[Webkit-unassigned] [Bug 44528] Improve FEGaussianBlur algorithm to access image buffer memory directly

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=44528





--- Comment #10 from Dirk Schulze <krit at webkit.org>  2010-10-25 05:46:04 PST ---
(In reply to comment #9)
> Thanks for the review krit.
> 
> (In reply to comment #8)
> > (From update of attachment 65698 [details] [details])
> > View in context: https://bugs.webkit.org/attachment.cgi?id=65698&action=review
> > 
> > Have you tried to use the original code, but take the WTF::ByteArray instead of the CanvasPixelArray? I'm interested in how much perf win this could bring. Surely not as much as your code, but still. Also it is not needed here to mark the surface as dirty. This is only necessary, if you want to access the context after pixel manipulations. That isn't the case here or on any other filter effect result.
> > You also ignored, that some platforms may use unpremultiplied colors. Would be great if we can find a way to manage pixel access with premultiplied and unpremultiplied colors, that would be a win for the other filter effects as well.
> > 
> 
> Yeah, you are right, I'll try to check if there is any option and check the performance in that case.

I already checked and tested ByteArray* an unsgined char* and couldn't see a difference in performance. I took the attached SVG on bug 48174 as reference. On Mac put/ and getImageData just take 0.8% of the time, GaussianBlur algo itself took 96% of the time. On my machine I could reduce the rendering time from 31s to 21s just by eliminating the -> operators and accessing the pixel data via ByteArray instead of CanvasPixelArray. 21s is still to much, but I don't see huge improvement by implementing direct pixel data access on ImageBuffer right now. Even so I'd support it!

Do you plan to investigate on this more?

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