[SciPy-dev] scipy distutils questions
Pearu Peterson
pearu at scipy.org
Wed Dec 14 09:35:00 CST 2005
On Wed, 14 Dec 2005, Arnd Baecker wrote:
> Is there a way to convince him to look for the 64 Bit variant
> under /opt/intel/mkl72/64 ??
In system_info.py mkl72/64 is used when cpu.is_Itanium() is true.
What is the output of
python scipy/distutils/cpuinfo.py
in your case?
> Presently the result is
>
> #------------------
> python scipy/distutils/system_info.py lapack_opt
> lapack_opt_info:
> lapack_mkl_info:
> mkl_info:
> FOUND:
> libraries = ['mkl_ia32', 'mkl', 'vml', 'guide', 'pthread']
> library_dirs = ['/opt/intel/mkl72/lib/32']
> include_dirs = ['/opt/intel/mkl72/include']
>
> FOUND:
> libraries = ['mkl_lapack', 'mkl_ia32', 'mkl', 'vml', 'guide',
> 'pthread']
> library_dirs = ['/opt/intel/mkl72/lib/32']
> include_dirs = ['/opt/intel/mkl72/include']
>
> ( library_dirs = /home/baecker/python2/lib:/usr/local/lib:/usr/lib )
> FOUND:
> libraries = ['mkl_lapack', 'mkl_ia32', 'mkl', 'vml', 'guide',
> 'pthread']
> library_dirs = ['/opt/intel/mkl72/lib/32']
> include_dirs = ['/opt/intel/mkl72/include']
> #----------------
>
>
> I tried
>
> [mkl_libs] or [mkl_lapack] or [mkl]
> library_dirs = /opt/intel/mkl72/lib/64
> include_dirs = /opt/intel/mkl72/include/
Hmm,
[mkl]
library_dirs = ..
include_dirs = ..
should have worked. Btw, though people use site.cfg, I don't use it
myself, it's better to fix/extend system_info.py so that people wouldn't
have to use site.cfg.
> but non worked?
>
>> To disable detecting mkl, define environment variable MKL=None.
>> For mkl 7.x versions one may need to fix library names (8.x does not have
>> ifcore, for instance) in system_info.py.
>
>
> For the full scipy I suspect that we will run into
> the single/double precision lapack routines problem
> (libmkl_lapack32.so provides all the s* and c*
> while libmkl_lapack64.so provides the d* and z* routines),
> but one thing after another ... ;-)
But isn't libmkl_lapack include both precision lapack routines?
This seems to be the case with mkl 8.x anyway. Could you send me the link
of mkl 7.x readme file that explains then contents of mkl?
Thanks,
Pearu
More information about the Scipy-dev
mailing list