Speed up the Dart_IsBlah functions by checking the class index. This

Speed up the Dart_IsBlah functions by checking the class index.  This
allows us to avoid creating handles (and therefore handle scopes) in
these functions.

Before:
UseDartApi : 66395

After:
UseDartApi : 42730

Main improvement comes from speeding up Dart_IsError.
Committed: https://code.google.com/p/dart/source/detail?r=7619

[Ivan Posva, 2012-05-09 23:28:16]

LGTM with comments.

-Ivan

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.cc
File vm/dart_api_impl.cc (right):

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.cc...
vm/dart_api_impl.cc:262: ASSERT(kError + 5 == kInstance);  // There are 4
subtypes of Error.
This is brittle in case we delete one of the error types and add a new non-error
type before kInstance. Should we just enumerate them here?

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.cc...
vm/dart_api_impl.cc:1083: ASSERT(kNumber + 6 == kString);  // There are 5
subtypes of Number.
ditto.

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.cc...
vm/dart_api_impl.cc:1093: ASSERT(kInteger + 4 == kDouble);  // There are 3
subtypes of Integer.
ditto.

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.cc...
vm/dart_api_impl.cc:1346: ASSERT(kString + 7 == kBool);  // There are 6 subtypes
of String.
ditto.

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.cc...
vm/dart_api_impl.cc:1644: ASSERT(kArray + 2 == kGrowableObjectArray);  // There
is 1 subtype of Array.
ditto.

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.cc...
vm/dart_api_impl.cc:1999: ASSERT(kByteArray + 21 == kClosure);  // There are 20
subtypes of ByteArray
ditto.

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.h
File vm/dart_api_impl.h (right):

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.h#...
vm/dart_api_impl.h:126: static intptr_t ClassIndex(Dart_Handle handle) {
There is an inconsistency naming this. In RawObject you call the value a
ClassTag, here it is a ClassIndex. My initial reaction was to use ClassIndex,
but on further reflection I do like distinguishing the difference of accessing
the tag of an Object (here) from getting the index of a Class.

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/raw_object.h
File vm/raw_object.h (right):

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/raw_object.h#new...
vm/raw_object.h:239: intptr_t GetClassTag() const {
Should we make this private?

[turnidge, 2012-05-10 17:49:38]

Ivan,

Please take another look.

Todd

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.cc
File vm/dart_api_impl.cc (right):

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.cc...
vm/dart_api_impl.cc:262: ASSERT(kError + 5 == kInstance);  // There are 4
subtypes of Error.
On 2012/05/09 23:28:16, Ivan Posva wrote:
> This is brittle in case we delete one of the error types and add a new
non-error
> type before kInstance. Should we just enumerate them here?

I have moved all of these predicates to raw_object.h.  I considered attaching
the functions to the Raw classes (like putting IsErrorClassIndex on RawError)
but it got odd in the case of builtin LIst types (they don't seem to have a
common supertype) so I left them as standalone functions.  I have expanded the
asserts to try to make them less brittle.

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.cc...
vm/dart_api_impl.cc:1288: ASSERT(kBool + 1 == kArray);  // No subtypes.
I dropped the asserts for types with no subtypes.  I felt that they were overly
paranoid -- let me know if you disagree.

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.h
File vm/dart_api_impl.h (right):

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/dart_api_impl.h#...
vm/dart_api_impl.h:126: static intptr_t ClassIndex(Dart_Handle handle) {
On 2012/05/09 23:28:16, Ivan Posva wrote:
> There is an inconsistency naming this. In RawObject you call the value a
> ClassTag, here it is a ClassIndex. My initial reaction was to use ClassIndex,
> but on further reflection I do like distinguishing the difference of accessing
> the tag of an Object (here) from getting the index of a Class.

I thought about this some more and decided to go with ClassIndex across the
board.  It just seems simpler.

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/raw_object.h
File vm/raw_object.h (right):

https://chromiumcodereview.appspot.com/10386060/diff/4001/vm/raw_object.h#new...
vm/raw_object.h:239: intptr_t GetClassTag() const {
On 2012/05/09 23:28:16, Ivan Posva wrote:
> Should we make this private?

Done.

[Ivan Posva, 2012-05-11 20:03:57]

LGTM -ip