Control Data Cyber 170 State Hardware Reference Manual page 190

CP Programming
Table 5-5.
Xj Divided by Xk (44,
45 Instructions)
+ c r ,
- a
0 0 IND
- s o
+ c r ,
0 0 IND
+O 0 0
IND IND
0 0
IND
-
- o ,
I - =
+ a
-
+ o r
IND IND IND
- -
+INDIIND IND IND IND IND IND IND
-
Normalized Numbers
A normalized floating-point number has as large
a
coefficient and as small an
exponent as possible.
A
floating-point number in packed format is normalized
if the coefficient sign bit is different from bit 47. This condition indicates
that the'coefficient has been left-shifted until bit
47
contains the most-
significant bit in the coefficient; therefore, the floating-point number has no
leading sign bits in the coefficient. The normalized instructions perform the
coefficient shift. The floating-multiply and floating-divide instructions
deliver normalized results when provided with normalized operands. The
floating-add instructions may deliver unnormalized results even when both
operands are normalized. Therefore, it is necessary to perform the normalize
operation after each sequence of floating-add or floating-subtract operations
if the result is to be kept in a normalized form.
Rounding
Floating-point instructions round the results in single-precision computation.
These instructions execute in the same amount of time as the unrounded
versions. The operands- are modified to accomplish the rounding function. The
amount of bias introduced by the rounding operation varies and is affected by
the coefficient value in the operands. The descriptions of the round
instructions define the effects of rounding in detail.
Double-Precision Results
The floating-point arithmetic ins truc tions generate double-precision results.
Use of unrounded instructions allows separate recovery of upper- and lower-half
results with proper exponents. Rounded instructions allow only upper-half
results to be obtained. Two instructions, one single-precision and one
double-precision, are required to retrieve an entire double-precision result.
60463560
A