apogee-software apogee compilers c & c++ & fortran 77 & fortran 90 User Manual

Page 94 Apogee User’s Manual

Consider the following program fragment:

A strict reading of the standard would indicate that

 and

 should receive values that are

the same as if the program had been written:

In other words,

 and

 should be stored with

 accuracy, while

should be stored with

 accuracy.

Unfortunately, this "strict reading" behavior is not the behavior expected of most FORTRAN
compilers by most FORTRAN programmers, who would instead expect

and

to hold

representations of the respective constants that were accurate to full

accuracy after the assignments.  While such behavior is not given by a strict reading of the
standard, neither is it forbidden by the standard, since the standard does not give rules
governing required accuracy.  Furthermore, such behavior is more "user-friendly" in most
circumstances.

In ANSI C, the situation is reversed, because the "natural" floating point constant (i.e., one
without a special suffix) is a double precision value.

The

 control-variable governs the stored accuracy of single precision floating point

constants.  Specifically:

For C, implement single precision floating point constants as indicated
by

.  For FORTRAN, implement single precision floating

point constants as indicated by

.

Implement single precision floating point constants with single
precision accuracy.

Implement single precision floating point constants with double
precision accuracy when they are used in a context in which the value
would be converted to double precision before being used, such as
assignment to a double precision variable, or use as one operand of an
operator whose other operand is a double precision variable, etc.

The

 control-variable has file scope and permits values of

,

, and

.  The first-

default-value is

 and the second-default-value is

.