[Webkit-unassigned] [Bug 49907] Better result passing in filter primitives

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=49907


Dirk Schulze <krit at webkit.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |oliver at apple.com,
                   |                            |simon.fraser at apple.com




--- Comment #6 from Dirk Schulze <krit at webkit.org>  2010-11-23 10:48:25 PST ---
(In reply to comment #5)
> > Even if I really like the idea, I fear that we and up with 3 image results per FilterEffect. But of course I see the performance win. But we also have to think about hardware acceleration. 
> 
> I have no strong opinion here. I know most filters (light, turbulence, gaussian blur, morphology, convolve matrix, displacement) will never be hw accelerated except the simplest ones (offset, merge). And yeah, we can have 3 results for each filters, although the filter itself only creates one, and the others are created on demand. But you should decide whether it is worth to work on this project. That is why I submit conceptual patches, since if this is not a good idea, at least I don't have to spend full days working on it.

Yes, thanks for the great proposal, but maybe others should also comment on this bug.

To the hwa. I know of OpenVG that they support fast filter effects for gaussian blur and convolve matrix, not sure how they calculate it. But I wouldn't say for sure that they can't be faster with hwa :-)
And I'm pretty sure that these filter can be a lot faster on a GPU because they are very good to parallelize.

Adding Simon and Oliver, since hwa is not my field of knowledge.

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