1706222 - Rendering artifact with test case of Bug 1678938

Status: NEW

(Core :: Graphics: WebRender, defect, P3)

Description

[Sotaro Ikeda [:sotaro]]

Found the problem during looking into crash test crash with sw-wr on Android.

• https://bugzilla.mozilla.org/show_bug.cgi?id=1706011#c2

When I tested test contents of Bug 1678938, I saw rendering artifact both with WebRender and sw-wr on Win10 and on Linux.

The problem did not happen by disabling WebRender and chrome browser.

Comment 1

[Sotaro Ikeda [:sotaro]]

Attachment: screenshot 1

Comment 2

[Sotaro Ikeda [:sotaro]]

Attachment: screenshot 2

Comment 3

[Sotaro Ikeda [:sotaro]]

:gw, can you comment to the bug?

Comment 4

[Timothy Nikkel (:tnikkel)]

I get this as a regression range, and I double checked it

https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=aefea952b6974c29eb0503b419c3f3f236165921&tochange=e6cd9ff9eca2b2b3c45abde7fd8b62e79cf1aa38

Comment 5

[Glenn Watson [:gw]]

The transform matrix on that border primitive is:

        m11: 2513,
        m12: 4.666667,
        m13: 3,
        m14: 2.6666667,
        m21: 5029,
        m22: 11.333334,
        m23: 7.0000005,
        m24: 5.3333335,
        m31: 7545,
        m32: 18,
        m33: 11,
        m34: 8,
        m41: -2357183.5,
        m42: -4370.333,
        m43: -2810.5,
        m44: -2501.3333,

I suspect that non-WR Gecko is probably rejecting that transform and skipping rendering of that content?

Matt, do you know if this might be a case where Gecko is skipping rendering of this primitive and if so, where any such code might exist?

Comment 6

[Matt Woodrow (:mattwoodrow)]

Sorry, totally forgot about this.

I think we cull out parts of polygons that cross through the w=0 plane, which might be what's needed here. https://searchfox.org/mozilla-central/rev/dedc585e3028e9b1ac9815e49b21e8f37e5ed3b7/gfx/2d/Polygon.h#314-316

Comment 7

[Pablo Huckins]

Attachment: same problem

Thanks from verbdimati1972.