Add OptimizedCodeList and DeoptimizedCodeList to native contexts. Both lists are weak. This makes i…

Add OptimizedCodeList and DeoptimizedCodeList to native contexts. Both lists are weak. This makes it possible to find optimized code that is not referred to by any function, but still needs to be deoptimized. It obsoletes the weak deoptimizing code list in the deoptimizer data and generally simplifies the process of deoptimizing code.

BUG=
R=mstarzinger@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=16530

[titzer, 2013-08-30 17:50:00]

Note that this CL solves the problem that the optimized code table (and
OptimizedCodeEntry) was supposed to solve. I will remove that OptimizedCodeEntry
in a future CL (and it was only a few days old!)

[Michael Starzinger, 2013-09-03 21:52:31]

Looking good. The only two issues are handlification in deoptimizer.cc and a
different choke-point in the compiler.cc file.

https://codereview.chromium.org/23444029/diff/1/src/compiler.cc
File src/compiler.cc (right):

https://codereview.chromium.org/23444029/diff/1/src/compiler.cc#newcode883
src/compiler.cc:883: info->context()->native_context()->AddOptimizedCode(*code);
This is a bad choking point to add code to the weak list for two reasons:
1) The name of this function implies that it only deals with the optimized code
map, and nothing else.
2) One of the two call-sites (i.e. Compiler::InstallOptimizedCode) actually only
calls it if there is not yet another optimized code object for the same native
context in the optimize code map.

Can we hoist this out to the two call-sites and leave this method untouched?

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc
File src/deoptimizer.cc (right):

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc#newcode419
src/deoptimizer.cc:419: DeoptimizeMarkedCodeForContext(native_context);
Since DeoptimizeMarkedCodeForContext() might cause a GC, all raw pointer here
(i.e. native_context) might actually be stale. Better handlify this function and
the "context" argument to DeoptimizeMarkedCodeForContext() as well.

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc#newcode434
src/deoptimizer.cc:434: DeoptimizeMarkedCodeForContext(native_context);
Likewise.

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc#newcode450
src/deoptimizer.cc:450: DeoptimizeMarkedCodeForContext(native_context);
Likewise.

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc#newcode477
src/deoptimizer.cc:477:
DeoptimizeMarkedCodeForContext(function->context()->native_context());
Likewise.

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc#newcode702
src/deoptimizer.cc:702: // count all entries in the deoptimizing code list of
every context.
nit: Capitalize sentence.

https://codereview.chromium.org/23444029/diff/1/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/23444029/diff/1/src/mark-compact.cc#newcode3454
src/mark-compact.cc:3454: // Update the heads of the native contexts list the
code to deoptimize list.
nit: Comment is outdated.

https://codereview.chromium.org/23444029/diff/1/src/objects.h
File src/objects.h (right):

https://codereview.chromium.org/23444029/diff/1/src/objects.h#newcode7131
src/objects.h:7131: // uses this link to chain together flushing candidates.
Treatly weakly
nit: s/Treatly/Treated/

[Michael Starzinger, 2013-09-03 22:03:47]

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc
File src/deoptimizer.cc (right):

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc#newcode398
src/deoptimizer.cc:398: // (This may actually cause a GC due to a stub being
generated.)
Is this actually true? Can this cause a GC, I cannot seem to find the path where
this happens. If it cannot happen, then my comments about handlification are
obsolete and we also could use a ZoneList<Code*> instead. If it can happen, then
there are many more stale pointers here.

[titzer, 2013-09-04 11:19:04]

https://codereview.chromium.org/23444029/diff/1/src/compiler.cc
File src/compiler.cc (right):

https://codereview.chromium.org/23444029/diff/1/src/compiler.cc#newcode883
src/compiler.cc:883: info->context()->native_context()->AddOptimizedCode(*code);
On 2013/09/03 21:52:31, Michael Starzinger wrote:
> This is a bad choking point to add code to the weak list for two reasons:
> 1) The name of this function implies that it only deals with the optimized
code
> map, and nothing else.
> 2) One of the two call-sites (i.e. Compiler::InstallOptimizedCode) actually
only
> calls it if there is not yet another optimized code object for the same native
> context in the optimize code map.
> 
> Can we hoist this out to the two call-sites and leave this method untouched?

I've moved it out to the point that the optimized code is transferred into the
CompilationInfo.

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc
File src/deoptimizer.cc (right):

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc#newcode398
src/deoptimizer.cc:398: // (This may actually cause a GC due to a stub being
generated.)
On 2013/09/03 22:03:47, Michael Starzinger wrote:
> Is this actually true? Can this cause a GC, I cannot seem to find the path
where
> this happens. If it cannot happen, then my comments about handlification are
> obsolete and we also could use a ZoneList<Code*> instead. If it can happen,
then
> there are many more stale pointers here.

Right...fixed. No GC tolerated.

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc#newcode419
src/deoptimizer.cc:419: DeoptimizeMarkedCodeForContext(native_context);
On 2013/09/03 21:52:31, Michael Starzinger wrote:
> Since DeoptimizeMarkedCodeForContext() might cause a GC, all raw pointer here
> (i.e. native_context) might actually be stale. Better handlify this function
and
> the "context" argument to DeoptimizeMarkedCodeForContext() as well.

So; you were right, we don't tolerate GC here--thankfully--so this code doesn't
need to be handlified.

https://codereview.chromium.org/23444029/diff/1/src/deoptimizer.cc#newcode702
src/deoptimizer.cc:702: // count all entries in the deoptimizing code list of
every context.
On 2013/09/03 21:52:31, Michael Starzinger wrote:
> nit: Capitalize sentence.

Done.

https://codereview.chromium.org/23444029/diff/1/src/mark-compact.cc
File src/mark-compact.cc (right):

https://codereview.chromium.org/23444029/diff/1/src/mark-compact.cc#newcode3454
src/mark-compact.cc:3454: // Update the heads of the native contexts list the
code to deoptimize list.
On 2013/09/03 21:52:31, Michael Starzinger wrote:
> nit: Comment is outdated.

Done.

https://codereview.chromium.org/23444029/diff/1/src/objects.h
File src/objects.h (right):

https://codereview.chromium.org/23444029/diff/1/src/objects.h#newcode7131
src/objects.h:7131: // uses this link to chain together flushing candidates.
Treatly weakly
On 2013/09/03 21:52:31, Michael Starzinger wrote:
> nit: s/Treatly/Treated/

This weakly treatly brought to you courtesy of a diseased, geistig behindert,
brain.

[Michael Starzinger, 2013-09-04 12:17:48]

LGTM (with final nit).

https://codereview.chromium.org/23444029/diff/23001/src/compiler.cc
File src/compiler.cc (right):

https://codereview.chromium.org/23444029/diff/23001/src/compiler.cc#newcode895
src/compiler.cc:895: if (FLAG_cache_optimized_code &&
nit: Can we keep this predicate in sync with the one in
InsertCodeIntoOptimizedCodeMap?

[titzer, 2013-09-04 13:53:40]

Committed patchset #5 manually as r16530 (presubmit successful).