Sounds like a good idea to me. On Tuesday 21 May 2013, Volker Braun wrote: > Reposted at the top to increase visiblitiy (see > https://groups.google.com/d/msg/sage-devel/5PdRIUic2Es/l0yDYLG_8qIJ) > > The sage matrices use unsigned long for the mod_int datatype. This does not > fit into a (signed) python int, so whenever it reaches Python (assigning to > a Python variable, or passing as an argument to a non-cdef function) it > gets converted to a Python arbitrary-precision integer. This wastes quite a > bit of time when creating it and gets worse once you start doing > arithmetic: > > -------------------------------------- > ctypedef unsigned long mod_int # this is how we define it > > cpdef caster_slow(): > cdef int i > foo = None > for i in range(10000000): > foo = <mod_int>(1.0) > > cpdef caster_fast(): > cdef int i > foo = None > for i in range(10000000): > foo = <int>(1.0) > -------------------------------------- > > Then we get > > sage: timeit('caster_slow()') > 5 loops, best of 3: 137 ms per loop > sage: timeit('caster_fast()') > 25 loops, best of 3: 34.4 ms per loop > > Should we just switch to (signed) long? This wastes one bit (factor of 2) > of maximum modulus length, but at least for now we define MAX_MODULUS = > 2**23 so this wouldn't be an issue.
Cheers, Martin -- name: Martin Albrecht _pgp: http://pgp.mit.edu:11371/pks/lookup?op=get&search=0x8EF0DC99 _otr: 47F43D1A 5D68C36F 468BAEBA 640E8856 D7951CCF _www: http://martinralbrecht.wordpress.com/ _jab: martinralbre...@jabber.ccc.de
signature.asc
Description: This is a digitally signed message part.
