develooper Front page | perl.perl6.internals | Postings from December 2001

Re: Just-in-Time Predereferencing Patch

Thread Previous
Gregor N. Purdy
December 11, 2001 08:52
Re: Just-in-Time Predereferencing Patch
Message ID:
Simon --

> I haven't tried this, so feel free to shoot me down, but is your 22%
> speed-up based on an unpatched Parrot, or a patched Parrot without the
> -P flag? I'm trying to work out whether or not this patch actually slows
> down the general case, which wouldn't be pleasant.

Based on patched without the -P flag. The only additional cost is that
of one additional test during setup. Its not in the inner loop. Its no
more expensive in the general case than any of the other setup flag
testing we do to choose, e.g. fast vs. slow core.

Just to be sure, I grabbed a fresh copy from CVS. I get 24.00 mops for
each of three trials each of unpatched and patched.

On patched with the -P flag, I get 30.26 mops on each of three trials.

At least on my system, I've not affected the general case one whit.

-- Gregor
/            Inspiration >> Innovation >> Excellence (TM)             \

   Gregor N. Purdy                
   Focus Research, Inc.      
   8080 Beckett Center Drive #203                   513-860-3570 vox
   West Chester, OH 45069                           513-860-3579 fax

[]$ ping
PING ( from : 56(84) bytes of
From Time to live exceeded

Thread Previous Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at | Group listing | About