Loading patent…
AssigneeHewlett-Packard Company
InventorsPeter Markstein, Alan H. Karp
Estimating the term…
A floating point arithmetic unit for correctly rounding a quotient or a square root of high precision numbers to the floating point number closest to the exact result is disclosed. The invention is generally applicable to round results to a precision greater than that provided by the floating point hardware. Prior to rounding, the hardware within the floating point unit produces a high precision mantissa with all but the last few digits correct. The rounding technique according to the invention is then used to produce a correctly rounded result using an enhanced Tuckerman test. Unlike a conventional Tuckerman test, the enhanced Tuckerman test determines the last few ULPs for both square root and division while checking for early termination. The advantage of checking for early termination is that the computation time needed to make the rounding decision can be significantly reduced.
A floating point arithmetic unit for performing at least one of a division operation to divide a dividend by a divisor to produce a correctly rounded quotient and a square root operation of a value to produce a correctly rounded square root, said arithmetic unit comprising:
A method of operating a data processing system to correctly round a high precision approximated result of a division or square root operation using an enhanced Tuckerman test to produce a correctly rounded result, said method being performed in a data processing system having an arithmetic logic unit and comprising:
A method for operating a data processing system having an arithmetic logic unit to correctly round a high precision approximated result of a division or square root operation using an enhanced Tuckerman test to produce a correctly rounded result, said method being performed in a data processing system and comprising:
A method for operating a data processing system to correctly round a high precision approximated result of a division or square root operation using an enhanced Tuckerman test to produce a correctly rounded result, said method being performed in a data processing system having an arithmetic logic unit and comprising:
In a floating point arithmetic circuit having an addition unit, a multiplication unit and a control unit, a Tuckerman test is used to correctly round an approximated result of a square root or division operation to produce a correctly rounded result, the improvement comprising modifying the Tuckerman test so as to check for early termination conditions by:
Dependencies are inferred from the claim text.