Buffer returned textures from render to VDA

Buffer returned textures from render to VDA

This is CrOS ARM only hot fix for tearing/out of order frames caused
by render/decode synchronization issues until we understand the
underlying problem (presumably in the driver) better.

The main effect of this change is to add an extra delay of 2 frames
between draw and decode to clearly separate both. We had a similar
hack in Flash when synchronization was broken (then in particular
 > 2 was used to achieve sufficient delay). We have a total of 8
PictureBuffers in flight but only need about 4 for smooth playback.
(With this change we are down to 6 which is still sufficient.)

BUG=chromium:225563
TEST=Ran on Daisy with different problematic videos and did not notice problems.
NOTRY=true

[ilja, 2013-04-13 01:09:52]

Please take a look.

[sheu, 2013-04-13 01:13:24]

lgtm

Nit: I think this change would be better described as:

Buffer returned textures from render to VDA

This is CrOS ARM only hot fix for tearing/out of order frames caused
by render/decode synchronization issues until we understand the
underlying problem (presumably in the driver) better.

[ilja, 2013-04-13 01:25:31]

Done.

Ami: could you please do an owner review?

[ilja, 2013-04-13 02:18:06]

I have increased the number of buffered frames to 2 as discussed with John. This
is to cover up latency from the Draw call to actually finishing it. Flash used
this number in the past (1 was not always sufficient then) and I feel it is the
safer change for now.

[ilja, 2013-04-13 04:32:18]

Looks like Ami is out. Dale/Xiaohan could one of you do an owners review? This
is supposed to land over the weekend.

[xhwang, 2013-04-13 18:59:11]

I am not very familiar with this code and the bug. It looks hacky to me since
it's timing related ("cover up latency from the Draw call to actually finishing
it"). But I'll rubberstamp LGTM if you really need/want to land it over the
weekend.

Questions and comments:
- Can we make the CL description more explicit about what we are doing? It's a
little bit confusing since "buffer" is also a noun.
- Add a BUG to the CL description?
- It's mentioned that Flash used 2 before. Can you put that part in the comment?
- So do you know the history why Flashed used 2 before but we changed it to 0?
Just to make sure we don't introduce other issues.

[ilja, 2013-04-14 17:58:13]

Sorry, I had a BUG and TEST field that deleted when updating the description.
Please let me know if the updated description doesn't address all your concerns.

[commit-bot: I haz the power, 2013-04-15 04:24:42]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/ihf@chromium.org/14238003/6001

[xhwang, 2013-04-15 18:53:12]

New CL description lgtm.

[ilja, 2013-04-16 01:06:38]

This was submitted by hshi as CL 194267 commit
224016e9a37f89694e8d15e320de261c1399bfe9 as the CQ keep being down.