[pypy-dev] rpython numpy
David Cournapeau
david at ar.media.kyoto-u.ac.jp
Wed Sep 26 04:22:33 CEST 2007
Simon Burton wrote:
> I've been having some fun working on a (tiny) rpython version of numpy.
That's great, this is really one of the most interesting part of pypy
for me.
>
>
> The above mentioned "efficient code" is perhaps a non-issue. Numpy
> code tends to become memory-bound, not cpu-bound.
I tend to agree, at least from my limited observation and hacking on
numpy: some (quite fundamental) operations are not cache friendly at
all. But I thought this was what rpython + pypy could improve quite a
lot, actually. For example, one problem with numpy, specially with
broadcasting and the ufunc machinery, is to decide on which dimension to
compute for maximum performances. See for example
http://www.mail-archive.com/numpy-discussion@scipy.org/msg03759.html.
David
More information about the Pypy-dev
mailing list