VM: [AOT] Make sure that we inline all resolved accessor invocations.

VM: [AOT] Make sure that we inline all resolved accessor invocations.

Previously we would just replace it with a static call and hope that inliner
handles it. However sometimes it happens too late in the pipeline and inliner
misses the opportunity.

Additionally drop some dead code from the aot_optimizer that was not doing
anything useful because FlowGraphInliner::TryInlineRecognizedMethod always
returns false in the AOT mode.

R=fschneider@google.com
BUG=

Committed: https://github.com/dart-lang/sdk/commit/31057dcec7419ef6b8ddc0d458989f7bad2fcc81

[Vyacheslav Egorov (Google), 2016-06-24 14:06:11]

A lot of the code in the AOT pipeline feels really clanky because we use ICData
to communicate inferred type information. I think we should refactor this stuff
away and use something else.

[Florian Schneider, 2016-06-24 16:09:58]

https://codereview.chromium.org/2098643003/diff/1/runtime/vm/aot_optimizer.cc
File runtime/vm/aot_optimizer.cc (left):

https://codereview.chromium.org/2098643003/diff/1/runtime/vm/aot_optimizer.cc...
runtime/vm/aot_optimizer.cc:741: bool
AotOptimizer::TryReplaceWithIndexedOp(InstanceCallInstr* call) {
I'd keep this code around:

FlowGraphInliner::TryInlineRecognizedMethod should inline certain methods even
in AOT.

It does not do it now - but that seems like an oversight.

Also by moving some of this code into FlowGraphInliner, we can avoid duplication
between jit_optimizer.cc and aot_optimizer.cc.

[Vyacheslav Egorov (Google), 2016-06-24 16:13:05]

https://codereview.chromium.org/2098643003/diff/1/runtime/vm/aot_optimizer.cc
File runtime/vm/aot_optimizer.cc (left):

https://codereview.chromium.org/2098643003/diff/1/runtime/vm/aot_optimizer.cc...
runtime/vm/aot_optimizer.cc:741: bool
AotOptimizer::TryReplaceWithIndexedOp(InstanceCallInstr* call) {
I think it does not make much sense to keep around the code that does not do
anything right now - it's just called and does essentially nothing (just
allocates some zone memory :))

Also as you say: when we actually start enabling it we should strive to unify it
as much as possible with the JIT - which means that this code will not survive
in its current form anyway.

[Florian Schneider, 2016-06-24 22:48:17]

https://codereview.chromium.org/2098643003/diff/1/runtime/vm/aot_optimizer.cc
File runtime/vm/aot_optimizer.cc (left):

https://codereview.chromium.org/2098643003/diff/1/runtime/vm/aot_optimizer.cc...
runtime/vm/aot_optimizer.cc:741: bool
AotOptimizer::TryReplaceWithIndexedOp(InstanceCallInstr* call) {
On 2016/06/24 16:13:04, Vyacheslav Egorov (Google) wrote:
> I think it does not make much sense to keep around the code that does not do
> anything right now - it's just called and does essentially nothing (just
> allocates some zone memory :))
> 
> Also as you say: when we actually start enabling it we should strive to unify
it
> as much as possible with the JIT - which means that this code will not survive
> in its current form anyway. 

We should make TryInlineRecognizedMethod work with the AOT optimizer. I can look
into that. I think it was just forgotten to enable this code...

[Vyacheslav Egorov (Google), 2016-06-27 11:01:22]

Please take another look. 

I restored all code that does nothing, will look into refactoring it in such a
way that it does something in a later change.

[Florian Schneider, 2016-06-27 14:17:26]

Lgtm.

btw: I can also look into enabling inlining of recognized methods for AOT.

[Vyacheslav Egorov (Google), 2016-06-27 15:15:30]

Description was changed from

==========
VM: [AOT] Make sure that we inline all resolved accessor invocations.

Previously we would just replace it with a static call and hope that inliner
handles it. However sometimes it happens too late in the pipeline and inliner
misses the opportunity.

Additionally drop some dead code from the aot_optimizer that was not doing
anything useful because FlowGraphInliner::TryInlineRecognizedMethod always
returns false in the AOT mode.

R=fschneider@google.com
BUG=
==========

to

==========
VM: [AOT] Make sure that we inline all resolved accessor invocations.

Previously we would just replace it with a static call and hope that inliner
handles it. However sometimes it happens too late in the pipeline and inliner
misses the opportunity.

Additionally drop some dead code from the aot_optimizer that was not doing
anything useful because FlowGraphInliner::TryInlineRecognizedMethod always
returns false in the AOT mode.

R=fschneider@google.com
BUG=

Committed:
https://github.com/dart-lang/sdk/commit/31057dcec7419ef6b8ddc0d458989f7bad2fcc81
==========

[Vyacheslav Egorov (Google), 2016-06-27 15:15:31]

Committed patchset #2 (id:20001) manually as
31057dcec7419ef6b8ddc0d458989f7bad2fcc81 (presubmit successful).