[Performance Issue] The godot ffi layer is a bottleneck for my use cases

[Ughuuu]

In my performance investigation I see a lot of usage of: PhysicsDirectBodyState2D -> get_transform, get_velocity ...

For these, the times are much higher than in the C++ equivalent. I even make the functions be no-op, but the overhead is still there.

What I see is Godot calls inside PhysicsDirectBodyState2D 5 times, with 5 different functions, and in 3D mode about 6 times. Old version of plugin was able to do 8800 circles, so that would be for 2D 8000 * 5 times per second.

The function right now looks like this:

#[godot_api]
impl IPhysicsDirectBodyState2DExtension for RapierDirectBodyState2D {
    fn get_transform(&self) -> Transform2D {
        Transform2D::IDENTITY
    }
}

I would like to have a more efficient way of doing this, if possible.