[pygame] Pygame and Psyco

[Pete Shinners <shredwheat@attbi.com>]

hey everyone, i've been playing a bit with pysco to see what it can do for 
pygame games. the good news is, psyco can have a huge speed boost. the 
not-so-good news is, it's not going to help everyone.

at this point there are a couple packages for python that promise speeding 
up python code. the thing i like about psyco is it requires no change to 
the python game. with a simple try:except: block, your game could get the 
extra speed for people with psyco, otherwise it will run as normal python 
speed for those without.

basically you just 'import psyco' and then tell it which functions you want 
it to optimize, and you are all set. because psyco takes a big chunk of 
memory, you mainly just optimize the parts of the code that are your 
bottleneck.

the first thing i did was to hook up the pygame sprite library to psyco, 
and test it with a fairly sprite intensive benchmark. (making good use of 
many of the different classes and collide functions). unfortunately there 
was barely 1 or 2 fps difference between them. i was pretty disapointed. i 
determined that even with the program running faster, most of the program 
was drawing graphics, which isn't going to get any faster. using a special 
version of the sprite library with a dummy "blit" function i noticed that 
psyco made the program about 20% faster, 700fps to 900fps.

then i decided i wanted to test some code that wasn't bottlenecked by 
graphics. i remembered that Jeroen was a little frustrated with the speed 
of the AI in his game "4st attack". i cranked up the AI to look 5 moves 
into the future, and noticed it took around 26 seconds for the ai to make 
each move. i imported psyco into the ai module and reran the test. it now 
takes almost exactly 2.6 seconds to compute each move. (argh the AI is way 
too hard at this point)

wow, i am still very impressed 10x speedup and i didn't have to change 
anything. this is the first 'python speedup' type thing i've tried that 
actually made any difference. i've played with "python to c" code 
converters and other sorts of optimizers, but nothing ever really gave me 
noticeable speedups.

well that's good news. worth anyone's time who is trying to get more speed 
out of 'fat' parts of their game. of course it will always be tricky when 
optimizing. you'll need to point psyco at parts of the game where 'python' 
is the actual bottleneck. it seems mainly in pygame games the most serious 
bottleneck is in the software graphics. but games with complex ai, physics, 
or object management may see similar benefits from psyco.

again, the coolest part is you don't have to change any of your code. your 
game will run with or without the user needing psyco. the psyco website is 
at http://psyco.sourceforge.net/ . also i've "cross-spammed" the author 
with this email, so if you have any questions for him as well, reply-all 
may catch his attention ;]


oh, and i hope that means we'll be seeing a new version of 4st attack 
someday sooner than later :]




____________________________________
pygame mailing list
pygame-users@seul.org
http://pygame.seul.org