On Oct 1, 2009, at 12:07 PM, Stefan Behnel wrote: > Hi all, > > for a talk I'm about to give, I ran the pybench benchmark with > cython-unstable. I had to fix up the sources a tiny bit (nothing that > impacts the results) and select the benchmarks that actually compile > (mostly due to inner classes being used). It's run like this: > > PYTHONPATH=. python2.6 -c 'import pybench; pybench.PyBenchCmdline()' \ > --debug -w 1 -n 5 -f results-cython.pybench -t \ > 'Arithmetic|IfThen|Loops|Dict|Try.*Except|Import|List|Compare| > String|Tuple|Unicode' > > FWIW, both the Python 2.6.2 interpreter and the Cython generated C > modules > were built on the same machine with the same CFLAGS. > > The sources I used are here: > > http://codespeak.net/lxml/pybench.tar.gz > > I attached the results. > > I left out the CompareIntegers/Floats/etc. benchmarks as they only > compare > numeric literals, which ends up in the C code as, well, a > comparison of > literals, so that the C compiler finds a lot of dead code to > discard. The > resulting 100% speedup isn't really that informative. > > Note that the numbers are for plain Python code without any type > annotations, so a lot of possible optimisations are not even used. > > However, the remaining total improvement of almost 33%, and the huge > speedups for control structures (especially loops) and builtin > types is > clearly worth a look.
Cool. - Robert _______________________________________________ Cython-dev mailing list [email protected] http://codespeak.net/mailman/listinfo/cython-dev
