A Small and Fast Leading One Predictor Corrector Circuit
C.N. Hinds, David Lutz · 2006
The use of leading one prediction (LOP) in floating-point addition is ubiquitous in high-performance designs. The common LOP designs are inexact in at most one position in their prediction with the indication of error difficult to detect and correct without adding significant delay or complexity. We present a survey of LOP designs and methods used to correct the prediction error, and a new method which resolves the error in parallel with the gross shifts of the normalization shifter using only the unshifted sum and a portion of the shift count. Our method requires very little area and provides a significant decrease in delay.