[Pythonmac-SIG] Compiling Scipy/available binaries for Universal Python 2.4?

[Robert Kern]

David Warde-Farley wrote:
> On 11-Jan-07, at 11:18 PM, Robert Kern wrote:
> 
>> Well, it's linking just fine, but vecLib removed the ATLAS version  
>> information
>> that the scipy build system uses to determine whether or not to  
>> build the
>> wrappers for the C versions of the BLAS subroutines that ATLAS and  
>> vecLib
>> provide. And the C versions of LAPACK subroutines are simply missing.
>>
>> Neither are necessary, though, just nice to have.
> 
> Eh, am I missing something or aren't the C versions of the BLAS  
> subroutines a good bit faster?

Okay, let me clarify. ATLAS is implemented entirely in C. vecLib is a slightly
modified version of ATLAS. ATLAS provides both the standard BLAS interface which
uses the FORTRAN calling conventions in terms of the names of symbols and
column-major arrays. It also provides versions which uses C calling conventions
and row-major arrays. The latter is what I am referring to when I say "C
versions of the BLAS subroutines."

If you only have "the FORTRAN versions of the subroutines," then you will incur
the cost of copying the numpy arrays (row major) to the column major format.
While that may not be negligible, it's not huge, either.

> Can/does gfortran take advance of  
> AltiVec/vecLib?

Yup. -framework Accelerate

-- 
Robert Kern

"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