Intel 64 and IA-32 Architectures Software Developers Manual Volume 1, Basic Architecture

8-20 Vol. 1
PROGRAMMING WITH THE X87 FPU
4-4 for the encoding of the integer indefinite, QNaN floating-point indefinite, and
packed BCD integer indefinite, respectively.
The binary integer encoding 100..00B represents either of two things, depending on
the circumstances of its use:
• The largest negative number supported by the format (–2
15
, –2
31
, or –2
63
)
• The integer indefinite value
If this encoding is used as a source operand (as in an integer load or integer arith-
metic instruction), the x87 FPU interprets it as the largest negative number repre-
sentable in the format being used. If the x87 FPU detects an invalid operation when
storing an integer value in memory with an FIST/FISTP instruction and the invalid-
operation exception is masked, the x87 FPU stores the integer indefinite encoding in
the destination operand as a masked response to the exception. In situations where
the origin of a value with this encoding may be ambiguous, the invalid-operation
exception flag can be examined to see if the value was produced as a response to an
exception.
8.2.2 Unsupported Double Extended-Precision
Floating-Point Encodings and Pseudo-Denormals
The double extended-precision floating-point format permits many encodings that
do not fall into any of the categories shown in Table 4-3. Table 8-3 shows these
unsupported encodings. Some of these encodings were supported by the Intel 287
math coprocessor; however, most of them are not supported by the Intel 387 math
coprocessor and later IA-32 processors. These encodings are no longer supported
due to changes made in the final version of IEEE Standard 754 that eliminated these
encodings.
Specifically, the categories of encodings formerly known as pseudo-NaNs, pseudo-
infinities, and un-normal numbers are not supported and should not be used as
operand values. The Intel 387 math coprocessor and later IA-32 processors generate
an invalid-operation exception when these encodings are encountered as operands.
Beginning with the Intel 387 math coprocessor, the encodings formerly known as
pseudo-denormal numbers are not generated by IA-32 processors. When encoun-
tered as operands, however, they are handled correctly; that is, they are treated as
denormals and a denormal exception is generated. Pseudo-denormal numbers
should not be used as operand values. They are supported by current IA-32 proces-
sors (as described here) to support legacy code.