Recompiling LAM with flags '-O3 -tpp7 -cpu-pentium4 -xN' did the trick. The
'-xN' tells ifort to only compile binaries for Pentium 4 processor. I
appreciate everyone's help with this and especially Brian's. Thank you so
much.
Robert LeBlanc
BioAg Computer Support
Brigham Young University
> -----Original Message-----
> From: lam-bounces_at_[hidden] [mailto:lam-bounces_at_[hidden]] On Behalf
> Of Brian Barrett
> Sent: Tuesday, February 28, 2006 6:58 PM
> To: General LAM/MPI mailing list
> Subject: Re: LAM: undefined references when compiling Charmm
>
> On Feb 28, 2006, at 3:54 PM, Robert LeBlanc wrote:
>
> > Ok, so made some progress. I was running 7.1.2b23 and upgraded to
> > 7.1.2b32
> > and now I only get the following messages:
>
> <snip>
>
> > I am using the following flags for C, C++ and Fortran when
> > compiling LAM,
> > would any of these cause the above problem?
> >
> > '-O3 -tpp7 -cpu=pentium4 -xN -ip -parallel'
>
> I asked Robert for some more information off the list and finally
> figured out what is going on. Intel's compilers implicitly add -
> lpthread if you give the -parallel option. The -parallel option
> enables auto-parallelization of loops and such. The LAM tests to
> figure out how to enable threads determined that no special flags
> were needed (since the -parallel option had included them for us).
> The build system doesn't copy most user CFLAGS/CXXFLAGS/FFLAGS/etc to
> the wrapper compilers, to -parallel was dropped. When the user code
> was compiled, the wrapper added neither -lpthread nor -parallel, and
> the missing symbol error occurred.
>
> Now, I didn't catch it before, but compiling LAM/MPI with the -
> parallel option is probably a really bad idea. It's not going to do
> much in terms of optimization, and has high potential to do some
> seriously bad things in parts of the code (particularly our shared
> memory device - I'm not sure the compiler can understand what is
> going on there). I'm not sure what the -xN flag does, but I would
> drop the -ip and -parallel options. It won't make LAM/MPI any
> slower, and will solve your linking issues.
>
> Brian
>
>
> --
> Brian Barrett
> LAM/MPI developer and all around nice guy
> Have a LAM/MPI day: http://www.lam-mpi.org/
>
>
> _______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
|
