[SciPy-user] Performance Python examples

David Cournapeau cournape@gmail....
Sun Jun 28 22:57:08 CDT 2009


On Mon, Jun 29, 2009 at 12:51 PM, David
Powell<davidanthonypowell@gmail.com> wrote:
>
>
> 2009/6/29 David Cournapeau <cournape@gmail.com>
>>
>> On Mon, Jun 29, 2009 at 11:41 AM, David
>> Powell<davidanthonypowell@gmail.com> wrote:
>>
>> >
>> > PS: I had a try at writing something at numexpr, but what I came up with
>> > was
>> > slower than just using numpy arrays.  So I'd probably stick to use
>> > weave.inline.  However I just realised that to get the full benefit of
>> > using
>> > openmp needs gcc 4.x, but I am not sure which version of mingw numpy and
>> > scipy are built with and whether this will cause compatibility problems.
>>
>> You will not be able to use openmp on windows, because scipy/numpy is
>> built with g77, and you can't mix g77 (fortran compiler of gcc 3.x)
>> and gfortran (fortran compiler of gcc 4.x). You will have to build
>> numpy and scipy by yourself in this case.
>>
>
> But if I calling C code with weave.inline, then this shouldn't have anything
> to do with the fortran compiler.

Since you were talking about f2py, I assumed you wanted to use openmp
in Fortran. I don't know if you can mix gcc 4 and gcc 3 even for C
code, especially on windows.

> As an aside, I notice that the windows 64 bit version of numpy 1.3 is built
> with mingw64, which as far as I can see uses gcc 4.4.  Can we expect that
> scipy 0.8 will be too?

The 64 bits version has many problems, and is unsuitable for real work
ATM. Official, "production-ready" versions may not be based on gcc 4.4
(or even gcc at all).

David


More information about the SciPy-user mailing list