Intel Itanium 9150M, Manual

Page: 73 / 420

3:64 Volume 3: Instruction Reference
fclass
fclass — Floating-point Class
Format: (
qp
) fclass.
fcrel
.
fctype p
1
,
p
2
=
f
2
,
fclass
9
F5
Description:
The contents of FR
f
2
are classified according to the
fclass
9
completer as shown in
Table 2-25 . This produces a boolean result based on whether the contents of FR
f
2
agrees with the floating-point number format specified by
fclass
9
, as specified by the
fcrel
completer. This result is written to the two predicate register destinations,
p
1
and
p
2
.
The result written to the destinations is determined by the compare type specified by
fctype
.
The allowed types are Normal (or none ) and unc. See Table 2-26 on page 3:67 . The
assembly syntax allows the specification of membership or non-membership and the
assembler swaps the target predicates to achieve the desired effect.
A number agrees with the pattern specified by
fclass
9
if:
• the number is NaTVal and
fclass
9
{8} is 1, or
• the number is a quiet NaN and
fclass
9
{7} is 1, or
• the number is a signaling NaN and
fclass
9
{6} is 1, or
• the sign of the number agrees with the sign specified by one of the two low-order
bits of
fclass
9
, and the type of the number (disregarding the sign) agrees with the
number-type specified by the next four bits of
fclass
9
, as shown in Table 2-25 .
Note: An
fclass
9
of 0x1FF is equivalent to testing for any supported operand.
The class names used in Table 2-25 are defined in Table 5-2, “Floating-point Register
Encodings” on page 1:86 .
Table 2-24. Floating-point Class Relations
fcrel Test Relation
m FR f
2
agrees with the pattern specified by fclass
9
(is a member)
nm FR f
2
does not agree with the pattern specified by fclass
9
(is not a member)
Table 2-25. Floating-point Classes
fclass
9
Class Mnemonic
Either these cases can be tested for
0x0100
NaTVal @nat
0x080 Quiet NaN @qnan
0x040 Signaling NaN @snan
or the OR of the following two cases
0x001
Positive @pos
0x002 Negative @neg
AND’ed with OR of the following four cases
0x004
Zero @zero
0x008 Unnormalized @unorm
0x010 Normalized @norm
0x020 Infinity @inf