Flush code faster when doing emergency GCs.

Flush code faster when doing emergency GCs.

R=ulan@chromium.org
BUG=

[Erik Corry Chromium.org, 2014-10-24 11:00:59]

This is a revival of https://codereview.chromium.org/10933124/ from 2 years ago.

[Erik Corry Chromium.org, 2014-10-24 11:07:02]

erikcorry@chromium.org changed reviewers:
+ rmcilroy@chromium.org

[ulan, 2014-10-24 11:18:57]

Hi Erik,

If I understand correctly, this flushes all flushable code including active
code. This could lead to recompile/flush cycles and could actually increase
memory usage because of fragmentation in code space.

Is there open bug for this? (or a way to measure improvement)

[rmcilroy, 2014-10-24 11:46:07]

Nice, it's always nice when a 2 year old CL comes back :).  LGTM, but please
wait for Ulan since I don't know the code you are touching in heap-inl.h very
well.

https://codereview.chromium.org/677043002/diff/1/src/heap/mark-compact.cc
File src/heap/mark-compact.cc (right):

https://codereview.chromium.org/677043002/diff/1/src/heap/mark-compact.cc#new...
src/heap/mark-compact.cc:1356: Smi::FromInt(JSRegExp::kUninitializedValue));
Could you split this out so that it's shared with the code in lines 	 1384-1387.

https://codereview.chromium.org/677043002/diff/1/test/cctest/test-api.cc
File test/cctest/test-api.cc (right):

https://codereview.chromium.org/677043002/diff/1/test/cctest/test-api.cc#newc...
test/cctest/test-api.cc:18738: CcTest::heap()->CollectAllAvailableGarbage("Swift
blow to evil", 1);
:)

[rmcilroy, 2014-10-24 11:48:02]

On 2014/10/24 11:18:57, ulan wrote:
> Hi Erik,
> 
> If I understand correctly, this flushes all flushable code including active
> code. This could lead to recompile/flush cycles and could actually increase
> memory usage because of fragmentation in code space.

Would this not happen currently for an emergency GC?  Presumably the code-aging
would get triggered 7 times during the 7 GCs, which would end up making the code
old before the end of the 7 GCs?

[ulan, 2014-10-24 11:55:44]

On 2014/10/24 11:48:02, rmcilroy wrote:
> On 2014/10/24 11:18:57, ulan wrote:
> > Hi Erik,
> > 
> > If I understand correctly, this flushes all flushable code including active
> > code. This could lead to recompile/flush cycles and could actually increase
> > memory usage because of fragmentation in code space.
> 
> Would this not happen currently for an emergency GC?  Presumably the
code-aging
> would get triggered 7 times during the 7 GCs, which would end up making the
code
> old before the end of the 7 GCs?

7 is the upper bound, we can stop earlier if there are no more pending weak
handles.
I checked few websites (gmail, google plus, google maps), and there low memory
notification triggers 2-3 GCs.