Bug 218614 – [Concurrent display lists] Support playback of display list items with cached resources

• Status: RESOLVED
• Resolution: FIXED
• Component: WebKit2

[Wenson Hsieh]

e.g. DrawGlyphs, DrawImage, etc.

[Wenson Hsieh]

Created attachment 413296 [details]
WIP

[Radar WebKit Bug Importer]

<rdar://problem/71326662>

[Wenson Hsieh]

This blocks making daringfireball.net work, with both canvas + DOM GPUP features enabled.

[Wenson Hsieh]

Created attachment 415692 [details]
Patch

[EWS]

Committed r270591: <https://trac.webkit.org/changeset/270591>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 415692 [details].

[Geoffrey Garen]

It feels kind of complicated for WebContent to send commands and resources out of order, and then rely on GPU Process to stitch them back into order. I wonder if there's a way to ensure that commands always arrive in order with their resources.

[Wenson Hsieh]

(In reply to Geoffrey Garen from comment #6)
> It feels kind of complicated for WebContent to send commands and resources
> out of order, and then rely on GPU Process to stitch them back into order. I
> wonder if there's a way to ensure that commands always arrive in order with
> their resources.

Yeah, I think that we should eventually implement caching for native images and image buffers through display list items (like Ryosuke is doing with fonts in https://bugs.webkit.org/show_bug.cgi?id=219672).

I think the underlying mechanism that I introduced to fix this will be generally useful though, for generally synchronizing with other IPC messages.

[Ryosuke Niwa]

(In reply to Wenson Hsieh from comment #7)
> (In reply to Geoffrey Garen from comment #6)
> > It feels kind of complicated for WebContent to send commands and resources
> > out of order, and then rely on GPU Process to stitch them back into order. I
> > wonder if there's a way to ensure that commands always arrive in order with
> > their resources.
> 
> Yeah, I think that we should eventually implement caching for native images
> and image buffers through display list items (like Ryosuke is doing with
> fonts in https://bugs.webkit.org/show_bug.cgi?id=219672).

We need some mechanism to map shared memory from display list if we're going on that route.