773460 - [Azure] pref on Azure canvas

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Nick Cameron [:nrc]]

We are going to land bug 764125 preffed off to a greater or lesser degree. We should at some point (or points) pref it on. Creating this bug to make a plan and so I don't forget.

Comment 1

[Nick Cameron [:nrc]]

We should land wihout 774775, but we can't pref on Android until 774775 lands

Comment 2

[Nick Cameron [:nrc]]

Attachment: turn on for Windows

Comment 3

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Comment on attachment 647399 [details] [diff] [review]
turn on for Windows

Review of attachment 647399 [details] [diff] [review]:
-----------------------------------------------------------------

Why are you removing the non-Windows non-Mac "pref("gfx.canvas.azure.enabled", false);"?

Comment 4

[Nick Cameron [:nrc]]

The plan is to turn Azure/Cairo on for Windows (as fallback if D2D is present, as preferred backend if not) now. We will turn on Cairo for Android, Linux, and Mac soon (as long as perf looks good), and switch to Skia (Windows XP at first) if the perf numbers are in favour (Joe is investigating).

Comment 5

[Nick Cameron [:nrc]]

(In reply to Robert O'Callahan (:roc) (Mozilla Corporation) from comment #3)
> Comment on attachment 647399 [details] [diff] [review]
> turn on for Windows
> 
> Review of attachment 647399 [details] [diff] [review]:
> -----------------------------------------------------------------
> 
> Why are you removing the non-Windows non-Mac
> "pref("gfx.canvas.azure.enabled", false);"?

I don't think it should be there - previously it was not there, I added it with one of the last set of patches, but I don't think I should have.

Comment 6

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Why shouldn't it be there? Something should define this pref for non-Win non-Mac platforms.

Comment 7

[Nick Cameron [:nrc]]

Because it wherever it is checked the default is false. Should I leave the pref in?

Comment 8

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Yes. Generally speaking all prefs should be in all.js. That helps people flip them in about:config.

Comment 9

[Nick Cameron [:nrc]]

Attachment: patch

leave the pref for non-Win, non-Mac

Comment 10

[Robert O'Callahan (:roc) (email my personal email if necessary)]

Comment on attachment 647418 [details] [diff] [review]
patch

Review of attachment 647418 [details] [diff] [review]:
-----------------------------------------------------------------

r+ with that

::: modules/libpref/src/init/all.js
@@ +236,5 @@
>  pref("gfx.canvas.azure.enabled", true);
>  pref("gfx.canvas.azure.backends", "cg");
>  #else
>  pref("gfx.canvas.azure.enabled", false);
> +pref("gfx.canvas.azure.backends", "");

make this "cairo" so that someone manually switching azure on will get something.

Comment 11

[Nick Cameron [:nrc]]

Attachment: patch: pref on Azure/Cairo for Windows

addressed comment, carrying r=roc

Comment 12

[Nick Cameron [:nrc]]

https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=5276035f74f3

Comment 13

[Nick Cameron [:nrc]]

backed out for Android XUL bustage (!): https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=8051eeb12d0c

Comment 14

[Nick Cameron [:nrc]]

https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=032ba64ab1f1

Comment 15

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/032ba64ab1f1

Comment 16

[Nick Cameron [:nrc]]

Attachment: patch: pref on Azure/Cairo for Android

Comment 17

[Nick Cameron [:nrc]]

Try push for Android (Android M1 broken): https://tbpl.mozilla.org/?tree=Try&rev=8305f258db00

With the fix: https://tbpl.mozilla.org/?tree=Try&rev=1d5599ccf295

Comment 18

[Nick Cameron [:nrc]]

Attachment: IsLinux should return false for Android

Comment 19

[Joe Drew (not getting mail)]

Comment on attachment 650433 [details] [diff] [review]
patch: pref on Azure/Cairo for Android

Review of attachment 650433 [details] [diff] [review]:
-----------------------------------------------------------------

Only concern here is that I believe this will also turn it on for B2G. Not too big a deal, though.

Comment 20

[Nick Cameron [:nrc]]

(In reply to Joe Drew (:JOEDREW!) from comment #19)
> Comment on attachment 650433 [details] [diff] [review]
> patch: pref on Azure/Cairo for Android
> 
> Review of attachment 650433 [details] [diff] [review]:
> -----------------------------------------------------------------
> 
> Only concern here is that I believe this will also turn it on for B2G. Not
> too big a deal, though.

Got to happend and some point, and hopefully it will get us fixing bugs quicker :-)

Comment 21

[Nick Cameron [:nrc]]

https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=5976c3d86e75

Comment 22

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/c25027433c15
https://hg.mozilla.org/mozilla-central/rev/2730ef5ea1d0

Comment 23

[Nick Cameron [:nrc]]

Attachment: patch: pref on for b2g

Comment 24

[Nick Cameron [:nrc]]

Turns out that Android patch won't turn on Azure/Cairo canvas for b2g (so I'm told), so the new patch hopefully does that.

Comment 25

[Nick Cameron [:nrc]]

Benchmarking
------------

Win7 no D2D, D3d9 Layers
------------------------

Cairo

Fish Bowl       - 10 fish, 8fps
Asteroids       - 47-52 fps
Paintball       - 55.68 secs, 93.76 pb/m
Webvizbench     - 3-5 fps


Skia

Fish Bowl       - 10 fish, 6fps
Asteroids       - 47-52 fps
Paintball       - 38.66 secs, 135.03 pb/m
Webvizbench     - 11-12 fps


Thebes

Fish Bowl       - 10 fish, 7-8fps
Asteroids       - 47-52 fps
Paintball       - 55.67 secs, 93.77 pb/m
Webvizbench     - 4-6 fps


Android, Nexus 7
----------------

(Fish bowl is pretty useless at 10 fish here)

Cairo

Fish Bowl       - 10 fish, 2fps
Asteroids       - 14-15 fps
http://www.smashcat.org/av/canvas_test/
                - 11 fps 

Skia

Fish Bowl       - 10 fish, 1fps
Asteroids       - 27-31 fps
http://www.smashcat.org/av/canvas_test/
                - 10 fps


Thebes

Fish Bowl       - 10 fish, 2fps
Asteroids       - 14-15 fps
http://www.smashcat.org/av/canvas_test/
                - 9-11 fps

                
Android, HTC One X
------------------

Cairo

Fish Bowl       - 10 fish, 1fps
Asteroids       - 11-16 fps
http://www.smashcat.org/av/canvas_test/
                - 10 fps
          

Skia

Fish Bowl       - 10 fish, 1fps
Asteroids       - 25-28 fps
http://www.smashcat.org/av/canvas_test/
                - 8 fps

          
Thebes

Fish Bowl       - 10 fish, 1fps
Asteroids       - 11-12 fps
http://www.smashcat.org/av/canvas_test/
                - 8-9 fps

Comment 26

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

Comment on attachment 651272 [details] [diff] [review]
patch: pref on for b2g

There are two gaia "demo games" that are fps sensitive.  One is broken (argh!), but the other shows no change in fps.

Comment 27

[Nick Cameron [:nrc]]

https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=5ac49037cb77

Comment 28

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/5ac49037cb77

Comment 29

[Michael Wu [:mwu]]

(In reply to Nick Cameron [:nrc] from comment #24)
> Turns out that Android patch won't turn on Azure/Cairo canvas for b2g (so
> I'm told), so the new patch hopefully does that.

It should and it appears to do so. ANDROID is defined on B2G.

Comment 30

[Nick Cameron [:nrc]]

(In reply to Michael Wu [:mwu] from comment #29)
> (In reply to Nick Cameron [:nrc] from comment #24)
> > Turns out that Android patch won't turn on Azure/Cairo canvas for b2g (so
> > I'm told), so the new patch hopefully does that.
> 
> It should and it appears to do so. ANDROID is defined on B2G.

I was told (possibly incorrectly) that b2g does not look at libpref/.../all.js, and so the patch is needed to add the pref to b2g.js. Are you saying that b2g uses both all.js and b2g.js? And therefore the pref can be removed from b2g.js? I'm afraid I have no idea about b2g, nor a working b2g build to test this with, so I am going on hearsay. If you could confirm what needs doing, that would be helpful.

Comment 32

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Attachment: perf on Azure/Cairo for everything else

Comment 33

[Nick Cameron [:nrc]]

Comment on attachment 654087 [details] [diff] [review]
perf on Azure/Cairo for everything else

Review of attachment 654087 [details] [diff] [review]:
-----------------------------------------------------------------

::: modules/libpref/src/init/all.js
@@ +241,5 @@
>  #ifdef ANDROID
>  pref("gfx.canvas.azure.enabled", true);
>  pref("gfx.canvas.azure.backends", "cairo");
>  #else
> +pref("gfx.canvas.azure.enabled", true);

Why not move this outside the ifdefs now that it holds everywhere?

Comment 34

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

I considered going that but figured that we're probably going to very soon move to skia on Android so I may as well just let sleeping dogs lie.

Comment 35

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Attachment: pref on Azure/Cairo for everything else

Comment 36

[Joe Drew (not getting mail)]

(In reply to Anthony Jones (:kentuckyfriedtakahe) from comment #34)
> I considered going that but figured that we're probably going to very soon
> move to skia on Android so I may as well just let sleeping dogs lie.

Only if it's faster!

Comment 37

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Attachment: pref on Azure/Cairo for everything v2

Comment 38

[Nick Cameron [:nrc]]

Attachment: remove the pref from b2g.js

As discussed with cjones on irc (and mentioned in comments above), b2g uses the ANDROID pref in all.js, so this change to b2g.js was unnecessary. Thus, removing it.

Comment 39

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

http://tbpl.mozilla.org/?tree=Try&rev=f4c7e27c718e

Comment 40

[Nick Cameron [:nrc]]

https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=caffdfa95b07

Comment 41

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

https://hg.mozilla.org/mozilla-central/rev/caffdfa95b07

Going to resolve this now, because the default has changed.

Comment 42

[Nick Cameron [:nrc]]

Reopening, because we need to backout the Linux pref patch, and put it back later.

Comment 43

[Nick Cameron [:nrc]]

Removing the pref from b2g.js:

https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=856299bce84f

Comment 44

[Nick Cameron [:nrc]]

Attachment: turn off for Linux again until we land performance improvements

[Approval Request Comment]
Bug caused by (feature/regressing bug #): attachment 655502 [details] [diff] [review]
User impact if declined: performance regressions
Testing completed (on m-c, etc.): n/a - just backing out a previous patch
Risk to taking this patch (and alternatives if risky): n/a
String or UUID changes made by this patch: none

The patch in 655502 shouldn't have made the Aurora uplift, other patches it depended on didn't make it.

Comment 45

[Nick Cameron [:nrc]]

Backing out the pref on for Linux:

https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=6885860f1319

Comment 46

[Nick Cameron [:nrc]]

Adding tracking flag because I need to pref off for Aurora (FF17), see comment 44, attachment 656705 [details] [diff] [review]

Comment 47

[Lukas Blakk [:lsblakk] use ?needinfo]

Comment on attachment 656705 [details] [diff] [review]
turn off for Linux again until we land performance improvements

Approving, please set the status flag for 17 to disabled once this has landed.

Comment 48

[Nick Cameron [:nrc]]

Turn off by default for Linux on Aurora: https://tbpl.mozilla.org/?tree=Mozilla-Aurora&rev=1bd8120a4e40

Comment 49

[Nick Cameron [:nrc]]

Try push for pref'ing Linux on by default: https://tbpl.mozilla.org/?tree=Try&rev=fec6e51fba98

Comment 50

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Both these try runs include this patch and are all green:

https://tbpl.mozilla.org/?tree=Try&rev=fec6e51fba98
https://tbpl.mozilla.org/?tree=Try&rev=dc20a1747a53

Comment 51

[Nick Cameron [:nrc]]

https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=1661cbbc3759

Comment 52

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/1661cbbc3759