Modified rendering to use cluster instead of foward

core/math/basis.cpp

@@ -244,6 +244,18 @@ void Basis::scale_local(const Vector3 &p_scale) {
 	*this = scaled_local(p_scale);
 }
 
+float Basis::get_uniform_scale() const {
+	return (elements[0].length() + elements[1].length() + elements[2].length()) / 3.0;
+}
+
+void Basis::make_scale_uniform() {
+	float l = (elements[0].length() + elements[1].length() + elements[2].length()) / 3.0;
+	for (int i = 0; i < 3; i++) {
+		elements[i].normalize();
+		elements[i] *= l;
+	}
+}
+
 Basis Basis::scaled_local(const Vector3 &p_scale) const {
 	Basis b;
 	b.set_diagonal(p_scale);

core/math/basis.h

@@ -108,6 +108,9 @@ public:
 	void scale_local(const Vector3 &p_scale);
 	Basis scaled_local(const Vector3 &p_scale) const;
 
+	void make_scale_uniform();
+	float get_uniform_scale() const;
+
 	Vector3 get_scale() const;
 	Vector3 get_scale_abs() const;
 	Vector3 get_scale_local() const;

core/math/camera_matrix.cpp

@@ -276,6 +276,36 @@ Vector2 CameraMatrix::get_viewport_half_extents() const {
 	return Vector2(res.x, res.y);
 }
 
+void CameraMatrix::get_far_plane_size(real_t &r_width, real_t &r_height) const {
+
+	const real_t *matrix = (const real_t *)this->matrix;
+	///////--- Far Plane ---///////
+	Plane far_plane = Plane(matrix[3] - matrix[2],
+			matrix[7] - matrix[6],
+			matrix[11] - matrix[10],
+			-matrix[15] + matrix[14]);
+	far_plane.normalize();
+
+	///////--- Right Plane ---///////
+	Plane right_plane = Plane(matrix[3] - matrix[0],
+			matrix[7] - matrix[4],
+			matrix[11] - matrix[8],
+			-matrix[15] + matrix[12]);
+	right_plane.normalize();
+
+	Plane top_plane = Plane(matrix[3] - matrix[1],
+			matrix[7] - matrix[5],
+			matrix[11] - matrix[9],
+			-matrix[15] + matrix[13]);
+	top_plane.normalize();
+
+	Vector3 res;
+	far_plane.intersect_3(right_plane, top_plane, &res);
+
+	r_width = res.x;
+	r_height = res.y;
+}
+
 bool CameraMatrix::get_endpoints(const Transform &p_transform, Vector3 *p_8points) const {
 
 	Vector<Plane> planes = get_projection_planes(Transform());

core/math/camera_matrix.h

@@ -75,6 +75,7 @@ struct CameraMatrix {
 
 	bool get_endpoints(const Transform &p_transform, Vector3 *p_8points) const;
 	Vector2 get_viewport_half_extents() const;
+	void get_far_plane_size(real_t &r_width, real_t &r_height) const;
 
 	void invert();
 	CameraMatrix inverse() const;
@@ -93,6 +94,21 @@ struct CameraMatrix {
 
 	void flip_y();
 
+	bool operator==(const CameraMatrix &p_cam) const {
+		for (uint32_t i = 0; i < 4; i++) {
+			for (uint32_t j = 0; j < 4; j++) {
+				if (matrix[i][j] != p_cam.matrix[i][j]) {
+					return false;
+				}
+			}
+		}
+		return true;
+	}
+
+	bool operator!=(const CameraMatrix &p_cam) const {
+		return !(*this == p_cam);
+	}
+
 	CameraMatrix();
 	CameraMatrix(const Transform &p_transform);
 	~CameraMatrix();

servers/visual/rasterizer_rd/light_cluster_builder.cpp

+#include "light_cluster_builder.h"
+
+void LightClusterBuilder::begin(const Transform &p_view_transform, const CameraMatrix &p_cam_projection) {
+	view_xform = p_view_transform;
+	projection = p_cam_projection;
+	z_near = -projection.get_z_near();
+	z_far = -projection.get_z_far();
+
+	//reset counts
+	light_count = 0;
+	refprobe_count = 0;
+	item_count = 0;
+	sort_id_count = 0;
+}
+
+void LightClusterBuilder::bake_cluster() {
+
+	float slice_depth = (z_near - z_far) / depth;
+
+	PoolVector<uint8_t>::Write cluster_dataw = cluster_data.write();
+	Cell *cluster_data_ptr = (Cell *)cluster_dataw.ptr();
+	//clear the cluster
+	zeromem(cluster_data_ptr, (width * height * depth * sizeof(Cell)));
+
+	/* Step 1, create cell positions and count them */
+
+	for (uint32_t i = 0; i < item_count; i++) {
+
+		const Item &item = items[i];
+
+		int from_slice = Math::floor((z_near - (item.aabb.position.z + item.aabb.size.z)) / slice_depth);
+		int to_slice = Math::floor((z_near - item.aabb.position.z) / slice_depth);
+
+		if (from_slice >= (int)depth || to_slice < 0) {
+			continue; //sorry no go
+		}
+
+		from_slice = MAX(0, from_slice);
+		to_slice = MIN((int)depth - 1, to_slice);
+
+		for (int j = from_slice; j <= to_slice; j++) {
+
+			Vector3 min = item.aabb.position;
+			Vector3 max = item.aabb.position + item.aabb.size;
+
+			float limit_near = MIN((z_near - slice_depth * j), max.z);
+			float limit_far = MAX((z_near - slice_depth * (j + 1)), min.z);
+
+			max.z = limit_near;
+			min.z = limit_near;
+
+			Vector3 proj_min = projection.xform(min);
+			Vector3 proj_max = projection.xform(max);
+
+			int near_from_x = int(Math::floor((proj_min.x * 0.5 + 0.5) * width));
+			int near_from_y = int(Math::floor((-proj_max.y * 0.5 + 0.5) * height));
+			int near_to_x = int(Math::floor((proj_max.x * 0.5 + 0.5) * width));
+			int near_to_y = int(Math::floor((-proj_min.y * 0.5 + 0.5) * height));
+
+			max.z = limit_far;
+			min.z = limit_far;
+
+			proj_min = projection.xform(min);
+			proj_max = projection.xform(max);
+
+			int far_from_x = int(Math::floor((proj_min.x * 0.5 + 0.5) * width));
+			int far_from_y = int(Math::floor((-proj_max.y * 0.5 + 0.5) * height));
+			int far_to_x = int(Math::floor((proj_max.x * 0.5 + 0.5) * width));
+			int far_to_y = int(Math::floor((-proj_min.y * 0.5 + 0.5) * height));
+
+			//print_line(itos(j) + " near - " + Vector2i(near_from_x, near_from_y) + " -> " + Vector2i(near_to_x, near_to_y));
+			//print_line(itos(j) + " far - " + Vector2i(far_from_x, far_from_y) + " -> " + Vector2i(far_to_x, far_to_y));
+
+			int from_x = MIN(near_from_x, far_from_x);
+			int from_y = MIN(near_from_y, far_from_y);
+			int to_x = MAX(near_to_x, far_to_x);
+			int to_y = MAX(near_to_y, far_to_y);
+
+			if (from_x >= (int)width || to_x < 0 || from_y >= (int)height || to_y < 0) {
+				continue;
+			}
+
+			int sx = MAX(0, from_x);
+			int sy = MAX(0, from_y);
+			int dx = MIN(width - 1, to_x);
+			int dy = MIN(height - 1, to_y);
+
+			//print_line(itos(j) + " - " + Vector2i(sx, sy) + " -> " + Vector2i(dx, dy));
+
+			for (int x = sx; x <= dx; x++) {
+				for (int y = sy; y <= dy; y++) {
+					uint32_t offset = j * (width * height) + y * width + x;
+
+					if (unlikely(sort_id_count == sort_id_max)) {
+						sort_id_max = nearest_power_of_2_templated(sort_id_max + 1);
+						sort_ids = (SortID *)memrealloc(sort_ids, sizeof(SortID) * sort_id_max);
+						if (ids.size()) {
+
+							ids.resize(sort_id_max);
+							RD::get_singleton()->free(items_buffer);
+							items_buffer = RD::get_singleton()->storage_buffer_create(sizeof(uint32_t) * sort_id_max);
+						}
+					}
+
+					sort_ids[sort_id_count].cell_index = offset;
+					sort_ids[sort_id_count].item_index = item.index;
+					sort_ids[sort_id_count].item_type = item.type;
+
+					sort_id_count++;
+
+					//for now, only count
+					cluster_data_ptr[offset].item_pointers[item.type]++;
+					//print_line("at offset " + itos(offset) + " value: " + itos(cluster_data_ptr[offset].item_pointers[item.type]));
+				}
+			}
+		}
+	}
+
+	/* Step 2, Assign pointers (and reset counters) */
+
+	uint32_t offset = 0;
+	for (uint32_t i = 0; i < (width * height * depth); i++) {
+		for (int j = 0; j < ITEM_TYPE_MAX; j++) {
+			uint32_t count = cluster_data_ptr[i].item_pointers[j]; //save count
+			cluster_data_ptr[i].item_pointers[j] = offset; //replace count by pointer
+			offset += count; //increase offset by count;
+		}
+	}
+
+	//print_line("offset: " + itos(offset));
+	/* Step 3, Place item lists */
+
+	PoolVector<uint32_t>::Write idsw = ids.write();
+	uint32_t *ids_ptr = idsw.ptr();
+
+	for (uint32_t i = 0; i < sort_id_count; i++) {
+		const SortID &id = sort_ids[i];
+		Cell &cell = cluster_data_ptr[id.cell_index];
+		uint32_t pointer = cell.item_pointers[id.item_type] & POINTER_MASK;
+		uint32_t counter = cell.item_pointers[id.item_type] >> COUNTER_SHIFT;
+		ids_ptr[pointer + counter] = id.item_index;
+
+		cell.item_pointers[id.item_type] = pointer | ((counter + 1) << COUNTER_SHIFT);
+	}
+
+	cluster_dataw = PoolVector<uint8_t>::Write();
+
+	RD::get_singleton()->texture_update(cluster_texture, 0, cluster_data, true);
+	RD::get_singleton()->buffer_update(items_buffer, 0, offset * sizeof(uint32_t), ids_ptr, true);
+
+	idsw = PoolVector<uint32_t>::Write();
+}
+
+void LightClusterBuilder::setup(uint32_t p_width, uint32_t p_height, uint32_t p_depth) {
+
+	if (width == p_width && height == p_height && depth == p_depth) {
+		return;
+	}
+	if (cluster_texture.is_valid()) {
+		RD::get_singleton()->free(cluster_texture);
+	}
+
+	width = p_width;
+	height = p_height;
+	depth = p_depth;
+
+	cluster_data.resize(width * height * depth * sizeof(Cell));
+
+	{
+		RD::TextureFormat tf;
+		tf.format = RD::DATA_FORMAT_R32G32B32A32_UINT;
+		tf.type = RD::TEXTURE_TYPE_3D;
+		tf.width = width;
+		tf.height = height;
+		tf.depth = depth;
+		tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_CAN_UPDATE_BIT;
+
+		cluster_texture = RD::get_singleton()->texture_create(tf, RD::TextureView());
+	}
+}
+
+RID LightClusterBuilder::get_cluster_texture() const {
+	return cluster_texture;
+}
+RID LightClusterBuilder::get_cluster_indices_buffer() const {
+	return items_buffer;
+}
+
+LightClusterBuilder::LightClusterBuilder() {
+	//initialize accumulators to something
+	lights = (LightData *)memalloc(sizeof(LightData) * 1024);
+	light_max = 1024;
+
+	refprobes = (OrientedBoxData *)memalloc(sizeof(OrientedBoxData) * 1024);
+	refprobe_max = 1024;
+
+	decals = (OrientedBoxData *)memalloc(sizeof(OrientedBoxData) * 1024);
+	decal_max = 1024;
+
+	items = (Item *)memalloc(sizeof(Item) * 1024);
+	item_max = 1024;
+
+	sort_ids = (SortID *)memalloc(sizeof(SortID) * 1024);
+	ids.resize(2014);
+	items_buffer = RD::get_singleton()->storage_buffer_create(sizeof(uint32_t) * 1024);
+	item_max = 1024;
+}
+LightClusterBuilder::~LightClusterBuilder() {
+
+	if (cluster_data.size()) {
+		RD::get_singleton()->free(cluster_texture);
+	}
+
+	if (lights) {
+		memfree(lights);
+	}
+	if (refprobes) {
+		memfree(refprobes);
+	}
+	if (decals) {
+		memfree(decals);
+	}
+	if (items) {
+		memfree(items);
+	}
+	if (sort_ids) {
+		memfree(sort_ids);
+		RD::get_singleton()->free(items_buffer);
+	}
+}

servers/visual/rasterizer_rd/light_cluster_builder.h

+#ifndef LIGHT_CLUSTER_BUILDER_H
+#define LIGHT_CLUSTER_BUILDER_H
+
+#include "servers/visual/rasterizer_rd/rasterizer_storage_rd.h"
+
+class LightClusterBuilder {
+public:
+	enum LightType {
+		LIGHT_TYPE_OMNI,
+		LIGHT_TYPE_SPOT
+	};
+
+	enum ItemType {
+		ITEM_TYPE_OMNI_LIGHT,
+		ITEM_TYPE_SPOT_LIGHT,
+		ITEM_TYPE_REFLECTION_PROBE,
+		ITEM_TYPE_DECAL,
+		ITEM_TYPE_MAX //should always be 4
+	};
+
+	enum {
+		COUNTER_SHIFT = 20, //one million total ids
+		POINTER_MASK = (1 << COUNTER_SHIFT) - 1,
+		COUNTER_MASK = 0xfff // 4096 items per cell
+	};
+
+private:
+	struct LightData {
+		float position[3];
+		uint32_t type;
+		float radius;
+		float spot_aperture;
+		uint32_t pad[2];
+	};
+
+	uint32_t light_count = 0;
+	uint32_t light_max = 0;
+	LightData *lights = nullptr;
+
+	struct OrientedBoxData {
+		float position[3];
+		uint32_t pad;
+		float x_axis[3];
+		uint32_t pad2;
+		float y_axis[3];
+		uint32_t pad3;
+		float z_axis[3];
+		uint32_t pad4;
+	};
+
+	uint32_t refprobe_count = 0;
+	uint32_t refprobe_max = 0;
+	OrientedBoxData *refprobes = nullptr;
+
+	uint32_t decal_count = 0;
+	uint32_t decal_max = 0;
+	OrientedBoxData *decals = nullptr;
+
+	struct Item {
+		AABB aabb;
+		ItemType type;
+		uint32_t index;
+	};
+
+	Item *items = nullptr;
+	uint32_t item_count = 0;
+	uint32_t item_max = 0;
+
+	uint32_t width = 0;
+	uint32_t height = 0;
+	uint32_t depth = 0;
+
+	struct Cell {
+		uint32_t item_pointers[ITEM_TYPE_MAX];
+	};
+
+	PoolVector<uint8_t> cluster_data;
+	RID cluster_texture;
+
+	struct SortID {
+		uint32_t cell_index;
+		uint32_t item_index;
+		ItemType item_type;
+	};
+
+	SortID *sort_ids = nullptr;
+	PoolVector<uint32_t> ids;
+	uint32_t sort_id_count = 0;
+	uint32_t sort_id_max = 0;
+	RID items_buffer;
+
+	Transform view_xform;
+	CameraMatrix projection;
+	float z_far = 0;
+	float z_near = 0;
+
+	_FORCE_INLINE_ void _add_item(const AABB &p_aabb, ItemType p_type, uint32_t p_index) {
+		if (unlikely(item_count == item_max)) {
+			item_max = nearest_power_of_2_templated(item_max + 1);
+			items = (Item *)memrealloc(items, sizeof(Item) * item_max);
+		}
+
+		Item &item = items[item_count];
+		item.aabb = p_aabb;
+		item.index = p_index;
+		item.type = p_type;
+		item_count++;
+	}
+
+public:
+	void begin(const Transform &p_view_transform, const CameraMatrix &p_cam_projection);
+
+	_FORCE_INLINE_ void add_light(LightType p_type, const Transform &p_transform, float p_radius, float p_spot_aperture) {
+		if (unlikely(light_count == light_max)) {
+			light_max = nearest_power_of_2_templated(light_max + 1);
+			lights = (LightData *)memrealloc(lights, sizeof(LightData) * light_max);
+		}
+
+		LightData &ld = lights[light_count];
+		ld.type = p_type;
+		ld.position[0] = p_transform.origin.x;
+		ld.position[1] = p_transform.origin.y;
+		ld.position[2] = p_transform.origin.z;
+		ld.radius = p_radius;
+		ld.spot_aperture = p_spot_aperture;
+
+		Transform xform = view_xform * p_transform;
+
+		ld.radius *= xform.basis.get_uniform_scale();
+
+		AABB aabb;
+
+		switch (p_type) {
+			case LIGHT_TYPE_OMNI: {
+				aabb.position = xform.origin;
+				aabb.size = Vector3(ld.radius, ld.radius, ld.radius);
+				aabb.position -= aabb.size;
+				aabb.size *= 2.0;
+
+				_add_item(aabb, ITEM_TYPE_OMNI_LIGHT, light_count);
+			} break;
+			case LIGHT_TYPE_SPOT: {
+				Vector3 v(0, 0, -1);
+				v.rotated(Vector3(0, 1, 0), Math::deg2rad(ld.spot_aperture)); //rotate in x-z
+				v.normalize();
+				v *= ld.radius;
+				v.y = v.x;
+
+				aabb.position = xform.origin;
+				aabb.expand_to(xform.xform(v));
+				aabb.expand_to(xform.xform(Vector3(-v.x, v.y, v.z)));
+				aabb.expand_to(xform.xform(Vector3(-v.x, -v.y, v.z)));
+				aabb.expand_to(xform.xform(Vector3(v.x, -v.y, v.z)));
+				_add_item(aabb, ITEM_TYPE_SPOT_LIGHT, light_count);
+			} break;
+		}
+
+		light_count++;
+	}
+
+	_FORCE_INLINE_ void add_reflection_probe(const Transform &p_transform, const Vector3 &p_half_extents) {
+
+		if (unlikely(refprobe_count == refprobe_max)) {
+			refprobe_max = nearest_power_of_2_templated(refprobe_max + 1);
+			refprobes = (OrientedBoxData *)memrealloc(refprobes, sizeof(OrientedBoxData) * refprobe_max);
+		}
+
+		OrientedBoxData &rp = refprobes[refprobe_count];
+		Vector3 origin = p_transform.origin;
+		rp.position[0] = origin.x;
+		rp.position[1] = origin.y;
+		rp.position[2] = origin.z;
+
+		Vector3 x_axis = p_transform.basis.get_axis(0) * p_half_extents.x;
+		rp.x_axis[0] = x_axis.x;
+		rp.x_axis[1] = x_axis.y;
+		rp.x_axis[2] = x_axis.z;
+
+		Vector3 y_axis = p_transform.basis.get_axis(1) * p_half_extents.y;
+		rp.y_axis[0] = y_axis.x;
+		rp.y_axis[1] = y_axis.y;
+		rp.y_axis[2] = y_axis.z;
+
+		Vector3 z_axis = p_transform.basis.get_axis(2) * p_half_extents.z;
+		rp.z_axis[0] = z_axis.x;
+		rp.z_axis[1] = z_axis.y;
+		rp.z_axis[2] = z_axis.z;
+
+		AABB aabb;
+
+		aabb.position = origin + x_axis + y_axis + z_axis;
+		aabb.expand_to(origin + x_axis + y_axis - z_axis);
+		aabb.expand_to(origin + x_axis - y_axis + z_axis);
+		aabb.expand_to(origin + x_axis - y_axis - z_axis);
+		aabb.expand_to(origin - x_axis + y_axis + z_axis);
+		aabb.expand_to(origin - x_axis + y_axis - z_axis);
+		aabb.expand_to(origin - x_axis - y_axis + z_axis);
+		aabb.expand_to(origin - x_axis - y_axis - z_axis);
+
+		_add_item(aabb, ITEM_TYPE_REFLECTION_PROBE, refprobe_count);
+
+		refprobe_count++;
+	}
+
+	_FORCE_INLINE_ void add_decal(const Transform &p_transform, const Vector2 &p_half_extents, float p_depth) {
+
+		if (unlikely(decal_count == decal_max)) {
+			decal_max = nearest_power_of_2_templated(decal_max + 1);
+			decals = (OrientedBoxData *)memrealloc(decals, sizeof(OrientedBoxData) * decal_max);
+		}
+
+		OrientedBoxData &dc = decals[decal_count];
+
+		Vector3 z_axis = -p_transform.basis.get_axis(2) * p_depth * 0.5;
+		dc.z_axis[0] = z_axis.x;
+		dc.z_axis[1] = z_axis.y;
+		dc.z_axis[2] = z_axis.z;
+
+		Vector3 origin = p_transform.origin - z_axis;
+		dc.position[0] = origin.x;
+		dc.position[1] = origin.y;
+		dc.position[2] = origin.z;
+
+		Vector3 x_axis = p_transform.basis.get_axis(0) * p_half_extents.x;
+		dc.x_axis[0] = x_axis.x;
+		dc.x_axis[1] = x_axis.y;
+		dc.x_axis[2] = x_axis.z;
+
+		Vector3 y_axis = p_transform.basis.get_axis(1) * p_half_extents.y;
+		dc.y_axis[0] = y_axis.x;
+		dc.y_axis[1] = y_axis.y;
+		dc.y_axis[2] = y_axis.z;
+
+		AABB aabb;
+
+		aabb.position = origin + x_axis + y_axis + z_axis;
+		aabb.expand_to(origin + x_axis + y_axis - z_axis);
+		aabb.expand_to(origin + x_axis - y_axis + z_axis);
+		aabb.expand_to(origin + x_axis - y_axis - z_axis);
+		aabb.expand_to(origin - x_axis + y_axis + z_axis);
+		aabb.expand_to(origin - x_axis + y_axis - z_axis);
+		aabb.expand_to(origin - x_axis - y_axis + z_axis);
+		aabb.expand_to(origin - x_axis - y_axis - z_axis);
+
+		_add_item(aabb, ITEM_TYPE_DECAL, decal_count);
+
+		decal_count++;
+	}
+
+	void bake_cluster();
+
+	void setup(uint32_t p_width, uint32_t p_height, uint32_t p_depth);
+
+	RID get_cluster_texture() const;
+	RID get_cluster_indices_buffer() const;
+
+	LightClusterBuilder();
+	~LightClusterBuilder();
+};
+
+#endif // LIGHT_CLUSTER_BUILDER_H

servers/visual/rasterizer_rd/rasterizer_rd.cpp

@@ -175,5 +175,5 @@ RasterizerRD::RasterizerRD() {
 
 	storage = memnew(RasterizerStorageRD);
 	canvas = memnew(RasterizerCanvasRD(storage));
-	scene = memnew(RasterizerSceneForwardRD(storage));
+	scene = memnew(RasterizerSceneHighEndRD(storage));
 }

servers/visual/rasterizer_rd/rasterizer_rd.h

@@ -35,14 +35,14 @@
 #include "core/thread_work_pool.h"
 #include "servers/visual/rasterizer.h"
 #include "servers/visual/rasterizer_rd/rasterizer_canvas_rd.h"
-#include "servers/visual/rasterizer_rd/rasterizer_scene_forward_rd.h"
+#include "servers/visual/rasterizer_rd/rasterizer_scene_high_end_rd.h"
 #include "servers/visual/rasterizer_rd/rasterizer_storage_rd.h"
 
 class RasterizerRD : public Rasterizer {
 protected:
 	RasterizerCanvasRD *canvas;
 	RasterizerStorageRD *storage;
-	RasterizerSceneForwardRD *scene;
+	RasterizerSceneHighEndRD *scene;
 
 	RID copy_viewports_rd_shader;
 	RID copy_viewports_rd_pipeline;

servers/visual/rasterizer_rd/rasterizer_scene_forward_rd.h → servers/visual/rasterizer_rd/rasterizer_scene_high_end_rd.h

@@ -28,15 +28,16 @@
 /* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.                */
 /*************************************************************************/
 
-#ifndef RASTERIZER_SCENE_FORWARD_RD_H
-#define RASTERIZER_SCENE_FORWARD_RD_H
+#ifndef RASTERIZER_SCENE_HIGHEND_RD_H
+#define RASTERIZER_SCENE_HIGHEND_RD_H
 
+#include "servers/visual/rasterizer_rd/light_cluster_builder.h"
 #include "servers/visual/rasterizer_rd/rasterizer_scene_rd.h"
 #include "servers/visual/rasterizer_rd/rasterizer_storage_rd.h"
 #include "servers/visual/rasterizer_rd/render_pipeline_vertex_format_cache_rd.h"
-#include "servers/visual/rasterizer_rd/shaders/scene_forward.glsl.gen.h"
+#include "servers/visual/rasterizer_rd/shaders/scene_high_end.glsl.gen.h"
 
-class RasterizerSceneForwardRD : public RasterizerSceneRD {
+class RasterizerSceneHighEndRD : public RasterizerSceneRD {
 
 	/* Shader */
 
@@ -56,7 +57,7 @@ class RasterizerSceneForwardRD : public RasterizerSceneRD {
 	};
 
 	struct {
-		SceneForwardShaderRD scene_shader;
+		SceneHighEndShaderRD scene_shader;
 		ShaderCompilerRD compiler;
 	} shader;
 
@@ -151,7 +152,7 @@ class RasterizerSceneForwardRD : public RasterizerSceneRD {
 
 	RasterizerStorageRD::ShaderData *_create_shader_func();
 	static RasterizerStorageRD::ShaderData *_create_shader_funcs() {
-		return static_cast<RasterizerSceneForwardRD *>(singleton)->_create_shader_func();
+		return static_cast<RasterizerSceneHighEndRD *>(singleton)->_create_shader_func();
 	}
 
 	struct MaterialData : public RasterizerStorageRD::MaterialData {
@@ -173,7 +174,7 @@ class RasterizerSceneForwardRD : public RasterizerSceneRD {
 
 	RasterizerStorageRD::MaterialData *_create_material_func(ShaderData *p_shader);
 	static RasterizerStorageRD::MaterialData *_create_material_funcs(RasterizerStorageRD::ShaderData *p_shader) {
-		return static_cast<RasterizerSceneForwardRD *>(singleton)->_create_material_func(static_cast<ShaderData *>(p_shader));
+		return static_cast<RasterizerSceneHighEndRD *>(singleton)->_create_material_func(static_cast<ShaderData *>(p_shader));
 	}
 
 	/* Push Constant */
@@ -282,11 +283,6 @@ class RasterizerSceneForwardRD : public RasterizerSceneRD {
 		uint32_t instance_ofs; //instance_offset in instancing/skeleton buffer
 		uint32_t gi_offset; //GI information when using lightmapping (VCT or lightmap)
 		uint32_t mask;
-
-		uint16_t reflection_probe_indices[8];
-		uint16_t omni_light_indices[8];
-		uint16_t spot_light_indices[8];
-		uint16_t decal_indices[8];
 	};
 
 	struct SceneState {
@@ -321,7 +317,7 @@ class RasterizerSceneForwardRD : public RasterizerSceneRD {
 			uint32_t directional_light_count;
 			float dual_paraboloid_side;
 			float z_far;
-			uint32_t pad[1];
+			float z_near;
 		};
 
 		UBO ubo;
@@ -497,7 +493,7 @@ class RasterizerSceneForwardRD : public RasterizerSceneRD {
 
 	RenderList render_list;
 
-	static RasterizerSceneForwardRD *singleton;
+	static RasterizerSceneHighEndRD *singleton;
 	uint64_t render_pass;
 	double time;
 	RID default_shader;
@@ -511,6 +507,8 @@ class RasterizerSceneForwardRD : public RasterizerSceneRD {
 	RID default_vec4_xform_buffer;
 	RID default_vec4_xform_uniform_set;
 
+	LightClusterBuilder cluster_builder;
+
 	enum PassMode {
 		PASS_MODE_COLOR,
 		PASS_MODE_COLOR_SPECULAR,
@@ -523,7 +521,7 @@ class RasterizerSceneForwardRD : public RasterizerSceneRD {
 		PASS_MODE_DEPTH_MATERIAL,
 	};
 
-	void _setup_environment(RID p_environment, const CameraMatrix &p_cam_projection, const Transform &p_cam_transform, RID p_reflection_probe, bool p_no_fog, const Size2 &p_screen_pixel_size, RID p_shadow_atlas, bool p_flip_y, const Color &p_default_bg_color);
+	void _setup_environment(RID p_environment, const CameraMatrix &p_cam_projection, const Transform &p_cam_transform, RID p_reflection_probe, bool p_no_fog, const Size2 &p_screen_pixel_size, RID p_shadow_atlas, bool p_flip_y, const Color &p_default_bg_color, float p_znear, float p_zfar);
 	void _setup_lights(RID *p_light_cull_result, int p_light_cull_count, const Transform &p_camera_inverse_transform, RID p_shadow_atlas, bool p_using_shadows);
 	void _setup_reflections(RID *p_reflection_probe_cull_result, int p_reflection_probe_cull_count, const Transform &p_camera_inverse_transform, RID p_environment);
 	void _setup_gi_probes(RID *p_gi_probe_probe_cull_result, int p_gi_probe_probe_cull_count, const Transform &p_camera_transform);
@@ -547,7 +545,7 @@ public:
 
 	virtual bool free(RID p_rid);
 
-	RasterizerSceneForwardRD(RasterizerStorageRD *p_storage);
-	~RasterizerSceneForwardRD();
+	RasterizerSceneHighEndRD(RasterizerStorageRD *p_storage);
+	~RasterizerSceneHighEndRD();
 };
-#endif // RASTERIZER_SCENE_FORWARD_RD_H
+#endif // RASTERIZER_SCENE_HIGHEND_RD_H

servers/visual/rasterizer_rd/shaders/SCsub

@@ -7,7 +7,7 @@ if 'RD_GLSL' in env['BUILDERS']:
     env.RD_GLSL('canvas_occlusion.glsl');
     env.RD_GLSL('blur.glsl');
     env.RD_GLSL('cubemap_roughness.glsl');
-    env.RD_GLSL('scene_forward.glsl');
+    env.RD_GLSL('scene_high_end.glsl');
     env.RD_GLSL('sky.glsl');
     env.RD_GLSL('tonemap.glsl');
     env.RD_GLSL('copy.glsl');

servers/visual/rasterizer_rd/shaders/scene_forward.glsl → servers/visual/rasterizer_rd/shaders/scene_high_end.glsl

@@ -5,7 +5,7 @@
 
 VERSION_DEFINES
 
-#include "scene_forward_inc.glsl"
+#include "scene_high_end_inc.glsl"
 
 /* INPUT ATTRIBS */
 
@@ -276,7 +276,7 @@ VERTEX_SHADER_CODE
 
 VERSION_DEFINES
 
-#include "scene_forward_inc.glsl"
+#include "scene_high_end_inc.glsl"
 
 /* Varyings */
 
@@ -1236,9 +1236,7 @@ void main() {
 
 	float normaldepth = 1.0;
 
-#if defined(SCREEN_UV_USED)
 	vec2 screen_uv = gl_FragCoord.xy * scene_data.screen_pixel_size;
-#endif
 
 	float sss_strength = 0.0;
 
@@ -1394,23 +1392,20 @@ FRAGMENT_SHADER_CODE
 		}
 	}
 #endif
+
+	uvec4 cluster_cell = texture(usampler3D(cluster_texture, material_samplers[SAMPLER_NEAREST_CLAMP]),vec3(screen_uv,(abs(vertex.z)-scene_data.z_near)/(scene_data.z_far-scene_data.z_near)));
+
 	{ // process reflections
 
 		vec4 reflection_accum = vec4(0.0, 0.0, 0.0, 0.0);
 		vec4 ambient_accum = vec4(0.0, 0.0, 0.0, 0.0);
 
-		uint reflection_probe_count = instances.data[instance_index].flags & INSTANCE_FLAGS_FORWARD_MASK;
+		uint reflection_probe_count = cluster_cell.z >> CLUSTER_COUNTER_SHIFT;
+		uint reflection_probe_pointer = cluster_cell.z & CLUSTER_POINTER_MASK;
 
 		for (uint i = 0; i < reflection_probe_count; i++) {
 
-			uint ref_index = instances.data[instance_index].reflection_probe_indices[i >> 1];
-
-			if (bool(i & 1)) {
-				ref_index >>= 16;
-			} else {
-				ref_index &= 0xFFFF;
-			}
-
+			uint ref_index = cluster_data.indices[reflection_probe_pointer + i];
 			reflection_process(ref_index, vertex, normal, roughness, ambient_light, specular_light, ambient_accum, reflection_accum);
 		}
 
@@ -1527,22 +1522,18 @@ FRAGMENT_SHADER_CODE
 	}
 
 	{ //omni lights
-		uint omni_light_count = (instances.data[instance_index].flags >> INSTANCE_FLAGS_FORWARD_OMNI_LIGHT_SHIFT) & INSTANCE_FLAGS_FORWARD_MASK;
+
+		uint omni_light_count = cluster_cell.x >> CLUSTER_COUNTER_SHIFT;
+		uint omni_light_pointer = cluster_cell.x & CLUSTER_POINTER_MASK;
+
 		for (uint i = 0; i < omni_light_count; i++) {
 
-			uint light_index = instances.data[instance_index].omni_light_indices[i >> 1];
+			uint light_index = cluster_data.indices[omni_light_pointer + i];
 
-			if (bool(i & 1)) {
-				light_index >>= 16;
-			} else {
-				light_index &= 0xFFFF;
+			if (!bool(lights.data[light_index].mask&instances.data[instance_index].layer_mask)) {
+				continue; //not masked
 			}
 
-			//this is done on CPU, so no need to do it here
-			//if (!bool(lights.data[light_index].mask&instances.data[instance_index].layer_mask)) {
-			//	continue; //not masked
-			//}
-
 			light_process_omni(light_index, vertex, view, normal, albedo, roughness, metallic, specular, specular_blob_intensity,
 #ifdef LIGHT_TRANSMISSION_USED
 					transmission,
@@ -1565,22 +1556,17 @@ FRAGMENT_SHADER_CODE
 	}
 
 	{ //spot lights
-		uint spot_light_count = (instances.data[instance_index].flags >> INSTANCE_FLAGS_FORWARD_SPOT_LIGHT_SHIFT) & INSTANCE_FLAGS_FORWARD_MASK;
+		uint spot_light_count = cluster_cell.y >> CLUSTER_COUNTER_SHIFT;
+		uint spot_light_pointer = cluster_cell.y & CLUSTER_POINTER_MASK;
+
 		for (uint i = 0; i < spot_light_count; i++) {
 
-			uint light_index = instances.data[instance_index].spot_light_indices[i >> 1];
+			uint light_index = cluster_data.indices[spot_light_pointer + i];
 
-			if (bool(i & 1)) {
-				light_index >>= 16;
-			} else {
-				light_index &= 0xFFFF;
+			if (!bool(lights.data[light_index].mask&instances.data[instance_index].layer_mask)) {
+				continue; //not masked
 			}
 
-			//this is done on CPU, so no need to do it here
-			//if (!bool(lights.data[light_index].mask&instances.data[instance_index].layer_mask)) {
-			//	continue; //not masked
-			//}
-
 			light_process_spot(light_index, vertex, view, normal, albedo, roughness, metallic, specular, specular_blob_intensity,
 #ifdef LIGHT_TRANSMISSION_USED
 					transmission,

servers/visual/rasterizer_rd/shaders/scene_forward_inc.glsl → servers/visual/rasterizer_rd/shaders/scene_high_end_inc.glsl

@@ -58,7 +58,7 @@ layout(set = 0, binding = 3, std140) uniform SceneData {
 	uint directional_light_count;
 	float dual_paraboloid_side;
 	float z_far;
-	uint pad0;
+	float z_near;
 
 #if 0
 	vec4 ambient_light_color;
@@ -118,11 +118,6 @@ struct InstanceData {
 	uint instance_ofs; //instance_offset in instancing/skeleton buffer
 	uint gi_offset; //GI information when using lightmapping (VCT or lightmap)
 	uint layer_mask;
-
-	uint reflection_probe_indices[4];
-	uint omni_light_indices[4];
-	uint spot_light_indices[4];
-	uint decal_indices[4];
 };
 
 layout(set = 0, binding = 4, std430) buffer Instances {
@@ -206,6 +201,16 @@ layout(set = 0, binding = 8, std140) uniform GIProbes {
 
 layout(set = 0, binding = 9) uniform texture3D gi_probe_textures[MAX_GI_PROBE_TEXTURES];
 
+#define CLUSTER_COUNTER_SHIFT 20
+#define CLUSTER_POINTER_MASK ((1 << CLUSTER_COUNTER_SHIFT) - 1)
+#define CLUSTER_COUNTER_MASK 0xfff
+
+layout(set = 0, binding = 10) uniform utexture3D cluster_texture;
+
+layout(set = 0, binding = 11, std430) buffer ClusterData {
+	uint indices[];
+} cluster_data;
+
 /* Set 1, Scene data that changes per render pass */
 
 layout(set = 1, binding = 0) uniform texture2D depth_buffer;

servers/visual_server.cpp

@@ -2302,7 +2302,7 @@ VisualServer::VisualServer() {
 
 	GLOBAL_DEF("rendering/quality/gi_probes/anisotropic", false);
 	GLOBAL_DEF("rendering/quality/gi_probes/quality", 1);
-	ProjectSettings::get_singleton()->set_custom_property_info("rendering/quality/gi_probes/quality", PropertyInfo(Variant::INT, "rendering/quality/gi_probes/quality", PROPERTY_HINT_ENUM, "Ultra-Low (1 cone, fastest),Medium (4 cones), High (6 cones, slowest)"));
+	ProjectSettings::get_singleton()->set_custom_property_info("rendering/quality/gi_probes/quality", PropertyInfo(Variant::INT, "rendering/quality/gi_probes/quality", PROPERTY_HINT_ENUM, "Ultra-Low (1 cone - fastest),Medium (4 cones), High (6 cones - slowest)"));
 
 	GLOBAL_DEF("rendering/quality/shading/force_vertex_shading", false);
 	GLOBAL_DEF("rendering/quality/shading/force_vertex_shading.mobile", true);