2011/9/30 Garo Garabedyan <g.a.garabedyan.gmail.com>
> Hi guys,
> After emails and investigations on the issue of compiling Amber11 with XL
> compilers on Blue Gene/P I would like to present my final conclusions and
> ask you about an Amber10 configure script.
> The Amber team changed their build process when moving from Amber10 to
> Amber11. In Amber10 there were two "configure" scripts, one for AmberTools
> and the other for Amber10. In Amber11, they moved to a single configure
> script under the AmberTools source. The output from this script is now used
> by the Amber11 makefiles. Unfortunately, a side-effect of this change is
> the
> removal of the XL compiler selection in the configure script. This was
> previously available in the Amber10 configure script!
>
This wasn't really a side-effect -- the XL compilers were knowingly excluded
from the list of optional compilers when the configure script was updated.
This was done because (I think) no developer has access to recent versions
of these compilers on any architecture that they will run on. There's a
chance that the source code for many of Amber/AmberTools' programs may need
to be changed slightly to accommodate those compilers, we simply don't know.
However, pmemd should still have the option of building with xlf90 via the
configure.advanced script. In all reality, I would suggest building
Amber/AmberTools with the gnu compilers (if available) if you even need the
rest of those programs and then building pmemd/pmemd.MPI with xlf90 using
"./configure.advanced aix xlf90 mpi". I'm guessing only pmemd will really
see benefit from commercial compiler optimization. Don't forget that if you
use this approach, you have to specify the advanced makefile via "make -f
Makefile.advanced parallel"
HTH,
Jason
I intend to run the Amber10 configuration script with xlf90_bluegene and
> replace configuration and makefiles and proceed with building Amber11.
> I am trying to install Amber11 with XL compilers on Blue Gene/P hosted in
> Bulgaria, Europe as part of my B.Sc. final project. After compiling Amber11
> with XL compilers I plan to instrument it with Scalasca, which is not
> possible when compiling with GNU.
>
>
> *Garo Garabedyan*
>
> Telephone 089 59-59-703
> garabedyan.wordpress.com
>
> Sofia, Bulgaria
>
> *Гаро Гарабедян*
>
> Телефон 089 59-59-703
> garabedyan.wordpress.com
>
> София, България
>
>
>
> 2011/9/14 Ross Walker <ross.rosswalker.co.uk>
>
> > Hi Garo,
> >
> > We got close to getting everything working with XLF90 and AMBERTools
> > 1.5/Amber11 for Power 7 but then IBM pulled out of the Blue Waters
> project.
> > I can try to dig out the notes and you can see if you can convert it over
> > to
> > BlueGene/P. The last BlueGene I had access to was a BlueGene/L in AMBER
> 10
> > days and the performance was so bad it wasn't worth the effort to keep
> > trying to update all the configure scripts for the cross compiling needed
> > etc. Blue Gene was like IBM tried their hardest to break existing
> configure
> > / build approaches so I pretty much gave up on it.
> >
> > You can look for the PMEMD instructions for AMBER 9 here:
> > http://ambermd.org/config.bgl.html
> >
> > One could probably hack up a PMEMD 11 build for BlueGene/P from these.
> Note
> > you would have to use the configure.advanced script in the PMEMD
> directory
> > rather than the AMBERTools configure script. Since PMEMD is the only code
> > worth running on BlueGene/P anyway, not building AMBERTools or sander
> > should
> > be too much of an issue.
> >
> > All the best
> > Ross
> >
> > > -----Original Message-----
> > > From: Garo Garabedyan [mailto:g.a.garabedyan.gmail.com]
> > > Sent: Wednesday, September 14, 2011 11:33 AM
> > > To: AMBER Mailing List
> > > Subject: Re: [AMBER] Compiling Amber11 on Blue Gene/P
> > >
> > > Hi,
> > > I would like to ask all here-Amber developers and users. Have someone
> > > built Amber11 with IBM XL compiler scripts? I have tried many times but
> > > with
> > > no success (errors look like due to Amber source code).
> > >
> > > *Garo Garabedyan*
> > >
> > > Telephone 089 59-59-703
> > > garabedyan.wordpress.com
> > >
> > > Sofia, Bulgaria
> > >
> > > *Гаро Гарабедян*
> > >
> > > Телефон 089 59-59-703
> > > garabedyan.wordpress.com
> > >
> > > София, България
> > >
> > >
> > >
> > > 2011/9/14 Garo Garabedyan <g.a.garabedyan.gmail.com>
> > >
> > > > Hi,
> > > > Thank you for your immediate help.
> > > > As I have wrote on my previous email I have tried to compile
> > > Amber11 with
> > > > XL compilers (both without and with AmberTools) but the errors seems
> > > to me
> > > > code specific according to the best of my knowledge. At
> > > >
> > >
> http://wiki.vlsci.unimelb.edu.au/foswiki/bin/view/Public/TamboApplicati
> > > ons seems
> > > > that someone have succeeded in building Amber with XL compiler
> > > scripts.
> > > >
> > > > *Garo Garabedyan*
> > > >
> > > > Telephone 089 59-59-703
> > > > garabedyan.wordpress.com
> > > >
> > > > Sofia, Bulgaria
> > > >
> > > > *Гаро Гарабедян*
> > > >
> > > > Телефон 089 59-59-703
> > > > garabedyan.wordpress.com
> > > >
> > > > София, България
> > > >
> > > >
> > > >
> > > > On Wed, Sep 14, 2011 at 00:09, Jed W Pitera <pitera.us.ibm.com>
> > > wrote:
> > > >
> > > >>
> > > >> Garo --
> > > >>
> > > >> 1. Many FORTRAN compilers add a trailing underscore to
> > > >> external
> > > >> names (e.g. functions & subroutines). This is a common issue when
> > > trying
> > > >> to call FORTRAN from C and vice versa. There is usually a compiler
> > > option
> > > >> to disable this feature. It is "-qnoextname" on IBM xlf compilers
> > > and
> > > >> "-fno-underscoring" for GCC/gfortran. On UNIX/Linux systems you can
> > > use
> > > >> "nm" to view the symbol tables in each object file to see what
> > > external
> > > >> names are actually being produced by the compiler.
> > > >>
> > > >> 2. It looks like you are trying to mix object files
> > > created by
> > > >> the
> > > >> GNU fortran compiler (wrapped by mpif90) with those built with
> > > mpxlf90.
> > > >> If
> > > >> you can, try building everything with mpxlf90.
> > > >>
> > > >> --Jed
> > > >> _______________________________________________
> > > >> AMBER mailing list
> > > >> AMBER.ambermd.org
> > > >> http://lists.ambermd.org/mailman/listinfo/amber
> > > >>
> > > >
> > > >
> > > _______________________________________________
> > > AMBER mailing list
> > > AMBER.ambermd.org
> > > http://lists.ambermd.org/mailman/listinfo/amber
> >
> >
> > _______________________________________________
> > AMBER mailing list
> > AMBER.ambermd.org
> > http://lists.ambermd.org/mailman/listinfo/amber
> >
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Sep 30 2011 - 07:00:02 PDT
