[Webkit-unassigned] [Bug 90166] NEON intrinsics should be used with gaussian blur filter

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=90166





--- Comment #3 from Gabor Rapcsanyi <rgabor at webkit.org>  2012-06-29 01:51:23 PST ---
(In reply to comment #2)
> (From update of attachment 149938 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=149938&action=review
> 
> Looks sane to me in general, but I prefer that Zoltan looks at the NEON stuff, he can give r+ or r- then :-)
> 
> > Source/WebCore/platform/graphics/filters/FEGaussianBlur.cpp:127
> > +                boxBlurNEON(src, dst, kernelSizeX, dxLeft, dxRight, 4, stride, paintSize.width(), paintSize.height(), isAlphaImage());
> > +            else
> > +                boxBlur(src, dst, kernelSizeX, dxLeft, dxRight, 4, stride, paintSize.width(), paintSize.height(), isAlphaImage());
> 
> You could expand isAlphaImage() here.
> 

Because this patch just speed up the 4 channel mode. The alpha image is much complicated to optimize. Now I'm working on that problem, but anyway the hand written assembly didn't improve performance to much so that's why I fall back to the origin boxBlur when the image is an alpha image.

> > Source/WebCore/platform/graphics/filters/FEGaussianBlur.cpp:138
> > +                boxBlur(src, dst, kernelSizeY, dyLeft, dyRight, stride, 4, paintSize.height(), paintSize.width(), isAlphaImage());
> 
> boxBlurNEON? Otherwise this makes no sense.
> You could also expand isAlphaImage() here.

Yes, you're right I mistyped it, this is boxBlurNEON.

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