[MITgcm-support] Bug R_X86_64_PC32

[Constantinos Evangelinos]

On Tue 05 Sep 2006 10:09, Martin Losch wrote:

> I guess you can try -mcmodel=large,
> but if that doesn't work, you'll have to reduce your number of nodes
> per CPU. What are your domain dimensions (that is, content of SIZE.h)?
>
> Martin
>
> On Sep 5, 2006, at 4:02 PM, Nataliya Stashchuk wrote:
> > Dear all,
> >
> > I added -mcmodel=medium to FFLAGS, but I still have "relocation
> > truncated to fit: R_X86_64_PC32 against symbol". What should I do
> > next?
> >
> > Regards,
> > Nataliya

So far as I know only the latest Intel compilers (version 9 and above) 
implement -mcmodel=large the reason being that -mcmodel=medium should cover 
most cases. An explanation (for Solaris x64 but things are similar for Linux 
x86_64) can be read in  
http://developers.sun.com/prodtech/cc/articles/mmodel.html

Things to keep in mind:
a) You might be better off if you simply compile everything with -fPIC (in the 
FFLAGS/CFLAGS) and link with the dynamic libraries for the Fortran runtime 
and MPI.
b) If the above is not possible you may have to recompile the MPI library to 
either use -fPIC or use the medium memory model (add a -fPIC or 
-mcmodel=medium to the CFLAGS etc. before running configure) and build using 
that MPI library (or ask your user administrator to do this for you).
 
Constantinos
-- 
Dr. Constantinos Evangelinos
Department of Earth, Atmospheric and Planetary Sciences
Massachusetts Institute of Technology