Once again here to tell that with
NUMA system with dual opterons,
Debian Linux amd64 lenny,
openmpi-1.2.6,
intel compilers and mkl libraries 10.0.1.014,
libnuma packages (1, dbg, dev) 1.0.2-1 ('ompi_info | grep libnuma' reported correctly)
all tests about the parallel compilation for the MM part PASSED.
With immediately subsequent QMMM test, after some tests PASSED, there was some problem with mpirun. Maybe someone is interested in looking at the output of the compilation (attached). To this concern, also the (renamed) config file is attached.
I checked with 'set | grep DO_PARALLEL' after the QMMM issue
that the env var was still active.
The openmpi was compiled (without error signaling) with the same Fortran, C, and C++ compilers used for amber.
Sorry again for redundant previous mails.
Thanks
francesco pietra
--- On Sat, 5/17/08, Francesco Pietra <chiendarret.yahoo.com> wrote:
> From: Francesco Pietra <chiendarret.yahoo.com>
> Subject: Fw: RE: AMBER: MKL libraries/Amber10
> To: "Amber" <amber.scripps.edu>
> Date: Saturday, May 17, 2008, 2:14 PM
> Here again (sorry) for the 3rd time, to say that
> config_amber.h seems to be correct as to mkl:
>
> # Loader:
> #------------------------------------------------------------------------------
> LOAD= ifort $(LOCALFLAGS) $(AMBERBUILDFLAGS)
> LOADCC= gcc $(LOCALFLAGS) $(AMBERBUILDFLAGS)
> LOADLIB= -L/opt/intel/mkl/10.0.1.014//lib/em64t -lvml
> -lmkl_lapack -lmkl -lguide -lpthread
> LM= -lm
> XHOME= /usr/X11R6
> XLIBS= -L/usr/X11R6/lib64 -L/usr/X11R6/lib
>
>
> --- On Sat, 5/17/08, Francesco Pietra
> <chiendarret.yahoo.com> wrote:
>
> > From: Francesco Pietra <chiendarret.yahoo.com>
> > Subject: Fw: RE: AMBER: MKL libraries/Amber10
> > To: "Amber" <amber.scripps.edu>
> > Date: Saturday, May 17, 2008, 10:33 AM
> > Just to add that libmkl_intel.so is present in
> >
> > /opt/intel/mkl/10.0.1.014/lib/32/
> >
> > not in /64 or em64t, where there is libmkl.so
> >
> >
> > --- On Sat, 5/17/08, Francesco Pietra
> > <chiendarret.yahoo.com> wrote:
> >
> > > From: Francesco Pietra
> <chiendarret.yahoo.com>
> > > Subject: RE: AMBER: MKL libraries/Amber10
> > > To: amber.scripps.edu
> > > Date: Saturday, May 17, 2008, 10:12 AM
> > > Hi Ross:
> > > With the attached config_amber.h, Amber10
> compiled
> > serial
> > > with all tests PASSED (except dftb as I had not
> > inserted
> > > the parameter files). I exported and sourced
> MKL_HOME
> > in my
> > > .bashrc as indicated below, however I missed to
> add
> > libmkl
> > > in the config.h. Examining the matter back, that
> > occurred
> > > because it was not clear to me how to add to the
> > config.h
> > > the two lines indicated below
> > >
> > > /opt/intel/mkl...
> > >
> > > (the path corresponds to my installation of mkl).
> > >
> > > In fact, trying to compile dock for mkl I got the
> > error:
> > > cannot find libmkl_intel.so
> > >
> > > Much indebted for plainly explaining how (and in
> place
> > of?)
> > > inserting the two lines. Apologiers for my lower
> than
> > low
> > > level in compiling. Is the double "//"
> > correct in
> > > those two lines?
> > >
> > > Thanks a lot
> > > francesco
> > >
> > >
> > >
> > > --- On Fri, 5/16/08, Ross Walker
> > > <ross.rosswalker.co.uk> wrote:
> > >
> > > > From: Ross Walker
> <ross.rosswalker.co.uk>
> > > > Subject: RE: AMBER: MKL libraries/Amber10
> > > > To: amber.scripps.edu
> > > > Date: Friday, May 16, 2008, 11:25 AM
> > > > Hi Francesco,
> > > >
> > > > > Before performing tests, may I ask why
> > during the
> > > > compilation both g++ and
> > > > > g77 were used. In particular about g77,
> > should I
> > > have
> > > > indicated ifort as
> > > > > compiler (as I'll do for Amber10
> and
> > > openmpi)? I
> > > > was unable to see an
> > > > > option for ifort in place of g77 for
> the
> > above
> > > > compilation.
> > > >
> > > > Ambertools always uses g77 for the Fortran
> > compiler.
> > > There
> > > > is no performance sensitive code in there
> that
> > would
> > > > benefit from ifort so in short don't
> worry
> > about
> > > it. If
> > > > it all compiled correctly and the tests pass
> you
> > are
> > > good to
> > > > go.
> > > >
> > > > > When all that is in order, I understand
> to
> > unpack
> > > > Amber10.tar.bz2 into
> > > > > /usr/local/amber10
> > > > > and proceed to compile.
> > > >
> > > > Yes and here you would benefit from
> specifying
> > ifort
> > > since
> > > > this is for the compilation of the MD engine
> code
> > > which is
> > > > more computationally intensive.
> > > >
> > > > > Incidentally (which I forgot to mention
> in
> > my
> > > original
> > > > post), are MKL
> > > > > libraries of help even for a NUMA-type
> > machine
> > > with
> > > > dual-opterons?
> > > >
> > > > Yes they can although it depends very much
> on the
> > type
> > > of
> > > > simulation you are running. QMMM runs
> (especially
> > if
> > > you
> > > > specify diag_routine=0 in &qmmm) can
> benefit
> > > enormously
> > > > from MKL. GB runs can also benefit quite a
> bit.
> > For
> > > regular
> > > > classical PME MD simulations the difference
> is
> > much
> > > > smaller, on the order of 5 to 10% or so but
> if
> > you
> > > have MKL
> > > > available you might as well use it.
> > > >
> > > > Good luck,
> > > > Ross
> > > >
> > > >
> > > > /\
> > > > \/
> > > > |\oss Walker
> > > >
> > > > | Assistant Research Professor |
> > > > | San Diego Supercomputer Center |
> > > > | Tel: +1 858 822 0854 | EMail:-
> > ross.rosswalker.co.uk
> > > |
> > > > | http://www.rosswalker.co.uk | PGP Key
> available
> > on
> > > > request |
> > > >
> > > > Note: Electronic Mail is not secure, has no
> > guarantee
> > > of
> > > > delivery, may not be read every day, and
> should
> > not be
> > > used
> > > > for urgent or sensitive issues.
> > > >
> > > >
> > > >
> > > >
> > > >
> > >
> >
> -----------------------------------------------------------------------
> > > > The AMBER Mail Reflector
> > > > To post, send mail to amber.scripps.edu
> > > > To unsubscribe, send "unsubscribe
> > amber" (in
> > > the
> > > > *body* of the email)
> > > > to majordomo.scripps.edu
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
to majordomo.scripps.edu
Received on Wed May 21 2008 - 06:07:10 PDT
