[GPU] GLES2 lost context recovery

Added support for lost context recovery on the client side. None of our service side GL implementations actually report lost contexts (yet).

Added pglGetError to PGL library.

pglSwapBuffers returns false on a lost context or other non-recoverable error and pglGetError reports PGL_CONTEXT_LOST.

Updated demo plugins to reset their PGL contexts on context lost. Standalone plugins cannot currently recover from lost context because they don't use PGL.

Added error code to NPDeviceContext3D for lost context.

TEST=none
BUG=none


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=38039

[alokp, 2010-02-02 23:47:34]

http://codereview.chromium.org/566021/diff/4001/5013
File gpu/demos/framework/plugin.cc (right):

http://codereview.chromium.org/566021/diff/4001/5013#newcode99
gpu/demos/framework/plugin.cc:99: pglMakeCurrent(NULL);
pglMakeCurrent(NULL) should be moved to DestroyContext()

http://codereview.chromium.org/566021/diff/4001/5013#newcode132
gpu/demos/framework/plugin.cc:132: for (int i = 0; i < 2; ++i) {
Is there anything special about TWO times? Can the context get lost anytime,
during any function call? Since we are drawing continuously anyway, I think it
might be better to just do something like this:
if (!pgl_context_) CreateContext();
Draw();
SwapBuffers();
if (context lost) DestroyContext();

http://codereview.chromium.org/566021/diff/4001/5013#newcode150
gpu/demos/framework/plugin.cc:150: DCHECK(!pgl_context_);
any partcular reason for using DCHECK instead of CHECK? The test is not very
expensive.

[alokp, 2010-02-02 23:50:13]

Would it be possible to instead generate an event when context is lost. It will
be much more easier for the apps to register a callback that gets called when
context is lost.

[greggman, 2010-02-02 23:59:33]

On Tue, Feb 2, 2010 at 3:50 PM, <alokp@chromium.org> wrote:

> Would it be possible to instead generate an event when context is lost. It
> will
> be much more easier for the apps to register a callback that gets called
> when
> context is lost.


I don't see how it would be easier. The app would have no idea what state
it's in when the callback comes in would it? Most apps would end up setting
a flag and then doing exactly what Al has done no?


>
>
> http://codereview.chromium.org/566021
>

[alokp, 2010-02-03 00:07:05]

Yes you right. I see that EGL follows the same pattern:

"Following a power management event, calls to eglSwapBuffers, eglCopy-
Buffer, or eglMakeCurrent will indicate failure by returning EGL_FALSE. The
error EGL_CONTEXT_LOST will be returned if a power management event has oc-
curred."

[apatrick_chromium, 2010-02-03 00:25:24]

http://codereview.chromium.org/566021/diff/4001/5013
File gpu/demos/framework/plugin.cc (right):

http://codereview.chromium.org/566021/diff/4001/5013#newcode99
gpu/demos/framework/plugin.cc:99: pglMakeCurrent(NULL);
On 2010/02/02 23:47:34, alokp wrote:
> pglMakeCurrent(NULL) should be moved to DestroyContext()

Why? I put it there to pair it with pglMakeCurrent(pgl_context_). The current
context should always be NULL in DestroyContext(). Also, pglDestroyContext sets
the current context to NULL if you try and destroy it, just like EGL.

http://codereview.chromium.org/566021/diff/4001/5013#newcode132
gpu/demos/framework/plugin.cc:132: for (int i = 0; i < 2; ++i) {
I could make it just try once. That would mean Paint() definitely would not
render anything the first frame after a lost context. I'm not quite sure what
the semantics are for EGL. It's possible trying twice might still fail to render
anything if the context is lost twice in quick succession. I don't think it
should keep trying indefinitely. Thoughts gman?

http://codereview.chromium.org/566021/diff/4001/5013#newcode150
gpu/demos/framework/plugin.cc:150: DCHECK(!pgl_context_);
On 2010/02/02 23:47:34, alokp wrote:
> any partcular reason for using DCHECK instead of CHECK? The test is not very
> expensive.

It seems more like a logic error than a runtime error. It should only fire if
there is a bug. A better thing to do would be to check if pglCreateContext fails
and return an error from CreateContext(). I can do that if you want.

[alokp, 2010-02-03 00:42:47]

http://codereview.chromium.org/566021/diff/4001/5013
File gpu/demos/framework/plugin.cc (right):

http://codereview.chromium.org/566021/diff/4001/5013#newcode99
gpu/demos/framework/plugin.cc:99: pglMakeCurrent(NULL);
oh ok. I did not know pglDestroyContext already does that. Usually with wgl, I
set the current context to NULL before destroying a context.

http://codereview.chromium.org/566021/diff/4001/5013#newcode132
gpu/demos/framework/plugin.cc:132: for (int i = 0; i < 2; ++i) {
Moving error-checking to make-current may have a higher probability of drawing:
bool context_lost = !pglMakeCurrent(context) && pglGetError() ==
PGL_CONTEXT_LOST;
if (context_lost) {recreateContext};
Draw();
SwapBuffers();

http://codereview.chromium.org/566021/diff/4001/5013#newcode150
gpu/demos/framework/plugin.cc:150: DCHECK(!pgl_context_);
No it is fine the way it is.

[apatrick_chromium, 2010-02-03 01:24:04]

Done

[alokp, 2010-02-03 01:40:04]

lgtm