This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: fortran codes for testing
On Mon, Apr 12, 2004 at 05:31:24PM +0100, Joost VandeVondele wrote:
> >
> > This depends on how you define a real world application. Several
> > of my real world application compile, run, and generate the expected
> > output. Admittedly, the codes use a fairly well defined subset
> > of Fortran 77 with only the most common extensions (e.g., IMPLICIT NONE,
> > lower case letters, and variable names longer than 6 characters, etc).
>
> My feeling is also that for 'clean F77' programs (I'd call them good style
> F77, not using stuff such as ENTRY/COMMON/DATA in all allowed strange
> ways) gfortran is doing best (but e.g. see the results of the NIST tests).
Bud Davis and I both run NIST through gfortran. Unfortunately,
only Bud has the capability of actually fixing problems. I'm
*slowly* grasping the gfortran code.
> Things are significantly worse looking at real F90 code.
Yes, I know. I have 500K LOC in my (private) testsuite where
about 100K lines are F90/F95. I add code as I find it on the
net.
> gfortran could not be really tested in 'F77 mode' on real programs until
> very recently due to some common stuff such as character arrays (e.g.
> LAPACK testsuite) not being implemented. But also 'just a few failures' in
> the LAPACK testsuite is problematic as they are a sign of miscompilations
> which is far more dangerous than an ICE for somebody using, rather than
> developing the compiler.
Yes, I know. I'll be looking at those failures this week to
distill concise test cases.
> Nevertheless, good to see that after a long time of little activity
> gfortran is starting to move again.
Help is always welcomed. I know you have a large code base that you
can run through gfortran. If you submit a bug report, mark the subject
line with [gfortran].
--
Steve