Enable weak embedded maps in optimized code.

Enable weak embedded maps in optimized code.

If the top optimized code in call stack is at the point that does not support
deoptimization, then treat the maps in the code as strong pointers.

Note that other optimized code in call stack must support deoptimization
because of the call instruction with side-effects.

BUG=217858, v8:2073
R=mstarzinger@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=15452

[ulan, 2013-06-13 15:42:39]

PTAL

https://chromiumcodereview.appspot.com/16955008/diff/2001/test/mjsunit/regres...
File test/mjsunit/regress/regress-crbug-217858.js (left):

https://chromiumcodereview.appspot.com/16955008/diff/2001/test/mjsunit/regres...
test/mjsunit/regress/regress-crbug-217858.js:29: 
This regression test wasn't working after analyze_environment_liveness flags was
introduced.

[Michael Starzinger, 2013-07-02 14:20:54]

Looking good, just a few minor comments.

https://codereview.chromium.org/16955008/diff/2001/src/isolate.cc
File src/isolate.cc (right):

https://codereview.chromium.org/16955008/diff/2001/src/isolate.cc#newcode727
src/isolate.cc:727: if (it.frame()->type() == StackFrame::OPTIMIZED) {
Shouldn't this actually just look whether the top-most frame is an optimized one
instead of searching for the first optimized frame?

As discussed offline: I am also fine with breaking the search at the first
encounter of a non-optimized frame.

https://codereview.chromium.org/16955008/diff/2001/src/isolate.h
File src/isolate.h (right):

https://codereview.chromium.org/16955008/diff/2001/src/isolate.h#newcode744
src/isolate.h:744: StackFrame* TopOptimizedFrame();
Can we inline this functionality into mark-compact.cc directly to where it's
being used, I think nobody should be tempted to use this.

https://codereview.chromium.org/16955008/diff/2001/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16955008/diff/2001/src/mark-compact.cc#newcod...
src/mark-compact.cc:2060: // otherwise a map can die and deoptimize the code.
Can we move this into MarkCompactCollector::MarkLiveObjects? It is not actually
part of the root-set, it is just a requirement from the mark-compact collector
(e.g. similar to PrepareForCodeFlushing).

[ulan, 2013-07-02 14:45:41]

Addresses comments, please take another look.

https://codereview.chromium.org/16955008/diff/2001/src/isolate.cc
File src/isolate.cc (right):

https://codereview.chromium.org/16955008/diff/2001/src/isolate.cc#newcode727
src/isolate.cc:727: if (it.frame()->type() == StackFrame::OPTIMIZED) {
Added a check for non-optimized frame.

On 2013/07/02 14:20:54, Michael Starzinger wrote:
> Shouldn't this actually just look whether the top-most frame is an optimized
one
> instead of searching for the first optimized frame?
> 
> As discussed offline: I am also fine with breaking the search at the first
> encounter of a non-optimized frame.

https://codereview.chromium.org/16955008/diff/2001/src/isolate.h
File src/isolate.h (right):

https://codereview.chromium.org/16955008/diff/2001/src/isolate.h#newcode744
src/isolate.h:744: StackFrame* TopOptimizedFrame();
On 2013/07/02 14:20:54, Michael Starzinger wrote:
> Can we inline this functionality into mark-compact.cc directly to where it's
> being used, I think nobody should be tempted to use this.

Done.

https://codereview.chromium.org/16955008/diff/2001/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16955008/diff/2001/src/mark-compact.cc#newcod...
src/mark-compact.cc:2060: // otherwise a map can die and deoptimize the code.
On 2013/07/02 14:20:54, Michael Starzinger wrote:
> Can we move this into MarkCompactCollector::MarkLiveObjects? It is not
actually
> part of the root-set, it is just a requirement from the mark-compact collector
> (e.g. similar to PrepareForCodeFlushing).

Done.

[Michael Starzinger, 2013-07-02 15:03:43]

LGTM with nits.

https://codereview.chromium.org/16955008/diff/9001/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16955008/diff/9001/src/mark-compact.cc#newcod...
src/mark-compact.cc:2218: if (frame) {
nit: s/frame/frame != NULL/

https://codereview.chromium.org/16955008/diff/9001/src/mark-compact.cc#newcod...
src/mark-compact.cc:2224: ProcessMarkingDeque();
nit: Move the processing of the marking deque into the condition, nothing to
process if nothing was marked.

[ulan, 2013-07-02 15:06:36]

Thank you. I will land after the next roll.

https://codereview.chromium.org/16955008/diff/9001/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/16955008/diff/9001/src/mark-compact.cc#newcod...
src/mark-compact.cc:2218: if (frame) {
On 2013/07/02 15:03:44, Michael Starzinger wrote:
> nit: s/frame/frame != NULL/

Done.

https://codereview.chromium.org/16955008/diff/9001/src/mark-compact.cc#newcod...
src/mark-compact.cc:2224: ProcessMarkingDeque();
On 2013/07/02 15:03:44, Michael Starzinger wrote:
> nit: Move the processing of the marking deque into the condition, nothing to
> process if nothing was marked.

Done.

[ulan, 2013-07-02 15:16:11]

Committed patchset #3 manually as r15452 (presubmit successful).