Function apply_random_float_error

pub(crate) fn apply_random_float_error<F: Float>(
    ecx: &mut MiriInterpCx<'_>,
    val: F,
    err_scale: i32,
) -> F

Disturbes a floating-point result by a relative error in the range (-2^scale, 2^scale).

For a 2^N ULP error, you can use an err_scale of -(F::PRECISION - 1 - N). In other words, a 1 ULP (absolute) error is the same as a 2^-(F::PRECISION-1) relative error. (Subtracting 1 compensates for the integer bit.)