[Numpy-discussion] efficient 3d histogram creation

[Charles R Harris]

2009/5/7 Dag Sverre Seljebotn <dagss at student.matnat.uio.no>

> Stéfan van der Walt wrote:
> > 2009/5/7 Chris Colbert <sccolbert at gmail.com>:
> >> This was really my first attempt at doing anything constructive with
> Cython.
> >> It was actually unbelievably easy to work with. I think i spent less
> time
> >> working on this, than I did trying to find an optimized solution using
> pure
> >> numpy and python.
> >
> > One aspect we often overlook is how easy it is to write a for-loop in
> > comparison to vectorisation.  Besides, for-loops are sometimes easier
> > to read as well!
> >
> > I think the Cython guys are planning some sort of templating, but I'll
> > CC Dag so that he can tell us more.
>
> We were discussing how it would/should look like, but noone's committed
> to implementing it so it's pretty much up in the blue I think -- someone
> might jump in and do it next week, or it might go another year, I can't
> tell.
>
> While I'm here, also note in that code Chris wrote that you want to pay
> attention to the change of default division semantics on Cython 0.12
> (especially for speed).
>
> http://wiki.cython.org/enhancements/division
>

Hi Dag,

Numpy can now do separate compilations with controlled export of symbols
when the object files are linked together to make a module. Does Cython have
anyway of controlling the visibility of symbols or should we just include
the right files in Numpy to get the needed macros?

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20090507/3843ae78/attachment.html>