On Tue, Feb 09, 2010 at 10:06:23AM -0500, case wrote:
> On Tue, Feb 09, 2010, MengJuei Hsieh wrote:
> > For nab, I have to use something like this:
> > $AMBERHOME/bin/nab -o test -lpbsa -lgfortran test.nab
> > This is due to the fact the linker doesn't know it's linking some Fortran
> > subroutines. I can imagine that it's going be a different library if I am
> > using Intel compiler. (like, -lifcore -lifport)
> > Any idea that could generalize this kind of thing in configure script or
> > just hack nab?
> The extra libraries that are needed should get put into the FLIBS variable
> in config.h; that way they will be automatically searched for. I think this
> is a better solution than trying to change nab itself.
> I've been travelling, and have not had a chance to look in any detail at what
> you have done with the sff interface. I'll try to get to that sometime this
> week.
I have been testing our pbsa integration in nab, and there is one
caveat: Icc/ifort are not very happy with
"$AMBERHOME/bin/nab -o test -lpbsa -lifcore -lifport test.nab".
This is due to the order of linker command and it looks like:
"-lnab -lpbsa".
Therefore it would be wise to archive libpbsa object files into
libnab.a, too. So it becomes a shorter command.
"AMBERHOME/bin/nab -o test -lifcore -lifport test.nab"
Instead of FLIB, my seggestion for solving the C linker on Fortran
routines is to use additional setting for linker like:
CFLINKER=$(CC) -lifcore -lifport # for intel
CFLINKER=$(CC) -lgfortran # for gfortran
In the same fashion, I can also solve my toy pbsa caller problem
with the setting:
FCLINKER=$(FC) -nofor_main # for intel
FCLINKER=$(FC) -Mnomain # for PGI
Just my $0.02 .
Yet another issues for pbsa integration is that I am still doing
single point calculation for every step of the nab minimization.
I.e. redundant initialization and deallocation will definitely
slow down the performance.
Sincerely,
--
Mengjuei Hsieh, Luo Research Group, Molecular Biology & Biochemistry,
University of California Irvine.
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Feb 09 2010 - 19:00:03 PST
