Re: IEEE 754 vs Fortran arithmetic

In article <BURLEY.90Oct24025053@world.std.com>, burley@world.std.com (James C Burley) writes:
>
> I don't know any references, but I do know we ran into this problem
> implementing Fortran on a machine using an IEEE 754 math chip:
>
> REAL R(...)
> DATA R/0.5,1.5,2.5,3.5,.../
> DO I=1,...
> PRINT *,NINT(R)
> END DO
> END
>
> Fortran specifies that the following values must be output:
>
> 1, 2, 3, 4,...


We have had this discussion before some here, but to be a bit nit-picking,
"Fortran" (i.e., the standard) does not specify such things, as it
does not even specify what + means. It certainly does encourage
such things and a vendor must be prepared to answer to the customer,
but not worry about strict standard conformace in this area.
>
> However, the IEEE 754 defines nearest-integer so that using its function
> instead of Fortran's definition of NINT produces:
>
> 0, 2, 2, 4,...
>
> Also, Fortran specifically prohibits zero from being negative (or being
> significantly negative


The appendix (not a legal part of the standard) says:


"A processor must not consider a negative zero to be different
from a positive zero."


I would take this as a SUGGESTION to make 0 and I-I compare true,
however the result of the subtraction is represented.


The standard (13.5.9) does say that a "negative signed zero"
must not be produced when doing numeric output into a formatted record.
So, from the point of view of the standard, printing 4-4 as 7 is OK,
but printing it as -0 (with say an I2 format) is not!
It's just a matter of who you run to if something doesn't work
the way you think it should.
--
Walt Brainerd Sun Microsystems, Inc.
wsb@eng.sun.com MS MTV 5-40
                                          Mountain View, CA 94043
                                          415/336-5991
--

Return to the comp.compilers page.
Search the comp.compilers archives again.