2009/8/15 rjf <fate...@gmail.com>: > > wow, I post one place and it comes out in two places... > > Here is an amplification about the comments on GMP and GMPY. > > From the perspective of Sage and python, I just took a look at the > current gmpy. > The documentation, which has not apparently been updated since 2003, > says > > "Early tests have shown that supporting Python 2's "in-place > operation" functionality (by making MPZ, MPF and MPQ Python objects > mutable) would offer a substantial performance boost. > > Despite this, widespread feeling among Python cognoscenti appears to > be against exposing such "mutable numbers". .... " > > ................ > > So my comment about the unsophisticated python wrappers for gmpy are > still correct for now, at least if the documentation is up to date. > > > If that is true, Sage, using this gmpy, imposes a significant > limitation to persons attempting > to use gmp in its most efficient manner, using mutable numbers.
Sage does not use gmpy. Sage also doesn't use GMP -- it uses a fork of GMP called MPIR, and has its own direct wrapping of the MPIR library (http://mpir.org/). This does support in place operations for speed reasons, though you have to explicitly call an "underscore method" to access them: sage: n = 15 sage: n._iadd_(5) 20 sage: n 20 They should be used only when people know what they are doing, since one can do stupid things like the following: sage: V = QQ^4; V Vector space of dimension 4 over Rational Field sage: a = V.dimension() sage: a._iadd_(5) 9 sage: V.dimension() 9 One can also just directly use the MPIR C api through Cython, and there is a lot of support for this. This is in fact how a substantial amount of Sage itself is written. It provides exactly the same performance as programming MPIR in C. To get a sense of how much this is used in Sage, note that there are nearly 5000 lines of code in Sage that contain mpz_ or mpq_: wst...@sage:~$ sage -grep mpz_ |wc -l 4021 wst...@sage:~$ sage -grep mpq_ |wc -l 888 > Some lisps that use gmp directly, may also avoid the parts that do > mutation, since that causes problems with the functional model that > prevails with the simplest viewpoint of lisp arithmetic. That does > NOT mean that lisp cannot use GMP and mutation. > > For examples of how I used gmp from Allegro Common Lisp, a lisp that > does not natively use GMP, see > http://www.cs.berkeley.edu/~fateman/generic > and especially... > http://www.cs.berkeley.edu/~fateman/generic/gmp.lisp > > I initially tried loading the python-packaged gmp library into lisp -- > it worked fine, but was limited. So I load the gmp library directly. > > > There are also comments in that file that give comparisons for > different hardware and different versions of gmp. > If you use the "generic pentium" gmp instead of the specific version > for pentium3 or pentium4 [depending on what > you are using!] then you suffer a penalty of 3X in speed. Or did so > when I ran those tests. > > I don't know if Sage is set up to provide linkages to the "most > appropriate" GMP for each architecture, but > there are substantial efficiency differences. Sage includes MPIR, and a large amount of work goes into making sure that MPIR is built optimally on each platform Sage supports. > To the extent that you find it acceptable to set up the foreign- > function interface, probably anything you can > import into python, you can import into lisp, though not uniformly for > all lisp implementations. And the > foreign-function interface is somewhat diddly, which is why there are > some utilities that will read a *.h file and generate > most of it automatically. I didn't use one myself, since I used only > a few entry points to start, and added > only those that I thought I needed. > > > > > > > > > -- William Stein Associate Professor of Mathematics University of Washington http://wstein.org --~--~---------~--~----~------------~-------~--~----~ To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel-unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~----------~----~----~----~------~----~------~--~---
