Make deoptimization architecture dependent (it depends on the frame layout).

Make deoptimization architecture dependent (it depends on the frame layout).

R=srdjan@google.com, zra@google.com

Committed: https://code.google.com/p/dart/source/detail?r=22719

[srdjan, 2013-05-14 22:16:39]

LGTM

https://codereview.chromium.org/15110003/diff/1/runtime/vm/deopt_instructions.cc
File runtime/vm/deopt_instructions.cc (right):

https://codereview.chromium.org/15110003/diff/1/runtime/vm/deopt_instructions...
runtime/vm/deopt_instructions.cc:494: // Callee's PC marker is not used anymore.
Set to 0.
When can this occur?

https://codereview.chromium.org/15110003/diff/1/runtime/vm/stack_frame_ia32.h
File runtime/vm/stack_frame_ia32.h (right):

https://codereview.chromium.org/15110003/diff/1/runtime/vm/stack_frame_ia32.h...
runtime/vm/stack_frame_ia32.h:37: static const int kSavedCallerPpSlotFromFp =
kSavedCallerFpSlotFromFp;
This looks fragile. Maybe implement as a getter instead and call UNIMPLEMENTED
for accesses in ia32, x64?

[zra, 2013-05-14 22:28:56]

lgtm

https://codereview.chromium.org/15110003/diff/1/runtime/vm/flow_graph_compile...
File runtime/vm/flow_graph_compiler_arm.cc (right):

https://codereview.chromium.org/15110003/diff/1/runtime/vm/flow_graph_compile...
runtime/vm/flow_graph_compiler_arm.cc:156: set_pc_offset(assem->CodeSize());
On ia32 and x64, there is a breakpoint here. Should there be one here, and in
the same place in MIPS?

[regis, 2013-05-14 23:23:42]

Thanks!

https://codereview.chromium.org/15110003/diff/1/runtime/vm/deopt_instructions.cc
File runtime/vm/deopt_instructions.cc (right):

https://codereview.chromium.org/15110003/diff/1/runtime/vm/deopt_instructions...
runtime/vm/deopt_instructions.cc:494: // Callee's PC marker is not used anymore.
Set to 0.
On 2013/05/14 22:16:39, srdjan wrote:
> When can this occur?
The Deoptimize stub sets up a Dart frame that is copied and filled by leaf
runtime calls. This pc marker slot is not used (pc of stub). The corresponding
deopt instruction is:
builder->AddPcMarker(Function::Handle(), slot_ix++);
I edited the comment to make it clearer.

https://codereview.chromium.org/15110003/diff/1/runtime/vm/flow_graph_compile...
File runtime/vm/flow_graph_compiler_arm.cc (right):

https://codereview.chromium.org/15110003/diff/1/runtime/vm/flow_graph_compile...
runtime/vm/flow_graph_compiler_arm.cc:156: set_pc_offset(assem->CodeSize());
On 2013/05/14 22:28:56, zra wrote:
> On ia32 and x64, there is a breakpoint here. Should there be one here, and in
> the same place in MIPS?

It is safer to have one, but it takes up space. I added it with a TODO to remove
it to save space later.

https://codereview.chromium.org/15110003/diff/1/runtime/vm/stack_frame_ia32.h
File runtime/vm/stack_frame_ia32.h (right):

https://codereview.chromium.org/15110003/diff/1/runtime/vm/stack_frame_ia32.h...
runtime/vm/stack_frame_ia32.h:37: static const int kSavedCallerPpSlotFromFp =
kSavedCallerFpSlotFromFp;
On 2013/05/14 22:16:39, srdjan wrote:
> This looks fragile. Maybe implement as a getter instead and call UNIMPLEMENTED
> for accesses in ia32, x64?

I added a getter in the common header and left these constants here. I cannot
declare a global getter here without running into duplicate definition issues.

[regis, 2013-05-14 23:24:02]

Committed patchset #2 manually as r22719 (presubmit successful).