[SciPy-user] Anybody used weave.blitz to speed NumPy up?

Fernando Perez fperez.net at gmail.com
Thu Jun 29 01:34:46 CDT 2006


On 6/28/06, Travis Oliphant <oliphant.travis at ieee.org> wrote:

> This is what I'm getting when running this script.   I don't understand
> what is going on?  The weave.blitz version is not helping at all.

I'm not really sure what's going on here either.

I've cleaned up the timing script to be a lot more careful, use
minimum timing, reduced overheads, etc.  I'm attaching that version.
And I'm also attaching timings for arrays of (linear) size 300, 500
and 1000.  It looks like cache effects throw the numbers around quite
a bit (as usual), but what I do consistently see is blitz outperform
numpy.

Blitz is heavily dependent on templates expanding properly to avoid
temporaries (that's its "big trick"); is there any chance the compiler
is messing up here?  I'm not sufficiently into C++ arcana to really
know if this can even happen...

Cheers,

f
-------------- next part --------------
A non-text attachment was scrubbed...
Name: weave_blitz.py
Type: text/x-python
Size: 2210 bytes
Desc: not available
Url : http://www.scipy.net/pipermail/scipy-user/attachments/20060629/74af2875/weave_blitz-0001.py
-------------- next part --------------
A non-text attachment was scrubbed...
Name: numpy-blitz_300.png
Type: image/png
Size: 11747 bytes
Desc: not available
Url : http://www.scipy.net/pipermail/scipy-user/attachments/20060629/74af2875/numpy-blitz_300-0001.png
-------------- next part --------------
A non-text attachment was scrubbed...
Name: numpy-blitz_500.png
Type: image/png
Size: 12350 bytes
Desc: not available
Url : http://www.scipy.net/pipermail/scipy-user/attachments/20060629/74af2875/numpy-blitz_500-0001.png
-------------- next part --------------
A non-text attachment was scrubbed...
Name: numpy-blitz_1000.png
Type: image/png
Size: 11947 bytes
Desc: not available
Url : http://www.scipy.net/pipermail/scipy-user/attachments/20060629/74af2875/numpy-blitz_1000-0001.png


More information about the SciPy-user mailing list