831525 - Enable Skia canvas rendering on windows xp

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Matt Woodrow (:mattwoodrow)]

We should just do this, it was definitely faster than GDI when I tested it last.

Unfortunately I can't find my benchmark results from last time, but I can probably repeat them if people need convincing. Please suggest specific canvas demos/benchmarks you'd like me to try if so.

Comment 1

[George Wright (:gw280) (needinfo me!)]

On unaccelerated windows xp, we already pass all canvas tests, so this is hopefully just a decision that needs to be made.

Comment 2

[Jet Villegas (inactive)]

What's involved in turning this on for XP? build-config? runtime pref?

Comment 3

[Matt Woodrow (:mattwoodrow)]

Just a run time pref.

The harder part is agreeing that it's what we want to do, and fixing the last remaining test failure.

Comment 4

[Joe Drew (not getting mail)]

Skia is definitely faster for any canvas users that stroke lots of paths, but the part I don't know is whether it's faster (or at least the same speed) for bitmap blitting and scaling. I'd like to see a couple of benchmarks on things like GUIMark and e.g. FishIE to see whether we're going to be taking a hit on blitting operations. If not, we should be good to go for unaccelerated Windows (and Linux probably too).

Comment 5

[Jet Villegas (inactive)]

(In reply to Joe Drew (:JOEDREW! \o/) from comment #4)
> Skia is definitely faster for any canvas users that stroke lots of paths,
> but the part I don't know is whether it's faster (or at least the same
> speed) for bitmap blitting and scaling. 

Can someone post the pref string? We can ask the Shumway team to compare bitmap & vector drawing perf.

Comment 6

[Joe Drew (not getting mail)]

Set gfx.canvas.azure.backends to the literal string "skia" (no quotes).

Comment 7

[Yury Delendik (:yury)]

FWIW, was playing with http://ie.microsoft.com/testdrive/Performance/SpeedReading/default.html, the speed for Windows 7 GDI: 15 fps, skia: 25 fps, new directx 60 fps

Comment 8

[Jet Villegas (inactive)]

Is there any reason to believe that our GDI vs. Skia numbers on XP will differ from what Yury found on Win7? Can we turn it on in Nightly and get more people to try it out?

Comment 9

[Jet Villegas (inactive)]

Attachment: Adds skia to the Windows XP canvas backend types

I'm not sure if this is the correct order of backends to enable. Please check!

Comment 10

[Matt Woodrow (:mattwoodrow)]

Comment on attachment 704690 [details] [diff] [review]
Adds skia to the Windows XP canvas backend types

Review of attachment 704690 [details] [diff] [review]:
-----------------------------------------------------------------

Looks good to me.

Note that we will fail 2 tests on tinderbox if we land this, we need to get bug 831529 fixed first. Or possibly mark the tests as known failures until it's fixed.

Comment 11

[Matt Woodrow (:mattwoodrow)]

Benchmarking results:

Skia

FishBowl - 38 fps
Asteroids - 2391, 50fps
WebVisBench - 4480, 16.59fps
Smashcat - 77fps

Cairo

FishBowl - 24fps
Asteroids - 2271, 48fps
WebVisBench - 3140, 5.03fps
Smashcat - 65fps

Comment 12

[Jeff Muizelaar [:jrmuizel]]

Upstream cairo is noticeably faster than the in-tree one. I'll try to do some measurements to compare where it's at.

Comment 13

[Matt Woodrow (:mattwoodrow)]

Jeff: Does that mean you're offering to do a cairo update if it is faster?

I'd still like to enable skia in the meantime unless it's going to happen quickly. Skia keeps rotting without being tested, and it's an easy win.

Comment 14

[Jeff Muizelaar [:jrmuizel]]

(In reply to Matt Woodrow (:mattwoodrow) from comment #13)
> Jeff: Does that mean you're offering to do a cairo update if it is faster?

Maybe.

> I'd still like to enable skia in the meantime unless it's going to happen
> quickly. Skia keeps rotting without being tested, and it's an easy win.

I agree. There's no reason to wait for the cairo results.

Comment 15

[Matt Woodrow (:mattwoodrow)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/a934441a70c2

Comment 16

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/a934441a70c2

Comment 17

[Jonathan Kew [:jfkthame]]

This bug was filed re Windows XP, but the pref change here will affect all Windows systems. Is that understood/intended/ok? See also bug 856394.