HP 3000 Manuals

HP FORTRAN 77/iX Reference

REAL*16 Format 

A REAL*16 datum is a processor approximation to a real number, whose
values can be positive, negative, or zero.  This is an extension to the
ANSI 77 standard.

An REAL*16 datum occupies four consecutive 32-bit words in memory, in
floating-point format.  It has an approximate normalized range of:

     0.0
     and
     +-3.362103143112093506262677817321753*10-4932 
     to
     +-1.189731495357231765085759326628007*10+4932 

In addition, it has an approximate denormalized range of:

     +-6.475175119438025110924438958227647*10-4966 
     to
     +-3.362103143112093506262677817321752*10-4932 


          Figure 10-6.  REAL*16 Format 

A REAL*16 datum is 64-bit-aligned; that is, its address is divisible by
eight.

The REAL*16 format has a 15-bit exponent and a 112-bit fraction.
Significance to the user is approximately 34 decimal digits.  The sign
bit is zero for plus, one for minus.  The exponent field contains 16,383
plus the actual exponent (power of 2) of the number.  Exponent fields
containing all zeros and all ones are "reserved." If the exponent is zero
and the fraction zero, the number is interpreted as a signed zero.  If
the exponent is zero and the fraction not zero, the number is called
"denormalized." A floating-point number stored in a "normalized" form has
a binary point to the left of the fraction field and an implied leading 1
to the left of the binary point; a denormalized number does not have this
implied leading 1 to the left of the binary point.


NOTE  If the ANSI directive is ON, the use of REAL*16 intrinsics,
      constants or directives produces an ANSI warning at compile time.


If the exponent is all ones and the fraction is zero, the number is
regarded as a signed infinity.  If the exponent is all ones and the
fraction is not zero, then the interpretation is "not-a-number" (NaN).
Attempts to operate on denormalized numbers, infinities, and NaNs cause a
system trap.

MPE/iX 5.0 Documentation