All operands for floating point instructions must be normalized or a floating pointzero. A floating point overflow shall be defined as exponent overflow if the exponentis greater than 7F16. The results of an operation which causes a floating point overflowshall be the largest positive number if the sign of the resulting mantissa was positive,or shall be the smallest negative number if the sign of the resulting mantissa wasnegative. Underflow shall be defined as exponent underflow if the exponent is lessthan 8016. The results of an operation which causes a floating point underflow shallbe floating point zero. Separate interrupts are set for overflow and underflow. Onlythe floating point instructions shall set the underflow interrupt.