RE: [AMBER-Developers] new locations for sander QM/MM routines

From: Ross Walker <ross.rosswalker.co.uk>
Date: Sat, 11 Jul 2009 01:52:07 +0100

Hi Dave, et al.

This seems to be causing a large number of build problems that I am having a
hard time tracking down, or fixing. Starting from a brand new CVS checkout:

1) Parallel building now appears to be broken.

./configure_amber -static -nopar -nosanderidc -nobintraj ifort
make -j8
...
...
fortcom: Error: _qm2_allocate_e_repul.f, line 41: Error in opening the
compiled module file. Check INCLUDE paths. [QMMM_MODULE]
  use qmmm_module, only : qmmm_nml,qm2_struct, qmmm_struct, qmmm_mpi
------^
fortcom: Error: _qm2_allocate_e_repul.f, line 50: Conflicting attributes or
multiple declaration of name. [QM2_STRUCT]
  allocate ( qm2_struct%qm_qm_2e_repul(n2el), stat=ier )
-------------^
...
...

This seems to be a dependency problem not being resolved properly but the
dependency file looks fine:

qm2_allocate_e_repul.o: \
                    copyright.h\
                    assert.h\
                    qmmm_module.o

This never used to be a problem, parallel builds worked fine, so I assume
this must be related to the libsqm.a changes.

Looking at it more closely I think the problem is with the depend file in
the sqm directory is completely screwed. The one in the CVS tree for this
kind of works but lacks the qmmm_module dependency for most of the files,
hence the issues. Trying to re-run make depend gives the error:

No file name associated with module qmmm_module

and doesn't list any of the QMMM files. Hence once cannot build anything. I
haven't been able to work out what is actually broken here. Has anybody else
had any joy with this?

If I use the depend file in the tree and make without the -j8 then it
compiles okay but the linking is hosed with loads of missing lapack
references unless I use MKL.

I will try to take a look at this over the weekend but for the moment the
CVS tree is hosed so I welcome any fixes.

All the best
Ross


> -----Original Message-----
> From: amber-developers-bounces.ambermd.org [mailto:amber-developers-
> bounces.ambermd.org] On Behalf Of case
> Sent: Tuesday, July 07, 2009 8:11 AM
> To: amber-developers.ambermd.org
> Subject: [AMBER-Developers] new locations for sander QM/MM routines
>
>
> Hi everyone,
>
> I have updated the Makefiles in amber11/src/sander and amber11/src/sqm,
> so
> that many of the qmmm routines (the ones that are almost independent of
> sander) are compiled in a separate directory, creating a libsqm.a
> library,
> which sander then links.
>
> I sure hope everything is still working, but let me know of any
> problems.
> There are no changes in code, just in how the Makefile links things
> together.
>
> NOTE: you may have to do "make clean; make depend" in
> amber11/src/sander
> and in amber11/src/sqm (after doing a CVS update).
>
> For now, the qm2* routines in both sander and sqm directories are
> derived from
> a common CVS source, so:
>
> a. You probably can still use the old Makefiles;
> b. You can still make changes to the qm2* routines in the sander
> directory,
> and commit them via CVS in the usual fashion. Committed files
> automagically show up in both sander and sqm working directories
> when you
> run "cvs update". This change should hence not affect any
> uncommitted
> changes you have lying around -- just commit them in the usual
> fashion.
>
> [After a testing period, we will remove the CVS pointers, and at that
> point,
> people will have to do their edits in the proper directories.]
>
> The whole point of this is that other codes will be able to link to
> libsqm.a.
> We still have some cleanup to do, since there are still some sander-
> specific
> references hanging around, so that the API is not a clean one. But I
> think
> this is a good first step.
>
> ...dac
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers


_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sat Jul 11 2009 - 01:10:54 PDT
Custom Search