NonStop S-Series Server Description Manual (G06.24+)

TNS Data Formats and Number Representations
HP NonStop S-Series Server Description Manual—520331-003
3-12
Instructions for Decimal and Floating-Point
Arithmetic
floating-point number affects only the high-order bit and not the fraction or exponent
fields.
The absolute-value range of 32-bit floating-point numbers is ± 2
-256
through
±(1–2
-2 3
)
*
2
256
. In decimal notation, this is approximately ± 8.64
*
10
-78
through
±1.15
*
10
77
.
For extended floating-point numbers (64 bits), the range is the same; only the precision
is increased: ± 2
-256
through ± (1 – 2
-55
)
*
2
256
. (Note, however, that the value +2
-256
is not representable; it would look like 0 in either floating-point or extended floating-
point format.)
For floating-point and extended floating-point arithmetic, the Overflow (V) indicator
is set if the exponent becomes either greater than +255 (exponent overflow) or less
than –256 (exponent underflow) during an attempt to represent the normalized result of
some operation. Otherwise, V is reset to 0. If the divisor in a divide operation is 0, the
Overflow indicator is also set. If any conversion instruction causes a numeric overflow
(“illegal conversion”), the Overflow indicator is set and the result (including Condition
Code) is undefined. If the result of some operation has a zero fraction and nonzero
exponent or sign, the value is forced to zero.
Table 3-2 defines termination conditions for various floating-point arithmetic errors.
(For further explanation of the Condition Code, refer to the topic The Environment
Register on page 6-6.)
Table 3-2. Termination Codes for Floating-Point Arithmetic Errors
Condition Overflow CC Result
Exponent overflow 1 00 Calculated result with exponent truncated
Exponent underflow 1 10 Calculated result with exponent truncated
Divide by zero 1 01 Zero
Illegal conversion 1 xx Undefined