Yes, makes sense to dig somewhere deeper. We lack a bit profiling of builtin functions, which makes it a bit tedious to compare two runs. On the other hand ctypes is a single point which is evident to consume a lot. Also, given bigger tables (like in tests) we consume way more time in ctypes than just 20%, this is minimal possible ctypes usage.
Cheers, fijal On Sun, Jun 22, 2008 at 12:12 PM, Armin Rigo <[EMAIL PROTECTED]> wrote: > Hi Maciej, > > On Sat, Jun 21, 2008 at 08:24:40PM +0200, Maciej Fijalkowski wrote: >> application. Anyway, it seems that our ctypes is huge problem there, >> calling of ctypes functions eats 20% of time overall (which is fairly >> a lot for 1-value table). > > Spending 20% of the time in ctypes is a lot, but something else is > consuming even more time than ctypes itself if I believe your numbers, > because in the good cases we are around 1.4x slower than CPython. So > with the admin app, say that CPython takes 1 sec; we would like to take > 1.4 sec; we really take 3 sec; we loose 1.6 sec somewhere, including 0.6 > sec in ctypes; so we loose 1 sec somewhere else. > > In other words, it is more urgent to figure out where we loose 1 sec > (33% of our time) than trying to reduce ctypes' 20%. > > > > A bientot, > > Armin. > _______________________________________________ [email protected] http://codespeak.net/mailman/listinfo/pypy-dev
