This is the mail archive of the fortran@gcc.gnu.org mailing list for the GNU Fortran project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Comments on gfortran performance


Jerry DeLisle wrote:
Hi all,

These were interesting comments from a gfortran user relative to a very large application, 500,000+ loc that now works:

"
  Works on both X86_64 and X86! Thanks a lot. In comparing g77 with
  4.1.1 of gfortran I noticed

    1.) gfortran (almost uniformly) runs longer on X86 than
        g77. Most of the 380 tests are small and the comparison
        of times is probably meaningless. For some large test,
        however, the times are 1.5 that of g77.

    2.) gfortran (almost uniformly) runs quicker on X86_64
        than g77! In fact, I observed almost the exact
        opposite of 1.)

  Frankly I was surprised that gfortran is a good as it is.
  If there is anything I can do to help you efforts, please
  let me know. My test may be valuable before you have a
  new release.

  Keep up the good work and thanks again,
"

I think the note of appreciation should extend to everyone.

I tested a 2,000,000 line commercial application on x86-64, using Intel MPI and MKL.
It could not be supported by g77. It ran correctly the first time (I did not try
QA suite). Execution time is about the same as ifort, except for the
sections which depend on vectorization or dealing with Xeon Write Combine buffers. So, I agree with the comments about outstanding work in gfortran.




Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]