Re: what's the ifc corresponding argument for -bp in absoft f90

From: *** Hendrickson (***.hendrickson_at_att.net)
Date: 10/20/03 

Date: Mon, 20 Oct 2003 19:34:13 GMT

Steve Lionel wrote:
>
> On Mon, 20 Oct 2003 14:01:07 -0400, "Shi Jin" <jinzishuai@hotmail.com> wrote:
>
> >But we have a way to overcome this problem under absoft f90 compiler --
> >using the -dp option. Here is the description from the man page:
> > -dp Causes variables declared in a DOUBLE PRECISION or DOUBLE
> > COMPLEX statement and constants specified with the D expo-
> > nent character to be converted to the default real kind.
> >
> >This works!
> >
> >But for the intel ifc compiler for linux, we are having trouble finding out
> >what's the corresponding switch .
> >Could anybody help us out here?
>
> I'm a bit confused. Typically, an option such as this makes the default real
> kind be double precision. The Intel equivalent is -r8. And I'd think this
> would solve your problem, But the Absoft man page description makes no sense
> to me - it is saying that things you declare as double precision get treated
> as if they were single precision! Is that really what you want? (And is that
> really what it does?) If so, Intel Fortran does not have such an option.
>

This is probably a hangover from the old Cray days. The Cray compiler
supported an OFF=p command line option that did treat double variables,
functions, etc., as if they were single. The reason was many codes came
from machines where 32 bits was the natural word size and the program
needed more precision to work correctly so variables were declared
double.
But, on a Cray 64 was natural and 128 bits was emulated in software
and tended to be rather slow (in fact, we were all under hypnotic
compulsion to suicide rather than discuss the double divide time ;-{ ).
A command line option to do everything as single worked out well
for a large set of programs.

*** Hendrickson

Loading