Introduce --job-based-recompilation flag

Introduce --job-based-recompilation flag

The implementation is not yet complete, it doesn't support blocking yet,
and doesn't collect statistics.

This things will be fixed in follow-up CLs.

BUG=v8:3608
R=yangguo@chromium.org
LOG=n

Committed: https://code.google.com/p/v8/source/detail?r=24425

[Yang, 2014-10-07 07:00:41]

LGTM with comment.

Support for flushing the queue would be great too (we do it when stopping, and
after a ContextDisposedNotification).

https://codereview.chromium.org/620093003/diff/1/src/optimizing-compiler-thre...
File src/optimizing-compiler-thread.cc (right):

https://codereview.chromium.org/620093003/diff/1/src/optimizing-compiler-thre...
src/optimizing-compiler-thread.cc:80: DisallowHandleDereference no_deref;
Please copy those Disallow* scopes over to CompileTask::Run

[jochen (gone - plz use gerrit), 2014-10-07 07:28:46]

On 2014/10/07 at 07:00:41, yangguo wrote:
> LGTM with comment.
> 
> Support for flushing the queue would be great too (we do it when stopping, and
after a ContextDisposedNotification).
> 

ack

>
https://codereview.chromium.org/620093003/diff/1/src/optimizing-compiler-thre...
> File src/optimizing-compiler-thread.cc (right):
> 
>
https://codereview.chromium.org/620093003/diff/1/src/optimizing-compiler-thre...
> src/optimizing-compiler-thread.cc:80: DisallowHandleDereference no_deref;
> Please copy those Disallow* scopes over to CompileTask::Run

done

[jochen (gone - plz use gerrit), 2014-10-07 07:29:36]

Committed patchset #2 (id:20001) manually as 24425 (presubmit successful).

[Yang, 2014-10-07 07:35:38]

https://codereview.chromium.org/620093003/diff/20001/src/optimizing-compiler-...
File src/optimizing-compiler-thread.cc (right):

https://codereview.chromium.org/620093003/diff/20001/src/optimizing-compiler-...
src/optimizing-compiler-thread.cc:28: Isolate::SetIsolateThreadLocals(isolate_,
NULL);
I guess we are assuming that the thread this tasks is assigned to may be one we
never used before and there is no other bottleneck we can put this, so that we
don't unnecessarily init thread locals twice?

[jochen (gone - plz use gerrit), 2014-10-07 07:58:32]

https://codereview.chromium.org/620093003/diff/20001/src/optimizing-compiler-...
File src/optimizing-compiler-thread.cc (right):

https://codereview.chromium.org/620093003/diff/20001/src/optimizing-compiler-...
src/optimizing-compiler-thread.cc:28: Isolate::SetIsolateThreadLocals(isolate_,
NULL);
On 2014/10/07 07:35:37, Yang wrote:
> I guess we are assuming that the thread this tasks is assigned to may be one
we
> never used before and there is no other bottleneck we can put this, so that we
> don't unnecessarily init thread locals twice?

we can't know which thread this will run on.

Ideally, we wouldn't set them at all, nothing should call Isolate::Current(),
right?

[Yang, 2014-10-07 08:23:21]

On 2014/10/07 07:58:32, jochen wrote:
>
https://codereview.chromium.org/620093003/diff/20001/src/optimizing-compiler-...
> File src/optimizing-compiler-thread.cc (right):
> 
>
https://codereview.chromium.org/620093003/diff/20001/src/optimizing-compiler-...
> src/optimizing-compiler-thread.cc:28:
Isolate::SetIsolateThreadLocals(isolate_,
> NULL);
> On 2014/10/07 07:35:37, Yang wrote:
> > I guess we are assuming that the thread this tasks is assigned to may be one
> we
> > never used before and there is no other bottleneck we can put this, so that
we
> > don't unnecessarily init thread locals twice?
> 
> we can't know which thread this will run on.
> 
> Ideally, we wouldn't set them at all, nothing should call Isolate::Current(),
> right?

I guess not. But the PerThreadAssertionScopes like those Disallow* scopes still
rely on the TLS. All of them are (currently) debug only though, so we could skip
this in release.

[jochen (gone - plz use gerrit), 2014-10-07 08:30:49]

On 2014/10/07 at 08:23:21, yangguo wrote:
> On 2014/10/07 07:58:32, jochen wrote:
> >
https://codereview.chromium.org/620093003/diff/20001/src/optimizing-compiler-...
> > File src/optimizing-compiler-thread.cc (right):
> > 
> >
https://codereview.chromium.org/620093003/diff/20001/src/optimizing-compiler-...
> > src/optimizing-compiler-thread.cc:28:
Isolate::SetIsolateThreadLocals(isolate_,
> > NULL);
> > On 2014/10/07 07:35:37, Yang wrote:
> > > I guess we are assuming that the thread this tasks is assigned to may be
one
> > we
> > > never used before and there is no other bottleneck we can put this, so
that we
> > > don't unnecessarily init thread locals twice?
> > 
> > we can't know which thread this will run on.
> > 
> > Ideally, we wouldn't set them at all, nothing should call
Isolate::Current(),
> > right?
> 
> I guess not. But the PerThreadAssertionScopes like those Disallow* scopes
still rely on the TLS. All of them are (currently) debug only though, so we
could skip this in release.

the assert scopes don't use the isolate thread locals, no?

[Yang, 2014-10-07 08:48:25]

On 2014/10/07 08:30:49, jochen wrote:
> On 2014/10/07 at 08:23:21, yangguo wrote:
> > On 2014/10/07 07:58:32, jochen wrote:
> > >
>
https://codereview.chromium.org/620093003/diff/20001/src/optimizing-compiler-...
> > > File src/optimizing-compiler-thread.cc (right):
> > > 
> > >
>
https://codereview.chromium.org/620093003/diff/20001/src/optimizing-compiler-...
> > > src/optimizing-compiler-thread.cc:28:
> Isolate::SetIsolateThreadLocals(isolate_,
> > > NULL);
> > > On 2014/10/07 07:35:37, Yang wrote:
> > > > I guess we are assuming that the thread this tasks is assigned to may be
> one
> > > we
> > > > never used before and there is no other bottleneck we can put this, so
> that we
> > > > don't unnecessarily init thread locals twice?
> > > 
> > > we can't know which thread this will run on.
> > > 
> > > Ideally, we wouldn't set them at all, nothing should call
> Isolate::Current(),
> > > right?
> > 
> > I guess not. But the PerThreadAssertionScopes like those Disallow* scopes
> still rely on the TLS. All of them are (currently) debug only though, so we
> could skip this in release.
> 
> the assert scopes don't use the isolate thread locals, no?

That's true. I was confusing things. I guess we won't need this at all then?