Concerns about performance w/Python, Pysco on Pentiums

[Peter Hansen]

Skip Montanaro wrote:
> 
>     Peter> Py2.2, list, inlined:        16468, 108%
>     Peter> Py2.3, list, inlined:        19074, 126%
>     Peter> Py2.2, list, inlined, Psyco:  9963,  66% (!)
> 
>     Peter> Not sure what that last one indicates, except that under certain
>     Peter> circumstances (perhaps when used by someone who hasn't even read
>     Peter> it's documentation? :-), Psyco can produce *worse* results.
> 
> How exactly are you using Psyco?  If you let it figure out which routines to
> specialize, it generally does a poor job.  If you know where your hotspots
> are, you can tell it to just specialize those routines.

At first I tried binding all the opcode handler functions, and the 
Opcode class, and the Cpu class.  A little experimentation showed that
ditching the opcode handlers increased performance, so now I just
do the two classes.  Are you suggesting picking only some methods 
in the classes would help?

At this point, actually, I need to do one thing before going any
further makes sense: profile the code.  And read the Psyco manual.
Two things I must do before it makes sense.  Profile, RTFM, and 
develop a _need_ for greater performance.  *Three* things... there
are three things I need.  Oh bugger...

-Peter