Ensure that the worker pool is running if we intend to use it

Ensure that the worker pool is running if we intend to use it

BUG=none
R=hpayer@chromium.org
LOG=n

Committed: https://code.google.com/p/v8/source/detail?r=19519

[tfarina, 2014-02-18 16:49:02]

https://codereview.chromium.org/170483002/diff/1/src/isolate.cc
File src/isolate.cc (right):

https://codereview.chromium.org/170483002/diff/1/src/isolate.cc#newcode42
src/isolate.cc:42: #ifdef V8_USE_DEFAULT_PLATFORM
separate this out, like we generally do in chromium? say, move to line 62?

[Hannes Payer (out of office), 2014-02-18 17:21:09]

What about changing the test expectations? I think we already did that when we
added sweeper threads.

[jochen (gone - plz use gerrit), 2014-02-18 21:39:01]

https://codereview.chromium.org/170483002/diff/1/src/isolate.cc
File src/isolate.cc (right):

https://codereview.chromium.org/170483002/diff/1/src/isolate.cc#newcode42
src/isolate.cc:42: #ifdef V8_USE_DEFAULT_PLATFORM
On 2014/02/18 16:49:02, tfarina wrote:
> separate this out, like we generally do in chromium? say, move to line 62?

in v8 the headers are sorted regardless of defines

[jochen (gone - plz use gerrit), 2014-02-18 21:39:46]

On 2014/02/18 17:21:09, Hannes Payer wrote:
> What about changing the test expectations? I think we already did that when we
> added sweeper threads.

starting the thread adds about 40MB, the current expectation is something like
4MB. If we increase the expectation to ~50MB, the test becomes kinda pointless

[Hannes Payer (out of office), 2014-02-19 21:12:44]

On 2014/02/18 21:39:46, jochen wrote:
> On 2014/02/18 17:21:09, Hannes Payer wrote:
> > What about changing the test expectations? I think we already did that when
we
> > added sweeper threads.
> 
> starting the thread adds about 40MB, the current expectation is something like
> 4MB. If we increase the expectation to ~50MB, the test becomes kinda pointless

Well, the platform threads are not really part of the isolate, they are part of
the platform so I understand that we do not want to account for the memory
overhead in the test.
40MB is a lot. What is our story for non Chrome embedders actually? Are we going
to provide a lightweight platform thread library? How does starting of the
platform threads effect the runtime of our tests?

[jochen (gone - plz use gerrit), 2014-02-20 08:40:47]

On 2014/02/19 21:12:44, Hannes Payer wrote:
> On 2014/02/18 21:39:46, jochen wrote:
> > On 2014/02/18 17:21:09, Hannes Payer wrote:
> > > What about changing the test expectations? I think we already did that
when
> we
> > > added sweeper threads.
> > 
> > starting the thread adds about 40MB, the current expectation is something
like
> > 4MB. If we increase the expectation to ~50MB, the test becomes kinda
pointless
> 
> Well, the platform threads are not really part of the isolate, they are part
of
> the platform so I understand that we do not want to account for the memory
> overhead in the test.
> 40MB is a lot. What is our story for non Chrome embedders actually? Are we
going
> to provide a lightweight platform thread library? How does starting of the
> platform threads effect the runtime of our tests?

10MB per thread is about as lightweight as threads get.

Note that this is a zero-sum change here anyway - if the platform threads are
spun up, we don't spin up the sweeper threads. They're exactly the same size.

[Hannes Payer (out of office), 2014-02-20 14:24:56]

Ah, one thread is 10MB. OK, makes sense. 

I guess we should change the test to account for that memory overhead. It would
be cleaner to start the threads in the platform and don't do the lazy starting
in the isolate. I would prefer that. WDYT?

[jochen (gone - plz use gerrit), 2014-02-20 14:50:23]

On 2014/02/20 14:24:56, Hannes Payer wrote:
> Ah, one thread is 10MB. OK, makes sense. 
> 
> I guess we should change the test to account for that memory overhead. It
would
> be cleaner to start the threads in the platform and don't do the lazy starting
> in the isolate. I would prefer that. WDYT?

hum, so actually, as long as it's behind a flag (which we want to avoid spinning
up both sweeper and platform threads), it doesn't really matter for the test.

PTAL

[Hannes Payer (out of office), 2014-02-20 18:07:37]

LGTM

[jochen (gone - plz use gerrit), 2014-02-20 19:32:35]

Committed patchset #2 manually as r19519 (presubmit successful).