).r;
+		group2.w = texelFetch(source_depth, pos, 7).r;
+	}
+	//16X
+	if (params.sample_count >= 16) {
+		group3.x = texelFetch(source_depth, pos, 8).r;
+		group3.y = texelFetch(source_depth, pos, 9).r;
+		group3.z = texelFetch(source_depth, pos, 10).r;
+		group3.w = texelFetch(source_depth, pos, 11).r;
+
+		group4.x = texelFetch(source_depth, pos, 12).r;
+		group4.y = texelFetch(source_depth, pos, 13).r;
+		group4.z = texelFetch(source_depth, pos, 14).r;
+		group4.w = texelFetch(source_depth, pos, 15).r;
+	}
+
+	if (params.sample_count == 2) {
+		best_index = (pos.x & 1) ^ ((pos.y >> 1) & 1); //not much can be done here
+	} else if (params.sample_count == 4) {
+		vec4 freq = vec4(equal(group1, vec4(group1.x)));
+		freq += vec4(equal(group1, vec4(group1.y)));
+		freq += vec4(equal(group1, vec4(group1.z)));
+		freq += vec4(equal(group1, vec4(group1.w)));
+
+		float min_f = freq.x;
+		best_index = 0;
+		if (freq.y < min_f) {
+			best_index = 1;
+			min_f = freq.y;
+		}
+		if (freq.z < min_f) {
+			best_index = 2;
+			min_f = freq.z;
+		}
+		if (freq.w < min_f) {
+			best_index = 3;
+		}
+	} else if (params.sample_count == 8) {
+		vec4 freq0 = vec4(equal(group1, vec4(group1.x)));
+		vec4 freq1 = vec4(equal(group2, vec4(group1.x)));
+		freq0 += vec4(equal(group1, vec4(group1.y)));
+		freq1 += vec4(equal(group2, vec4(group1.y)));
+		freq0 += vec4(equal(group1, vec4(group1.z)));
+		freq1 += vec4(equal(group2, vec4(group1.z)));
+		freq0 += vec4(equal(group1, vec4(group1.w)));
+		freq1 += vec4(equal(group2, vec4(group1.w)));
+		freq0 += vec4(equal(group1, vec4(group2.x)));
+		freq1 += vec4(equal(group2, vec4(group2.x)));
+		freq0 += vec4(equal(group1, vec4(group2.y)));
+		freq1 += vec4(equal(group2, vec4(group2.y)));
+		freq0 += vec4(equal(group1, vec4(group2.z)));
+		freq1 += vec4(equal(group2, vec4(group2.z)));
+		freq0 += vec4(equal(group1, vec4(group2.w)));
+		freq1 += vec4(equal(group2, vec4(group2.w)));
+
+		float min_f0 = freq0.x;
+		int best_index0 = 0;
+		if (freq0.y < min_f0) {
+			best_index0 = 1;
+			min_f0 = freq0.y;
+		}
+		if (freq0.z < min_f0) {
+			best_index0 = 2;
+			min_f0 = freq0.z;
+		}
+		if (freq0.w < min_f0) {
+			best_index0 = 3;
+			min_f0 = freq0.w;
+		}
+
+		float min_f1 = freq1.x;
+		int best_index1 = 4;
+		if (freq1.y < min_f1) {
+			best_index1 = 5;
+			min_f1 = freq1.y;
+		}
+		if (freq1.z < min_f1) {
+			best_index1 = 6;
+			min_f1 = freq1.z;
+		}
+		if (freq1.w < min_f1) {
+			best_index1 = 7;
+			min_f1 = freq1.w;
+		}
+
+		best_index = mix(best_index0, best_index1, min_f0 < min_f1);
+	}
+
+#else
+	float depths[16];
+	int depth_indices[16];
+	int depth_amount[16];
+	int depth_count = 0;
+
+	for (int i = 0; i < params.sample_count; i++) {
+		float depth = texelFetch(source_depth, pos, i).r;
+		int depth_index = -1;
+		for (int j = 0; j < depth_count; j++) {
+			if (abs(depths[j] - depth) < 0.000001) {
+				depth_index = j;
+				break;
+			}
+		}
+
+		if (depth_index == -1) {
+			depths[depth_count] = depth;
+			depth_indices[depth_count] = i;
+			depth_amount[depth_count] = 1;
+			depth_count += 1;
+		} else {
+			depth_amount[depth_index] += 1;
+		}
+	}
+
+	int depth_least = 0xFFFF;
+	int best_index = 0;
+	for (int j = 0; j < depth_count; j++) {
+		if (depth_amount[j] < depth_least) {
+			best_index = depth_indices[j];
+			depth_least = depth_amount[j];
+		}
+	}
+#endif
+	best_depth = texelFetch(source_depth, pos, best_index).r;
+	best_normal_roughness = texelFetch(source_normal_roughness, pos, best_index);
+#ifdef VOXEL_GI_RESOLVE
+	best_voxel_gi = texelFetch(source_voxel_gi, pos, best_index).rg;
+#endif
+
+#endif
+
+	imageStore(dest_depth, pos, vec4(best_depth));
+	imageStore(dest_normal_roughness, pos, vec4(best_normal_roughness));
+#ifdef VOXEL_GI_RESOLVE
+	imageStore(dest_voxel_gi, pos, uvec4(best_voxel_gi, 0, 0));
+#endif
+
+#endif
+}

servers/rendering/renderer_rd/shaders/effects/roughness_limiter.glsl

@@ -0,0 +1,70 @@
+#[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_normal;
+layout(r8, set = 1, binding = 0) uniform restrict writeonly image2D dest_roughness;
+
+layout(push_constant, std430) uniform Params {
+	ivec2 screen_size;
+	float curve;
+	uint pad;
+}
+params;
+
+#define HALF_PI 1.5707963267948966
+
+void main() {
+	// Pixel being shaded
+	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
+	if (any(greaterThan(pos, params.screen_size))) { //too large, do nothing
+		return;
+	}
+
+	vec3 normal_accum = vec3(0.0);
+	float accum = 0.0;
+	for (int i = 0; i <= 1; i++) {
+		for (int j = 0; j <= 1; j++) {
+			normal_accum += normalize(texelFetch(source_normal, pos + ivec2(i, j), 0).xyz * 2.0 - 1.0);
+			accum += 1.0;
+		}
+	}
+
+	normal_accum /= accum;
+
+	float r = length(normal_accum);
+
+	float limit;
+
+	if (r < 1.0) {
+		float threshold = 0.4;
+
+		/*
+		//Formula from Filament, does not make sense to me.
+
+		float r2 = r * r;
+		float kappa = (3.0f * r - r * r2) / (1.0f - r2);
+		float variance = 0.25f / kappa;
+		limit = sqrt(min(2.0f * variance, threshold * threshold));
+		*/
+		/*
+		//Formula based on probability distribution graph
+
+		float width = acos(max(0.0,r)); // convert to angle (width)
+		float roughness = pow(width,1.7)*0.854492; //approximate (crappy) formula to convert to roughness
+		limit = min(sqrt(roughness), threshold); //convert to perceptual roughness and apply threshold
+		*/
+
+		limit = min(sqrt(pow(acos(max(0.0, r)) / HALF_PI, params.curve)), threshold); //convert to perceptual roughness and apply threshold
+
+		//limit = 0.5;
+	} else {
+		limit = 0.0;
+	}
+
+	imageStore(dest_roughness, pos, vec4(limit));
+}

servers/rendering/renderer_rd/shaders/effects/screen_space_reflection.glsl

@@ -0,0 +1,297 @@
+#[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 = normalize(normal_roughness.xyz * 2.0 - 1.0);
+	float roughness = normal_roughness.w;
+	if (roughness > 0.5) {
+		roughness = 1.0 - roughness;
+	}
+	roughness /= (127.0 / 255.0);
+
+	// 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 = params.orthogonal ? vec3(0.0, 0.0, -1.0) : 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 final_pos = pos;
+
+		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;
+		}
+
+		// Fade In / Fade Out
+		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;
+
+		// Ensure that precision errors do not introduce any fade. Even if it is just slightly below 1.0,
+		// strong specular light can leak through the reflection.
+		if (fade > 0.999) {
+			fade = 1.0;
+		}
+
+		// 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 - roughness);
+
+		// Schlick term.
+		float metallic = texelFetch(source_metallic, ssC << 1, 0).w;
+
+		// F0 is the reflectance of normally incident light (perpendicular to the surface).
+		// Dielectric materials have a widely accepted default value of 0.04. We assume that metals reflect all light, so their F0 is 1.0.
+		float f0 = mix(0.04, 1.0, metallic);
+		float m = clamp(1.0 - dot(normal, -view_dir), 0.0, 1.0);
+		float m2 = m * m;
+		m = m2 * m2 * m; // pow(m,5)
+		float fresnel_term = f0 + (1.0 - f0) * m; // Fresnel Schlick term.
+
+		// The alpha value of final_color controls the blending with specular light in specular_merge.glsl.
+		// Note that the Fresnel term is multiplied with the RGB color instead of being a part of the alpha value.
+		// There is a key difference:
+		// - multiplying a term with RGB darkens the SSR light without introducing/taking away specular light.
+		// - combining a term into the Alpha value introduces specular light at the expense of the SSR light.
+		vec4 final_color = vec4(imageLoad(source_diffuse, ivec2(final_pos - 0.5)).rgb * fresnel_term, fade * margin_blend * roughness_fade);
+
+		imageStore(ssr_image, ssC, 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
+
+	} else {
+#ifdef MODE_ROUGH
+		imageStore(blur_radius_image, ssC, vec4(0.0));
+#endif
+		imageStore(ssr_image, ssC, vec4(0.0));
+	}
+}

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
+}

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);
+		}
+	}
+}

servers/rendering/renderer_rd/shaders/effects/screen_space_reflection_scale.glsl

@@ -0,0 +1,112 @@
+#[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 artifacts 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 += normalize(nr.xyz * 2.0 - 1.0);
+			float roughness = normal.w;
+			if (roughness > 0.5) {
+				roughness = 1.0 - roughness;
+			}
+			roughness /= (127.0 / 255.0);
+			normal.w += roughness;
+
+			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;
+		normal.w = normal.w * (127.0 / 255.0);
+	} 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)) / 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);
+}

servers/rendering/renderer_rd/shaders/effects/shadow_frustum.glsl

@@ -0,0 +1,41 @@
+/* clang-format off */
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+/* clang-format on */
+
+layout(push_constant, std430) uniform Info {
+	mat4 mvp;
+	vec4 color;
+}
+info;
+
+layout(location = 0) in vec3 vertex_attrib;
+
+void main() {
+	vec4 vertex = info.mvp * vec4(vertex_attrib, 1.0);
+	vertex.xyz /= vertex.w;
+	gl_Position = vec4(vertex.xy, 0.0, 1.0);
+}
+
+/* clang-format off */
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(push_constant, std430) uniform Info {
+	mat4 mvp;
+	vec4 color;
+}
+info;
+
+layout(location = 0) out vec4 frag_color;
+
+void main() {
+	frag_color = info.color;
+}

servers/rendering/renderer_rd/shaders/effects/smaa_blending.glsl

@@ -0,0 +1,150 @@
+/**
+ * Copyright (C) 2013 Jorge Jimenez (jorge@iryoku.com)
+ * Copyright (C) 2013 Jose I. Echevarria (joseignacioechevarria@gmail.com)
+ * Copyright (C) 2013 Belen Masia (bmasia@unizar.es)
+ * Copyright (C) 2013 Fernando Navarro (fernandn@microsoft.com)
+ * Copyright (C) 2013 Diego Gutierrez (diegog@unizar.es)
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * 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. As clarification, there
+ * is no requirement that the copyright notice and permission be included in
+ * binary distributions 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.
+ */
+
+#[vertex]
+#version 450
+
+layout(location = 0) out vec2 tex_coord;
+layout(location = 1) out vec4 offset;
+
+layout(push_constant, std430) uniform Params {
+	vec2 inv_size;
+	vec2 pad;
+}
+params;
+
+void main() {
+	vec2 vertex_base;
+	if (gl_VertexIndex == 0) {
+		vertex_base = vec2(-1.0, -1.0);
+	} else if (gl_VertexIndex == 1) {
+		vertex_base = vec2(-1.0, 3.0);
+	} else {
+		vertex_base = vec2(3.0, -1.0);
+	}
+	gl_Position = vec4(vertex_base, 0.0, 1.0);
+	tex_coord = clamp(vertex_base, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+	offset = fma(params.inv_size.xyxy, vec4(1.0, 0.0, 0.0, 1.0), tex_coord.xyxy);
+}
+
+#[fragment]
+#version 450
+
+layout(location = 0) in vec2 tex_coord;
+layout(location = 1) in vec4 offset;
+layout(set = 0, binding = 0) uniform sampler2D color_tex;
+layout(set = 1, binding = 0) uniform sampler2D blend_tex;
+
+layout(location = 0) out vec4 out_color;
+
+#define FLAG_USE_8_BIT_DEBANDING (1 << 0)
+#define FLAG_USE_10_BIT_DEBANDING (1 << 1)
+
+layout(push_constant, std430) uniform Params {
+	vec2 inv_size;
+	uint flags;
+	float pad;
+}
+params;
+
+#define textureLinear(tex, uv) srgb_to_linear(textureLod(tex, uv, 0.0).rgb)
+
+vec3 linear_to_srgb(vec3 color) {
+	// If going to srgb, clamp from 0 to 1.
+	color = clamp(color, vec3(0.0), vec3(1.0));
+	const vec3 a = vec3(0.055f);
+	return mix((vec3(1.0f) + a) * pow(color.rgb, vec3(1.0f / 2.4f)) - a, 12.92f * color.rgb, lessThan(color.rgb, vec3(0.0031308f)));
+}
+
+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 SMAAMovc(bvec2 cond, inout vec2 variable, vec2 value) {
+	if (cond.x) {
+		variable.x = value.x;
+	}
+	if (cond.y) {
+		variable.y = value.y;
+	}
+}
+
+void SMAAMovc(bvec4 cond, inout vec4 variable, vec4 value) {
+	SMAAMovc(cond.xy, variable.xy, value.xy);
+	SMAAMovc(cond.zw, variable.zw, value.zw);
+}
+
+// From https://alex.vlachos.com/graphics/Alex_Vlachos_Advanced_VR_Rendering_GDC2015.pdf
+// and https://www.shadertoy.com/view/MslGR8 (5th one starting from the bottom)
+// NOTE: `frag_coord` is in pixels (i.e. not normalized UV).
+// This dithering must be applied after encoding changes (linear/nonlinear) have been applied
+// as the final step before quantization from floating point to integer values.
+vec3 screen_space_dither(vec2 frag_coord, float bit_alignment_diviser) {
+	// Iestyn's RGB dither (7 asm instructions) from Portal 2 X360, slightly modified for VR.
+	// Removed the time component to avoid passing time into this shader.
+	vec3 dither = vec3(dot(vec2(171.0, 231.0), frag_coord));
+	dither.rgb = fract(dither.rgb / vec3(103.0, 71.0, 97.0));
+
+	// Subtract 0.5 to avoid slightly brightening the whole viewport.
+	// Use a dither strength of 100% rather than the 37.5% suggested by the original source.
+	return (dither.rgb - 0.5) / bit_alignment_diviser;
+}
+
+void main() {
+	vec4 a;
+	a.x = texture(blend_tex, offset.xy).a;
+	a.y = texture(blend_tex, offset.zw).g;
+	a.wz = texture(blend_tex, tex_coord).xz;
+
+	if (dot(a, vec4(1.0, 1.0, 1.0, 1.0)) < 1e-5) {
+		out_color = textureLod(color_tex, tex_coord, 0.0);
+	} else {
+		bool h = max(a.x, a.z) > max(a.y, a.w);
+
+		vec4 blending_offset = vec4(0.0, a.y, 0.0, a.w);
+		vec2 blending_weight = a.yw;
+
+		SMAAMovc(bvec4(h, h, h, h), blending_offset, vec4(a.x, 0.0, a.z, 0.0));
+		SMAAMovc(bvec2(h, h), blending_weight, a.xz);
+		blending_weight /= dot(blending_weight, vec2(1.0, 1.0));
+
+		vec4 blending_coord = fma(blending_offset, vec4(params.inv_size.xy, -params.inv_size.xy), tex_coord.xyxy);
+
+		out_color.rgb = blending_weight.x * textureLinear(color_tex, blending_coord.xy);
+		out_color.rgb += blending_weight.y * textureLinear(color_tex, blending_coord.zw);
+		out_color.rgb = linear_to_srgb(out_color.rgb);
+		out_color.a = texture(color_tex, tex_coord).a;
+	}
+	if (bool(params.flags & FLAG_USE_8_BIT_DEBANDING)) {
+		// Divide by 255 to align to 8-bit quantization.
+		out_color.rgb += screen_space_dither(gl_FragCoord.xy, 255.0);
+	} else if (bool(params.flags & FLAG_USE_10_BIT_DEBANDING)) {
+		// Divide by 1023 to align to 10-bit quantization.
+		out_color.rgb += screen_space_dither(gl_FragCoord.xy, 1023.0);
+	}
+}

servers/rendering/renderer_rd/shaders/effects/smaa_edge_detection.glsl

@@ -0,0 +1,120 @@
+/**
+ * Copyright (C) 2013 Jorge Jimenez (jorge@iryoku.com)
+ * Copyright (C) 2013 Jose I. Echevarria (joseignacioechevarria@gmail.com)
+ * Copyright (C) 2013 Belen Masia (bmasia@unizar.es)
+ * Copyright (C) 2013 Fernando Navarro (fernandn@microsoft.com)
+ * Copyright (C) 2013 Diego Gutierrez (diegog@unizar.es)
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * 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. As clarification, there
+ * is no requirement that the copyright notice and permission be included in
+ * binary distributions 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.
+ */
+
+#[vertex]
+#version 450
+
+layout(location = 0) out vec2 tex_coord;
+layout(location = 1) out vec4 offset[3];
+
+layout(push_constant, std430) uniform Params {
+	vec2 inv_size;
+	float threshold;
+	float reserved;
+}
+params;
+
+void main() {
+	vec2 vertex_base;
+	if (gl_VertexIndex == 0) {
+		vertex_base = vec2(-1.0, -1.0);
+	} else if (gl_VertexIndex == 1) {
+		vertex_base = vec2(-1.0, 3.0);
+	} else {
+		vertex_base = vec2(3.0, -1.0);
+	}
+	gl_Position = vec4(vertex_base, 0.0, 1.0);
+	tex_coord = clamp(vertex_base, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+
+	offset[0] = fma(params.inv_size.xyxy, vec4(-1.0, 0.0, 0.0, -1.0), tex_coord.xyxy);
+	offset[1] = fma(params.inv_size.xyxy, vec4(1.0, 0.0, 0.0, 1.0), tex_coord.xyxy);
+	offset[2] = fma(params.inv_size.xyxy, vec4(-2.0, 0.0, 0.0, -2.0), tex_coord.xyxy);
+}
+
+#[fragment]
+#version 450
+
+layout(location = 0) in vec2 tex_coord;
+layout(location = 1) in vec4 offset[3];
+
+layout(set = 0, binding = 0) uniform sampler2D color_tex;
+
+layout(location = 0) out vec2 edges;
+
+layout(push_constant, std430) uniform Params {
+	vec2 inv_size;
+	float threshold;
+	float reserved;
+}
+params;
+
+#define SMAA_LOCAL_CONTRAST_ADAPTATION_FACTOR 2.0
+
+void main() {
+	vec2 threshold = vec2(params.threshold);
+
+	vec4 delta;
+	vec3 C = texture(color_tex, tex_coord).rgb;
+
+	vec3 Cleft = texture(color_tex, offset[0].xy).rgb;
+	vec3 t = abs(C - Cleft);
+	delta.x = max(max(t.r, t.g), t.b);
+
+	vec3 Ctop = texture(color_tex, offset[0].zw).rgb;
+	t = abs(C - Ctop);
+	delta.y = max(max(t.r, t.g), t.b);
+
+	edges = step(threshold, delta.xy);
+
+	if (dot(edges, vec2(1.0, 1.0)) == 0.0) {
+		discard;
+	}
+
+	vec3 Cright = texture(color_tex, offset[1].xy).rgb;
+	t = abs(C - Cright);
+	delta.z = max(max(t.r, t.g), t.b);
+
+	vec3 Cbottom = texture(color_tex, offset[1].zw).rgb;
+	t = abs(C - Cbottom);
+	delta.w = max(max(t.r, t.g), t.b);
+
+	vec2 max_delta = max(delta.xy, delta.zw);
+
+	vec3 Cleftleft = texture(color_tex, offset[2].xy).rgb;
+	t = abs(Cleft - Cleftleft);
+	delta.z = max(max(t.r, t.g), t.b);
+
+	vec3 Ctoptop = texture(color_tex, offset[2].zw).rgb;
+	t = abs(Ctop - Ctoptop);
+	delta.w = max(max(t.r, t.g), t.b);
+
+	max_delta = max(max_delta.xy, delta.zw);
+	float final_delta = max(max_delta.x, max_delta.y);
+
+	edges.xy *= step(final_delta, SMAA_LOCAL_CONTRAST_ADAPTATION_FACTOR * delta.xy);
+}

servers/rendering/renderer_rd/shaders/effects/smaa_weight_calculation.glsl

@@ -0,0 +1,376 @@
+/**
+ * Copyright (C) 2013 Jorge Jimenez (jorge@iryoku.com)
+ * Copyright (C) 2013 Jose I. Echevarria (joseignacioechevarria@gmail.com)
+ * Copyright (C) 2013 Belen Masia (bmasia@unizar.es)
+ * Copyright (C) 2013 Fernando Navarro (fernandn@microsoft.com)
+ * Copyright (C) 2013 Diego Gutierrez (diegog@unizar.es)
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * 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. As clarification, there
+ * is no requirement that the copyright notice and permission be included in
+ * binary distributions 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.
+ */
+
+#[vertex]
+#version 450
+
+layout(location = 0) out vec2 tex_coord;
+layout(location = 1) out vec2 pix_coord;
+layout(location = 2) out vec4 offset[3];
+
+layout(push_constant, std430) uniform Params {
+	vec2 inv_size;
+	ivec2 size;
+
+	vec4 subsample_indices;
+}
+params;
+
+#define SMAA_MAX_SEARCH_STEPS 32
+
+void main() {
+	vec2 vertex_base;
+	if (gl_VertexIndex == 0) {
+		vertex_base = vec2(-1.0, -1.0);
+	} else if (gl_VertexIndex == 1) {
+		vertex_base = vec2(-1.0, 3.0);
+	} else {
+		vertex_base = vec2(3.0, -1.0);
+	}
+	gl_Position = vec4(vertex_base, 0.0, 1.0);
+	tex_coord = clamp(vertex_base, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+	pix_coord = tex_coord * params.size.xy;
+
+	offset[0] = fma(params.inv_size.xyxy, vec4(-0.25, -0.125, 1.25, -0.125), tex_coord.xyxy);
+	offset[1] = fma(params.inv_size.xyxy, vec4(-0.125, -0.25, -0.125, 1.25), tex_coord.xyxy);
+	offset[2] = fma(params.inv_size.xxyy,
+			vec4(-2.0, 2.0, -2.0, 2.0) * SMAA_MAX_SEARCH_STEPS,
+			vec4(offset[0].xz, offset[1].yw));
+}
+
+#[fragment]
+#version 450
+
+layout(location = 0) in vec2 tex_coord;
+layout(location = 1) in vec2 pix_coord;
+layout(location = 2) in vec4 offset[3];
+layout(set = 0, binding = 0) uniform sampler2D edges_tex;
+layout(set = 1, binding = 0) uniform sampler2D area_tex;
+layout(set = 1, binding = 1) uniform sampler2D search_tex;
+layout(location = 0) out vec4 weights;
+
+layout(push_constant, std430) uniform Params {
+	vec2 inv_size;
+	ivec2 size;
+
+	vec4 subsample_indices;
+}
+params;
+
+#define SMAA_MAX_SEARCH_STEPS 32
+#define SMAA_MAX_SEARCH_STEPS_DIAG 16
+#define SMAA_CORNER_ROUNDING 25
+
+#ifndef SMAA_AREATEX_SELECT
+#define SMAA_AREATEX_SELECT(sample) sample.rg
+#endif
+
+#ifndef SMAA_SEARCHTEX_SELECT
+#define SMAA_SEARCHTEX_SELECT(sample) sample.r
+#endif
+
+#define SMAA_AREATEX_MAX_DISTANCE 16
+#define SMAA_AREATEX_MAX_DISTANCE_DIAG 20
+#define SMAA_AREATEX_PIXEL_SIZE (1.0 / vec2(160.0, 560.0))
+#define SMAA_AREATEX_SUBTEX_SIZE (1.0 / 7.0)
+#define SMAA_SEARCHTEX_SIZE vec2(66.0, 33.0)
+#define SMAA_SEARCHTEX_PACKED_SIZE vec2(64.0, 16.0)
+#define SMAA_CORNER_ROUNDING_NORM (float(SMAA_CORNER_ROUNDING) / 100.0)
+
+void SMAAMovc(bvec2 cond, inout vec2 variable, vec2 value) {
+	if (cond.x) {
+		variable.x = value.x;
+	}
+	if (cond.y) {
+		variable.y = value.y;
+	}
+}
+
+vec2 SMAADecodeDiagBilinearAccess(vec2 e) {
+	e.r = e.r * abs(5.0 * e.r - 5.0 * 0.75);
+	return round(e);
+}
+
+vec4 SMAADecodeDiagBilinearAccess(vec4 e) {
+	e.rb = e.rb * abs(5.0 * e.rb - 5.0 * 0.75);
+	return round(e);
+}
+
+vec2 SMAASearchDiag1(vec2 tex_coord, vec2 dir, out vec2 e) {
+	vec4 coord = vec4(tex_coord, -1.0, 1.0);
+	vec3 t = vec3(params.inv_size.xy, 1.0);
+	while (coord.z < float(SMAA_MAX_SEARCH_STEPS_DIAG - 1) &&
+			coord.w > 0.9) {
+		coord.xyz = fma(t, vec3(dir, 1.0), coord.xyz);
+		e = textureLod(edges_tex, coord.xy, 0.0).rg;
+		coord.w = dot(e, vec2(0.5, 0.5));
+	}
+	return coord.zw;
+}
+
+vec2 SMAASearchDiag2(vec2 tex_coord, vec2 dir, out vec2 e) {
+	vec4 coord = vec4(tex_coord, -1.0, 1.0);
+	coord.x += 0.25 * params.inv_size.x;
+	vec3 t = vec3(params.inv_size.xy, 1.0);
+	while (coord.z < float(SMAA_MAX_SEARCH_STEPS_DIAG - 1) &&
+			coord.w > 0.9) {
+		coord.xyz = fma(t, vec3(dir, 1.0), coord.xyz);
+
+		e = textureLod(edges_tex, coord.xy, 0.0).rg;
+		e = SMAADecodeDiagBilinearAccess(e);
+
+		coord.w = dot(e, vec2(0.5, 0.5));
+	}
+	return coord.zw;
+}
+
+vec2 SMAAAreaDiag(vec2 dist, vec2 e, float offset) {
+	vec2 coord = fma(vec2(SMAA_AREATEX_MAX_DISTANCE_DIAG, SMAA_AREATEX_MAX_DISTANCE_DIAG), e, dist);
+
+	coord = fma(SMAA_AREATEX_PIXEL_SIZE, coord, 0.5 * SMAA_AREATEX_PIXEL_SIZE);
+
+	coord.x += 0.5;
+
+	coord.y += SMAA_AREATEX_SUBTEX_SIZE * offset;
+
+	return SMAA_AREATEX_SELECT(textureLod(area_tex, coord, 0.0));
+}
+
+vec2 SMAACalculateDiagWeights(vec2 tex_coord, vec2 e, vec4 subsample_indices) {
+	vec2 weights = vec2(0.0, 0.0);
+
+	vec4 d;
+	vec2 end;
+	if (e.r > 0.0) {
+		d.xz = SMAASearchDiag1(tex_coord, vec2(-1.0, 1.0), end);
+		d.x += float(end.y > 0.9);
+	} else {
+		d.xz = vec2(0.0, 0.0);
+	}
+	d.yw = SMAASearchDiag1(tex_coord, vec2(1.0, -1.0), end);
+
+	if (d.x + d.y > 2.0) {
+		vec4 coords = fma(vec4(-d.x + 0.25, d.x, d.y, -d.y - 0.25), params.inv_size.xyxy, tex_coord.xyxy);
+		vec4 c;
+		c.xy = textureLodOffset(edges_tex, coords.xy, 0.0, ivec2(-1, 0)).rg;
+		c.zw = textureLodOffset(edges_tex, coords.zw, 0.0, ivec2(1, 0)).rg;
+		c.yxwz = SMAADecodeDiagBilinearAccess(c.xyzw);
+
+		vec2 cc = fma(vec2(2.0, 2.0), c.xz, c.yw);
+
+		SMAAMovc(bvec2(step(0.9, d.zw)), cc, vec2(0.0, 0.0));
+
+		weights += SMAAAreaDiag(d.xy, cc, subsample_indices.z);
+	}
+
+	d.xz = SMAASearchDiag2(tex_coord, vec2(-1.0, -1.0), end);
+	if (textureLodOffset(edges_tex, tex_coord, 0.0, ivec2(1, 0)).r > 0.0) {
+		d.yw = SMAASearchDiag2(tex_coord, vec2(1.0, 1.0), end);
+		d.y += float(end.y > 0.9);
+	} else {
+		d.yw = vec2(0.0, 0.0);
+	}
+
+	if (d.x + d.y > 2.0) {
+		vec4 coords = fma(vec4(-d.x, -d.x, d.y, d.y), params.inv_size.xyxy, tex_coord.xyxy);
+		vec4 c;
+		c.x = textureLodOffset(edges_tex, coords.xy, 0.0, ivec2(-1, 0)).g;
+		c.y = textureLodOffset(edges_tex, coords.xy, 0.0, ivec2(0, -1)).r;
+		c.zw = textureLodOffset(edges_tex, coords.zw, 0.0, ivec2(1, 0)).gr;
+		vec2 cc = fma(vec2(2.0, 2.0), c.xz, c.yw);
+
+		SMAAMovc(bvec2(step(0.9, d.zw)), cc, vec2(0.0, 0.0));
+
+		weights += SMAAAreaDiag(d.xy, cc, subsample_indices.w).gr;
+	}
+
+	return weights;
+}
+
+float SMAASearchLength(vec2 e, float offset) {
+	vec2 scale = SMAA_SEARCHTEX_SIZE * vec2(0.5, -1.0);
+	vec2 bias = SMAA_SEARCHTEX_SIZE * vec2(offset, 1.0);
+
+	scale += vec2(-1.0, 1.0);
+	bias += vec2(0.5, -0.5);
+
+	scale *= 1.0 / SMAA_SEARCHTEX_PACKED_SIZE;
+	bias *= 1.0 / SMAA_SEARCHTEX_PACKED_SIZE;
+
+	return SMAA_SEARCHTEX_SELECT(textureLod(search_tex, fma(scale, e, bias), 0.0));
+}
+
+float SMAASearchXLeft(vec2 tex_coord, float end) {
+	vec2 e = vec2(0.0, 1.0);
+	while (tex_coord.x > end &&
+			e.g > 0.8281 &&
+			e.r == 0.0) {
+		e = textureLod(edges_tex, tex_coord, 0.0).rg;
+		tex_coord = fma(-vec2(2.0, 0.0), params.inv_size.xy, tex_coord);
+	}
+
+	float offset = fma(-(255.0 / 127.0), SMAASearchLength(e, 0.0), 3.25);
+	return fma(params.inv_size.x, offset, tex_coord.x);
+}
+
+float SMAASearchXRight(vec2 tex_coord, float end) {
+	vec2 e = vec2(0.0, 1.0);
+	while (tex_coord.x < end &&
+			e.g > 0.8281 &&
+			e.r == 0.0) {
+		e = textureLod(edges_tex, tex_coord, 0.0).rg;
+		tex_coord = fma(vec2(2.0, 0.0), params.inv_size.xy, tex_coord);
+	}
+
+	float offset = fma(-(255.0 / 127.0), SMAASearchLength(e, 0.5), 3.25);
+	return fma(-params.inv_size.x, offset, tex_coord.x);
+}
+
+float SMAASearchYUp(vec2 tex_coord, float end) {
+	vec2 e = vec2(1.0, 0.0);
+	while (tex_coord.y > end &&
+			e.r > 0.8281 &&
+			e.g == 0.0) {
+		e = textureLod(edges_tex, tex_coord, 0.0).rg;
+		tex_coord = fma(-vec2(0.0, 2.0), params.inv_size.xy, tex_coord);
+	}
+
+	float offset = fma(-(255.0 / 127.0), SMAASearchLength(e.gr, 0.0), 3.25);
+	return fma(params.inv_size.y, offset, tex_coord.y);
+}
+
+float SMAASearchYDown(vec2 tex_coord, float end) {
+	vec2 e = vec2(1.0, 0.0);
+	while (tex_coord.y < end &&
+			e.r > 0.8281 &&
+			e.g == 0.0) {
+		e = textureLod(edges_tex, tex_coord, 0.0).rg;
+		tex_coord = fma(vec2(0.0, 2.0), params.inv_size.xy, tex_coord);
+	}
+
+	float offset = fma(-(255.0 / 127.0), SMAASearchLength(e.gr, 0.5), 3.25);
+	return fma(-params.inv_size.y, offset, tex_coord.y);
+}
+
+vec2 SMAAArea(vec2 dist, float e1, float e2, float offset) {
+	vec2 tex_coord = fma(vec2(SMAA_AREATEX_MAX_DISTANCE, SMAA_AREATEX_MAX_DISTANCE), round(4.0 * vec2(e1, e2)), dist);
+
+	tex_coord = fma(SMAA_AREATEX_PIXEL_SIZE, tex_coord, 0.5 * SMAA_AREATEX_PIXEL_SIZE);
+
+	tex_coord.y = fma(SMAA_AREATEX_SUBTEX_SIZE, offset, tex_coord.y);
+
+	return SMAA_AREATEX_SELECT(textureLod(area_tex, tex_coord, 0.0));
+}
+
+void SMAADetectHorizontalCornerPattern(inout vec2 weights, vec4 coord, vec2 d) {
+	vec2 left_right = step(d.xy, d.yx);
+	vec2 rounding = (1.0 - SMAA_CORNER_ROUNDING_NORM) * left_right;
+
+	rounding /= left_right.x + left_right.y;
+
+	vec2 factor = vec2(1.0, 1.0);
+	factor.x -= rounding.x * textureLodOffset(edges_tex, coord.xy, 0.0, ivec2(0, 1)).r;
+	factor.x -= rounding.y * textureLodOffset(edges_tex, coord.zw, 0.0, ivec2(1, 1)).r;
+	factor.y -= rounding.x * textureLodOffset(edges_tex, coord.xy, 0.0, ivec2(0, -2)).r;
+	factor.y -= rounding.y * textureLodOffset(edges_tex, coord.zw, 0.0, ivec2(1, -2)).r;
+
+	weights *= clamp(factor, 0.0, 1.0);
+}
+
+void SMAADetectVerticalCornerPattern(inout vec2 weights, vec4 coord, vec2 d) {
+	vec2 left_right = step(d.xy, d.yx);
+	vec2 rounding = (1.0 - SMAA_CORNER_ROUNDING_NORM) * left_right;
+
+	rounding /= left_right.x + left_right.y;
+
+	vec2 factor = vec2(1.0, 1.0);
+	factor.x -= rounding.x * textureLodOffset(edges_tex, coord.xy, 0.0, ivec2(1, 0)).g;
+	factor.x -= rounding.y * textureLodOffset(edges_tex, coord.zw, 0.0, ivec2(1, 1)).g;
+	factor.y -= rounding.x * textureLodOffset(edges_tex, coord.xy, 0.0, ivec2(-2, 0)).g;
+	factor.y -= rounding.y * textureLodOffset(edges_tex, coord.zw, 0.0, ivec2(-2, 1)).g;
+
+	weights *= clamp(factor, 0.0, 1.0);
+}
+
+void main() {
+	weights = vec4(0.0, 0.0, 0.0, 0.0);
+	vec2 e = textureLod(edges_tex, tex_coord, 0.0).rg;
+
+	if (e.g > 0.0) { // Edge at north.
+		weights.rg = SMAACalculateDiagWeights(tex_coord, e, params.subsample_indices);
+		if (weights.r == -weights.g) {
+			vec2 d;
+			vec3 coords;
+			coords.x = SMAASearchXLeft(offset[0].xy, offset[2].x);
+			coords.y = offset[1].y;
+			d.x = coords.x;
+
+			float e1 = textureLod(edges_tex, coords.xy, 0.0).r;
+
+			coords.z = SMAASearchXRight(offset[0].zw, offset[2].y);
+			d.y = coords.z;
+
+			d = abs(round(fma(params.size.xx, d, -pix_coord.xx)));
+
+			vec2 sqrt_d = sqrt(d);
+
+			float e2 = textureLodOffset(edges_tex, coords.zy, 0.0, ivec2(1, 0)).r;
+
+			weights.rg = SMAAArea(sqrt_d, e1, e2, params.subsample_indices.y);
+
+			coords.y = tex_coord.y;
+			SMAADetectHorizontalCornerPattern(weights.rg, coords.xyzy, d);
+		} else {
+			e.r = 0.0;
+		}
+	}
+
+	if (e.r > 0.0) { // Edge at west.
+		vec2 d;
+		vec3 coords;
+		coords.y = SMAASearchYUp(offset[1].xy, offset[2].z);
+		coords.x = offset[0].x;
+		d.x = coords.y;
+
+		float e1 = textureLod(edges_tex, coords.xy, 0.0).g;
+
+		coords.z = SMAASearchYDown(offset[1].zw, offset[2].w);
+		d.y = coords.z;
+
+		d = abs(round(fma(params.size.yy, d, -pix_coord.yy)));
+
+		vec2 sqrt_d = sqrt(d);
+
+		float e2 = textureLodOffset(edges_tex, coords.xz, 0.0, ivec2(0, 1)).g;
+
+		weights.ba = SMAAArea(sqrt_d, e1, e2, params.subsample_indices.x);
+
+		coords.x = tex_coord.x;
+		SMAADetectVerticalCornerPattern(weights.ba, coords.xyxz, d);
+	}
+}

servers/rendering/renderer_rd/shaders/effects/sort.glsl

@@ -0,0 +1,205 @@
+#[compute]
+
+#version 450
+
+#VERSION_DEFINES
+
+// Original version here:
+// https://github.com/GPUOpen-LibrariesAndSDKs/GPUParticles11/blob/master/gpuparticles11/src/Shaders
+
+//
+// Copyright (c) 2016 Advanced Micro Devices, Inc. All rights reserved.
+//
+// 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.
+//
+
+#define SORT_SIZE 512
+#define NUM_THREADS (SORT_SIZE / 2)
+#define INVERSION (16 * 2 + 8 * 3)
+#define ITERATIONS 1
+
+layout(local_size_x = NUM_THREADS, local_size_y = 1, local_size_z = 1) in;
+
+#ifndef MODE_SORT_STEP
+
+shared vec2 g_LDS[SORT_SIZE];
+
+#endif
+
+layout(set = 1, binding = 0, std430) restrict buffer SortBuffer {
+	vec2 data[];
+}
+sort_buffer;
+
+layout(push_constant, std430) uniform Params {
+	uint total_elements;
+	uint pad[3];
+	ivec4 job_params;
+}
+params;
+
+void main() {
+#ifdef MODE_SORT_BLOCK
+
+	uvec3 Gid = gl_WorkGroupID;
+	uvec3 DTid = gl_GlobalInvocationID;
+	uvec3 GTid = gl_LocalInvocationID;
+	uint GI = gl_LocalInvocationIndex;
+
+	int GlobalBaseIndex = int((Gid.x * SORT_SIZE) + GTid.x);
+	int LocalBaseIndex = int(GI);
+	int numElementsInThreadGroup = int(min(SORT_SIZE, params.total_elements - (Gid.x * SORT_SIZE)));
+
+	// Load shared data
+
+	int i;
+	for (i = 0; i < 2 * ITERATIONS; ++i) {
+		if (GI + i * NUM_THREADS < numElementsInThreadGroup) {
+			g_LDS[LocalBaseIndex + i * NUM_THREADS] = sort_buffer.data[GlobalBaseIndex + i * NUM_THREADS];
+		}
+	}
+
+	groupMemoryBarrier();
+	barrier();
+
+	// Bitonic sort
+	for (int nMergeSize = 2; nMergeSize <= SORT_SIZE; nMergeSize = nMergeSize * 2) {
+		for (int nMergeSubSize = nMergeSize >> 1; nMergeSubSize > 0; nMergeSubSize = nMergeSubSize >> 1) {
+			for (i = 0; i < ITERATIONS; ++i) {
+				int tmp_index = int(GI + NUM_THREADS * i);
+				int index_low = tmp_index & (nMergeSubSize - 1);
+				int index_high = 2 * (tmp_index - index_low);
+				int index = index_high + index_low;
+
+				int nSwapElem = nMergeSubSize == nMergeSize >> 1 ? index_high + (2 * nMergeSubSize - 1) - index_low : index_high + nMergeSubSize + index_low;
+				if (nSwapElem < numElementsInThreadGroup) {
+					vec2 a = g_LDS[index];
+					vec2 b = g_LDS[nSwapElem];
+
+					if (a.x > b.x) {
+						g_LDS[index] = b;
+						g_LDS[nSwapElem] = a;
+					}
+				}
+				groupMemoryBarrier();
+				barrier();
+			}
+		}
+	}
+
+	// Store shared data
+	for (i = 0; i < 2 * ITERATIONS; ++i) {
+		if (GI + i * NUM_THREADS < numElementsInThreadGroup) {
+			sort_buffer.data[GlobalBaseIndex + i * NUM_THREADS] = g_LDS[LocalBaseIndex + i * NUM_THREADS];
+		}
+	}
+
+#endif
+
+#ifdef MODE_SORT_STEP
+
+	uvec3 Gid = gl_WorkGroupID;
+	uvec3 GTid = gl_LocalInvocationID;
+
+	ivec4 tgp;
+
+	tgp.x = int(Gid.x) * 256;
+	tgp.y = 0;
+	tgp.z = int(params.total_elements);
+	tgp.w = min(512, max(0, tgp.z - int(Gid.x) * 512));
+
+	uint localID = int(tgp.x) + GTid.x; // calculate threadID within this sortable-array
+
+	uint index_low = localID & (params.job_params.x - 1);
+	uint index_high = 2 * (localID - index_low);
+
+	uint index = tgp.y + index_high + index_low;
+	uint nSwapElem = tgp.y + index_high + params.job_params.y + params.job_params.z * index_low;
+
+	if (nSwapElem < tgp.y + tgp.z) {
+		vec2 a = sort_buffer.data[index];
+		vec2 b = sort_buffer.data[nSwapElem];
+
+		if (a.x > b.x) {
+			sort_buffer.data[index] = b;
+			sort_buffer.data[nSwapElem] = a;
+		}
+	}
+
+#endif
+
+#ifdef MODE_SORT_INNER
+
+	uvec3 Gid = gl_WorkGroupID;
+	uvec3 DTid = gl_GlobalInvocationID;
+	uvec3 GTid = gl_LocalInvocationID;
+	uint GI = gl_LocalInvocationIndex;
+
+	ivec4 tgp;
+
+	tgp.x = int(Gid.x * 256);
+	tgp.y = 0;
+	tgp.z = int(params.total_elements.x);
+	tgp.w = int(min(512, max(0, params.total_elements - Gid.x * 512)));
+
+	int GlobalBaseIndex = int(tgp.y + tgp.x * 2 + GTid.x);
+	int LocalBaseIndex = int(GI);
+	int i;
+
+	// Load shared data
+	for (i = 0; i < 2; ++i) {
+		if (GI + i * NUM_THREADS < tgp.w) {
+			g_LDS[LocalBaseIndex + i * NUM_THREADS] = sort_buffer.data[GlobalBaseIndex + i * NUM_THREADS];
+		}
+	}
+
+	groupMemoryBarrier();
+	barrier();
+
+	// sort threadgroup shared memory
+	for (int nMergeSubSize = SORT_SIZE >> 1; nMergeSubSize > 0; nMergeSubSize = nMergeSubSize >> 1) {
+		int tmp_index = int(GI);
+		int index_low = tmp_index & (nMergeSubSize - 1);
+		int index_high = 2 * (tmp_index - index_low);
+		int index = index_high + index_low;
+
+		int nSwapElem = index_high + nMergeSubSize + index_low;
+
+		if (nSwapElem < tgp.w) {
+			vec2 a = g_LDS[index];
+			vec2 b = g_LDS[nSwapElem];
+
+			if (a.x > b.x) {
+				g_LDS[index] = b;
+				g_LDS[nSwapElem] = a;
+			}
+		}
+		groupMemoryBarrier();
+		barrier();
+	}
+
+	// Store shared data
+	for (i = 0; i < 2; ++i) {
+		if (GI + i * NUM_THREADS < tgp.w) {
+			sort_buffer.data[GlobalBaseIndex + i * NUM_THREADS] = g_LDS[LocalBaseIndex + i * NUM_THREADS];
+		}
+	}
+
+#endif
+}

servers/rendering/renderer_rd/shaders/effects/specular_merge.glsl

@@ -0,0 +1,80 @@
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+#if defined(USE_MULTIVIEW)
+#extension GL_EXT_multiview : enable
+#define ViewIndex gl_ViewIndex
+#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[3] = vec2[](vec2(-1.0, -1.0), vec2(-1.0, 3.0), vec2(3.0, -1.0));
+	gl_Position = vec4(base_arr[gl_VertexIndex], 0.0, 1.0);
+	uv_interp.xy = clamp(gl_Position.xy, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+#ifdef USE_MULTIVIEW
+	uv_interp.z = ViewIndex;
+#endif
+}
+
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#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
+}

servers/rendering/renderer_rd/shaders/effects/ss_effects_downsample.glsl

@@ -0,0 +1,224 @@
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// 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)) / 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)) / 2.0;
+	}
+
+	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;
+
+	if (still_alive) {
+		// Use the previous average, not ideal, but still not bad.
+		float sample_00 = depth_buffer[depth_array_index][buffer_coord.x + 0][buffer_coord.y + 0];
+		imageStore(dest_image4, ivec3(p_output_coord.x, p_output_coord.y, depth_array_index), vec4(sample_00));
+	}
+#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
+// Half buffers means that we divide depth into two half res buffers (we only capture 1/4 of pixels).
+#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 //USE_HALF_SIZE
+#ifdef GENERATE_MIPS
+	prepare_depths_and_mips(samples, output_coord, gl_LocalInvocationID.xy);
+#else
+	prepare_depths(samples, gl_GlobalInvocationID.xy);
+#endif
+#endif //USE_HALF_BUFFERS
+}

servers/rendering/renderer_rd/shaders/effects/ssao.glsl

@@ -0,0 +1,481 @@
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// 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
+//
+// 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_ENABLE_AT_QUALITY_PRESET (2)
+#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 load_normal(ivec2 p_pos) {
+	vec3 encoded_normal = normalize(imageLoad(source_normal, p_pos).xyz * 2.0 - 1.0);
+	encoded_normal.z = -encoded_normal.z;
+	return encoded_normal;
+}
+
+vec3 load_normal(ivec2 p_pos, ivec2 p_offset) {
+	vec3 encoded_normal = normalize(imageLoad(source_normal, p_pos + p_offset).xyz * 2.0 - 1.0);
+	encoded_normal.z = -encoded_normal.z;
+	return 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 reduce = max(0, -hit_delta.z);
+		reduce = clamp(reduce * params.neg_inv_radius + 2.0, 0.0, 1.0);
+		weight = SSAO_HALOING_REDUCTION_AMOUNT * reduce + (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.
+	pix_center_pos *= 0.99;
+
+	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
+}

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));
+}

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
+}

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
+}

servers/rendering/renderer_rd/shaders/effects/ssil.glsl

@@ -0,0 +1,440 @@
+///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
+// 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 load_normal(ivec2 p_pos) {
+	vec3 encoded_normal = normalize(imageLoad(source_normal, p_pos).xyz * 2.0 - 1.0);
+	encoded_normal.z = -encoded_normal.z;
+	return encoded_normal;
+}
+
+vec3 load_normal(ivec2 p_pos, ivec2 p_offset) {
+	vec3 encoded_normal = normalize(imageLoad(source_normal, p_pos + p_offset).xyz * 2.0 - 1.0);
+	encoded_normal.z = -encoded_normal.z;
+	return 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 reduce = max(0, -hit_delta.z);
+		reduce = clamp(reduce * params.neg_inv_radius + 2.0, 0.0, 1.0);
+		weight = SSIL_HALOING_REDUCTION_AMOUNT * reduce + (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.
+	pix_center_pos *= 0.99;
+
+	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
+}

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);
+}

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
+}

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
+}

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)) / 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);
+}

servers/rendering/renderer_rd/shaders/effects/taa_resolve.glsl

@@ -0,0 +1,381 @@
+///////////////////////////////////////////////////////////////////////////////////
+// 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>
+
+#define GROUP_SIZE 8
+#define FLT_MIN 0.00000001
+#define FLT_MAX 32767.0
+#define RPC_9 0.11111111111
+#define RPC_16 0.0625
+
+layout(local_size_x = GROUP_SIZE, local_size_y = GROUP_SIZE, local_size_z = 1) in;
+
+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;
+}
+
+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();
+}
+
+/*------------------------------------------------------------------------------
+								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);
+}
+
+// This is "velocity disocclusion" as described by https://www.elopezr.com/temporal-aa-and-the-quest-for-the-holy-trail/.
+// We use texel space, so our scale and threshold differ.
+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 immediately.
+		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() {
+	populate_group_shared_memory(gl_WorkGroupID.xy, gl_LocalInvocationIndex);
+
+	// Out of bounds check
+	if (any(greaterThanEqual(vec2(gl_GlobalInvocationID.xy), params.resolution))) {
+		return;
+	}
+
+	const uvec2 pos_group = gl_LocalInvocationID.xy;
+	const uvec2 pos_group_top_left = gl_WorkGroupID.xy * kGroupSize - kBorderSize;
+	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));
+}

servers/rendering/renderer_rd/shaders/effects/tonemap.glsl

@@ -0,0 +1,930 @@
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+layout(location = 0) out vec2 uv_interp;
+
+void main() {
+	// old code, ARM driver bug on Mali-GXXx GPUs and Vulkan API 1.3.xxx
+	// https://github.com/godotengine/godot/pull/92817#issuecomment-2168625982
+	//vec2 base_arr[3] = vec2[](vec2(-1.0, -1.0), vec2(-1.0, 3.0), vec2(3.0, -1.0));
+	//gl_Position = vec4(base_arr[gl_VertexIndex], 0.0, 1.0);
+	//uv_interp = clamp(gl_Position.xy, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+
+	vec2 vertex_base;
+	if (gl_VertexIndex == 0) {
+		vertex_base = vec2(-1.0, -1.0);
+	} else if (gl_VertexIndex == 1) {
+		vertex_base = vec2(-1.0, 3.0);
+	} else {
+		vertex_base = vec2(3.0, -1.0);
+	}
+	gl_Position = vec4(vertex_base, 0.0, 1.0);
+	uv_interp = clamp(vertex_base, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+}
+
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef USE_MULTIVIEW
+#extension GL_EXT_multiview : enable
+#define ViewIndex gl_ViewIndex
+#endif //USE_MULTIVIEW
+
+layout(location = 0) in vec2 uv_interp;
+
+#ifdef SUBPASS
+layout(input_attachment_index = 0, set = 0, binding = 0) uniform subpassInput input_color;
+#elif defined(USE_MULTIVIEW)
+layout(set = 0, binding = 0) uniform sampler2DArray source_color;
+#else
+layout(set = 0, binding = 0) uniform sampler2D source_color;
+#endif
+
+layout(set = 1, binding = 0) uniform sampler2D source_auto_exposure;
+#ifdef USE_MULTIVIEW
+layout(set = 2, binding = 0) uniform sampler2DArray source_glow;
+#else
+layout(set = 2, binding = 0) uniform sampler2D source_glow;
+#endif
+layout(set = 2, binding = 1) uniform sampler2D glow_map;
+
+#ifdef USE_1D_LUT
+layout(set = 3, binding = 0) uniform sampler2D source_color_correction;
+#else
+layout(set = 3, binding = 0) uniform sampler3D source_color_correction;
+#endif
+
+#define FLAG_USE_BCS (1 << 0)
+#define FLAG_USE_GLOW (1 << 1)
+#define FLAG_USE_AUTO_EXPOSURE (1 << 2)
+#define FLAG_USE_COLOR_CORRECTION (1 << 3)
+#define FLAG_USE_FXAA (1 << 4)
+#define FLAG_USE_8_BIT_DEBANDING (1 << 5)
+#define FLAG_USE_10_BIT_DEBANDING (1 << 6)
+#define FLAG_CONVERT_TO_SRGB (1 << 7)
+
+layout(push_constant, std430) uniform Params {
+	vec3 bcs;
+	uint flags;
+
+	vec2 pixel_size;
+	uint tonemapper;
+	uint pad;
+
+	uvec2 glow_texture_size;
+	float glow_intensity;
+	float glow_map_strength;
+
+	uint glow_mode;
+	float glow_levels[7];
+
+	float exposure;
+	float white;
+	float auto_exposure_scale;
+	float luminance_multiplier;
+}
+params;
+
+layout(location = 0) out vec4 frag_color;
+
+#ifdef USE_GLOW_FILTER_BICUBIC
+// w0, w1, w2, and w3 are the four cubic B-spline basis functions
+float w0(float a) {
+	return (1.0f / 6.0f) * (a * (a * (-a + 3.0f) - 3.0f) + 1.0f);
+}
+
+float w1(float a) {
+	return (1.0f / 6.0f) * (a * a * (3.0f * a - 6.0f) + 4.0f);
+}
+
+float w2(float a) {
+	return (1.0f / 6.0f) * (a * (a * (-3.0f * a + 3.0f) + 3.0f) + 1.0f);
+}
+
+float w3(float a) {
+	return (1.0f / 6.0f) * (a * a * a);
+}
+
+// g0 and g1 are the two amplitude functions
+float g0(float a) {
+	return w0(a) + w1(a);
+}
+
+float g1(float a) {
+	return w2(a) + w3(a);
+}
+
+// h0 and h1 are the two offset functions
+float h0(float a) {
+	return -1.0f + w1(a) / (w0(a) + w1(a));
+}
+
+float h1(float a) {
+	return 1.0f + w3(a) / (w2(a) + w3(a));
+}
+
+#ifdef USE_MULTIVIEW
+vec4 texture2D_bicubic(sampler2DArray tex, vec2 uv, int p_lod) {
+	float lod = float(p_lod);
+	vec2 tex_size = vec2(params.glow_texture_size >> p_lod);
+	vec2 pixel_size = vec2(1.0f) / tex_size;
+
+	uv = uv * tex_size + vec2(0.5f);
+
+	vec2 iuv = floor(uv);
+	vec2 fuv = fract(uv);
+
+	float g0x = g0(fuv.x);
+	float g1x = g1(fuv.x);
+	float h0x = h0(fuv.x);
+	float h1x = h1(fuv.x);
+	float h0y = h0(fuv.y);
+	float h1y = h1(fuv.y);
+
+	vec3 p0 = vec3((vec2(iuv.x + h0x, iuv.y + h0y) - vec2(0.5f)) * pixel_size, ViewIndex);
+	vec3 p1 = vec3((vec2(iuv.x + h1x, iuv.y + h0y) - vec2(0.5f)) * pixel_size, ViewIndex);
+	vec3 p2 = vec3((vec2(iuv.x + h0x, iuv.y + h1y) - vec2(0.5f)) * pixel_size, ViewIndex);
+	vec3 p3 = vec3((vec2(iuv.x + h1x, iuv.y + h1y) - vec2(0.5f)) * pixel_size, ViewIndex);
+
+	return (g0(fuv.y) * (g0x * textureLod(tex, p0, lod) + g1x * textureLod(tex, p1, lod))) +
+			(g1(fuv.y) * (g0x * textureLod(tex, p2, lod) + g1x * textureLod(tex, p3, lod)));
+}
+
+#define GLOW_TEXTURE_SAMPLE(m_tex, m_uv, m_lod) texture2D_bicubic(m_tex, m_uv, m_lod)
+#else // USE_MULTIVIEW
+
+vec4 texture2D_bicubic(sampler2D tex, vec2 uv, int p_lod) {
+	float lod = float(p_lod);
+	vec2 tex_size = vec2(params.glow_texture_size >> p_lod);
+	vec2 pixel_size = vec2(1.0f) / tex_size;
+
+	uv = uv * tex_size + vec2(0.5f);
+
+	vec2 iuv = floor(uv);
+	vec2 fuv = fract(uv);
+
+	float g0x = g0(fuv.x);
+	float g1x = g1(fuv.x);
+	float h0x = h0(fuv.x);
+	float h1x = h1(fuv.x);
+	float h0y = h0(fuv.y);
+	float h1y = h1(fuv.y);
+
+	vec2 p0 = (vec2(iuv.x + h0x, iuv.y + h0y) - vec2(0.5f)) * pixel_size;
+	vec2 p1 = (vec2(iuv.x + h1x, iuv.y + h0y) - vec2(0.5f)) * pixel_size;
+	vec2 p2 = (vec2(iuv.x + h0x, iuv.y + h1y) - vec2(0.5f)) * pixel_size;
+	vec2 p3 = (vec2(iuv.x + h1x, iuv.y + h1y) - vec2(0.5f)) * pixel_size;
+
+	return (g0(fuv.y) * (g0x * textureLod(tex, p0, lod) + g1x * textureLod(tex, p1, lod))) +
+			(g1(fuv.y) * (g0x * textureLod(tex, p2, lod) + g1x * textureLod(tex, p3, lod)));
+}
+
+#define GLOW_TEXTURE_SAMPLE(m_tex, m_uv, m_lod) texture2D_bicubic(m_tex, m_uv, m_lod)
+#endif // !USE_MULTIVIEW
+
+#else // USE_GLOW_FILTER_BICUBIC
+
+#ifdef USE_MULTIVIEW
+#define GLOW_TEXTURE_SAMPLE(m_tex, m_uv, m_lod) textureLod(m_tex, vec3(m_uv, ViewIndex), float(m_lod))
+#else // USE_MULTIVIEW
+#define GLOW_TEXTURE_SAMPLE(m_tex, m_uv, m_lod) textureLod(m_tex, m_uv, float(m_lod))
+#endif // !USE_MULTIVIEW
+
+#endif // !USE_GLOW_FILTER_BICUBIC
+
+// Based on Reinhard's extended formula, see equation 4 in https://doi.org/cjbgrt
+vec3 tonemap_reinhard(vec3 color, float white) {
+	float white_squared = white * white;
+	vec3 white_squared_color = white_squared * color;
+	// Equivalent to color * (1 + color / white_squared) / (1 + color)
+	return (white_squared_color + color * color) / (white_squared_color + white_squared);
+}
+
+vec3 tonemap_filmic(vec3 color, float white) {
+	// exposure bias: input scale (color *= bias, white *= bias) to make the brightness consistent with other tonemappers
+	// also useful to scale the input to the range that the tonemapper is designed for (some require very high input values)
+	// has no effect on the curve's general shape or visual properties
+	const float exposure_bias = 2.0f;
+	const float A = 0.22f * exposure_bias * exposure_bias; // bias baked into constants for performance
+	const float B = 0.30f * exposure_bias;
+	const float C = 0.10f;
+	const float D = 0.20f;
+	const float E = 0.01f;
+	const float F = 0.30f;
+
+	vec3 color_tonemapped = ((color * (A * color + C * B) + D * E) / (color * (A * color + B) + D * F)) - E / F;
+	float white_tonemapped = ((white * (A * white + C * B) + D * E) / (white * (A * white + B) + D * F)) - E / F;
+
+	return color_tonemapped / white_tonemapped;
+}
+
+// Adapted from https://github.com/TheRealMJP/BakingLab/blob/master/BakingLab/ACES.hlsl
+// (MIT License).
+vec3 tonemap_aces(vec3 color, float white) {
+	const float exposure_bias = 1.8f;
+	const float A = 0.0245786f;
+	const float B = 0.000090537f;
+	const float C = 0.983729f;
+	const float D = 0.432951f;
+	const float E = 0.238081f;
+
+	// Exposure bias baked into transform to save shader instructions. Equivalent to `color *= exposure_bias`
+	const mat3 rgb_to_rrt = mat3(
+			vec3(0.59719f * exposure_bias, 0.35458f * exposure_bias, 0.04823f * exposure_bias),
+			vec3(0.07600f * exposure_bias, 0.90834f * exposure_bias, 0.01566f * exposure_bias),
+			vec3(0.02840f * exposure_bias, 0.13383f * exposure_bias, 0.83777f * exposure_bias));
+
+	const mat3 odt_to_rgb = mat3(
+			vec3(1.60475f, -0.53108f, -0.07367f),
+			vec3(-0.10208f, 1.10813f, -0.00605f),
+			vec3(-0.00327f, -0.07276f, 1.07602f));
+
+	color *= rgb_to_rrt;
+	vec3 color_tonemapped = (color * (color + A) - B) / (color * (C * color + D) + E);
+	color_tonemapped *= odt_to_rgb;
+
+	white *= exposure_bias;
+	float white_tonemapped = (white * (white + A) - B) / (white * (C * white + D) + E);
+
+	return color_tonemapped / white_tonemapped;
+}
+
+// Polynomial approximation of EaryChow's AgX sigmoid curve.
+// x must be within the range [0.0, 1.0]
+vec3 agx_contrast_approx(vec3 x) {
+	// Generated with Excel trendline
+	// Input data: Generated using python sigmoid with EaryChow's configuration and 57 steps
+	// Additional padding values were added to give correct intersections at 0.0 and 1.0
+	// 6th order, intercept of 0.0 to remove an operation and ensure intersection at 0.0
+	vec3 x2 = x * x;
+	vec3 x4 = x2 * x2;
+	return 0.021 * x + 4.0111 * x2 - 25.682 * x2 * x + 70.359 * x4 - 74.778 * x4 * x + 27.069 * x4 * x2;
+}
+
+// This is an approximation and simplification of EaryChow's AgX implementation that is used by Blender.
+// This code is based off of the script that generates the AgX_Base_sRGB.cube LUT that Blender uses.
+// Source: https://github.com/EaryChow/AgX_LUT_Gen/blob/main/AgXBasesRGB.py
+vec3 tonemap_agx(vec3 color) {
+	// Combined linear sRGB to linear Rec 2020 and Blender AgX inset matrices:
+	const mat3 srgb_to_rec2020_agx_inset_matrix = mat3(
+			0.54490813676363087053, 0.14044005884001287035, 0.088827411851915368603,
+			0.37377945959812267119, 0.75410959864013760045, 0.17887712465043811023,
+			0.081384976686407536266, 0.10543358536857773485, 0.73224999956948382528);
+
+	// Combined inverse AgX outset matrix and linear Rec 2020 to linear sRGB matrices.
+	const mat3 agx_outset_rec2020_to_srgb_matrix = mat3(
+			1.9645509602733325934, -0.29932243390911083839, -0.16436833806080403409,
+			-0.85585845117807513559, 1.3264510741502356555, -0.23822464068860595117,
+			-0.10886710826831608324, -0.027084020983874825605, 1.402665347143271889);
+
+	// LOG2_MIN      = -10.0
+	// LOG2_MAX      =  +6.5
+	// MIDDLE_GRAY   =  0.18
+	const float min_ev = -12.4739311883324; // log2(pow(2, LOG2_MIN) * MIDDLE_GRAY)
+	const float max_ev = 4.02606881166759; // log2(pow(2, LOG2_MAX) * MIDDLE_GRAY)
+
+	// Large negative values in one channel and large positive values in other
+	// channels can result in a colour that appears darker and more saturated than
+	// desired after passing it through the inset matrix. For this reason, it is
+	// best to prevent negative input values.
+	// This is done before the Rec. 2020 transform to allow the Rec. 2020
+	// transform to be combined with the AgX inset matrix. This results in a loss
+	// of color information that could be correctly interpreted within the
+	// Rec. 2020 color space as positive RGB values, but it is less common for Godot
+	// to provide this function with negative sRGB values and therefore not worth
+	// the performance cost of an additional matrix multiplication.
+	// A value of 2e-10 intentionally introduces insignificant error to prevent
+	// log2(0.0) after the inset matrix is applied; color will be >= 1e-10 after
+	// the matrix transform.
+	color = max(color, 2e-10);
+
+	// Do AGX in rec2020 to match Blender and then apply inset matrix.
+	color = srgb_to_rec2020_agx_inset_matrix * color;
+
+	// Log2 space encoding.
+	// Must be clamped because agx_contrast_approx may not work
+	// well with values outside of the range [0.0, 1.0]
+	color = clamp(log2(color), min_ev, max_ev);
+	color = (color - min_ev) / (max_ev - min_ev);
+
+	// Apply sigmoid function approximation.
+	color = agx_contrast_approx(color);
+
+	// Convert back to linear before applying outset matrix.
+	color = pow(color, vec3(2.4));
+
+	// Apply outset to make the result more chroma-laden and then go back to linear sRGB.
+	color = agx_outset_rec2020_to_srgb_matrix * color;
+
+	// Blender's lusRGB.compensate_low_side is too complex for this shader, so
+	// simply return the color, even if it has negative components. These negative
+	// components may be useful for subsequent color adjustments.
+	return color;
+}
+
+vec3 linear_to_srgb(vec3 color) {
+	// Clamping is not strictly necessary for floating point nonlinear sRGB encoding,
+	// but many cases that call this function need the result clamped.
+	color = clamp(color, vec3(0.0), vec3(1.0));
+	const vec3 a = vec3(0.055f);
+	return mix((vec3(1.0f) + a) * pow(color.rgb, vec3(1.0f / 2.4f)) - a, 12.92f * color.rgb, lessThan(color.rgb, vec3(0.0031308f)));
+}
+
+#define TONEMAPPER_LINEAR 0
+#define TONEMAPPER_REINHARD 1
+#define TONEMAPPER_FILMIC 2
+#define TONEMAPPER_ACES 3
+#define TONEMAPPER_AGX 4
+
+vec3 apply_tonemapping(vec3 color, float white) { // inputs are LINEAR
+	// Ensure color values passed to tonemappers are positive.
+	// They can be negative in the case of negative lights, which leads to undesired behavior.
+	if (params.tonemapper == TONEMAPPER_LINEAR) {
+		return color;
+	} else if (params.tonemapper == TONEMAPPER_REINHARD) {
+		return tonemap_reinhard(max(vec3(0.0f), color), white);
+	} else if (params.tonemapper == TONEMAPPER_FILMIC) {
+		return tonemap_filmic(max(vec3(0.0f), color), white);
+	} else if (params.tonemapper == TONEMAPPER_ACES) {
+		return tonemap_aces(max(vec3(0.0f), color), white);
+	} else { // TONEMAPPER_AGX
+		return tonemap_agx(color);
+	}
+}
+
+#ifdef USE_MULTIVIEW
+vec3 gather_glow(sampler2DArray tex, vec2 uv) { // sample all selected glow levels, view is added to uv later
+#else
+vec3 gather_glow(sampler2D tex, vec2 uv) { // sample all selected glow levels
+#endif // defined(USE_MULTIVIEW)
+	vec3 glow = vec3(0.0f);
+
+	if (params.glow_levels[0] > 0.0001) {
+		glow += GLOW_TEXTURE_SAMPLE(tex, uv, 0).rgb * params.glow_levels[0];
+	}
+
+	if (params.glow_levels[1] > 0.0001) {
+		glow += GLOW_TEXTURE_SAMPLE(tex, uv, 1).rgb * params.glow_levels[1];
+	}
+
+	if (params.glow_levels[2] > 0.0001) {
+		glow += GLOW_TEXTURE_SAMPLE(tex, uv, 2).rgb * params.glow_levels[2];
+	}
+
+	if (params.glow_levels[3] > 0.0001) {
+		glow += GLOW_TEXTURE_SAMPLE(tex, uv, 3).rgb * params.glow_levels[3];
+	}
+
+	if (params.glow_levels[4] > 0.0001) {
+		glow += GLOW_TEXTURE_SAMPLE(tex, uv, 4).rgb * params.glow_levels[4];
+	}
+
+	if (params.glow_levels[5] > 0.0001) {
+		glow += GLOW_TEXTURE_SAMPLE(tex, uv, 5).rgb * params.glow_levels[5];
+	}
+
+	if (params.glow_levels[6] > 0.0001) {
+		glow += GLOW_TEXTURE_SAMPLE(tex, uv, 6).rgb * params.glow_levels[6];
+	}
+
+	return glow;
+}
+
+#define GLOW_MODE_ADD 0
+#define GLOW_MODE_SCREEN 1
+#define GLOW_MODE_SOFTLIGHT 2
+#define GLOW_MODE_REPLACE 3
+#define GLOW_MODE_MIX 4
+
+vec3 apply_glow(vec3 color, vec3 glow) { // apply glow using the selected blending mode
+	if (params.glow_mode == GLOW_MODE_ADD) {
+		return color + glow;
+	} else if (params.glow_mode == GLOW_MODE_SCREEN) {
+		// Needs color clamping.
+		glow.rgb = clamp(glow.rgb, vec3(0.0f), vec3(1.0f));
+		return max((color + glow) - (color * glow), vec3(0.0));
+	} else if (params.glow_mode == GLOW_MODE_SOFTLIGHT) {
+		// Needs color clamping.
+		glow.rgb = clamp(glow.rgb, vec3(0.0f), vec3(1.0f));
+		glow = glow * vec3(0.5f) + vec3(0.5f);
+
+		color.r = (glow.r <= 0.5f) ? (color.r - (1.0f - 2.0f * glow.r) * color.r * (1.0f - color.r)) : (((glow.r > 0.5f) && (color.r <= 0.25f)) ? (color.r + (2.0f * glow.r - 1.0f) * (4.0f * color.r * (4.0f * color.r + 1.0f) * (color.r - 1.0f) + 7.0f * color.r)) : (color.r + (2.0f * glow.r - 1.0f) * (sqrt(color.r) - color.r)));
+		color.g = (glow.g <= 0.5f) ? (color.g - (1.0f - 2.0f * glow.g) * color.g * (1.0f - color.g)) : (((glow.g > 0.5f) && (color.g <= 0.25f)) ? (color.g + (2.0f * glow.g - 1.0f) * (4.0f * color.g * (4.0f * color.g + 1.0f) * (color.g - 1.0f) + 7.0f * color.g)) : (color.g + (2.0f * glow.g - 1.0f) * (sqrt(color.g) - color.g)));
+		color.b = (glow.b <= 0.5f) ? (color.b - (1.0f - 2.0f * glow.b) * color.b * (1.0f - color.b)) : (((glow.b > 0.5f) && (color.b <= 0.25f)) ? (color.b + (2.0f * glow.b - 1.0f) * (4.0f * color.b * (4.0f * color.b + 1.0f) * (color.b - 1.0f) + 7.0f * color.b)) : (color.b + (2.0f * glow.b - 1.0f) * (sqrt(color.b) - color.b)));
+		return color;
+	} else { //replace
+		return glow;
+	}
+}
+
+vec3 apply_bcs(vec3 color, vec3 bcs) {
+	color = mix(vec3(0.0f), color, bcs.x);
+	color = mix(vec3(0.5f), color, bcs.y);
+	color = mix(vec3(dot(vec3(1.0f), color) * 0.33333f), color, bcs.z);
+
+	return color;
+}
+#ifdef USE_1D_LUT
+vec3 apply_color_correction(vec3 color) {
+	color.r = texture(source_color_correction, vec2(color.r, 0.0f)).r;
+	color.g = texture(source_color_correction, vec2(color.g, 0.0f)).g;
+	color.b = texture(source_color_correction, vec2(color.b, 0.0f)).b;
+	return color;
+}
+#else
+vec3 apply_color_correction(vec3 color) {
+	return textureLod(source_color_correction, color, 0.0).rgb;
+}
+#endif
+
+#ifndef SUBPASS
+
+// FXAA 3.11 compact, Ported from https://github.com/kosua20/Rendu/blob/master/resources/common/shaders/screens/fxaa.frag
+///////////////////////////////////////////////////////////////////////////////////
+// MIT License
+//
+// Copyright (c) 2017 Simon Rodriguez
+//
+// 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.
+///////////////////////////////////////////////////////////////////////////////////
+
+// Nvidia Original FXAA 3.11 License
+//----------------------------------------------------------------------------------
+// File:        es3-kepler\FXAA/FXAA3_11.h
+// SDK Version: v3.00
+// Email:       gameworks@nvidia.com
+// Site:        http://developer.nvidia.com/
+//
+// Copyright (c) 2014-2015, NVIDIA CORPORATION. All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions
+// are met:
+//  * Redistributions of source code must retain the above copyright
+//    notice, this list of conditions and the following disclaimer.
+//  * Redistributions in binary form must reproduce the above copyright
+//    notice, this list of conditions and the following disclaimer in the
+//    documentation and/or other materials provided with the distribution.
+//  * Neither the name of NVIDIA CORPORATION nor the names of its
+//    contributors may be used to endorse or promote products derived
+//    from this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
+// EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+// PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR
+// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
+// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+//
+//----------------------------------------------------------------------------------
+//
+//                    NVIDIA FXAA 3.11 by TIMOTHY LOTTES
+//
+//----------------------------------------------------------------------------------
+
+float QUALITY(float q) {
+	return (q < 5 ? 1.0 : (q > 5 ? (q < 10 ? 2.0 : (q < 11 ? 4.0 : 8.0)) : 1.5));
+}
+
+float rgb2luma(vec3 rgb) {
+	return sqrt(dot(rgb, vec3(0.299, 0.587, 0.114)));
+}
+
+vec3 do_fxaa(vec3 color, float exposure, vec2 uv_interp) {
+	const float EDGE_THRESHOLD_MIN = 0.0312;
+	const float EDGE_THRESHOLD_MAX = 0.125;
+	const int ITERATIONS = 12;
+	const float SUBPIXEL_QUALITY = 0.75;
+
+#ifdef USE_MULTIVIEW
+	float lumaUp = rgb2luma(textureLodOffset(source_color, vec3(uv_interp, ViewIndex), 0.0, ivec2(0, 1)).xyz * exposure * params.luminance_multiplier);
+	float lumaDown = rgb2luma(textureLodOffset(source_color, vec3(uv_interp, ViewIndex), 0.0, ivec2(0, -1)).xyz * exposure * params.luminance_multiplier);
+	float lumaLeft = rgb2luma(textureLodOffset(source_color, vec3(uv_interp, ViewIndex), 0.0, ivec2(-1, 0)).xyz * exposure * params.luminance_multiplier);
+	float lumaRight = rgb2luma(textureLodOffset(source_color, vec3(uv_interp, ViewIndex), 0.0, ivec2(1, 0)).xyz * exposure * params.luminance_multiplier);
+
+	float lumaCenter = rgb2luma(color);
+
+	float lumaMin = min(lumaCenter, min(min(lumaUp, lumaDown), min(lumaLeft, lumaRight)));
+	float lumaMax = max(lumaCenter, max(max(lumaUp, lumaDown), max(lumaLeft, lumaRight)));
+
+	float lumaRange = lumaMax - lumaMin;
+
+	if (lumaRange < max(EDGE_THRESHOLD_MIN, lumaMax * EDGE_THRESHOLD_MAX)) {
+		return color;
+	}
+
+	float lumaDownLeft = rgb2luma(textureLodOffset(source_color, vec3(uv_interp, ViewIndex), 0.0, ivec2(-1, -1)).xyz * exposure * params.luminance_multiplier);
+	float lumaUpRight = rgb2luma(textureLodOffset(source_color, vec3(uv_interp, ViewIndex), 0.0, ivec2(1, 1)).xyz * exposure * params.luminance_multiplier);
+	float lumaUpLeft = rgb2luma(textureLodOffset(source_color, vec3(uv_interp, ViewIndex), 0.0, ivec2(-1, 1)).xyz * exposure * params.luminance_multiplier);
+	float lumaDownRight = rgb2luma(textureLodOffset(source_color, vec3(uv_interp, ViewIndex), 0.0, ivec2(1, -1)).xyz * exposure * params.luminance_multiplier);
+
+	float lumaDownUp = lumaDown + lumaUp;
+	float lumaLeftRight = lumaLeft + lumaRight;
+
+	float lumaLeftCorners = lumaDownLeft + lumaUpLeft;
+	float lumaDownCorners = lumaDownLeft + lumaDownRight;
+	float lumaRightCorners = lumaDownRight + lumaUpRight;
+	float lumaUpCorners = lumaUpRight + lumaUpLeft;
+
+	float edgeHorizontal = abs(-2.0 * lumaLeft + lumaLeftCorners) + abs(-2.0 * lumaCenter + lumaDownUp) * 2.0 + abs(-2.0 * lumaRight + lumaRightCorners);
+	float edgeVertical = abs(-2.0 * lumaUp + lumaUpCorners) + abs(-2.0 * lumaCenter + lumaLeftRight) * 2.0 + abs(-2.0 * lumaDown + lumaDownCorners);
+
+	bool isHorizontal = (edgeHorizontal >= edgeVertical);
+
+	float stepLength = isHorizontal ? params.pixel_size.y : params.pixel_size.x;
+
+	float luma1 = isHorizontal ? lumaDown : lumaLeft;
+	float luma2 = isHorizontal ? lumaUp : lumaRight;
+	float gradient1 = luma1 - lumaCenter;
+	float gradient2 = luma2 - lumaCenter;
+
+	bool is1Steepest = abs(gradient1) >= abs(gradient2);
+
+	float gradientScaled = 0.25 * max(abs(gradient1), abs(gradient2));
+
+	float lumaLocalAverage = 0.0;
+	if (is1Steepest) {
+		stepLength = -stepLength;
+		lumaLocalAverage = 0.5 * (luma1 + lumaCenter);
+	} else {
+		lumaLocalAverage = 0.5 * (luma2 + lumaCenter);
+	}
+
+	vec2 currentUv = uv_interp;
+	if (isHorizontal) {
+		currentUv.y += stepLength * 0.5;
+	} else {
+		currentUv.x += stepLength * 0.5;
+	}
+
+	vec2 offset = isHorizontal ? vec2(params.pixel_size.x, 0.0) : vec2(0.0, params.pixel_size.y);
+	vec3 uv1 = vec3(currentUv - offset * QUALITY(0), ViewIndex);
+	vec3 uv2 = vec3(currentUv + offset * QUALITY(0), ViewIndex);
+
+	float lumaEnd1 = rgb2luma(textureLod(source_color, uv1, 0.0).xyz * exposure * params.luminance_multiplier);
+	float lumaEnd2 = rgb2luma(textureLod(source_color, uv2, 0.0).xyz * exposure * params.luminance_multiplier);
+	lumaEnd1 -= lumaLocalAverage;
+	lumaEnd2 -= lumaLocalAverage;
+
+	bool reached1 = abs(lumaEnd1) >= gradientScaled;
+	bool reached2 = abs(lumaEnd2) >= gradientScaled;
+	bool reachedBoth = reached1 && reached2;
+
+	if (!reached1) {
+		uv1 -= vec3(offset * QUALITY(1), 0.0);
+	}
+	if (!reached2) {
+		uv2 += vec3(offset * QUALITY(1), 0.0);
+	}
+
+	if (!reachedBoth) {
+		for (int i = 2; i < ITERATIONS; i++) {
+			if (!reached1) {
+				lumaEnd1 = rgb2luma(textureLod(source_color, uv1, 0.0).xyz * exposure * params.luminance_multiplier);
+				lumaEnd1 = lumaEnd1 - lumaLocalAverage;
+			}
+			if (!reached2) {
+				lumaEnd2 = rgb2luma(textureLod(source_color, uv2, 0.0).xyz * exposure * params.luminance_multiplier);
+				lumaEnd2 = lumaEnd2 - lumaLocalAverage;
+			}
+			reached1 = abs(lumaEnd1) >= gradientScaled;
+			reached2 = abs(lumaEnd2) >= gradientScaled;
+			reachedBoth = reached1 && reached2;
+			if (!reached1) {
+				uv1 -= vec3(offset * QUALITY(i), 0.0);
+			}
+			if (!reached2) {
+				uv2 += vec3(offset * QUALITY(i), 0.0);
+			}
+			if (reachedBoth) {
+				break;
+			}
+		}
+	}
+
+	float distance1 = isHorizontal ? (uv_interp.x - uv1.x) : (uv_interp.y - uv1.y);
+	float distance2 = isHorizontal ? (uv2.x - uv_interp.x) : (uv2.y - uv_interp.y);
+
+	bool isDirection1 = distance1 < distance2;
+	float distanceFinal = min(distance1, distance2);
+
+	float edgeThickness = (distance1 + distance2);
+
+	bool isLumaCenterSmaller = lumaCenter < lumaLocalAverage;
+
+	bool correctVariation1 = (lumaEnd1 < 0.0) != isLumaCenterSmaller;
+	bool correctVariation2 = (lumaEnd2 < 0.0) != isLumaCenterSmaller;
+
+	bool correctVariation = isDirection1 ? correctVariation1 : correctVariation2;
+
+	float pixelOffset = -distanceFinal / edgeThickness + 0.5;
+
+	float finalOffset = correctVariation ? pixelOffset : 0.0;
+
+	float lumaAverage = (1.0 / 12.0) * (2.0 * (lumaDownUp + lumaLeftRight) + lumaLeftCorners + lumaRightCorners);
+
+	float subPixelOffset1 = clamp(abs(lumaAverage - lumaCenter) / lumaRange, 0.0, 1.0);
+	float subPixelOffset2 = (-2.0 * subPixelOffset1 + 3.0) * subPixelOffset1 * subPixelOffset1;
+
+	float subPixelOffsetFinal = subPixelOffset2 * subPixelOffset2 * SUBPIXEL_QUALITY;
+
+	finalOffset = max(finalOffset, subPixelOffsetFinal);
+
+	vec3 finalUv = vec3(uv_interp, ViewIndex);
+	if (isHorizontal) {
+		finalUv.y += finalOffset * stepLength;
+	} else {
+		finalUv.x += finalOffset * stepLength;
+	}
+
+	vec3 finalColor = textureLod(source_color, finalUv, 0.0).xyz * exposure * params.luminance_multiplier;
+	return finalColor;
+
+#else
+	float lumaUp = rgb2luma(textureLodOffset(source_color, uv_interp, 0.0, ivec2(0, 1)).xyz * exposure * params.luminance_multiplier);
+	float lumaDown = rgb2luma(textureLodOffset(source_color, uv_interp, 0.0, ivec2(0, -1)).xyz * exposure * params.luminance_multiplier);
+	float lumaLeft = rgb2luma(textureLodOffset(source_color, uv_interp, 0.0, ivec2(-1, 0)).xyz * exposure * params.luminance_multiplier);
+	float lumaRight = rgb2luma(textureLodOffset(source_color, uv_interp, 0.0, ivec2(1, 0)).xyz * exposure * params.luminance_multiplier);
+
+	float lumaCenter = rgb2luma(color);
+
+	float lumaMin = min(lumaCenter, min(min(lumaUp, lumaDown), min(lumaLeft, lumaRight)));
+	float lumaMax = max(lumaCenter, max(max(lumaUp, lumaDown), max(lumaLeft, lumaRight)));
+
+	float lumaRange = lumaMax - lumaMin;
+
+	if (lumaRange < max(EDGE_THRESHOLD_MIN, lumaMax * EDGE_THRESHOLD_MAX)) {
+		return color;
+	}
+
+	float lumaDownLeft = rgb2luma(textureLodOffset(source_color, uv_interp, 0.0, ivec2(-1, -1)).xyz * exposure * params.luminance_multiplier);
+	float lumaUpRight = rgb2luma(textureLodOffset(source_color, uv_interp, 0.0, ivec2(1, 1)).xyz * exposure * params.luminance_multiplier);
+	float lumaUpLeft = rgb2luma(textureLodOffset(source_color, uv_interp, 0.0, ivec2(-1, 1)).xyz * exposure * params.luminance_multiplier);
+	float lumaDownRight = rgb2luma(textureLodOffset(source_color, uv_interp, 0.0, ivec2(1, -1)).xyz * exposure * params.luminance_multiplier);
+
+	float lumaDownUp = lumaDown + lumaUp;
+	float lumaLeftRight = lumaLeft + lumaRight;
+
+	float lumaLeftCorners = lumaDownLeft + lumaUpLeft;
+	float lumaDownCorners = lumaDownLeft + lumaDownRight;
+	float lumaRightCorners = lumaDownRight + lumaUpRight;
+	float lumaUpCorners = lumaUpRight + lumaUpLeft;
+
+	float edgeHorizontal = abs(-2.0 * lumaLeft + lumaLeftCorners) + abs(-2.0 * lumaCenter + lumaDownUp) * 2.0 + abs(-2.0 * lumaRight + lumaRightCorners);
+	float edgeVertical = abs(-2.0 * lumaUp + lumaUpCorners) + abs(-2.0 * lumaCenter + lumaLeftRight) * 2.0 + abs(-2.0 * lumaDown + lumaDownCorners);
+
+	bool isHorizontal = (edgeHorizontal >= edgeVertical);
+
+	float stepLength = isHorizontal ? params.pixel_size.y : params.pixel_size.x;
+
+	float luma1 = isHorizontal ? lumaDown : lumaLeft;
+	float luma2 = isHorizontal ? lumaUp : lumaRight;
+	float gradient1 = luma1 - lumaCenter;
+	float gradient2 = luma2 - lumaCenter;
+
+	bool is1Steepest = abs(gradient1) >= abs(gradient2);
+
+	float gradientScaled = 0.25 * max(abs(gradient1), abs(gradient2));
+
+	float lumaLocalAverage = 0.0;
+	if (is1Steepest) {
+		stepLength = -stepLength;
+		lumaLocalAverage = 0.5 * (luma1 + lumaCenter);
+	} else {
+		lumaLocalAverage = 0.5 * (luma2 + lumaCenter);
+	}
+
+	vec2 currentUv = uv_interp;
+	if (isHorizontal) {
+		currentUv.y += stepLength * 0.5;
+	} else {
+		currentUv.x += stepLength * 0.5;
+	}
+
+	vec2 offset = isHorizontal ? vec2(params.pixel_size.x, 0.0) : vec2(0.0, params.pixel_size.y);
+	vec2 uv1 = currentUv - offset * QUALITY(0);
+	vec2 uv2 = currentUv + offset * QUALITY(0);
+
+	float lumaEnd1 = rgb2luma(textureLod(source_color, uv1, 0.0).xyz * exposure * params.luminance_multiplier);
+	float lumaEnd2 = rgb2luma(textureLod(source_color, uv2, 0.0).xyz * exposure * params.luminance_multiplier);
+	lumaEnd1 -= lumaLocalAverage;
+	lumaEnd2 -= lumaLocalAverage;
+
+	bool reached1 = abs(lumaEnd1) >= gradientScaled;
+	bool reached2 = abs(lumaEnd2) >= gradientScaled;
+	bool reachedBoth = reached1 && reached2;
+
+	if (!reached1) {
+		uv1 -= offset * QUALITY(1);
+	}
+	if (!reached2) {
+		uv2 += offset * QUALITY(1);
+	}
+
+	if (!reachedBoth) {
+		for (int i = 2; i < ITERATIONS; i++) {
+			if (!reached1) {
+				lumaEnd1 = rgb2luma(textureLod(source_color, uv1, 0.0).xyz * exposure * params.luminance_multiplier);
+				lumaEnd1 = lumaEnd1 - lumaLocalAverage;
+			}
+			if (!reached2) {
+				lumaEnd2 = rgb2luma(textureLod(source_color, uv2, 0.0).xyz * exposure * params.luminance_multiplier);
+				lumaEnd2 = lumaEnd2 - lumaLocalAverage;
+			}
+			reached1 = abs(lumaEnd1) >= gradientScaled;
+			reached2 = abs(lumaEnd2) >= gradientScaled;
+			reachedBoth = reached1 && reached2;
+			if (!reached1) {
+				uv1 -= offset * QUALITY(i);
+			}
+			if (!reached2) {
+				uv2 += offset * QUALITY(i);
+			}
+			if (reachedBoth) {
+				break;
+			}
+		}
+	}
+
+	float distance1 = isHorizontal ? (uv_interp.x - uv1.x) : (uv_interp.y - uv1.y);
+	float distance2 = isHorizontal ? (uv2.x - uv_interp.x) : (uv2.y - uv_interp.y);
+
+	bool isDirection1 = distance1 < distance2;
+	float distanceFinal = min(distance1, distance2);
+
+	float edgeThickness = (distance1 + distance2);
+
+	bool isLumaCenterSmaller = lumaCenter < lumaLocalAverage;
+
+	bool correctVariation1 = (lumaEnd1 < 0.0) != isLumaCenterSmaller;
+	bool correctVariation2 = (lumaEnd2 < 0.0) != isLumaCenterSmaller;
+
+	bool correctVariation = isDirection1 ? correctVariation1 : correctVariation2;
+
+	float pixelOffset = -distanceFinal / edgeThickness + 0.5;
+
+	float finalOffset = correctVariation ? pixelOffset : 0.0;
+
+	float lumaAverage = (1.0 / 12.0) * (2.0 * (lumaDownUp + lumaLeftRight) + lumaLeftCorners + lumaRightCorners);
+
+	float subPixelOffset1 = clamp(abs(lumaAverage - lumaCenter) / lumaRange, 0.0, 1.0);
+	float subPixelOffset2 = (-2.0 * subPixelOffset1 + 3.0) * subPixelOffset1 * subPixelOffset1;
+
+	float subPixelOffsetFinal = subPixelOffset2 * subPixelOffset2 * SUBPIXEL_QUALITY;
+
+	finalOffset = max(finalOffset, subPixelOffsetFinal);
+
+	vec2 finalUv = uv_interp;
+	if (isHorizontal) {
+		finalUv.y += finalOffset * stepLength;
+	} else {
+		finalUv.x += finalOffset * stepLength;
+	}
+
+	vec3 finalColor = textureLod(source_color, finalUv, 0.0).xyz * exposure * params.luminance_multiplier;
+	return finalColor;
+
+#endif
+}
+#endif // !SUBPASS
+
+// From https://alex.vlachos.com/graphics/Alex_Vlachos_Advanced_VR_Rendering_GDC2015.pdf
+// and https://www.shadertoy.com/view/MslGR8 (5th one starting from the bottom)
+// NOTE: `frag_coord` is in pixels (i.e. not normalized UV).
+// This dithering must be applied after encoding changes (linear/nonlinear) have been applied
+// as the final step before quantization from floating point to integer values.
+vec3 screen_space_dither(vec2 frag_coord, float bit_alignment_diviser) {
+	// Iestyn's RGB dither (7 asm instructions) from Portal 2 X360, slightly modified for VR.
+	// Removed the time component to avoid passing time into this shader.
+	vec3 dither = vec3(dot(vec2(171.0, 231.0), frag_coord));
+	dither.rgb = fract(dither.rgb / vec3(103.0, 71.0, 97.0));
+
+	// Subtract 0.5 to avoid slightly brightening the whole viewport.
+	// Use a dither strength of 100% rather than the 37.5% suggested by the original source.
+	return (dither.rgb - 0.5) / bit_alignment_diviser;
+}
+
+void main() {
+#ifdef SUBPASS
+	// SUBPASS and USE_MULTIVIEW can be combined but in that case we're already reading from the correct layer
+#ifdef USE_MULTIVIEW
+	// In order to ensure the `SpvCapabilityMultiView` is included in the SPIR-V capabilities, gl_ViewIndex must
+	// be read in the shader. Without this, transpilation to Metal fails to include the multi-view variant.
+	uint vi = ViewIndex;
+#endif
+	vec4 color = subpassLoad(input_color);
+#elif defined(USE_MULTIVIEW)
+	vec4 color = textureLod(source_color, vec3(uv_interp, ViewIndex), 0.0f);
+#else
+	vec4 color = textureLod(source_color, uv_interp, 0.0f);
+#endif
+	color.rgb *= params.luminance_multiplier;
+
+	// Exposure
+
+	float exposure = params.exposure;
+
+#ifndef SUBPASS
+	if (bool(params.flags & FLAG_USE_AUTO_EXPOSURE)) {
+		exposure *= 1.0 / (texelFetch(source_auto_exposure, ivec2(0, 0), 0).r * params.luminance_multiplier / params.auto_exposure_scale);
+	}
+#endif
+
+	color.rgb *= exposure;
+
+	// Early Tonemap & SRGB Conversion
+#ifndef SUBPASS
+	if (bool(params.flags & FLAG_USE_FXAA)) {
+		// FXAA must be performed before glow to preserve the "bleed" effect of glow.
+		color.rgb = do_fxaa(color.rgb, exposure, uv_interp);
+	}
+
+	if (bool(params.flags & FLAG_USE_GLOW) && params.glow_mode == GLOW_MODE_MIX) {
+		vec3 glow = gather_glow(source_glow, uv_interp) * params.luminance_multiplier;
+		if (params.glow_map_strength > 0.001) {
+			glow = mix(glow, texture(glow_map, uv_interp).rgb * glow, params.glow_map_strength);
+		}
+		color.rgb = mix(color.rgb, glow, params.glow_intensity);
+	}
+#endif
+
+	color.rgb = apply_tonemapping(color.rgb, params.white);
+
+	bool convert_to_srgb = bool(params.flags & FLAG_CONVERT_TO_SRGB);
+	if (convert_to_srgb) {
+		color.rgb = linear_to_srgb(color.rgb); // Regular linear -> SRGB conversion.
+	}
+#ifndef SUBPASS
+	// Glow
+	if (bool(params.flags & FLAG_USE_GLOW) && params.glow_mode != GLOW_MODE_MIX) {
+		vec3 glow = gather_glow(source_glow, uv_interp) * params.glow_intensity * params.luminance_multiplier;
+		if (params.glow_map_strength > 0.001) {
+			glow = mix(glow, texture(glow_map, uv_interp).rgb * glow, params.glow_map_strength);
+		}
+
+		// high dynamic range -> SRGB
+		glow = apply_tonemapping(glow, params.white);
+		if (convert_to_srgb) {
+			glow = linear_to_srgb(glow);
+		}
+
+		color.rgb = apply_glow(color.rgb, glow);
+	}
+#endif
+
+	// Additional effects
+
+	if (bool(params.flags & FLAG_USE_BCS)) {
+		color.rgb = apply_bcs(color.rgb, params.bcs);
+	}
+
+	if (bool(params.flags & FLAG_USE_COLOR_CORRECTION)) {
+		// apply_color_correction requires nonlinear sRGB encoding
+		if (!convert_to_srgb) {
+			color.rgb = linear_to_srgb(color.rgb);
+		}
+		color.rgb = apply_color_correction(color.rgb);
+		// When convert_to_srgb is false, there is no need to convert back to
+		// linear because the color correction texture sampling does this for us.
+	}
+
+	// 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.
+
+	if (bool(params.flags & FLAG_USE_8_BIT_DEBANDING)) {
+		// Divide by 255 to align to 8-bit quantization.
+		color.rgb += screen_space_dither(gl_FragCoord.xy, 255.0);
+	} else if (bool(params.flags & FLAG_USE_10_BIT_DEBANDING)) {
+		// Divide by 1023 to align to 10-bit quantization.
+		color.rgb += screen_space_dither(gl_FragCoord.xy, 1023.0);
+	}
+
+	frag_color = color;
+}

servers/rendering/renderer_rd/shaders/effects/vrs.glsl

@@ -0,0 +1,93 @@
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef USE_MULTIVIEW
+#extension GL_EXT_multiview : enable
+#define ViewIndex gl_ViewIndex
+#endif //USE_MULTIVIEW
+
+#ifdef USE_MULTIVIEW
+layout(location = 0) out vec3 uv_interp;
+#else
+layout(location = 0) out vec2 uv_interp;
+#endif
+
+layout(push_constant, std430) uniform Params {
+	float max_texel_factor;
+	float res1;
+	float res2;
+	float res3;
+}
+params;
+
+void main() {
+	vec2 base_arr[3] = vec2[](vec2(-1.0, -1.0), vec2(-1.0, 3.0), vec2(3.0, -1.0));
+	gl_Position = vec4(base_arr[gl_VertexIndex], 0.0, 1.0);
+	uv_interp.xy = clamp(gl_Position.xy, vec2(0.0, 0.0), vec2(1.0, 1.0)) * 2.0; // saturate(x) * 2.0
+#ifdef USE_MULTIVIEW
+	uv_interp.z = ViewIndex;
+#endif
+}
+
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef USE_MULTIVIEW
+layout(location = 0) in vec3 uv_interp;
+layout(set = 0, binding = 0) uniform sampler2DArray source_color;
+#else /* USE_MULTIVIEW */
+layout(location = 0) in vec2 uv_interp;
+layout(set = 0, binding = 0) uniform sampler2D source_color;
+#endif /* USE_MULTIVIEW */
+
+#ifdef SPLIT_RG
+layout(location = 0) out vec2 frag_color;
+#else
+layout(location = 0) out uint frag_color;
+#endif
+
+layout(push_constant, std430) uniform Params {
+	float max_texel_factor;
+	float res1;
+	float res2;
+	float res3;
+}
+params;
+
+void main() {
+#ifdef USE_MULTIVIEW
+	vec3 uv = uv_interp;
+#else
+	vec2 uv = uv_interp;
+#endif
+
+	// Input is standardized. R for X, G for Y, 0.0 (0) = 1, 0.33 (85) = 2, 0.66 (170) = 3, 1.0 (255) = 8
+	vec4 color = textureLod(source_color, uv, 0.0);
+
+#ifdef SPLIT_RG
+	// Density map for VRS according to VK_EXT_fragment_density_map, we can use as is.
+	frag_color = max(vec2(1.0f) - color.rg, vec2(1.0f / 255.0f));
+#else
+	// Output image shading rate image for VRS according to VK_KHR_fragment_shading_rate.
+	color.r = clamp(floor(color.r * params.max_texel_factor + 0.1), 0.0, params.max_texel_factor);
+	color.g = clamp(floor(color.g * params.max_texel_factor + 0.1), 0.0, params.max_texel_factor);
+
+	// Note 1x4, 4x1, 1x8, 8x1, 2x8 and 8x2 are not supported:
+	if (color.r < (color.g - 1.0)) {
+		color.r = color.g - 1.0;
+	}
+	if (color.g < (color.r - 1.0)) {
+		color.g = color.r - 1.0;
+	}
+
+	// Encode to frag_color;
+	frag_color = int(color.r + 0.1) << 2;
+	frag_color += int(color.g + 0.1);
+#endif
+}