| 摘要 |
PURPOSE:To display maximum speediness while maintaining constant accuracy by inputting accuracy information as data and outputting the result value of arithmetic wherein accuracy is considered and its accuracy as output data. CONSTITUTION:When arithmetic data (a) and (b) are passed to a before- arithmetic normalizer 1, the before-arithmetic normalizer 1 passes an accuracy part (h) to a mantissa part computing element 3 and an accuracy part computing element 4. The mantissa part computing part 3 passes accuracy correction information (k) such as a digit overflow to the accuracy part computing element 4, which passes the arithmetic result to an after-arithmetic normalizer 5, so that the finally obtained result is normalized and outputted. Consequently, when one word consists of, for example, 32 bits, i.e. a 1-bit sign, a 7-bit exponent part, a 20-bit mantissa part, and a 4-bit accuracy part, the value that the data represents including an error is (-1)<s>X1.mX2<e>+ or -2<e+p-21> [where (s) is 0 or 1, (e) an integer, (p) an integer larger than 0, and (m) binary expression below the decimal point], so that the error of the arithmetic result due to a digit overflow is accurately found. |
