Motorola DSP96002 - Manual (Page 738)

MOTOROLA DSP96002 USER’S MANUAL C-15
C.1.5.1.1.3 Denormalized Numbers In Double Precision (DP)
This section describes what happens when a 64-bit denormalized double precision number is written by a
double precision floating-point move, into a Data ALU floating-point register D0-D9. Following the above
operation, the Data ALU register will be read first by a single precision and then by a double precision float-
ing-point move.
The denormalized double precision data is stored in the Data ALU register with the V tag set and the ex-
ponent set to $000 (always). The V-TAG set indicates that floating-point multiply operations will require
extra cycles to wrap it ("normalize") before using it as an operand. Double precision moves will yield correct
results when reading the denormalized DP from the register to memory (the V-TAG will also be set when
a single extended denormalized result is obtained from a Data ALU operation).
Here is an example of a double precision denormalized number:
- 64 bit data from source is 0004000000000000 (= 2**(-1024))
- exp = $000 (11-bit bias)
- mantissa = $4000000000000 (the hidden bit is zero)
- data stored in the register
- e = 000 (correct representation with 11-bit bias)
- I = 0 (the number is not normalized)
- U-TAG = 0 (cleared; the number can be used in computations
as it is by the adder)
- V-TAG = 1 (set; it indicates a denormalized number in DP,
requiring extra cycles for denormalization in
multiply operations)
- fraction = 40000000
- mantissa = 0.010...00