Implicit conversion from float -0.810810810810807 to int loses precision