ARM: Reduce the stack requirements of GetNoCodeAgeSequence.

ARM: Reduce the stack requirements of GetNoCodeAgeSequence.

Allocate the patcher object on the heap, to avoid occasional stack
overflows on QNX/ARM when entering GetNoCodeAgeSequence.

BUG=v8:3111
LOG=y

Patch from Cosmin Truta <ctruta@blackberry.com>;.

R=svenpanne@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=18815

[c.truta, 2014-01-22 11:50:12]

Hi, may I please a review for bugfix 3111?

Although I replaced a fast automatic stack allocation with a heap allocation, I
considered it to be ok because it's only a one-time thing. Having a much smaller
stack growth might actually be a benefit in all subsequent calls, due to
potentially lesser paging involved in stress conditions.

[Sven Panne, 2014-01-22 13:03:52]

Quick DBC...

https://codereview.chromium.org/144933002/diff/1/src/arm/codegen-arm.cc
File src/arm/codegen-arm.cc (right):

https://codereview.chromium.org/144933002/diff/1/src/arm/codegen-arm.cc#newco...
src/arm/codegen-arm.cc:862: CodePatcher* patcher =
Use "SmartPointer<CodePatcher> patcher ="...

https://codereview.chromium.org/144933002/diff/1/src/arm/codegen-arm.cc#newco...
src/arm/codegen-arm.cc:870: delete patcher;
... and remove this line.

[c.truta, 2014-01-22 14:54:59]

On 2014/01/22 13:03:52, Sven Panne wrote:
> src/arm/codegen-arm.cc:862: CodePatcher* patcher =
> Use "SmartPointer<CodePatcher> patcher ="...

Aha, indeed! Fixed. I thought SmartPointer is a shared pointer, but haven't
looked more closely. SmartPointer is like auto_ptr, and I guess when the
codebase will move to C++11, this will become like unique_ptr (or even be
replaced by unique_ptr), right?

I noticed the pattern SmartPointer<CompilationInfo> info(new ...), and I used
that one, for consistency.
Or if you preferred SmartPointer<CodePatcher> patcher = new ... instead, I will
go back and correct both my code and that old code; whichever you prefer.

[Sven Panne, 2014-01-22 15:02:30]

On 2014/01/22 14:54:59, cosmin.truta wrote:
> [...] (or even be replaced by unique_ptr), right?

That's at least my hope... ;-)

> I noticed the pattern SmartPointer<CompilationInfo> info(new ...), and I used
> that one, for consistency.
> Or if you preferred SmartPointer<CodePatcher> patcher = new ... instead, I
will
> go back and correct both my code and that old code; whichever you prefer.

Your version is actually nicer, although I don't have a strong preference
regarding this.

[Sven Panne, 2014-01-22 15:03:09]

LGTM, I'll land this tomorrow when nobody else did it in the meantime (hint,
hint, ...)

[Sven Panne, 2014-01-24 11:48:16]

Committed patchset #2 manually as r18815 (presubmit successful).