On Fri, 13 Oct 2006 15:43:47 -0700, Martin Albrecht <[EMAIL PROTECTED]> wrote:
> > On Friday 13 October 2006 16:29, William Stein wrote: >> omalloc is now GPL'd! See below. > > Excellent! Let's make sure the Debian people are informed. They might > reconsider packaging Singular (which would mean that they threw their > autoconf wisdom in). The relevant Debian thread is here: http://www.mail-archive.com/debian-mentors@lists.debian.org/msg46083.html However, that thread hides email addresses, and I don't know how or want to sign up for that list. So I actually have no idea how to inform the Debian people. Any ideas? > Also, I've built a shared library of omalloc (by hand, but easily > automated), > and here are the timing results: > > sage: x = 3; y = 5 # malloc (GMP default) > sage: timeit _ = x + y > 100000 loops, best of 3: 8.14 µs per loop > sage: timeit _ = x + y > 100000 loops, best of 3: 8.05 µs per loop > > sage: sage.rings.integer.pmem_malloc() > > sage: timeit _ = x + y # PyMem_Malloc > 100000 loops, best of 3: 8.27 µs per loop > sage: x = 3; y = 5 > sage: timeit _ = x + y > 100000 loops, best of 3: 8.21 µs per loop > > sage: omalloc_malloc() > > sage: timeit _ = x + y # omalloc > 100000 loops, best of 3: 7.55 µs per loop > sage: timeit _ = x + y > 100000 loops, best of 3: 7.55 µs per loop > > My conclusion for small integers: PyMem_Malloc is slower than malloc or > omalloc. omalloc is slightly faster than malloc. However I find these > benchmarks not very expressive (Python overhead, little difference etc.) > > Larger integers: > > <restart> > sage: x = > 45234723458934634056310653251305043576435013985134515643531450714359863148659136951369851836545643565 > > sage: y = > 23423964523462384234234234872345623645623452345283548234652365482348562346523642348263486234954534554 > > sage: timeit _ = x + y #malloc > 100000 loops, best of 3: 8.3 µs per loop > > sage: sage.rings.integer.pmem_malloc() > > sage: timeit _ = x + y #PyMem_Malloc > 100000 loops, best of 3: 8.58 µs per loop > > sage: omalloc_malloc() > > sage: timeit _ = x + y # omalloc > 100000 loops, best of 3: 7.58 µs per loop > > Conclusion for larger integers: These benchmarks seem to reinforce the > previous conclusion. > > Thoughts? Could you do some simpler benchmarks that compare malloc and omalloc, and don't have anything to do with Python? E.g., direct C-level mpz_init's, etc.? And allocation of int*'s, etc. That's *all* very relevant to implementing SAGE. william --~--~---------~--~----~------------~-------~--~----~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to [EMAIL PROTECTED] For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://sage.scipy.org/sage/ and http://modular.math.washington.edu/sage/ -~----------~----~----~----~------~----~------~--~---
