将双精度舍入到最近的非次正规表示

Question

In C++ there are (sometimes) subnormal doubles that are finite, not nan and not zero.

How can I round them to nearest non subnormal double representation?

For example: std::isnormal(DBL_MIN/2.0) = false

Answer 1

All subnormal double magnitudes are between 0 and DBL_MIN . If a value is less than DBL_MIN , just round it to one of those, based on comparing it to DBL_MIN/2 . (To match usual IEEE-754 rounding behavior, DBL_MIN/2 itself should be rounded to 0.)