Freescale Semiconductor MCF5480 - Manual (Page 194)

MCF548x Reference Manual, Rev. 3
6-4 Freescale Semiconductor
yields a signed, two’s complement power of two. This represents the magnitude of a normalized
floating-point number when multiplied by the mantissa.
By definition, a normalized mantissa always takes values starting from 1.0 and going up to, but not
including, 2.0; that is, [1.0...2.0).
6.2.3 Floating-Point Data Types
Each floating-point data format supports five unique data types: normalized numbers, zeros, infinities,
NANs, and denormalized numbers. The normalized data type, Figure 6-3 , never uses the maximum or
minimum exponent value for a given format.
6.2.3.1 Normalized Numbers
Normalized numbers include all positive or negative numbers with exponents between the maximum and
minimum values. For single- and double-precision normalized numbers, the implied integer bit is one and
the exponent can be zero.
Figure 6-3. Normalized Number Format
6.2.3.2 Zeros
Zeros can be positive or negative and represent real values, + 0.0 and – 0.0. See Figure 6-4 .
Figure 6-4. Zero Format
6.2.3.3 Infinities
Infinities can be positive or negative and represent real values that exceed the overflow threshold. A
result’s exponent greater than or equal to the maximum exponent value indicates an overflow for a given
data format and operation. This overflow description ignores the effects of rounding and the
user-selectable rounding models. For single- and double-precision infinities, the fraction is a zero. See
Figure 6-5 .
Figure 6-5. Infinity Format
Min
< Exponent < Max
Fraction = Any bit pattern
Sign of Mantissa, 0 or 1
Exponent = 0 Fraction = 0
Sign of Mantissa, 0 or 1
Exponent = Maximum Fraction = 0
Sign of Mantissa, 0 or 1