[SciPy-Dev] Work on fwrap and SciPy

Matthew Brett matthew.brett@gmail....
Thu Nov 4 13:41:37 CDT 2010


Hi,

On Thu, Nov 4, 2010 at 11:11 AM, Jason McCampbell
<jmccampbell@enthought.com> wrote:
>> Thanks for the useful summary.   Do you anticipate any performance
>> changes?  I seem to remember we're a bit short on benchmarks - is that
>> correct?
>
> The intent is to that performance changes should be minimal if any.  All of
> the tight ufunc loops are unchanged, except for some handling of object
> arrays, and those might be slightly faster.  There is an extra level of
> indirection because the Py_OBJECT_HEAD structure had to be separated from
> the main structure, so there could be a slight amount of overhead there.  We
> still need to do some benchmarks runs now that things are pieced back
> together.

(I nearly accidentally deleted your name from the top of the email but
remembered just in time!)

Am I right in thinking that our current benchmark suite is not very
comprehensive?

I see we have <numpy>/benchmarks - does that cover most or all of the
cases you want to test?

Maybe that in itself could be an orthogonal piece of work?  I hesitate
to ask - but can we (I) help in that?

Best,

Matthew


More information about the SciPy-Dev mailing list