535412 - Figure out how to do blur in hardware

Status: NEW

(Core :: Graphics, defect)

Description

[Jeff Muizelaar [:jrmuizel]]

We need to figure out whether we want to do blur/shadow at the cairo level or at the layers level. I'm sort of leaning toward the layer level as that will give us more flexibility and avoids adding api.

However, I wonder if doing things in cairo will let us be more efficient...

Also, quartz does support native shadow drawing, though I'm not sure we want to use that.

Thoughts anyone?

Comment 1

[Robert O'Callahan (:roc) (email my personal email if necessary)]

I was thinking about this this morning.

There can be a lot of shadows. E.g. if you have a lot of text with a blur-shadow of small radius, I don't think you want to do it all in hardware with a lot of layers.

I think the approach we have right now where we render a mask to an image surface and then blur it on the CPU is quite good and we can make it a lot better, per https://bugzilla.mozilla.org/show_bug.cgi?id=509052#c3. Another thing I realized is that for small radii, we should probably just apply an explicit Gaussian kernel instead of the 6-pass (3 horizontal, 3 vertical) box-blur that we currently do. I'd like to push on those optimizations before we switch over to hardware.

Comment 2

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Hrm, I would actually prefer to do this in hardware, but maybe we just want both.  A lot of text with blur-shadow of small radius I would think would be exactly what you want in hardware -- it's not a lot of layers, it's just one layer, right?

Comment 3

[Robert O'Callahan (:roc) (email my personal email if necessary)]

No, it's a lot of layers, because according to CSS the shadow gets painted directly under each text node. If the shadow overlaps adjacent text nodes (common), you're stuck with <shadow1> <text1> <shadow2> <text2> ...