Control Data Cyber 170 State Hardware Reference Manual page 102

CP Floating-Point Arithmetic Instructions
33ijk Floating double-precision
difference of (Xj) and
(Xk)
to Xi
DXi Xj
-
Xk
14 9 8 6 5 3 2
0
I
33
i j k
This instruction reads operands from two X registers, operates on them to form
a double-precision, floating-point difference, and delivers the lower half of
this result to a third
X
register. The operands for this instruction are in Xj
and X k . These operands are in floating-point format and are not necessarily
normalized. The result of subtracting the quantity in Xk from the quantity in
Xj is delivered to Xi in floating-point format and is not necessarily
normalized.
The two operands are unpacked from floating-point format, and the exponents are
compared. The coefficient with the smaller exponent is right-shifted by the
difference of the two exponents such that both coefficients are the same
significance. The Xk coefficient is then subtracted from the Xj coefficient to
form a 96-bit result. The lower half of the result is then selected and packed
along with the larger exponent minus 4 8 (decimal) to form the result sent to
Xi. If coefficient overflow occurs, the result is right-shifted one place, and
the exponent is increased by one.
Infinite (3777xxx
...
x or 4000xxx...x) or indefinite (1777x9~
...
x or
6OOOx xx...
x) operands cause corresponding exit conditions to set in the CP for
exit mode action.
For further information, refer to Floating-point Arithmetic under CP
Programming in chapter 5.