
Further Readings 313
5.17
5.18
1
How does it compare with the case ~ < d < 1 in which three bits ofd and
six bits of the shifted residual are required?
(c) Summarize the effect of using the divisor range [1, 2). Is it a good idea?
[Scaling and selection by rounding] Consider a high radix r digit recurrence
division method. Assume that residuals are in nonredundant form. The quotient
digit q j+l s {-a, ..., a }, a < r - 1, is selected as the integer part of the rounded
shifted partial residual. That is,
qj+l -- integer(rw[j] + 0.5)
For convergence, such an algorithm requires that the divisor be in the range
l<d<l+~, fl~_0 ...