On Thu, Apr 7, 2011 at 2:48 PM, Thomas Cheatham III <tec3.utah.edu> wrote:
>
> > Can you check the date of amber11/lib/libfftw3.a? It must be left over
> from
> > the intel compile. When you change compilers, you currently have to do a
> > "make uninstall" (not just a "make clean").
> >
> > 1. Can you see if "make uninstall" solves the problem?
> > 2. We should either document this, or change the "make clean" target.
> > I rather favor the former: asking people to do a "make uninstall"
> before
> > changing compilers makes sense. Note that most people will actually
> go
> > to a different tree to test different compilers.
>
> Ah, the problem is likely that the *tar.bz has the memory of previous
> Intel compiles in there; prior to tar'ing it up for release perhaps "make
> uninstall" should likely be run? What I did was simply extract and do
> ./configure followed by make install...
>
> I agree with Jason that make clean should be sufficient to fully clean; I
> never would have thought to look for a make uninstall feature (especially
> when installing for the first time from a release tarball). Or perhaps
> make veryclean could be put in to match make uninstall.
>
> OK, on to results: make uninstall, configure pgi, make install works on
> kraken. For sander/pmemd compile the mpif90, mpicc need to be replaced
> with ftn, cc. Perhaps this -crayxt5 option to configure can do this to
> the config.h* files? I did it by hand.
>
I added the -crayxt5 to do that early on in the build process -- this makes
sure that NetCDF and pnetcdf (and every other package for whom Amber's
configure calls *their* configure) are built with the appropriate compiler
wrappers. Changing those wrappers after-the-fact causes those packages to
get installed with the raw compilers.
>
> The at15-05apr11 tarball still has that python issue, as does an
> updated compile out of my at15-with-patches git tree... I did
> make uninstall; make clean; configure intel; make parallel
>
> make[1]: Entering directory `/uufs/
> chpc.utah.edu/common/home/u0181375/AMBER/git/amber/AmberTools/src/mmpbsa_py
> '
> /bin/bash setup.sh
> Using system python
> Error importing MMPBSA python modules! MMPBSA.py will not work.
>
Hmm... I don't see this behavior. Can you edit
$AMBERHOME/AmberTools/src/mmpbsa_py/setup.sh line 102 ($pythonexe -c "from
MMPBSA_mods import *" > /dev/null 2>&1) and take out the stdout/stderr
redirection and tell me what you get when you just run that script?
Thanks!
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Thu Apr 07 2011 - 15:00:04 PDT
