962517 - add a way to create canvas without hardware acceleration (chrome only)

Status: RESOLVED DUPLICATE of bug 884226

(Core :: Graphics, defect)

Description

[Andreas Gal :gal]

We use skiaGL on FFOS by default. As a result every 2d canvas we create is hardware accelerated. In some cases that doesn't pay off. For example, we snapshot an app during app switching. For this we create a canvas which forces the GL library to be loaded during a very GPU resource intensive animation moment (old app is fading out, new app is being drawn). Spinning up a GL canvas doesn't help here actually. We are repainting the content right now in software and then blit it onto the gralloc-backed canvas.

I propose a simple fix here to allow chrome to create a software-backed canvas. I want to do this instead of demote() to avoid loading the GL libraries, especially during the transition itself.

syntax: canvas.getContext("software-2d")

Again, this is chrome only and won't affect content. For content we probably should queue draw calls and then device whether using GL is worth it and do all of this on a separate thread to boot.

Comment 1

[Andreas Gal :gal]

Attachment: Allow chrome to create a software-only canvas.

Comment 2

[Andreas Gal :gal]

Attachment: Allow chrome to create a software-only canvas.

Comment 3

[Andreas Gal :gal]

This makes the app transition animation homescreen -> settings app butter smooth on my nexus 5. The problem here was really loading the GL drivers into the content process and/or creating the new GL canvas for the snapshot. That was locking the GPU somehow. This will likely also hurt on other GPUs and drivers.

Comment 4

[Benoit Girard (:BenWa)]

Comment on attachment 8363590 [details] [diff] [review]
Allow chrome to create a software-only canvas.

Review of attachment 8363590 [details] [diff] [review]:
-----------------------------------------------------------------

bjacob is a better fit for Canvas reviews.

Comment 5

[Benoit Jacob [:bjacob] (mostly away)]

According to hg log, I've only ever reviewed 3 patches touching that file, all specific to Skia/GL :-)

This piqued my curiosity though, as I had previously been wondering who would be a good reviewer for general Canvas2D change, so here are the 10 top reviewers for this file over the past year:

$ hg log -d -365 content/canvas/src/CanvasRenderingContext2D.cpp | grep ^summary | grep 'r=' | sed 's|.*r=||g' | sort |uniq -c | sort -rn | head -n10
     21 roc
     18 bz
     15 mattwoodrow
     10 smaug
      6 Ms2ger
      5 peterv
      5 Bas
      4 roc,bz
      4 bz.
      3 snorp

Given the contents of the present patch, adding a new canvas context ID, Roc would probably be the right reviewer.

That said, adding a new context ID is a really important basic change that needs agreeing on on the conceptual level before reviewing. Roc is again a person to be asked about that, but I would also suggest involving :jrmuizel in this conversation.

Comment 6

[Jeff Muizelaar [:jrmuizel]]

It seems like this just a dup of bug 884226

Comment 7

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

Yes, we're using this in particular to avoid creating SkiaGL canvas in the scenario where doing so would have us OOM, without much benefit if we don't, because we have to read back.

Comment 8

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

This has been uplifted to 1.2.

Comment 9

[Andreas Gal :gal]

Can someone land the rest of the patch please that uses this for screenshots? Might be worth using this elsewhere too in Gaia. Milan please find the right help if your team can't do the changes.

Comment 10

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

Sure, I'll create a Gaia bug for this, so that we don't try and morph this into it.