This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: [Fortran-CAF][Fortran-DEV] Merge from the trunk into the branch
- From: Paul Richard Thomas <paul dot richard dot thomas at gmail dot com>
- To: Dominique Dhumieres <dominiq at lps dot ens dot fr>
- Cc: Tobias Burnus <burnus at net-b dot de>, "fortran at gcc dot gnu dot org" <fortran at gcc dot gnu dot org>
- Date: Wed, 11 Jun 2014 09:37:30 +0200
- Subject: Re: [Fortran-CAF][Fortran-DEV] Merge from the trunk into the branch
- Authentication-results: sourceware.org; auth=none
- References: <53975721 dot 7040101 at net-b dot de> <CAGkQGiKNUV7GxPsyVpJBFJ3sG5XE=PLg2SvTx_xJ8_ddqf8uPw at mail dot gmail dot com> <20140610223557 dot 96BCF105 at mailhost dot lps dot ens dot fr>
Dear Dominique,
I wonder why trunk and fortran-dev diverge with the options that I
used? I think that it is important to understand this.
However, what is much more important is that fortran-dev not only
compiles the testsuite but that pbvalid verifies that the results are
correct and that there does not appear to be any serious performance
penalty.
Cheers
Paul
On 11 June 2014 00:35, Dominique Dhumieres <dominiq@lps.ens.fr> wrote:
> Hi Paul,
>
>> The only test that stands out is fatigue2 ...
>
> With your options (-march=native -ffast-math -static-libgfortran -funroll-loops -O3 + -Wa,-q)
> the executable compiled with trunk runs in ~84s and compiled with fortran-dev in ~160s.
> If I add the option -fwhole-program, the times are ~52s in both cases. If I compile with
> -funroll-loops -Ofast -fwhole-program the times are ~28.5s in both cases (more or less
> the times I have posted).
>
> The fatigue* programs are a bit peculiar as the optimizer is able to inline the main routine
> with -fwhole-program, leading to almost a factor two speedup.
>
> Cheers,
>
> Dominique
--
The knack of flying is learning how to throw yourself at the ground and miss.
--Hitchhikers Guide to the Galaxy