Adding flow traces for blink scheduler events

Adding flow traces for blink scheduler events

BUG=391005
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=181908

[picksi1, 2014-08-20 11:50:34]

This code adds flow events to tasks scheduled using the Blink scheduler to aid
visualization.

The TracedTask class has grown (a bit) and now has more code in the .cpp file;
I'm wondering if TracedTask should have its own .h & .cpp file, especially as
I'm declaring a couple of static member variables in the cpp?

[Sami, 2014-08-20 14:20:57]

Good idea about moving TracedTask out of the header. We can add a new .cpp and
.h for it and still keep the constructor private so only Scheduler can use it.

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
File Source/platform/scheduler/Scheduler.cpp (right):

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:96: 
nit: extra whitespace

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:250: "src_file", m_location.fileName(),
No need to report the same src_file and src_func here since we do that below.

Chromium also reports the queuing duration here[1]. I wonder if we could do the
same?

[1]
https://code.google.com/p/chromium/codesearch#chromium/src/base/debug/task_an...

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:267: :
m_FlowTraceID(s_NextFlowTraceID++)
Unfortunately TracedTasks can get constructed on arbitrary threads, so
incrementing the counter like this isn't thread safe. I think this needs to be
an atomic increment. Or better yet, we can make the counter a member of
Scheduler and increment it while m_pendingTaskMutex is held.

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
File Source/platform/scheduler/Scheduler.h (right):

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.h:21: typedef long long unsigned
FlowTraceID;
Probably better to use int64_t for this. Actually, it would also be simpler just
to use a naked int64_t since that's what the rest of chromium does.

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.h:90: FlowTraceID m_FlowTraceID;
s/m_FlowTraceID/m_flowTraceID/

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.h:95: Type m_type;
Instead of having a type enum, could we just pass the trace name as pointer to
this class? The string being pointed to is immutable so there are no lifetime
issues. This way we don't need to set up the pointers at construction time
either.

[picksi1, 2014-08-29 14:11:29]

Thread safety updates + enums removed & other issues addressed.

I'll add the queuing duration in a separate CL to prevent bloating this one.

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
File Source/platform/scheduler/Scheduler.cpp (right):

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:96: 
On 2014/08/20 14:20:57, Sami wrote:
> nit: extra whitespace

Done.

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:250: "src_file", m_location.fileName(),
On 2014/08/20 14:20:57, Sami wrote:
> No need to report the same src_file and src_func here since we do that below.
> 
> Chromium also reports the queuing duration here[1]. I wonder if we could do
the
> same?
> 
> [1]
>
https://code.google.com/p/chromium/codesearch#chromium/src/base/debug/task_an...

file and func removed, will add queue duration as a separate patch to avoid
bloating.

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:267: :
m_FlowTraceID(s_NextFlowTraceID++)
On 2014/08/20 14:20:57, Sami wrote:
> Unfortunately TracedTasks can get constructed on arbitrary threads, so
> incrementing the counter like this isn't thread safe. I think this needs to be
> an atomic increment. Or better yet, we can make the counter a member of
> Scheduler and increment it while m_pendingTaskMutex is held.

Done.

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
File Source/platform/scheduler/Scheduler.h (right):

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.h:21: typedef long long unsigned
FlowTraceID;
On 2014/08/20 14:20:57, Sami wrote:
> Probably better to use int64_t for this. Actually, it would also be simpler
just
> to use a naked int64_t since that's what the rest of chromium does.

Changed to an int so keep atomicincrement happy

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.h:90: FlowTraceID m_FlowTraceID;
On 2014/08/20 14:20:57, Sami wrote:
> s/m_FlowTraceID/m_flowTraceID/

This is no longer static in latest version

https://codereview.chromium.org/490913002/diff/40001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.h:95: Type m_type;
On 2014/08/20 14:20:57, Sami wrote:
> Instead of having a type enum, could we just pass the trace name as pointer to
> this class? The string being pointed to is immutable so there are no lifetime
> issues. This way we don't need to set up the pointers at construction time
> either.

Done.

[Sami, 2014-08-29 14:50:25]

Functionally looks pretty good -- just some style nits.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
File Source/platform/scheduler/Scheduler.cpp (right):

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:76: const Scheduler::Task& task, const
TraceLocation& location, const char * name, int flowTraceID)
No space before "*".

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:123: intptr_t thisAsPointer =
reinterpret_cast<intptr_t>(this);
Maybe it's just me, but I think this is a little verbose -- I'd just write the
conversion into a single line in the member initialization.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:146: void
Scheduler::postAsHighPriorityTask(const TraceLocation& location, const Task&
task, const char * traceName)
nit: postHighPriorityTask to match the wording of the other functions.

No space before "*".

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
File Source/platform/scheduler/Scheduler.h (right):

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.h:63: void postAsHighPriorityTask(const
TraceLocation&, const Task&, const char*);
It's not clear what the char* parameter is so it should probably have a name
here.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
File Source/platform/scheduler/TracedTask.h (right):

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/TracedTask.h:24: // class hidden to everyone apart
from the scheduler.
This is a fairly common pattern in Blink so I don't think it needs a comment.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/TracedTask.h:25: friend Scheduler;
nit: People tend to write "friend class Scheduler".

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/TracedTask.h:26: TracedTask(const Task&, const
TraceLocation&, const char*, int);
Both const char* and int should have a name here.

|int| should match the type of m_flowTraceID.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/TracedTask.h:28: unsigned long long m_FlowTraceID;
Lower caps for members: m_flowTraceID

This should probably be int64_t to match Chromium.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/TracedTask.h:31: const char * m_name;
No space before "*".

[picksi1, 2014-09-01 11:06:51]

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
File Source/platform/scheduler/Scheduler.cpp (right):

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:76: const Scheduler::Task& task, const
TraceLocation& location, const char * name, int flowTraceID)
On 2014/08/29 14:50:25, Sami wrote:
> No space before "*".

Done.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:123: intptr_t thisAsPointer =
reinterpret_cast<intptr_t>(this);
On 2014/08/29 14:50:25, Sami wrote:
> Maybe it's just me, but I think this is a little verbose -- I'd just write the
> conversion into a single line in the member initialization.

Done.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.cpp:146: void
Scheduler::postAsHighPriorityTask(const TraceLocation& location, const Task&
task, const char * traceName)
I'd named this differently on purpose as it is an internal (i.e. private)
function and has different parameters from the other PostTask functions (which
all take a location and a task).

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
File Source/platform/scheduler/Scheduler.h (right):

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/Scheduler.h:63: void postAsHighPriorityTask(const
TraceLocation&, const Task&, const char*);
On 2014/08/29 14:50:25, Sami wrote:
> It's not clear what the char* parameter is so it should probably have a name
> here.

Done.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
File Source/platform/scheduler/TracedTask.h (right):

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/TracedTask.h:24: // class hidden to everyone apart
from the scheduler.
On 2014/08/29 14:50:25, Sami wrote:
> This is a fairly common pattern in Blink so I don't think it needs a comment.

Done.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/TracedTask.h:25: friend Scheduler;
On 2014/08/29 14:50:25, Sami wrote:
> nit: People tend to write "friend class Scheduler".

Done.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/TracedTask.h:26: TracedTask(const Task&, const
TraceLocation&, const char*, int);
I'm keen to encapsulate the fact that a task ID ends up as an uint64_t inside
the TracedTask, as it feels like an implementation detail forced on us by the
MANGLE macro. I've kept the parameter an int and am explicitly casting it into a
uint64_t in the constructor.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/TracedTask.h:28: unsigned long long m_FlowTraceID;
On 2014/08/29 14:50:25, Sami wrote:
> Lower caps for members: m_flowTraceID
> 
> This should probably be int64_t to match Chromium.

Done.

https://codereview.chromium.org/490913002/diff/80001/Source/platform/schedule...
Source/platform/scheduler/TracedTask.h:31: const char * m_name;
On 2014/08/29 14:50:25, Sami wrote:
> No space before "*".

Done.

[alexclarke, 2014-09-01 11:24:20]

alexclarke@google.com changed reviewers:
+ alexclarke@google.com

[alexclarke, 2014-09-01 11:24:20]

https://codereview.chromium.org/490913002/diff/100001/Source/platform/TraceEv...
File Source/platform/TraceEvent.h (right):

https://codereview.chromium.org/490913002/diff/100001/Source/platform/TraceEv...
Source/platform/TraceEvent.h:458: // - category and name strings must have
application lifetime (statics or
|category|? and |name|?

https://codereview.chromium.org/490913002/diff/100001/Source/platform/TraceEv...
Source/platform/TraceEvent.h:459: //   literals). They may not include " chars.
inverted commas?

https://codereview.chromium.org/490913002/diff/100001/Source/platform/schedul...
File Source/platform/scheduler/Scheduler.cpp (right):

https://codereview.chromium.org/490913002/diff/100001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.cpp:147: TracedTask tracedTask =
TracedTask(task, location, traceName, generateFlowTraceID());
How about:

m_pendingHighPriorityTasks.append(TracedTask(task, location, traceName,
generateFlowTraceID()));

[picksi1, 2014-09-01 13:33:48]

https://codereview.chromium.org/490913002/diff/100001/Source/platform/TraceEv...
File Source/platform/TraceEvent.h (right):

https://codereview.chromium.org/490913002/diff/100001/Source/platform/TraceEv...
Source/platform/TraceEvent.h:458: // - category and name strings must have
application lifetime (statics or
You are correct, but as this code is cut-and-pasted from
src/base/debug/TraceEvent.h in chromium it is probably best to keep both
in-sync. As discussed I believe this file will be removed once Blink and
Chromium repo's are merged.

https://codereview.chromium.org/490913002/diff/100001/Source/platform/TraceEv...
Source/platform/TraceEvent.h:459: //   literals). They may not include " chars.
As above

https://codereview.chromium.org/490913002/diff/100001/Source/platform/schedul...
File Source/platform/scheduler/Scheduler.cpp (right):

https://codereview.chromium.org/490913002/diff/100001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.cpp:147: TracedTask tracedTask =
TracedTask(task, location, traceName, generateFlowTraceID());
In general I like to break complex lines down to aid readability, but I agree
splitting this into two merely bloats the code as this line is fairly self
explanatory. Done.

[Sami, 2014-09-01 13:51:29]

Thanks Simon, lgtm with a couple of nits.

https://codereview.chromium.org/490913002/diff/120001/Source/platform/schedul...
File Source/platform/scheduler/Scheduler.cpp (right):

https://codereview.chromium.org/490913002/diff/120001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.cpp:118: // flow trace ID begins as the
'this' pointer to help ensure different scheduler
nit: Start with capital 'f'.

https://codereview.chromium.org/490913002/diff/120001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.cpp:143: void
Scheduler::postHighPriorityTaskInternal(const TraceLocation& location, const
Task& task, const char * traceName)
No space before "*". I think "git cl format" may be able to sort these out
automatically.

[picksi1, 2014-09-01 15:02:32]

Fixed Sami's nits and ran git cl format over the code, which removed some blank
lines.

https://codereview.chromium.org/490913002/diff/120001/Source/platform/schedul...
File Source/platform/scheduler/Scheduler.cpp (right):

https://codereview.chromium.org/490913002/diff/120001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.cpp:118: // flow trace ID begins as the
'this' pointer to help ensure different scheduler
On 2014/09/01 13:51:29, Sami wrote:
> nit: Start with capital 'f'.

Done.

https://codereview.chromium.org/490913002/diff/120001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.cpp:143: void
Scheduler::postHighPriorityTaskInternal(const TraceLocation& location, const
Task& task, const char * traceName)
On 2014/09/01 13:51:29, Sami wrote:
> No space before "*". I think "git cl format" may be able to sort these out
> automatically.

Done. FYI "git cl format" messed up indentation of follow-on lines (e.g. when
parameters to a function are split across lines), indenting the additional
parameters to line up with the first parameter, not a 4-space indentation.

[eseidel, 2014-09-02 16:19:25]

we shoudl avoid teh atomicIncrement but otherwise lgtm.

https://codereview.chromium.org/490913002/diff/140001/Source/platform/schedul...
File Source/platform/scheduler/Scheduler.cpp (right):

https://codereview.chromium.org/490913002/diff/140001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.cpp:147:
m_pendingHighPriorityTasks.append(TracedTask(task, location, traceName,
generateFlowTraceID()));
Unfortunate we always generate these trace ids.  atomicIncrement is non-trivial
cost no?

https://codereview.chromium.org/490913002/diff/140001/Source/platform/schedul...
File Source/platform/scheduler/Scheduler.h (right):

https://codereview.chromium.org/490913002/diff/140001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.h:78: int generateFlowTraceID();
No type?

[picksi1, 2014-09-03 16:32:57]

AtomicIncrement is now only performed if tracing is enabled.

I've also moved the ID generation down into TracedTask which (1) keeps it nicely
encapsulated and (2) means IDs will remain unique even if multiple instances of
the scheduler are creating tasks.

https://codereview.chromium.org/490913002/diff/140001/Source/platform/schedul...
File Source/platform/scheduler/Scheduler.cpp (right):

https://codereview.chromium.org/490913002/diff/140001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.cpp:147:
m_pendingHighPriorityTasks.append(TracedTask(task, location, traceName,
generateFlowTraceID()));
I've moved ID generation into TracedTasks and only do the atomicIncrement if
tracing is enabled.

https://codereview.chromium.org/490913002/diff/140001/Source/platform/schedul...
File Source/platform/scheduler/Scheduler.h (right):

https://codereview.chromium.org/490913002/diff/140001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.h:78: int generateFlowTraceID();
I don't know what you are asking for here? Are you suggesting different IDs for
different types of events?

[eseidel, 2014-09-04 06:49:13]

lgtm

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
File Source/platform/scheduler/TracedTask.cpp (right):

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
Source/platform/scheduler/TracedTask.cpp:29:
TRACE_EVENT_CATEGORY_GROUP_ENABLED("blink", &tracingEnabled);
Is there also a check for flow events?  My understnding was that flow events
were a separate checkbox?

[Sami, 2014-09-04 09:51:07]

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
File Source/platform/scheduler/Scheduler.h (right):

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.h:94: // Declared volatile as they are
atomically incremented.
nit: s/they are/this is/.

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
File Source/platform/scheduler/TracedTask.cpp (right):

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
Source/platform/scheduler/TracedTask.cpp:29:
TRACE_EVENT_CATEGORY_GROUP_ENABLED("blink", &tracingEnabled);
On 2014/09/04 06:49:13, eseidel wrote:
> Is there also a check for flow events?  My understnding was that flow events
> were a separate checkbox?

Only whole categories can be enabled or disabled. There isn't a way to select
individual event types within a category.

If this ends up being too noisy, we could move everything here under
"blink.scheduler".

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
File Source/platform/scheduler/TracedTask.h (right):

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
Source/platform/scheduler/TracedTask.h:24: static int s_nextFlowTraceID;
Should be volatile.

[eseidel, 2014-09-04 16:23:53]

I see.  I was thinking of toplevel.flow

On Thu, Sep 4, 2014 at 2:51 AM,  <skyostil@chromium.org> wrote:
>
>
https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
> File Source/platform/scheduler/Scheduler.h (right):
>
>
https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
> Source/platform/scheduler/Scheduler.h:94: // Declared volatile as they
> are atomically incremented.
> nit: s/they are/this is/.
>
>
>
https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
> File Source/platform/scheduler/TracedTask.cpp (right):
>
>
https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
> Source/platform/scheduler/TracedTask.cpp:29:
> TRACE_EVENT_CATEGORY_GROUP_ENABLED("blink", &tracingEnabled);
> On 2014/09/04 06:49:13, eseidel wrote:
>>
>> Is there also a check for flow events?  My understnding was that flow
>
> events
>>
>> were a separate checkbox?
>
>
> Only whole categories can be enabled or disabled. There isn't a way to
> select individual event types within a category.
>
> If this ends up being too noisy, we could move everything here under
> "blink.scheduler".
>
>
https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
> File Source/platform/scheduler/TracedTask.h (right):
>
>
https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
> Source/platform/scheduler/TracedTask.h:24: static int s_nextFlowTraceID;
> Should be volatile.
>
> https://codereview.chromium.org/490913002/

To unsubscribe from this group and stop receiving emails from it, send an email
to blink-reviews+unsubscribe@chromium.org.

[picksi1, 2014-09-05 08:28:33]

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
File Source/platform/scheduler/Scheduler.h (right):

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
Source/platform/scheduler/Scheduler.h:94: // Declared volatile as they are
atomically incremented.
On 2014/09/04 09:51:07, Sami wrote:
> nit: s/they are/this is/.

Done.

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
File Source/platform/scheduler/TracedTask.h (right):

https://codereview.chromium.org/490913002/diff/160001/Source/platform/schedul...
Source/platform/scheduler/TracedTask.h:24: static int s_nextFlowTraceID;
On 2014/09/04 09:51:07, Sami wrote:
> Should be volatile.

Done.

[Sami, 2014-09-05 10:00:06]

Thanks Simon, lgtm!

[picksi1, 2014-09-05 12:27:53]

The CQ bit was checked by picksi@chromium.org

[commit-bot: I haz the power, 2014-09-05 12:28:36]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/picksi@chromium.org/490913002/180001

[commit-bot: I haz the power, 2014-09-05 12:42:36]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2014-09-05 12:42:37]

Try jobs failed on following builders:
  linux_gpu on tryserver.chromium.gpu
(http://build.chromium.org/p/tryserver.chromium.gpu/builders/linux_gpu/builds/...)

[picksi1, 2014-09-05 13:41:00]

The CQ bit was checked by picksi@chromium.org

[commit-bot: I haz the power, 2014-09-05 13:41:36]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/picksi@chromium.org/490913002/200001

[commit-bot: I haz the power, 2014-09-05 14:01:35]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2014-09-05 14:01:36]

Try jobs failed on following builders:
  win_gpu on tryserver.chromium.gpu
(http://build.chromium.org/p/tryserver.chromium.gpu/builders/win_gpu/builds/56999)

[picksi1, 2014-09-05 16:31:18]

The CQ bit was checked by picksi@chromium.org

[commit-bot: I haz the power, 2014-09-05 16:32:03]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/picksi@chromium.org/490913002/220001

[commit-bot: I haz the power, 2014-09-05 16:52:57]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2014-09-05 16:52:58]

Try jobs failed on following builders:
  win_gpu on tryserver.chromium.gpu
(http://build.chromium.org/p/tryserver.chromium.gpu/builders/win_gpu/builds/57061)

[picksi1, 2014-09-12 13:45:44]

The CQ bit was checked by picksi@chromium.org

[commit-bot: I haz the power, 2014-09-12 13:46:05]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patchset/490913002/300001

[commit-bot: I haz the power, 2014-09-12 14:53:54]

Committed patchset #16 (id:300001) as 181908