[SciPy-user] building numpy/scipy on Solaris

raphael langella raphael.langella@steria....
Thu Mar 22 04:39:42 CDT 2007


---- Messages d´origine ----
De: "David M. Cooke" <cookedm@physics.mcmaster.ca>
Date: jeudi, mars 22, 2007 9:41 am
Objet: Re: [SciPy-user] building numpy/scipy on Solaris

> Ahh, just realized. We can do this completely with the C compiler; the
> -xlic_lib=sunperf -xarch=v9b is only being used for Fortran. Try
> 
> CFLAGS='-xlic_lib=sunperf -xarch=v9b' CPPFLAGS='-
> DNO_APPEND_FORTRAN' python setup.py install
> 
> Since sunperf has C bindings, we don't need the Fortran compiler at 
> allfor Numpy (and I don't think it was being used in the first place).

well, options for C don't have the same syntax as Fortran, so I used
'-mcpu=v9 -lsunperf' and got :

ImportError: ld.so.1: python: fatal: relocation error: file
/Produits/sun/forte/7/SUNWspro/lib/libsunperf.so.4: symbol __f95_sign:
referenced symbol not found

but then, I realized I was using the default libsunperf (v8). As I'm
building on sparc III, I want v9b optimizations, so I did :

LD_LIBRARY_PATH=/Produits/sun/forte/7/SUNWspro/lib/v9b:$LD_LIBRARY_PATH

and got :

ld: fatal: file /Produits/sun/forte/7/SUNWspro/lib/v9b/libsunperf.so:
wrong ELF class: ELFCLASS64

but I'm building 64 bits code, am I not? I added the -m64 option and got :

/Produits/publics/sparc.SunOS.5.8/python/2.3.4/include/python2.3/pyport.h:554:2:
#error "LONG_BIT definition appears wrong for platform (bad gcc/glibc
config?)."

Damn, it looks like my python was compiled in 32 bits. Well, if I have
to keep to 32 bits, so be it. Going back a few lines up, I have an
__f95_sign symbol not found (removing -mcpu=v9 doesn't change anything).


More information about the SciPy-user mailing list