Bug 69866 – Chromium needs support for border radius clipping

• Status: RESOLVED
• Resolution: FIXED
• Component: New Bugs

[Tom Hudson]

The speed of the current soft clipping implementation in Chromium is insufficient to support border radius clipping in websites that make heavy use of layers (https://bugs.webkit.org/show_bug.cgi?id=68733). Border radius clipping is turned off for chromium in https://bugs.webkit.org/show_bug.cgi?id=69844, but it needs to be reenabled - either when the fast path in #68733 is shown to work well enough in practice, or when the Windows/GDI/Skia issues blocking soft clipping for Chromium are resolved.

[Tom Hudson]

Status update: the first pass at anti-aliased clipping in Skia has landed and gets performance on targeted pages up to 10-15fps when we reenable border radius clipping; this is far better than it was when border radius clipping was disabled, but still a factor of two below our target. We're investigating.

[Peter Beverloo]

For reference: I just tried this with ToT Chromium and WebKit, unfortunately I'm still getting 80-100% CPU usage and only 15-20 fps on Twitter.

[Tom Hudson]

(In reply to comment #2)
> For reference: I just tried this with ToT Chromium and WebKit, unfortunately I'm still getting 80-100% CPU usage and only 15-20 fps on Twitter.

Actually, that's the pre-existing performance, and what we'll get after this bug is resolved.

The Skia optimizations required to support WebKit's border radius clipping in Chromium have landed, and enabling that feature produces no significant difference in the frame rate or profile, but we're rendering wrong for one of the four layout tests (fast/clip/overflow-border-radius-composited).

The Chromium CPU profile doesn't suggest any more obvious targets for microoptimzation in Skia; we're busy doing the direct work of clearing and blitting - further speedup for Twitter is likely going to take changing how much WebKit tries to draw.

Created attachment 166792 [details]
Patch

The test noted above (fast/clip/overflow-border-radius-composited) is still failing, but there is a separate issue for this: http://webkit.org/b/68196

A number of media control tests have subtle differences due to their uses of border-radius, I've included rebaselines for linux and will update expectations for the other platforms after landing.

[noel gordon]

LGTM.

aisde: maybe file a bug about video-controls-rendering.html and it's use of scroll bars?

[noel gordon]

Mr White: review please?

[Stephen White]

This looks ok, but note that antialiased clipping is not (quite) implemented in the GPU path in Skia.  That's probably not an issue right now, but I'd like to get a thumbs-up from reed@ and robertphillips@ before we turn this on.

[Mike Reed]

aaclipping is used for all other soft-clip requests in PlatformContext, so I see no reason to not use it here (from skia's perspective).

[Stephen White]

Comment on attachment 166792 [details]
Patch

Thanks.  robertphillips@ gave me a thumbs-up offline, so r=me.

[WebKit Review Bot]

Comment on attachment 166792 [details]
Patch

Clearing flags on attachment: 166792

Committed r130355: <http://trac.webkit.org/changeset/130355>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.

[Yury Semikhatsky]

The patch caused webkit_lint failure: http://build.chromium.org/p/chromium.webkit/builders/Webkit%20Win/builds/18128/steps/webkit_lint/logs/stdio

[Yury Semikhatsky]

Updated expectations file in http://trac.webkit.org/changeset/130368.

Landed mac/win rebaselines in http://trac.webkit.org/changeset/130447