On Tue, Oct 20, 2015 at 10:08 PM, Ross Walker <ross.rosswalker.co.uk> wrote:
>
> > On Oct 20, 2015, at 17:55, Jason Swails <jason.swails.gmail.com> wrote:
> >
> > On Tue, Oct 20, 2015 at 8:07 PM, case <dacase.scarletmail.rutgers.edu>
> > wrote:
> >
> >> On Tue, Oct 20, 2015, Ross Walker wrote:
> >>>
> >>> I am trying to get an AWS AMI of AMBER up and running. This involves
> >>> using Amazon's hacked up version of Linux which is loosely based on
> >>> Redhat but has then been 'tweaked' in tons of ways. Anyway, long story
> >>> short AmberTools 15 configure is dying:
> >>>
> >>> /usr/bin/ld:
> >> /usr/lib/gcc/x86_64-amazon-linux/4.8.3/libgfortran.a(reshape_i4.o):
> >> relocation R_X86_64_32 against `.rodata.str1.8' can not be used when
> making
> >> a shared object; recompile with -fPIC
> >>> /usr/lib/gcc/x86_64-amazon-linux/4.8.3/libgfortran.a: could not read
> >> symbols: Bad value
> >>
> >> Oooh...it doesn't like libgfortran.a! Seems hard for us to fix.
> >>
> >> Can you play around with things like "yum install netcdf" (or
> netcdf-devel,
> >> etc.)? Then use the --with-netcdf flag to point configure to that.
> >>
> >
> > I'm with Dave here. The problem seems to be that PIC is really not
> > supported by gfortran on that platform. It's the first Linux I've seen
> of
> > its kind (any chance of convincing Amazon to start supporting PIC in
> their
> > platform?)
>
> Ha ha ha ha... ROFLMAO... Convince Amazon??? That's the funniest thing I
> ever heard. ;-)
>
> > I don't know that simply yum installing netcdf is going to work
> > -- it'll get past the configure stage, but it may promptly choke on
> linking
> > any executable (at the very least, it will choke on any of the shared
> > objects that it tries to build, like libsander.so).
>
> There isn't a netcdf package available in the yum repository they provide
> so I can't 'easily' yum install on it - I could download a Redhat RPM and
> try it but I think that would be hacky and unlikely to work. It is a weird
> weird weird Linux... They have Redhat AMI's which I could use - and already
> did successfully (without the GPU support) only to find out that only the
> Amazon Linux images support spot pricing (which is like 10% of on demand
> pricing) so other os's are not really a viable option.
>
> > Another thing to try -- do a global delete of "-fPIC" in configure2 and
> try
> > that. That will prevent the sander API, MDGX API, pysander, and pytraj
> > codes from working on that platform. But if that doesn't matter, the
> rest
> > of AmberTools should still work.
> >
>
> Yeah I tried this already but it didn't seem to work - at least making
> sure it doesn't get passed to the netcdf configure doesn't help - the auto
> config still detects support for -fPIC in the compiler and tries to use it
> - I need to look more closely at the various arguments that can be passed
> to configure. I think the issue is that some of the system libraries
> clearly somehow weren't compiled with -fPIC. I might be able to hack
> through some more and see if I can really remove everything.
>
> The other option 'maybe' would be to try a completely different compiler
> suite. Any idea if that would work? - I could also try forcing on an older
> copy of devtools or manually replacing the suspect library since I have
> root access to the image. Any ideas on the best place to start?
>
> The odd thing is that netcdf-4.3.0 seems to compile fine, it is just the
> netcdf-fortan-4.2 that fails. How hard would it be to switch to
> netcdf-4.3.0 for everything?
>
This is already solved, but I'll answer this question anyway. The answer
is 'no'. netcdf-4.3.0 only provides the C API. The Fortran and C++ APIs
are split out into separate packages. If we don't install the Fortran API
for Amber, then neither sander nor pmemd can write NetCDF trajectories
without significant work to move those routines into C.
All the best,
Jason
--
Jason M. Swails
BioMaPS,
Rutgers University
Postdoctoral Researcher
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Oct 21 2015 - 05:00:03 PDT
