Add deoptimization translations.

Add deoptimization translations.

BUG=
R=bmeurer@chromium.org, mstarzinger@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=22924

[Jarin, 2014-08-06 09:28:20]

Could you have a look? (You already LGTMed this in the Turbofan branch, but I
did not commit before the merge.)

[Benedikt Meurer, 2014-08-06 09:41:00]

https://codereview.chromium.org/442253002/diff/1/src/compiler/instruction-sel...
File src/compiler/instruction-selector.cc (right):

https://codereview.chromium.org/442253002/diff/1/src/compiler/instruction-sel...
src/compiler/instruction-selector.cc:1003: descriptor->size(), &inputs.front(),
0, NULL);
Nit: use inputs.size() instead of descriptor->size(). And maybe add a check that
they're equal.

https://codereview.chromium.org/442253002/diff/1/src/compiler/instruction.h
File src/compiler/instruction.h (right):

https://codereview.chromium.org/442253002/diff/1/src/compiler/instruction.h#n...
src/compiler/instruction.h:731: typedef std::vector<FrameStateDescriptor*,
Why store pointers? FrameStateDescriptor is small, copying is cheap, so using
pointers here just wastes memory. Or am I missing something?

[Jarin, 2014-08-06 10:05:05]

https://codereview.chromium.org/442253002/diff/1/src/compiler/instruction-sel...
File src/compiler/instruction-selector.cc (right):

https://codereview.chromium.org/442253002/diff/1/src/compiler/instruction-sel...
src/compiler/instruction-selector.cc:1003: descriptor->size(), &inputs.front(),
0, NULL);
On 2014/08/06 09:41:00, Benedikt Meurer wrote:
> Nit: use inputs.size() instead of descriptor->size(). And maybe add a check
that
> they're equal.

Done.

https://codereview.chromium.org/442253002/diff/1/src/compiler/instruction.h
File src/compiler/instruction.h (right):

https://codereview.chromium.org/442253002/diff/1/src/compiler/instruction.h#n...
src/compiler/instruction.h:731: typedef std::vector<FrameStateDescriptor*,
On 2014/08/06 09:41:00, Benedikt Meurer wrote:
> Why store pointers? FrameStateDescriptor is small, copying is cheap, so using
> pointers here just wastes memory. Or am I missing something?

Yeah, at the moment it is four ints. However, with inlining we will add a
pointer to the parent FrameStateDescriptor, so pointers will make the
representation more consistent. If you feel strongly about this, I am fine with
changing back to storing values.

[Benedikt Meurer, 2014-08-06 10:47:30]

lgtm

[Michael Starzinger, 2014-08-06 11:40:42]

LGTM.

[Jarin, 2014-08-06 11:49:17]

Committed patchset #3 manually as 22924 (presubmit successful).