Bug 232469

Summary:

[GPU Process] [Filters 4/23] Make CSSFilter be a composite of FilterFunctions

Product:

WebKit

Reporter:

Said Abou-Hallawa <sabouhallawa>

Component:

Layout and Rendering

Assignee:

Said Abou-Hallawa <sabouhallawa>

Status:

RESOLVED FIXED

Severity:

Normal

CC:

bfulgham, changseok, dino, eric.carlson, esprehn+autocc, ews-watchlist, fmalita, fred.wang, glenn, gyuyoung.kim, hta, jer.noble, kondapallykalyan, macpherson, menard, pdr, philipj, schenney, sergio, simon.fraser, tommyw, webkit-bug-importer, zalan

Priority:

Keywords:

InRadar

Version:

WebKit Nightly Build

Hardware:

Unspecified

OS:

Unspecified

See Also:

https://bugs.webkit.org/show_bug.cgi?id=233008
https://bugs.webkit.org/show_bug.cgi?id=234918

Bug Depends on:

Bug Blocks:

231253

Attachments:

Description	Flags
Patch	ews-feeder: commit-queue-
Patch for review	simon.fraser: review+
Patch	ews-feeder: commit-queue-
Patch	ews-feeder: commit-queue-
Patch	none
Patch	none

Said Abou-Hallawa

Reported 2021-10-28 22:35:49 PDT

This will make CSSFilter have an item of SVGFilter in this list of FilterFunctions.

Attachments
Patch (151.60 KB, patch) 2021-10-28 22:55 PDT, Said Abou-Hallawa	ews-feeder: commit-queue-	Details Formatted Diff Diff
Patch for review (56.67 KB, patch) 2021-10-28 22:59 PDT, Said Abou-Hallawa	simon.fraser: review+	Details Formatted Diff Diff
Patch (155.29 KB, patch) 2021-10-31 18:01 PDT, Said Abou-Hallawa	ews-feeder: commit-queue-	Details Formatted Diff Diff
Patch (69.94 KB, patch) 2021-11-10 12:39 PST, Said Abou-Hallawa	ews-feeder: commit-queue-	Details Formatted Diff Diff
Patch (66.43 KB, patch) 2021-11-10 14:38 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Patch (66.49 KB, patch) 2021-11-10 19:58 PST, Said Abou-Hallawa	no flags	Details Formatted Diff Diff
Show Obsolete (4) View All Add attachment proposed patch, testcase, etc.

Said Abou-Hallawa

Comment 1 2021-10-28 22:55:27 PDT

Created attachment 442784 [details] Patch

Said Abou-Hallawa

Comment 2 2021-10-28 22:59:10 PDT

Created attachment 442785 [details] Patch for review

Simon Fraser (smfr)

Comment 3 2021-10-29 12:03:34 PDT

Comment on attachment 442785 [details] Patch for review View in context: https://bugs.webkit.org/attachment.cgi?id=442785&action=review > Source/WebCore/ChangeLog:8 > + In this patch, the CSS referenced filter is built as an SVGFilter and it "reference filter" > Source/WebCore/css/CSSFilterImageValue.cpp:120 > + auto texture = ImageBuffer::create(size, renderingMode, 1, DestinationColorSpace::SRGB(), PixelFormat::BGRA8); texture is a weird name for this. > Source/WebCore/css/CSSFilterImageValue.cpp:130 > + auto cssFilter = CSSFilter::create(m_filterOperations, renderingMode); > + > cssFilter->setSourceImageRect(imageRect); > if (!cssFilter->build(renderer, m_filterOperations, FilterConsumer::FilterFunction)) Why do we have to pass m_filterOperations both to the constructor, and to the build() function? > Source/WebCore/rendering/CSSFilter.cpp:223 > + auto builder = makeUnique<SVGFilterBuilder>(); Can SVGFilterBuilder just be on the stack? > Source/WebCore/rendering/RenderLayerFilters.h:66 > + GraphicsContext* beginFilterEffect(RenderElement&, const LayoutRect& filterBoxRect, const LayoutRect& dirtyRect, const LayoutRect& layerRepaintRect, GraphicsContext& destinationContext); I tend to prefer GraphicsContext& destinationContext as the first argument. > Source/WebCore/rendering/svg/SVGRenderTreeAsText.cpp:436 > + auto builder = makeUnique<SVGFilterBuilder>(); Can this live on the stack?

Said Abou-Hallawa

Comment 4 2021-10-31 17:18:41 PDT

Some of the layout tests are failing because of the need to set the sourceImageRect before building the CSSFilter. The sourceImageRect has to include the Outsets of the FilterEffects. So currently to fix these tests we need to set the sourceImageRect twice and to build the CSSFilter twice. The plan to remove this dependency and be able to build the CSSFilter without having to set sourceImageRect and have the FilterEffects be able to calculate the outsets without knowing the sourceImageRect. The reset of the failures are happing because of the color space which should be fixed once the calculation of the result color space out of the operating color space. So for now, I am going to skip these few tests and remove them by the end.

Said Abou-Hallawa

Comment 5 2021-10-31 18:01:03 PDT

Created attachment 442946 [details] Patch

Radar WebKit Bug Importer

Comment 6 2021-11-04 22:36:17 PDT

<rdar://problem/85047148>

Said Abou-Hallawa

Comment 7 2021-11-10 12:39:25 PST

Created attachment 443851 [details] Patch

Said Abou-Hallawa

Comment 8 2021-11-10 14:38:21 PST

Created attachment 443864 [details] Patch