NORMALIZING DEVICE USING REDUNDANT SHIFT NUMBER PREDICTION AND SHIFT ERROR CORRECTION

摘要

PURPOSE: To enable the fast execution of floating-point arithmetic operation by shortening the time required for a normalizing process as one of principal processes of the floating-point arithmetic operation. ;CONSTITUTION: This device is provided with a means 11 which predicts a shift number required to normalize an addition/subtraction result within an n-bit error range by using input data of addition and subtraction, a means 13 which detects the difference between the predicted shift number and a correct shift number, and a means 14 which shifts the result of normalization with the predicted shift number optionally by 0-n bits. Consequently, even if the shift number is mispredicted, the correct normalization result can be obtained. An adding subtracting circuit 10 and a shift number redundant predicting circuit 11 can operate at the same time, so the total processing time of the floating- point arithmetic operation is shortened to speed up the operation.;COPYRIGHT: (C)1996,JPO