Add EGL_ARM_implicit_external_sync extension support

Add EGL_ARM_implicit_external_sync extension support

And use it to flush rendering before page flip.

BUG=412508, 442605
TEST=build and run Chrome on jerry/freon. It should render one frame.
R=alexst@chromium.org
R=dnicoara@chromium.org
R=kbr@chromium.org
R=piman@chromium.org

Signed-off-by: Dominik Behr <dbehr@chromium.org>;

Committed: https://crrev.com/5addbd55b7adfd6c158ee169d5e56a5c3567a06b
Cr-Commit-Position: refs/heads/master@{#315171}

[dbehr, 2015-02-07 00:44:51]

dbehr@chromium.org changed reviewers:
+ kbr@chromium.org
- kbr@chromiun.org

[piman, 2015-02-07 01:13:17]

LGTM, but would like dnicoara@ to take a look too.

https://codereview.chromium.org/903063003/diff/20001/ui/gl/gl_surface_ozone.cc
File ui/gl/gl_surface_ozone.cc (right):

https://codereview.chromium.org/903063003/diff/20001/ui/gl/gl_surface_ozone.c...
ui/gl/gl_surface_ozone.cc:178: EGL_SYNC_FLUSH_COMMANDS_BIT_KHR,
EGL_FOREVER_KHR);
nit: if we're doing EGL_SYNC_FLUSH_COMMANDS_BIT_KHR, we don't need the glFlush
on l.166. Maybe we can change the logic as 
if (has_implicit_external_sync_) {
  // Do fence logic
} else {
  glFlush();
}

https://codereview.chromium.org/903063003/diff/20001/ui/gl/gl_surface_ozone.c...
ui/gl/gl_surface_ozone.cc:178: EGL_SYNC_FLUSH_COMMANDS_BIT_KHR,
EGL_FOREVER_KHR);
As a potential optimization, should we delay the ClientWaitSync until we're
about to submit the flip? In case there's already a pending flip we will delay
scheduling this flip, so we could delay the ClientWaitSync for better overlap.
Maybe keep as a TODO / file a bug?

[dnicoara, 2015-02-07 02:03:36]

lgtm

https://codereview.chromium.org/903063003/diff/20001/ui/gl/gl_surface_ozone.cc
File ui/gl/gl_surface_ozone.cc (right):

https://codereview.chromium.org/903063003/diff/20001/ui/gl/gl_surface_ozone.c...
ui/gl/gl_surface_ozone.cc:178: EGL_SYNC_FLUSH_COMMANDS_BIT_KHR,
EGL_FOREVER_KHR);
On 2015/02/07 01:13:17, piman (Very slow to review) wrote:
> As a potential optimization, should we delay the ClientWaitSync until we're
> about to submit the flip? In case there's already a pending flip we will delay
> scheduling this flip, so we could delay the ClientWaitSync for better overlap.
> Maybe keep as a TODO / file a bug?

+1 for TODO and bug. I was talking to Alex about this a few weeks back. We were
discussing optimizing this so we wouldn't block the GPU main thread for this
operation.

[dbehr, 2015-02-07 02:08:02]

I do not think glFlush does anything heavy weight in Mali driver so this
wouldn't help much.

This fence just flushes the work to kernel, I do not think this wait actually
waits for anything so moving it would not achieve much or anything at all. And
the following page flip is only several function calls away.

[dbehr, 2015-02-07 02:20:18]

New patchsets have been uploaded after l-g-t-m from
piman@chromium.org,dnicoara@chromium.org

[piman, 2015-02-07 02:30:07]

lgtm

[dnicoara, 2015-02-07 02:31:49]

On 2015/02/07 02:08:02, dbehr wrote:
> I do not think glFlush does anything heavy weight in Mali driver so this
> wouldn't help much.

I think this is more for correctness purposes (especially if some future
platform does perform something more heavyweight in glFlush()).

> 
> This fence just flushes the work to kernel, I do not think this wait actually
> waits for anything so moving it would not achieve much or anything at all. And
> the following page flip is only several function calls away.

The docs mention that the calling thread blocks on eglClientWaitSyncKHR(), so
there is some room for optimization. Though we can address that later if needed.

[dbehr, 2015-02-07 02:32:21]

The CQ bit was checked by dbehr@chromium.org

[commit-bot: I haz the power, 2015-02-07 02:32:59]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/903063003/60001

[commit-bot: I haz the power, 2015-02-07 03:18:19]

Committed patchset #4 (id:60001)

[commit-bot: I haz the power, 2015-02-07 03:19:37]

Patchset 4 (id:??) landed as
https://crrev.com/5addbd55b7adfd6c158ee169d5e56a5c3567a06b
Cr-Commit-Position: refs/heads/master@{#315171}