901225 - WebGL rendering seems do not lock gralloc buffers correctly on b2g master

Status: RESOLVED DUPLICATE of bug 917444

(Core :: Graphics: CanvasWebGL, defect)

Description

[Sotaro Ikeda [:sotaro]]

When using CrystalSkull App, the app's rendering often have a vertical and horizontal line.

Comment 1

[Sotaro Ikeda [:sotaro]]

Attachment: screenshot

screen shot of the problem. Color is also incorrect. The color problem seems Bug 886415.

Comment 2

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

Peter, what do you think?

Comment 3

[Peter Chang[:pchang]]

(In reply to Milan Sreckovic [:milan] from comment #2)
> Peter, what do you think?

I think this issue was related to bug 894847.
https://bugzilla.mozilla.org/show_bug.cgi?id=894847#c2

If we change texture swap of CanvasGL from async to sync, it will become better.
http://mxr.mozilla.org/mozilla-central/source/gfx/layers/client/CanvasClient.cpp#123

Because async way couldn't guarantee the buffer synchronization between content and chrome process.

Comment 4

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

Wait for the needinfo on bug 894847.  Sotaro, this only happens when HWC is not available?

Comment 5

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

Also, is sync texture swap the only way to deal with this?

Comment 6

[Sotaro Ikeda [:sotaro]]

It happens only on HwComposer rendering. I do not see this on OpenGL rendering. But on OpenGL rendering, performance is too bad(Bug 894847);

Comment 7

[Sotaro Ikeda [:sotaro]]

This rendering problem might not be related to genlock.

Comment 8

[Peter Chang[:pchang]]

(In reply to Milan Sreckovic [:milan] from comment #5)
> Also, is sync texture swap the only way to deal with this?

In my opinion, I think sync texture swap for b2g only could be a short-term solution.

For long-term solution, we still need to consider buffer management between content and chrome process to guarantee compositor doesn't use the same buffer as content process.

Comment 9

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

Still can't decide if this is a blocker for 1.2.  What is it about CrystalScull app that triggers this?  Where else would we expect to see this kind of a problem?

Comment 10

[Sotaro Ikeda [:sotaro]]

The root cause of the problem seems same as Bug 917444. It is already koi+. By changing "gfx.gralloc.fence-with-readpixels" to true, the problem seems to be fixewd as in Bug 917444 Comment 2.
- Bug 917444 - [B2G][Gallery][Edit]Effects only filter part of an image and sometimes both parts incorrectly

Comment 11

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

Let's aim for koi+, and once we verify the same root cause, we can dupe this bug to 917444.

Comment 12

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

Diego, we think this is the same as bug 917444 that you have.

Comment 13

[Peter Chang[:pchang]]

(In reply to Sotaro Ikeda [:sotaro] from comment #10)
> The root cause of the problem seems same as Bug 917444. It is already koi+.
> By changing "gfx.gralloc.fence-with-readpixels" to true, the problem seems
> to be fixewd as in Bug 917444 Comment 2.
> - Bug 917444 - [B2G][Gallery][Edit]Effects only filter part of an image and
> sometimes both parts incorrectly

The result for "UpdateTextureSync" or "force readpixel" on content side both could fix the problem because they will wait for a period time for draw calls.

But "force readpixel" method may apply to content app which already calls "UpdateTextureSync" and it is not necessary for it.

Comment 14

[Sotaro Ikeda [:sotaro]]

It is a dup of Bug 917444. And discussion is going on there.