This shouldn't really come as a surprise. From the Cython documentation, "This is about 20 times slower, but still about 10 times faster than the original Python-only integration code. This shows how large the speed-ups can easily be when whole loops are moved from Python code into a Cython module."
http://docs.cython.org/src/tutorial/cdef_classes.html On Tue, Apr 9, 2013 at 12:58 PM, Nils Bruin <nbr...@sfu.ca> wrote: > Sometimes one ends up with implementing really small access routines, > just to do a lookup for something. In cython you might even want to do > that with some "cdef" methods. For general accessibility, you might > consider slapping a "cpdef" on rather than a "cdef". That may come at > a serious price, because "cpdef" methods also check if they have been > overridden on the instance or via subclassing if they get called via > their cdef interface: > > with the following code we get: > > cython(""" > cdef class A(object): > cpdef int a(self): > return 0 > cdef int b(self): > return 0 > cpdef int test_a(self, long N): > cdef long i > for i in xrange(N): > self.a() > return 0 > cpdef int test_b(self, long N): > cdef long i > for i in xrange(N): > self.b() > return 0 > """) > > t=A() > timeit('t.test_a(10000000)') > timeit('t.test_b(10000000)') > class B(A): > pass > > s=B() > timeit('s.test_a(10000000)') > timeit('s.test_b(10000000)') > > 25 loops, best of 3: 21.2 ms per loop > 25 loops, best of 3: 15.9 ms per loop > 5 loops, best of 3: 507 ms per loop > 25 loops, best of 3: 15.9 ms per loop > > As you can see for A it's not so bad, because there's no instance > dict. However, for the subclassed s there is an instance dict. Now the > cpdef method s.a (even inside the cython!) checks every time if it has > been overridden with a full attribute lookup. > > Timings improve with attribute caching turned on (of course only for > s.test_a, because the other tests hardly do any attribute lookup) > > 25 loops, best of 3: 21.2 ms per loop > 25 loops, best of 3: 15.9 ms per loop > 5 loops, best of 3: 303 ms per loop > 25 loops, best of 3: 16 ms per loop > > but the penalty is still significant if your routine does very little > by itself. This was certainly unexpected for me. I was expecting that, > as far as cdef is concerned, the p in cpdef would have no effect. > > -- > You received this message because you are subscribed to the Google Groups > "sage-devel" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to sage-devel+unsubscr...@googlegroups.com. > To post to this group, send email to sage-devel@googlegroups.com. > Visit this group at http://groups.google.com/group/sage-devel?hl=en. > For more options, visit https://groups.google.com/groups/opt_out. > > -- You received this message because you are subscribed to the Google Groups "sage-devel" group. To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+unsubscr...@googlegroups.com. To post to this group, send email to sage-devel@googlegroups.com. Visit this group at http://groups.google.com/group/sage-devel?hl=en. For more options, visit https://groups.google.com/groups/opt_out.
