[Webkit-unassigned] [Bug 68353] [Qt] Apply ParallelJobs for ImageBuffer::platformTransformColorSpace method

Tue Oct 18 00:01:37 PDT 2011

https://bugs.webkit.org/show_bug.cgi?id=68353

Simon Hausmann <hausmann at webkit.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hausmann at webkit.org

--- Comment #27 from Simon Hausmann <hausmann at webkit.org>  2011-10-18 00:01:37 PST ---
(In reply to comment #18)
[...]
> But never mind that. So what we're talking about here is a 3% overall improvement on a test suite that exercises a myriad of functionality, not necessarily even hitting platformTransformColorSpace() in all tests etc? That's not a bad thing in itself, but we should know how much faster the new platformTransformColorSpace() is compared to the old version.

I share a bit of Andreas' sentiment here.

It would be good to know how much this particular function improved and much more importantly it would be good to know _why_. "Blind" optimizations are dangerous :)

It is not apparent to me why distributing the function over different cores improves performance. From looking at the function and its lack of a computational kernel I _suspect_ that the bottleneck here is actually memory throughput. If that is the case, then it may be worth researching other ways of improving the performance because they may lead to even bigger wins with less heuristics and the thread start-up overhead.

I'm saying this not to criticize your code, but I'd just like to stress the importance of explaining what you saw in oprofile and why your patch works. And if it is a memory throughput problem, then I recommend trying some other cache optimization techniques in comparision to see which one is best.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.