[pypy-dev] Current test_all on windows

Christian Tismer tismer at stackless.com
Tue Sep 20 03:40:41 CEST 2005


Hola Ben,

> It really looks like your optimization work is progressing! At this rate 
> you'll be faster than CPython by Christmas. I just had a couple of 
> questions from looking over the current changes.

Yes, but with no idea of the kind of courve we are sliding along,
I would not want to fix the year of this Christmas. :-)

> Is there a plan to write an API for graph transformations that stops you 
> from performing invalid changes? That would be really useful considering 
> the amount and subtlety of the changes that are being made. 

No idea. All changes are supported by adding tests, and when the tests
are not enough, I'm yelling out loud if my exe doesn't work.

> Also, do you really need to do inlining to do malloc removal? If you 
> annotate whether a function takes any ownership over an argument (would 
> need to propagate up from the leaf functions) then you can tell whether 
> you can just stack allocate the argument and just pass it in with & (Or is 
> that the plan anyway?).

This is an alternative to the current concept, that is considered.
Inlining was mostly done to have a cheap way to see what saving
the mallocs can do. In fact, saving malloc and inlining work together,
but are not really related and should be better handled seperately
at some point. ATM we get very much inlining which is not desirable.
Ideally, we would run some simulation to find the true candidates,
and I'm personally even pushing to do the opposite: Move common code
which is not frequently used into artificial functions, and re-unify
specializations which are not worth it, performance-wise.

> Now as I was told to continue my nagging emails here is a latest list of 
> test failures on Windows. They have all been failing for a week or so (I 
> think). Interestingly the "inf" one only started failing around the time 
> you set more compliant calculations on for Windows 

:-) ok, ok, I was lazy and busy with timings.
If nobody else takes it, who is the windows head here?

yes, keep naggin', thanks so much -- chris

-- 
Christian Tismer             :^)   <mailto:tismer at stackless.com>
tismerysoft GmbH             :     Have a break! Take a ride on Python's
Johannes-Niemeyer-Weg 9A     :    *Starship* http://starship.python.net/
14109 Berlin                 :     PGP key -> http://wwwkeys.pgp.net/
work +49 30 802 86 56  mobile +49 173 24 18 776  fax +49 30 80 90 57 05
PGP 0x57F3BF04       9064 F4E1 D754 C2FF 1619  305B C09C 5A3B 57F3 BF04
      whom do you want to sponsor today?   http://www.stackless.com/



More information about the Pypy-dev mailing list