Improve queuing for concurrent OSR.

Improve queuing for concurrent OSR.

Specifically, this includes:
- Encapsulating data structure for osr buffer into CyclicBuffer
- Use the new CyclicQueue instead of UnboundedQueue to queue new jobs.
  We can enqueue and dequeue a CyclicQueue on both ends in O(1).
  This allows us to add OSR jobs to the front for lower compile latency.
- Dispose osr buffer by one stale job per GC to avoid leak


R=titzer@chromium.org
BUG=

Committed: https://code.google.com/p/v8/source/detail?r=17244

[titzer, 2013-10-10 09:41:57]

https://codereview.chromium.org/25505002/diff/1/src/optimizing-compiler-thread.h
File src/optimizing-compiler-thread.h (right):

https://codereview.chromium.org/25505002/diff/1/src/optimizing-compiler-threa...
src/optimizing-compiler-thread.h:55: input_queue_semaphore_(0),
The external synchronization on these buffers is starting to be a drag. Now that
we are using mutexes again, can we please encapsulate the synchronization into
the buffers somehow?

https://codereview.chromium.org/25505002/diff/1/src/optimizing-compiler-threa...
src/optimizing-compiler-thread.h:79: inline void DrainOsrBuffer() {
I like the idea of dropping one OSR entry per GC, but this method could use a
better name. "Drain" to me implies that it empties the entire buffer.

https://codereview.chromium.org/25505002/diff/1/src/utils.h
File src/utils.h (right):

https://codereview.chromium.org/25505002/diff/1/src/utils.h#newcode1144
src/utils.h:1144: class CircularBuffer {
This class needs some unit tests.

https://codereview.chromium.org/25505002/diff/1/src/utils.h#newcode1199
src/utils.h:1199: class CircularQueue {
And this one too.

[Yang, 2013-10-15 10:03:13]

On 2013/10/10 09:41:57, titzer wrote:
>
https://codereview.chromium.org/25505002/diff/1/src/optimizing-compiler-thread.h
> File src/optimizing-compiler-thread.h (right):
> 
>
https://codereview.chromium.org/25505002/diff/1/src/optimizing-compiler-threa...
> src/optimizing-compiler-thread.h:55: input_queue_semaphore_(0),
> The external synchronization on these buffers is starting to be a drag. Now
that
> we are using mutexes again, can we please encapsulate the synchronization into
> the buffers somehow?
> 
>
https://codereview.chromium.org/25505002/diff/1/src/optimizing-compiler-threa...
> src/optimizing-compiler-thread.h:79: inline void DrainOsrBuffer() {
> I like the idea of dropping one OSR entry per GC, but this method could use a
> better name. "Drain" to me implies that it empties the entire buffer.
> 
> https://codereview.chromium.org/25505002/diff/1/src/utils.h
> File src/utils.h (right):
> 
> https://codereview.chromium.org/25505002/diff/1/src/utils.h#newcode1144
> src/utils.h:1144: class CircularBuffer {
> This class needs some unit tests.
> 
> https://codereview.chromium.org/25505002/diff/1/src/utils.h#newcode1199
> src/utils.h:1199: class CircularQueue {
> And this one too.

Addressed those comments.

Please give it a second look.

[Yang, 2013-10-16 13:17:23]

On 2013/10/15 10:03:13, Yang wrote:
> On 2013/10/10 09:41:57, titzer wrote:
> >
>
https://codereview.chromium.org/25505002/diff/1/src/optimizing-compiler-thread.h
> > File src/optimizing-compiler-thread.h (right):
> > 
> >
>
https://codereview.chromium.org/25505002/diff/1/src/optimizing-compiler-threa...
> > src/optimizing-compiler-thread.h:55: input_queue_semaphore_(0),
> > The external synchronization on these buffers is starting to be a drag. Now
> that
> > we are using mutexes again, can we please encapsulate the synchronization
into
> > the buffers somehow?
> > 
> >
>
https://codereview.chromium.org/25505002/diff/1/src/optimizing-compiler-threa...
> > src/optimizing-compiler-thread.h:79: inline void DrainOsrBuffer() {
> > I like the idea of dropping one OSR entry per GC, but this method could use
a
> > better name. "Drain" to me implies that it empties the entire buffer.
> > 
> > https://codereview.chromium.org/25505002/diff/1/src/utils.h
> > File src/utils.h (right):
> > 
> > https://codereview.chromium.org/25505002/diff/1/src/utils.h#newcode1144
> > src/utils.h:1144: class CircularBuffer {
> > This class needs some unit tests.
> > 
> > https://codereview.chromium.org/25505002/diff/1/src/utils.h#newcode1199
> > src/utils.h:1199: class CircularQueue {
> > And this one too.
> 
> Addressed those comments.
> 
> Please give it a second look.

As discussed offline, I removed the data structures and inlined them into the
OptimizingCompilerThread class.

[titzer, 2013-10-16 14:02:10]

lgtm

https://codereview.chromium.org/25505002/diff/11001/src/optimizing-compiler-t...
File src/optimizing-compiler-thread.h (right):

https://codereview.chromium.org/25505002/diff/11001/src/optimizing-compiler-t...
src/optimizing-compiler-thread.h:120: int InputQueueIndex(int i) {
inline annotation?

[Yang, 2013-10-16 14:47:29]

Committed patchset #4 manually as r17244 (presubmit successful).