Optimize Skeleton3D

[TokageItLab]

co-authored-by: @Ryan-000

• Split from Optimize animation performance and improve internals #112308

[Ivorforce]

Core-wise (adding the SKELETON_3D ancestral class, and how it's used) this is fine.

I can't comment on whether this makes a difference in Skeleton3D code performance, but we have ~10 bits left, so we don't need to be stingy with it.
So I'll leave it to the animation team to test the impact and review the functionality of this change.

[TokageItLab]

What concerns me is that cases where all loc/rot/scale components are enable are quite rare.

In cases where both position and rotation are active, setting them simultaneously might potentially lead to improvements. So if I were to implement this, I would try passing loc/rot/scale as a bitmask as an argument and performing conditional branching within the set_bone_pose_components() method to determine what to set and always use that method, like:

void Skeleton3D::set_bone_pose_components(int p_bone, int p_component_mask, const Vector3 &p_position, const Quaternion &p_rotation, const Vector3 &p_scale) {
	const int bone_size = bones.size();
	ERR_FAIL_INDEX(p_bone, bone_size);

	if (p_component_mask & COMPONENT_POSITION) {
		bones[p_bone].pose_position = p_position;
	}
	if (p_component_mask & COMPONENT_ROTATION) {
		bones[p_bone].pose_rotation = p_rotation;
	}
	if (p_component_mask & COMPONENT_SCALE) {
		bones[p_bone].pose_scale = p_scale;
	}

But I'm not sure which cost is lower until I test it.

[Ryan-000]

I agree and think this is a good idea since most skeletal animations only use position and rotation.

If performance is a concern, marking the set_bone_pose_components method as inline should allow the compiler to eliminate the branches when it is in Skeleton3D::set_bone_pose.

Or we can have just both methods, one with flags and one without.