Abstract: In an apparatus and method for computing inverses and square roots, a highly accurate initial approximation is computed using a second order polynomial equation, the coefficients of which are stored in a ROM. The most significant bits of an operand are used to address a ROM to select coefficients, providing different coefficients for different operand ranges. The remaining lesser significant operand bits are used in the computation; the coefficient values already account for the bits used to address them. The result is in single precision accuracy. For double precision, the polynomial results are used as the first approximation for a Newton-Raphson iteration. The multiplier has a split array mode to speed up the calculation of the polynomial, whereby two lesser precision values can be computed at once. The size of the coefficients is tailored to produce the proper precision result for each of the elements of Ax.sup.2 +Bx+C.

Abstract: In an apparatus and method for computing inverses and square roots a highly accurate initial approximation is computed using a second order polynomial equation, the coefficients of which are stored in a ROM. The most significant bits of an operand are used to address a ROM to select coefficients, providing different coefficients for different operand ranges. The remaining lesser significant operand bits are used in the computation; the coefficient values already account for the bits used to address them. The result is in single precision accuracy. For double precision, the polynomial results are used as the first approximation for a Newton-Raphson iteration. The multiplier has a split array mode to speed up the calculation of the polynomial, whereby two lesser precision values can be computed at once. The size of the coefficients is tailored to produce the proper precision result for each of the elements of Ax.sup.2 +Bx+C.

Abstract: A leading zero predictor (LZP) in parallel with the full subtraction operation correctly predicts the exact number of leading zeros of a subtraction result. Once the full subtraction operation is completed, the necessary shifts may be performed immediately, without a delay to determine the presence of leading zeros, and without need for a normalization corrector.

Abstract: A method and apparatus for improving the speed of a floating-point arithmetic logic unit (ALU) by arranging the logic to provide two parallel paths, each performing four functions. Six different functions are performed, and thus there is a duplication of two functions. However, each path requires only four functions, thus reducing the throughput from six to four functions. Logic circuitry is provided to determine whether the exponents of the operands are close or not, and thus select one or the other of the two paths. The fractions of the operand are processed on the two paths in parallel while the logic is determining which path to select. This determination can thus be done in parallel, with the selection being done by a multiplexer at the end of the two parallel paths.

Abstract: This invention relates to a system having a coprocessor being utilized by a processor for floating point double precision operations. The coprocessor utilizes one format for storing double precision data, the processor utilizes a second format for storing double precision data. The communication between the coprocessor and the processor limited to one half of a double precision data at a time. The processor utilizes a loop instruction to generate repetitive commands with incrementing source and destination addresses. Each command transfers data from a source address in the processor to a destination address in the coprocessor and may perform a double precision operation at the destination address. Each command also provides a double precision indication with the command.

Abstract: A programmable latency (a programmable number of clock cycles) needed for an operation completion. The required latency for a pipe is determined from a formula including the system clock cycle time which the unit will be specified to operate under. The latency is preprogrammed by setting the count of a timer accordingly to provide at least the minimum number of clock cycles necessary to cover the time required to do the computation. Separate timers are independently set for arithmetic logic unit (ALU) operations, multiply operations, logical operations and divide and square root operations.

Abstract: An improved logic structure and a method for implementing the same to perform division and square-root operations for radix four and higher is disclosed. The divsion and square-root bits are generated by a non-restoring method with the partial remainder, partial radicand, quotient and root all in redundant form. The partial remainder/radicand is stored in a series of sum and carry registers. The upper bits from these registers are supplied to a carry look-ahead adder for conversion to non-redundant form. These upper bits are then used to select a next divisor or root from a prediction programmable logic array (PLA). The output of the prediction PLA is supplied to a quotient/root register and a divisor/root multiple selector. The output of the selector is supplied to a carry save adder which has its output provided back to the input of the partial remainder/radicand sum and carry registers. The system of the present invention allows both division and square root calculations to be done with the same hardware.

Abstract: The present invention optimizes the number and ratio of cycles required among the divide/square root unit, multiplier unit and ALU. An intermediate latch with its own clock is provided at the output of the multiplier half-array in the intermediate stage to feed back data for a second pass for double-precision numbers. The multiplier can then be adjusted for either two-cycle latency mode (for optimizing double-precision multiplies) or three-cycle latency mode (for optimizing single-precision multiplies). A separate divide clock is used for the divide/square root unit, and is synchronized with the multiplier cycle clock on input and output. This allows the divide time to be optimized so that it requires fewer clock cycles when a longer multiplier clock cycle time is used.

Abstract: A method and apparatus for combining the multiply and ALU functions for floating point numbers to enable the completion of a multiply-accumulate operation in a shorter time. The multiplied fraction is left in sum and carry form and is provided in this form to the ALU, eliminating the CP adder from the multiplier. The normalization of the fraction and the corresponding changes to the exponent in the multiplier are also eliminated. The ALU can combine the sum and carry of the product fraction simultaneously if the exponents are sufficiently similar. Otherwise, the sum and carry of the fraction product is combined first and compared with the new fraction, with the smaller of the fractions being right shifted prior to their combination.