The ansif and ansid libraries use no rounding. Note that no rounding might yield different results compared to truncation. The compliant libraries use only one rounding mode, namely the "round to nearest even number". Although there is only one rounding mode, it provides the best precision and is the standard choice for the vast majority of use cases. With the exception of implicit floating point to integer conversions, all functions are correctly rounded.
When no rounding mode is used, precision loss is generally 1ulp. However, in the case of subnormal numbers, the precision loss may be greater.