This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: LAPACK further results
- From: Tobias Schlüter <tobias dot schlueter at physik dot uni-muenchen dot de>
- To: Jerry DeLisle <jvdelisle at verizon dot net>
- Cc: fortran at gcc dot gnu dot org
- Date: Thu, 23 Sep 2004 20:32:45 +0200
- Subject: Re: LAPACK further results
- References: <1095922416.25788.28.camel@localhost.localdomain>
Jerry DeLisle wrote:
> Using -O2 -malign-double:
>
> sdrvst.f takes a very very long time to compile. It does eventually
> complete compiling, but something on the order of a minute.
I looked into that a few weeks ago. Most of the time is lost when the compiler
verifies the validity of its symbol tables. This runs much faster if the
ocmpielr is built with '--disable-checking'
> [ infinite loop ]
> Some internal math error could simply be preventing the solution from
> converging?
Maybe. Unfortunately, without a reduced testcase we'll have a hard time
finding the problem, and I know that reducing LAPACK tests is very difficult.
> Any suggestions for further testing here.
Try -ffast-math. That should change convergence behavior. I would also be
interested in how it affects number of failures.
Thanks,
- Tobi
>