Hi,
> > > On Tue, May 16, 2017, D'Auria, Raffaella wrote:
> > > > Is
> > > > there any option that will allow me to get rid of it without having to
> > > > edit any config/makefile where xHost may be?
Yes, define the environment variable SSE_TYPES to the set of
CPU types for your heterogeneous cluster and then run configure.
See the intel compiler man page for the -ax option for details.
But note that this works:
setenv SSE_TYPES " "
./configure intel
make
It does produce innocuous warnings, eg:
icpc ... -ax -fPIC ...
icpc: command line warning #10155: ignoring option '-ax'; argument required
As an aside since you have intel compilers, I recommend this:
./configure -mkl intel
Note that the default with -mkl is dynamic linking so there shouldn't
be problems on a heterogeneous cluster.
Sorry for not carefully reading your post the first time.
scott
On Wed, May 17, 2017 at 01:40:08PM -0400, Jason Swails wrote:
> One possible concern is that configure actually runs some other configure
> files as well (fftw, netcdf, cpptraj, etc.), so even if you change config.h
> to remove -xHost after running configure, you may still wind up with some
> binaries that cannot easily run on multiple CPU architectures. I don't
> remember if this was a problem or not.
>
> In the rare cases that I needed to build for a heterogeneous cluster, I'd
> just remove all instances of "-xHost" from the configure2 file
> (AmberTools/src/configure2) and run configure after I do that (a helpful
> hint is to back up configure2 first lest you lose the original). In most
> cases, -xHost is the right thing to do, though.
>
> HTH,
> Jason
>
> On Wed, May 17, 2017 at 12:21 PM, Scott Brozell <sbrozell.rci.rutgers.edu>
> wrote:
>
> > On Wed, May 17, 2017 at 08:52:30AM -0400, David Case wrote:
> > > On Tue, May 16, 2017, D'Auria, Raffaella wrote:
> > > >
> > > > While trying to build Amber 16 with the intel compiler I realized
> > > > that the xHost flag is now sprinkled in many configure files and
> > > > makefiles. Since I need to build an executable that will run across
> > > > different processors (on a heterogeneous cluster) the xHost flag is a
> > > > problem (since it will generate a processor-specific executable). Is
> > > > there any option that will allow me to get rid of it without having to
> > > > edit any config/makefile where xHost may be?
> > >
> > > Can you be more specific? The xHost flag should only be in the single
> > > config.h created by the configure script (which actually calls
> > > AmberTools/src/configure2 to do the real work.) Note that a couple of
> > > other "config.h" scripts are just links to the master one, so that
> > editing
> > > the config.h in $AMBERHOME changes all of them.
> > >
> > > Bottom line: you *should* be able to just hand-edit $AMBERHOME/config.h
> > to
> > > remove the xHost invocations after the ./configure step but before the
> > > "make install" step. It's certainly worth a try.
> > >
> > > You mention that xHost shows up in Makefiles as well, but I don't
> > readily see
> > > that, but I haven't examined every one. Try the suggestion above, and
> > let us
> > > know details of what (if anything) goes wrong.
> >
> > In the repo amber tree, I grepped all 368 Makefiles and all 701 *akefile*
> > files - no xhost, ignoring case, in any of them.
> >
> > One file that is not used in most (all?) situations has an -march:
> > ./AmberTools/src/gleap/freelib/gtkglext/examples/Makefile.mingw:15:OPTIMIZE
> > = -march=pentium -O2
> >
> > There are 11 config.h files. But as Dave indicated ./config.h
> > is the only one that should matter.
> > There is 1 that hard codes xhost,
> > but that file is not used and probably should be removed:
> > ./AmberTools/src/cphstats/config.h
> >
> >
> > scott
> >
> > ps
> > A Unix operating system is an awesome tool.
> > When it comes to finding things in textual files, some of the greatest
> > programmers of all time (several of them Turing award winners), have
> > created softwares (and textbooks) that are still the best tools.
> > Learn find and grep.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed May 17 2017 - 11:30:02 PDT
