| 摘要 |
A method and a digital computer for providing the result of a calculation performed upon data having a floating point representation, with the number of exact significant figures in this result. With every mantissa of initial data is associated an item of error information having a floating point representation and which has a value of zero when the real value of the initial data and the value thereof in its digital representation coincide and which does not have a value of zero when the real value of the initial data and the value thereof in its digital representation do not coincide. The first operation is carried out by employing the digital representation and the item of error information of the mantissa in order to calculate a mean value of the mantissa of the result and the item of error information associated with the mean value. Each other operation is carried out by employing, in case of need, as a non-initial piece of data, the result of a preceding operation in the form of its calculated mantissa mean value and item of error information. The number of exact significant figures in the final result is determined from the calculated values of the mean value and error information of the mantissa of the final result. |
