Fix deoptimization for out-of-line constant pool.

Fix deoptimization for out-of-line constant pool.

Ensure that the stack contains the correct constant pool pointer when a
function deopts.

This CL depends on https://codereview.chromium.org/183803022/ landing first.

R=ulan@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=19940

[rmcilroy, 2014-03-07 18:53:29]

Part of the out-of-line constant pool change - PTAL.

[ulan, 2014-03-11 16:14:19]

The first three files in patch set 2 didn't upload correctly.

[rmcilroy, 2014-03-11 16:17:43]

On 2014/03/11 16:14:19, ulan wrote:
> The first three files in patch set 2 didn't upload correctly.

Re-uploaded, should be fine now.  Sorry about that.

[ulan, 2014-03-13 13:35:22]

https://codereview.chromium.org/188063002/diff/40001/src/deoptimizer.cc
File src/deoptimizer.cc (right):

https://codereview.chromium.org/188063002/diff/40001/src/deoptimizer.cc#newco...
src/deoptimizer.cc:1136: // Our constant pool is the same as our caller's, get
it from the previous
Why ArgumentsAdaptorTrampoline and other stubs below have the same constant pool
as their callers?

[rmcilroy, 2014-03-14 12:35:57]

https://codereview.chromium.org/188063002/diff/40001/src/deoptimizer.cc
File src/deoptimizer.cc (right):

https://codereview.chromium.org/188063002/diff/40001/src/deoptimizer.cc#newco...
src/deoptimizer.cc:1136: // Our constant pool is the same as our caller's, get
it from the previous
On 2014/03/13 13:35:22, ulan wrote:
> Why ArgumentsAdaptorTrampoline and other stubs below have the same constant
pool
> as their callers?

You are right - originally stubs didn't load the constant pool pointer so it was
the case that their cp was the same as their callers, but now with the
FrameAndConstantPoolScope they can load their own constant pool.  I've made this
change (which also required updating LeaveArgumentsAdapterFrame to restore the
pp register).

[ulan, 2014-03-14 12:45:27]

On 2014/03/14 12:35:57, rmcilroy wrote:
> https://codereview.chromium.org/188063002/diff/40001/src/deoptimizer.cc
> File src/deoptimizer.cc (right):
> 
>
https://codereview.chromium.org/188063002/diff/40001/src/deoptimizer.cc#newco...
> src/deoptimizer.cc:1136: // Our constant pool is the same as our caller's, get
> it from the previous
> On 2014/03/13 13:35:22, ulan wrote:
> > Why ArgumentsAdaptorTrampoline and other stubs below have the same constant
> pool
> > as their callers?
> 
> You are right - originally stubs didn't load the constant pool pointer so it
was
> the case that their cp was the same as their callers, but now with the
> FrameAndConstantPoolScope they can load their own constant pool.  I've made
this
> change (which also required updating LeaveArgumentsAdapterFrame to restore the
> pp register).

Thanks for clarification! LGTM

[rmcilroy, 2014-03-14 15:12:10]

Committed patchset #4 manually as r19940 (presubmit successful).