[SciPy-user] SciPy with gcc4 and gfortran on OS X

Robert Kern robert.kern at gmail.com
Sat Apr 8 16:19:28 CDT 2006


Zachary Pincus wrote:
>  >Robert
> 
>>The scipy-gfortran.patch that Zach provided adds extra_compile_args= 
>>['-O'] to
>>the all of the mach files regardless of what compiler is compiling  
>>them.
>>
>>I'm also wondering about Windows Fortran compilers which may try to  
>>mimic MSVC's
>>/slash /style /arguments.
> 
> Pearu added a more sophisticated patch to the SVN that passes a '-- 
> noopt' argument to config_fc, instead of the more crude job that I  
> had initially done with forcibly passing '-O' to the compiler. So  
> that's good. It's still for all compilers/platforms, but at least it  
> won't be passing overtly wrong arguments to non-gfortran compilers.

Oh good. I didn't see that before posting.

>  >Pearu
> 
>>When I compiled blas sources with gfortran and disabled ATLAS (export
>>ATLAS=None; export BLAS=None; export BLAS_OPT=None; export
>>BLAS_SRC=/path/to/blas/sources/)
> 
> Hmm, I've been using the Apple-supplied optimized whatever that's in  
> their VecLib, so I haven't bothered configuring or compiling any blas  
> or atlas sources. I do get the following warning on scipy.test():
> 
>>WARNING: cblas module is empty
> 
> ...
> 
>>* If atlas library is not found by numpy/distutils/system_info.py,
>>  then scipy uses fblas instead of cblas.
> 
> Should I try anything in particular to see about tracking the  
> problems down on my box? Maybe tell scipy to not use VecLib or  
> whatever (how?) ?

Well, that behavior is a bit wrong since the VecLib framework does in fact
contain all of the ATLAS CBLAS functions. I don't think it contains the
optimized C versions of the LAPACK functions, though.

-- 
Robert Kern
robert.kern at gmail.com

"I have come to believe that the whole world is an enigma, a harmless enigma
 that is made terrible by our own mad attempt to interpret it as though it had
 an underlying truth."
  -- Umberto Eco



More information about the SciPy-user mailing list