Implicit conversion from float -0.675844806007504 to int loses precision