Finish implementing lazy deoptimization (ia32, x64). Ran tests with --deoptimize-alot.

Finish implementing lazy deoptimization (ia32, x64). Ran tests with --deoptimize-alot.
Committed: https://code.google.com/p/dart/source/detail?r=12093

[siva, 2012-09-07 23:16:16]

lgtm

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/code_gen...
File runtime/vm/code_generator.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/code_gen...
runtime/vm/code_generator.cc:1460: // is not a performance issue).
You could avoid this multiple patch by first checking if the code is alive
right. If the code has already been marked as
not alive it means it has already been patched.

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/code_pat...
File runtime/vm/code_patcher_arm.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/code_pat...
runtime/vm/code_patcher_arm.cc:63: }
Needs
void CodePatcher::InsertCallAt(uword start, uword target) {
 UNIMPLEMENTED();
}

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/dart_api...
File runtime/vm/dart_api_impl.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/dart_api...
runtime/vm/dart_api_impl.cc:3785: // Deoptimize live frames.
Deoptimize all live frames?

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/object.cc
File runtime/vm/object.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/object.c...
runtime/vm/object.cc:6880: case PcDescriptors::kLazyDeoptJump: return "lazy     
   ";
why not make it "lazy-deopt"
would be more explicit than just "lazy"

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/stub_cod...
File runtime/vm/stub_code_ia32.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/stub_cod...
runtime/vm/stub_code_ia32.cc:557: __ LeaveFrame();
Could you add a comment here or above stating that the code
above should not have any GC in it and that runtime calls
kDeoptimizeCopyFrameRuntimeEntry and
kDeoptimizeFillFrameEntry
are leaf runtime calls

would make it easier when reading through the code.

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/stub_cod...
File runtime/vm/stub_code_x64.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/stub_cod...
runtime/vm/stub_code_x64.cc:549: __ LeaveFrame();
Ditto comment regarding GC:

Could you add a comment here or above stating that the code
above should not have any GC in it and that runtime calls
kDeoptimizeCopyFrameRuntimeEntry and
kDeoptimizeFillFrameEntry
are leaf runtime calls

would make it easier when reading through the code.

[srdjan, 2012-09-12 06:15:26]

Oops, forgot to send response.

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/code_gen...
File runtime/vm/code_generator.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/code_gen...
runtime/vm/code_generator.cc:1460: // is not a performance issue).
On 2012/09/07 23:16:17, siva wrote:
> You could avoid this multiple patch by first checking if the code is alive
> right. If the code has already been marked as
> not alive it means it has already been patched.

But is is not patched at all possible return points. Two frames from two
different calls from the same code may be alive. In that case I patch the same
code twice but at different addresses.

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/code_pat...
File runtime/vm/code_patcher_arm.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/code_pat...
runtime/vm/code_patcher_arm.cc:63: }
On 2012/09/07 23:16:17, siva wrote:
> Needs
> void CodePatcher::InsertCallAt(uword start, uword target) {
>  UNIMPLEMENTED();
> }

Done.

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/dart_api...
File runtime/vm/dart_api_impl.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/dart_api...
runtime/vm/dart_api_impl.cc:3785: // Deoptimize live frames.
On 2012/09/07 23:16:17, siva wrote:
> Deoptimize all live frames?

Done.

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/object.cc
File runtime/vm/object.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/object.c...
runtime/vm/object.cc:6880: case PcDescriptors::kLazyDeoptJump: return "lazy     
   ";
On 2012/09/07 23:16:17, siva wrote:
> why not make it "lazy-deopt"
> would be more explicit than just "lazy"

Done.

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/stub_cod...
File runtime/vm/stub_code_ia32.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/stub_cod...
runtime/vm/stub_code_ia32.cc:557: __ LeaveFrame();
On 2012/09/07 23:16:17, siva wrote:
> Could you add a comment here or above stating that the code
> above should not have any GC in it and that runtime calls
> kDeoptimizeCopyFrameRuntimeEntry and
> kDeoptimizeFillFrameEntry
> are leaf runtime calls
> 
> would make it easier when reading through the code.

Done.

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/stub_cod...
File runtime/vm/stub_code_x64.cc (right):

https://chromiumcodereview.appspot.com/10912146/diff/9002/runtime/vm/stub_cod...
runtime/vm/stub_code_x64.cc:549: __ LeaveFrame();
On 2012/09/07 23:16:17, siva wrote:
> Ditto comment regarding GC:
> 
> Could you add a comment here or above stating that the code
> above should not have any GC in it and that runtime calls
> kDeoptimizeCopyFrameRuntimeEntry and
> kDeoptimizeFillFrameEntry
> are leaf runtime calls
> 
> would make it easier when reading through the code.

Done.