| Re: Praise for Gfortran (finally) |
|
 |
|
 |
|
 |
|
 |
Group: comp.lang.fortran · Group Profile
Author: glen herrmannsfeldtglen herrmannsfeldt Date: Sep 20, 2008 13:07
Richard Maine wrote:
(snip)
> For example, there
> was concern in the past about cross compilers having to do things like
> implement all the intrinsics twice - once for the target environment and
> once for the compilation environment. That concern contributed towards
> keeping the list of intrinsics allowed in initialization expressions
> down. Today, I think the assumption is that if such duplication is
> needed, it can be done.
It does get interesting when the source and target systems have
different floating point formats. Also, even on the same system
some do constant arithmetic in software. When the pentium FDIV
bug came out, some test programs failed to detect it, doing the
constant division expression as software emulation.
I could imagine more differences in the results of compile
time and run time intrinsic functions. Another reason not
to do equality testing of floating point values.
-- glen
| 
