- Create isolate specific resuable handles and use them in the hot lookup paths.

- Create isolate specific resuable handles and use them in the hot lookup paths.
- Create a ResuableHandleScope class which ensures that we do not end up
  recursively reusing handles leading to corruption.

This change shows Dart2JSCompileAll runtime change from 739931 to about
645035 on my local machine.

R=iposva@google.com

Committed: https://code.google.com/p/dart/source/detail?r=24373

[siva, 2013-06-14 23:42:38]

synched up to top of tree

[Ivan Posva, 2013-06-16 11:15:26]

We should discuss this change and ways to make it safe in person.

-Ivan

https://codereview.chromium.org/16174008/diff/15001/runtime/vm/handles.h
File runtime/vm/handles.h (right):

https://codereview.chromium.org/16174008/diff/15001/runtime/vm/handles.h#newc...
runtime/vm/handles.h:380: // Macro to start a no handles scope in the code.
Comment does not match code.

[siva, 2013-06-19 20:41:55]

As discussed offline I moved all the isolate specific handle accessors to the
ReusableHandleScope object. This ensures that access to these handles is
possible only if ReusableHandleScope is used. This ensures guarding against
recursive use of these handles.

https://codereview.chromium.org/16174008/diff/15001/runtime/vm/handles.h
File runtime/vm/handles.h (right):

https://codereview.chromium.org/16174008/diff/15001/runtime/vm/handles.h#newc...
runtime/vm/handles.h:380: // Macro to start a no handles scope in the code.
On 2013/06/16 11:15:26, Ivan Posva wrote:
> Comment does not match code.

Done.

[Ivan Posva, 2013-06-22 01:01:36]

LGTM -ip

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/isolate.h
File runtime/vm/isolate.h (right):

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/isolate.h#newc...
runtime/vm/isolate.h:672: VMHandles handles_;
reusable_handles_?

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/isolate.h#newc...
runtime/vm/isolate.h:724: #endif  // defined(DEBUG)
To not keep objects alive longer than needed we should probably set all of the
handles to pointing to null.

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/object.h
File runtime/vm/object.h (right):

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/object.h#newco...
runtime/vm/object.h:5816: inline void Object::SetRaw(RawObject* value) {
DART_FORCE_INLINE would be a good alternative here.

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/object.h#newco...
runtime/vm/object.h:5823: vtable_ptr = Smi::handle_vtable_;
--verify_handles will fail below when being passed a Smi handle.

[siva, 2013-06-24 22:41:07]

Committed patchset #4 manually as r24373 (presubmit successful).

[siva, 2013-07-18 20:39:23]

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/isolate.h
File runtime/vm/isolate.h (right):

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/isolate.h#newc...
runtime/vm/isolate.h:672: VMHandles handles_;
On 2013/06/22 01:01:36, Ivan Posva wrote:
> reusable_handles_?

Done.

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/isolate.h#newc...
runtime/vm/isolate.h:724: #endif  // defined(DEBUG)
On 2013/06/22 01:01:36, Ivan Posva wrote:
> To not keep objects alive longer than needed we should probably set all of the
> handles to pointing to null.

Done.

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/object.h
File runtime/vm/object.h (right):

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/object.h#newco...
runtime/vm/object.h:5816: inline void Object::SetRaw(RawObject* value) {
On 2013/06/22 01:01:36, Ivan Posva wrote:
> DART_FORCE_INLINE would be a good alternative here.

Done.

https://codereview.chromium.org/16174008/diff/25001/runtime/vm/object.h#newco...
runtime/vm/object.h:5823: vtable_ptr = Smi::handle_vtable_;
Reverted this change as DART_FORCE_INLINE should make sure this is inlined.

On 2013/06/22 01:01:36, Ivan Posva wrote:
> --verify_handles will fail below when being passed a Smi handle.