Replace BRDF approximation with a DFG LUT

doc/classes/BaseMaterial3D.xml

@@ -723,6 +723,8 @@
 		</constant>
 		<constant name="SPECULAR_SCHLICK_GGX" value="0" enum="SpecularMode">
 			Default specular blob.
+			[b]Note:[/b] Forward+ uses multiscattering for more accurate reflections, although the impact of multiscattering is more noticeable on rough metallic surfaces than on smooth, non-metallic surfaces.
+			[b]Note:[/b] Mobile and Compatibility don't perform multiscattering for performance reasons. Instead, they perform single scattering, which means rough metallic surfaces may look slightly darker than intended.
 		</constant>
 		<constant name="SPECULAR_TOON" value="1" enum="SpecularMode">
 			Toon blob which changes size based on roughness.

servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp

@@ -3205,6 +3205,14 @@ void RenderForwardClustered::_update_render_base_uniform_set() {
 			uniforms.push_back(u);
 		}
 
+		{
+			RD::Uniform u;
+			u.binding = 16;
+			u.uniform_type = RD::UNIFORM_TYPE_TEXTURE;
+			u.append_id(dfg_lut.texture);
+			uniforms.push_back(u);
+		}
+
 		render_base_uniform_set = RD::get_singleton()->uniform_set_create(uniforms, scene_shader.default_shader_rd, SCENE_UNIFORM_SET);
 	}
 }
@@ -4946,6 +4954,45 @@ RenderForwardClustered::RenderForwardClustered() {
 		best_fit_normal.shader.version_free(best_fit_normal.shader_version);
 	}
 
+	/* DFG LUT */
+	{
+		Vector<String> modes;
+		modes.push_back("\n");
+		dfg_lut.shader.initialize(modes);
+		dfg_lut.shader_version = dfg_lut.shader.version_create();
+		dfg_lut.pipeline = RD::get_singleton()->compute_pipeline_create(dfg_lut.shader.version_get_shader(dfg_lut.shader_version, 0));
+
+		RD::TextureFormat tformat;
+		tformat.format = RD::DATA_FORMAT_R16G16B16A16_SFLOAT;
+		tformat.width = 128;
+		tformat.height = 128;
+		tformat.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
+		tformat.texture_type = RD::TEXTURE_TYPE_2D;
+		dfg_lut.texture = RD::get_singleton()->texture_create(tformat, RD::TextureView());
+
+		RID shader = dfg_lut.shader.version_get_shader(dfg_lut.shader_version, 0);
+		ERR_FAIL_COND(shader.is_null());
+
+		Vector<RD::Uniform> uniforms;
+
+		{
+			RD::Uniform u;
+			u.binding = 0;
+			u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
+			u.append_id(dfg_lut.texture);
+			uniforms.push_back(u);
+		}
+		RID uniform_set = RD::get_singleton()->uniform_set_create(uniforms, shader, 0);
+
+		RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin();
+		RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, dfg_lut.pipeline);
+		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, uniform_set, 0);
+		RD::get_singleton()->compute_list_dispatch_threads(compute_list, tformat.width, tformat.height, 1);
+		RD::get_singleton()->compute_list_end();
+
+		dfg_lut.shader.version_free(dfg_lut.shader_version);
+	}
+
 	_update_shader_quality_settings();
 	_update_global_pipeline_data_requirements_from_project();
 
@@ -4995,6 +5042,7 @@ RenderForwardClustered::~RenderForwardClustered() {
 	RD::get_singleton()->free(shadow_sampler);
 	RSG::light_storage->directional_shadow_atlas_set_size(0);
 	RD::get_singleton()->free(best_fit_normal.texture);
+	RD::get_singleton()->free(dfg_lut.texture);
 
 	{
 		for (const RID &rid : scene_state.uniform_buffers) {

servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.h

@@ -43,6 +43,7 @@
 #include "servers/rendering/renderer_rd/forward_clustered/scene_shader_forward_clustered.h"
 #include "servers/rendering/renderer_rd/renderer_scene_render_rd.h"
 #include "servers/rendering/renderer_rd/shaders/forward_clustered/best_fit_normal.glsl.gen.h"
+#include "servers/rendering/renderer_rd/shaders/forward_clustered/integrate_dfg.glsl.gen.h"
 
 #define RB_SCOPE_FORWARD_CLUSTERED SNAME("forward_clustered")
 
@@ -180,6 +181,13 @@ private:
 		RID texture;
 	} best_fit_normal;
 
+	struct IntegrateDFG {
+		IntegrateDfgShaderRD shader;
+		RID shader_version;
+		RID pipeline;
+		RID texture;
+	} dfg_lut;
+
 	enum PassMode {
 		PASS_MODE_COLOR,
 		PASS_MODE_SHADOW,

servers/rendering/renderer_rd/shaders/forward_clustered/integrate_dfg.glsl

@@ -0,0 +1,156 @@
+#[compute]
+#version 450
+
+// References:
+// https://www.gamedevs.org/uploads/real-shading-in-unreal-engine-4.pdf
+// https://google.github.io/filament/Filament.html
+// https://learnopengl.com/PBR/IBL/Specular-IBL
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+layout(rgba16f, set = 0, binding = 0) uniform restrict writeonly image2D current_image;
+
+#define M_PI 3.14159265359
+#define SAMPLE_COUNT 1024
+#define SIZE 128
+
+#define saturate(x) clamp(x, 0, 1)
+
+// http://holger.dammertz.org/stuff/notes_HammersleyOnHemisphere.html
+// efficient VanDerCorpus calculation
+float radical_inverse_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, float n) {
+	return vec2(float(i) / n, radical_inverse_vdc(i));
+}
+
+vec3 importance_sample_ggx(vec2 Xi, vec3 N, float roughness) {
+	float a = roughness * roughness;
+
+	float phi = 2.0 * M_PI * Xi.x;
+	float cosTheta = sqrt((1.0 - Xi.y) / (1.0 + (a * a - 1.0) * Xi.y));
+	float sinTheta = sqrt(1.0 - cosTheta * cosTheta);
+
+	// from spherical coordinates to cartesian coordinates - halfway vector
+	vec3 H;
+	H.x = cos(phi) * sinTheta;
+	H.y = sin(phi) * sinTheta;
+	H.z = cosTheta;
+
+	// from tangent-space H vector to world-space sample vector
+	vec3 up = abs(N.z) < 0.999 ? vec3(0.0, 0.0, 1.0) : vec3(1.0, 0.0, 0.0);
+	vec3 tangent = normalize(cross(up, N));
+	vec3 bitangent = cross(N, tangent);
+
+	vec3 sampleVec = tangent * H.x + bitangent * H.y + N * H.z;
+	return normalize(sampleVec);
+}
+
+float geometry_schlick_ggx(float NdotV, float roughness) {
+	// note that we use a different k for IBL
+	float a = roughness;
+	float k = (a * a) / 2.0;
+
+	float nom = NdotV;
+	float denom = NdotV * (1.0 - k) + k;
+
+	return nom / denom;
+}
+
+float geometry_smith(vec3 N, vec3 V, vec3 L, float roughness) {
+	float NdotV = saturate(dot(N, V));
+	float NdotL = saturate(dot(N, L));
+	float ggx2 = geometry_schlick_ggx(NdotV, roughness);
+	float ggx1 = geometry_schlick_ggx(NdotL, roughness);
+
+	return ggx1 * ggx2;
+}
+
+vec3 importance_uniform_sample(vec2 u) {
+	float phi = 2.0f * M_PI * u.x;
+	float cosTheta = 1 - u.y;
+	float sinTheta = sqrt(1 - cosTheta * cosTheta);
+	return vec3(sinTheta * cos(phi), sinTheta * sin(phi), cosTheta);
+}
+
+float distribution_charlie(float NoH, float roughness) {
+	// Estevez and Kulla 2017, "Production Friendly Microfacet Sheen BRDF"
+	float a = roughness * roughness;
+	float invAlpha = 1 / a;
+	float cos2h = NoH * NoH;
+	float sin2h = 1 - cos2h;
+	return (2.0f + invAlpha) * pow(sin2h, invAlpha * 0.5f) / (2.0f * M_PI);
+}
+
+float visibility_ashikhmin(float NoV, float NoL) {
+	// Neubelt and Pettineo 2013, "Crafting a Next-gen Material Pipeline for The Order: 1886"
+	return 1 / (4 * (NoL + NoV - NoL * NoV));
+}
+
+void integrate_brdfs(float n_dot_v, float roughness, out vec2 brdf, out float cloth_brdf) {
+	vec3 v = vec3(sqrt(1.0 - n_dot_v * n_dot_v), 0, n_dot_v);
+	vec3 n = vec3(0.0f, 0.0f, 1.0f);
+	float A = 0.0f;
+	float B = 0.0f;
+	float C = 0.0f;
+
+	for (uint i = 0; i < SAMPLE_COUNT; ++i) {
+		vec2 Xi = hammersley(i, SAMPLE_COUNT);
+		vec3 h = importance_sample_ggx(Xi, n, roughness);
+		vec3 l = normalize(2.0 * dot(v, h) * h - v);
+
+		float n_dot_l = saturate(l.z);
+		float n_dot_h = saturate(h.z);
+		float v_dot_h = saturate(dot(v, h));
+
+		if (n_dot_l > 0.0) {
+			float G = geometry_smith(n, v, l, roughness);
+			float G_Vis = (G * v_dot_h) / (n_dot_h * n_dot_v);
+			float Fc = pow(1.0 - v_dot_h, 5.0);
+
+			// LDFG term for multiscattering
+			// https://google.github.io/filament/Filament.html#toc5.3.4.7
+			A += Fc * G_Vis;
+			B += G_Vis;
+		}
+
+		// Cloth BRDF calculations
+		// https://github.com/google/filament/blob/main/libs/ibl/src/CubemapIBL.cpp#L856-L874
+		vec3 h_cloth = importance_uniform_sample(Xi);
+		vec3 l_cloth = normalize(2.0 * dot(v, h_cloth) * h_cloth - v);
+		float n_dot_l_cloth = saturate(l_cloth.z);
+		float n_dot_h_cloth = saturate(h_cloth.z);
+		float v_dot_h_cloth = saturate(dot(v, h_cloth));
+
+		if (n_dot_l_cloth > 0.0) {
+			float v_cloth = visibility_ashikhmin(n_dot_v, n_dot_l_cloth);
+			float d_cloth = distribution_charlie(n_dot_h_cloth, roughness);
+			C += v_cloth * d_cloth * n_dot_l_cloth * v_dot_h_cloth;
+		}
+	}
+
+	A /= float(SAMPLE_COUNT);
+	B /= float(SAMPLE_COUNT);
+	C *= (4.0 * 2.0 * M_PI / SAMPLE_COUNT);
+
+	brdf = vec2(A, B);
+	cloth_brdf = C;
+}
+
+void main() {
+	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
+	float roughness = float(pos.y + 0.5f) / SIZE;
+	float NdotV = float(pos.x + 0.5f) / SIZE;
+	vec2 brdf;
+	float cloth_brdf;
+	integrate_brdfs(NdotV, roughness, brdf, cloth_brdf);
+	ivec2 out_pos = ivec2(pos.x, (SIZE - 1) - pos.y);
+	imageStore(current_image, out_pos, vec4(brdf, cloth_brdf, 1.0));
+}

servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered.glsl

@@ -1151,6 +1151,7 @@ void fragment_shader(in SceneData scene_data) {
 	float clearcoat_roughness = 0.0;
 	float anisotropy = 0.0;
 	vec2 anisotropy_flow = vec2(1.0, 0.0);
+	vec3 energy_compensation = vec3(1.0);
 #ifndef FOG_DISABLED
 	vec4 fog = vec4(0.0);
 #endif // !FOG_DISABLED
@@ -1994,19 +1995,15 @@ void fragment_shader(in SceneData scene_data) {
 		//simplify for toon, as
 		specular_light *= specular * metallic * albedo * 2.0;
 #else
+		// Base Layer
+		float NdotV = clamp(dot(normal, view), 0.0001, 1.0);
+		vec2 envBRDF = prefiltered_dfg(roughness, NdotV).xy;
+		// Multiscattering
+		energy_compensation = get_energy_compensation(f0, envBRDF.y);
 
-		// scales the specular reflections, needs to be computed before lighting happens,
+		// cheap luminance approximation
-		// but after environment, GI, and reflection probes are added
+		float f90 = clamp(50.0 * f0.g, metallic, 1.0);
-		// Environment brdf approximation (Lazarov 2013)
+		specular_light *= energy_compensation * (f90 * envBRDF.x + f0 * envBRDF.y);
-		// see https://www.unrealengine.com/en-US/blog/physically-based-shading-on-mobile
-		const vec4 c0 = vec4(-1.0, -0.0275, -0.572, 0.022);
-		const vec4 c1 = vec4(1.0, 0.0425, 1.04, -0.04);
-		vec4 r = roughness * c0 + c1;
-		float ndotv = clamp(dot(normal, view), 0.0, 1.0);
-		float a004 = min(r.x * r.x, exp2(-9.28 * ndotv)) * r.x + r.y;
-		vec2 env = vec2(-1.04, 1.04) * a004 + r.zw;
-
-		specular_light *= env.x * f0 + env.y * clamp(50.0 * f0.g, metallic, 1.0);
 #endif
 	}
 
@@ -2405,7 +2402,7 @@ void fragment_shader(in SceneData scene_data) {
 #else
 					directional_lights.data[i].color * directional_lights.data[i].energy * tint,
 #endif
-					true, shadow, f0, orms, directional_lights.data[i].specular, albedo, alpha, screen_uv,
+					true, shadow, f0, orms, directional_lights.data[i].specular, albedo, alpha, screen_uv, energy_compensation,
 #ifdef LIGHT_BACKLIGHT_USED
 					backlight,
 #endif
@@ -2475,7 +2472,7 @@ void fragment_shader(in SceneData scene_data) {
 					continue; // Statically baked light and object uses lightmap, skip
 				}
 
-				light_process_omni(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, scene_data.taa_frame_count, albedo, alpha, screen_uv,
+				light_process_omni(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, scene_data.taa_frame_count, albedo, alpha, screen_uv, energy_compensation,
 #ifdef LIGHT_BACKLIGHT_USED
 						backlight,
 #endif
@@ -2543,7 +2540,7 @@ void fragment_shader(in SceneData scene_data) {
 					continue; // Statically baked light and object uses lightmap, skip
 				}
 
-				light_process_spot(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, scene_data.taa_frame_count, albedo, alpha, screen_uv,
+				light_process_spot(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, scene_data.taa_frame_count, albedo, alpha, screen_uv, energy_compensation,
 #ifdef LIGHT_BACKLIGHT_USED
 						backlight,
 #endif

servers/rendering/renderer_rd/shaders/forward_clustered/scene_forward_clustered_inc.glsl

@@ -280,6 +280,8 @@ layout(set = 0, binding = 14) uniform sampler DEFAULT_SAMPLER_LINEAR_WITH_MIPMAP
 
 layout(set = 0, binding = 15) uniform texture2D best_fit_normal_texture;
 
+layout(set = 0, binding = 16) uniform texture2D dfg;
+
 /* Set 1: Render Pass (changes per render pass) */
 
 layout(set = 1, binding = 0, std140) uniform SceneDataBlock {
@@ -454,6 +456,18 @@ vec4 normal_roughness_compatibility(vec4 p_normal_roughness) {
 	return vec4(normalize(p_normal_roughness.xyz * 2.0 - 1.0) * 0.5 + 0.5, roughness);
 }
 
+// https://google.github.io/filament/Filament.html#toc5.3.4.7
+// Note: The roughness value is inverted
+vec3 prefiltered_dfg(float lod, float NoV) {
+	return textureLod(sampler2D(dfg, SAMPLER_LINEAR_CLAMP), vec2(NoV, 1.0 - lod), 0.0).rgb;
+}
+
+// Compute multiscatter compensation
+// https://google.github.io/filament/Filament.html#listing_energycompensationimpl
+vec3 get_energy_compensation(vec3 f0, float env) {
+	return 1.0 + f0 * (1.0 / env - 1.0);
+}
+
 /* Set 2 Skeleton & Instancing (can change per item) */
 
 layout(set = 2, binding = 0, std430) restrict readonly buffer Transforms {

servers/rendering/renderer_rd/shaders/forward_mobile/scene_forward_mobile.glsl

@@ -1705,7 +1705,8 @@ void main() {
 
 			light_compute(normal, directional_lights.data[i].direction, view, size_A,
 					directional_lights.data[i].color * directional_lights.data[i].energy * tint,
-					true, shadow, f0, orms, directional_lights.data[i].specular, albedo, alpha, screen_uv,
+					true, shadow, f0, orms, directional_lights.data[i].specular, albedo, alpha,
+					screen_uv, vec3(1.0),
 #ifdef LIGHT_BACKLIGHT_USED
 					backlight,
 #endif
@@ -1736,7 +1737,7 @@ void main() {
 	uvec2 omni_indices = instances.data[draw_call.instance_index].omni_lights;
 	for (uint i = 0; i < sc_omni_lights(); i++) {
 		uint light_index = (i > 3) ? ((omni_indices.y >> ((i - 4) * 8)) & 0xFF) : ((omni_indices.x >> (i * 8)) & 0xFF);
-		light_process_omni(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, scene_data.taa_frame_count, albedo, alpha, screen_uv,
+		light_process_omni(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, scene_data.taa_frame_count, albedo, alpha, screen_uv, vec3(1.0),
 #ifdef LIGHT_BACKLIGHT_USED
 				backlight,
 #endif
@@ -1764,7 +1765,7 @@ void main() {
 	uvec2 spot_indices = instances.data[draw_call.instance_index].spot_lights;
 	for (uint i = 0; i < sc_spot_lights(); i++) {
 		uint light_index = (i > 3) ? ((spot_indices.y >> ((i - 4) * 8)) & 0xFF) : ((spot_indices.x >> (i * 8)) & 0xFF);
-		light_process_spot(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, scene_data.taa_frame_count, albedo, alpha, screen_uv,
+		light_process_spot(light_index, vertex, view, normal, vertex_ddx, vertex_ddy, f0, orms, scene_data.taa_frame_count, albedo, alpha, screen_uv, vec3(1.0),
 #ifdef LIGHT_BACKLIGHT_USED
 				backlight,
 #endif

servers/rendering/renderer_rd/shaders/scene_forward_lights_inc.glsl

@@ -52,7 +52,7 @@ vec3 F0(float metallic, float specular, vec3 albedo) {
 	return mix(vec3(dielectric), albedo, vec3(metallic));
 }
 
-void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, bool is_directional, float attenuation, vec3 f0, uint orms, float specular_amount, vec3 albedo, inout float alpha, vec2 screen_uv,
+void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, bool is_directional, float attenuation, vec3 f0, uint orms, float specular_amount, vec3 albedo, inout float alpha, vec2 screen_uv, vec3 energy_compensation,
 #ifdef LIGHT_BACKLIGHT_USED
 		vec3 backlight,
 #endif
@@ -230,7 +230,7 @@ void light_compute(vec3 N, vec3 L, vec3 V, float A, vec3 light_color, bool is_di
 			// https://google.github.io/filament/Filament.html#lighting/occlusion/specularocclusion
 			float f90 = clamp(dot(f0, vec3(50.0 * 0.33)), metallic, 1.0);
 			vec3 F = f0 + (f90 - f0) * cLdotH5;
-			vec3 specular_brdf_NL = cNdotL * D * F * G;
+			vec3 specular_brdf_NL = energy_compensation * (cNdotL * D * F * G);
 			specular_light += specular_brdf_NL * light_color * attenuation * specular_amount;
 #endif
 		}
@@ -405,7 +405,7 @@ float get_omni_attenuation(float distance, float inv_range, float decay) {
 	return nd * pow(max(distance, 0.0001), -decay);
 }
 
-void light_process_omni(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 vertex_ddx, vec3 vertex_ddy, vec3 f0, uint orms, float taa_frame_count, vec3 albedo, inout float alpha, vec2 screen_uv,
+void light_process_omni(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 vertex_ddx, vec3 vertex_ddy, vec3 f0, uint orms, float taa_frame_count, vec3 albedo, inout float alpha, vec2 screen_uv, vec3 energy_compensation,
 #ifdef LIGHT_BACKLIGHT_USED
 		vec3 backlight,
 #endif
@@ -667,7 +667,7 @@ void light_process_omni(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v
 	}
 
 	vec3 light_rel_vec_norm = light_rel_vec / light_length;
-	light_compute(normal, light_rel_vec_norm, eye_vec, size, color, false, omni_attenuation * shadow, f0, orms, omni_lights.data[idx].specular_amount, albedo, alpha, screen_uv,
+	light_compute(normal, light_rel_vec_norm, eye_vec, size, color, false, omni_attenuation * shadow, f0, orms, omni_lights.data[idx].specular_amount, albedo, alpha, screen_uv, energy_compensation,
 #ifdef LIGHT_BACKLIGHT_USED
 			backlight,
 #endif
@@ -702,7 +702,7 @@ vec2 normal_to_panorama(vec3 n) {
 	return panorama_coords;
 }
 
-void light_process_spot(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 vertex_ddx, vec3 vertex_ddy, vec3 f0, uint orms, float taa_frame_count, vec3 albedo, inout float alpha, vec2 screen_uv,
+void light_process_spot(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 vertex_ddx, vec3 vertex_ddy, vec3 f0, uint orms, float taa_frame_count, vec3 albedo, inout float alpha, vec2 screen_uv, vec3 energy_compensation,
 #ifdef LIGHT_BACKLIGHT_USED
 		vec3 backlight,
 #endif
@@ -871,7 +871,7 @@ void light_process_spot(uint idx, vec3 vertex, vec3 eye_vec, vec3 normal, vec3 v
 		}
 	}
 
-	light_compute(normal, light_rel_vec_norm, eye_vec, size, color, false, spot_attenuation * shadow, f0, orms, spot_lights.data[idx].specular_amount, albedo, alpha, screen_uv,
+	light_compute(normal, light_rel_vec_norm, eye_vec, size, color, false, spot_attenuation * shadow, f0, orms, spot_lights.data[idx].specular_amount, albedo, alpha, screen_uv, energy_compensation,
 #ifdef LIGHT_BACKLIGHT_USED
 			backlight,
 #endif