219468 – [GPU Process] Record changes in GraphicsContextState as individual DisplayList items

RESOLVED WONTFIX 219468

[GPU Process] Record changes in GraphicsContextState as individual DisplayList items

https://bugs.webkit.org/show_bug.cgi?id=219468

Summary [GPU Process] Record changes in GraphicsContextState as individual DisplayLis...

Tim Horton

Reported 2020-12-02 17:55:16 PST

SHOULD NEVER BE REACHED Source/WebKit/Platform/IPC/Encoder.h(110) : static RefPtr<WebCore::SharedBuffer> IPC::Encoder::encodeSingleObject(const T &) [T = WebCore::DisplayList::SetState] WTFCrash WTFCrashWithInfo(int, char const*, char const*, int) WTF::RefPtr<WebCore::SharedBuffer, WTF::RawPtrTraits<WebCore::SharedBuffer>, WTF::DefaultRefDerefTraits<WebCore::SharedBuffer> > IPC::Encoder::encodeSingleObject<WebCore::DisplayList::SetState>(WebCore::DisplayList::SetState const&) WebKit::RemoteImageBufferProxy<WebKit::ImageBufferShareableMappedIOSurfaceBackend>::encodeItem(WebCore::DisplayList::ItemHandle) const non-virtual thunk to WebKit::RemoteImageBufferProxy<WebKit::ImageBufferShareableMappedIOSurfaceBackend>::encodeItem(WebCore::DisplayList::ItemHandle) const WebCore::DisplayList::ItemBuffer::appendEncodedData(WebCore::DisplayList::ItemHandle) void WebCore::DisplayList::ItemBuffer::append<WebCore::DisplayList::SetState, WebCore::GraphicsContextState const&, WTF::OptionSet<WebCore::GraphicsContextState::Change>&>(WebCore::GraphicsContextState const&, WTF::OptionSet<WebCore::GraphicsContextState::Change>&) void WebCore::DisplayList::DisplayList::append<WebCore::DisplayList::SetState, WebCore::GraphicsContextState const&, WTF::OptionSet<WebCore::GraphicsContextState::Change>&>(WebCore::GraphicsContextState const&, WTF::OptionSet<WebCore::GraphicsContextState::Change>&) void WebCore::DisplayList::Recorder::append<WebCore::DisplayList::SetState, WebCore::GraphicsContextState const&, WTF::OptionSet<WebCore::GraphicsContextState::Change>&>(WebCore::GraphicsContextState const&, WTF::OptionSet<WebCore::GraphicsContextState::Change>&) WebCore::DisplayList::Recorder::appendStateChangeItem(WebCore::GraphicsContextStateChange const&, WTF::OptionSet<WebCore::GraphicsContextState::Change>)

Attachments
WIP (112.13 KB, patch) 2020-12-19 12:06 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
WIP (135.41 KB, patch) 2020-12-20 11:00 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Patch (134.89 KB, patch) 2020-12-21 01:33 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
WIP (117.26 KB, patch) 2021-02-05 01:48 PST, Said Abou-Hallawa	ews-feeder: commit-queue-	Details Formatted Diff Diff
WIP (118.24 KB, patch) 2021-02-05 02:06 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
WIP (123.74 KB, patch) 2021-02-05 11:22 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
WIP (122.25 KB, patch) 2021-02-05 14:48 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Patch (127.67 KB, patch) 2021-02-05 21:57 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Patch (129.03 KB, patch) 2021-02-06 00:50 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Patch (42.74 KB, patch) 2021-02-14 00:50 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Patch (46.25 KB, patch) 2021-02-14 12:52 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Patch (45.72 KB, patch) 2021-02-15 11:38 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Patch (95.95 KB, patch) 2021-02-16 22:53 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Patch (97.52 KB, patch) 2021-02-17 01:45 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Show Obsolete (13) View All Add attachment proposed patch, testcase, etc.

Tim Horton

Comment 1 2020-12-02 17:55:32 PST

For example, canvas/philip/tests/2d.composite.globalAlpha.canvaspattern.html

Tim Horton

Comment 2 2020-12-02 17:56:04 PST

The problem is, SetState is encoded using encodeSingleObject, but it has things like Pattern hanging off of it.

Radar WebKit Bug Importer

Comment 3 2020-12-02 19:22:48 PST

<rdar://problem/71921036>

Tim Horton

Comment 4 2020-12-02 19:28:23 PST

And Pattern has an Image that it encodes with ImageHandle. We either need a subdisplaylist for the pattern or to instead cache that image ahead of time.

Said Abou-Hallawa

Comment 5 2020-12-19 12:06:45 PST

Created attachment 416572 [details] WIP

Said Abou-Hallawa

Comment 6 2020-12-20 11:00:50 PST

Created attachment 416584 [details] WIP

Said Abou-Hallawa

Comment 7 2020-12-21 01:33:12 PST

Created attachment 416597 [details] Patch

Said Abou-Hallawa

Comment 8 2021-02-05 01:48:59 PST

Created attachment 419374 [details] WIP

Said Abou-Hallawa

Comment 9 2021-02-05 02:06:37 PST

Created attachment 419375 [details] WIP

Said Abou-Hallawa

Comment 10 2021-02-05 11:22:11 PST

Created attachment 419437 [details] WIP

Said Abou-Hallawa

Comment 11 2021-02-05 14:48:41 PST

Created attachment 419468 [details] WIP

Said Abou-Hallawa

Comment 12 2021-02-05 21:57:26 PST

Created attachment 419493 [details] Patch

Said Abou-Hallawa

Comment 13 2021-02-06 00:50:25 PST

Created attachment 419497 [details] Patch

Wenson Hsieh

Comment 14 2021-02-08 14:02:14 PST

Comment on attachment 419497 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=419497&action=review > Source/WebCore/platform/graphics/displaylists/DisplayListItems.h:170 > +class SetInlineState { It looks like SetInlineState doesn't include RGBA (non-extended) stroke color or fill changes, which (IIRC) make up the bulk of state changes in Paths, Lines and Arcs in MotionMark. I think this could encompass changes to fill and stroke color that only use SRGBA<uint8_t> as well, to keep most of the state change items in these fully inline in those three subtests. Though, perhaps this could be done as a followup.

Said Abou-Hallawa

Comment 15 2021-02-09 12:44:15 PST

Comment on attachment 419497 [details] Patch View in context: https://bugs.webkit.org/attachment.cgi?id=419497&action=review >> Source/WebCore/platform/graphics/displaylists/DisplayListItems.h:170 >> +class SetInlineState { > > It looks like SetInlineState doesn't include RGBA (non-extended) stroke color or fill changes, which (IIRC) make up the bulk of state changes in Paths, Lines and Arcs in MotionMark. I think this could encompass changes to fill and stroke color that only use SRGBA<uint8_t> as well, to keep most of the state change items in these fully inline in those three subtests. > > Though, perhaps this could be done as a followup. Yes this is right. I wrote a similar comment in the ChangeLog above. "Future work is to make SetInlineState handle more cases for colors and gradients." But I will try to do this inline changes in the patch of this bug to avoid introducing regression.

Said Abou-Hallawa

Comment 16 2021-02-14 00:50:20 PST

Created attachment 420236 [details] Patch

Said Abou-Hallawa

Comment 17 2021-02-14 12:52:29 PST

Created attachment 420249 [details] Patch

Said Abou-Hallawa

Comment 18 2021-02-15 11:38:32 PST

Created attachment 420343 [details] Patch

Said Abou-Hallawa

Comment 19 2021-02-16 22:53:45 PST

Created attachment 420601 [details] Patch

Said Abou-Hallawa

Comment 20 2021-02-17 01:45:58 PST

Created attachment 420617 [details] Patch

Note You need to log in before you can comment on or make changes to this bug.

Status RESOLVED

Resolution WONTFIX

Priority P2

Severity Normal

Classification Unclassified

Version Other

Hardware Unspecified

OS Unspecified

Product WebKit

Component Canvas

Assignee

Said Abou-Hallawa

Reported

2020-12-02 17:55 PST

Modified

2022-02-10 16:16 PST History

CC List

23 users Show

URL

Keywords InRadar

Depends on

220079 221550 221637

Blocks

Dependencies

tree graph

annulen
bfulgham
changseok
dino
esprehn+autocc
ews-watchlist
fmalita
fred.wang
glenn
gyuyoung.kim
hi
joepeck
kondapallykalyan
pdr
ryuan.choi
sabouhallawa
schenney
sergio
simon.fraser
thorton
webkit-bug-importer
wenson_hsieh
zalan