Canvas item hierarchical culling

Adds optional hierarchical culling to the 2D rendering (within VisualServer).

Each canvas item maintains a bound in local space of the item itself and all child / grandchild items. This allows branches to be culled at once when they don't intersect a viewport.

Also caches global transforms and rects.

doc/classes/ProjectSettings.xml

@@ -1438,6 +1438,10 @@
 			[b]Experimental.[/b] If set to on, uses the [code]GL_STREAM_DRAW[/code] flag for legacy buffer uploads. If off, uses the [code]GL_DYNAMIC_DRAW[/code] flag.
 			[b]Note:[/b] Use with care. You are advised to leave this as default for exports. A non-default setting that works better on your machine may adversely affect performance for end users.
 		</member>
+		<member name="rendering/2d/options/culling_mode" type="int" setter="" getter="" default="1">
+			The culling mode determines the method used for rejecting canvas items that are outside a viewport. The visual result should be identical, but some modes may be faster for a particular project.
+			You can either cull items individually ([code]Item mode[/code]), or use hierarchical culling ([code]Node mode[/code]) which has a little more housekeeping but can increase performance by culling large numbers of items at once.
+		</member>
 		<member name="rendering/2d/options/ninepatch_mode" type="int" setter="" getter="" default="1">
 			Choose between fixed mode where corner scalings are preserved matching the artwork, and scaling mode.
 			Not available in GLES3 when [member rendering/batching/options/use_batching] is off.

scene/2d/canvas_item.cpp

@@ -39,6 +39,7 @@
 #include "scene/resources/style_box.h"
 #include "scene/resources/texture.h"
 #include "scene/scene_string_names.h"
+#include "servers/visual/visual_server_constants.h"
 #include "servers/visual/visual_server_raster.h"
 #include "servers/visual_server.h"
 
@@ -616,6 +617,17 @@ void CanvasItem::_notification(int p_what) {
 	}
 }
 
+#ifdef DEV_ENABLED
+void CanvasItem::_name_changed_notify() {
+	// Even in DEV builds, there is no point in calling this unless we are debugging
+	// canvas item names. Even calling the stub function will be expensive, as there
+	// are a lot of canvas items.
+#ifdef VISUAL_SERVER_CANVAS_DEBUG_ITEM_NAMES
+	VisualServer::get_singleton()->canvas_item_set_name(canvas_item, get_name());
+#endif
+}
+#endif
+
 void CanvasItem::update() {
 	if (!is_inside_tree()) {
 		return;

scene/2d/canvas_item.h

@@ -240,6 +240,10 @@ class CanvasItem : public Node {
 	void _notification(int p_what);
 	static void _bind_methods();
 
+#ifdef DEV_ENABLED
+	virtual void _name_changed_notify();
+#endif
+
 public:
 	enum {
 		NOTIFICATION_TRANSFORM_CHANGED = SceneTree::NOTIFICATION_TRANSFORM_CHANGED, //unique

scene/main/node.cpp

@@ -1051,7 +1051,15 @@ StringName Node::get_name() const {
 
 void Node::_set_name_nocheck(const StringName &p_name) {
 	data.name = p_name;
+#ifdef DEV_ENABLED
+	_name_changed_notify();
+#endif
+}
+
+#ifdef DEV_ENABLED
+void Node::_name_changed_notify() {
 }
+#endif
 
 void Node::set_name(const String &p_name) {
 	String name = p_name.validate_node_name();
@@ -1078,6 +1086,10 @@ void Node::set_name(const String &p_name) {
 		get_tree()->node_renamed(this);
 		get_tree()->tree_changed();
 	}
+
+#ifdef DEV_ENABLED
+	_name_changed_notify();
+#endif
 }
 
 static bool node_hrcr = false;
@@ -1262,7 +1274,7 @@ void Node::_generate_serial_child_name(const Node *p_child, StringName &name) co
 void Node::_add_child_nocheck(Node *p_child, const StringName &p_name) {
 	//add a child node quickly, without name validation
 
-	p_child->data.name = p_name;
+	p_child->_set_name_nocheck(p_name);
 	p_child->data.pos = data.children.size();
 	data.children.push_back(p_child);
 	p_child->data.parent = this;

scene/main/node.h

@@ -238,6 +238,9 @@ class Node : public Object {
 	virtual void remove_child_notify(Node *p_child);
 	virtual void move_child_notify(Node *p_child);
 	virtual void owner_changed_notify();
+#ifdef DEV_ENABLED
+	virtual void _name_changed_notify();
+#endif
 
 	virtual void _physics_interpolated_changed();
 

servers/visual/rasterizer.h

@@ -956,6 +956,8 @@ class RasterizerCanvas {
 		bool light_masked : 1;
 		mutable bool custom_rect : 1;
 		mutable bool rect_dirty : 1;
+		mutable bool xform_dirty : 1;
+		mutable bool bound_dirty : 1;
 
 		Vector<Command *> commands;
 		mutable Rect2 rect;
@@ -984,6 +986,10 @@ class RasterizerCanvas {
 
 		Rect2 global_rect_cache;
 
+		// the rect containing this item and all children,
+		// in local space.
+		Rect2 local_bound;
+
 		const Rect2 &get_rect() const {
 			if (custom_rect) {
 				return rect;
@@ -1185,6 +1191,7 @@ class RasterizerCanvas {
 			commands.clear();
 			clip = false;
 			rect_dirty = true;
+			xform_dirty = true;
 			final_clip_owner = nullptr;
 			material_owner = nullptr;
 			light_masked = false;
@@ -1199,6 +1206,8 @@ class RasterizerCanvas {
 			final_modulate = Color(1, 1, 1, 1);
 			visible = true;
 			rect_dirty = true;
+			xform_dirty = true;
+			bound_dirty = true;
 			custom_rect = false;
 			behind = false;
 			material_owner = nullptr;