[SciPy-User] [build] visual studio 2008 and intel fortran 11.1 compiler
Tue Dec 15 05:53:37 CST 2009
David Cournapeau wrote:
> HI Eloi,
> On Tue, Dec 15, 2009 at 3:37 PM, Eloi Gaudry <email@example.com> wrote:
>> I think that the numpy/distutils/intel.py module is not up-to-date as it
>> doesn't support Intel Visual Fortran for 64-bits apps (windows).
> If you have IFort and the MKL, I advise you to use numscons to build
> and install numpy and scipy on windows 64. This is much easier, and it
> should work today (if it doesn't, I will fix the problems, whereas I
> have no interest in fixing distutils at this point).
> You only need to first install numscons (pure python, so just do
> python setup.py install):
> Then, you build numpy and scipy as follows (inside their respective
> source tree):
> python setupscons.py scons -b --fcompiler=ifort --compiler=msvc install
> - setupscons.py instead of setup.py tells distutils to use numscons
> to build extensions
> - -b tells numscons to bypass distutils compiler detection, so that
> numscons does its own detection internally
> - in bypass mode, you need to say which C and Fortran compiler you want to use.
> To tell numscons where to look for the MKL, you need to create a
> numscons.cfg file inside the source trees, something like:
> include_dirs = C:¥Program¥ Files¥ (x86)¥Intel¥compilers¥....¥MKL¥include
> library_dirs = C:¥Prorgram¥ Files¥ (x86)¥Intel......¥MKL¥ia32¥lib
> Both numpy and scipy test suites pass almost every test on windows 64 bits.
> SciPy-User mailing list
thanks for this advise.
Can I use numscons to build scipy-0.7.1 and numpy-1.3.0 or should I use
their current svn version ?
Another question, using numscons to build scipy implicitly means that
numpy should have been built using numscons too (cf. compilers options
that might be different, etc.) ?
Free Field Technologies
Axis Park Louvain-la-Neuve
Rue Emile Francqui, 1
B-1435 Mont-Saint Guibert
Company Phone: +32 10 487 959
Company Fax: +32 10 454 626
More information about the SciPy-User