On 07/03/2016 12:19, Fabien wrote:
On 03/07/2016 12:38 PM, BartC wrote:
(Although competing with CPython is too easy. PyPy is more of a problem.
With the Jpeg benchmark I mentioned, I can beat PyPy up to 6Mpix, but
then PyPy starts to get faster. At 80Mpix, PyPy is 60% faster.)
Just out of curiosity: are you also competing against numpy/scipy?
No, I only compare basic language functions. I understand that Python
depends on complex built-in functions, and external libraries such as
numpy, for it to be used viably. But I'm also interested in using such
languages directly.
Take the jpeg benchmark. Of course both Python and my language are
hopelessly slow and impractical compared with a C implementation, but
this is still a useful test (and in fact the interpreted version was
used to more easily develop a streamlined decoder that was then
back-ported to C, doubling its speed).
(The Python version of that program is here:
http://pastebin.com/cHx3UhQb. It should work with any Python.)
For example, suppose you wanted to do crude animation by loading a
series of small 640x480 images. Loading such an image gives you the
following frame rates, excluding the extra overheads of displaying the
results:
Python 2.7: 0.33 fps
Python 3.4: 0.25 fps
PyPy: 0.55 fps (don't known what version)
Mine: 3.77 fps
You can't really call the first three animated (more like a slide show),
but 4 fps just about cuts it!
Sometimes pushing an interpreted language a bit further can be useful
(hence the existence of the PyPy project).
--
Bartc
--
https://mail.python.org/mailman/listinfo/python-list
