On Mon, Mar 8, 2010 at 3:40 PM, Maciej Fijalkowski <fij...@gmail.com> wrote:
> On Mon, Mar 8, 2010 at 1:55 PM, Ondrej Certik <ond...@certik.cz> wrote:
>> On Mon, Mar 8, 2010 at 11:53 AM, Maciej Fijalkowski <fij...@gmail.com> wrote:
>>> Hello.
>>>
>>> Half a year ago, I was asking about sympy benchmarks. So here is some
>>> feedback on what I got using the JIT. Running the very simple
>>> benchmark, like this:
>>>
>>> import sympy
>>> import time
>>> x, y = sympy.symbols('x, y')
>>>
>>> for i in range(10):
>>> t0 = time.time()
>>> sympy.factor(x**20 - y**20)
>>> print time.time() - t0
>>>
>>> (I run it 10 times for JIT to warm up)
>>>
>>> I get the following results:
>>>
>>> CPython 2.6.4:
>>>
>>> 0.894039154053
>>> 0.738809108734
>>> 0.749840021133
>>> 0.817865848541
>>> 0.72897195816
>>> 0.771667957306
>>> 0.777043104172
>>> 0.770677089691
>>> 0.770140886307
>>> 0.877714157104
>>>
>>> PyPy trunk (rev 71774):
>>>
>>> 1.19711208344
>>> 0.634068012238
>>> 0.557782888412
>>> 0.614212036133
>>> 0.521821022034
>>> 0.577347993851
>>> 0.481374979019
>>> 0.466244935989
>>> 0.444444894791
>>> 0.520822048187
>>>
>>> So overal we get some (not too much) speedup, after warmup, without
>>> even profiling that case. I suppose we can do much better if we try
>>> slightly harder.
>>
>> Also note that sympy uses caching, so you might try it with caching turned
>> off:
>>
>> http://wiki.sympy.org/wiki/FAQ#How_do_I_turn_off_caching.3F
>
> It seems to be irrelevant for this particular example.
>
>>
>>>
>>> What do you think? Is it an interesting direction to pursue? Are
>>> people interested?
>>
>> I am interested. What kind of things can pypy compile in time using JIT?
>
> In general all Python. It can perform better on some stuff and worse
> on some stuff (combinations of map/reduce being particularly nasty,
> loop being much much faster for example).
>
> I can perform some more benchmarking, if you happen to have some more
> comprehensive bench suite than just this :-)
Try divisors() from sympy/ntheory/factor_.py. I tried that with Cython
(pure Python mode, so that you have the same source code, that works
both in Python and in Cython) and I forgot which speedup I got, but at
least 10x, maybe up to 25x. Should be in the archives of this list.
Ondrej
--
You received this message because you are subscribed to the Google Groups
"sympy" group.
To post to this group, send email to sy...@googlegroups.com.
To unsubscribe from this group, send email to
sympy+unsubscr...@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/sympy?hl=en.
