Hi Stas,
Can we see any difference with optimal ZEND_ADD_(STRING|CHAR|VAR)? :-)
Well, multiple ADD_STRINGs are one thing that optimizers/accelerators would
stitch back together, because it's a "major optimization." ;-) I didn't
specifically mention it in my previous messages, but since these particular
opcodes are now optimal, it's one step that those optimizers would no longer
need to do.
I mentioned some performance figures in my original message, but here's a
few again... Should be a pretty easy thing for anyone to benchmark. I was
surprised by the difference using ApacheBench (no accelerator of course) on
some different applications/scripts: 5-20% *overall* speedup, with 10-15%
being typical. I'm curious whether others would find the same on their
stuff??
When talking about the difference on specific code affected by the
optimizations, let's look at the NO_PCRE_ERROR heredoc from run-tests.php
which Sara referred to in her blog entry. The compile/cleanup time on that
(tested by eval()'ing many in an if (0) { ... } block) is about 2.5 times
faster. And because it's only a few opcodes now instead of 70-something, it
runs *15 TIMES* faster!
Now take a fairly short interpolated string, "This is $var string" ($var =
'some'), and its compile/cleanup and run times are about 40% and 60% faster,
respectively (I think I previously measured 50% and 55%, so somewhere around
there :-)).
It all depends on how many "pieces" (opcodes) were created by the strings
the old way and new way... But it's easy to test your own code, like I
said.
Finally, since any 1 character "pieces" that would've previously used
ADD_STRING are now made to use ADD_CHAR (only certain things would before),
that makes a measurable difference too. For example, with "$var." or
"<$var>" I see a ~20% speedup.
Hope that helps. :-)
Matt
----- Original Message -----
From: "Stanislav Malyshev"
Sent: Thursday, May 17, 2007
> > The patch mainly improves not the parser, but generated code.
> > It builds near optimal ZEND_ADD_STRING, ZEND_ADD_VAR sequence for quoted
> > strings and heredoc.
>
> OK, I understand that, the question is can we see any difference? We are
> doing quite major overhaul of the parser, so there should be a reason
> for it - like the code gets to be that and that faster, etc. Is there a
> way to see it?
>
> > You can see a small difference even on Zend/bench.php.
>
> Actually I don't :) Not beyond the margin of error at least. I'm not
> very surprised since bench.php doesn't do much with interpolated strings
> as I see - but I would like to verify that this patch is really a "major
> optimization" ;)
>
> --
> Stanislav Malyshev, Zend Products Engineer
> [EMAIL PROTECTED] http://www.zend.com/
--
PHP Internals - PHP Runtime Development Mailing List
To unsubscribe, visit: http://www.php.net/unsub.php
