Abstract: A floating-point unit constructed of at least two identical modules. Each module contains registers for storing floating-point data, a sign and exponent processing unit for processing the sign and exponent portions of floating-point values, and a mantissa processing unit for processing the mantissa portion. Buses allow transfer of operands from the registers to the mantissa and sign and exponent processing units and the return of the result to the registers. Interconnections between the modules and configuration logic on each module enable the modules to function as a single floating-point unit. The interconnections include connections between corresponding buses of the modules and connections between corresponding mantissa processing units. The configuration logic is responsive to position signals indicating the module's position relative to other modules in the floating-point unit and precision signals indicating the precision of the floating-point data being processed by the unit.