You seem to include a fortran77 version of xerbla and
lsame in the binary builds, and these versions are
compiled with g77, so they need libg2c.a.
This causes linking problems, since if I compile my
code with a different version of gcc/g77 than you used
to build ATLAS, my routines need the installed version
of libg2c, but these two atlas routines need sometimes
another version of libg2c. I have found that libg2c
changes quite often, so I usually have these problems.
I can fix these problems in two ways, but you can do it
in a much better way. What I can do is either compile
ATLAS myself, or I can link with my own version of
xerbla and lsame.
What I suggest is that you simply replace the fortran
routines with C routines that are compiled by gcc
rather than g77. Both are two-liners, so this should
not be a real problem. This will simplify linking for
people who download the binary versions.
Regards and thanks for a wonderful code, Sivan Toledo,
Tel-Aviv University.
Logged In: YES
user_id=182470
Sivan,
Long time no hear! I've moved this to the "feature request"
tracker. I'll need to think about the issues before I take
any action. My initial reaction is that I'm pretty happy
with the way the f77 interface is presently handled (it's
very clean in it's use
of F77), and I hesitate to change it based on binary-only
issues.
As for the big problem you are having, it seems to me there
is a third and much simpler thing for *you* to do. Can't
you simply grab xerbla and lsame, compile it with your new
g77, and ar it into the ATLAS lib to fix this problem?
Further, are you sure lsame causes the problem? Maybe it's
the use of ICHAR? Otherwise, I would figure its IO that's
doing it.
Finally, if you are using the stable release, I guess you
know that the stable is substantially slower than developer,
as shown:
http://math-atlas.sourceforge.net/timing/
Let me know,
Clint