Implicit conversion from float -99.21319796954316 to int loses precision