+ /*
+ the result is stored before return for correct precision and exceptions
+
+ one corner case is when the underflow flag should be raised because
+ the precise result is an inexact subnormal double, but the calculated
+ long double result is an exact subnormal double
+ (so rounding to double does not raise exceptions)
+
+ in nearest rounding mode dadd takes care of this: the last bit of the
+ result is adjusted so rounding sees an inexact value when it should
+
+ in non-nearest rounding mode fenv is used for the workaround
+ */