Implicit conversion from float -127.19999999999999 to int loses precision