Analog Devices ADSP-21261 SHARC Hardware Reference Manual page 66

Numeric Formats
Table 2-2. FPACK Operations
Condition
135 < exp
120 < exp  135
109 < exp  120
exp < 110
exp = source exponent
sign bit remains the same in all cases
Table 2-3. FUNPACK Operations
Condition
0 < exp  15
exp = 0
exp = source exponent
sign bit remains the same in all cases
The short float type supports gradual underflow. This method sacrifices
precision for dynamic range. When packing a number which would have
underflowed, the exponent is set to zero and the mantissa (including "hid-
den" 1) is right-shifted the appropriate amount. The packed result is a
denormal, which can be unpacked into a normal IEEE floating-point
number.
2-8
Result
Largest magnitude representation.
Exponent is Most Significant Bit (MSB) of source exponent concatenated
with the three Least Significant Bits (LSBs) of source exponent. The
packed fraction is the rounded upper 11 bits of the source fraction.
Exponent = 0. Packed fraction is the upper bits (source exponent – 110)
of the source fraction prefixed by zeros and the "hidden" one. The packed
fraction is rounded.
Packed word is all zeros.
Result
Exponent is the 3 LSBs of the source exponent prefixed by the MSB of
the source exponent and four copies of the complement of the MSB. The
unpacked fraction is the source fraction with 12 zeros appended.
Exponent is (120 – N) where N is the number of leading zeros in the
source fraction. The unpacked fraction is the remainder of the source
fraction with zeros appended to pad it and the "hidden" one stripped
away.
ADSP-2126x SHARC Processor Hardware Reference