To fix that you can write simple BP function that receives floating points, calculates this operation in double precision - and then returns results as float again. Problem disappears.
To fix that you can write simple BP function that receives floating points, calculates this operation in double precision - and then returns results as float again. Problem disappears.