[SciPy-user] Mac OS X question: using Apple's vecLib

[Andrew Straw]

On Thursday, November 14, 2002, at 07:17  PM, Pearu Peterson wrote:
> On Thu, 14 Nov 2002, Andrew Straw wrote:
>
>> It appears that Apple has given Mac OS X (10.2) a pre-installed LAPACK
>> in the vecLib framework.  (See
>> http://developer.apple.com/techpubs/macosx/ReleaseNotes/vecLib.html )
>> Apparently, this is a c version of the library (clapack).
>>
>> Is there any reason why scipy shouldn't use the clapack installed with
>> Mac OS X 10.2 rather than ATLAS/LAPACK installed from sources as
>> suggested in the scipy docs?
>
> The only reason is that ATLAS/LAPACK would be upto 10 times faster than
> CLAPACK.

My impression (although it's not stated explicitly from the webpage I 
posted above) was that Apple's CLAPACK in vecLib would be optimized for 
the platform, although perhaps not to the level of ATLAS.

>> If not, I may try to create a patch for scipy to use Apple's
>> pre-installed library.
>
> Though I would recommend getting ATLAS/LAPACK to work on Mac OS X as
> the patch for CLAPACK may not be trivial, mostly because "there is a
> difference in the definition of a two-dimensional array in Fortran and
> C". This requires that all 2-dimensional arguments in linalg/lapack.pyf
> must have intent(c) attribute. Adding these attributes is simple but
> detecting when this attribute is acctually needed, is probably not.

That looks like the killer for me...

Cheers!
Andrew