Add support for leaf runtime calls

Add support for leaf runtime calls
- use a NoGCScope around these calls
- do not save and restore exit frame info, context register etc.
- Do not create a zone and handlescope by default
Committed: https://code.google.com/p/dart/source/detail?r=8699

[srdjan, 2012-06-12 20:10:41]

DBC

http://codereview.chromium.org/10542123/diff/6002/vm/runtime_entry.h
File vm/runtime_entry.h (right):

http://codereview.chromium.org/10542123/diff/6002/vm/runtime_entry.h#newcode45
vm/runtime_entry.h:45: int argument_count_;
const

http://codereview.chromium.org/10542123/diff/6002/vm/runtime_entry.h#newcode46
vm/runtime_entry.h:46: bool is_leaf_;
const

http://codereview.chromium.org/10542123/diff/6002/vm/stub_code_ia32.cc
File vm/stub_code_ia32.cc (right):

http://codereview.chromium.org/10542123/diff/6002/vm/stub_code_ia32.cc#newcod...
vm/stub_code_ia32.cc:104: __ EnterFrame(0);
Why do you need the frame enter and leave? No stack traversal can occur here

[siva, 2012-06-12 21:18:04]

http://codereview.chromium.org/10542123/diff/6002/vm/runtime_entry.h
File vm/runtime_entry.h (right):

http://codereview.chromium.org/10542123/diff/6002/vm/runtime_entry.h#newcode45
vm/runtime_entry.h:45: int argument_count_;
On 2012/06/12 20:10:41, srdjan wrote:
> const

Done.

http://codereview.chromium.org/10542123/diff/6002/vm/runtime_entry.h#newcode46
vm/runtime_entry.h:46: bool is_leaf_;
On 2012/06/12 20:10:41, srdjan wrote:
> const

Done.

http://codereview.chromium.org/10542123/diff/6002/vm/stub_code_ia32.cc
File vm/stub_code_ia32.cc (right):

http://codereview.chromium.org/10542123/diff/6002/vm/stub_code_ia32.cc#newcod...
vm/stub_code_ia32.cc:104: __ EnterFrame(0);
Yes, I would love to get rid of this too, but the stack alignment done below
leaves us with a not so easy way of restoring the stack pointer after the call,
hence the EnterFrame.



On 2012/06/12 20:10:41, srdjan wrote:
> Why do you need the frame enter and leave? No stack traversal can occur here

[Ivan Posva, 2012-06-14 04:09:35]

LGTM with comment.

-Ivan

https://chromiumcodereview.appspot.com/10542123/diff/11012/vm/runtime_entry_t...
File vm/runtime_entry_test.cc (right):

https://chromiumcodereview.appspot.com/10542123/diff/11012/vm/runtime_entry_t...
vm/runtime_entry_test.cc:54: return Smi::New(result);
I would avoid any calls to handle based methods here, even if they are known not
to involve handles or allocation.

How about:
intptr_t result = reinterpret_cast<intptr_t>(args.At(0)) +
reinterpret_cast<intptr_t>(args.At(1));
return result;

[siva, 2012-06-15 00:09:41]

https://chromiumcodereview.appspot.com/10542123/diff/11012/vm/runtime_entry_t...
File vm/runtime_entry_test.cc (right):

https://chromiumcodereview.appspot.com/10542123/diff/11012/vm/runtime_entry_t...
vm/runtime_entry_test.cc:54: return Smi::New(result);
On 2012/06/14 04:09:35, Ivan Posva wrote:
> I would avoid any calls to handle based methods here, even if they are known
not
> to involve handles or allocation.
> 
> How about:
> intptr_t result = reinterpret_cast<intptr_t>(args.At(0)) +
> reinterpret_cast<intptr_t>(args.At(1));
> return result;

Done.