Merge pull request #242 from jss2a98aj/backport-assorted

[backport] Assorted fixes and optimizations

Author: jss2a98aj

core/math/color.h

@@ -187,16 +187,16 @@ struct [[nodiscard]] Color {
 
 	_FORCE_INLINE_ Color srgb_to_linear() const {
 		return Color(
-				r < 0.04045f ? r * (1.0f / 12.92f) : Math::pow((r + 0.055f) * (float)(1.0 / (1.0 + 0.055)), 2.4f),
-				g < 0.04045f ? g * (1.0f / 12.92f) : Math::pow((g + 0.055f) * (float)(1.0 / (1.0 + 0.055)), 2.4f),
-				b < 0.04045f ? b * (1.0f / 12.92f) : Math::pow((b + 0.055f) * (float)(1.0 / (1.0 + 0.055)), 2.4f),
+				r < 0.04045f ? r * (1.0f / 12.92f) : Math::pow(float((r + 0.055) * (1.0 / (1.0 + 0.055))), 2.4f),
+				g < 0.04045f ? g * (1.0f / 12.92f) : Math::pow(float((g + 0.055) * (1.0 / (1.0 + 0.055))), 2.4f),
+				b < 0.04045f ? b * (1.0f / 12.92f) : Math::pow(float((b + 0.055) * (1.0 / (1.0 + 0.055))), 2.4f),
 				a);
 	}
 	_FORCE_INLINE_ Color linear_to_srgb() const {
 		return Color(
-				r < 0.0031308f ? 12.92f * r : (1.0f + 0.055f) * Math::pow(r, 1.0f / 2.4f) - 0.055f,
-				g < 0.0031308f ? 12.92f * g : (1.0f + 0.055f) * Math::pow(g, 1.0f / 2.4f) - 0.055f,
-				b < 0.0031308f ? 12.92f * b : (1.0f + 0.055f) * Math::pow(b, 1.0f / 2.4f) - 0.055f, a);
+				r < 0.0031308f ? 12.92f * r : (1.0 + 0.055) * Math::pow(r, 1.0f / 2.4f) - 0.055,
+				g < 0.0031308f ? 12.92f * g : (1.0 + 0.055) * Math::pow(g, 1.0f / 2.4f) - 0.055,
+				b < 0.0031308f ? 12.92f * b : (1.0 + 0.055) * Math::pow(b, 1.0f / 2.4f) - 0.055, a);
 	}
 
 	static Color hex(uint32_t p_hex);

core/os/thread.cpp

@@ -38,8 +38,8 @@
 #include "core/templates/safe_refcount.h"
 
 SafeNumeric<uint64_t> Thread::id_counter(1); // The first value after .increment() is 2, hence by default the main thread ID should be 1.
+thread_local Thread::ID Thread::caller_id = Thread::id_counter.increment();
 
-thread_local Thread::ID Thread::caller_id = Thread::UNASSIGNED_ID;
 #endif
 
 Thread::PlatformFunctions Thread::platform_functions;

core/os/thread.h

@@ -107,24 +107,22 @@ class Thread {
 	static PlatformFunctions platform_functions;
 
 	ID id = UNASSIGNED_ID;
+
 	static SafeNumeric<uint64_t> id_counter;
 	static thread_local ID caller_id;
 	THREADING_NAMESPACE::thread thread;
 
 	static void callback(ID p_caller_id, const Settings &p_settings, Thread::Callback p_callback, void *p_userdata);
 
 	static void make_main_thread() { caller_id = MAIN_ID; }
-	static void release_main_thread() { caller_id = UNASSIGNED_ID; }
+	static void release_main_thread() { caller_id = id_counter.increment(); }
 
 public:
 	static void _set_platform_functions(const PlatformFunctions &p_functions);
 
 	_FORCE_INLINE_ ID get_id() const { return id; }
 	// get the ID of the caller thread
 	_FORCE_INLINE_ static ID get_caller_id() {
-		if (unlikely(caller_id == UNASSIGNED_ID)) {
-			caller_id = id_counter.increment();
-		}
 		return caller_id;
 	}
 	// get the ID of the main thread

core/variant/variant.h

@@ -258,53 +258,53 @@ class Variant {
 
 	void _clear_internal();
 
-	_FORCE_INLINE_ void clear() {
-		static const bool needs_deinit[Variant::VARIANT_MAX] = {
-			false, //NIL,
-			false, //BOOL,
-			false, //INT,
-			false, //FLOAT,
-			true, //STRING,
-			false, //VECTOR2,
-			false, //VECTOR2I,
-			false, //RECT2,
-			false, //RECT2I,
-			false, //VECTOR3,
-			false, //VECTOR3I,
-			true, //TRANSFORM2D,
-			false, //VECTOR4,
-			false, //VECTOR4I,
-			false, //PLANE,
-			false, //QUATERNION,
-			true, //AABB,
-			true, //BASIS,
-			true, //TRANSFORM,
-			true, //PROJECTION,
-
-			// misc types
-			false, //COLOR,
-			true, //STRING_NAME,
-			true, //NODE_PATH,
-			false, //RID,
-			true, //OBJECT,
-			true, //CALLABLE,
-			true, //SIGNAL,
-			true, //DICTIONARY,
-			true, //ARRAY,
-
-			// typed arrays
-			true, //PACKED_BYTE_ARRAY,
-			true, //PACKED_INT32_ARRAY,
-			true, //PACKED_INT64_ARRAY,
-			true, //PACKED_FLOAT32_ARRAY,
-			true, //PACKED_FLOAT64_ARRAY,
-			true, //PACKED_STRING_ARRAY,
-			true, //PACKED_VECTOR2_ARRAY,
-			true, //PACKED_VECTOR3_ARRAY,
-			true, //PACKED_COLOR_ARRAY,
-			true, //PACKED_VECTOR4_ARRAY,
-		};
+	static constexpr bool needs_deinit[Variant::VARIANT_MAX] = {
+		false, //NIL,
+		false, //BOOL,
+		false, //INT,
+		false, //FLOAT,
+		true, //STRING,
+		false, //VECTOR2,
+		false, //VECTOR2I,
+		false, //RECT2,
+		false, //RECT2I,
+		false, //VECTOR3,
+		false, //VECTOR3I,
+		true, //TRANSFORM2D,
+		false, //VECTOR4,
+		false, //VECTOR4I,
+		false, //PLANE,
+		false, //QUATERNION,
+		true, //AABB,
+		true, //BASIS,
+		true, //TRANSFORM,
+		true, //PROJECTION,
+
+		// misc types
+		false, //COLOR,
+		true, //STRING_NAME,
+		true, //NODE_PATH,
+		false, //RID,
+		true, //OBJECT,
+		true, //CALLABLE,
+		true, //SIGNAL,
+		true, //DICTIONARY,
+		true, //ARRAY,
 
+		// typed arrays
+		true, //PACKED_BYTE_ARRAY,
+		true, //PACKED_INT32_ARRAY,
+		true, //PACKED_INT64_ARRAY,
+		true, //PACKED_FLOAT32_ARRAY,
+		true, //PACKED_FLOAT64_ARRAY,
+		true, //PACKED_STRING_ARRAY,
+		true, //PACKED_VECTOR2_ARRAY,
+		true, //PACKED_VECTOR3_ARRAY,
+		true, //PACKED_COLOR_ARRAY,
+		true, //PACKED_VECTOR4_ARRAY,
+	};
+
+	_FORCE_INLINE_ void clear() {
 		if (unlikely(needs_deinit[type])) { // Make it fast for types that don't need deinit.
 			_clear_internal();
 		}
@@ -800,7 +800,9 @@ class Variant {
 	_FORCE_INLINE_ Variant() :
 			type(NIL) {}
 	_FORCE_INLINE_ ~Variant() {
-		clear();
+		if (unlikely(needs_deinit[type])) { // Make it fast for types that don't need deinit.
+			_clear_internal();
+		}
 	}
 };
 

doc/classes/Basis.xml

@@ -138,7 +138,7 @@
 			<return type="Quaternion" />
 			<description>
 				Returns this basis's rotation as a [Quaternion].
-				[b]Note:[/b] Quatenions are much more suitable for 3D math but are less intuitive. For user interfaces, consider using the [method get_euler] method, which returns Euler angles.
+				[b]Note:[/b] Quaternions are much more suitable for 3D math but are less intuitive. For user interfaces, consider using the [method get_euler] method, which returns Euler angles.
 			</description>
 		</method>
 		<method name="get_scale" qualifiers="const">

doc/classes/CollisionPolygon2D.xml

@@ -5,7 +5,7 @@
 	</brief_description>
 	<description>
 		A node that provides a polygon shape to a [CollisionObject2D] parent and allows to edit it. The polygon can be concave or convex. This can give a detection shape to an [Area2D], turn [PhysicsBody2D] into a solid object, or give a hollow shape to a [StaticBody2D].
-		[b]Warning:[/b] A non-uniformly scaled [CollisionShape2D] will likely not behave as expected. Make sure to keep its scale the same on all axes and adjust its shape resource instead.
+		[b]Warning:[/b] A non-uniformly scaled [CollisionPolygon2D] will likely not behave as expected. Make sure to keep its scale the same on all axes and adjust its polygon instead.
 	</description>
 	<tutorials>
 	</tutorials>

editor/plugins/path_3d_editor_plugin.cpp

@@ -336,16 +336,19 @@ void Path3DGizmo::redraw() {
 			// Path3D as a ribbon.
 			ribbon_ptr[i] = p1;
 
-			// Fish Bone.
-			const Vector3 p_left = p1 + (side + forward - up * 0.3) * 0.06;
-			const Vector3 p_right = p1 + (-side + forward - up * 0.3) * 0.06;
-
-			const int bone_idx = i * 4;
-
-			bones_ptr[bone_idx] = p1;
-			bones_ptr[bone_idx + 1] = p_left;
-			bones_ptr[bone_idx + 2] = p1;
-			bones_ptr[bone_idx + 3] = p_right;
+			if (i % 4 == 0) {
+				// Draw fish bone every 4 points to reduce visual noise and performance impact
+				// (compared to drawing it for every point).
+				const Vector3 p_left = p1 + (side + forward - up * 0.3) * 0.06;
+				const Vector3 p_right = p1 + (-side + forward - up * 0.3) * 0.06;
+
+				const int bone_idx = i * 4;
+
+				bones_ptr[bone_idx] = p1;
+				bones_ptr[bone_idx + 1] = p_left;
+				bones_ptr[bone_idx + 2] = p1;
+				bones_ptr[bone_idx + 3] = p_right;
+			}
 		}
 
 		add_collision_segments(_collision_segments);

scene/2d/line_2d.cpp

@@ -41,13 +41,13 @@ Rect2 Line2D::_edit_get_rect() const {
 	if (_points.size() == 0) {
 		return Rect2(0, 0, 0, 0);
 	}
-	Vector2 d = Vector2(_width, _width);
-	Rect2 bounding_rect = Rect2(_points[0] - d, 2 * d);
+	Vector2 min = _points[0];
+	Vector2 max = min;
 	for (int i = 1; i < _points.size(); i++) {
-		bounding_rect.expand_to(_points[i] - d);
-		bounding_rect.expand_to(_points[i] + d);
+		min = min.min(_points[i]);
+		max = max.max(_points[i]);
 	}
-	return bounding_rect;
+	return Rect2(min, max - min).grow(_width);
 }
 
 bool Line2D::_edit_use_rect() const {

scene/2d/path_2d.cpp

@@ -138,13 +138,14 @@ void Path2D::_notification(int p_what) {
 					draw_polyline(v2p, get_tree()->get_debug_paths_color(), line_width, false);
 				}
 
-				// Draw fish bones
+				// Draw fish bone every 4 points to reduce visual noise and performance impact
+				// (compared to drawing it for every point).
 				{
 					PackedVector2Array v2p;
 					v2p.resize(3);
 					Vector2 *w = v2p.ptrw();
 
-					for (int i = 0; i < sample_count; i++) {
+					for (int i = 0; i < sample_count; i += 4) {
 						const Vector2 p = r[i].get_origin();
 						const Vector2 side = r[i].columns[1];
 						const Vector2 forward = r[i].columns[0];

scene/3d/path_3d.cpp

@@ -131,16 +131,19 @@ void Path3D::_update_debug_mesh() {
 		// Path3D as a ribbon.
 		ribbon_ptr[i] = p1;
 
-		// Fish Bone.
-		const Vector3 p_left = p1 + (side + forward - up * 0.3) * 0.06;
-		const Vector3 p_right = p1 + (-side + forward - up * 0.3) * 0.06;
-
-		const int bone_idx = i * 4;
-
-		bones_ptr[bone_idx] = p1;
-		bones_ptr[bone_idx + 1] = p_left;
-		bones_ptr[bone_idx + 2] = p1;
-		bones_ptr[bone_idx + 3] = p_right;
+		if (i % 4 == 0) {
+			// Draw fish bone every 4 points to reduce visual noise and performance impact
+			// (compared to drawing it for every point).
+			const Vector3 p_left = p1 + (side + forward - up * 0.3) * 0.06;
+			const Vector3 p_right = p1 + (-side + forward - up * 0.3) * 0.06;
+
+			const int bone_idx = i * 4;
+
+			bones_ptr[bone_idx] = p1;
+			bones_ptr[bone_idx + 1] = p_left;
+			bones_ptr[bone_idx + 2] = p1;
+			bones_ptr[bone_idx + 3] = p_right;
+		}
 	}
 
 	Array ribbon_array;

servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp

@@ -2261,7 +2261,7 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
 			for (uint32_t v = 0; v < rb->get_view_count(); v++) {
 				real_t fov = p_render_data->scene_data->cam_projection.get_fov();
 				real_t aspect = p_render_data->scene_data->cam_projection.get_aspect();
-				real_t fovy = p_render_data->scene_data->cam_projection.get_fovy(fov, aspect);
+				real_t fovy = p_render_data->scene_data->cam_projection.get_fovy(fov, 1.0 / aspect);
 				Vector2 jitter = p_render_data->scene_data->taa_jitter * Vector2(rb->get_internal_size()) * 0.5f;
 				RendererRD::FSR2Effect::Parameters params;
 				params.context = rb_data->get_fsr2_context();

servers/rendering/rendering_device_driver.cpp

@@ -259,7 +259,7 @@ Error RenderingDeviceDriver::_reflect_spirv(VectorView<ShaderStageSPIRVData> p_s
 							}
 						}
 
-						if (existing > 0) {
+						if (existing >= 0) {
 							r_reflection.specialization_constants.write[existing].stages.set_flag(stage_flag);
 						} else {
 							r_reflection.specialization_constants.push_back(sconst);

tests/core/math/test_color.h

@@ -160,6 +160,12 @@ TEST_CASE("[Color] Linear <-> sRGB conversion") {
 	CHECK_MESSAGE(
 			color_srgb.srgb_to_linear().is_equal_approx(Color(0.35, 0.5, 0.6, 0.7)),
 			"The sRGB color converted back to linear color space should match the expected value.");
+	CHECK_MESSAGE(
+			Color(1.0, 1.0, 1.0, 1.0).srgb_to_linear() == (Color(1.0, 1.0, 1.0, 1.0)),
+			"White converted from sRGB to linear should remain white.");
+	CHECK_MESSAGE(
+			Color(1.0, 1.0, 1.0, 1.0).linear_to_srgb() == (Color(1.0, 1.0, 1.0, 1.0)),
+			"White converted from linear to sRGB should remain white.");
 }
 
 TEST_CASE("[Color] Named colors") {