1195788 - Reftest analyzer does not highlight differences on Windows 10

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Mason Chang [Inactive] [:mchang]]

The reftest analyzer does not highlight pixel differences with nightly on Windows 10. Clicking show differences does nothing.

STR:
1) Get a reftest failure in the reftest analyzer
2) click a failed test.
3) Click highlight differences
4) See that nothing is highlighted.

Comment 1

[Mason Chang [Inactive] [:mchang]]

Attachment: screenshot

Screenshot from the reftest analyzer.

An example reftest analyzer link:

http://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=http://ftp.mozilla.org/pub/mozilla.org/firefox/try-builds/[e-mail]-bf08c0c4dbed/try-linux64-debug/try_ubuntu64_vm-debug_test-reftest-1-bm116-tests1-linux64-build415.txt.gz&only_show_unexpected=1

Comment 2

[Jeff Muizelaar [:jrmuizel]]

Weird. Disabling hw acceleration fixes the problem to me which suggests that it's related to the d2d filter code.

Comment 3

[Jeff Muizelaar [:jrmuizel]]

Someone should check on a different version of Windows to see if it's only a Win10 issue

Comment 4

[Mason Chang [Inactive] [:mchang]]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #3)
> Someone should check on a different version of Windows to see if it's only a
> Win10 issue

This works fine for me on Windows 7.

Comment 5

[Bas Schouten (:bas.schouten)]

A reduced testcase for this filtergraph would be really really nice...

Comment 6

[Milan Sreckovic [:milan] (needinfo for best results)]

(In reply to Bas Schouten (:bas.schouten) from comment #5)
> A reduced testcase for this filtergraph would be really really nice...

If one doesn't show up, lets try without it.

Comment 7

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

I'm seeing this on Windows 10 for this failure: http://hg.mozilla.org/mozilla-central/raw-file/tip/layout/tools/reftest/reftest-analyzer.xhtml#logurl=http://ftp.mozilla.org/pub/mozilla.org/firefox/tinderbox-builds/mozilla-inbound-macosx64/1441997691/mozilla-inbound_snowleopard_test-reftest-bm106-tests1-macosx-build2754.txt.gz&only_show_unexpected=1

I'll have to go see if I still have any Windows 8/8.1 devices that I can test this on...

Comment 8

[Xidorn Quan [:xidorn] UTC+10]

It seems to me the feComponentTransfer nodes c1->w1 and c2->w2 doesn't take any effect.

Comment 9

[Xidorn Quan [:xidorn] UTC+10]

Well, not "doesn't take any effect", but has a wrong effect...

Comment 10

[Xidorn Quan [:xidorn] UTC+10]

c1 and w2 are identical, and c2 and w1 are identical. It seems to me there is some mis-optimization happens.

Bas, hopefully this information is enough for you to figure out what's wrong there.

Comment 11

[Xidorn Quan [:xidorn] UTC+10]

So basically the formulas are
w1 = img1 + ~img2
w2 = img2 + ~img1
c1 = ~w1
c2 = ~w2

Given c1 == w2 and c2 == w1, it seems some optimizer collapses the formulas together, so
c1 = ~(img1 + ~img2) = ~img1 + img2 = w2
c2 = ~(img2 + ~img1) = ~img2 + img1 = w1

These equations look reasonable at first glance, but they are incorrect because arithmetic compositer clamps the result to [0..1] which is what the analyzer relies on.

Comment 12

[Xidorn Quan [:xidorn] UTC+10]

Attachment: patch

Comment 13

[Xidorn Quan [:xidorn] UTC+10]

Attachment: patch

Comment 14

[Xidorn Quan [:xidorn] UTC+10]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=a24aec074c9b

Comment 15

[Bas Schouten (:bas.schouten)]

Comment on attachment 8701370 [details] [diff] [review]
patch

Review of attachment 8701370 [details] [diff] [review]:
-----------------------------------------------------------------

Good catch!

Comment 16

[Xidorn Quan [:xidorn] UTC+10]

https://hg.mozilla.org/integration/mozilla-inbound/rev/343ef099a285ee73e4fb6fa1a4f896da8f6cc0f2
Bug 1195788 - Set arithmetic composite clamp to true for D2D filter node. r=bas

Comment 17

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/343ef099a285

Comment 18

[Xidorn Quan [:xidorn] UTC+10]

Comment on attachment 8701370 [details] [diff] [review]
patch

Approval Request Comment
[Feature/regressing bug #]: unknown, hw acceleration related issue
[User impact if declined]: may see incorrect rendering from svg filter
[Describe test coverage new/current, TreeHerder]: new test added with the patch
[Risks and why]: low risk. This patch just sets the clamp property on the d2d impl which is required by the svg filter spec. So it doesn't affect anything other than windows hw accel part. And this patch itself is simple and straightforward enough.
[String/UUID change made/needed]: n/a

Comment 19

[Sylvestre Ledru [:Sylvestre]]

Comment on attachment 8701370 [details] [diff] [review]
patch

With test! Thanks :)

Comment 20

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/releases/mozilla-aurora/rev/df26cbee3d0c