[TYPO3-dev] FYI: comparing execution speed of some php functions

[Ries van Twisk]

On Jun 28, 2009, at 9:59 AM, Steffen Kamper wrote:

> Hi Niels,
>
> no - it'S a different issue.
>
> I tried with adding this to Rupi's script:
>
> $GLOBALS['myClass'] = new  mytestclass();
> $S = microtime(TRUE);
> for ($i = 0; $i < $num; ++ $i) {
> 	$GLOBALS['myClass']->testfunc(1);
> }
> $E = microtime(TRUE) - $S;
> echo "global object call took \t" . number_format(floatval($E * 1000),
> 5) . " ms\n";
>
> these are the results:
> calling methods in classes:
> ---------------------------------------------
> object call took 		347.18299 ms
> derived object call took 	359.87806 ms
> static call took 		523.85497 ms
> global object call took 	375.33998 ms
>
> (oh god, my numbers are 8 times slower than Rupis, what's wrong with  
> my
> machine?)
>
> but, after isolating the object calls (removing all other tests) i got
> incredible different results:
>
> object call took 		0.02289 ms
> derived object call took 	0.01001 ms
> static call took 		0.00811 ms
> global object call took 	0.00811 ms
>
> and this is completely the other way showing that static class calls  
> are
> faster than object calls.
>
> As this example shows, don't trust results before you did an isolate
> test. The performance of php completely change after filling memory  
> etc.
>
> vg Steffen


I believe that static methods will be compiled in-line if possible,
however I think this is still up to the compiler to make a final  
decision.

Creating a static method and just return 1 is not really a good test
without knowing what decisions a compiler can take to compile or
not compile the method in-line or not, this was shown in the PDF  
here : http://ilia.ws/files/phptek2007_performance.pdf

I can imagine that if a function is called many times from different  
locations
and the static method is large, the compiler might not decide to compile
the final code in-line and decide to reduce file size instead.

So what steffen points out is right, be careful with such simple tests
and don't draw conclusions without testing it in a real world scenarios.

Ries