Add a stub_code_space in the heap alongside code_space so that stub code generation happens here an…

Add a stub_code_space in the heap alongside code_space so that stub code generation happens here and it is easy to determine if a PC is a stub code or not.

This gets rid of the temporary change made in code_generator.cc for fast access to the top dart frame. In addition it will not pollute the pc ==> code cache that I plan to add next as we don't have to store stub pcs in that table.
Committed: https://code.google.com/p/dart/source/detail?r=7033

[srdjan, 2012-04-25 23:05:34]

LGTM if the performance has not changed

[Ivan Posva, 2012-04-26 15:09:01]

I think this is fine for now, but longer-term we need to find a better solution
than burning yet more pages to quickly identifying stub frames from regular dart
frames.

We could for example add a special pattern after the call to mark stub frames or
always have a marker to the point to the beginning of the instructions object.

-Ivan

https://chromiumcodereview.appspot.com/10223015/diff/13/vm/heap.cc
File vm/heap.cc (right):

https://chromiumcodereview.appspot.com/10223015/diff/13/vm/heap.cc#newcode37
vm/heap.cc:37: "e.g: --code_heap_size=256 allocates a 256KB stub code heap");
Mismatch in flag comment.

[siva, 2012-04-26 16:51:42]

I agree that we should strongly consider emitting the instruction object after a
'call' instruction that is the most efficient form of identifying the frame.

Once the new codegen settles, I will do this experiment to see what the size
impact would be on adding this.

https://chromiumcodereview.appspot.com/10223015/diff/13/vm/heap.cc
File vm/heap.cc (right):

https://chromiumcodereview.appspot.com/10223015/diff/13/vm/heap.cc#newcode37
vm/heap.cc:37: "e.g: --code_heap_size=256 allocates a 256KB stub code heap");
On 2012/04/26 15:09:01, Ivan Posva wrote:
> Mismatch in flag comment.

Done.