1507227 - Android: Photo of firefox logo on mozilla.org has visual artifacts [Adreno 3xx]

Status: RESOLVED WORKSFORME

(Core :: Graphics: WebRender, defect, P2)

Description

[Kartikaya Gupta (email:[e-mail])]

Start the geckoview example app with WR enabled.
Wait for it to load mozilla.org
Scroll down to the second headline ("Firefox on the Cheddar") and look at the image above it (The logo on the Firefox monument in SF)

Expected:
- image renders properly

Actual:
- image has visual artifacts

Tested on a Sony Z3C

I'll attach screenshots in a bit, currently attempting a bisection, since this regressed pretty recently.

Comment 1

[Kartikaya Gupta (email:[e-mail])]

https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=e1143f990aba5b34db7cde4e31b91f72f049bbe3&tochange=e92582fa73c6fc6902cb6236bd404b2eaae896cb

Regression from servo/webrender#3283.

Comment 2

[Kartikaya Gupta (email:[e-mail])]

Attachment: One instance of this bug

Comment 3

[Kartikaya Gupta (email:[e-mail])]

Attachment: Another instance of this bug

Comment 4

[Kartikaya Gupta (email:[e-mail])]

(In reply to Kartikaya Gupta (email:[e-mail]) from comment #1)
> Regression from servo/webrender#3283.

Err that should be servo/webrender#3282.

Comment 5

[Bobby Holley (:bholley)]

This seems likely related to the issue in bug 1505508?

Comment 6

[Kartikaya Gupta (email:[e-mail])]

They might be related, but bug 1505508 was present even before servo/webrender#3282 got merged (in fact it's been present ever since servo/webrender#3239 made the glyphs render at all).

If it's helpful I can bisect this bug further to identify which of the commits in servo/webrender#3282 caused the regression.

Comment 7

[Bobby Holley (:bholley)]

(In reply to Kartikaya Gupta (email:[e-mail]) from comment #6)
> They might be related, but bug 1505508 was present even before
> servo/webrender#3282 got merged (in fact it's been present ever since
> servo/webrender#3239 made the glyphs render at all).
> 
> If it's helpful I can bisect this bug further to identify which of the
> commits in servo/webrender#3282 caused the regression.

My suspicion is that it's the "Dynamically grow cache textures" commit. That would definitely cause breakage if the blit behavior identified in bug 1505508 is actually happening. Worth confirming it's that one, and not "Use one layer per region in the texture cache" (which was the cause of that mac bug we had recently).

Comment 8

[Kartikaya Gupta (email:[e-mail])]

Yup, verified.

https://treeherder.mozilla.org/#/jobs?repo=try&group_state=expanded&revision=9effd37da2c3bd5fbd46f8f1c7af3ceea3fd9480 is good
https://treeherder.mozilla.org/#/jobs?repo=try&group_state=expanded&revision=9a9be08d27b0b969122052fb0cb26ff45849f558 is bad

So that's from commit 45f3af1cabc645d54f1167207c10216b52c72cb7 in WR, which is "Dynamically grow cache textures"

Comment 9

[Bobby Holley (:bholley)]

Ok, that's the patch that introduced the blit, so that seems very likely to be the same issue as bug 1505508.

Comment 10

[Kartikaya Gupta (email:[e-mail])]

ni to myself to retest

Comment 11

[Kartikaya Gupta (email:[e-mail])]

mozilla.org still renders with all sorts of garbage on the Z3C

Comment 12

[Jamie Nicol [:jnicol]]

This no longer seems to apply (probably due to no longer using texture arrays). We can always re-open if we see something similar again.