Auto LOD fixes and improvements

* Fixed LODs for shadow meshes.
* Added a merging step before simplification. This helps with tesselated
  meshes that were previously left untouched. The angle difference at
  wich edges ar considered "hard" can be tweaked as an import setting.
* LODs will now start with the highest decimation possible and keep
  doubling (approximately) the number of triangles from there. This
  makes sure that very low triangle counts are included when possible.
* Given more weight to normal preservation.
* Modified MeshOptimizer to report distance-based error instead of
  including attributes in the reported metrics.
* Added attribute transference between the original mesh and the
  various LODs. Right now only normals are taken into account,
  but it could be expanded to other attributes in the future.

15 files changed, 1044 insertions, 99 deletions

diff --git a/core/math/static_raycaster.cpp b/core/math/static_raycaster.cpp
new file mode 100644
index 0000000000..da05d49428
--- /dev/null
+++ b/core/math/static_raycaster.cpp
@@ -0,0 +1,40 @@
+/*************************************************************************/
+/*  static_raycaster.cpp                                                 */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2021 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2021 Godot Engine contributors (cf. AUTHORS.md).   */
+/*                                                                       */
+/* 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.                */
+/*************************************************************************/
+
+#include "static_raycaster.h"
+
+StaticRaycaster *(*StaticRaycaster::create_function)() = nullptr;
+
+Ref<StaticRaycaster> StaticRaycaster::create() {
+	if (create_function) {
+		return Ref<StaticRaycaster>(create_function());
+	}
+	return Ref<StaticRaycaster>();
+}
diff --git a/core/math/static_raycaster.h b/core/math/static_raycaster.h
new file mode 100644
index 0000000000..3759c788a7
--- /dev/null
+++ b/core/math/static_raycaster.h
@@ -0,0 +1,111 @@
+/*************************************************************************/
+/*  static_raycaster.h                                                   */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2021 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2021 Godot Engine contributors (cf. AUTHORS.md).   */
+/*                                                                       */
+/* 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.                */
+/*************************************************************************/
+
+#ifndef STATIC_RAYCASTER_H
+#define STATIC_RAYCASTER_H
+
+#include "core/object/ref_counted.h"
+
+#if !defined(__aligned)
+
+#if defined(_WIN32) && defined(_MSC_VER)
+#define __aligned(...) __declspec(align(__VA_ARGS__))
+#else
+#define __aligned(...) __attribute__((aligned(__VA_ARGS__)))
+#endif
+
+#endif
+
+class StaticRaycaster : public RefCounted {
+	GDCLASS(StaticRaycaster, RefCounted)
+protected:
+	static StaticRaycaster *(*create_function)();
+
+public:
+	// compatible with embree3 rays
+	struct __aligned(16) Ray {
+		const static unsigned int INVALID_GEOMETRY_ID = ((unsigned int)-1); // from rtcore_common.h
+
+		/*! Default construction does nothing. */
+		_FORCE_INLINE_ Ray() :
+				geomID(INVALID_GEOMETRY_ID) {}
+
+		/*! Constructs a ray from origin, direction, and ray segment. Near
+		 *  has to be smaller than far. */
+		_FORCE_INLINE_ Ray(const Vector3 &org,
+				const Vector3 &dir,
+				float tnear = 0.0f,
+				float tfar = INFINITY) :
+				org(org),
+				tnear(tnear),
+				dir(dir),
+				time(0.0f),
+				tfar(tfar),
+				mask(-1),
+				u(0.0),
+				v(0.0),
+				primID(INVALID_GEOMETRY_ID),
+				geomID(INVALID_GEOMETRY_ID),
+				instID(INVALID_GEOMETRY_ID) {}
+
+		/*! Tests if we hit something. */
+		_FORCE_INLINE_ explicit operator bool() const { return geomID != INVALID_GEOMETRY_ID; }
+
+	public:
+		Vector3 org; //!< Ray origin + tnear
+		float tnear; //!< Start of ray segment
+		Vector3 dir; //!< Ray direction + tfar
+		float time; //!< Time of this ray for motion blur.
+		float tfar; //!< End of ray segment
+		unsigned int mask; //!< used to mask out objects during traversal
+		unsigned int id; //!< ray ID
+		unsigned int flags; //!< ray flags
+
+		Vector3 normal; //!< Not normalized geometry normal
+		float u; //!< Barycentric u coordinate of hit
+		float v; //!< Barycentric v coordinate of hit
+		unsigned int primID; //!< primitive ID
+		unsigned int geomID; //!< geometry ID
+		unsigned int instID; //!< instance ID
+	};
+
+	virtual bool intersect(Ray &p_ray) = 0;
+	virtual void intersect(Vector<Ray> &r_rays) = 0;
+
+	virtual void add_mesh(const PackedVector3Array &p_vertices, const PackedInt32Array &p_indices, unsigned int p_id) = 0;
+	virtual void commit() = 0;
+
+	virtual void set_mesh_filter(const Set<int> &p_mesh_ids) = 0;
+	virtual void clear_mesh_filter() = 0;
+
+	static Ref<StaticRaycaster> create();
+};
+
+#endif // STATIC_RAYCASTER_H
diff --git a/editor/import/resource_importer_scene.cpp b/editor/import/resource_importer_scene.cpp
index c48d9bb117..2c9bc7dadf 100644
--- a/editor/import/resource_importer_scene.cpp
+++ b/editor/import/resource_importer_scene.cpp
@@ -980,6 +980,8 @@ void ResourceImporterScene::get_internal_import_options(InternalImportCategory p
 			r_options->push_back(ImportOption(PropertyInfo(Variant::INT, "generate/shadow_meshes", PROPERTY_HINT_ENUM, "Default,Enable,Disable"), 0));
 			r_options->push_back(ImportOption(PropertyInfo(Variant::INT, "generate/lightmap_uv", PROPERTY_HINT_ENUM, "Default,Enable,Disable"), 0));
 			r_options->push_back(ImportOption(PropertyInfo(Variant::INT, "generate/lods", PROPERTY_HINT_ENUM, "Default,Enable,Disable"), 0));
+			r_options->push_back(ImportOption(PropertyInfo(Variant::FLOAT, "lods/normal_split_angle", PROPERTY_HINT_RANGE, "0,180,0.1,degrees"), 25.0f));
+			r_options->push_back(ImportOption(PropertyInfo(Variant::FLOAT, "lods/normal_merge_angle", PROPERTY_HINT_RANGE, "0,180,0.1,degrees"), 60.0f));
 		} break;
 		case INTERNAL_IMPORT_CATEGORY_MATERIAL: {
 			r_options->push_back(ImportOption(PropertyInfo(Variant::BOOL, "use_external/enabled", PROPERTY_HINT_NONE, "", PROPERTY_USAGE_DEFAULT | PROPERTY_USAGE_UPDATE_ALL_IF_MODIFIED), false));
@@ -1259,6 +1261,8 @@ void ResourceImporterScene::_generate_meshes(Node *p_node, const Dictionary &p_m
 				//do mesh processing
 
 				bool generate_lods = p_generate_lods;
+				float split_angle = 25.0f;
+				float merge_angle = 60.0f;
 				bool create_shadow_meshes = p_create_shadow_meshes;
 				bool bake_lightmaps = p_light_bake_mode == LIGHT_BAKE_STATIC_LIGHTMAPS;
 				String save_to_file;
@@ -1301,6 +1305,14 @@ void ResourceImporterScene::_generate_meshes(Node *p_node, const Dictionary &p_m
 						}
 					}
 
+					if (mesh_settings.has("lods/normal_split_angle")) {
+						split_angle = mesh_settings["lods/normal_split_angle"];
+					}
+
+					if (mesh_settings.has("lods/normal_merge_angle")) {
+						merge_angle = mesh_settings["lods/normal_merge_angle"];
+					}
+
 					if (mesh_settings.has("save_to_file/enabled") && bool(mesh_settings["save_to_file/enabled"]) && mesh_settings.has("save_to_file/path")) {
 						save_to_file = mesh_settings["save_to_file/path"];
 						if (!save_to_file.is_resource_file()) {
@@ -1310,8 +1322,9 @@ void ResourceImporterScene::_generate_meshes(Node *p_node, const Dictionary &p_m
 				}
 
 				if (generate_lods) {
-					src_mesh_node->get_mesh()->generate_lods();
+					src_mesh_node->get_mesh()->generate_lods(merge_angle, split_angle);
 				}
+
 				if (create_shadow_meshes) {
 					src_mesh_node->get_mesh()->create_shadow_mesh();
 				}
diff --git a/editor/import/scene_importer_mesh.cpp b/editor/import/scene_importer_mesh.cpp
index 5e6dd08e79..f83a7d046a 100644
--- a/editor/import/scene_importer_mesh.cpp
+++ b/editor/import/scene_importer_mesh.cpp
@@ -30,11 +30,99 @@
 
 #include "scene_importer_mesh.h"
 
-#include "core/math/math_defs.h"
+#include "core/math/random_pcg.h"
+#include "core/math/static_raycaster.h"
 #include "scene/resources/surface_tool.h"
 
 #include <cstdint>
 
+void EditorSceneImporterMesh::Surface::split_normals(const LocalVector<int> &p_indices, const LocalVector<Vector3> &p_normals) {
+	ERR_FAIL_COND(arrays.size() != RS::ARRAY_MAX);
+
+	const PackedVector3Array &vertices = arrays[RS::ARRAY_VERTEX];
+	int current_vertex_count = vertices.size();
+	int new_vertex_count = p_indices.size();
+	int final_vertex_count = current_vertex_count + new_vertex_count;
+	const int *indices_ptr = p_indices.ptr();
+
+	for (int i = 0; i < arrays.size(); i++) {
+		if (i == RS::ARRAY_INDEX) {
+			continue;
+		}
+
+		if (arrays[i].get_type() == Variant::NIL) {
+			continue;
+		}
+
+		switch (arrays[i].get_type()) {
+			case Variant::PACKED_VECTOR3_ARRAY: {
+				PackedVector3Array data = arrays[i];
+				data.resize(final_vertex_count);
+				Vector3 *data_ptr = data.ptrw();
+				if (i == RS::ARRAY_NORMAL) {
+					const Vector3 *normals_ptr = p_normals.ptr();
+					memcpy(&data_ptr[current_vertex_count], normals_ptr, sizeof(Vector3) * new_vertex_count);
+				} else {
+					for (int j = 0; j < new_vertex_count; j++) {
+						data_ptr[current_vertex_count + j] = data_ptr[indices_ptr[j]];
+					}
+				}
+				arrays[i] = data;
+			} break;
+			case Variant::PACKED_VECTOR2_ARRAY: {
+				PackedVector2Array data = arrays[i];
+				data.resize(final_vertex_count);
+				Vector2 *data_ptr = data.ptrw();
+				for (int j = 0; j < new_vertex_count; j++) {
+					data_ptr[current_vertex_count + j] = data_ptr[indices_ptr[j]];
+				}
+				arrays[i] = data;
+			} break;
+			case Variant::PACKED_FLOAT32_ARRAY: {
+				PackedFloat32Array data = arrays[i];
+				int elements = data.size() / current_vertex_count;
+				data.resize(final_vertex_count * elements);
+				float *data_ptr = data.ptrw();
+				for (int j = 0; j < new_vertex_count; j++) {
+					memcpy(&data_ptr[(current_vertex_count + j) * elements], &data_ptr[indices_ptr[j] * elements], sizeof(float) * elements);
+				}
+				arrays[i] = data;
+			} break;
+			case Variant::PACKED_INT32_ARRAY: {
+				PackedInt32Array data = arrays[i];
+				int elements = data.size() / current_vertex_count;
+				data.resize(final_vertex_count * elements);
+				int32_t *data_ptr = data.ptrw();
+				for (int j = 0; j < new_vertex_count; j++) {
+					memcpy(&data_ptr[(current_vertex_count + j) * elements], &data_ptr[indices_ptr[j] * elements], sizeof(int32_t) * elements);
+				}
+				arrays[i] = data;
+			} break;
+			case Variant::PACKED_BYTE_ARRAY: {
+				PackedByteArray data = arrays[i];
+				int elements = data.size() / current_vertex_count;
+				data.resize(final_vertex_count * elements);
+				uint8_t *data_ptr = data.ptrw();
+				for (int j = 0; j < new_vertex_count; j++) {
+					memcpy(&data_ptr[(current_vertex_count + j) * elements], &data_ptr[indices_ptr[j] * elements], sizeof(uint8_t) * elements);
+				}
+				arrays[i] = data;
+			} break;
+			case Variant::PACKED_COLOR_ARRAY: {
+				PackedColorArray data = arrays[i];
+				data.resize(final_vertex_count);
+				Color *data_ptr = data.ptrw();
+				for (int j = 0; j < new_vertex_count; j++) {
+					data_ptr[current_vertex_count + j] = data_ptr[indices_ptr[j]];
+				}
+			} break;
+			default: {
+				ERR_FAIL_MSG("Uhandled array type.");
+			} break;
+		}
+	}
+}
+
 void EditorSceneImporterMesh::add_blend_shape(const String &p_name) {
 	ERR_FAIL_COND(surfaces.size() > 0);
 	blend_shapes.push_back(p_name);
@@ -157,29 +245,14 @@ void EditorSceneImporterMesh::set_surface_material(int p_surface, const Ref<Mate
 	mesh.unref();
 }
 
-Basis EditorSceneImporterMesh::compute_rotation_matrix_from_ortho_6d(Vector3 p_x_raw, Vector3 p_y_raw) {
-	Vector3 x = p_x_raw.normalized();
-	Vector3 z = x.cross(p_y_raw);
-	z = z.normalized();
-	Vector3 y = z.cross(x);
-	Basis basis;
-	basis.set_axis(Vector3::AXIS_X, x);
-	basis.set_axis(Vector3::AXIS_Y, y);
-	basis.set_axis(Vector3::AXIS_Z, z);
-	return basis;
-}
-
-void EditorSceneImporterMesh::generate_lods() {
-	if (!SurfaceTool::simplify_func) {
-		return;
-	}
+void EditorSceneImporterMesh::generate_lods(float p_normal_merge_angle, float p_normal_split_angle) {
 	if (!SurfaceTool::simplify_scale_func) {
 		return;
 	}
-	if (!SurfaceTool::simplify_sloppy_func) {
+	if (!SurfaceTool::simplify_with_attrib_func) {
 		return;
 	}
-	if (!SurfaceTool::simplify_with_attrib_func) {
+	if (!SurfaceTool::optimize_vertex_cache_func) {
 		return;
 	}
 
@@ -190,67 +263,343 @@ void EditorSceneImporterMesh::generate_lods() {
 
 		surfaces.write[i].lods.clear();
 		Vector<Vector3> vertices = surfaces[i].arrays[RS::ARRAY_VERTEX];
-		Vector<int> indices = surfaces[i].arrays[RS::ARRAY_INDEX];
-		if (indices.size() == 0) {
+		PackedInt32Array indices = surfaces[i].arrays[RS::ARRAY_INDEX];
+		Vector<Vector3> normals = surfaces[i].arrays[RS::ARRAY_NORMAL];
+		Vector<Vector2> uvs = surfaces[i].arrays[RS::ARRAY_TEX_UV];
+
+		unsigned int index_count = indices.size();
+		unsigned int vertex_count = vertices.size();
+
+		if (index_count == 0) {
 			continue; //no lods if no indices
 		}
-		Vector<Vector3> normals = surfaces[i].arrays[RS::ARRAY_NORMAL];
-		uint32_t vertex_count = vertices.size();
+
 		const Vector3 *vertices_ptr = vertices.ptr();
-		Vector<float> attributes;
-		Vector<float> normal_weights;
-		int32_t attribute_count = 6;
-		if (normals.size()) {
-			attributes.resize(normals.size() * attribute_count);
-			for (int32_t normal_i = 0; normal_i < normals.size(); normal_i++) {
-				Basis basis;
-				basis.set_euler(normals[normal_i]);
-				Vector3 basis_x = basis.get_axis(0);
-				Vector3 basis_y = basis.get_axis(1);
-				basis = compute_rotation_matrix_from_ortho_6d(basis_x, basis_y);
-				basis_x = basis.get_axis(0);
-				basis_y = basis.get_axis(1);
-				attributes.write[normal_i * attribute_count + 0] = basis_x.x;
-				attributes.write[normal_i * attribute_count + 1] = basis_x.y;
-				attributes.write[normal_i * attribute_count + 2] = basis_x.z;
-				attributes.write[normal_i * attribute_count + 3] = basis_y.x;
-				attributes.write[normal_i * attribute_count + 4] = basis_y.y;
-				attributes.write[normal_i * attribute_count + 5] = basis_y.z;
-			}
-			normal_weights.resize(vertex_count);
-			for (int32_t weight_i = 0; weight_i < normal_weights.size(); weight_i++) {
-				normal_weights.write[weight_i] = 1.0;
+		const int *indices_ptr = indices.ptr();
+
+		if (normals.is_empty()) {
+			normals.resize(vertices.size());
+			Vector3 *n_ptr = normals.ptrw();
+			for (unsigned int j = 0; j < index_count; j += 3) {
+				const Vector3 &v0 = vertices_ptr[indices_ptr[j + 0]];
+				const Vector3 &v1 = vertices_ptr[indices_ptr[j + 1]];
+				const Vector3 &v2 = vertices_ptr[indices_ptr[j + 2]];
+				Vector3 n = vec3_cross(v0 - v2, v0 - v1).normalized();
+				n_ptr[j + 0] = n;
+				n_ptr[j + 1] = n;
+				n_ptr[j + 2] = n;
 			}
-		} else {
-			attribute_count = 0;
 		}
-		const int min_indices = 10;
-		const float error_tolerance = 1.44224'95703; // Cube root of 3
-		const float threshold = 1.0 / error_tolerance;
-		int index_target = indices.size() * threshold;
-		float max_mesh_error_percentage = 1e0f;
+
+		float normal_merge_threshold = Math::cos(Math::deg2rad(p_normal_merge_angle));
+		float normal_pre_split_threshold = Math::cos(Math::deg2rad(MIN(180.0f, p_normal_split_angle * 2.0f)));
+		float normal_split_threshold = Math::cos(Math::deg2rad(p_normal_split_angle));
+		const Vector3 *normals_ptr = normals.ptr();
+
+		Map<Vector3, LocalVector<Pair<int, int>>> unique_vertices;
+
+		LocalVector<int> vertex_remap;
+		LocalVector<int> vertex_inverse_remap;
+		LocalVector<Vector3> merged_vertices;
+		LocalVector<Vector3> merged_normals;
+		LocalVector<int> merged_normals_counts;
+		const Vector2 *uvs_ptr = uvs.ptr();
+
+		for (unsigned int j = 0; j < vertex_count; j++) {
+			const Vector3 &v = vertices_ptr[j];
+			const Vector3 &n = normals_ptr[j];
+
+			Map<Vector3, LocalVector<Pair<int, int>>>::Element *E = unique_vertices.find(v);
+
+			if (E) {
+				const LocalVector<Pair<int, int>> &close_verts = E->get();
+
+				bool found = false;
+				for (unsigned int k = 0; k < close_verts.size(); k++) {
+					const Pair<int, int> &idx = close_verts[k];
+
+					// TODO check more attributes?
+					if ((!uvs_ptr || uvs_ptr[j].distance_squared_to(uvs_ptr[idx.second]) < CMP_EPSILON2) && normals[idx.second].dot(n) > normal_merge_threshold) {
+						vertex_remap.push_back(idx.first);
+						merged_normals[idx.first] += normals[idx.second];
+						merged_normals_counts[idx.first]++;
+						found = true;
+						break;
+					}
+				}
+
+				if (!found) {
+					int vcount = merged_vertices.size();
+					unique_vertices[v].push_back(Pair<int, int>(vcount, j));
+					vertex_inverse_remap.push_back(j);
+					merged_vertices.push_back(v);
+					vertex_remap.push_back(vcount);
+					merged_normals.push_back(normals_ptr[j]);
+					merged_normals_counts.push_back(1);
+				}
+			} else {
+				int vcount = merged_vertices.size();
+				unique_vertices[v] = LocalVector<Pair<int, int>>();
+				unique_vertices[v].push_back(Pair<int, int>(vcount, j));
+				vertex_inverse_remap.push_back(j);
+				merged_vertices.push_back(v);
+				vertex_remap.push_back(vcount);
+				merged_normals.push_back(normals_ptr[j]);
+				merged_normals_counts.push_back(1);
+			}
+		}
+
+		LocalVector<int> merged_indices;
+		merged_indices.resize(index_count);
+		for (unsigned int j = 0; j < index_count; j++) {
+			merged_indices[j] = vertex_remap[indices[j]];
+		}
+
+		unsigned int merged_vertex_count = merged_vertices.size();
+		const Vector3 *merged_vertices_ptr = merged_vertices.ptr();
+		const int32_t *merged_indices_ptr = merged_indices.ptr();
+
+		{
+			const int *counts_ptr = merged_normals_counts.ptr();
+			Vector3 *merged_normals_ptrw = merged_normals.ptr();
+			for (unsigned int j = 0; j < merged_vertex_count; j++) {
+				merged_normals_ptrw[j] /= counts_ptr[j];
+			}
+		}
+
+		LocalVector<float> normal_weights;
+		normal_weights.resize(merged_vertex_count);
+		for (unsigned int j = 0; j < merged_vertex_count; j++) {
+			normal_weights[j] = 2.0; // Give some weight to normal preservation, may be worth exposing as an import setting
+		}
+
+		const float max_mesh_error = FLT_MAX; // We don't want to limit by error, just by index target
+		float scale = SurfaceTool::simplify_scale_func((const float *)merged_vertices_ptr, merged_vertex_count, sizeof(Vector3));
 		float mesh_error = 0.0f;
-		float scale = SurfaceTool::simplify_scale_func((const float *)vertices_ptr, vertex_count, sizeof(Vector3));
-		while (index_target > min_indices) {
-			Vector<int> new_indices;
-			new_indices.resize(indices.size());
-			size_t new_len = SurfaceTool::simplify_with_attrib_func((unsigned int *)new_indices.ptrw(), (const unsigned int *)indices.ptr(), indices.size(), (const float *)vertices_ptr, vertex_count, sizeof(Vector3), index_target, max_mesh_error_percentage, &mesh_error, (float *)attributes.ptrw(), normal_weights.ptrw(), attribute_count);
-			if ((int)new_len > (index_target * error_tolerance)) {
-				break;
+
+		unsigned int index_target = 12; // Start with the smallest target, 4 triangles
+		unsigned int last_index_count = 0;
+
+		int split_vertex_count = vertex_count;
+		LocalVector<Vector3> split_vertex_normals;
+		LocalVector<int> split_vertex_indices;
+		split_vertex_normals.reserve(index_count / 3);
+		split_vertex_indices.reserve(index_count / 3);
+
+		RandomPCG pcg;
+		pcg.seed(123456789); // Keep seed constant across imports
+
+		Ref<StaticRaycaster> raycaster = StaticRaycaster::create();
+		if (raycaster.is_valid()) {
+			raycaster->add_mesh(vertices, indices, 0);
+			raycaster->commit();
+		}
+
+		while (index_target < index_count) {
+			PackedInt32Array new_indices;
+			new_indices.resize(index_count);
+
+			size_t new_index_count = SurfaceTool::simplify_with_attrib_func((unsigned int *)new_indices.ptrw(), (const uint32_t *)merged_indices_ptr, index_count, (const float *)merged_vertices_ptr, merged_vertex_count, sizeof(Vector3), index_target, max_mesh_error, &mesh_error, (float *)merged_normals.ptr(), normal_weights.ptr(), 3);
+
+			if (new_index_count < last_index_count * 1.5f) {
+				index_target = index_target * 1.5f;
+				continue;
 			}
-			Surface::LOD lod;
-			lod.distance = mesh_error * scale;
-			if (Math::is_zero_approx(mesh_error)) {
+
+			if (new_index_count <= 0 || (new_index_count >= (index_count * 0.75f))) {
 				break;
 			}
-			if (new_len <= 0) {
-				break;
+
+			new_indices.resize(new_index_count);
+
+			LocalVector<LocalVector<int>> vertex_corners;
+			vertex_corners.resize(vertex_count);
+			{
+				int *ptrw = new_indices.ptrw();
+				for (unsigned int j = 0; j < new_index_count; j++) {
+					const int &remapped = vertex_inverse_remap[ptrw[j]];
+					vertex_corners[remapped].push_back(j);
+					ptrw[j] = remapped;
+				}
 			}
-			new_indices.resize(new_len);
+
+			if (raycaster.is_valid()) {
+				float error_factor = 1.0f / (scale * MAX(mesh_error, 0.15));
+				const float ray_bias = 0.05;
+				float ray_length = ray_bias + mesh_error * scale * 3.0f;
+
+				Vector<StaticRaycaster::Ray> rays;
+				LocalVector<Vector2> ray_uvs;
+
+				int32_t *new_indices_ptr = new_indices.ptrw();
+
+				int current_ray_count = 0;
+				for (unsigned int j = 0; j < new_index_count; j += 3) {
+					const Vector3 &v0 = vertices_ptr[new_indices_ptr[j + 0]];
+					const Vector3 &v1 = vertices_ptr[new_indices_ptr[j + 1]];
+					const Vector3 &v2 = vertices_ptr[new_indices_ptr[j + 2]];
+					Vector3 face_normal = vec3_cross(v0 - v2, v0 - v1);
+					float face_area = face_normal.length(); // Actually twice the face area, since it's the same error_factor on all faces, we don't care
+
+					Vector3 dir = face_normal / face_area;
+					int ray_count = CLAMP(5.0 * face_area * error_factor, 16, 64);
+
+					rays.resize(current_ray_count + ray_count);
+					StaticRaycaster::Ray *rays_ptr = rays.ptrw();
+
+					ray_uvs.resize(current_ray_count + ray_count);
+					Vector2 *ray_uvs_ptr = ray_uvs.ptr();
+
+					for (int k = 0; k < ray_count; k++) {
+						float u = pcg.randf();
+						float v = pcg.randf();
+
+						if (u + v >= 1.0f) {
+							u = 1.0f - u;
+							v = 1.0f - v;
+						}
+
+						u = 0.9f * u + 0.05f / 3.0f; // Give barycentric coordinates some padding, we don't want to sample right on the edge
+						v = 0.9f * v + 0.05f / 3.0f; // v = (v - one_third) * 0.95f + one_third;
+						float w = 1.0f - u - v;
+
+						Vector3 org = v0 * w + v1 * u + v2 * v;
+						org -= dir * ray_bias;
+						rays_ptr[current_ray_count + k] = StaticRaycaster::Ray(org, dir, 0.0f, ray_length);
+						rays_ptr[current_ray_count + k].id = j / 3;
+						ray_uvs_ptr[current_ray_count + k] = Vector2(u, v);
+					}
+
+					current_ray_count += ray_count;
+				}
+
+				raycaster->intersect(rays);
+
+				LocalVector<Vector3> ray_normals;
+				LocalVector<float> ray_normal_weights;
+
+				ray_normals.resize(new_index_count);
+				ray_normal_weights.resize(new_index_count);
+
+				for (unsigned int j = 0; j < new_index_count; j++) {
+					ray_normal_weights[j] = 0.0f;
+				}
+
+				const StaticRaycaster::Ray *rp = rays.ptr();
+				for (int j = 0; j < rays.size(); j++) {
+					if (rp[j].geomID != 0) { // Ray missed
+						continue;
+					}
+
+					if (rp[j].normal.normalized().dot(rp[j].dir) > 0.0f) { // Hit a back face.
+						continue;
+					}
+
+					const float &u = rp[j].u;
+					const float &v = rp[j].v;
+					const float w = 1.0f - u - v;
+
+					const unsigned int &hit_tri_id = rp[j].primID;
+					const unsigned int &orig_tri_id = rp[j].id;
+
+					const Vector3 &n0 = normals_ptr[indices_ptr[hit_tri_id * 3 + 0]];
+					const Vector3 &n1 = normals_ptr[indices_ptr[hit_tri_id * 3 + 1]];
+					const Vector3 &n2 = normals_ptr[indices_ptr[hit_tri_id * 3 + 2]];
+					Vector3 normal = n0 * w + n1 * u + n2 * v;
+
+					Vector2 orig_uv = ray_uvs[j];
+					float orig_bary[3] = { 1.0f - orig_uv.x - orig_uv.y, orig_uv.x, orig_uv.y };
+					for (int k = 0; k < 3; k++) {
+						int idx = orig_tri_id * 3 + k;
+						float weight = orig_bary[k];
+						ray_normals[idx] += normal * weight;
+						ray_normal_weights[idx] += weight;
+					}
+				}
+
+				for (unsigned int j = 0; j < new_index_count; j++) {
+					if (ray_normal_weights[j] < 1.0f) { // Not enough data, the new normal would be just a bad guess
+						ray_normals[j] = Vector3();
+					} else {
+						ray_normals[j] /= ray_normal_weights[j];
+					}
+				}
+
+				LocalVector<LocalVector<int>> normal_group_indices;
+				LocalVector<Vector3> normal_group_averages;
+				normal_group_indices.reserve(24);
+				normal_group_averages.reserve(24);
+
+				for (unsigned int j = 0; j < vertex_count; j++) {
+					const LocalVector<int> &corners = vertex_corners[j];
+					const Vector3 &vertex_normal = normals_ptr[j];
+
+					for (unsigned int k = 0; k < corners.size(); k++) {
+						const int &corner_idx = corners[k];
+						const Vector3 &ray_normal = ray_normals[corner_idx];
+
+						if (ray_normal.length_squared() < CMP_EPSILON2) {
+							continue;
+						}
+
+						bool found = false;
+						for (unsigned int l = 0; l < normal_group_indices.size(); l++) {
+							LocalVector<int> &group_indices = normal_group_indices[l];
+							Vector3 n = normal_group_averages[l] / group_indices.size();
+							if (n.dot(ray_normal) > normal_pre_split_threshold) {
+								found = true;
+								group_indices.push_back(corner_idx);
+								normal_group_averages[l] += ray_normal;
+								break;
+							}
+						}
+
+						if (!found) {
+							LocalVector<int> new_group;
+							new_group.push_back(corner_idx);
+							normal_group_indices.push_back(new_group);
+							normal_group_averages.push_back(ray_normal);
+						}
+					}
+
+					for (unsigned int k = 0; k < normal_group_indices.size(); k++) {
+						LocalVector<int> &group_indices = normal_group_indices[k];
+						Vector3 n = normal_group_averages[k] / group_indices.size();
+
+						if (vertex_normal.dot(n) < normal_split_threshold) {
+							split_vertex_indices.push_back(j);
+							split_vertex_normals.push_back(n);
+							int new_idx = split_vertex_count++;
+							for (unsigned int l = 0; l < group_indices.size(); l++) {
+								new_indices_ptr[group_indices[l]] = new_idx;
+							}
+						}
+					}
+
+					normal_group_indices.clear();
+					normal_group_averages.clear();
+				}
+			}
+
+			Surface::LOD lod;
+			lod.distance = MAX(mesh_error * scale, CMP_EPSILON2);
 			lod.indices = new_indices;
-			print_line("Lod " + itos(surfaces.write[i].lods.size()) + " begin with " + itos(indices.size() / 3) + " triangles and shoot for " + itos(index_target / 3) + " triangles. Got " + itos(new_len / 3) + " triangles. Lod screen ratio " + rtos(lod.distance));
 			surfaces.write[i].lods.push_back(lod);
-			index_target *= threshold;
+			index_target = MAX(new_index_count, index_target) * 2;
+			last_index_count = new_index_count;
+
+			if (mesh_error == 0.0f) {
+				break;
+			}
+		}
+
+		surfaces.write[i].split_normals(split_vertex_indices, split_vertex_normals);
+		surfaces.write[i].lods.sort_custom<Surface::LODComparator>();
+
+		for (int j = 0; j < surfaces.write[i].lods.size(); j++) {
+			Surface::LOD &lod = surfaces.write[i].lods.write[j];
+			unsigned int *lod_indices_ptr = (unsigned int *)lod.indices.ptrw();
+			SurfaceTool::optimize_vertex_cache_func(lod_indices_ptr, lod_indices_ptr, lod.indices.size(), split_vertex_count);
 		}
 	}
 }
@@ -347,7 +696,7 @@ void EditorSceneImporterMesh::create_shadow_mesh() {
 			Map<Vector3, int> unique_vertices;
 			const Vector3 *vptr = vertices.ptr();
 			for (int j = 0; j < vertex_count; j++) {
-				Vector3 v = vptr[j];
+				const Vector3 &v = vptr[j];
 
 				Map<Vector3, int>::Element *E = unique_vertices.find(v);
 
@@ -397,9 +746,9 @@ void EditorSceneImporterMesh::create_shadow_mesh() {
 				index_wptr = new_indices.ptrw();
 
 				for (int k = 0; k < index_count; k++) {
-					int index = index_rptr[j];
+					int index = index_rptr[k];
 					ERR_FAIL_INDEX(index, vertex_count);
-					index_wptr[j] = vertex_remap[index];
+					index_wptr[k] = vertex_remap[index];
 				}
 
 				lods[surfaces[i].lods[j].distance] = new_indices;
@@ -436,9 +785,9 @@ void EditorSceneImporterMesh::_set_data(const Dictionary &p_data) {
 			if (s.has("lods")) {
 				lods = s["lods"];
 			}
-			Array blend_shapes;
-			if (s.has("blend_shapes")) {
-				blend_shapes = s["blend_shapes"];
+			Array b_shapes;
+			if (s.has("b_shapes")) {
+				b_shapes = s["b_shapes"];
 			}
 			Ref<Material> material;
 			if (s.has("material")) {
@@ -448,7 +797,7 @@ void EditorSceneImporterMesh::_set_data(const Dictionary &p_data) {
 			if (s.has("flags")) {
 				flags = s["flags"];
 			}
-			add_surface(prim, arr, blend_shapes, lods, material, name, flags);
+			add_surface(prim, arr, b_shapes, lods, material, name, flags);
 		}
 	}
 }
diff --git a/editor/import/scene_importer_mesh.h b/editor/import/scene_importer_mesh.h
index d32b1fdf74..111b191cae 100644
--- a/editor/import/scene_importer_mesh.h
+++ b/editor/import/scene_importer_mesh.h
@@ -32,6 +32,7 @@
 #define EDITOR_SCENE_IMPORTER_MESH_H
 
 #include "core/io/resource.h"
+#include "core/templates/local_vector.h"
 #include "scene/resources/concave_polygon_shape_3d.h"
 #include "scene/resources/convex_polygon_shape_3d.h"
 #include "scene/resources/mesh.h"
@@ -55,12 +56,20 @@ class EditorSceneImporterMesh : public Resource {
 		Vector<BlendShape> blend_shape_data;
 		struct LOD {
 			Vector<int> indices;
-			float distance;
+			float distance = 0.0f;
 		};
 		Vector<LOD> lods;
 		Ref<Material> material;
 		String name;
 		uint32_t flags = 0;
+
+		struct LODComparator {
+			_FORCE_INLINE_ bool operator()(const LOD &l, const LOD &r) const {
+				return l.distance < r.distance;
+			}
+		};
+
+		void split_normals(const LocalVector<int> &p_indices, const LocalVector<Vector3> &p_normals);
 	};
 	Vector<Surface> surfaces;
 	Vector<String> blend_shapes;
@@ -71,7 +80,6 @@ class EditorSceneImporterMesh : public Resource {
 	Ref<EditorSceneImporterMesh> shadow_mesh;
 
 	Size2i lightmap_size_hint;
-	Basis compute_rotation_matrix_from_ortho_6d(Vector3 p_x_raw, Vector3 y_raw);
 
 protected:
 	void _set_data(const Dictionary &p_data);
@@ -103,7 +111,7 @@ public:
 
 	void set_surface_material(int p_surface, const Ref<Material> &p_material);
 
-	void generate_lods();
+	void generate_lods(float p_normal_merge_angle, float p_normal_split_angle);
 
 	void create_shadow_mesh();
 	Ref<EditorSceneImporterMesh> get_shadow_mesh() const;
diff --git a/editor/plugins/node_3d_editor_plugin.cpp b/editor/plugins/node_3d_editor_plugin.cpp
index d52a633fed..f5d3352299 100644
--- a/editor/plugins/node_3d_editor_plugin.cpp
+++ b/editor/plugins/node_3d_editor_plugin.cpp
@@ -2601,6 +2601,9 @@ void Node3DEditorViewport::_project_settings_changed() {
 
 	const bool use_occlusion_culling = GLOBAL_GET("rendering/occlusion_culling/use_occlusion_culling");
 	viewport->set_use_occlusion_culling(use_occlusion_culling);
+
+	const float lod_threshold = GLOBAL_GET("rendering/mesh_lod/lod_change/threshold_pixels");
+	viewport->set_lod_threshold(lod_threshold);
 }
 
 void Node3DEditorViewport::_notification(int p_what) {
diff --git a/modules/raycast/lightmap_raycaster.cpp b/modules/raycast/lightmap_raycaster.cpp
index 0583acc119..fdcf509da8 100644
--- a/modules/raycast/lightmap_raycaster.cpp
+++ b/modules/raycast/lightmap_raycaster.cpp
@@ -168,7 +168,7 @@ void LightmapRaycasterEmbree::clear_mesh_filter() {
 	filter_meshes.clear();
 }
 
-void embree_error_handler(void *p_user_data, RTCError p_code, const char *p_str) {
+void embree_lm_error_handler(void *p_user_data, RTCError p_code, const char *p_str) {
 	print_error("Embree error: " + String(p_str));
 }
 
@@ -179,16 +179,11 @@ LightmapRaycasterEmbree::LightmapRaycasterEmbree() {
 #endif
 
 	embree_device = rtcNewDevice(nullptr);
-	rtcSetDeviceErrorFunction(embree_device, &embree_error_handler, nullptr);
+	rtcSetDeviceErrorFunction(embree_device, &embree_lm_error_handler, nullptr);
 	embree_scene = rtcNewScene(embree_device);
 }
 
 LightmapRaycasterEmbree::~LightmapRaycasterEmbree() {
-#ifdef __SSE2__
-	_MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_OFF);
-	_MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_OFF);
-#endif
-
 	if (embree_scene != nullptr) {
 		rtcReleaseScene(embree_scene);
 	}
diff --git a/modules/raycast/lightmap_raycaster.h b/modules/raycast/lightmap_raycaster.h
index 4c3de27837..290b0a1cf3 100644
--- a/modules/raycast/lightmap_raycaster.h
+++ b/modules/raycast/lightmap_raycaster.h
@@ -30,9 +30,9 @@
 
 #ifdef TOOLS_ENABLED
 
+#include "core/io/image.h"
 #include "core/object/object.h"
 #include "scene/3d/lightmapper.h"
-#include "scene/resources/mesh.h"
 
 #include <embree3/rtcore.h>
 
diff --git a/modules/raycast/register_types.cpp b/modules/raycast/register_types.cpp
index 78ca91309f..ed99e635e1 100644
--- a/modules/raycast/register_types.cpp
+++ b/modules/raycast/register_types.cpp
@@ -32,12 +32,14 @@
 
 #include "lightmap_raycaster.h"
 #include "raycast_occlusion_cull.h"
+#include "static_raycaster.h"
 
 RaycastOcclusionCull *raycast_occlusion_cull = nullptr;
 
 void register_raycast_types() {
 #ifdef TOOLS_ENABLED
 	LightmapRaycasterEmbree::make_default_raycaster();
+	StaticRaycasterEmbree::make_default_raycaster();
 #endif
 	raycast_occlusion_cull = memnew(RaycastOcclusionCull);
 }
@@ -46,4 +48,7 @@ void unregister_raycast_types() {
 	if (raycast_occlusion_cull) {
 		memdelete(raycast_occlusion_cull);
 	}
+#ifdef TOOLS_ENABLED
+	StaticRaycasterEmbree::free();
+#endif
 }
diff --git a/modules/raycast/static_raycaster.cpp b/modules/raycast/static_raycaster.cpp
new file mode 100644
index 0000000000..2ba65eebf8
--- /dev/null
+++ b/modules/raycast/static_raycaster.cpp
@@ -0,0 +1,137 @@
+/*************************************************************************/
+/*  static_raycaster.cpp                                                 */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2021 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2021 Godot Engine contributors (cf. AUTHORS.md).   */
+/*                                                                       */
+/* 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.                */
+/*************************************************************************/
+
+#ifdef TOOLS_ENABLED
+
+#include "static_raycaster.h"
+
+#ifdef __SSE2__
+#include <pmmintrin.h>
+#endif
+
+RTCDevice StaticRaycasterEmbree::embree_device;
+
+StaticRaycaster *StaticRaycasterEmbree::create_embree_raycaster() {
+	return memnew(StaticRaycasterEmbree);
+}
+
+void StaticRaycasterEmbree::make_default_raycaster() {
+	create_function = create_embree_raycaster;
+}
+
+void StaticRaycasterEmbree::free() {
+	if (embree_device) {
+		rtcReleaseDevice(embree_device);
+	}
+}
+
+bool StaticRaycasterEmbree::intersect(Ray &r_ray) {
+	RTCIntersectContext context;
+	rtcInitIntersectContext(&context);
+	rtcIntersect1(embree_scene, &context, (RTCRayHit *)&r_ray);
+	return r_ray.geomID != RTC_INVALID_GEOMETRY_ID;
+}
+
+void StaticRaycasterEmbree::intersect(Vector<Ray> &r_rays) {
+	Ray *rays = r_rays.ptrw();
+	for (int i = 0; i < r_rays.size(); ++i) {
+		intersect(rays[i]);
+	}
+}
+
+void StaticRaycasterEmbree::add_mesh(const PackedVector3Array &p_vertices, const PackedInt32Array &p_indices, unsigned int p_id) {
+	RTCGeometry embree_mesh = rtcNewGeometry(embree_device, RTC_GEOMETRY_TYPE_TRIANGLE);
+
+	int vertex_count = p_vertices.size();
+
+	Vector3 *embree_vertices = (Vector3 *)rtcSetNewGeometryBuffer(embree_mesh, RTC_BUFFER_TYPE_VERTEX, 0, RTC_FORMAT_FLOAT3, sizeof(Vector3), vertex_count);
+	memcpy(embree_vertices, p_vertices.ptr(), sizeof(Vector3) * vertex_count);
+
+	if (p_indices.is_empty()) {
+		ERR_FAIL_COND(vertex_count % 3 != 0);
+		uint32_t *embree_triangles = (uint32_t *)rtcSetNewGeometryBuffer(embree_mesh, RTC_BUFFER_TYPE_INDEX, 0, RTC_FORMAT_UINT3, sizeof(uint32_t) * 3, vertex_count / 3);
+		for (int i = 0; i < vertex_count; i++) {
+			embree_triangles[i] = i;
+		}
+	} else {
+		uint32_t *embree_triangles = (uint32_t *)rtcSetNewGeometryBuffer(embree_mesh, RTC_BUFFER_TYPE_INDEX, 0, RTC_FORMAT_UINT3, sizeof(uint32_t) * 3, p_indices.size() / 3);
+		memcpy(embree_triangles, p_indices.ptr(), sizeof(uint32_t) * p_indices.size());
+	}
+
+	rtcCommitGeometry(embree_mesh);
+	rtcAttachGeometryByID(embree_scene, embree_mesh, p_id);
+	rtcReleaseGeometry(embree_mesh);
+}
+
+void StaticRaycasterEmbree::commit() {
+	rtcCommitScene(embree_scene);
+}
+
+void StaticRaycasterEmbree::set_mesh_filter(const Set<int> &p_mesh_ids) {
+	for (Set<int>::Element *E = p_mesh_ids.front(); E; E = E->next()) {
+		rtcDisableGeometry(rtcGetGeometry(embree_scene, E->get()));
+	}
+	rtcCommitScene(embree_scene);
+	filter_meshes = p_mesh_ids;
+}
+
+void StaticRaycasterEmbree::clear_mesh_filter() {
+	for (Set<int>::Element *E = filter_meshes.front(); E; E = E->next()) {
+		rtcEnableGeometry(rtcGetGeometry(embree_scene, E->get()));
+	}
+	rtcCommitScene(embree_scene);
+	filter_meshes.clear();
+}
+
+void embree_error_handler(void *p_user_data, RTCError p_code, const char *p_str) {
+	print_error("Embree error: " + String(p_str));
+}
+
+StaticRaycasterEmbree::StaticRaycasterEmbree() {
+#ifdef __SSE2__
+	_MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON);
+	_MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON);
+#endif
+
+	if (!embree_device) {
+		embree_device = rtcNewDevice(nullptr);
+		rtcSetDeviceErrorFunction(embree_device, &embree_error_handler, nullptr);
+	}
+
+	embree_scene = rtcNewScene(embree_device);
+}
+
+StaticRaycasterEmbree::~StaticRaycasterEmbree() {
+	if (embree_scene != nullptr) {
+		rtcReleaseScene(embree_scene);
+	}
+}
+
+#endif
diff --git a/modules/raycast/static_raycaster.h b/modules/raycast/static_raycaster.h
new file mode 100644
index 0000000000..6b13ecf690
--- /dev/null
+++ b/modules/raycast/static_raycaster.h
@@ -0,0 +1,64 @@
+/*************************************************************************/
+/*  static_raycaster.h                                                   */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2021 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2021 Godot Engine contributors (cf. AUTHORS.md).   */
+/*                                                                       */
+/* 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.                */
+/*************************************************************************/
+
+#ifdef TOOLS_ENABLED
+
+#include "core/math/static_raycaster.h"
+
+#include <embree3/rtcore.h>
+
+class StaticRaycasterEmbree : public StaticRaycaster {
+	GDCLASS(StaticRaycasterEmbree, StaticRaycaster);
+
+private:
+	static RTCDevice embree_device;
+	RTCScene embree_scene;
+
+	Set<int> filter_meshes;
+
+public:
+	virtual bool intersect(Ray &p_ray) override;
+	virtual void intersect(Vector<Ray> &r_rays) override;
+
+	virtual void add_mesh(const PackedVector3Array &p_vertices, const PackedInt32Array &p_indices, unsigned int p_id) override;
+	virtual void commit() override;
+
+	virtual void set_mesh_filter(const Set<int> &p_mesh_ids) override;
+	virtual void clear_mesh_filter() override;
+
+	static StaticRaycaster *create_embree_raycaster();
+	static void make_default_raycaster();
+	static void free();
+
+	StaticRaycasterEmbree();
+	~StaticRaycasterEmbree();
+};
+
+#endif
diff --git a/servers/rendering_server.h b/servers/rendering_server.h
index 1b04a6e5e2..579f8abbe6 100644
--- a/servers/rendering_server.h
+++ b/servers/rendering_server.h
@@ -295,7 +295,7 @@ public:
 
 		AABB aabb;
 		struct LOD {
-			float edge_length;
+			float edge_length = 0.0f;
 			Vector<uint8_t> index_data;
 		};
 		Vector<LOD> lods;
diff --git a/thirdparty/README.md b/thirdparty/README.md
index 990bd30320..0c458bef7a 100644
--- a/thirdparty/README.md
+++ b/thirdparty/README.md
@@ -371,7 +371,9 @@ Files extracted from upstream repository:
 - `LICENSE.md`.
 
 An [experimental upstream feature](https://github.com/zeux/meshoptimizer/tree/simplify-attr),
-has been backported, see patch in `patches` directory.
+has been backported. On top of that, it was modified to report only distance error metrics 
+instead of a combination of distance and attribute errors. Patches for both changes can be
+found in the `patches` directory.
 
 
 ## miniupnpc
diff --git a/thirdparty/meshoptimizer/patches/attribute-aware-simplify-distance-only-metric.patch b/thirdparty/meshoptimizer/patches/attribute-aware-simplify-distance-only-metric.patch
new file mode 100644
index 0000000000..54132a6c86
--- /dev/null
+++ b/thirdparty/meshoptimizer/patches/attribute-aware-simplify-distance-only-metric.patch
@@ -0,0 +1,176 @@
+diff --git a/thirdparty/meshoptimizer/simplifier.cpp b/thirdparty/meshoptimizer/simplifier.cpp
+index 0f10ebef4b..cf5db4e119 100644
+--- a/thirdparty/meshoptimizer/simplifier.cpp
++++ b/thirdparty/meshoptimizer/simplifier.cpp
+@@ -20,7 +20,7 @@
+ #define TRACESTATS(i) (void)0
+ #endif
+ 
+-#define ATTRIBUTES 8
++#define ATTRIBUTES 3
+ 
+ // This work is based on:
+ // Michael Garland and Paul S. Heckbert. Surface simplification using quadric error metrics. 1997
+@@ -445,6 +445,7 @@ struct Collapse
+ 		float error;
+ 		unsigned int errorui;
+ 	};
++	float distance_error;
+ };
+ 
+ static float normalize(Vector3& v)
+@@ -525,6 +526,34 @@ static float quadricError(const Quadric& Q, const Vector3& v)
+ 	return fabsf(r) * s;
+ }
+ 
++static float quadricErrorNoAttributes(const Quadric& Q, const Vector3& v)
++{
++	float rx = Q.b0;
++	float ry = Q.b1;
++	float rz = Q.b2;
++
++	rx += Q.a10 * v.y;
++	ry += Q.a21 * v.z;
++	rz += Q.a20 * v.x;
++
++	rx *= 2;
++	ry *= 2;
++	rz *= 2;
++
++	rx += Q.a00 * v.x;
++	ry += Q.a11 * v.y;
++	rz += Q.a22 * v.z;
++
++	float r = Q.c;
++	r += rx * v.x;
++	r += ry * v.y;
++	r += rz * v.z;
++
++	float s = Q.w == 0.f ? 0.f : 1.f / Q.w;
++
++	return fabsf(r) * s;
++}
++
+ static void quadricFromPlane(Quadric& Q, float a, float b, float c, float d, float w)
+ {
+ 	float aw = a * w;
+@@ -680,7 +709,7 @@ static void quadricUpdateAttributes(Quadric& Q, const Vector3& p0, const Vector3
+ }
+ #endif
+ 
+-static void fillFaceQuadrics(Quadric* vertex_quadrics, const unsigned int* indices, size_t index_count, const Vector3* vertex_positions, const unsigned int* remap)
++static void fillFaceQuadrics(Quadric* vertex_quadrics, Quadric* vertex_no_attrib_quadrics, const unsigned int* indices, size_t index_count, const Vector3* vertex_positions, const unsigned int* remap)
+ {
+ 	for (size_t i = 0; i < index_count; i += 3)
+ 	{
+@@ -690,6 +719,9 @@ static void fillFaceQuadrics(Quadric* vertex_quadrics, const unsigned int* indic
+ 
+ 		Quadric Q;
+ 		quadricFromTriangle(Q, vertex_positions[i0], vertex_positions[i1], vertex_positions[i2], 1.f);
++		quadricAdd(vertex_no_attrib_quadrics[remap[i0]], Q);
++		quadricAdd(vertex_no_attrib_quadrics[remap[i1]], Q);
++		quadricAdd(vertex_no_attrib_quadrics[remap[i2]], Q);
+ 
+ #if ATTRIBUTES
+ 		quadricUpdateAttributes(Q, vertex_positions[i0], vertex_positions[i1], vertex_positions[i2], Q.w);
+@@ -700,7 +732,7 @@ static void fillFaceQuadrics(Quadric* vertex_quadrics, const unsigned int* indic
+ 	}
+ }
+ 
+-static void fillEdgeQuadrics(Quadric* vertex_quadrics, const unsigned int* indices, size_t index_count, const Vector3* vertex_positions, const unsigned int* remap, const unsigned char* vertex_kind, const unsigned int* loop, const unsigned int* loopback)
++static void fillEdgeQuadrics(Quadric* vertex_quadrics, Quadric* vertex_no_attrib_quadrics, const unsigned int* indices, size_t index_count, const Vector3* vertex_positions, const unsigned int* remap, const unsigned char* vertex_kind, const unsigned int* loop, const unsigned int* loopback)
+ {
+ 	for (size_t i = 0; i < index_count; i += 3)
+ 	{
+@@ -744,6 +776,9 @@ static void fillEdgeQuadrics(Quadric* vertex_quadrics, const unsigned int* indic
+ 
+ 			quadricAdd(vertex_quadrics[remap[i0]], Q);
+ 			quadricAdd(vertex_quadrics[remap[i1]], Q);
++
++			quadricAdd(vertex_no_attrib_quadrics[remap[i0]], Q);
++			quadricAdd(vertex_no_attrib_quadrics[remap[i1]], Q);
+ 		}
+ 	}
+ }
+@@ -848,7 +883,7 @@ static size_t pickEdgeCollapses(Collapse* collapses, const unsigned int* indices
+ 	return collapse_count;
+ }
+ 
+-static void rankEdgeCollapses(Collapse* collapses, size_t collapse_count, const Vector3* vertex_positions, const Quadric* vertex_quadrics, const unsigned int* remap)
++static void rankEdgeCollapses(Collapse* collapses, size_t collapse_count, const Vector3* vertex_positions, const Quadric* vertex_quadrics, const Quadric* vertex_no_attrib_quadrics, const unsigned int* remap)
+ {
+ 	for (size_t i = 0; i < collapse_count; ++i)
+ 	{
+@@ -868,10 +903,14 @@ static void rankEdgeCollapses(Collapse* collapses, size_t collapse_count, const
+ 		float ei = quadricError(qi, vertex_positions[i1]);
+ 		float ej = quadricError(qj, vertex_positions[j1]);
+ 
++		const Quadric& naqi = vertex_no_attrib_quadrics[remap[i0]];
++		const Quadric& naqj = vertex_no_attrib_quadrics[remap[j0]];
++
+ 		// pick edge direction with minimal error
+ 		c.v0 = ei <= ej ? i0 : j0;
+ 		c.v1 = ei <= ej ? i1 : j1;
+ 		c.error = ei <= ej ? ei : ej;
++		c.distance_error = ei <= ej ? quadricErrorNoAttributes(naqi, vertex_positions[i1]) :  quadricErrorNoAttributes(naqj, vertex_positions[j1]);
+ 	}
+ }
+ 
+@@ -968,7 +1007,7 @@ static void sortEdgeCollapses(unsigned int* sort_order, const Collapse* collapse
+ 	}
+ }
+ 
+-static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char* collapse_locked, Quadric* vertex_quadrics, const Collapse* collapses, size_t collapse_count, const unsigned int* collapse_order, const unsigned int* remap, const unsigned int* wedge, const unsigned char* vertex_kind, const Vector3* vertex_positions, const EdgeAdjacency& adjacency, size_t triangle_collapse_goal, float error_limit, float& result_error)
++static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char* collapse_locked, Quadric* vertex_quadrics, Quadric* vertex_no_attrib_quadrics, const Collapse* collapses, size_t collapse_count, const unsigned int* collapse_order, const unsigned int* remap, const unsigned int* wedge, const unsigned char* vertex_kind, const Vector3* vertex_positions, const EdgeAdjacency& adjacency, size_t triangle_collapse_goal, float error_limit, float& result_error)
+ {
+ 	size_t edge_collapses = 0;
+ 	size_t triangle_collapses = 0;
+@@ -1030,6 +1069,7 @@ static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char*
+ 		assert(collapse_remap[r1] == r1);
+ 
+ 		quadricAdd(vertex_quadrics[r1], vertex_quadrics[r0]);
++		quadricAdd(vertex_no_attrib_quadrics[r1], vertex_no_attrib_quadrics[r0]);
+ 
+ 		if (vertex_kind[i0] == Kind_Complex)
+ 		{
+@@ -1067,7 +1107,7 @@ static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char*
+ 		triangle_collapses += (vertex_kind[i0] == Kind_Border) ? 1 : 2;
+ 		edge_collapses++;
+ 
+-		result_error = result_error < c.error ? c.error : result_error;
++		result_error = result_error < c.distance_error ? c.distance_error : result_error;
+ 	}
+ 
+ #if TRACE
+@@ -1455,9 +1495,11 @@ size_t meshopt_simplifyWithAttributes(unsigned int* destination, const unsigned
+ 
+ 	Quadric* vertex_quadrics = allocator.allocate<Quadric>(vertex_count);
+ 	memset(vertex_quadrics, 0, vertex_count * sizeof(Quadric));
++	Quadric* vertex_no_attrib_quadrics = allocator.allocate<Quadric>(vertex_count);
++	memset(vertex_no_attrib_quadrics, 0, vertex_count * sizeof(Quadric));
+ 
+-	fillFaceQuadrics(vertex_quadrics, indices, index_count, vertex_positions, remap);
+-	fillEdgeQuadrics(vertex_quadrics, indices, index_count, vertex_positions, remap, vertex_kind, loop, loopback);
++	fillFaceQuadrics(vertex_quadrics, vertex_no_attrib_quadrics, indices, index_count, vertex_positions, remap);
++	fillEdgeQuadrics(vertex_quadrics, vertex_no_attrib_quadrics, indices, index_count, vertex_positions, remap, vertex_kind, loop, loopback);
+ 
+ 	if (result != indices)
+ 		memcpy(result, indices, index_count * sizeof(unsigned int));
+@@ -1488,7 +1530,7 @@ size_t meshopt_simplifyWithAttributes(unsigned int* destination, const unsigned
+ 		if (edge_collapse_count == 0)
+ 			break;
+ 
+-		rankEdgeCollapses(edge_collapses, edge_collapse_count, vertex_positions, vertex_quadrics, remap);
++		rankEdgeCollapses(edge_collapses, edge_collapse_count, vertex_positions, vertex_quadrics, vertex_no_attrib_quadrics, remap);
+ 
+ #if TRACE > 1
+ 		dumpEdgeCollapses(edge_collapses, edge_collapse_count, vertex_kind);
+@@ -1507,7 +1549,7 @@ size_t meshopt_simplifyWithAttributes(unsigned int* destination, const unsigned
+ 		printf("pass %d: ", int(pass_count++));
+ #endif
+ 
+-		size_t collapses = performEdgeCollapses(collapse_remap, collapse_locked, vertex_quadrics, edge_collapses, edge_collapse_count, collapse_order, remap, wedge, vertex_kind, vertex_positions, adjacency, triangle_collapse_goal, error_limit, result_error);
++		size_t collapses = performEdgeCollapses(collapse_remap, collapse_locked, vertex_quadrics, vertex_no_attrib_quadrics, edge_collapses, edge_collapse_count, collapse_order, remap, wedge, vertex_kind, vertex_positions, adjacency, triangle_collapse_goal, error_limit, result_error);
+ 
+ 		// no edges can be collapsed any more due to hitting the error limit or triangle collapse limit
+ 		if (collapses == 0)
diff --git a/thirdparty/meshoptimizer/simplifier.cpp b/thirdparty/meshoptimizer/simplifier.cpp
index 0f10ebef4b..cf5db4e119 100644
--- a/thirdparty/meshoptimizer/simplifier.cpp
+++ b/thirdparty/meshoptimizer/simplifier.cpp
@@ -20,7 +20,7 @@
 #define TRACESTATS(i) (void)0
 #endif
 
-#define ATTRIBUTES 8
+#define ATTRIBUTES 3
 
 // This work is based on:
 // Michael Garland and Paul S. Heckbert. Surface simplification using quadric error metrics. 1997
@@ -445,6 +445,7 @@ struct Collapse
 		float error;
 		unsigned int errorui;
 	};
+	float distance_error;
 };
 
 static float normalize(Vector3& v)
@@ -525,6 +526,34 @@ static float quadricError(const Quadric& Q, const Vector3& v)
 	return fabsf(r) * s;
 }
 
+static float quadricErrorNoAttributes(const Quadric& Q, const Vector3& v)
+{
+	float rx = Q.b0;
+	float ry = Q.b1;
+	float rz = Q.b2;
+
+	rx += Q.a10 * v.y;
+	ry += Q.a21 * v.z;
+	rz += Q.a20 * v.x;
+
+	rx *= 2;
+	ry *= 2;
+	rz *= 2;
+
+	rx += Q.a00 * v.x;
+	ry += Q.a11 * v.y;
+	rz += Q.a22 * v.z;
+
+	float r = Q.c;
+	r += rx * v.x;
+	r += ry * v.y;
+	r += rz * v.z;
+
+	float s = Q.w == 0.f ? 0.f : 1.f / Q.w;
+
+	return fabsf(r) * s;
+}
+
 static void quadricFromPlane(Quadric& Q, float a, float b, float c, float d, float w)
 {
 	float aw = a * w;
@@ -680,7 +709,7 @@ static void quadricUpdateAttributes(Quadric& Q, const Vector3& p0, const Vector3
 }
 #endif
 
-static void fillFaceQuadrics(Quadric* vertex_quadrics, const unsigned int* indices, size_t index_count, const Vector3* vertex_positions, const unsigned int* remap)
+static void fillFaceQuadrics(Quadric* vertex_quadrics, Quadric* vertex_no_attrib_quadrics, const unsigned int* indices, size_t index_count, const Vector3* vertex_positions, const unsigned int* remap)
 {
 	for (size_t i = 0; i < index_count; i += 3)
 	{
@@ -690,6 +719,9 @@ static void fillFaceQuadrics(Quadric* vertex_quadrics, const unsigned int* indic
 
 		Quadric Q;
 		quadricFromTriangle(Q, vertex_positions[i0], vertex_positions[i1], vertex_positions[i2], 1.f);
+		quadricAdd(vertex_no_attrib_quadrics[remap[i0]], Q);
+		quadricAdd(vertex_no_attrib_quadrics[remap[i1]], Q);
+		quadricAdd(vertex_no_attrib_quadrics[remap[i2]], Q);
 
 #if ATTRIBUTES
 		quadricUpdateAttributes(Q, vertex_positions[i0], vertex_positions[i1], vertex_positions[i2], Q.w);
@@ -700,7 +732,7 @@ static void fillFaceQuadrics(Quadric* vertex_quadrics, const unsigned int* indic
 	}
 }
 
-static void fillEdgeQuadrics(Quadric* vertex_quadrics, const unsigned int* indices, size_t index_count, const Vector3* vertex_positions, const unsigned int* remap, const unsigned char* vertex_kind, const unsigned int* loop, const unsigned int* loopback)
+static void fillEdgeQuadrics(Quadric* vertex_quadrics, Quadric* vertex_no_attrib_quadrics, const unsigned int* indices, size_t index_count, const Vector3* vertex_positions, const unsigned int* remap, const unsigned char* vertex_kind, const unsigned int* loop, const unsigned int* loopback)
 {
 	for (size_t i = 0; i < index_count; i += 3)
 	{
@@ -744,6 +776,9 @@ static void fillEdgeQuadrics(Quadric* vertex_quadrics, const unsigned int* indic
 
 			quadricAdd(vertex_quadrics[remap[i0]], Q);
 			quadricAdd(vertex_quadrics[remap[i1]], Q);
+
+			quadricAdd(vertex_no_attrib_quadrics[remap[i0]], Q);
+			quadricAdd(vertex_no_attrib_quadrics[remap[i1]], Q);
 		}
 	}
 }
@@ -848,7 +883,7 @@ static size_t pickEdgeCollapses(Collapse* collapses, const unsigned int* indices
 	return collapse_count;
 }
 
-static void rankEdgeCollapses(Collapse* collapses, size_t collapse_count, const Vector3* vertex_positions, const Quadric* vertex_quadrics, const unsigned int* remap)
+static void rankEdgeCollapses(Collapse* collapses, size_t collapse_count, const Vector3* vertex_positions, const Quadric* vertex_quadrics, const Quadric* vertex_no_attrib_quadrics, const unsigned int* remap)
 {
 	for (size_t i = 0; i < collapse_count; ++i)
 	{
@@ -868,10 +903,14 @@ static void rankEdgeCollapses(Collapse* collapses, size_t collapse_count, const
 		float ei = quadricError(qi, vertex_positions[i1]);
 		float ej = quadricError(qj, vertex_positions[j1]);
 
+		const Quadric& naqi = vertex_no_attrib_quadrics[remap[i0]];
+		const Quadric& naqj = vertex_no_attrib_quadrics[remap[j0]];
+
 		// pick edge direction with minimal error
 		c.v0 = ei <= ej ? i0 : j0;
 		c.v1 = ei <= ej ? i1 : j1;
 		c.error = ei <= ej ? ei : ej;
+		c.distance_error = ei <= ej ? quadricErrorNoAttributes(naqi, vertex_positions[i1]) :  quadricErrorNoAttributes(naqj, vertex_positions[j1]);
 	}
 }
 
@@ -968,7 +1007,7 @@ static void sortEdgeCollapses(unsigned int* sort_order, const Collapse* collapse
 	}
 }
 
-static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char* collapse_locked, Quadric* vertex_quadrics, const Collapse* collapses, size_t collapse_count, const unsigned int* collapse_order, const unsigned int* remap, const unsigned int* wedge, const unsigned char* vertex_kind, const Vector3* vertex_positions, const EdgeAdjacency& adjacency, size_t triangle_collapse_goal, float error_limit, float& result_error)
+static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char* collapse_locked, Quadric* vertex_quadrics, Quadric* vertex_no_attrib_quadrics, const Collapse* collapses, size_t collapse_count, const unsigned int* collapse_order, const unsigned int* remap, const unsigned int* wedge, const unsigned char* vertex_kind, const Vector3* vertex_positions, const EdgeAdjacency& adjacency, size_t triangle_collapse_goal, float error_limit, float& result_error)
 {
 	size_t edge_collapses = 0;
 	size_t triangle_collapses = 0;
@@ -1030,6 +1069,7 @@ static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char*
 		assert(collapse_remap[r1] == r1);
 
 		quadricAdd(vertex_quadrics[r1], vertex_quadrics[r0]);
+		quadricAdd(vertex_no_attrib_quadrics[r1], vertex_no_attrib_quadrics[r0]);
 
 		if (vertex_kind[i0] == Kind_Complex)
 		{
@@ -1067,7 +1107,7 @@ static size_t performEdgeCollapses(unsigned int* collapse_remap, unsigned char*
 		triangle_collapses += (vertex_kind[i0] == Kind_Border) ? 1 : 2;
 		edge_collapses++;
 
-		result_error = result_error < c.error ? c.error : result_error;
+		result_error = result_error < c.distance_error ? c.distance_error : result_error;
 	}
 
 #if TRACE
@@ -1455,9 +1495,11 @@ size_t meshopt_simplifyWithAttributes(unsigned int* destination, const unsigned
 
 	Quadric* vertex_quadrics = allocator.allocate<Quadric>(vertex_count);
 	memset(vertex_quadrics, 0, vertex_count * sizeof(Quadric));
+	Quadric* vertex_no_attrib_quadrics = allocator.allocate<Quadric>(vertex_count);
+	memset(vertex_no_attrib_quadrics, 0, vertex_count * sizeof(Quadric));
 
-	fillFaceQuadrics(vertex_quadrics, indices, index_count, vertex_positions, remap);
-	fillEdgeQuadrics(vertex_quadrics, indices, index_count, vertex_positions, remap, vertex_kind, loop, loopback);
+	fillFaceQuadrics(vertex_quadrics, vertex_no_attrib_quadrics, indices, index_count, vertex_positions, remap);
+	fillEdgeQuadrics(vertex_quadrics, vertex_no_attrib_quadrics, indices, index_count, vertex_positions, remap, vertex_kind, loop, loopback);
 
 	if (result != indices)
 		memcpy(result, indices, index_count * sizeof(unsigned int));
@@ -1488,7 +1530,7 @@ size_t meshopt_simplifyWithAttributes(unsigned int* destination, const unsigned
 		if (edge_collapse_count == 0)
 			break;
 
-		rankEdgeCollapses(edge_collapses, edge_collapse_count, vertex_positions, vertex_quadrics, remap);
+		rankEdgeCollapses(edge_collapses, edge_collapse_count, vertex_positions, vertex_quadrics, vertex_no_attrib_quadrics, remap);
 
 #if TRACE > 1
 		dumpEdgeCollapses(edge_collapses, edge_collapse_count, vertex_kind);
@@ -1507,7 +1549,7 @@ size_t meshopt_simplifyWithAttributes(unsigned int* destination, const unsigned
 		printf("pass %d: ", int(pass_count++));
 #endif
 
-		size_t collapses = performEdgeCollapses(collapse_remap, collapse_locked, vertex_quadrics, edge_collapses, edge_collapse_count, collapse_order, remap, wedge, vertex_kind, vertex_positions, adjacency, triangle_collapse_goal, error_limit, result_error);
+		size_t collapses = performEdgeCollapses(collapse_remap, collapse_locked, vertex_quadrics, vertex_no_attrib_quadrics, edge_collapses, edge_collapse_count, collapse_order, remap, wedge, vertex_kind, vertex_positions, adjacency, triangle_collapse_goal, error_limit, result_error);
 
 		// no edges can be collapsed any more due to hitting the error limit or triangle collapse limit
 		if (collapses == 0)