Deletion barrier: Distinguish+verify field initialization in ia32 generated code.

Distinguish between initialization and updates to object fields. Currently only for DEBUG mode verification, but will be needed for deletion barrier.

Split up initialization flag in StoreInstanceFieldInstr for needed more precise semantics.

Zap swept old space with same value as uninitialized new in debug mode.

R=fschneider@google.com

Committed: https://code.google.com/p/dart/source/detail?r=42717

[koda, 2015-01-05 20:06:21]

koda@google.com changed reviewers:
+ iposva@google.com

[koda, 2015-01-08 15:42:43]

koda@google.com changed reviewers:
+ fschneider@google.com

[koda, 2015-01-08 15:45:05]

fschneider: Could you please have a quick look just at the changes to
intermediate_language*? I've refined the initialization flag, as I previously
discussed with you.

[Florian Schneider, 2015-01-08 15:56:44]

intermediate_language* lgtm.

[Ivan Posva, 2015-01-08 19:34:52]

LGTMwC

-Ivan

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia3...
File runtime/vm/assembler_ia32.cc (right):

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia3...
runtime/vm/assembler_ia32.cc:2290: testb(dest, Immediate(kHeapObjectTag));
// Check for Smi.

or as above explain the "// Non-smi case: ..." which would be even easier to
follow.

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia3...
runtime/vm/assembler_ia32.cc:2400: VerifySmi(dest, "New value must be Smi.");
Why are you checking that the value is Smi after the write? Shouldn't we check
before so that we have a chance to see what the value is that was overwritten?

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia3...
runtime/vm/assembler_ia32.cc:2405: void Assembler::ZeroSmiField(const Address&
dest) {
Is this only used in initializing writes?

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia32.h
File runtime/vm/assembler_ia32.h (right):

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia3...
runtime/vm/assembler_ia32.h:675: kEmptySmiOrNull,  // Empty = garbage/zapped in
release/debug mode.
Does this mean an empty Smi or kEmptyOrSmiOrNull?

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/flow_graph_op...
File runtime/vm/flow_graph_optimizer.cc (right):

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/flow_graph_op...
runtime/vm/flow_graph_optimizer.cc:7142: return
!store_instance->is_potential_unboxed_initialization() &&
Personally I find this variant easier to read:
!(is_potential_unboxed_initialization || is_object_reference_initialization)

[koda, 2015-01-08 20:25:34]

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia3...
File runtime/vm/assembler_ia32.cc (right):

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia3...
runtime/vm/assembler_ia32.cc:2290: testb(dest, Immediate(kHeapObjectTag));
On 2015/01/08 19:34:51, Ivan Posva wrote:
> // Check for Smi.
> 
> or as above explain the "// Non-smi case: ..." which would be even easier to
> follow.

Done.

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia3...
runtime/vm/assembler_ia32.cc:2400: VerifySmi(dest, "New value must be Smi.");
On 2015/01/08 19:34:51, Ivan Posva wrote:
> Why are you checking that the value is Smi after the write? Shouldn't we check
> before so that we have a chance to see what the value is that was overwritten?

Done.

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia3...
runtime/vm/assembler_ia32.cc:2405: void Assembler::ZeroSmiField(const Address&
dest) {
On 2015/01/08 19:34:51, Ivan Posva wrote:
> Is this only used in initializing writes?

Yes. Renamed to clarify this.

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia32.h
File runtime/vm/assembler_ia32.h (right):

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/assembler_ia3...
runtime/vm/assembler_ia32.h:675: kEmptySmiOrNull,  // Empty = garbage/zapped in
release/debug mode.
On 2015/01/08 19:34:51, Ivan Posva wrote:
> Does this mean an empty Smi or kEmptyOrSmiOrNull?

Renamed to clarify.

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/flow_graph_op...
File runtime/vm/flow_graph_optimizer.cc (right):

https://codereview.chromium.org/836593002/diff/80001/runtime/vm/flow_graph_op...
runtime/vm/flow_graph_optimizer.cc:7142: return
!store_instance->is_potential_unboxed_initialization() &&
On 2015/01/08 19:34:52, Ivan Posva wrote:
> Personally I find this variant easier to read:
> !(is_potential_unboxed_initialization || is_object_reference_initialization)

Done.

[koda, 2015-01-08 20:45:52]

Committed patchset #6 (id:100001) manually as r42717 (presubmit successful).