On Feb 27, 1:16 pm, Ondrej Certik <ond...@certik.cz> wrote:
> On Fri, Feb 27, 2009 at 12:12 PM, mabshoff <mabsh...@googlemail.com> wrote:
<SNIP>
> > ATLAS does not depends on CBLAS because it uses the CBLAS interface.
This should read "ATLAS does depends ..." obviously ;)
> > But if you need the F77 BLAS interface you just need to add f77blas to
> > the libraries to link.
>
> Wow, you are great, this line did it:
>
> target_link_libraries(openmx fftw3 lapack cblas f77blas atlas)
>
> Now it links. :) Btw, wouldn't the same trick fix this problem here:
This does only work in this particular case since we link the given
fortran runtime statically into libatlas.so. If you used a "plain"
atlas this would blow up.
> http://www.mail-archive.com/numpy-discuss...@scipy.org/msg12593.html
>
> ?
No, numpy/scipy has the problem that they use either the system
provided BLAS/ATLAS/UMFPACK or whatever and too many times they are
compiled using different Fortran ABIs. The end result is the too
common *boom* and due to python's extension loading mechanism you find
out when it is too late :).
One can learn about all this in general by doing a little RTFM and can
trvially side step the issue. The ultimate irony is that linking C and
Fortran code on Windows "just" works unlike Linux, OSX or Unix systems
where and endless number of incompatible Fortran ABIs make your life a
living hell unless you know what you are doing. With the dominance the
gfortran hopefully g77 and g95 will quickly go away.
<SNIP>
> > I am working on a rework of the Fortran support in Sage and in the not
> > too distant future (Sage 4.0 probably) you will just call some script
> > and it will tell you the Fortran runtime you will need to link. As is
> > the Fortran support is rather messy in Sage and I need to fix it for
> > the Cygwin port, so it will be a focus for me once I am at UW after SD
> > 13 :)
>
> Thanks for help, now everything works. I am really a beginner in this,
> but I just learned a new trick. :)
Well, this is numerical software development 101 and I am glad you
learned something :). You would be surprised how many people do not
even grasp this simple concept despite me explaining to them this in
minute details again and again over a couple weeks.
> Ondrej
Cheers,
Michael
--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to sage-devel@googlegroups.com
To unsubscribe from this group, send email to
sage-devel-unsubscr...@googlegroups.com
For more options, visit this group at http://groups.google.com/group/sage-devel
URLs: http://www.sagemath.org
-~----------~----~----~----~------~----~------~--~---
