Hi Jason,
Yah, I noticed all of those extra libraries when I did my search:
libnetcdf.a
libnetcdf_c++.a
libnetcdff.a
I do have the shared objects for each of these as well. I checked the libnetcdff.a and it does have routines AMBER is calling for. For Ubuntu with the Synaptic package manager, Netcdf has libnetcdf, libnetcdff, and libnetcdf_c++ all bundled into one package.
What's odd to me is that I've had these libraries installed for a few months (checked my package history) and AMBER never tried to pull from them till now. I imagine if the libnetcdf.so from Ubuntu was pulled it would have absolutely no idea what to do with the fortran code since those have been moved into libnetcdff.so
...Now this leads me to another question. Why did it get confused with the libraries only when I built with the cuda flag and not with serial or mpi? strange
~Matt
Matthew.Wampole.jefferson.edu> wrote:
> Hi Dan,
>
> I've attached a log of libnetcdf.a from the command you gave. Everything
> looks to be in order for libnetcdf.a that is contained in amber12.
> Branching from your idea, I searched my computer for other copies of
> libnetcdf.a and found one in my usr/lib directory installed with the
> libnetcdf developer kit. That version of libnetcdf.a does not contain nf90
> routines so I'm wondering if AMBER could be using the Synaptic installed
> version instead of the one built by AMBER.
>
> I removed the developer kit of libnetcdf from my Ubuntu machine and was
> able to install like normal without removing the bintraj functionality.
> I've had the developer kit installed for a while for other programs and
> this is the first time this has occurred, curious but it provided a
> temporary fix.
>
This actually makes a lot of sense now. It was definitely pulling the
libnetcdf.a file from /usr/lib (maybe it was pulling libnetcdf.so -- do you
have the shared objects installed, too?) More recent versions of NetCDF
actually bundle the Fortran and C symbols in separate libraries. The
Fortran symbols are found in libnetcdff.a (and/or libnetcdff.so). (In
fact, there may be a separate Fortran NetCDF package, depending on your
particular system and package manager).
You can actually skip the Amber NetCDF build altogether with the
--with-netcdf flag to configure and just use the existing NetCDF on the
system.
I don't know that I would necessarily suggest this route, though
(especially now that you seem to have fixed it).
HTH,
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
The information contained in this transmission contains privileged and confidential information. It is intended only for the use of the person named above. If you are not the intended recipient, you are hereby notified that any review, dissemination, distribution or duplication of this communication is strictly prohibited. If you are not the intended recipient, please contact the sender by reply email and destroy all copies of the original message.
CAUTION: Intended recipients should NOT use email communication for emergent or urgent health care matters.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Jun 14 2013 - 08:00:03 PDT
