Xerox 560 Reference Manual page 83

The following table shows the possible cases:
Postnorma I i zati on
Pre-alignment
(exponents
I)
Scale
Answer
left
Scale
Answer
Right Guard Digit Action
o o
o
o
o
o
o
o
o
(Guard digit
=
0.)
Round on guard digit.
<D
o
Guard digit left shifted
into low end of
register.@
o
o
Not possible.
Round on guard digit. <D
Round on guard digit.<D
Guard digit left shifted
into low end of
register. @
Not possible.
Notes:
<DIncrement fraction if guard digit
~
8.
<YContents of guard digit become zero on first
left shift.
Normally; there is no time penalty for the rounding opera-
tion. However, if the intermediate value is . FFFFFF and
the guard digit is 8-F after postnormalization, a right
alignment is done after rounding. (See the example below.)
Example
. FFFFFFF
1.000000
. 100000
(intermediate result before rounding)
(resu
I
t after roundi ng and truncati ng -
not valid)
(result after postrounding alignment)
FN Floating normalize:
FN = 0 The results of additions and subtractions are
to be postnormalized. If characteristic under-
flow occurs, if the result is zero, or if more
than two postnormalization hexadecimal shifts
are required, the settings for FZ and FS de-
termine the resultant action. If none of the
above conditions occurs, the condition code
is set to 0010 if the result is positive, or
to 0001 if the result is negative.
FN
=
1 Inhibit postnormalization of the result of ad-
ditions and subtractions. The settings of FZ
FZ
and FS have no effect on the instruction
operation. If the result is zero, the result
is set to "true" zero and the condition code
is set to 0000. If the result is positive, the
condition code is set to 0010. If the re-
sult is negative, the condition code is set
to 0001.
Floating zero: (applies only if FN
=
0)
FZ
=
0
FZ = 1
If the final result of en addition or subtrac-
tion operation cannot be expressed in normal-
ized form because of the characteristic being
reduced below zero, underflow has occurred,
in which case the result is set equal to "true"
zero and the condition code is set to 1100.
(Exception: if a trap results from significance
checking with FS
=
1
and FZ
=
0,
an under-
flow generated in the process of postnormal-
izing is ignored.)
Characteristic underflow causes the basic pro-
cessor to trap to location X ' 44 1 with the
contents of the general registers unchanged.
If the result is positive, the condition code
is set to 1110. If the result is negative, the
condition code is set to 1101.
FS Floating significance: (applies only if FN
=
0)
FS
=
0
Inhibit significance trap. If the result of an
addition or subtraction is zero, the result is
set equal to "true" zero, the condition code
is set to 1000, and the basic processor exe-
cutes the next instruction in sequence. If
more than two hexadecimal places of post-
normalization shifting are required and char-
acteri sti c underflow does not occur, the
condition code is set to 1010 if the result is
positive, or to 1001 if the result is negative;
then, the basi c processor executes the next
instruction in sequence. (Exception: if
characteristic underflow occurs with FS
=
0,
FZ determines the resultant action. )
FS = 1 The basic processor traps to location X ' 44 1 if
more than two hexadecimal places of post-
normalization shifting are required or
if
the
result is zero. The condition code is set
to 1000 if the result is zero, to 1010 if the
result is positive, or to 1001 if the result is
negative; however, the contents of the gen-
era I registers are not changed.
(Exception:
if a trap resu I ts from characteri sti c underflow
with FZ
=
1, the results of significance test-
ing are ignored.)
Floating-Point Arithmetic Instructions 77