This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
Re: Fork
On Sun, May 23, 2004 at 01:13:43PM -0400, William M. Quarles wrote:
> BTW, I have a suggestion for GFortran (although if this has been
> suggested before, count it as a vote). A lot of universities like to
> use the F and Lahey's ELF90 compilers because they teach students to use
> the newer better code of FORTRAN 90, while leaving the redundant and
> deprecated FORTRAN 77 commands behind (also good because people tell me
> that some 77 code can't communicate with 90/95 code).
Hmm. FORTRAN 77 is a subset of Fortran 90/95, so certainly the two
languages should be able to communicate. You might be thinking of the
I guess rather typical situation where somebody installs some Linux
distribution and some Fortran libraries such as BLAS and LAPACK from
the distribution. Then the try to use these libraries (compiled with
g77) with their commercial F90 compiler, which doesn't necessarily
work due to different linking conventions. Note that this is a problem
due to using different compilers, not something having to do with the
language itself.
> These compilers
> also run a lot faster than full-fledged FORTRAN compilers and can
> produce better, faster programs because they don't have to worry about
> all of that extra baggage.
Sorry, I don't buy that argument. Sure, there might be a small
performance benefit from skipping deprecated F77 syntax due to smaller
code size in the compiler, but I'd be very surprised if that
difference would actually be measurable. The speed of the compiler
depends on the internal architecture of the compiler, as well as the
amount of optimization done. Given that F and ELF90 are targeted at
the educational market, I'd guess that they haven't put so much
emphasis on optimization as other compilers. Set the optimization
level to -O0 and you might compile an order of magnitude faster than
with -O4 (or whatever high optimization setting you fancy).
> It would be nice if GFortran included a
> commandline option for Essential FOTRAN 90/95 compilation.
In principle I like this idea. There is however the problem of
choosing which subset to support, F, ELF90 or some gfortran specific?
> This would
> fill the niche for these universities, and also provide a mode for
> gfortran that compiles faster for modern code.
As I said, I don't think the difference in compilation speed would
even be measurable.
--
Janne Blomqvist