Binary arithmetic optimization (Was: Why notan__assign__method?)
Robin Thomas
robin900 at yahoo.com
Wed Apr 11 21:03:23 EDT 2001
Update on this patch:
1) With the patch, the performance loss on non-in-place built-in types is
enough to offset the gain from built-in types that do support in-place ops.
Roeland's bench test, run on my box, showed a smaller-than-expected gain
for list concat/repeat, and a larger-than-expected loss for string
concat/repeat, and about the expected (small) loss for numerics.
2) Even worse, pystone showed a average loss of 180 pystones/sec with the
patch. That's on my box.
3) For small operations like [None] + [None] + [None], the patch is
actually a little slower, despite its more efficient creation/destruction
of objects.
4) The patch is posted anyway: patch ID 414391 on sourceforge. Use it
against latest CVS.
http://sourceforge.net/tracker/index.php?func=detail&aid=414391&group_id=5470&atid=305470
Sigh. Oh well, I learned a little more about CPython. :)
--
robin900 at yahoo.com
_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com
More information about the Python-list
mailing list