AN OPERATION UNIT FOR FLOATING POINT DATA HAVING A VARIABLE LENGTH EXPONENT PART

A floating-point number operating unit utilizes a floating-point representation comprising a variable exponent part and a variable mantissa part and having a fixed length as a whole. The exponent part comprises a first bit string having the value corresponding to the exponent and a bit string comprising 1s or 0s having the length depending on the length of the first bit string; the latter bit string comprises 1s or 0s depending on a combination of the sign of the numerical value and the sign of the exponent. Moreover, the mantissa part has the length obtained by subtracting the length (one bit) of the sign bit and that of the exponent part from the fixed length. The floating-point number operating unit is provided with a first circuit for creating the exponent and mantissa parts in the fixed-length representation from a floating-point number expressed according to the floating-point representation explained above, a second circuit for performing a floating-point number operation based on the exponent and mantissa parts expressed according to the fixed-length representation, and a third circuit for creating a variable-length exponent part and a variable-length mantissa part from the result of the floating-point number operation.