AMBER Archive (2005)

Subject: Re: AMBER: Amber 8 and LAM compilation problem

From: Robert Duke (rduke_at_email.unc.edu)
Date: Thu May 19 2005 - 07:05:48 CDT


Glen -
It looks to me like your major problem should be solved by bugfix.35. This
is in bugfix.all, but the patch application must have failed somehow; in my
experience, applying patches is not always smooth (I have had problems in
other parts of amber that I don't make much use of, so I have not tracked
them down). I presume that when patching fails, it is generally a matter of
other modifications existing in the file, or the patches not being applied
in a given order. Anyway, I know that bugfix 35 has not been applied
because the ierr variable in the opnmrg call is still ierr instead of having
been renamed to ierr_nmr. I presume what is going on here is that ifort has
picked up the more strict f95 fortran definition, and now it will not allow
an internal scope ierr when there is an external scope ierr also. So apply
patch 35 to nmr_calls.f90, and confirm that this ierr (and its definition in
the subroutine scope) are renamed. As far as your LAM build flags, I can't
vouch for them; it could be all the problems with static linkage and various
linux distributions (red hat I know about) have been fixed, but if you have
problems, I would revert to dynamic linkage. If there are static threads
library problems you will see sigsegv's (but you can see this too if you
don't set stacksize unlimited). Other than that you look pretty up-to-date
with you build effort here. By the way, the reason the serial build works
is that the external ierr is only there in the parallel code.

Good luck; let me know if you have further problems. I have built with lam
6.5.9 on rhel3 with ifort 8.1.025, no problems (I presume your ifort is
later).

Folks - there may be a more general need to apply bugfix.35 to pmemd as
various compilers get updated, and applying bugfix.all may not successfully
complete down to this bugfix. I would recommend applying it individually if
you have problems, and look for ierr_nmr in the new source. Sorry for the
grief.

Regards - Bob Duke

----- Original Message -----
From: "Glen Otero" <glen_at_callident.com>
To: <amber_at_scripps.edu>
Sent: Thursday, May 19, 2005 1:52 AM
Subject: AMBER: Amber 8 and LAM compilation problem

> I'm having problems building pmemd with LAM and the Intel-8.1 compilers
> on a p4. I can build a serial pmemd that completes the factor_ix, hb, and
> jac benchmarks fine. But building with LAM (7.0.4 or 7.1.1) results in
> this error:
>
> <snip>
> ../Compile OPT_HI -P nmr_calls.f90
> cat nmr_calls.f90 | /lib/cpp -traditional -I/opt/lam/7.0.4/intel/8.0/
> include -P -DDIRFRC_VECT_OPT -DMPI -DSLOW_NONBLOCKING_MPI >
> _nmr_calls_.f90
> ifort -c -auto -tpp7 -xN -ip -O3 _nmr_calls_.f90
> fortcom: Error: _nmr_calls_.f90, line 3005: The same named entity from
> different modules and/or program units cannot be referenced. [IERR]
> call opnmrg(redir(2)(1:iredir(2)), iscopn, 2, iout, ierr)
> ------------------------------------------------------------^
> compilation aborted for _nmr_calls_.f90 (code 1)
> make[1]: *** [nmr_calls.o] Error 1
> make[1]: Leaving directory `/opt/amber8/src/pmemd/src'
> make: *** [install] Error 2
>
> I've seen a similar problem reported on the mailing list when using the
> PGI compilers, and I'm wondering whether this can be a similar problem,
> even though I patched Amber with bugfix.all.
>
> I built LAM with the Intel-8.1 compilers like so:
>
> export CC=icc
> export CXX=icpc
> export FC=ifort
> export F77=ifort
> export LDFLAGS="-all-static"
> export FFLAGS="-static"
> export CFLAGS="-static"
> export CXXFLAGS="-static"
> export CXXLDFLAGS="-static"
>
> ./configure --without-romio --prefix=/opt/lam/7.0.4/intel/8.0
> make
> make install
>
> export AMBERHOME=/opt/amber8
> export PATH=$AMBERHOME/exe:$PATH
>
> I'm using the new_configure script like so:
> ./new_configure linux_p4 ifort lam
>
> Suggestions?
>
> Thanks!
>
> Glen
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu