Clean PrepareForScavenge

Clean PrepareForScavenge

This method requires pretty much cost(milliseconds per 1MB), but no
more needed. Until the last year, marking deque used from-space for
marking_deque_memory. So, cleaning from-space was needed. But now,
marking deque does not use from-space any more.

previous patch:
Use just one marking deque.
https://codereview.chromium.org/770453003

1. Two marking deque(incremental, mark-compact) merged into one.
2. The memory space is changed from from-space to new virtual
   memory.

BUG=

[sejunho, 2015-05-27 05:50:15]

sejunho@gmail.com changed reviewers:
+ hpayer@chromium.org

[sejunho, 2015-05-27 05:51:59]

Hi payer, please take a look :)

[sejunho, 2015-05-28 02:42:52]

Hello,
I worked on GC, and send my patch for review.
Take a look, please :)

[Hannes Payer (out of office), 2015-05-28 07:48:40]

no lgtm, the marking deque has (and had) nothing to do with the mark bits.

[Hannes Payer (out of office), 2015-05-28 07:50:38]

On 2015/05/28 07:48:40, Hannes Payer wrote:
> no lgtm, the marking deque has (and had) nothing to do with the mark bits.

not lgtm

[sejunho, 2015-05-28 08:37:30]

On 2015/05/28 07:50:38, Hannes Payer wrote:
> On 2015/05/28 07:48:40, Hannes Payer wrote:
> > no lgtm, the marking deque has (and had) nothing to do with the mark bits.
> 
> not lgtm

Thank you for taking the time to review. I will keep trying.

[Hannes Payer (out of office), 2015-05-28 09:38:33]

On 2015/05/28 08:37:30, sejunho wrote:
> On 2015/05/28 07:50:38, Hannes Payer wrote:
> > On 2015/05/28 07:48:40, Hannes Payer wrote:
> > > no lgtm, the marking deque has (and had) nothing to do with the mark bits.
> > 
> > not lgtm
> 
> Thank you for taking the time to review. I will keep trying.

But thanks for looking into this. After staring at the code I realized that we
clear the mark bits on two locations, which is not needed. We should fix that.

[sejunho, 2015-05-28 09:49:51]

On 2015/05/28 09:38:33, Hannes Payer wrote:
> On 2015/05/28 08:37:30, sejunho wrote:
> > On 2015/05/28 07:50:38, Hannes Payer wrote:
> > > On 2015/05/28 07:48:40, Hannes Payer wrote:
> > > > no lgtm, the marking deque has (and had) nothing to do with the mark
bits.
> > > 
> > > not lgtm
> > 
> > Thank you for taking the time to review. I will keep trying.
> 
> But thanks for looking into this. After staring at the code I realized that we
> clear the mark bits on two locations, which is not needed. We should fix that.


Is 'two locations' means PrepareForScavenge in incremental-marking and
ResetAllocationInfo
in NewSpace? If that right, I will work for fix that. I will modify description
and submit
another CL  :)