Replace deopt stubs location/register shuffling with using deopt-info.

Replace deopt stubs location/register shuffling with using deopt-info. Added deopt instruction commands. Also fixes a crash in IA32 with use_ssa where the stack is incorrect after deoptimization. Tested IA32 with and without use-ssa.
Committed: https://code.google.com/p/dart/source/detail?r=10469

[siva, 2012-08-09 21:38:19]

lgtm

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/code_gen...
File runtime/vm/code_generator.cc (right):

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/code_gen...
runtime/vm/code_generator.cc:1420: intptr_t* registers_copy = new
intptr_t[kNumberOfCpuRegisters];
Why did you choose to make the register copy an allocated object instead of a
buffer in the isolate object?

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/code_gen...
runtime/vm/code_generator.cc:1427:
Isolate::Current()->set_deopt_registers_copy(registers_copy);
We should add an assert in set_deopt_registers_copy 
ASSERT(deopt_registers_copy_ == NULL);
to ensure that we don't go and stomp an existing value.

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/code_gen...
runtime/vm/code_generator.cc:1454:
Isolate::Current()->SetDeoptFrameCopy(frame_copy, frame_copy_size);
Similarly assert
ASSERT(deopt_frame_copy_ == NULL);
in SetDeoptFrameCopy(..);

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
File runtime/vm/deopt_instructions.cc (right):

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
runtime/vm/deopt_instructions.cc:132:
deopt_context->ObjectAt(object_table_index_));
There will be many ConstInstr instructions so it might be more efficient to
create the object handle in deopt_context once and use it here.

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
File runtime/vm/deopt_instructions.h (right):

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
runtime/vm/deopt_instructions.h:21: // 'frame_start' points to the return
address just below the frame's
'frame_start' -> 'to_frame_start'

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
runtime/vm/deopt_instructions.h:26: intptr_t num_args);
why explicit?

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
runtime/vm/deopt_instructions.h:34: ASSERT(0 <= index);
Is there a way to assert that index is valid on the higher side.

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/flow_gra...
File runtime/vm/flow_graph_compiler_x64.cc (right):

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/flow_gra...
runtime/vm/flow_graph_compiler_x64.cc:45: // TMP is never part od deoptimization
environment.
od -> of

[srdjan, 2012-08-09 23:28:30]

Thanks!

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/code_gen...
File runtime/vm/code_generator.cc (right):

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/code_gen...
runtime/vm/code_generator.cc:1420: intptr_t* registers_copy = new
intptr_t[kNumberOfCpuRegisters];
On 2012/08/09 21:38:19, asiva wrote:
> Why did you choose to make the register copy an allocated object instead of a
> buffer in the isolate object?

I did not want to always allocate one per Isolate. In near future we will also
have XMM registers in the buffer.

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/code_gen...
runtime/vm/code_generator.cc:1427:
Isolate::Current()->set_deopt_registers_copy(registers_copy);
On 2012/08/09 21:38:19, asiva wrote:
> We should add an assert in set_deopt_registers_copy 
> ASSERT(deopt_registers_copy_ == NULL);
> to ensure that we don't go and stomp an existing value.

And prevents memory leak ... good idea, done.

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/code_gen...
runtime/vm/code_generator.cc:1454:
Isolate::Current()->SetDeoptFrameCopy(frame_copy, frame_copy_size);
On 2012/08/09 21:38:19, asiva wrote:
> Similarly assert
> ASSERT(deopt_frame_copy_ == NULL);
> in SetDeoptFrameCopy(..);

Done.

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
File runtime/vm/deopt_instructions.cc (right):

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
runtime/vm/deopt_instructions.cc:132:
deopt_context->ObjectAt(object_table_index_));
On 2012/08/09 21:38:19, asiva wrote:
> There will be many ConstInstr instructions so it might be more efficient to
> create the object handle in deopt_context once and use it here.

Discussed off-line, caching isolate instead.

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
File runtime/vm/deopt_instructions.h (right):

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
runtime/vm/deopt_instructions.h:21: // 'frame_start' points to the return
address just below the frame's
On 2012/08/09 21:38:19, asiva wrote:
> 'frame_start' -> 'to_frame_start'

Done.

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
runtime/vm/deopt_instructions.h:26: intptr_t num_args);
On 2012/08/09 21:38:19, asiva wrote:
> why explicit?

Removed

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/deopt_in...
runtime/vm/deopt_instructions.h:34: ASSERT(0 <= index);
On 2012/08/09 21:38:19, asiva wrote:
> Is there a way to assert that index is valid on the higher side.

Passing additional to_frame_size and checking it here.

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/flow_gra...
File runtime/vm/flow_graph_compiler_x64.cc (right):

https://chromiumcodereview.appspot.com/10825236/diff/6007/runtime/vm/flow_gra...
runtime/vm/flow_graph_compiler_x64.cc:45: // TMP is never part od deoptimization
environment.
On 2012/08/09 21:38:19, asiva wrote:
> od -> of

Done.