[webkit-dev] sunspider slow-down

[Mark Rowe]

On Jul 18, 2008, at 8:00 AM, Zoltan Herczeg wrote:

> Hi webkit developers!
>
> I have recently updated my repository from revision 35033 to revision
> 35231, and noticed a great (20%) perfromance loss for sunspider. My  
> test
> system has a Dual-core 2.2 GHz AMD opteron cpu. Since this is a big  
> range,
> I have tried some other revisions as well to find out the reason of  
> this
> big jump. It seems the change was heppened between 35208 and 35231.
>
> Sunspider results:
> revision 35033 - 3168.0ms +/- 0.8%
> revision 35231 - 3172.0ms +/- 0.7%
> revision 35208 - 3782.8ms +/- 0.6%
> revision 35231 - 3779.2ms +/- 0.6%
>
> *1.193x as slow* according to sunspider-compare-results

To clarify, is the standard SunSpider slower or is a version you have  
extended with custom tests slower?

> We have collected more than a dozen other js programs to extend the  
> jsc
> benchmark set, and it seems "bubble sort" is slowed down to 3.4 sec  
> from 2
> esec, which is about 70% performance loss.


> The results and this benchmark are attached to this mail.
>
> Do you know about it? Is there a reason behind it or shall I make a
> bug-report?

Performance regressions are definitely worth a bug report.  Be sure to  
include some information that you did not include in this email, such  
as which port of WebKit you are building, which compiler version you  
are using, and how you are running the tests.

Kind regards,

Mark Rowe