[Numpy-discussion] parallel numpy (by Brian Granger) - any info?

[David Cournapeau]

Matthieu Brucher wrote:
>
>         MKL does the multithreading on its own for level 3 BLAS
>         instructions (OpenMP). For ACML, the problem is that AMD does
>         not provide a CBLAS interface and is not interested in doing
>         so. With ACML, the compilation fails with the current Numpy,
>         but hopefully with Scons it will work, at least for the LAPACK
>         part.
>
>
>     [..]
>
>     That is news to me.  I compile NumPy with ACML regularly.  I
>     haven't tried with the current trunk but it worked recently.  I
>     have not tried the parallel versions, though.
>
>
> Could you share with us your setup.cfg and what you did to make Numpy 
> work with ACML ? Last time I tried (one month ago), I had several 
> unresolved externals (and I'm sure ACML does not provide the CBLAS 
> interface which is needed by Numpy).
The CBLAS is not needed by numpy, only the fortran interface is [1]. But 
some  functions need the CBLAS interface to be optimized, like numpy.dot.

cheers,

David

[1] actually, no blas is needed for numpy, but to be usable with scipy, 
you need one. I think, without being sure, that this is so such as numpy 
can be built wo any fortran compiler.