Merge pull request #89919 from BlueCube3310/bicubic-lightmap

Implement bicubic sampling for lightmaps

drivers/gles3/rasterizer_scene_gles3.cpp

@@ -3210,6 +3210,10 @@ void RasterizerSceneGLES3::_render_list_template(RenderListParameters *p_params,
 							if (lm->uses_spherical_harmonics) {
 								spec_constants |= SceneShaderGLES3::USE_SH_LIGHTMAP;
 							}
+
+							if (lightmap_bicubic_upscale) {
+								spec_constants |= SceneShaderGLES3::LIGHTMAP_BICUBIC_FILTER;
+							}
 						} else if (inst->lightmap_sh) {
 							spec_constants |= SceneShaderGLES3::USE_LIGHTMAP_CAPTURE;
 						} else {
@@ -3352,6 +3356,11 @@ void RasterizerSceneGLES3::_render_list_template(RenderListParameters *p_params,
 						Vector4 uv_scale(inst->lightmap_uv_scale.position.x, inst->lightmap_uv_scale.position.y, inst->lightmap_uv_scale.size.x, inst->lightmap_uv_scale.size.y);
 						material_storage->shaders.scene_shader.version_set_uniform(SceneShaderGLES3::LIGHTMAP_UV_SCALE, uv_scale, shader->version, instance_variant, spec_constants);
 
+						if (lightmap_bicubic_upscale) {
+							Vector2 light_texture_size(lm->light_texture_size.x, lm->light_texture_size.y);
+							material_storage->shaders.scene_shader.version_set_uniform(SceneShaderGLES3::LIGHTMAP_TEXTURE_SIZE, light_texture_size, shader->version, instance_variant, spec_constants);
+						}
+
 						float exposure_normalization = 1.0;
 						if (p_render_data->camera_attributes.is_valid()) {
 							float enf = RSG::camera_attributes->camera_attributes_get_exposure_normalization_factor(p_render_data->camera_attributes);
@@ -4047,6 +4056,10 @@ void RasterizerSceneGLES3::decals_set_filter(RS::DecalFilter p_filter) {
 void RasterizerSceneGLES3::light_projectors_set_filter(RS::LightProjectorFilter p_filter) {
 }
 
+void RasterizerSceneGLES3::lightmaps_set_bicubic_filter(bool p_enable) {
+	lightmap_bicubic_upscale = p_enable;
+}
+
 RasterizerSceneGLES3::RasterizerSceneGLES3() {
 	singleton = this;
 
@@ -4060,6 +4073,7 @@ RasterizerSceneGLES3::RasterizerSceneGLES3() {
 
 	positional_soft_shadow_filter_set_quality((RS::ShadowQuality)(int)GLOBAL_GET("rendering/lights_and_shadows/positional_shadow/soft_shadow_filter_quality"));
 	directional_soft_shadow_filter_set_quality((RS::ShadowQuality)(int)GLOBAL_GET("rendering/lights_and_shadows/directional_shadow/soft_shadow_filter_quality"));
+	lightmaps_set_bicubic_filter(GLOBAL_GET("rendering/lightmapping/lightmap_gi/use_bicubic_filter"));
 
 	{
 		// Setup Lights

drivers/gles3/rasterizer_scene_gles3.h

@@ -680,6 +680,8 @@ protected:
 	bool glow_bicubic_upscale = false;
 	RS::EnvironmentSSRRoughnessQuality ssr_roughness_quality = RS::ENV_SSR_ROUGHNESS_QUALITY_LOW;
 
+	bool lightmap_bicubic_upscale = false;
+
 	/* Sky */
 
 	struct SkyGlobals {
@@ -863,6 +865,7 @@ public:
 
 	void decals_set_filter(RS::DecalFilter p_filter) override;
 	void light_projectors_set_filter(RS::LightProjectorFilter p_filter) override;
+	virtual void lightmaps_set_bicubic_filter(bool p_enable) override;
 
 	RasterizerSceneGLES3();
 	~RasterizerSceneGLES3();

drivers/gles3/shaders/scene.glsl

@@ -36,6 +36,7 @@ ADDITIVE_OMNI = false
 ADDITIVE_SPOT = false
 RENDER_MATERIAL = false
 SECOND_REFLECTION_PROBE = false
+LIGHTMAP_BICUBIC_FILTER = false
 
 
 #[vertex]
@@ -925,6 +926,10 @@ uniform lowp uint lightmap_slice;
 uniform highp vec4 lightmap_uv_scale;
 uniform float lightmap_exposure_normalization;
 
+#ifdef LIGHTMAP_BICUBIC_FILTER
+uniform highp vec2 lightmap_texture_size;
+#endif
+
 #ifdef USE_SH_LIGHTMAP
 uniform mediump mat3 lightmap_normal_xform;
 #endif // USE_SH_LIGHTMAP
@@ -1417,6 +1422,67 @@ void reflection_process(samplerCube reflection_map,
 
 #endif // !MODE_RENDER_DEPTH
 
+#ifdef LIGHTMAP_BICUBIC_FILTER
+// w0, w1, w2, and w3 are the four cubic B-spline basis functions
+float w0(float a) {
+	return (1.0 / 6.0) * (a * (a * (-a + 3.0) - 3.0) + 1.0);
+}
+
+float w1(float a) {
+	return (1.0 / 6.0) * (a * a * (3.0 * a - 6.0) + 4.0);
+}
+
+float w2(float a) {
+	return (1.0 / 6.0) * (a * (a * (-3.0 * a + 3.0) + 3.0) + 1.0);
+}
+
+float w3(float a) {
+	return (1.0 / 6.0) * (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 -1.0 + w1(a) / (w0(a) + w1(a));
+}
+
+float h1(float a) {
+	return 1.0 + w3(a) / (w2(a) + w3(a));
+}
+
+vec4 textureArray_bicubic(sampler2DArray tex, vec3 uv, vec2 texture_size) {
+	vec2 texel_size = vec2(1.0) / texture_size;
+
+	uv.xy = uv.xy * texture_size + vec2(0.5);
+
+	vec2 iuv = floor(uv.xy);
+	vec2 fuv = fract(uv.xy);
+
+	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(0.5)) * texel_size;
+	vec2 p1 = (vec2(iuv.x + h1x, iuv.y + h0y) - vec2(0.5)) * texel_size;
+	vec2 p2 = (vec2(iuv.x + h0x, iuv.y + h1y) - vec2(0.5)) * texel_size;
+	vec2 p3 = (vec2(iuv.x + h1x, iuv.y + h1y) - vec2(0.5)) * texel_size;
+
+	return (g0(fuv.y) * (g0x * texture(tex, vec3(p0, uv.z)) + g1x * texture(tex, vec3(p1, uv.z)))) +
+			(g1(fuv.y) * (g0x * texture(tex, vec3(p2, uv.z)) + g1x * texture(tex, vec3(p3, uv.z))));
+}
+#endif //LIGHTMAP_BICUBIC_FILTER
+
 void main() {
 	//lay out everything, whatever is unused is optimized away anyway
 	vec3 vertex = vertex_interp;
@@ -1732,10 +1798,18 @@ void main() {
 
 #ifdef USE_SH_LIGHTMAP
 		uvw.z *= 4.0; // SH textures use 4 times more data.
+
+#ifdef LIGHTMAP_BICUBIC_FILTER
+		vec3 lm_light_l0 = textureArray_bicubic(lightmap_textures, uvw + vec3(0.0, 0.0, 0.0), lightmap_texture_size).rgb;
+		vec3 lm_light_l1n1 = textureArray_bicubic(lightmap_textures, uvw + vec3(0.0, 0.0, 1.0), lightmap_texture_size).rgb;
+		vec3 lm_light_l1_0 = textureArray_bicubic(lightmap_textures, uvw + vec3(0.0, 0.0, 2.0), lightmap_texture_size).rgb;
+		vec3 lm_light_l1p1 = textureArray_bicubic(lightmap_textures, uvw + vec3(0.0, 0.0, 3.0), lightmap_texture_size).rgb;
+#else
 		vec3 lm_light_l0 = textureLod(lightmap_textures, uvw + vec3(0.0, 0.0, 0.0), 0.0).rgb;
 		vec3 lm_light_l1n1 = textureLod(lightmap_textures, uvw + vec3(0.0, 0.0, 1.0), 0.0).rgb;
 		vec3 lm_light_l1_0 = textureLod(lightmap_textures, uvw + vec3(0.0, 0.0, 2.0), 0.0).rgb;
 		vec3 lm_light_l1p1 = textureLod(lightmap_textures, uvw + vec3(0.0, 0.0, 3.0), 0.0).rgb;
+#endif
 
 		vec3 n = normalize(lightmap_normal_xform * normal);
 
@@ -1749,8 +1823,12 @@ void main() {
 			specular_light += lm_light_l1_0 * 0.32573 * r.z * lightmap_exposure_normalization;
 			specular_light += lm_light_l1p1 * 0.32573 * r.x * lightmap_exposure_normalization;
 		}
+#else
+#ifdef LIGHTMAP_BICUBIC_FILTER
+		ambient_light += textureArray_bicubic(lightmap_textures, uvw, lightmap_texture_size).rgb * lightmap_exposure_normalization;
 #else
 		ambient_light += textureLod(lightmap_textures, uvw, 0.0).rgb * lightmap_exposure_normalization;
+#endif
 #endif
 	}
 #endif // USE_LIGHTMAP

drivers/gles3/storage/light_storage.cpp

@@ -1046,6 +1046,9 @@ void LightStorage::lightmap_set_textures(RID p_lightmap, RID p_light, bool p_use
 	lightmap->light_texture = p_light;
 	lightmap->uses_spherical_harmonics = p_uses_spherical_haromics;
 
+	Vector3i light_texture_size = GLES3::TextureStorage::get_singleton()->texture_get_size(lightmap->light_texture);
+	lightmap->light_texture_size = Vector2i(light_texture_size.x, light_texture_size.y);
+
 	GLuint tex = GLES3::TextureStorage::get_singleton()->texture_get_texid(lightmap->light_texture);
 	glBindTexture(GL_TEXTURE_2D_ARRAY, tex);
 	glTexParameteri(GL_TEXTURE_2D_ARRAY, GL_TEXTURE_MAG_FILTER, GL_LINEAR);

drivers/gles3/storage/light_storage.h

@@ -180,6 +180,7 @@ struct Lightmap {
 	bool interior = false;
 	AABB bounds = AABB(Vector3(), Vector3(1, 1, 1));
 	float baked_exposure = 1.0;
+	Vector2i light_texture_size;
 	int32_t array_index = -1; //unassigned
 	PackedVector3Array points;
 	PackedColorArray point_sh;

drivers/gles3/storage/texture_storage.cpp

@@ -1680,6 +1680,14 @@ uint32_t TextureStorage::texture_get_texid(RID p_texture) const {
 	return texture->tex_id;
 }
 
+Vector3i TextureStorage::texture_get_size(RID p_texture) const {
+	Texture *texture = texture_owner.get_or_null(p_texture);
+
+	ERR_FAIL_NULL_V(texture, Vector3i(0, 0, 0));
+
+	return Vector3i(texture->width, texture->height, texture->depth);
+}
+
 uint32_t TextureStorage::texture_get_width(RID p_texture) const {
 	Texture *texture = texture_owner.get_or_null(p_texture);
 

drivers/gles3/storage/texture_storage.h

@@ -553,6 +553,7 @@ public:
 	void texture_set_data(RID p_texture, const Ref<Image> &p_image, int p_layer = 0);
 	virtual Image::Format texture_get_format(RID p_texture) const override;
 	uint32_t texture_get_texid(RID p_texture) const;
+	Vector3i texture_get_size(RID p_texture) const;
 	uint32_t texture_get_width(RID p_texture) const;
 	uint32_t texture_get_height(RID p_texture) const;
 	uint32_t texture_get_depth(RID p_texture) const;